diff options
| author | Linus Torvalds | 2019-03-09 15:53:03 -0800 |
|---|---|---|
| committer | Linus Torvalds | 2019-03-09 15:53:03 -0800 |
| commit | a50243b1ddcdd766d0d17fbfeeb1a22e62fdc461 (patch) | |
| tree | 3dbf847105558eaac3658a46c4934df503c866a2 | |
| parent | 2901752c14b8e1b7dd898d2e5245c93e531aa624 (diff) | |
| parent | fca22e7e595f1799cfbfdfa13e16d48ece0d136c (diff) | |
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"This has been a slightly more active cycle than normal with ongoing
core changes and quite a lot of collected driver updates.
- Various driver fixes for bnxt_re, cxgb4, hns, mlx5, pvrdma, rxe
- A new data transfer mode for HFI1 giving higher performance
- Significant functional and bug fix update to the mlx5
On-Demand-Paging MR feature
- A chip hang reset recovery system for hns
- Change mm->pinned_vm to an atomic64
- Update bnxt_re to support a new 57500 chip
- A sane netlink 'rdma link add' method for creating rxe devices and
fixing the various unregistration race conditions in rxe's
unregister flow
- Allow lookup up objects by an ID over netlink
- Various reworking of the core to driver interface:
- drivers should not assume umem SGLs are in PAGE_SIZE chunks
- ucontext is accessed via udata not other means
- start to make the core code responsible for object memory
allocation
- drivers should convert struct device to struct ib_device via a
helper
- drivers have more tools to avoid use after unregister problems"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (280 commits)
net/mlx5: ODP support for XRC transport is not enabled by default in FW
IB/hfi1: Close race condition on user context disable and close
RDMA/umem: Revert broken 'off by one' fix
RDMA/umem: minor bug fix in error handling path
RDMA/hns: Use GFP_ATOMIC in hns_roce_v2_modify_qp
cxgb4: kfree mhp after the debug print
IB/rdmavt: Fix concurrency panics in QP post_send and modify to error
IB/rdmavt: Fix loopback send with invalidate ordering
IB/iser: Fix dma_nents type definition
IB/mlx5: Set correct write permissions for implicit ODP MR
bnxt_re: Clean cq for kernel consumers only
RDMA/uverbs: Don't do double free of allocated PD
RDMA: Handle ucontext allocations by IB/core
RDMA/core: Fix a WARN() message
bnxt_re: fix the regression due to changes in alloc_pbl
IB/mlx4: Increase the timeout for CM cache
IB/core: Abort page fault handler silently during owning process exit
IB/mlx5: Validate correct PD before prefetch MR
IB/mlx5: Protect against prefetch of invalid MR
RDMA/uverbs: Store PR pointer before it is overwritten
...
264 files changed, 16710 insertions, 5014 deletions
diff --git a/.clang-format b/.clang-format index bc2ffb2a0b53..201a4f531b90 100644 --- a/.clang-format +++ b/.clang-format @@ -240,6 +240,7 @@ ForEachMacros: - 'for_each_set_bit' - 'for_each_set_bit_from' - 'for_each_sg' + - 'for_each_sg_dma_page' - 'for_each_sg_page' - 'for_each_sibling_event' - '__for_each_thread' @@ -360,6 +361,7 @@ ForEachMacros: - 'radix_tree_for_each_slot' - 'radix_tree_for_each_tagged' - 'rbtree_postorder_for_each_entry_safe' + - 'rdma_for_each_port' - 'resource_list_for_each_entry' - 'resource_list_for_each_entry_safe' - 'rhl_for_each_entry_rcu' diff --git a/Documentation/infiniband/user_verbs.txt b/Documentation/infiniband/user_verbs.txt index df049b9f5b6e..47ebf2f80b2b 100644 --- a/Documentation/infiniband/user_verbs.txt +++ b/Documentation/infiniband/user_verbs.txt @@ -46,11 +46,11 @@ Memory pinning I/O targets be kept resident at the same physical address. The ib_uverbs module manages pinning and unpinning memory regions via get_user_pages() and put_page() calls. It also accounts for the - amount of memory pinned in the process's locked_vm, and checks that + amount of memory pinned in the process's pinned_vm, and checks that unprivileged processes do not exceed their RLIMIT_MEMLOCK limit. Pages that are pinned multiple times are counted each time they are - pinned, so the value of locked_vm may be an overestimate of the + pinned, so the value of pinned_vm may be an overestimate of the number of pages pinned by a process. /dev files diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c index 31786b200afc..a3357ff7540d 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c @@ -311,7 +311,13 @@ static dma_addr_t __vmw_piter_dma_addr(struct vmw_piter *viter) static dma_addr_t __vmw_piter_sg_addr(struct vmw_piter *viter) { - return sg_page_iter_dma_address(&viter->iter); + /* + * FIXME: This driver wrongly mixes DMA and CPU SG list iteration and + * needs revision. See + * https://lore.kernel.org/lkml/20190104223531.GA1705@ziepe.ca/ + */ + return sg_page_iter_dma_address( + container_of(&viter->iter, struct sg_dma_page_iter, base)); } diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig index 0a3ec7c726ec..a1fb840de45d 100644 --- a/drivers/infiniband/Kconfig +++ b/drivers/infiniband/Kconfig @@ -89,6 +89,7 @@ config INFINIBAND_ADDR_TRANS_CONFIGFS This allows the user to config the default GID type that the CM uses for each device, when initiaing new connections. +if INFINIBAND_USER_ACCESS || !INFINIBAND_USER_ACCESS source "drivers/infiniband/hw/mthca/Kconfig" source "drivers/infiniband/hw/qib/Kconfig" source "drivers/infiniband/hw/cxgb3/Kconfig" @@ -101,6 +102,12 @@ source "drivers/infiniband/hw/ocrdma/Kconfig" source "drivers/infiniband/hw/vmw_pvrdma/Kconfig" source "drivers/infiniband/hw/usnic/Kconfig" source "drivers/infiniband/hw/hns/Kconfig" +source "drivers/infiniband/hw/bnxt_re/Kconfig" +source "drivers/infiniband/hw/hfi1/Kconfig" +source "drivers/infiniband/hw/qedr/Kconfig" +source "drivers/infiniband/sw/rdmavt/Kconfig" +source "drivers/infiniband/sw/rxe/Kconfig" +endif source "drivers/infiniband/ulp/ipoib/Kconfig" @@ -111,13 +118,5 @@ source "drivers/infiniband/ulp/iser/Kconfig" source "drivers/infiniband/ulp/isert/Kconfig" source "drivers/infiniband/ulp/opa_vnic/Kconfig" -source "drivers/infiniband/sw/rdmavt/Kconfig" -source "drivers/infiniband/sw/rxe/Kconfig" - -source "drivers/infiniband/hw/hfi1/Kconfig" - -source "drivers/infiniband/hw/qedr/Kconfig" - -source "drivers/infiniband/hw/bnxt_re/Kconfig" endif # INFINIBAND diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile index 69dee36e0e89..313f2349b518 100644 --- a/drivers/infiniband/core/Makefile +++ b/drivers/infiniband/core/Makefile @@ -15,8 +15,6 @@ ib_core-y := packer.o ud_header.o verbs.o cq.o rw.o sysfs.o \ nldev.o restrack.o ib_core-$(CONFIG_SECURITY_INFINIBAND) += security.o -ib_core-$(CONFIG_INFINIBAND_USER_MEM) += umem.o -ib_core-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += umem_odp.o ib_core-$(CONFIG_CGROUP_RDMA) += cgroup.o ib_cm-y := cm.o @@ -39,3 +37,5 @@ ib_uverbs-y := uverbs_main.o uverbs_cmd.o uverbs_marshall.o \ uverbs_std_types_flow_action.o uverbs_std_types_dm.o \ uverbs_std_types_mr.o uverbs_std_types_counters.o \ uverbs_uapi.o uverbs_std_types_device.o +ib_uverbs-$(CONFIG_INFINIBAND_USER_MEM) += umem.o +ib_uverbs-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += umem_odp.o diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c index 7b04590f307f..43c67e5f43c6 100644 --- a/drivers/infiniband/core/cache.c +++ b/drivers/infiniband/core/cache.c @@ -185,7 +185,7 @@ EXPORT_SYMBOL(ib_cache_gid_parse_type_str); static struct ib_gid_table *rdma_gid_table(struct ib_device *device, u8 port) { - return device->cache.ports[port - rdma_start_port(device)].gid; + return device->port_data[port].cache.gid; } static bool is_gid_entry_free(const struct ib_gid_table_entry *entry) @@ -547,21 +547,19 @@ int ib_cache_gid_add(struct ib_device *ib_dev, u8 port, unsigned long mask; int ret; - if (ib_dev->ops.get_netdev) { - idev = ib_dev->ops.get_netdev(ib_dev, port); - if (idev && attr->ndev != idev) { - union ib_gid default_gid; + idev = ib_device_get_netdev(ib_dev, port); + if (idev && attr->ndev != idev) { + union ib_gid default_gid; - /* Adding default GIDs in not permitted */ - make_default_gid(idev, &default_gid); - if (!memcmp(gid, &default_gid, sizeof(*gid))) { - dev_put(idev); - return -EPERM; - } - } - if (idev) + /* Adding default GIDs is not permitted */ + make_default_gid(idev, &default_gid); + if (!memcmp(gid, &default_gid, sizeof(*gid))) { dev_put(idev); + return -EPERM; + } } + if (idev) + dev_put(idev); mask = GID_ATTR_FIND_MASK_GID | GID_ATTR_FIND_MASK_GID_TYPE | @@ -765,7 +763,7 @@ err_free_table: return NULL; } -static void release_gid_table(struct ib_device *device, u8 port, +static void release_gid_table(struct ib_device *device, struct ib_gid_table *table) { bool leak = false; @@ -863,31 +861,27 @@ static void gid_table_reserve_default(struct ib_device *ib_dev, u8 port, static void gid_table_release_one(struct ib_device *ib_dev) { - struct ib_gid_table *table; - u8 port; + unsigned int p; - for (port = 0; port < ib_dev->phys_port_cnt; port++) { - table = ib_dev->cache.ports[port].gid; - release_gid_table(ib_dev, port, table); - ib_dev->cache.ports[port].gid = NULL; + rdma_for_each_port (ib_dev, p) { + release_gid_table(ib_dev, ib_dev->port_data[p].cache.gid); + ib_dev->port_data[p].cache.gid = NULL; } } static int _gid_table_setup_one(struct ib_device *ib_dev) { - u8 port; struct ib_gid_table *table; + unsigned int rdma_port; - for (port = 0; port < ib_dev->phys_port_cnt; port++) { - u8 rdma_port = port + rdma_start_port(ib_dev); - - table = alloc_gid_table( - ib_dev->port_immutable[rdma_port].gid_tbl_len); + rdma_for_each_port (ib_dev, rdma_port) { + table = alloc_gid_table( + ib_dev->port_data[rdma_port].immutable.gid_tbl_len); if (!table) goto rollback_table_setup; gid_table_reserve_default(ib_dev, rdma_port, table); - ib_dev->cache.ports[port].gid = table; + ib_dev->port_data[rdma_port].cache.gid = table; } return 0; @@ -898,14 +892,11 @@ rollback_table_setup: static void gid_table_cleanup_one(struct ib_device *ib_dev) { - struct ib_gid_table *table; - u8 port; + unsigned int p; - for (port = 0; port < ib_dev->phys_port_cnt; port++) { - table = ib_dev->cache.ports[port].gid; - cleanup_gid_table_port(ib_dev, port + rdma_start_port(ib_dev), - table); - } + rdma_for_each_port (ib_dev, p) + cleanup_gid_table_port(ib_dev, p, + ib_dev->port_data[p].cache.gid); } static int gid_table_setup_one(struct ib_device *ib_dev) @@ -983,17 +974,17 @@ const struct ib_gid_attr *rdma_find_gid(struct ib_device *device, unsigned long mask = GID_ATTR_FIND_MASK_GID | GID_ATTR_FIND_MASK_GID_TYPE; struct ib_gid_attr gid_attr_val = {.ndev = ndev, .gid_type = gid_type}; - u8 p; + unsigned int p; if (ndev) mask |= GID_ATTR_FIND_MASK_NETDEV; - for (p = 0; p < device->phys_port_cnt; p++) { + rdma_for_each_port(device, p) { struct ib_gid_table *table; unsigned long flags; int index; - table = device->cache.ports[p].gid; + table = device->port_data[p].cache.gid; read_lock_irqsave(&table->rwlock, flags); index = find_gid(table, gid, &gid_attr_val, false, mask, NULL); if (index >= 0) { @@ -1025,7 +1016,7 @@ int ib_get_cached_pkey(struct ib_device *device, read_lock_irqsave(&device->cache.lock, flags); - cache = device->cache.ports[port_num - rdma_start_port(device)].pkey; + cache = device->port_data[port_num].cache.pkey; if (index < 0 || index >= cache->table_len) ret = -EINVAL; @@ -1043,14 +1034,12 @@ int ib_get_cached_subnet_prefix(struct ib_device *device, u64 *sn_pfx) { unsigned long flags; - int p; if (!rdma_is_port_valid(device, port_num)) return -EINVAL; - p = port_num - rdma_start_port(device); read_lock_irqsave(&device->cache.lock, flags); - *sn_pfx = device->cache.ports[p].subnet_prefix; + *sn_pfx = device->port_data[port_num].cache.subnet_prefix; read_unlock_irqrestore(&device->cache.lock, flags); return 0; @@ -1073,7 +1062,7 @@ int ib_find_cached_pkey(struct ib_device *device, read_lock_irqsave(&device->cache.lock, flags); - cache = device->cache.ports[port_num - rdma_start_port(device)].pkey; + cache = device->port_data[port_num].cache.pkey; *index = -1; @@ -1113,7 +1102,7 @@ int ib_find_exact_cached_pkey(struct ib_device *device, read_lock_irqsave(&device->cache.lock, flags); - cache = device->cache.ports[port_num - rdma_start_port(device)].pkey; + cache = device->port_data[port_num].cache.pkey; *index = -1; @@ -1141,7 +1130,7 @@ int ib_get_cached_lmc(struct ib_device *device, return -EINVAL; read_lock_irqsave(&device->cache.lock, flags); - *lmc = device->cache.ports[port_num - rdma_start_port(device)].lmc; + *lmc = device->port_data[port_num].cache.lmc; read_unlock_irqrestore(&device->cache.lock, flags); return ret; @@ -1159,8 +1148,7 @@ int ib_get_cached_port_state(struct ib_device *device, return -EINVAL; read_lock_irqsave(&device->cache.lock, flags); - *port_state = device->cache.ports[port_num - - rdma_start_port(device)].port_state; + *port_state = device->port_data[port_num].cache.port_state; read_unlock_irqrestore(&device->cache.lock, flags); return ret; @@ -1361,16 +1349,13 @@ static void ib_cache_update(struct ib_device *device, write_lock_irq(&device->cache.lock); - old_pkey_cache = device->cache.ports[port - - rdma_start_port(device)].pkey; + old_pkey_cache = device->port_data[port].cache.pkey; - device->cache.ports[port - rdma_start_port(device)].pkey = pkey_cache; - device->cache.ports[port - rdma_start_port(device)].lmc = tprops->lmc; - device->cache.ports[port - rdma_start_port(device)].port_state = - tprops->state; + device->port_data[port].cache.pkey = pkey_cache; + device->port_data[port].cache.lmc = tprops->lmc; + device->port_data[port].cache.port_state = tprops->state; - device->cache.ports[port - rdma_start_port(device)].subnet_prefix = - tprops->subnet_prefix; + device->port_data[port].cache.subnet_prefix = tprops->subnet_prefix; write_unlock_irq(&device->cache.lock); if (enforce_security) @@ -1428,27 +1413,17 @@ static void ib_cache_event(struct ib_event_handler *handler, int ib_cache_setup_one(struct ib_device *device) { - int p; + unsigned int p; int err; rwlock_init(&device->cache.lock); - device->cache.ports = - kcalloc(rdma_end_port(device) - rdma_start_port(device) + 1, - sizeof(*device->cache.ports), - GFP_KERNEL); - if (!device->cache.ports) - return -ENOMEM; - err = gid_table_setup_one(device); - if (err) { - kfree(device->cache.ports); - device->cache.ports = NULL; + if (err) return err; - } - for (p = 0; p <= rdma_end_port(device) - rdma_start_port(device); ++p) - ib_cache_update(device, p + rdma_start_port(device), true); + rdma_for_each_port (device, p) + ib_cache_update(device, p, true); INIT_IB_EVENT_HANDLER(&device->cache.event_handler, device, ib_cache_event); @@ -1458,7 +1433,7 @@ int ib_cache_setup_one(struct ib_device *device) void ib_cache_release_one(struct ib_device *device) { - int p; + unsigned int p; /* * The release function frees all the cache elements. @@ -1466,11 +1441,10 @@ void ib_cache_release_one(struct ib_device *device) * all the device's resources when the cache could no * longer be accessed. */ - for (p = 0; p <= rdma_end_port(device) - rdma_start_port(device); ++p) - kfree(device->cache.ports[p].pkey); + rdma_for_each_port (device, p) + kfree(device->port_data[p].cache.pkey); gid_table_release_one(device); - kfree(device->cache.ports); } void ib_cache_cleanup_one(struct ib_device *device) diff --git a/drivers/infiniband/core/cgroup.c b/drivers/infiniband/core/cgroup.c index 126ac5f99db7..388fd04e5f63 100644 --- a/drivers/infiniband/core/cgroup.c +++ b/drivers/infiniband/core/cgroup.c @@ -21,12 +21,11 @@ * Register with the rdma cgroup. Should be called before * exposing rdma device to user space applications to avoid * resource accounting leak. - * Returns 0 on success or otherwise failure code. */ -int ib_device_register_rdmacg(struct ib_device *device) +void ib_device_register_rdmacg(struct ib_device *device) { device->cg_device.name = device->name; - return rdmacg_register_device(&device->cg_device); + rdmacg_register_device(&device->cg_device); } /** diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c index 37980c7564c0..b9416a6fca36 100644 --- a/drivers/infiniband/core/cm.c +++ b/drivers/infiniband/core/cm.c @@ -4052,8 +4052,7 @@ static void cm_recv_handler(struct ib_mad_agent *mad_agent, atomic_long_inc(&port->counter_group[CM_RECV]. counter[attr_id - CM_ATTR_ID_OFFSET]); - work = kmalloc(sizeof(*work) + sizeof(struct sa_path_rec) * paths, - GFP_KERNEL); + work = kmalloc(struct_size(work, path, paths), GFP_KERNEL); if (!work) { ib_free_recv_mad(mad_recv_wc); return; diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 84f077b2b90a..68c997be2429 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -659,7 +659,7 @@ static int cma_acquire_dev_by_src_ip(struct rdma_id_private *id_priv) struct cma_device *cma_dev; enum ib_gid_type gid_type; int ret = -ENODEV; - u8 port; + unsigned int port; if (dev_addr->dev_type != ARPHRD_INFINIBAND && id_priv->id.ps == RDMA_PS_IPOIB) @@ -673,8 +673,7 @@ static int cma_acquire_dev_by_src_ip(struct rdma_id_private *id_priv) mutex_lock(&lock); list_for_each_entry(cma_dev, &dev_list, list) { - for (port = rdma_start_port(cma_dev->device); - port <= rdma_end_port(cma_dev->device); port++) { + rdma_for_each_port (cma_dev->device, port) { gidp = rdma_protocol_roce(cma_dev->device, port) ? &iboe_gid : &gid; gid_type = cma_dev->default_gid_type[port - 1]; @@ -888,6 +887,7 @@ struct rdma_cm_id *__rdma_create_id(struct net *net, id_priv->id.ps = ps; id_priv->id.qp_type = qp_type; id_priv->tos_set = false; + id_priv->timeout_set = false; id_priv->gid_type = IB_GID_TYPE_IB; spin_lock_init(&id_priv->lock); mutex_init(&id_priv->qp_mutex); @@ -1130,6 +1130,9 @@ int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr, } else ret = -ENOSYS; + if ((*qp_attr_mask & IB_QP_TIMEOUT) && id_priv->timeout_set) + qp_attr->timeout = id_priv->timeout; + return ret; } EXPORT_SYMBOL(rdma_init_qp_attr); @@ -2410,6 +2413,7 @@ static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog) return PTR_ERR(id); id->tos = id_priv->tos; + id->tos_set = id_priv->tos_set; id_priv->cm_id.iw = id; memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv), @@ -2462,6 +2466,8 @@ static void cma_listen_on_dev(struct rdma_id_private *id_priv, atomic_inc(&id_priv->refcount); dev_id_priv->internal_id = 1; dev_id_priv->afonly = id_priv->afonly; + dev_id_priv->tos_set = id_priv->tos_set; + dev_id_priv->tos = id_priv->tos; ret = rdma_listen(id, id_priv->backlog); if (ret) @@ -2490,6 +2496,34 @@ void rdma_set_service_type(struct rdma_cm_id *id, int tos) } EXPORT_SYMBOL(rdma_set_service_type); +/** + * rdma_set_ack_timeout() - Set the ack timeout of QP associated + * with a connection identifier. + * @id: Communication identifier to associated with service type. + * @timeout: Ack timeout to set a QP, expressed as 4.096 * 2^(timeout) usec. + * + * This function should be called before rdma_connect() on active side, + * and on passive side before rdma_accept(). It is applicable to primary + * path only. The timeout will affect the local side of the QP, it is not + * negotiated with remote side and zero disables the timer. + * + * Return: 0 for success + */ +int rdma_set_ack_timeout(struct rdma_cm_id *id, u8 timeout) +{ + struct rdma_id_private *id_priv; + + if (id->qp_type != IB_QPT_RC) + return -EINVAL; + + id_priv = container_of(id, struct rdma_id_private, id); + id_priv->timeout = timeout; + id_priv->timeout_set = true; + + return 0; +} +EXPORT_SYMBOL(rdma_set_ack_timeout); + static void cma_query_handler(int status, struct sa_path_rec *path_rec, void *context) { @@ -2966,13 +3000,22 @@ static void addr_handler(int status, struct sockaddr *src_addr, { struct rdma_id_private *id_priv = context; struct rdma_cm_event event = {}; + struct sockaddr *addr; + struct sockaddr_storage old_addr; mutex_lock(&id_priv->handler_mutex); if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_RESOLVED)) goto out; - memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr)); + /* + * Store the previous src address, so that if we fail to acquire + * matching rdma device, old address can be restored back, which helps + * to cancel the cma listen operation correctly. + */ + addr = cma_src_addr(id_priv); + memcpy(&old_addr, addr, rdma_addr_size(addr)); + memcpy(addr, src_addr, rdma_addr_size(src_addr)); if (!status && !id_priv->cma_dev) { status = cma_acquire_dev_by_src_ip(id_priv); if (status) @@ -2983,6 +3026,8 @@ static void addr_handler(int status, struct sockaddr *src_addr, } if (status) { + memcpy(addr, &old_addr, + rdma_addr_size((struct sockaddr *)&old_addr)); if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ADDR_BOUND)) goto out; @@ -3798,6 +3843,7 @@ static int cma_connect_iw(struct rdma_id_private *id_priv, return PTR_ERR(cm_id); cm_id->tos = id_priv->tos; + cm_id->tos_set = id_priv->tos_set; id_priv->cm_id.iw = cm_id; memcpy(&cm_id->local_addr, cma_src_addr(id_priv), @@ -4501,7 +4547,7 @@ static void cma_add_one(struct ib_device *device) if (!cma_dev->default_roce_tos) goto free_gid_type; - for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) { + rdma_for_each_port (device, i) { supported_gids = roce_gid_type_mask_support(device, i); WARN_ON(!supported_gids); if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE)) @@ -4605,85 +4651,6 @@ static void cma_remove_one(struct ib_device *device, void *client_data) kfree(cma_dev); } -static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb) -{ - struct nlmsghdr *nlh; - struct rdma_cm_id_stats *id_stats; - struct rdma_id_private *id_priv; - struct rdma_cm_id *id = NULL; - struct cma_device *cma_dev; - int i_dev = 0, i_id = 0; - - /* - * We export all of the IDs as a sequence of messages. Each - * ID gets its own netlink message. - */ - mutex_lock(&lock); - - list_for_each_entry(cma_dev, &dev_list, list) { - if (i_dev < cb->args[0]) { - i_dev++; - continue; - } - - i_id = 0; - list_for_each_entry(id_priv, &cma_dev->id_list, list) { - if (i_id < cb->args[1]) { - i_id++; - continue; - } - - id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq, - sizeof *id_stats, RDMA_NL_RDMA_CM, - RDMA_NL_RDMA_CM_ID_STATS, - NLM_F_MULTI); - if (!id_stats) - goto out; - - memset(id_stats, 0, sizeof *id_stats); - id = &id_priv->id; - id_stats->node_type = id->route.addr.dev_addr.dev_type; - id_stats->port_num = id->port_num; - id_stats->bound_dev_if = - id->route.addr.dev_addr.bound_dev_if; - - if (ibnl_put_attr(skb, nlh, - rdma_addr_size(cma_src_addr(id_priv)), - cma_src_addr(id_priv), - RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) - goto out; - if (ibnl_put_attr(skb, nlh, - rdma_addr_size(cma_dst_addr(id_priv)), - cma_dst_addr(id_priv), - RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) - goto out; - - id_stats->pid = task_pid_vnr(id_priv->res.task); - id_stats->port_space = id->ps; - id_stats->cm_state = id_priv->state; - id_stats->qp_num = id_priv->qp_num; - id_stats->qp_type = id->qp_type; - - i_id++; - nlmsg_end(skb, nlh); - } - - cb->args[1] = 0; - i_dev++; - } - -out: - mutex_unlock(&lock); - cb->args[0] = i_dev; - cb->args[1] = i_id; - - return skb->len; -} - -static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = { - [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats}, -}; - static int cma_init_net(struct net *net) { struct cma_pernet *pernet = cma_pernet(net); @@ -4732,7 +4699,6 @@ static int __init cma_init(void) if (ret) goto err; - rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table); cma_configfs_init(); return 0; @@ -4748,7 +4714,6 @@ err_wq: static void __exit cma_cleanup(void) { cma_configfs_exit(); - rdma_nl_unregister(RDMA_NL_RDMA_CM); ib_unregister_client(&cma_client); unregister_netdevice_notifier(&cma_nb); ib_sa_unregister_client(&sa_client); @@ -4756,7 +4721,5 @@ static void __exit cma_cleanup(void) destroy_workqueue(cma_wq); } -MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1); - module_init(cma_init); module_exit(cma_cleanup); diff --git a/drivers/infiniband/core/cma_priv.h b/drivers/infiniband/core/cma_priv.h index cf47c69436a7..ca7307277518 100644 --- a/drivers/infiniband/core/cma_priv.h +++ b/drivers/infiniband/core/cma_priv.h @@ -84,9 +84,11 @@ struct rdma_id_private { u32 options; u8 srq; u8 tos; - bool tos_set; + u8 tos_set:1; + u8 timeout_set:1; u8 reuseaddr; u8 afonly; + u8 timeout; enum ib_gid_type gid_type; /* diff --git a/drivers/infiniband/core/core_priv.h b/drivers/infiniband/core/core_priv.h index 616734313f0c..08c690249594 100644 --- a/drivers/infiniband/core/core_priv.h +++ b/drivers/infiniband/core/core_priv.h @@ -54,9 +54,9 @@ struct pkey_index_qp_list { struct list_head qp_list; }; -int ib_device_register_sysfs(struct ib_device *device, - int (*port_callback)(struct ib_device *, - u8, struct kobject *)); +extern const struct attribute_group ib_dev_attr_group; + +int ib_device_register_sysfs(struct ib_device *device); void ib_device_unregister_sysfs(struct ib_device *device); int ib_device_rename(struct ib_device *ibdev, const char *name); @@ -66,6 +66,9 @@ typedef void (*roce_netdev_callback)(struct ib_device *device, u8 port, typedef bool (*roce_netdev_filter)(struct ib_device *device, u8 port, struct net_device *idev, void *cookie); +struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, + unsigned int port); + void ib_enum_roce_netdev(struct ib_device *ib_dev, roce_netdev_filter filter, void *filter_cookie, @@ -117,7 +120,7 @@ void ib_cache_cleanup_one(struct ib_device *device); void ib_cache_release_one(struct ib_device *device); #ifdef CONFIG_CGROUP_RDMA -int ib_device_register_rdmacg(struct ib_device *device); +void ib_device_register_rdmacg(struct ib_device *device); void ib_device_unregister_rdmacg(struct ib_device *device); int ib_rdmacg_try_charge(struct ib_rdmacg_object *cg_obj, @@ -128,21 +131,26 @@ void ib_rdmacg_uncharge(struct ib_rdmacg_object *cg_obj, struct ib_device *device, enum rdmacg_resource_type resource_index); #else -static inline int ib_device_register_rdmacg(struct ib_device *device) -{ return 0; } +static inline void ib_device_register_rdmacg(struct ib_device *device) +{ +} static inline void ib_device_unregister_rdmacg(struct ib_device *device) -{ } +{ +} static inline int ib_rdmacg_try_charge(struct ib_rdmacg_object *cg_obj, struct ib_device *device, enum rdmacg_resource_type resource_index) -{ return 0; } +{ + return 0; +} static inline void ib_rdmacg_uncharge(struct ib_rdmacg_object *cg_obj, struct ib_device *device, enum rdmacg_resource_type resource_index) -{ } +{ +} #endif static inline bool rdma_is_upper_dev_rcu(struct net_device *dev, @@ -178,7 +186,7 @@ int ib_get_cached_subnet_prefix(struct ib_device *device, u64 *sn_pfx); #ifdef CONFIG_SECURITY_INFINIBAND -void ib_security_destroy_port_pkey_list(struct ib_device *device); +void ib_security_release_port_pkey_list(struct ib_device *device); void ib_security_cache_change(struct ib_device *device, u8 port_num, @@ -199,8 +207,9 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent, enum ib_qp_type qp_type); void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent); int ib_mad_enforce_security(struct ib_mad_agent_private *map, u16 pkey_index); +void ib_mad_agent_security_change(void); #else -static inline void ib_security_destroy_port_pkey_list(struct ib_device *device) +static inline void ib_security_release_port_pkey_list(struct ib_device *device) { } @@ -264,6 +273,10 @@ static inline int ib_mad_enforce_security(struct ib_mad_agent_private *map, { return 0; } + +static inline void ib_mad_agent_security_change(void) +{ +} #endif struct ib_device *ib_device_get_by_index(u32 ifindex); diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c index 238ec42778ef..a9f29156e486 100644 --- a/drivers/infiniband/core/device.c +++ b/drivers/infiniband/core/device.c @@ -37,54 +37,111 @@ #include <linux/kernel.h> #include <linux/slab.h> #include <linux/init.h> -#include <linux/mutex.h> #include <linux/netdevice.h> #include <linux/security.h> #include <linux/notifier.h> +#include <linux/hashtable.h> #include <rdma/rdma_netlink.h> #include <rdma/ib_addr.h> #include <rdma/ib_cache.h> #include "core_priv.h" +#include "restrack.h" MODULE_AUTHOR("Roland Dreier"); MODULE_DESCRIPTION("core kernel InfiniBand API"); MODULE_LICENSE("Dual BSD/GPL"); -struct ib_client_data { - struct list_head list; - struct ib_client *client; - void * data; - /* The device or client is going down. Do not call client or device - * callbacks other than remove(). */ - bool going_down; -}; - struct workqueue_struct *ib_comp_wq; struct workqueue_struct *ib_comp_unbound_wq; struct workqueue_struct *ib_wq; EXPORT_SYMBOL_GPL(ib_wq); -/* The device_list and client_list contain devices and clients after their - * registration has completed, and the devices and clients are removed - * during unregistration. */ -static LIST_HEAD(device_list); -static LIST_HEAD(client_list); +/* + * Each of the three rwsem locks (devices, clients, client_data) protects the + * xarray of the same name. Specifically it allows the caller to assert that + * the MARK will/will not be changing under the lock, and for devices and + * clients, that the value in the xarray is still a valid pointer. Change of + * the MARK is linked to the object state, so holding the lock and testing the + * MARK also asserts that the contained object is in a certain state. + * + * This is used to build a two stage register/unregister flow where objects + * can continue to be in the xarray even though they are still in progress to + * register/unregister. + * + * The xarray itself provides additional locking, and restartable iteration, + * which is also relied on. + * + * Locks should not be nested, with the exception of client_data, which is + * allowed to nest under the read side of the other two locks. + * + * The devices_rwsem also protects the device name list, any change or + * assignment of device name must also hold the write side to guarantee unique + * names. + */ /* - * device_mutex and lists_rwsem protect access to both device_list and - * client_list. device_mutex protects writer access by device and client - * registration / de-registration. lists_rwsem protects reader access to - * these lists. Iterators of these lists must lock it for read, while updates - * to the lists must be done with a write lock. A special case is when the - * device_mutex is locked. In this case locking the lists for read access is - * not necessary as the device_mutex implies it. + * devices contains devices that have had their names assigned. The + * devices may not be registered. Users that care about the registration + * status need to call ib_device_try_get() on the device to ensure it is + * registered, and keep it registered, for the required duration. * - * lists_rwsem also protects access to the client data list. */ -static DEFINE_MUTEX(device_mutex); -static DECLARE_RWSEM(lists_rwsem); +static DEFINE_XARRAY_FLAGS(devices, XA_FLAGS_ALLOC); +static DECLARE_RWSEM(devices_rwsem); +#define DEVICE_REGISTERED XA_MARK_1 +static LIST_HEAD(client_list); +#define CLIENT_REGISTERED XA_MARK_1 +static DEFINE_XARRAY_FLAGS(clients, XA_FLAGS_ALLOC); +static DECLARE_RWSEM(clients_rwsem); + +/* + * If client_data is registered then the corresponding client must also still + * be registered. + */ +#define CLIENT_DATA_REGISTERED XA_MARK_1 +/* + * xarray has this behavior where it won't iterate over NULL values stored in + * allocated arrays. So we need our own iterator to see all values stored in + * the array. This does the same thing as xa_for_each except that it also + * returns NULL valued entries if the array is allocating. Simplified to only + * work on simple xarrays. + */ +static void *xan_find_marked(struct xarray *xa, unsigned long *indexp, + xa_mark_t filter) +{ + XA_STATE(xas, xa, *indexp); + void *entry; + + rcu_read_lock(); + do { + entry = xas_find_marked(&xas, ULONG_MAX, filter); + if (xa_is_zero(entry)) + break; + } while (xas_retry(&xas, entry)); + rcu_read_unlock(); + + if (entry) { + *indexp = xas.xa_index; + if (xa_is_zero(entry)) + return NULL; + return entry; + } + return XA_ERROR(-ENOENT); +} +#define xan_for_each_marked(xa, index, entry, filter) \ + for (index = 0, entry = xan_find_marked(xa, &(index), filter); \ + !xa_is_err(entry); \ + (index)++, entry = xan_find_marked(xa, &(index), filter)) + +/* RCU hash table mapping netdevice pointers to struct ib_port_data */ +static DEFINE_SPINLOCK(ndev_hash_lock); +static DECLARE_HASHTABLE(ndev_hash, 5); + +static void free_netdevs(struct ib_device *ib_dev); +static void ib_unregister_work(struct work_struct *work); +static void __ib_unregister_device(struct ib_device *device); static int ib_security_change(struct notifier_block *nb, unsigned long event, void *lsm_data); static void ib_policy_change_task(struct work_struct *work); @@ -94,6 +151,12 @@ static struct notifier_block ibdev_lsm_nb = { .notifier_call = ib_security_change, }; +/* Pointer to the RCU head at the start of the ib_port_data array */ +struct ib_port_data_rcu { + struct rcu_head rcu_head; + struct ib_port_data pdata[]; +}; + static int ib_device_check_mandatory(struct ib_device *device) { #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device_ops, x), #x } @@ -121,30 +184,18 @@ static int ib_device_check_mandatory(struct ib_device *device) }; int i; + device->kverbs_provider = true; for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { if (!*(void **) ((void *) &device->ops + mandatory_table[i].offset)) { - dev_warn(&device->dev, - "Device is missing mandatory function %s\n", - mandatory_table[i].name); - return -EINVAL; + device->kverbs_provider = false; + break; } } return 0; } -static struct ib_device *__ib_device_get_by_index(u32 index) -{ - struct ib_device *device; - - list_for_each_entry(device, &device_list, core_list) - if (device->index == index) - return device; - - return NULL; -} - /* * Caller must perform ib_device_put() to return the device reference count * when ib_device_get_by_index() returns valid device pointer. @@ -153,13 +204,13 @@ struct ib_device *ib_device_get_by_index(u32 index) { struct ib_device *device; - down_read(&lists_rwsem); - device = __ib_device_get_by_index(index); + down_read(&devices_rwsem); + device = xa_load(&devices, index); if (device) { if (!ib_device_try_get(device)) device = NULL; } - up_read(&lists_rwsem); + up_read(&devices_rwsem); return device; } @@ -180,28 +231,56 @@ EXPORT_SYMBOL(ib_device_put); static struct ib_device *__ib_device_get_by_name(const char *name) { struct ib_device *device; + unsigned long index; - list_for_each_entry(device, &device_list, core_list) + xa_for_each (&devices, index, device) if (!strcmp(name, dev_name(&device->dev))) return device; return NULL; } -int ib_device_rename(struct ib_device *ibdev, const char *name) +/** + * ib_device_get_by_name - Find an IB device by name + * @name: The name to look for + * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) + * + * Find and hold an ib_device by its name. The caller must call + * ib_device_put() on the returned pointer. + */ +struct ib_device *ib_device_get_by_name(const char *name, + enum rdma_driver_id driver_id) { struct ib_device *device; - int ret = 0; - if (!strcmp(name, dev_name(&ibdev->dev))) - return ret; + down_read(&devices_rwsem); + device = __ib_device_get_by_name(name); + if (device && driver_id != RDMA_DRIVER_UNKNOWN && + device->driver_id != driver_id) + device = NULL; - mutex_lock(&device_mutex); - list_for_each_entry(device, &device_list, core_list) { - if (!strcmp(name, dev_name(&device->dev))) { - ret = -EEXIST; - goto out; - } + if (device) { + if (!ib_device_try_get(device)) + device = NULL; + } + up_read(&devices_rwsem); + return device; +} +EXPORT_SYMBOL(ib_device_get_by_name); + +int ib_device_rename(struct ib_device *ibdev, const char *name) +{ + int ret; + + down_write(&devices_rwsem); + if (!strcmp(name, dev_name(&ibdev->dev))) { + ret = 0; + goto out; + } + + if (__ib_device_get_by_name(name)) { + ret = -EEXIST; + goto out; } ret = device_rename(&ibdev->dev, name); @@ -209,53 +288,60 @@ int ib_device_rename(struct ib_device *ibdev, const char *name) goto out; strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); out: - mutex_unlock(&device_mutex); + up_write(&devices_rwsem); return ret; } static int alloc_name(struct ib_device *ibdev, const char *name) { - unsigned long *inuse; struct ib_device *device; + unsigned long index; + struct ida inuse; + int rc; int i; - inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL); - if (!inuse) - return -ENOMEM; - - list_for_each_entry(device, &device_list, core_list) { + lockdep_assert_held_exclusive(&devices_rwsem); + ida_init(&inuse); + xa_for_each (&devices, index, device) { char buf[IB_DEVICE_NAME_MAX]; if (sscanf(dev_name(&device->dev), name, &i) != 1) continue; - if (i < 0 || i >= PAGE_SIZE * 8) + if (i < 0 || i >= INT_MAX) continue; snprintf(buf, sizeof buf, name, i); - if (!strcmp(buf, dev_name(&device->dev))) - set_bit(i, inuse); + if (strcmp(buf, dev_name(&device->dev)) != 0) + continue; + + rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); + if (rc < 0) + goto out; } - i = find_first_zero_bit(inuse, PAGE_SIZE * 8); - free_page((unsigned long) inuse); + rc = ida_alloc(&inuse, GFP_KERNEL); + if (rc < 0) + goto out; - return dev_set_name(&ibdev->dev, name, i); + rc = dev_set_name(&ibdev->dev, name, rc); +out: + ida_destroy(&inuse); + return rc; } static void ib_device_release(struct device *device) { struct ib_device *dev = container_of(device, struct ib_device, dev); - WARN_ON(dev->reg_state == IB_DEV_REGISTERED); - if (dev->reg_state == IB_DEV_UNREGISTERED) { - /* - * In IB_DEV_UNINITIALIZED state, cache or port table - * is not even created. Free cache and port table only when - * device reaches UNREGISTERED state. - */ - ib_cache_release_one(dev); - kfree(dev->port_immutable); - } - kfree(dev); + free_netdevs(dev); + WARN_ON(refcount_read(&dev->refcount)); + ib_cache_release_one(dev); + ib_security_release_port_pkey_list(dev); + xa_destroy(&dev->client_data); + if (dev->port_data) + kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, + pdata[0]), + rcu_head); + kfree_rcu(dev, rcu_head); } static int ib_device_uevent(struct device *device, @@ -278,7 +364,7 @@ static struct class ib_class = { }; /** - * ib_alloc_device - allocate an IB device struct + * _ib_alloc_device - allocate an IB device struct * @size:size of structure to allocate * * Low-level drivers should use ib_alloc_device() to allocate &struct @@ -287,7 +373,7 @@ static struct class ib_class = { * ib_dealloc_device() must be used to free structures allocated with * ib_alloc_device(). */ -struct ib_device *ib_alloc_device(size_t size) +struct ib_device *_ib_alloc_device(size_t size) { struct ib_device *device; @@ -298,23 +384,32 @@ struct ib_device *ib_alloc_device(size_t size) if (!device) return NULL; - rdma_restrack_init(&device->res); + if (rdma_restrack_init(device)) { + kfree(device); + return NULL; + } device->dev.class = &ib_class; + device->groups[0] = &ib_dev_attr_group; + device->dev.groups = device->groups; device_initialize(&device->dev); - dev_set_drvdata(&device->dev, device); - INIT_LIST_HEAD(&device->event_handler_list); spin_lock_init(&device->event_handler_lock); - rwlock_init(&device->client_data_lock); - INIT_LIST_HEAD(&device->client_data_list); + mutex_init(&device->unregistration_lock); + /* + * client_data needs to be alloc because we don't want our mark to be + * destroyed if the user stores NULL in the client data. + */ + xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); + init_rwsem(&device->client_data_rwsem); INIT_LIST_HEAD(&device->port_list); init_completion(&device->unreg_completion); + INIT_WORK(&device->unregistration_work, ib_unregister_work); return device; } -EXPORT_SYMBOL(ib_alloc_device); +EXPORT_SYMBOL(_ib_alloc_device); /** * ib_dealloc_device - free an IB device struct @@ -324,32 +419,153 @@ EXPORT_SYMBOL(ib_alloc_device); */ void ib_dealloc_device(struct ib_device *device) { - WARN_ON(!list_empty(&device->client_data_list)); - WARN_ON(device->reg_state != IB_DEV_UNREGISTERED && - device->reg_state != IB_DEV_UNINITIALIZED); - rdma_restrack_clean(&device->res); + if (device->ops.dealloc_driver) + device->ops.dealloc_driver(device); + + /* + * ib_unregister_driver() requires all devices to remain in the xarray + * while their ops are callable. The last op we call is dealloc_driver + * above. This is needed to create a fence on op callbacks prior to + * allowing the driver module to unload. + */ + down_write(&devices_rwsem); + if (xa_load(&devices, device->index) == device) + xa_erase(&devices, device->index); + up_write(&devices_rwsem); + + /* Expedite releasing netdev references */ + free_netdevs(device); + + WARN_ON(!xa_empty(&device->client_data)); + WARN_ON(refcount_read(&device->refcount)); + rdma_restrack_clean(device); + /* Balances with device_initialize */ put_device(&device->dev); } EXPORT_SYMBOL(ib_dealloc_device); -static int add_client_context(struct ib_device *device, struct ib_client *client) +/* + * add_client_context() and remove_client_context() must be safe against + * parallel calls on the same device - registration/unregistration of both the + * device and client can be occurring in parallel. + * + * The routines need to be a fence, any caller must not return until the add + * or remove is fully completed. + */ +static int add_client_context(struct ib_device *device, + struct ib_client *client) { - struct ib_client_data *context; + int ret = 0; - context = kmalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return -ENOMEM; + if (!device->kverbs_provider && !client->no_kverbs_req) + return 0; + + down_write(&device->client_data_rwsem); + /* + * Another caller to add_client_context got here first and has already + * completely initialized context. + */ + if (xa_get_mark(&device->client_data, client->client_id, + CLIENT_DATA_REGISTERED)) + goto out; + + ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, + GFP_KERNEL)); + if (ret) + goto out; + downgrade_write(&device->client_data_rwsem); + if (client->add) + client->add(device); + + /* Readers shall not see a client until add has been completed */ + xa_set_mark(&device->client_data, client->client_id, + CLIENT_DATA_REGISTERED); + up_read(&device->client_data_rwsem); + return 0; - context->client = client; - context->data = NULL; - context->going_down = false; +out: + up_write(&device->client_data_rwsem); + return ret; +} - down_write(&lists_rwsem); - write_lock_irq(&device->client_data_lock); - list_add(&context->list, &device->client_data_list); - write_unlock_irq(&device->client_data_lock); - up_write(&lists_rwsem); +static void remove_client_context(struct ib_device *device, + unsigned int client_id) +{ + struct ib_client *client; + void *client_data; + + down_write(&device->client_data_rwsem); + if (!xa_get_mark(&device->client_data, client_id, + CLIENT_DATA_REGISTERED)) { + up_write(&device->client_data_rwsem); + return; + } + client_data = xa_load(&device->client_data, client_id); + xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); + client = xa_load(&clients, client_id); + downgrade_write(&device->client_data_rwsem); + + /* + * Notice we cannot be holding any exclusive locks when calling the + * remove callback as the remove callback can recurse back into any + * public functions in this module and thus try for any locks those + * functions take. + * + * For this reason clients and drivers should not call the + * unregistration functions will holdling any locks. + * + * It tempting to drop the client_data_rwsem too, but this is required + * to ensure that unregister_client does not return until all clients + * are completely unregistered, which is required to avoid module + * unloading races. + */ + if (client->remove) + client->remove(device, client_data); + + xa_erase(&device->client_data, client_id); + up_read(&device->client_data_rwsem); +} + +static int alloc_port_data(struct ib_device *device) +{ + struct ib_port_data_rcu *pdata_rcu; + unsigned int port; + + if (device->port_data) + return 0; + /* This can only be called once the physical port range is defined */ + if (WARN_ON(!device->phys_port_cnt)) + return -EINVAL; + + /* + * device->port_data is indexed directly by the port number to make + * access to this data as efficient as possible. + * + * Therefore port_data is declared as a 1 based array with potential + * empty slots at the beginning. + */ + pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, + rdma_end_port(device) + 1), + GFP_KERNEL); + if (!pdata_rcu) + return -ENOMEM; + /* + * The rcu_head is put in front of the port data array and the stored + * pointer is adjusted since we never need to see that member until + * kfree_rcu. + */ + device->port_data = pdata_rcu->pdata; + + rdma_for_each_port (device, port) { + struct ib_port_data *pdata = &device->port_data[port]; + + pdata->ib_dev = device; + spin_lock_init(&pdata->pkey_list_lock); + INIT_LIST_HEAD(&pdata->pkey_list); + spin_lock_init(&pdata->netdev_lock); + INIT_HLIST_NODE(&pdata->ndev_hash_link); + } return 0; } @@ -359,29 +575,20 @@ static int verify_immutable(const struct ib_device *dev, u8 port) rdma_max_mad_size(dev, port) != 0); } -static int read_port_immutable(struct ib_device *device) +static int setup_port_data(struct ib_device *device) { + unsigned int port; int ret; - u8 start_port = rdma_start_port(device); - u8 end_port = rdma_end_port(device); - u8 port; - /** - * device->port_immutable is indexed directly by the port number to make - * access to this data as efficient as possible. - * - * Therefore port_immutable is declared as a 1 based array with - * potential empty slots at the beginning. - */ - device->port_immutable = kcalloc(end_port + 1, - sizeof(*device->port_immutable), - GFP_KERNEL); - if (!device->port_immutable) - return -ENOMEM; + ret = alloc_port_data(device); + if (ret) + return ret; - for (port = start_port; port <= end_port; ++port) { - ret = device->ops.get_port_immutable( - device, port, &device->port_immutable[port]); + rdma_for_each_port (device, port) { + struct ib_port_data *pdata = &device->port_data[port]; + + ret = device->ops.get_port_immutable(device, port, + &pdata->immutable); if (ret) return ret; @@ -400,39 +607,16 @@ void ib_get_device_fw_str(struct ib_device *dev, char *str) } EXPORT_SYMBOL(ib_get_device_fw_str); -static int setup_port_pkey_list(struct ib_device *device) -{ - int i; - - /** - * device->port_pkey_list is indexed directly by the port number, - * Therefore it is declared as a 1 based array with potential empty - * slots at the beginning. - */ - device->port_pkey_list = kcalloc(rdma_end_port(device) + 1, - sizeof(*device->port_pkey_list), - GFP_KERNEL); - - if (!device->port_pkey_list) - return -ENOMEM; - - for (i = 0; i < (rdma_end_port(device) + 1); i++) { - spin_lock_init(&device->port_pkey_list[i].list_lock); - INIT_LIST_HEAD(&device->port_pkey_list[i].pkey_list); - } - - return 0; -} - static void ib_policy_change_task(struct work_struct *work) { struct ib_device *dev; + unsigned long index; - down_read(&lists_rwsem); - list_for_each_entry(dev, &device_list, core_list) { - int i; + down_read(&devices_rwsem); + xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { + unsigned int i; - for (i = rdma_start_port(dev); i <= rdma_end_port(dev); i++) { + rdma_for_each_port (dev, i) { u64 sp; int ret = ib_get_cached_subnet_prefix(dev, i, @@ -445,7 +629,7 @@ static void ib_policy_change_task(struct work_struct *work) ib_security_cache_change(dev, i, sp); } } - up_read(&lists_rwsem); + up_read(&devices_rwsem); } static int ib_security_change(struct notifier_block *nb, unsigned long event, @@ -455,32 +639,52 @@ static int ib_security_change(struct notifier_block *nb, unsigned long event, return NOTIFY_DONE; schedule_work(&ib_policy_change_work); + ib_mad_agent_security_change(); return NOTIFY_OK; } -/** - * __dev_new_index - allocate an device index - * - * Returns a suitable unique value for a new device interface - * number. It assumes that there are less than 2^32-1 ib devices - * will be present in the system. +/* + * Assign the unique string device name and the unique device index. This is + * undone by ib_dealloc_device. */ -static u32 __dev_new_index(void) +static int assign_name(struct ib_device *device, const char *name) { - /* - * The device index to allow stable naming. - * Similar to struct net -> ifindex. - */ - static u32 index; + static u32 last_id; + int ret; - for (;;) { - if (!(++index)) - index = 1; + down_write(&devices_rwsem); + /* Assign a unique name to the device */ + if (strchr(name, '%')) + ret = alloc_name(device, name); + else + ret = dev_set_name(&device->dev, name); + if (ret) + goto out; + + if (__ib_device_get_by_name(dev_name(&device->dev))) { + ret = -ENFILE; + goto out; + } + strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); - if (!__ib_device_get_by_index(index)) - return index; + /* Cyclically allocate a user visible ID for the device */ + device->index = last_id; + ret = xa_alloc(&devices, &device->index, INT_MAX, device, GFP_KERNEL); + if (ret == -ENOSPC) { + device->index = 0; + ret = xa_alloc(&devices, &device->index, INT_MAX, device, + GFP_KERNEL); } + if (ret) + goto out; + last_id = device->index + 1; + + ret = 0; + +out: + up_write(&devices_rwsem); + return ret; } static void setup_dma_device(struct ib_device *device) @@ -518,27 +722,25 @@ static void setup_dma_device(struct ib_device *device) } } -static void cleanup_device(struct ib_device *device) -{ - ib_cache_cleanup_one(device); - ib_cache_release_one(device); - kfree(device->port_pkey_list); - kfree(device->port_immutable); -} - +/* + * setup_device() allocates memory and sets up data that requires calling the + * device ops, this is the only reason these actions are not done during + * ib_alloc_device. It is undone by ib_dealloc_device(). + */ static int setup_device(struct ib_device *device) { struct ib_udata uhw = {.outlen = 0, .inlen = 0}; int ret; + setup_dma_device(device); + ret = ib_device_check_mandatory(device); if (ret) return ret; - ret = read_port_immutable(device); + ret = setup_port_data(device); if (ret) { - dev_warn(&device->dev, - "Couldn't create per port immutable data\n"); + dev_warn(&device->dev, "Couldn't create per-port data\n"); return ret; } @@ -547,27 +749,76 @@ static int setup_device(struct ib_device *device) if (ret) { dev_warn(&device->dev, "Couldn't query the device attributes\n"); - goto port_cleanup; + return ret; } - ret = setup_port_pkey_list(device); - if (ret) { - dev_warn(&device->dev, "Couldn't create per port_pkey_list\n"); - goto port_cleanup; + return 0; +} + +static void disable_device(struct ib_device *device) +{ + struct ib_client *client; + + WARN_ON(!refcount_read(&device->refcount)); + + down_write(&devices_rwsem); + xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); + up_write(&devices_rwsem); + + down_read(&clients_rwsem); + list_for_each_entry_reverse(client, &client_list, list) + remove_client_context(device, client->client_id); + up_read(&clients_rwsem); + + /* Pairs with refcount_set in enable_device */ + ib_device_put(device); + wait_for_completion(&device->unreg_completion); + + /* Expedite removing unregistered pointers from the hash table */ + free_netdevs(device); +} + +/* + * An enabled device is visible to all clients and to all the public facing + * APIs that return a device pointer. This always returns with a new get, even + * if it fails. + */ +static int enable_device_and_get(struct ib_device *device) +{ + struct ib_client *client; + unsigned long index; + int ret = 0; + + /* + * One ref belongs to the xa and the other belongs to this + * thread. This is needed to guard against parallel unregistration. + */ + refcount_set(&device->refcount, 2); + down_write(&devices_rwsem); + xa_set_mark(&devices, device->index, DEVICE_REGISTERED); + + /* + * By using downgrade_write() we ensure that no other thread can clear + * DEVICE_REGISTERED while we are completing the client setup. + */ + downgrade_write(&devices_rwsem); + + if (device->ops.enable_driver) { + ret = device->ops.enable_driver(device); + if (ret) + goto out; } - ret = ib_cache_setup_one(device); - if (ret) { - dev_warn(&device->dev, - "Couldn't set up InfiniBand P_Key/GID cache\n"); - goto pkey_cleanup; + down_read(&clients_rwsem); + xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { + ret = add_client_context(device, client); + if (ret) + break; } - return 0; + up_read(&clients_rwsem); -pkey_cleanup: - kfree(device->port_pkey_list); -port_cleanup: - kfree(device->port_immutable); +out: + up_read(&devices_rwsem); return ret; } @@ -579,133 +830,253 @@ port_cleanup: * devices with the IB core. All registered clients will receive a * callback for each device that is added. @device must be allocated * with ib_alloc_device(). + * + * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() + * asynchronously then the device pointer may become freed as soon as this + * function returns. */ -int ib_register_device(struct ib_device *device, const char *name, - int (*port_callback)(struct ib_device *, u8, - struct kobject *)) +int ib_register_device(struct ib_device *device, const char *name) { int ret; - struct ib_client *client; - - setup_dma_device(device); - - mutex_lock(&device_mutex); - if (strchr(name, '%')) { - ret = alloc_name(device, name); - if (ret) - goto out; - } else { - ret = dev_set_name(&device->dev, name); - if (ret) - goto out; - } - if (__ib_device_get_by_name(dev_name(&device->dev))) { - ret = -ENFILE; - goto out; - } - strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); + ret = assign_name(device, name); + if (ret) + return ret; ret = setup_device(device); if (ret) - goto out; - - device->index = __dev_new_index(); + return ret; - ret = ib_device_register_rdmacg(device); + ret = ib_cache_setup_one(device); if (ret) { dev_warn(&device->dev, - "Couldn't register device with rdma cgroup\n"); - goto dev_cleanup; + "Couldn't set up InfiniBand P_Key/GID cache\n"); + return ret; } - ret = ib_device_register_sysfs(device, port_callback); + ib_device_register_rdmacg(device); + + ret = device_add(&device->dev); + if (ret) + goto cg_cleanup; + + ret = ib_device_register_sysfs(device); if (ret) { dev_warn(&device->dev, "Couldn't register device with driver model\n"); - goto cg_cleanup; + goto dev_cleanup; } - refcount_set(&device->refcount, 1); - device->reg_state = IB_DEV_REGISTERED; + ret = enable_device_and_get(device); + if (ret) { + void (*dealloc_fn)(struct ib_device *); - list_for_each_entry(client, &client_list, list) - if (!add_client_context(device, client) && client->add) - client->add(device); + /* + * If we hit this error flow then we don't want to + * automatically dealloc the device since the caller is + * expected to call ib_dealloc_device() after + * ib_register_device() fails. This is tricky due to the + * possibility for a parallel unregistration along with this + * error flow. Since we have a refcount here we know any + * parallel flow is stopped in disable_device and will see the + * NULL pointers, causing the responsibility to + * ib_dealloc_device() to revert back to this thread. + */ + dealloc_fn = device->ops.dealloc_driver; + device->ops.dealloc_driver = NULL; + ib_device_put(device); + __ib_unregister_device(device); + device->ops.dealloc_driver = dealloc_fn; + return ret; + } + ib_device_put(device); - down_write(&lists_rwsem); - list_add_tail(&device->core_list, &device_list); - up_write(&lists_rwsem); - mutex_unlock(&device_mutex); return 0; +dev_cleanup: + device_del(&device->dev); cg_cleanup: ib_device_unregister_rdmacg(device); -dev_cleanup: - cleanup_device(device); -out: - mutex_unlock(&device_mutex); + ib_cache_cleanup_one(device); return ret; } EXPORT_SYMBOL(ib_register_device); +/* Callers must hold a get on the device. */ +static void __ib_unregister_device(struct ib_device *ib_dev) +{ + /* + * We have a registration lock so that all the calls to unregister are + * fully fenced, once any unregister returns the device is truely + * unregistered even if multiple callers are unregistering it at the + * same time. This also interacts with the registration flow and + * provides sane semantics if register and unregister are racing. + */ + mutex_lock(&ib_dev->unregistration_lock); + if (!refcount_read(&ib_dev->refcount)) + goto out; + + disable_device(ib_dev); + ib_device_unregister_sysfs(ib_dev); + device_del(&ib_dev->dev); + ib_device_unregister_rdmacg(ib_dev); + ib_cache_cleanup_one(ib_dev); + + /* + * Drivers using the new flow may not call ib_dealloc_device except + * in error unwind prior to registration success. + */ + if (ib_dev->ops.dealloc_driver) { + WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); + ib_dealloc_device(ib_dev); + } +out: + mutex_unlock(&ib_dev->unregistration_lock); +} + /** * ib_unregister_device - Unregister an IB device - * @device:Device to unregister + * @device: The device to unregister * * Unregister an IB device. All clients will receive a remove callback. + * + * Callers should call this routine only once, and protect against races with + * registration. Typically it should only be called as part of a remove + * callback in an implementation of driver core's struct device_driver and + * related. + * + * If ops.dealloc_driver is used then ib_dev will be freed upon return from + * this function. */ -void ib_unregister_device(struct ib_device *device) +void ib_unregister_device(struct ib_device *ib_dev) { - struct ib_client_data *context, *tmp; - unsigned long flags; + get_device(&ib_dev->dev); + __ib_unregister_device(ib_dev); + put_device(&ib_dev->dev); +} +EXPORT_SYMBOL(ib_unregister_device); - /* - * Wait for all netlink command callers to finish working on the - * device. - */ - ib_device_put(device); - wait_for_completion(&device->unreg_completion); +/** + * ib_unregister_device_and_put - Unregister a device while holding a 'get' + * device: The device to unregister + * + * This is the same as ib_unregister_device(), except it includes an internal + * ib_device_put() that should match a 'get' obtained by the caller. + * + * It is safe to call this routine concurrently from multiple threads while + * holding the 'get'. When the function returns the device is fully + * unregistered. + * + * Drivers using this flow MUST use the driver_unregister callback to clean up + * their resources associated with the device and dealloc it. + */ +void ib_unregister_device_and_put(struct ib_device *ib_dev) +{ + WARN_ON(!ib_dev->ops.dealloc_driver); + get_device(&ib_dev->dev); + ib_device_put(ib_dev); + __ib_unregister_device(ib_dev); + put_device(&ib_dev->dev); +} +EXPORT_SYMBOL(ib_unregister_device_and_put); - mutex_lock(&device_mutex); +/** + * ib_unregister_driver - Unregister all IB devices for a driver + * @driver_id: The driver to unregister + * + * This implements a fence for device unregistration. It only returns once all + * devices associated with the driver_id have fully completed their + * unregistration and returned from ib_unregister_device*(). + * + * If device's are not yet unregistered it goes ahead and starts unregistering + * them. + * + * This does not block creation of new devices with the given driver_id, that + * is the responsibility of the caller. + */ +void ib_unregister_driver(enum rdma_driver_id driver_id) +{ + struct ib_device *ib_dev; + unsigned long index; - down_write(&lists_rwsem); - list_del(&device->core_list); - write_lock_irq(&device->client_data_lock); - list_for_each_entry(context, &device->client_data_list, list) - context->going_down = true; - write_unlock_irq(&device->client_data_lock); - downgrade_write(&lists_rwsem); + down_read(&devices_rwsem); + xa_for_each (&devices, index, ib_dev) { + if (ib_dev->driver_id != driver_id) + continue; - list_for_each_entry(context, &device->client_data_list, list) { - if (context->client->remove) - context->client->remove(device, context->data); + get_device(&ib_dev->dev); + up_read(&devices_rwsem); + + WARN_ON(!ib_dev->ops.dealloc_driver); + __ib_unregister_device(ib_dev); + + put_device(&ib_dev->dev); + down_read(&devices_rwsem); } - up_read(&lists_rwsem); + up_read(&devices_rwsem); +} +EXPORT_SYMBOL(ib_unregister_driver); - ib_device_unregister_sysfs(device); - ib_device_unregister_rdmacg(device); +static void ib_unregister_work(struct work_struct *work) +{ + struct ib_device *ib_dev = + container_of(work, struct ib_device, unregistration_work); - mutex_unlock(&device_mutex); + __ib_unregister_device(ib_dev); + put_device(&ib_dev->dev); +} - ib_cache_cleanup_one(device); +/** + * ib_unregister_device_queued - Unregister a device using a work queue + * device: The device to unregister + * + * This schedules an asynchronous unregistration using a WQ for the device. A + * driver should use this to avoid holding locks while doing unregistration, + * such as holding the RTNL lock. + * + * Drivers using this API must use ib_unregister_driver before module unload + * to ensure that all scheduled unregistrations have completed. + */ +void ib_unregister_device_queued(struct ib_device *ib_dev) +{ + WARN_ON(!refcount_read(&ib_dev->refcount)); + WARN_ON(!ib_dev->ops.dealloc_driver); + get_device(&ib_dev->dev); + if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) + put_device(&ib_dev->dev); +} +EXPORT_SYMBOL(ib_unregister_device_queued); - ib_security_destroy_port_pkey_list(device); - kfree(device->port_pkey_list); +static int assign_client_id(struct ib_client *client) +{ + int ret; - down_write(&lists_rwsem); - write_lock_irqsave(&device->client_data_lock, flags); - list_for_each_entry_safe(context, tmp, &device->client_data_list, - list) { - list_del(&context->list); - kfree(context); - } - write_unlock_irqrestore(&device->client_data_lock, flags); - up_write(&lists_rwsem); + down_write(&clients_rwsem); + /* + * The add/remove callbacks must be called in FIFO/LIFO order. To + * achieve this we assign client_ids so they are sorted in + * registration order, and retain a linked list we can reverse iterate + * to get the LIFO order. The extra linked list can go away if xarray + * learns to reverse iterate. + */ + if (list_empty(&client_list)) + client->client_id = 0; + else + client->client_id = + list_last_entry(&client_list, struct ib_client, list) + ->client_id; + ret = xa_alloc(&clients, &client->client_id, INT_MAX, client, + GFP_KERNEL); + if (ret) + goto out; + + xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); + list_add_tail(&client->list, &client_list); - device->reg_state = IB_DEV_UNREGISTERED; +out: + up_write(&clients_rwsem); + return ret; } -EXPORT_SYMBOL(ib_unregister_device); /** * ib_register_client - Register an IB client @@ -723,19 +1094,23 @@ EXPORT_SYMBOL(ib_unregister_device); int ib_register_client(struct ib_client *client) { struct ib_device *device; + unsigned long index; + int ret; - mutex_lock(&device_mutex); - - list_for_each_entry(device, &device_list, core_list) - if (!add_client_context(device, client) && client->add) - client->add(device); - - down_write(&lists_rwsem); - list_add_tail(&client->list, &client_list); - up_write(&lists_rwsem); - - mutex_unlock(&device_mutex); + ret = assign_client_id(client); + if (ret) + return ret; + down_read(&devices_rwsem); + xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { + ret = add_client_context(device, client); + if (ret) { + up_read(&devices_rwsem); + ib_unregister_client(client); + return ret; + } + } + up_read(&devices_rwsem); return 0; } EXPORT_SYMBOL(ib_register_client); @@ -747,108 +1122,56 @@ EXPORT_SYMBOL(ib_register_client); * Upper level users use ib_unregister_client() to remove their client * registration. When ib_unregister_client() is called, the client * will receive a remove callback for each IB device still registered. + * + * This is a full fence, once it returns no client callbacks will be called, + * or are running in another thread. */ void ib_unregister_client(struct ib_client *client) { - struct ib_client_data *context; struct ib_device *device; + unsigned long index; - mutex_lock(&device_mutex); + down_write(&clients_rwsem); + xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); + up_write(&clients_rwsem); + /* + * Every device still known must be serialized to make sure we are + * done with the client callbacks before we return. + */ + down_read(&devices_rwsem); + xa_for_each (&devices, index, device) + remove_client_context(device, client->client_id); + up_read(&devices_rwsem); - down_write(&lists_rwsem); + down_write(&clients_rwsem); list_del(&client->list); - up_write(&lists_rwsem); - - list_for_each_entry(device, &device_list, core_list) { - struct ib_client_data *found_context = NULL; - - down_write(&lists_rwsem); - write_lock_irq(&device->client_data_lock); - list_for_each_entry(context, &device->client_data_list, list) - if (context->client == client) { - context->going_down = true; - found_context = context; - break; - } - write_unlock_irq(&device->client_data_lock); - up_write(&lists_rwsem); - - if (client->remove) - client->remove(device, found_context ? - found_context->data : NULL); - - if (!found_context) { - dev_warn(&device->dev, - "No client context found for %s\n", - client->name); - continue; - } - - down_write(&lists_rwsem); - write_lock_irq(&device->client_data_lock); - list_del(&found_context->list); - write_unlock_irq(&device->client_data_lock); - up_write(&lists_rwsem); - kfree(found_context); - } - - mutex_unlock(&device_mutex); + xa_erase(&clients, client->client_id); + up_write(&clients_rwsem); } EXPORT_SYMBOL(ib_unregister_client); /** - * ib_get_client_data - Get IB client context - * @device:Device to get context for - * @client:Client to get context for - * - * ib_get_client_data() returns client context set with - * ib_set_client_data(). - */ -void *ib_get_client_data(struct ib_device *device, struct ib_client *client) -{ - struct ib_client_data *context; - void *ret = NULL; - unsigned long flags; - - read_lock_irqsave(&device->client_data_lock, flags); - list_for_each_entry(context, &device->client_data_list, list) - if (context->client == client) { - ret = context->data; - break; - } - read_unlock_irqrestore(&device->client_data_lock, flags); - - return ret; -} -EXPORT_SYMBOL(ib_get_client_data); - -/** * ib_set_client_data - Set IB client context * @device:Device to set context for * @client:Client to set context for * @data:Context to set * - * ib_set_client_data() sets client context that can be retrieved with - * ib_get_client_data(). + * ib_set_client_data() sets client context data that can be retrieved with + * ib_get_client_data(). This can only be called while the client is + * registered to the device, once the ib_client remove() callback returns this + * cannot be called. */ void ib_set_client_data(struct ib_device *device, struct ib_client *client, void *data) { - struct ib_client_data *context; - unsigned long flags; - - write_lock_irqsave(&device->client_data_lock, flags); - list_for_each_entry(context, &device->client_data_list, list) - if (context->client == client) { - context->data = data; - goto out; - } + void *rc; - dev_warn(&device->dev, "No client context found for %s\n", - client->name); + if (WARN_ON(IS_ERR(data))) + data = NULL; -out: - write_unlock_irqrestore(&device->client_data_lock, flags); + rc = xa_store(&device->client_data, client->client_id, data, + GFP_KERNEL); + WARN_ON(xa_is_err(rc)); } EXPORT_SYMBOL(ib_set_client_data); @@ -947,6 +1270,185 @@ int ib_query_port(struct ib_device *device, } EXPORT_SYMBOL(ib_query_port); +static void add_ndev_hash(struct ib_port_data *pdata) +{ + unsigned long flags; + + might_sleep(); + + spin_lock_irqsave(&ndev_hash_lock, flags); + if (hash_hashed(&pdata->ndev_hash_link)) { + hash_del_rcu(&pdata->ndev_hash_link); + spin_unlock_irqrestore(&ndev_hash_lock, flags); + /* + * We cannot do hash_add_rcu after a hash_del_rcu until the + * grace period + */ + synchronize_rcu(); + spin_lock_irqsave(&ndev_hash_lock, flags); + } + if (pdata->netdev) + hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, + (uintptr_t)pdata->netdev); + spin_unlock_irqrestore(&ndev_hash_lock, flags); +} + +/** + * ib_device_set_netdev - Associate the ib_dev with an underlying net_device + * @ib_dev: Device to modify + * @ndev: net_device to affiliate, may be NULL + * @port: IB port the net_device is connected to + * + * Drivers should use this to link the ib_device to a netdev so the netdev + * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be + * affiliated with any port. + * + * The caller must ensure that the given ndev is not unregistered or + * unregistering, and that either the ib_device is unregistered or + * ib_device_set_netdev() is called with NULL when the ndev sends a + * NETDEV_UNREGISTER event. + */ +int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, + unsigned int port) +{ + struct net_device *old_ndev; + struct ib_port_data *pdata; + unsigned long flags; + int ret; + + /* + * Drivers wish to call this before ib_register_driver, so we have to + * setup the port data early. + */ + ret = alloc_port_data(ib_dev); + if (ret) + return ret; + + if (!rdma_is_port_valid(ib_dev, port)) + return -EINVAL; + + pdata = &ib_dev->port_data[port]; + spin_lock_irqsave(&pdata->netdev_lock, flags); + old_ndev = rcu_dereference_protected( + pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); + if (old_ndev == ndev) { + spin_unlock_irqrestore(&pdata->netdev_lock, flags); + return 0; + } + + if (ndev) + dev_hold(ndev); + rcu_assign_pointer(pdata->netdev, ndev); + spin_unlock_irqrestore(&pdata->netdev_lock, flags); + + add_ndev_hash(pdata); + if (old_ndev) + dev_put(old_ndev); + + return 0; +} +EXPORT_SYMBOL(ib_device_set_netdev); + +static void free_netdevs(struct ib_device *ib_dev) +{ + unsigned long flags; + unsigned int port; + + rdma_for_each_port (ib_dev, port) { + struct ib_port_data *pdata = &ib_dev->port_data[port]; + struct net_device *ndev; + + spin_lock_irqsave(&pdata->netdev_lock, flags); + ndev = rcu_dereference_protected( + pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); + if (ndev) { + spin_lock(&ndev_hash_lock); + hash_del_rcu(&pdata->ndev_hash_link); + spin_unlock(&ndev_hash_lock); + + /* + * If this is the last dev_put there is still a + * synchronize_rcu before the netdev is kfreed, so we + * can continue to rely on unlocked pointer + * comparisons after the put + */ + rcu_assign_pointer(pdata->netdev, NULL); + dev_put(ndev); + } + spin_unlock_irqrestore(&pdata->netdev_lock, flags); + } +} + +struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, + unsigned int port) +{ + struct ib_port_data *pdata; + struct net_device *res; + + if (!rdma_is_port_valid(ib_dev, port)) + return NULL; + + pdata = &ib_dev->port_data[port]; + + /* + * New drivers should use ib_device_set_netdev() not the legacy + * get_netdev(). + */ + if (ib_dev->ops.get_netdev) + res = ib_dev->ops.get_netdev(ib_dev, port); + else { + spin_lock(&pdata->netdev_lock); + res = rcu_dereference_protected( + pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); + if (res) + dev_hold(res); + spin_unlock(&pdata->netdev_lock); + } + + /* + * If we are starting to unregister expedite things by preventing + * propagation of an unregistering netdev. + */ + if (res && res->reg_state != NETREG_REGISTERED) { + dev_put(res); + return NULL; + } + + return res; +} + +/** + * ib_device_get_by_netdev - Find an IB device associated with a netdev + * @ndev: netdev to locate + * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) + * + * Find and hold an ib_device that is associated with a netdev via + * ib_device_set_netdev(). The caller must call ib_device_put() on the + * returned pointer. + */ +struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, + enum rdma_driver_id driver_id) +{ + struct ib_device *res = NULL; + struct ib_port_data *cur; + + rcu_read_lock(); + hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, + (uintptr_t)ndev) { + if (rcu_access_pointer(cur->netdev) == ndev && + (driver_id == RDMA_DRIVER_UNKNOWN || + cur->ib_dev->driver_id == driver_id) && + ib_device_try_get(cur->ib_dev)) { + res = cur->ib_dev; + break; + } + } + rcu_read_unlock(); + + return res; +} +EXPORT_SYMBOL(ib_device_get_by_netdev); + /** * ib_enum_roce_netdev - enumerate all RoCE ports * @ib_dev : IB device we want to query @@ -965,21 +1467,12 @@ void ib_enum_roce_netdev(struct ib_device *ib_dev, roce_netdev_callback cb, void *cookie) { - u8 port; + unsigned int port; - for (port = rdma_start_port(ib_dev); port <= rdma_end_port(ib_dev); - port++) + rdma_for_each_port (ib_dev, port) if (rdma_protocol_roce(ib_dev, port)) { - struct net_device *idev = NULL; - - if (ib_dev->ops.get_netdev) - idev = ib_dev->ops.get_netdev(ib_dev, port); - - if (idev && - idev->reg_state >= NETREG_UNREGISTERED) { - dev_put(idev); - idev = NULL; - } + struct net_device *idev = + ib_device_get_netdev(ib_dev, port); if (filter(ib_dev, port, idev, filter_cookie)) cb(ib_dev, port, idev, cookie); @@ -1006,11 +1499,12 @@ void ib_enum_all_roce_netdevs(roce_netdev_filter filter, void *cookie) { struct ib_device *dev; + unsigned long index; - down_read(&lists_rwsem); - list_for_each_entry(dev, &device_list, core_list) + down_read(&devices_rwsem); + xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); - up_read(&lists_rwsem); + up_read(&devices_rwsem); } /** @@ -1022,19 +1516,19 @@ void ib_enum_all_roce_netdevs(roce_netdev_filter filter, int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, struct netlink_callback *cb) { + unsigned long index; struct ib_device *dev; unsigned int idx = 0; int ret = 0; - down_read(&lists_rwsem); - list_for_each_entry(dev, &device_list, core_list) { + down_read(&devices_rwsem); + xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { ret = nldev_cb(dev, skb, cb, idx); if (ret) break; idx++; } - - up_read(&lists_rwsem); + up_read(&devices_rwsem); return ret; } @@ -1121,13 +1615,15 @@ int ib_find_gid(struct ib_device *device, union ib_gid *gid, u8 *port_num, u16 *index) { union ib_gid tmp_gid; - int ret, port, i; + unsigned int port; + int ret, i; - for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) { + rdma_for_each_port (device, port) { if (!rdma_protocol_ib(device, port)) continue; - for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) { + for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; + ++i) { ret = rdma_query_gid(device, port, i, &tmp_gid); if (ret) return ret; @@ -1159,7 +1655,8 @@ int ib_find_pkey(struct ib_device *device, u16 tmp_pkey; int partial_ix = -1; - for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) { + for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; + ++i) { ret = ib_query_pkey(device, port_num, i, &tmp_pkey); if (ret) return ret; @@ -1192,6 +1689,7 @@ EXPORT_SYMBOL(ib_find_pkey); * @gid: A GID that the net_dev uses to communicate. * @addr: Contains the IP address that the request specified as its * destination. + * */ struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, @@ -1200,29 +1698,30 @@ struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, const struct sockaddr *addr) { struct net_device *net_dev = NULL; - struct ib_client_data *context; + unsigned long index; + void *client_data; if (!rdma_protocol_ib(dev, port)) return NULL; - down_read(&lists_rwsem); - - list_for_each_entry(context, &dev->client_data_list, list) { - struct ib_client *client = context->client; + /* + * Holding the read side guarantees that the client will not become + * unregistered while we are calling get_net_dev_by_params() + */ + down_read(&dev->client_data_rwsem); + xan_for_each_marked (&dev->client_data, index, client_data, + CLIENT_DATA_REGISTERED) { + struct ib_client *client = xa_load(&clients, index); - if (context->going_down) + if (!client || !client->get_net_dev_by_params) continue; - if (client->get_net_dev_by_params) { - net_dev = client->get_net_dev_by_params(dev, port, pkey, - gid, addr, - context->data); - if (net_dev) - break; - } + net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, + addr, client_data); + if (net_dev) + break; } - - up_read(&lists_rwsem); + up_read(&dev->client_data_rwsem); return net_dev; } @@ -1238,6 +1737,8 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) (ptr)->name = ops->name; \ } while (0) +#define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) + SET_DEVICE_OP(dev_ops, add_gid); SET_DEVICE_OP(dev_ops, advise_mr); SET_DEVICE_OP(dev_ops, alloc_dm); @@ -1261,6 +1762,7 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) SET_DEVICE_OP(dev_ops, create_srq); SET_DEVICE_OP(dev_ops, create_wq); SET_DEVICE_OP(dev_ops, dealloc_dm); + SET_DEVICE_OP(dev_ops, dealloc_driver); SET_DEVICE_OP(dev_ops, dealloc_fmr); SET_DEVICE_OP(dev_ops, dealloc_mw); SET_DEVICE_OP(dev_ops, dealloc_pd); @@ -1281,6 +1783,8 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) SET_DEVICE_OP(dev_ops, disassociate_ucontext); SET_DEVICE_OP(dev_ops, drain_rq); SET_DEVICE_OP(dev_ops, drain_sq); + SET_DEVICE_OP(dev_ops, enable_driver); + SET_DEVICE_OP(dev_ops, fill_res_entry); SET_DEVICE_OP(dev_ops, get_dev_fw_str); SET_DEVICE_OP(dev_ops, get_dma_mr); SET_DEVICE_OP(dev_ops, get_hw_stats); @@ -1290,6 +1794,7 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) SET_DEVICE_OP(dev_ops, get_vector_affinity); SET_DEVICE_OP(dev_ops, get_vf_config); SET_DEVICE_OP(dev_ops, get_vf_stats); + SET_DEVICE_OP(dev_ops, init_port); SET_DEVICE_OP(dev_ops, map_mr_sg); SET_DEVICE_OP(dev_ops, map_phys_fmr); SET_DEVICE_OP(dev_ops, mmap); @@ -1325,6 +1830,9 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) SET_DEVICE_OP(dev_ops, set_vf_guid); SET_DEVICE_OP(dev_ops, set_vf_link_state); SET_DEVICE_OP(dev_ops, unmap_fmr); + + SET_OBJ_SIZE(dev_ops, ib_pd); + SET_OBJ_SIZE(dev_ops, ib_ucontext); } EXPORT_SYMBOL(ib_set_device_ops); @@ -1443,6 +1951,9 @@ static void __exit ib_core_cleanup(void) destroy_workqueue(ib_comp_wq); /* Make sure that any pending umem accounting work is done. */ destroy_workqueue(ib_wq); + flush_workqueue(system_unbound_wq); + WARN_ON(!xa_empty(&clients)); + WARN_ON(!xa_empty(&devices)); } MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); diff --git a/drivers/infiniband/core/iwcm.c b/drivers/infiniband/core/iwcm.c index 476abc74178e..732637c913d9 100644 --- a/drivers/infiniband/core/iwcm.c +++ b/drivers/infiniband/core/iwcm.c @@ -87,7 +87,8 @@ static struct rdma_nl_cbs iwcm_nl_cb_table[RDMA_NL_IWPM_NUM_OPS] = { [RDMA_NL_IWPM_REMOTE_INFO] = {.dump = iwpm_remote_info_cb}, [RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb}, [RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb}, - [RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb} + [RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}, + [RDMA_NL_IWPM_HELLO] = {.dump = iwpm_hello_cb} }; static struct workqueue_struct *iwcm_wq; @@ -504,7 +505,7 @@ static int iw_cm_map(struct iw_cm_id *cm_id, bool active) { const char *devname = dev_name(&cm_id->device->dev); const char *ifname = cm_id->device->iwcm->ifname; - struct iwpm_dev_data pm_reg_msg; + struct iwpm_dev_data pm_reg_msg = {}; struct iwpm_sa_data pm_msg; int status; @@ -515,8 +516,8 @@ static int iw_cm_map(struct iw_cm_id *cm_id, bool active) cm_id->m_local_addr = cm_id->local_addr; cm_id->m_remote_addr = cm_id->remote_addr; - strncpy(pm_reg_msg.dev_name, devname, sizeof(pm_reg_msg.dev_name)); - strncpy(pm_reg_msg.if_name, ifname, sizeof(pm_reg_msg.if_name)); + strcpy(pm_reg_msg.dev_name, devname); + strcpy(pm_reg_msg.if_name, ifname); if (iwpm_register_pid(&pm_reg_msg, RDMA_NL_IWCM) || !iwpm_valid_pid()) @@ -525,6 +526,8 @@ static int iw_cm_map(struct iw_cm_id *cm_id, bool active) cm_id->mapped = true; pm_msg.loc_addr = cm_id->local_addr; pm_msg.rem_addr = cm_id->remote_addr; + pm_msg.flags = (cm_id->device->iwcm->driver_flags & IW_F_NO_PORT_MAP) ? + IWPM_FLAGS_NO_PORT_MAP : 0; if (active) status = iwpm_add_and_query_mapping(&pm_msg, RDMA_NL_IWCM); @@ -543,7 +546,7 @@ static int iw_cm_map(struct iw_cm_id *cm_id, bool active) return iwpm_create_mapinfo(&cm_id->local_addr, &cm_id->m_local_addr, - RDMA_NL_IWCM); + RDMA_NL_IWCM, pm_msg.flags); } /* diff --git a/drivers/infiniband/core/iwpm_msg.c b/drivers/infiniband/core/iwpm_msg.c index 8861c052155a..2452b0ddcf0d 100644 --- a/drivers/infiniband/core/iwpm_msg.c +++ b/drivers/infiniband/core/iwpm_msg.c @@ -34,18 +34,25 @@ #include "iwpm_util.h" static const char iwpm_ulib_name[IWPM_ULIBNAME_SIZE] = "iWarpPortMapperUser"; -static int iwpm_ulib_version = 3; +u16 iwpm_ulib_version = IWPM_UABI_VERSION_MIN; static int iwpm_user_pid = IWPM_PID_UNDEFINED; static atomic_t echo_nlmsg_seq; +/** + * iwpm_valid_pid - Check if the userspace iwarp port mapper pid is valid + * + * Returns true if the pid is greater than zero, otherwise returns false + */ int iwpm_valid_pid(void) { return iwpm_user_pid > 0; } -/* - * iwpm_register_pid - Send a netlink query to user space - * for the iwarp port mapper pid +/** + * iwpm_register_pid - Send a netlink query to userspace + * to get the iwarp port mapper pid + * @pm_msg: Contains driver info to send to the userspace port mapper + * @nl_client: The index of the netlink client * * nlmsg attributes: * [IWPM_NLA_REG_PID_SEQ] @@ -124,12 +131,19 @@ pid_query_error: return ret; } -/* - * iwpm_add_mapping - Send a netlink add mapping message - * to the port mapper +/** + * iwpm_add_mapping - Send a netlink add mapping request to + * the userspace port mapper + * @pm_msg: Contains the local ip/tcp address info to send + * @nl_client: The index of the netlink client + * * nlmsg attributes: * [IWPM_NLA_MANAGE_MAPPING_SEQ] * [IWPM_NLA_MANAGE_ADDR] + * [IWPM_NLA_MANAGE_FLAGS] + * + * If the request is successful, the pm_msg stores + * the port mapper response (mapped address info) */ int iwpm_add_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client) { @@ -173,6 +187,18 @@ int iwpm_add_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client) if (ret) goto add_mapping_error; + /* If flags are required and we're not V4, then return a quiet error */ + if (pm_msg->flags && iwpm_ulib_version == IWPM_UABI_VERSION_MIN) { + ret = -EINVAL; + goto add_mapping_error_nowarn; + } + if (iwpm_ulib_version > IWPM_UABI_VERSION_MIN) { + ret = ibnl_put_attr(skb, nlh, sizeof(u32), &pm_msg->flags, + IWPM_NLA_MANAGE_FLAGS); + if (ret) + goto add_mapping_error; + } + nlmsg_end(skb, nlh); nlmsg_request->req_buffer = pm_msg; @@ -187,6 +213,7 @@ int iwpm_add_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client) return ret; add_mapping_error: pr_info("%s: %s (client = %d)\n", __func__, err_str, nl_client); +add_mapping_error_nowarn: if (skb) dev_kfree_skb(skb); if (nlmsg_request) @@ -194,13 +221,17 @@ add_mapping_error: return ret; } -/* - * iwpm_add_and_query_mapping - Send a netlink add and query - * mapping message to the port mapper +/** + * iwpm_add_and_query_mapping - Process the port mapper response to + * iwpm_add_and_query_mapping request + * @pm_msg: Contains the local ip/tcp address info to send + * @nl_client: The index of the netlink client + * * nlmsg attributes: * [IWPM_NLA_QUERY_MAPPING_SEQ] * [IWPM_NLA_QUERY_LOCAL_ADDR] * [IWPM_NLA_QUERY_REMOTE_ADDR] + * [IWPM_NLA_QUERY_FLAGS] */ int iwpm_add_and_query_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client) { @@ -251,6 +282,18 @@ int iwpm_add_and_query_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client) if (ret) goto query_mapping_error; + /* If flags are required and we're not V4, then return a quite error */ + if (pm_msg->flags && iwpm_ulib_version == IWPM_UABI_VERSION_MIN) { + ret = -EINVAL; + goto query_mapping_error_nowarn; + } + if (iwpm_ulib_version > IWPM_UABI_VERSION_MIN) { + ret = ibnl_put_attr(skb, nlh, sizeof(u32), &pm_msg->flags, + IWPM_NLA_QUERY_FLAGS); + if (ret) + goto query_mapping_error; + } + nlmsg_end(skb, nlh); nlmsg_request->req_buffer = pm_msg; @@ -264,6 +307,7 @@ int iwpm_add_and_query_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client) return ret; query_mapping_error: pr_info("%s: %s (client = %d)\n", __func__, err_str, nl_client); +query_mapping_error_nowarn: if (skb) dev_kfree_skb(skb); if (nlmsg_request) @@ -271,9 +315,13 @@ query_mapping_error: return ret; } -/* - * iwpm_remove_mapping - Send a netlink remove mapping message - * to the port mapper +/** + * iwpm_remove_mapping - Send a netlink remove mapping request + * to the userspace port mapper + * + * @local_addr: Local ip/tcp address to remove + * @nl_client: The index of the netlink client + * * nlmsg attributes: * [IWPM_NLA_MANAGE_MAPPING_SEQ] * [IWPM_NLA_MANAGE_ADDR] @@ -344,9 +392,14 @@ static const struct nla_policy resp_reg_policy[IWPM_NLA_RREG_PID_MAX] = { [IWPM_NLA_RREG_PID_ERR] = { .type = NLA_U16 } }; -/* - * iwpm_register_pid_cb - Process a port mapper response to - * iwpm_register_pid() +/** + * iwpm_register_pid_cb - Process the port mapper response to + * iwpm_register_pid query + * @skb: + * @cb: Contains the received message (payload and netlink header) + * + * If successful, the function receives the userspace port mapper pid + * which is used in future communication with the port mapper */ int iwpm_register_pid_cb(struct sk_buff *skb, struct netlink_callback *cb) { @@ -379,7 +432,7 @@ int iwpm_register_pid_cb(struct sk_buff *skb, struct netlink_callback *cb) /* check device name, ulib name and version */ if (strcmp(pm_msg->dev_name, dev_name) || strcmp(iwpm_ulib_name, iwpm_name) || - iwpm_version != iwpm_ulib_version) { + iwpm_version < IWPM_UABI_VERSION_MIN) { pr_info("%s: Incorrect info (dev = %s name = %s version = %d)\n", __func__, dev_name, iwpm_name, iwpm_version); @@ -387,6 +440,10 @@ int iwpm_register_pid_cb(struct sk_buff *skb, struct netlink_callback *cb) goto register_pid_response_exit; } iwpm_user_pid = cb->nlh->nlmsg_pid; + iwpm_ulib_version = iwpm_version; + if (iwpm_ulib_version < IWPM_UABI_VERSION) + pr_warn_once("%s: Down level iwpmd/pid %u. Continuing...", + __func__, iwpm_user_pid); atomic_set(&echo_nlmsg_seq, cb->nlh->nlmsg_seq); pr_debug("%s: iWarp Port Mapper (pid = %d) is available!\n", __func__, iwpm_user_pid); @@ -403,15 +460,19 @@ register_pid_response_exit: /* netlink attribute policy for the received response to add mapping request */ static const struct nla_policy resp_add_policy[IWPM_NLA_RMANAGE_MAPPING_MAX] = { - [IWPM_NLA_MANAGE_MAPPING_SEQ] = { .type = NLA_U32 }, - [IWPM_NLA_MANAGE_ADDR] = { .len = sizeof(struct sockaddr_storage) }, - [IWPM_NLA_MANAGE_MAPPED_LOC_ADDR] = { .len = sizeof(struct sockaddr_storage) }, - [IWPM_NLA_RMANAGE_MAPPING_ERR] = { .type = NLA_U16 } + [IWPM_NLA_RMANAGE_MAPPING_SEQ] = { .type = NLA_U32 }, + [IWPM_NLA_RMANAGE_ADDR] = { + .len = sizeof(struct sockaddr_storage) }, + [IWPM_NLA_RMANAGE_MAPPED_LOC_ADDR] = { + .len = sizeof(struct sockaddr_storage) }, + [IWPM_NLA_RMANAGE_MAPPING_ERR] = { .type = NLA_U16 } }; -/* - * iwpm_add_mapping_cb - Process a port mapper response to - * iwpm_add_mapping() +/** + * iwpm_add_mapping_cb - Process the port mapper response to + * iwpm_add_mapping request + * @skb: + * @cb: Contains the received message (payload and netlink header) */ int iwpm_add_mapping_cb(struct sk_buff *skb, struct netlink_callback *cb) { @@ -430,7 +491,7 @@ int iwpm_add_mapping_cb(struct sk_buff *skb, struct netlink_callback *cb) atomic_set(&echo_nlmsg_seq, cb->nlh->nlmsg_seq); - msg_seq = nla_get_u32(nltb[IWPM_NLA_MANAGE_MAPPING_SEQ]); + msg_seq = nla_get_u32(nltb[IWPM_NLA_RMANAGE_MAPPING_SEQ]); nlmsg_request = iwpm_find_nlmsg_request(msg_seq); if (!nlmsg_request) { pr_info("%s: Could not find a matching request (seq = %u)\n", @@ -439,9 +500,9 @@ int iwpm_add_mapping_cb(struct sk_buff *skb, struct netlink_callback *cb) } pm_msg = nlmsg_request->req_buffer; local_sockaddr = (struct sockaddr_storage *) - nla_data(nltb[IWPM_NLA_MANAGE_ADDR]); + nla_data(nltb[IWPM_NLA_RMANAGE_ADDR]); mapped_sockaddr = (struct sockaddr_storage *) - nla_data(nltb[IWPM_NLA_MANAGE_MAPPED_LOC_ADDR]); + nla_data(nltb[IWPM_NLA_RMANAGE_MAPPED_LOC_ADDR]); if (iwpm_compare_sockaddr(local_sockaddr, &pm_msg->loc_addr)) { nlmsg_request->err_code = IWPM_USER_LIB_INFO_ERR; @@ -472,17 +533,23 @@ add_mapping_response_exit: /* netlink attribute policy for the response to add and query mapping request * and response with remote address info */ static const struct nla_policy resp_query_policy[IWPM_NLA_RQUERY_MAPPING_MAX] = { - [IWPM_NLA_QUERY_MAPPING_SEQ] = { .type = NLA_U32 }, - [IWPM_NLA_QUERY_LOCAL_ADDR] = { .len = sizeof(struct sockaddr_storage) }, - [IWPM_NLA_QUERY_REMOTE_ADDR] = { .len = sizeof(struct sockaddr_storage) }, - [IWPM_NLA_RQUERY_MAPPED_LOC_ADDR] = { .len = sizeof(struct sockaddr_storage) }, - [IWPM_NLA_RQUERY_MAPPED_REM_ADDR] = { .len = sizeof(struct sockaddr_storage) }, + [IWPM_NLA_RQUERY_MAPPING_SEQ] = { .type = NLA_U32 }, + [IWPM_NLA_RQUERY_LOCAL_ADDR] = { + .len = sizeof(struct sockaddr_storage) }, + [IWPM_NLA_RQUERY_REMOTE_ADDR] = { + .len = sizeof(struct sockaddr_storage) }, + [IWPM_NLA_RQUERY_MAPPED_LOC_ADDR] = { + .len = sizeof(struct sockaddr_storage) }, + [IWPM_NLA_RQUERY_MAPPED_REM_ADDR] = { + .len = sizeof(struct sockaddr_storage) }, [IWPM_NLA_RQUERY_MAPPING_ERR] = { .type = NLA_U16 } }; -/* - * iwpm_add_and_query_mapping_cb - Process a port mapper response to - * iwpm_add_and_query_mapping() +/** + * iwpm_add_and_query_mapping_cb - Process the port mapper response to + * iwpm_add_and_query_mapping request + * @skb: + * @cb: Contains the received message (payload and netlink header) */ int iwpm_add_and_query_mapping_cb(struct sk_buff *skb, struct netlink_callback *cb) @@ -502,7 +569,7 @@ int iwpm_add_and_query_mapping_cb(struct sk_buff *skb, return -EINVAL; atomic_set(&echo_nlmsg_seq, cb->nlh->nlmsg_seq); - msg_seq = nla_get_u32(nltb[IWPM_NLA_QUERY_MAPPING_SEQ]); + msg_seq = nla_get_u32(nltb[IWPM_NLA_RQUERY_MAPPING_SEQ]); nlmsg_request = iwpm_find_nlmsg_request(msg_seq); if (!nlmsg_request) { pr_info("%s: Could not find a matching request (seq = %u)\n", @@ -511,9 +578,9 @@ int iwpm_add_and_query_mapping_cb(struct sk_buff *skb, } pm_msg = nlmsg_request->req_buffer; local_sockaddr = (struct sockaddr_storage *) - nla_data(nltb[IWPM_NLA_QUERY_LOCAL_ADDR]); + nla_data(nltb[IWPM_NLA_RQUERY_LOCAL_ADDR]); remote_sockaddr = (struct sockaddr_storage *) - nla_data(nltb[IWPM_NLA_QUERY_REMOTE_ADDR]); + nla_data(nltb[IWPM_NLA_RQUERY_REMOTE_ADDR]); mapped_loc_sockaddr = (struct sockaddr_storage *) nla_data(nltb[IWPM_NLA_RQUERY_MAPPED_LOC_ADDR]); mapped_rem_sockaddr = (struct sockaddr_storage *) @@ -560,9 +627,13 @@ query_mapping_response_exit: return 0; } -/* - * iwpm_remote_info_cb - Process a port mapper message, containing - * the remote connecting peer address info +/** + * iwpm_remote_info_cb - Process remote connecting peer address info, which + * the port mapper has received from the connecting peer + * @skb: + * @cb: Contains the received message (payload and netlink header) + * + * Stores the IPv4/IPv6 address info in a hash table */ int iwpm_remote_info_cb(struct sk_buff *skb, struct netlink_callback *cb) { @@ -588,9 +659,9 @@ int iwpm_remote_info_cb(struct sk_buff *skb, struct netlink_callback *cb) atomic_set(&echo_nlmsg_seq, cb->nlh->nlmsg_seq); local_sockaddr = (struct sockaddr_storage *) - nla_data(nltb[IWPM_NLA_QUERY_LOCAL_ADDR]); + nla_data(nltb[IWPM_NLA_RQUERY_LOCAL_ADDR]); remote_sockaddr = (struct sockaddr_storage *) - nla_data(nltb[IWPM_NLA_QUERY_REMOTE_ADDR]); + nla_data(nltb[IWPM_NLA_RQUERY_REMOTE_ADDR]); mapped_loc_sockaddr = (struct sockaddr_storage *) nla_data(nltb[IWPM_NLA_RQUERY_MAPPED_LOC_ADDR]); mapped_rem_sockaddr = (struct sockaddr_storage *) @@ -635,8 +706,14 @@ static const struct nla_policy resp_mapinfo_policy[IWPM_NLA_MAPINFO_REQ_MAX] = { [IWPM_NLA_MAPINFO_ULIB_VER] = { .type = NLA_U16 } }; -/* - * iwpm_mapping_info_cb - Process a port mapper request for mapping info +/** + * iwpm_mapping_info_cb - Process a notification that the userspace + * port mapper daemon is started + * @skb: + * @cb: Contains the received message (payload and netlink header) + * + * Using the received port mapper pid, send all the local mapping + * info records to the userspace port mapper */ int iwpm_mapping_info_cb(struct sk_buff *skb, struct netlink_callback *cb) { @@ -655,7 +732,7 @@ int iwpm_mapping_info_cb(struct sk_buff *skb, struct netlink_callback *cb) iwpm_name = (char *)nla_data(nltb[IWPM_NLA_MAPINFO_ULIB_NAME]); iwpm_version = nla_get_u16(nltb[IWPM_NLA_MAPINFO_ULIB_VER]); if (strcmp(iwpm_ulib_name, iwpm_name) || - iwpm_version != iwpm_ulib_version) { + iwpm_version < IWPM_UABI_VERSION_MIN) { pr_info("%s: Invalid port mapper name = %s version = %d\n", __func__, iwpm_name, iwpm_version); return ret; @@ -669,6 +746,11 @@ int iwpm_mapping_info_cb(struct sk_buff *skb, struct netlink_callback *cb) iwpm_set_registration(nl_client, IWPM_REG_INCOMPL); atomic_set(&echo_nlmsg_seq, cb->nlh->nlmsg_seq); iwpm_user_pid = cb->nlh->nlmsg_pid; + + if (iwpm_ulib_version < IWPM_UABI_VERSION) + pr_warn_once("%s: Down level iwpmd/pid %u. Continuing...", + __func__, iwpm_user_pid); + if (!iwpm_mapinfo_available()) return 0; pr_debug("%s: iWarp Port Mapper (pid = %d) is available!\n", @@ -684,9 +766,11 @@ static const struct nla_policy ack_mapinfo_policy[IWPM_NLA_MAPINFO_NUM_MAX] = { [IWPM_NLA_MAPINFO_ACK_NUM] = { .type = NLA_U32 } }; -/* - * iwpm_ack_mapping_info_cb - Process a port mapper ack for - * the provided mapping info records +/** + * iwpm_ack_mapping_info_cb - Process the port mapper ack for + * the provided local mapping info records + * @skb: + * @cb: Contains the received message (payload and netlink header) */ int iwpm_ack_mapping_info_cb(struct sk_buff *skb, struct netlink_callback *cb) { @@ -712,8 +796,11 @@ static const struct nla_policy map_error_policy[IWPM_NLA_ERR_MAX] = { [IWPM_NLA_ERR_CODE] = { .type = NLA_U16 }, }; -/* - * iwpm_mapping_error_cb - Process a port mapper error message +/** + * iwpm_mapping_error_cb - Process port mapper notification for error + * + * @skb: + * @cb: Contains the received message (payload and netlink header) */ int iwpm_mapping_error_cb(struct sk_buff *skb, struct netlink_callback *cb) { @@ -748,3 +835,46 @@ int iwpm_mapping_error_cb(struct sk_buff *skb, struct netlink_callback *cb) up(&nlmsg_request->sem); return 0; } + +/* netlink attribute policy for the received hello request */ +static const struct nla_policy hello_policy[IWPM_NLA_HELLO_MAX] = { + [IWPM_NLA_HELLO_ABI_VERSION] = { .type = NLA_U16 } +}; + +/** + * iwpm_hello_cb - Process a hello message from iwpmd + * + * @skb: + * @cb: Contains the received message (payload and netlink header) + * + * Using the received port mapper pid, send the kernel's abi_version + * after adjusting it to support the iwpmd version. + */ +int iwpm_hello_cb(struct sk_buff *skb, struct netlink_callback *cb) +{ + struct nlattr *nltb[IWPM_NLA_HELLO_MAX]; + const char *msg_type = "Hello request"; + u8 nl_client; + u16 abi_version; + int ret = -EINVAL; + + if (iwpm_parse_nlmsg(cb, IWPM_NLA_HELLO_MAX, hello_policy, nltb, + msg_type)) { + pr_info("%s: Unable to parse nlmsg\n", __func__); + return ret; + } + abi_version = nla_get_u16(nltb[IWPM_NLA_HELLO_ABI_VERSION]); + nl_client = RDMA_NL_GET_CLIENT(cb->nlh->nlmsg_type); + if (!iwpm_valid_client(nl_client)) { + pr_info("%s: Invalid port mapper client = %d\n", + __func__, nl_client); + return ret; + } + iwpm_set_registration(nl_client, IWPM_REG_INCOMPL); + atomic_set(&echo_nlmsg_seq, cb->nlh->nlmsg_seq); + iwpm_ulib_version = min_t(u16, IWPM_UABI_VERSION, abi_version); + pr_debug("Using ABI version %u\n", iwpm_ulib_version); + iwpm_user_pid = cb->nlh->nlmsg_pid; + ret = iwpm_send_hello(nl_client, iwpm_user_pid, iwpm_ulib_version); + return ret; +} diff --git a/drivers/infiniband/core/iwpm_util.c b/drivers/infiniband/core/iwpm_util.c index cdb63f3f4de7..a5d2a20ee697 100644 --- a/drivers/infiniband/core/iwpm_util.c +++ b/drivers/infiniband/core/iwpm_util.c @@ -51,6 +51,12 @@ static DEFINE_SPINLOCK(iwpm_reminfo_lock); static DEFINE_MUTEX(iwpm_admin_lock); static struct iwpm_admin_data iwpm_admin; +/** + * iwpm_init - Allocate resources for the iwarp port mapper + * @nl_client: The index of the netlink client + * + * Should be called when network interface goes up. + */ int iwpm_init(u8 nl_client) { int ret = 0; @@ -87,6 +93,12 @@ init_exit: static void free_hash_bucket(void); static void free_reminfo_bucket(void); +/** + * iwpm_exit - Deallocate resources for the iwarp port mapper + * @nl_client: The index of the netlink client + * + * Should be called when network interface goes down. + */ int iwpm_exit(u8 nl_client) { @@ -112,9 +124,17 @@ int iwpm_exit(u8 nl_client) static struct hlist_head *get_mapinfo_hash_bucket(struct sockaddr_storage *, struct sockaddr_storage *); +/** + * iwpm_create_mapinfo - Store local and mapped IPv4/IPv6 address + * info in a hash table + * @local_addr: Local ip/tcp address + * @mapped_addr: Mapped local ip/tcp address + * @nl_client: The index of the netlink client + * @map_flags: IWPM mapping flags + */ int iwpm_create_mapinfo(struct sockaddr_storage *local_sockaddr, struct sockaddr_storage *mapped_sockaddr, - u8 nl_client) + u8 nl_client, u32 map_flags) { struct hlist_head *hash_bucket_head = NULL; struct iwpm_mapping_info *map_info; @@ -132,6 +152,7 @@ int iwpm_create_mapinfo(struct sockaddr_storage *local_sockaddr, memcpy(&map_info->mapped_sockaddr, mapped_sockaddr, sizeof(struct sockaddr_storage)); map_info->nl_client = nl_client; + map_info->map_flags = map_flags; spin_lock_irqsave(&iwpm_mapinfo_lock, flags); if (iwpm_hash_bucket) { @@ -150,6 +171,15 @@ int iwpm_create_mapinfo(struct sockaddr_storage *local_sockaddr, return ret; } +/** + * iwpm_remove_mapinfo - Remove local and mapped IPv4/IPv6 address + * info from the hash table + * @local_addr: Local ip/tcp address + * @mapped_local_addr: Mapped local ip/tcp address + * + * Returns err code if mapping info is not found in the hash table, + * otherwise returns 0 + */ int iwpm_remove_mapinfo(struct sockaddr_storage *local_sockaddr, struct sockaddr_storage *mapped_local_addr) { @@ -250,6 +280,17 @@ void iwpm_add_remote_info(struct iwpm_remote_info *rem_info) spin_unlock_irqrestore(&iwpm_reminfo_lock, flags); } +/** + * iwpm_get_remote_info - Get the remote connecting peer address info + * + * @mapped_loc_addr: Mapped local address of the listening peer + * @mapped_rem_addr: Mapped remote address of the connecting peer + * @remote_addr: To store the remote address of the connecting peer + * @nl_client: The index of the netlink client + * + * The remote address info is retrieved and provided to the client in + * the remote_addr. After that it is removed from the hash table + */ int iwpm_get_remote_info(struct sockaddr_storage *mapped_loc_addr, struct sockaddr_storage *mapped_rem_addr, struct sockaddr_storage *remote_addr, @@ -686,6 +727,14 @@ int iwpm_send_mapinfo(u8 nl_client, int iwpm_pid) if (ret) goto send_mapping_info_unlock; + if (iwpm_ulib_version > IWPM_UABI_VERSION_MIN) { + ret = ibnl_put_attr(skb, nlh, sizeof(u32), + &map_info->map_flags, + IWPM_NLA_MAPINFO_FLAGS); + if (ret) + goto send_mapping_info_unlock; + } + nlmsg_end(skb, nlh); iwpm_print_sockaddr(&map_info->local_sockaddr, @@ -754,3 +803,38 @@ int iwpm_mapinfo_available(void) spin_unlock_irqrestore(&iwpm_mapinfo_lock, flags); return full_bucket; } + +int iwpm_send_hello(u8 nl_client, int iwpm_pid, u16 abi_version) +{ + struct sk_buff *skb = NULL; + struct nlmsghdr *nlh; + const char *err_str = ""; + int ret = -EINVAL; + + skb = iwpm_create_nlmsg(RDMA_NL_IWPM_HELLO, &nlh, nl_client); + if (!skb) { + err_str = "Unable to create a nlmsg"; + goto hello_num_error; + } + nlh->nlmsg_seq = iwpm_get_nlmsg_seq(); + err_str = "Unable to put attribute of abi_version into nlmsg"; + ret = ibnl_put_attr(skb, nlh, sizeof(u16), &abi_version, + IWPM_NLA_HELLO_ABI_VERSION); + if (ret) + goto hello_num_error; + nlmsg_end(skb, nlh); + + ret = rdma_nl_unicast(skb, iwpm_pid); + if (ret) { + skb = NULL; + err_str = "Unable to send a nlmsg"; + goto hello_num_error; + } + pr_debug("%s: Sent hello abi_version = %u\n", __func__, abi_version); + return 0; +hello_num_error: + pr_info("%s: %s\n", __func__, err_str); + if (skb) + dev_kfree_skb(skb); + return ret; +} diff --git a/drivers/infiniband/core/iwpm_util.h b/drivers/infiniband/core/iwpm_util.h index af1fc14a0d3d..7e2bcc72f66c 100644 --- a/drivers/infiniband/core/iwpm_util.h +++ b/drivers/infiniband/core/iwpm_util.h @@ -78,6 +78,7 @@ struct iwpm_mapping_info { struct sockaddr_storage local_sockaddr; struct sockaddr_storage mapped_sockaddr; u8 nl_client; + u32 map_flags; }; struct iwpm_remote_info { @@ -266,4 +267,15 @@ int iwpm_parse_nlmsg(struct netlink_callback *cb, int policy_max, * @msg: Message to print */ void iwpm_print_sockaddr(struct sockaddr_storage *sockaddr, char *msg); + +/** + * iwpm_send_hello - Send hello response to iwpmd + * + * @nl_client: The index of the netlink client + * @abi_version: The kernel's abi_version + * + * Returns 0 on success or a negative error code + */ +int iwpm_send_hello(u8 nl_client, int iwpm_pid, u16 abi_version); +extern u16 iwpm_ulib_version; #endif diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c index 7870823bac47..e742a6a2c138 100644 --- a/drivers/infiniband/core/mad.c +++ b/drivers/infiniband/core/mad.c @@ -3326,9 +3326,9 @@ error: static void ib_mad_remove_device(struct ib_device *device, void *client_data) { - int i; + unsigned int i; - for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) { + rdma_for_each_port (device, i) { if (!rdma_cap_ib_mad(device, i)) continue; diff --git a/drivers/infiniband/core/netlink.c b/drivers/infiniband/core/netlink.c index 724f5a62e82f..eecfc0b377c9 100644 --- a/drivers/infiniband/core/netlink.c +++ b/drivers/infiniband/core/netlink.c @@ -56,7 +56,6 @@ EXPORT_SYMBOL(rdma_nl_chk_listeners); static bool is_nl_msg_valid(unsigned int type, unsigned int op) { static const unsigned int max_num_ops[RDMA_NL_NUM_CLIENTS] = { - [RDMA_NL_RDMA_CM] = RDMA_NL_RDMA_CM_NUM_OPS, [RDMA_NL_IWCM] = RDMA_NL_IWPM_NUM_OPS, [RDMA_NL_LS] = RDMA_NL_LS_NUM_OPS, [RDMA_NL_NLDEV] = RDMA_NLDEV_NUM_OPS, @@ -181,8 +180,7 @@ static int rdma_nl_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, return -EINVAL; } /* FIXME: Convert IWCM to properly handle doit callbacks */ - if ((nlh->nlmsg_flags & NLM_F_DUMP) || index == RDMA_NL_RDMA_CM || - index == RDMA_NL_IWCM) { + if ((nlh->nlmsg_flags & NLM_F_DUMP) || index == RDMA_NL_IWCM) { struct netlink_dump_control c = { .dump = cb_table[op].dump, }; diff --git a/drivers/infiniband/core/nldev.c b/drivers/infiniband/core/nldev.c index 3c97a8b6bf1e..11ed58d3fce5 100644 --- a/drivers/infiniband/core/nldev.c +++ b/drivers/infiniband/core/nldev.c @@ -33,12 +33,14 @@ #include <linux/module.h> #include <linux/pid.h> #include <linux/pid_namespace.h> +#include <linux/mutex.h> #include <net/netlink.h> #include <rdma/rdma_cm.h> #include <rdma/rdma_netlink.h> #include "core_priv.h" #include "cma_priv.h" +#include "restrack.h" static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = { [RDMA_NLDEV_ATTR_DEV_INDEX] = { .type = NLA_U32 }, @@ -107,6 +109,13 @@ static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = { [RDMA_NLDEV_ATTR_DRIVER_U32] = { .type = NLA_U32 }, [RDMA_NLDEV_ATTR_DRIVER_S64] = { .type = NLA_S64 }, [RDMA_NLDEV_ATTR_DRIVER_U64] = { .type = NLA_U64 }, + [RDMA_NLDEV_ATTR_RES_PDN] = { .type = NLA_U32 }, + [RDMA_NLDEV_ATTR_RES_CQN] = { .type = NLA_U32 }, + [RDMA_NLDEV_ATTR_RES_MRN] = { .type = NLA_U32 }, + [RDMA_NLDEV_ATTR_RES_CM_IDN] = { .type = NLA_U32 }, + [RDMA_NLDEV_ATTR_RES_CTXN] = { .type = NLA_U32 }, + [RDMA_NLDEV_ATTR_LINK_TYPE] = { .type = NLA_NUL_STRING, + .len = RDMA_NLDEV_ATTR_ENTRY_STRLEN }, }; static int put_driver_name_print_type(struct sk_buff *msg, const char *name, @@ -262,9 +271,7 @@ static int fill_port_info(struct sk_buff *msg, if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_PHYS_STATE, attr.phys_state)) return -EMSGSIZE; - if (device->ops.get_netdev) - netdev = device->ops.get_netdev(device, port); - + netdev = ib_device_get_netdev(device, port); if (netdev && net_eq(dev_net(netdev), net)) { ret = nla_put_u32(msg, RDMA_NLDEV_ATTR_NDEV_INDEX, netdev->ifindex); @@ -314,7 +321,6 @@ static int fill_res_info(struct sk_buff *msg, struct ib_device *device) [RDMA_RESTRACK_CTX] = "ctx", }; - struct rdma_restrack_root *res = &device->res; struct nlattr *table_attr; int ret, i, curr; @@ -328,7 +334,8 @@ static int fill_res_info(struct sk_buff *msg, struct ib_device *device) for (i = 0; i < RDMA_RESTRACK_MAX; i++) { if (!names[i]) continue; - curr = rdma_restrack_count(res, i, task_active_pid_ns(current)); + curr = rdma_restrack_count(device, i, + task_active_pid_ns(current)); ret = fill_res_info_entry(msg, names[i], curr); if (ret) goto err; @@ -361,13 +368,20 @@ static int fill_res_name_pid(struct sk_buff *msg, return 0; } -static int fill_res_qp_entry(struct sk_buff *msg, struct netlink_callback *cb, +static bool fill_res_entry(struct ib_device *dev, struct sk_buff *msg, + struct rdma_restrack_entry *res) +{ + if (!dev->ops.fill_res_entry) + return false; + return dev->ops.fill_res_entry(msg, res); +} + +static int fill_res_qp_entry(struct sk_buff *msg, bool has_cap_net_admin, struct rdma_restrack_entry *res, uint32_t port) { struct ib_qp *qp = container_of(res, struct ib_qp, res); - struct rdma_restrack_root *resroot = &qp->device->res; + struct ib_device *dev = qp->device; struct ib_qp_init_attr qp_init_attr; - struct nlattr *entry_attr; struct ib_qp_attr qp_attr; int ret; @@ -376,11 +390,7 @@ static int fill_res_qp_entry(struct sk_buff *msg, struct netlink_callback *cb, return ret; if (port && port != qp_attr.port_num) - return 0; - - entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP_ENTRY); - if (!entry_attr) - goto out; + return -EAGAIN; /* In create_qp() port is not set yet */ if (qp_attr.port_num && @@ -412,38 +422,32 @@ static int fill_res_qp_entry(struct sk_buff *msg, struct netlink_callback *cb, if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_STATE, qp_attr.qp_state)) goto err; + if (!rdma_is_kernel_res(res) && + nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, qp->pd->res.id)) + goto err; + if (fill_res_name_pid(msg, res)) goto err; - if (resroot->fill_res_entry(msg, res)) + if (fill_res_entry(dev, msg, res)) goto err; - nla_nest_end(msg, entry_attr); return 0; -err: - nla_nest_cancel(msg, entry_attr); -out: - return -EMSGSIZE; +err: return -EMSGSIZE; } -static int fill_res_cm_id_entry(struct sk_buff *msg, - struct netlink_callback *cb, +static int fill_res_cm_id_entry(struct sk_buff *msg, bool has_cap_net_admin, struct rdma_restrack_entry *res, uint32_t port) { struct rdma_id_private *id_priv = container_of(res, struct rdma_id_private, res); - struct rdma_restrack_root *resroot = &id_priv->id.device->res; + struct ib_device *dev = id_priv->id.device; struct rdma_cm_id *cm_id = &id_priv->id; - struct nlattr *entry_attr; if (port && port != cm_id->port_num) return 0; - entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY); - if (!entry_attr) - goto out; - if (cm_id->port_num && nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, cm_id->port_num)) goto err; @@ -472,31 +476,25 @@ static int fill_res_cm_id_entry(struct sk_buff *msg, &cm_id->route.addr.dst_addr)) goto err; + if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CM_IDN, res->id)) + goto err; + if (fill_res_name_pid(msg, res)) goto err; - if (resroot->fill_res_entry(msg, res)) + if (fill_res_entry(dev, msg, res)) goto err; - nla_nest_end(msg, entry_attr); return 0; -err: - nla_nest_cancel(msg, entry_attr); -out: - return -EMSGSIZE; +err: return -EMSGSIZE; } -static int fill_res_cq_entry(struct sk_buff *msg, struct netlink_callback *cb, +static int fill_res_cq_entry(struct sk_buff *msg, bool has_cap_net_admin, struct rdma_restrack_entry *res, uint32_t port) { struct ib_cq *cq = container_of(res, struct ib_cq, res); - struct rdma_restrack_root *resroot = &cq->device->res; - struct nlattr *entry_attr; - - entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_CQ_ENTRY); - if (!entry_attr) - goto out; + struct ib_device *dev = cq->device; if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQE, cq->cqe)) goto err; @@ -509,33 +507,31 @@ static int fill_res_cq_entry(struct sk_buff *msg, struct netlink_callback *cb, nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_POLL_CTX, cq->poll_ctx)) goto err; + if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQN, res->id)) + goto err; + if (!rdma_is_kernel_res(res) && + nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN, + cq->uobject->context->res.id)) + goto err; + if (fill_res_name_pid(msg, res)) goto err; - if (resroot->fill_res_entry(msg, res)) + if (fill_res_entry(dev, msg, res)) goto err; - nla_nest_end(msg, entry_attr); return 0; -err: - nla_nest_cancel(msg, entry_attr); -out: - return -EMSGSIZE; +err: return -EMSGSIZE; } -static int fill_res_mr_entry(struct sk_buff *msg, struct netlink_callback *cb, +static int fill_res_mr_entry(struct sk_buff *msg, bool has_cap_net_admin, struct rdma_restrack_entry *res, uint32_t port) { struct ib_mr *mr = container_of(res, struct ib_mr, res); - struct rdma_restrack_root *resroot = &mr->pd->device->res; - struct nlattr *entry_attr; - - entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_MR_ENTRY); - if (!entry_attr) - goto out; + struct ib_device *dev = mr->pd->device; - if (netlink_capable(cb->skb, CAP_NET_ADMIN)) { + if (has_cap_net_admin) { if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RKEY, mr->rkey)) goto err; if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LKEY, mr->lkey)) @@ -546,33 +542,31 @@ static int fill_res_mr_entry(struct sk_buff *msg, struct netlink_callback *cb, RDMA_NLDEV_ATTR_PAD)) goto err; + if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_MRN, res->id)) + goto err; + + if (!rdma_is_kernel_res(res) && + nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, mr->pd->res.id)) + goto err; + if (fill_res_name_pid(msg, res)) goto err; - if (resroot->fill_res_entry(msg, res)) + if (fill_res_entry(dev, msg, res)) goto err; - nla_nest_end(msg, entry_attr); return 0; -err: - nla_nest_cancel(msg, entry_attr); -out: - return -EMSGSIZE; +err: return -EMSGSIZE; } -static int fill_res_pd_entry(struct sk_buff *msg, struct netlink_callback *cb, +static int fill_res_pd_entry(struct sk_buff *msg, bool has_cap_net_admin, struct rdma_restrack_entry *res, uint32_t port) { struct ib_pd *pd = container_of(res, struct ib_pd, res); - struct rdma_restrack_root *resroot = &pd->device->res; - struct nlattr *entry_attr; + struct ib_device *dev = pd->device; - entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_PD_ENTRY); - if (!entry_attr) - goto out; - - if (netlink_capable(cb->skb, CAP_NET_ADMIN)) { + if (has_cap_net_admin) { if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LOCAL_DMA_LKEY, pd->local_dma_lkey)) goto err; @@ -585,19 +579,23 @@ static int fill_res_pd_entry(struct sk_buff *msg, struct netlink_callback *cb, atomic_read(&pd->usecnt), RDMA_NLDEV_ATTR_PAD)) goto err; + if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, res->id)) + goto err; + + if (!rdma_is_kernel_res(res) && + nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN, + pd->uobject->context->res.id)) + goto err; + if (fill_res_name_pid(msg, res)) goto err; - if (resroot->fill_res_entry(msg, res)) + if (fill_res_entry(dev, msg, res)) goto err; - nla_nest_end(msg, entry_attr); return 0; -err: - nla_nest_cancel(msg, entry_attr); -out: - return -EMSGSIZE; +err: return -EMSGSIZE; } static int nldev_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh, @@ -777,7 +775,7 @@ static int nldev_port_get_dumpit(struct sk_buff *skb, u32 idx = 0; u32 ifindex; int err; - u32 p; + unsigned int p; err = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, nldev_policy, NULL); @@ -789,7 +787,7 @@ static int nldev_port_get_dumpit(struct sk_buff *skb, if (!device) return -EINVAL; - for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) { + rdma_for_each_port (device, p) { /* * The dumpit function returns all information from specific * index. This specific index is taken from the netlink @@ -905,10 +903,17 @@ static int nldev_res_get_dumpit(struct sk_buff *skb, } struct nldev_fill_res_entry { - int (*fill_res_func)(struct sk_buff *msg, struct netlink_callback *cb, + int (*fill_res_func)(struct sk_buff *msg, bool has_cap_net_admin, struct rdma_restrack_entry *res, u32 port); enum rdma_nldev_attr nldev_attr; enum rdma_nldev_command nldev_cmd; + u8 flags; + u32 entry; + u32 id; +}; + +enum nldev_res_flags { + NLDEV_PER_DEV = 1 << 0, }; static const struct nldev_fill_res_entry fill_entries[RDMA_RESTRACK_MAX] = { @@ -916,29 +921,136 @@ static const struct nldev_fill_res_entry fill_entries[RDMA_RESTRACK_MAX] = { .fill_res_func = fill_res_qp_entry, .nldev_cmd = RDMA_NLDEV_CMD_RES_QP_GET, .nldev_attr = RDMA_NLDEV_ATTR_RES_QP, + .entry = RDMA_NLDEV_ATTR_RES_QP_ENTRY, + .id = RDMA_NLDEV_ATTR_RES_LQPN, }, [RDMA_RESTRACK_CM_ID] = { .fill_res_func = fill_res_cm_id_entry, .nldev_cmd = RDMA_NLDEV_CMD_RES_CM_ID_GET, .nldev_attr = RDMA_NLDEV_ATTR_RES_CM_ID, + .entry = RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY, + .id = RDMA_NLDEV_ATTR_RES_CM_IDN, }, [RDMA_RESTRACK_CQ] = { .fill_res_func = fill_res_cq_entry, .nldev_cmd = RDMA_NLDEV_CMD_RES_CQ_GET, .nldev_attr = RDMA_NLDEV_ATTR_RES_CQ, + .flags = NLDEV_PER_DEV, + .entry = RDMA_NLDEV_ATTR_RES_CQ_ENTRY, + .id = RDMA_NLDEV_ATTR_RES_CQN, }, [RDMA_RESTRACK_MR] = { .fill_res_func = fill_res_mr_entry, .nldev_cmd = RDMA_NLDEV_CMD_RES_MR_GET, .nldev_attr = RDMA_NLDEV_ATTR_RES_MR, + .flags = NLDEV_PER_DEV, + .entry = RDMA_NLDEV_ATTR_RES_MR_ENTRY, + .id = RDMA_NLDEV_ATTR_RES_MRN, }, [RDMA_RESTRACK_PD] = { .fill_res_func = fill_res_pd_entry, .nldev_cmd = RDMA_NLDEV_CMD_RES_PD_GET, .nldev_attr = RDMA_NLDEV_ATTR_RES_PD, + .flags = NLDEV_PER_DEV, + .entry = RDMA_NLDEV_ATTR_RES_PD_ENTRY, + .id = RDMA_NLDEV_ATTR_RES_PDN, }, }; +static bool is_visible_in_pid_ns(struct rdma_restrack_entry *res) +{ + /* + * 1. Kern resources should be visible in init name space only + * 2. Present only resources visible in the current namespace + */ + if (rdma_is_kernel_res(res)) + return task_active_pid_ns(current) == &init_pid_ns; + return task_active_pid_ns(current) == task_active_pid_ns(res->task); +} + +static int res_get_common_doit(struct sk_buff *skb, struct nlmsghdr *nlh, + struct netlink_ext_ack *extack, + enum rdma_restrack_type res_type) +{ + const struct nldev_fill_res_entry *fe = &fill_entries[res_type]; + struct nlattr *tb[RDMA_NLDEV_ATTR_MAX]; + struct rdma_restrack_entry *res; + struct ib_device *device; + u32 index, id, port = 0; + bool has_cap_net_admin; + struct sk_buff *msg; + int ret; + + ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, + nldev_policy, extack); + if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !fe->id || !tb[fe->id]) + return -EINVAL; + + index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]); + device = ib_device_get_by_index(index); + if (!device) + return -EINVAL; + + if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) { + port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]); + if (!rdma_is_port_valid(device, port)) { + ret = -EINVAL; + goto err; + } + } + + if ((port && fe->flags & NLDEV_PER_DEV) || + (!port && ~fe->flags & NLDEV_PER_DEV)) { + ret = -EINVAL; + goto err; + } + + id = nla_get_u32(tb[fe->id]); + res = rdma_restrack_get_byid(device, res_type, id); + if (IS_ERR(res)) { + ret = PTR_ERR(res); + goto err; + } + + if (!is_visible_in_pid_ns(res)) { + ret = -ENOENT; + goto err_get; + } + + msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); + if (!msg) { + ret = -ENOMEM; + goto err; + } + + nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, + RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, fe->nldev_cmd), + 0, 0); + + if (fill_nldev_handle(msg, device)) { + ret = -EMSGSIZE; + goto err_free; + } + + has_cap_net_admin = netlink_capable(skb, CAP_NET_ADMIN); + ret = fe->fill_res_func(msg, has_cap_net_admin, res, port); + rdma_restrack_put(res); + if (ret) + goto err_free; + + nlmsg_end(msg, nlh); + ib_device_put(device); + return rdma_nl_unicast(msg, NETLINK_CB(skb).portid); + +err_free: + nlmsg_free(msg); +err_get: + rdma_restrack_put(res); +err: + ib_device_put(device); + return ret; +} + static int res_get_common_dumpit(struct sk_buff *skb, struct netlink_callback *cb, enum rdma_restrack_type res_type) @@ -946,11 +1058,15 @@ static int res_get_common_dumpit(struct sk_buff *skb, const struct nldev_fill_res_entry *fe = &fill_entries[res_type]; struct nlattr *tb[RDMA_NLDEV_ATTR_MAX]; struct rdma_restrack_entry *res; + struct rdma_restrack_root *rt; int err, ret = 0, idx = 0; struct nlattr *table_attr; + struct nlattr *entry_attr; struct ib_device *device; int start = cb->args[0]; + bool has_cap_net_admin; struct nlmsghdr *nlh; + unsigned long id; u32 index, port = 0; bool filled = false; @@ -998,55 +1114,51 @@ static int res_get_common_dumpit(struct sk_buff *skb, goto err; } - down_read(&device->res.rwsem); - hash_for_each_possible(device->res.hash, res, node, res_type) { - if (idx < start) - goto next; + has_cap_net_admin = netlink_capable(cb->skb, CAP_NET_ADMIN); - if ((rdma_is_kernel_res(res) && - task_active_pid_ns(current) != &init_pid_ns) || - (!rdma_is_kernel_res(res) && task_active_pid_ns(current) != - task_active_pid_ns(res->task))) - /* - * 1. Kern resources should be visible in init - * namspace only - * 2. Present only resources visible in the current - * namespace - */ - goto next; + rt = &device->res[res_type]; + xa_lock(&rt->xa); + /* + * FIXME: if the skip ahead is something common this loop should + * use xas_for_each & xas_pause to optimize, we can have a lot of + * objects. + */ + xa_for_each(&rt->xa, id, res) { + if (!is_visible_in_pid_ns(res)) + continue; - if (!rdma_restrack_get(res)) - /* - * Resource is under release now, but we are not - * relesing lock now, so it will be released in - * our next pass, once we will get ->next pointer. - */ + if (idx < start || !rdma_restrack_get(res)) goto next; + xa_unlock(&rt->xa); + filled = true; - up_read(&device->res.rwsem); - ret = fe->fill_res_func(skb, cb, res, port); - down_read(&device->res.rwsem); - /* - * Return resource back, but it won't be released till - * the &device->res.rwsem will be released for write. - */ + entry_attr = nla_nest_start(skb, fe->entry); + if (!entry_attr) { + ret = -EMSGSIZE; + rdma_restrack_put(res); + goto msg_full; + } + + ret = fe->fill_res_func(skb, has_cap_net_admin, res, port); rdma_restrack_put(res); - if (ret == -EMSGSIZE) - /* - * There is a chance to optimize here. - * It can be done by using list_prepare_entry - * and list_for_each_entry_continue afterwards. - */ - break; - if (ret) + if (ret) { + nla_nest_cancel(skb, entry_attr); + if (ret == -EMSGSIZE) + goto msg_full; + if (ret == -EAGAIN) + goto again; goto res_err; + } + nla_nest_end(skb, entry_attr); +again: xa_lock(&rt->xa); next: idx++; } - up_read(&device->res.rwsem); + xa_unlock(&rt->xa); +msg_full: nla_nest_end(skb, table_attr); nlmsg_end(skb, nlh); cb->args[0] = idx; @@ -1063,7 +1175,6 @@ next: idx++; res_err: nla_nest_cancel(skb, table_attr); - up_read(&device->res.rwsem); err: nlmsg_cancel(skb, nlh); @@ -1073,34 +1184,132 @@ err_index: return ret; } -static int nldev_res_get_qp_dumpit(struct sk_buff *skb, - struct netlink_callback *cb) +#define RES_GET_FUNCS(name, type) \ + static int nldev_res_get_##name##_dumpit(struct sk_buff *skb, \ + struct netlink_callback *cb) \ + { \ + return res_get_common_dumpit(skb, cb, type); \ + } \ + static int nldev_res_get_##name##_doit(struct sk_buff *skb, \ + struct nlmsghdr *nlh, \ + struct netlink_ext_ack *extack) \ + { \ + return res_get_common_doit(skb, nlh, extack, type); \ + } + +RES_GET_FUNCS(qp, RDMA_RESTRACK_QP); +RES_GET_FUNCS(cm_id, RDMA_RESTRACK_CM_ID); +RES_GET_FUNCS(cq, RDMA_RESTRACK_CQ); +RES_GET_FUNCS(pd, RDMA_RESTRACK_PD); +RES_GET_FUNCS(mr, RDMA_RESTRACK_MR); + +static LIST_HEAD(link_ops); +static DECLARE_RWSEM(link_ops_rwsem); + +static const struct rdma_link_ops *link_ops_get(const char *type) { - return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_QP); + const struct rdma_link_ops *ops; + + list_for_each_entry(ops, &link_ops, list) { + if (!strcmp(ops->type, type)) + goto out; + } + ops = NULL; +out: + return ops; } -static int nldev_res_get_cm_id_dumpit(struct sk_buff *skb, - struct netlink_callback *cb) +void rdma_link_register(struct rdma_link_ops *ops) { - return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_CM_ID); + down_write(&link_ops_rwsem); + if (WARN_ON_ONCE(link_ops_get(ops->type))) + goto out; + list_add(&ops->list, &link_ops); +out: + up_write(&link_ops_rwsem); } +EXPORT_SYMBOL(rdma_link_register); -static int nldev_res_get_cq_dumpit(struct sk_buff *skb, - struct netlink_callback *cb) +void rdma_link_unregister(struct rdma_link_ops *ops) { - return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_CQ); + down_write(&link_ops_rwsem); + list_del(&ops->list); + up_write(&link_ops_rwsem); } +EXPORT_SYMBOL(rdma_link_unregister); -static int nldev_res_get_mr_dumpit(struct sk_buff *skb, - struct netlink_callback *cb) +static int nldev_newlink(struct sk_buff *skb, struct nlmsghdr *nlh, + struct netlink_ext_ack *extack) { - return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_MR); + struct nlattr *tb[RDMA_NLDEV_ATTR_MAX]; + char ibdev_name[IB_DEVICE_NAME_MAX]; + const struct rdma_link_ops *ops; + char ndev_name[IFNAMSIZ]; + struct net_device *ndev; + char type[IFNAMSIZ]; + int err; + + err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, + nldev_policy, extack); + if (err || !tb[RDMA_NLDEV_ATTR_DEV_NAME] || + !tb[RDMA_NLDEV_ATTR_LINK_TYPE] || !tb[RDMA_NLDEV_ATTR_NDEV_NAME]) + return -EINVAL; + + nla_strlcpy(ibdev_name, tb[RDMA_NLDEV_ATTR_DEV_NAME], + sizeof(ibdev_name)); + if (strchr(ibdev_name, '%')) + return -EINVAL; + + nla_strlcpy(type, tb[RDMA_NLDEV_ATTR_LINK_TYPE], sizeof(type)); + nla_strlcpy(ndev_name, tb[RDMA_NLDEV_ATTR_NDEV_NAME], + sizeof(ndev_name)); + + ndev = dev_get_by_name(&init_net, ndev_name); + if (!ndev) + return -ENODEV; + + down_read(&link_ops_rwsem); + ops = link_ops_get(type); +#ifdef CONFIG_MODULES + if (!ops) { + up_read(&link_ops_rwsem); + request_module("rdma-link-%s", type); + down_read(&link_ops_rwsem); + ops = link_ops_get(type); + } +#endif + err = ops ? ops->newlink(ibdev_name, ndev) : -EINVAL; + up_read(&link_ops_rwsem); + dev_put(ndev); + + return err; } -static int nldev_res_get_pd_dumpit(struct sk_buff *skb, - struct netlink_callback *cb) +static int nldev_dellink(struct sk_buff *skb, struct nlmsghdr *nlh, + struct netlink_ext_ack *extack) { - return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_PD); + struct nlattr *tb[RDMA_NLDEV_ATTR_MAX]; + struct ib_device *device; + u32 index; + int err; + + err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, + nldev_policy, extack); + if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX]) + return -EINVAL; + + index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]); + device = ib_device_get_by_index(index); + if (!device) + return -EINVAL; + + if (!(device->attrs.device_cap_flags & IB_DEVICE_ALLOW_USER_UNREG)) { + ib_device_put(device); + return -EINVAL; + } + + ib_unregister_device_and_put(device); + return 0; } static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = { @@ -1112,6 +1321,14 @@ static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = { .doit = nldev_set_doit, .flags = RDMA_NL_ADMIN_PERM, }, + [RDMA_NLDEV_CMD_NEWLINK] = { + .doit = nldev_newlink, + .flags = RDMA_NL_ADMIN_PERM, + }, + [RDMA_NLDEV_CMD_DELLINK] = { + .doit = nldev_dellink, + .flags = RDMA_NL_ADMIN_PERM, + }, [RDMA_NLDEV_CMD_PORT_GET] = { .doit = nldev_port_get_doit, .dump = nldev_port_get_dumpit, @@ -1121,28 +1338,23 @@ static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = { .dump = nldev_res_get_dumpit, }, [RDMA_NLDEV_CMD_RES_QP_GET] = { + .doit = nldev_res_get_qp_doit, .dump = nldev_res_get_qp_dumpit, - /* - * .doit is not implemented yet for two reasons: - * 1. It is not needed yet. - * 2. There is a need to provide identifier, while it is easy - * for the QPs (device index + port index + LQPN), it is not - * the case for the rest of resources (PD and CQ). Because it - * is better to provide similar interface for all resources, - * let's wait till we will have other resources implemented - * too. - */ }, [RDMA_NLDEV_CMD_RES_CM_ID_GET] = { + .doit = nldev_res_get_cm_id_doit, .dump = nldev_res_get_cm_id_dumpit, }, [RDMA_NLDEV_CMD_RES_CQ_GET] = { + .doit = nldev_res_get_cq_doit, .dump = nldev_res_get_cq_dumpit, }, [RDMA_NLDEV_CMD_RES_MR_GET] = { + .doit = nldev_res_get_mr_doit, .dump = nldev_res_get_mr_dumpit, }, [RDMA_NLDEV_CMD_RES_PD_GET] = { + .doit = nldev_res_get_pd_doit, .dump = nldev_res_get_pd_dumpit, }, }; diff --git a/drivers/infiniband/core/rdma_core.c b/drivers/infiniband/core/rdma_core.c index 6c4747e61d2b..778375ff664e 100644 --- a/drivers/infiniband/core/rdma_core.c +++ b/drivers/infiniband/core/rdma_core.c @@ -438,6 +438,38 @@ free: uverbs_uobject_put(uobj); return ERR_PTR(ret); } +struct ib_uobject *_uobj_get_read(enum uverbs_default_objects type, + u32 object_id, + struct uverbs_attr_bundle *attrs) +{ + struct ib_uobject *uobj; + + uobj = rdma_lookup_get_uobject(uobj_get_type(attrs, type), attrs->ufile, + object_id, UVERBS_LOOKUP_READ); + if (IS_ERR(uobj)) + return uobj; + + attrs->context = uobj->context; + + return uobj; +} + +struct ib_uobject *_uobj_get_write(enum uverbs_default_objects type, + u32 object_id, + struct uverbs_attr_bundle *attrs) +{ + struct ib_uobject *uobj; + + uobj = rdma_lookup_get_uobject(uobj_get_type(attrs, type), attrs->ufile, + object_id, UVERBS_LOOKUP_WRITE); + + if (IS_ERR(uobj)) + return uobj; + + attrs->context = uobj->context; + + return uobj; +} static struct ib_uobject * alloc_begin_idr_uobject(const struct uverbs_api_object *obj, @@ -801,6 +833,7 @@ void uverbs_close_fd(struct file *f) /* Pairs with filp->private_data in alloc_begin_fd_uobject */ uverbs_uobject_put(uobj); } +EXPORT_SYMBOL(uverbs_close_fd); /* * Drop the ucontext off the ufile and completely disconnect it from the @@ -811,7 +844,6 @@ static void ufile_destroy_ucontext(struct ib_uverbs_file *ufile, { struct ib_ucontext *ucontext = ufile->ucontext; struct ib_device *ib_dev = ucontext->device; - int ret; /* * If we are closing the FD then the user mmap VMAs must have @@ -829,12 +861,8 @@ static void ufile_destroy_ucontext(struct ib_uverbs_file *ufile, rdma_restrack_del(&ucontext->res); - /* - * FIXME: Drivers are not permitted to fail dealloc_ucontext, remove - * the error return. - */ - ret = ib_dev->ops.dealloc_ucontext(ucontext); - WARN_ON(ret); + ib_dev->ops.dealloc_ucontext(ucontext); + kfree(ucontext); ufile->ucontext = NULL; } diff --git a/drivers/infiniband/core/restrack.c b/drivers/infiniband/core/restrack.c index 46a5c553c624..fa804093fafb 100644 --- a/drivers/infiniband/core/restrack.c +++ b/drivers/infiniband/core/restrack.c @@ -11,17 +11,51 @@ #include <linux/pid_namespace.h> #include "cma_priv.h" +#include "restrack.h" -static int fill_res_noop(struct sk_buff *msg, - struct rdma_restrack_entry *entry) +static int rt_xa_alloc_cyclic(struct xarray *xa, u32 *id, void *entry, + u32 *next) { - return 0; + int err; + + *id = *next; + if (*next == U32_MAX) + *id = 0; + + xa_lock(xa); + err = __xa_alloc(xa, id, U32_MAX, entry, GFP_KERNEL); + if (err && *next != U32_MAX) { + *id = 0; + err = __xa_alloc(xa, id, *next, entry, GFP_KERNEL); + } + + if (!err) + *next = *id + 1; + xa_unlock(xa); + return err; } -void rdma_restrack_init(struct rdma_restrack_root *res) +/** + * rdma_restrack_init() - initialize and allocate resource tracking + * @dev: IB device + * + * Return: 0 on success + */ +int rdma_restrack_init(struct ib_device *dev) { - init_rwsem(&res->rwsem); - res->fill_res_entry = fill_res_noop; + struct rdma_restrack_root *rt; + int i; + + dev->res = kcalloc(RDMA_RESTRACK_MAX, sizeof(*rt), GFP_KERNEL); + if (!dev->res) + return -ENOMEM; + + rt = dev->res; + + for (i = 0; i < RDMA_RESTRACK_MAX; i++) + xa_init_flags(&rt[i].xa, XA_FLAGS_ALLOC); + + return 0; } static const char *type2str(enum rdma_restrack_type type) @@ -38,55 +72,79 @@ static const char *type2str(enum rdma_restrack_type type) return names[type]; }; -void rdma_restrack_clean(struct rdma_restrack_root *res) +/** + * rdma_restrack_clean() - clean resource tracking + * @dev: IB device + */ +void rdma_restrack_clean(struct ib_device *dev) { + struct rdma_restrack_root *rt = dev->res; struct rdma_restrack_entry *e; char buf[TASK_COMM_LEN]; - struct ib_device *dev; + bool found = false; const char *owner; - int bkt; - - if (hash_empty(res->hash)) - return; - - dev = container_of(res, struct ib_device, res); - pr_err("restrack: %s", CUT_HERE); - dev_err(&dev->dev, "BUG: RESTRACK detected leak of resources\n"); - hash_for_each(res->hash, bkt, e, node) { - if (rdma_is_kernel_res(e)) { - owner = e->kern_name; - } else { - /* - * There is no need to call get_task_struct here, - * because we can be here only if there are more - * get_task_struct() call than put_task_struct(). - */ - get_task_comm(buf, e->task); - owner = buf; + int i; + + for (i = 0 ; i < RDMA_RESTRACK_MAX; i++) { + struct xarray *xa = &dev->res[i].xa; + + if (!xa_empty(xa)) { + unsigned long index; + + if (!found) { + pr_err("restrack: %s", CUT_HERE); + dev_err(&dev->dev, "BUG: RESTRACK detected leak of resources\n"); + } + xa_for_each(xa, index, e) { + if (rdma_is_kernel_res(e)) { + owner = e->kern_name; + } else { + /* + * There is no need to call get_task_struct here, + * because we can be here only if there are more + * get_task_struct() call than put_task_struct(). + */ + get_task_comm(buf, e->task); + owner = buf; + } + + pr_err("restrack: %s %s object allocated by %s is not freed\n", + rdma_is_kernel_res(e) ? "Kernel" : + "User", + type2str(e->type), owner); + } + found = true; } - - pr_err("restrack: %s %s object allocated by %s is not freed\n", - rdma_is_kernel_res(e) ? "Kernel" : "User", - type2str(e->type), owner); + xa_destroy(xa); } - pr_err("restrack: %s", CUT_HERE); + if (found) + pr_err("restrack: %s", CUT_HERE); + + kfree(rt); } -int rdma_restrack_count(struct rdma_restrack_root *res, - enum rdma_restrack_type type, +/** + * rdma_restrack_count() - the current usage of specific object + * @dev: IB device + * @type: actual type of object to operate + * @ns: PID namespace + */ +int rdma_restrack_count(struct ib_device *dev, enum rdma_restrack_type type, struct pid_namespace *ns) { + struct rdma_restrack_root *rt = &dev->res[type]; struct rdma_restrack_entry *e; + XA_STATE(xas, &rt->xa, 0); u32 cnt = 0; - down_read(&res->rwsem); - hash_for_each_possible(res->hash, e, node, type) { + xa_lock(&rt->xa); + xas_for_each(&xas, e, U32_MAX) { if (ns == &init_pid_ns || (!rdma_is_kernel_res(e) && ns == task_active_pid_ns(e->task))) cnt++; } - up_read(&res->rwsem); + xa_unlock(&rt->xa); return cnt; } EXPORT_SYMBOL(rdma_restrack_count); @@ -157,28 +215,28 @@ EXPORT_SYMBOL(rdma_restrack_set_task); static void rdma_restrack_add(struct rdma_restrack_entry *res) { struct ib_device *dev = res_to_dev(res); + struct rdma_restrack_root *rt; + int ret; if (!dev) return; - if (res->type != RDMA_RESTRACK_CM_ID || rdma_is_kernel_res(res)) - res->task = NULL; - - if (!rdma_is_kernel_res(res)) { - if (!res->task) - rdma_restrack_set_task(res, NULL); - res->kern_name = NULL; - } else { - set_kern_name(res); - } + rt = &dev->res[res->type]; kref_init(&res->kref); init_completion(&res->comp); - res->valid = true; + if (res->type != RDMA_RESTRACK_QP) + ret = rt_xa_alloc_cyclic(&rt->xa, &res->id, res, &rt->next_id); + else { + /* Special case to ensure that LQPN points to right QP */ + struct ib_qp *qp = container_of(res, struct ib_qp, res); + + ret = xa_insert(&rt->xa, qp->qp_num, res, GFP_KERNEL); + res->id = ret ? 0 : qp->qp_num; + } - down_write(&dev->res.rwsem); - hash_add(dev->res.hash, &res->node, res->type); - up_write(&dev->res.rwsem); + if (!ret) + res->valid = true; } /** @@ -187,6 +245,8 @@ static void rdma_restrack_add(struct rdma_restrack_entry *res) */ void rdma_restrack_kadd(struct rdma_restrack_entry *res) { + res->task = NULL; + set_kern_name(res); res->user = false; rdma_restrack_add(res); } @@ -198,6 +258,13 @@ EXPORT_SYMBOL(rdma_restrack_kadd); */ void rdma_restrack_uadd(struct rdma_restrack_entry *res) { + if (res->type != RDMA_RESTRACK_CM_ID) + res->task = NULL; + + if (!res->task) + rdma_restrack_set_task(res, NULL); + res->kern_name = NULL; + res->user = true; rdma_restrack_add(res); } @@ -209,6 +276,31 @@ int __must_check rdma_restrack_get(struct rdma_restrack_entry *res) } EXPORT_SYMBOL(rdma_restrack_get); +/** + * rdma_restrack_get_byid() - translate from ID to restrack object + * @dev: IB device + * @type: resource track type + * @id: ID to take a look + * + * Return: Pointer to restrack entry or -ENOENT in case of error. + */ +struct rdma_restrack_entry * +rdma_restrack_get_byid(struct ib_device *dev, + enum rdma_restrack_type type, u32 id) +{ + struct rdma_restrack_root *rt = &dev->res[type]; + struct rdma_restrack_entry *res; + + xa_lock(&rt->xa); + res = xa_load(&rt->xa, id); + if (!res || !rdma_restrack_get(res)) + res = ERR_PTR(-ENOENT); + xa_unlock(&rt->xa); + + return res; +} +EXPORT_SYMBOL(rdma_restrack_get_byid); + static void restrack_release(struct kref *kref) { struct rdma_restrack_entry *res; @@ -225,23 +317,25 @@ EXPORT_SYMBOL(rdma_restrack_put); void rdma_restrack_del(struct rdma_restrack_entry *res) { + struct rdma_restrack_entry *old; + struct rdma_restrack_root *rt; struct ib_device *dev; if (!res->valid) goto out; dev = res_to_dev(res); - if (!dev) + if (WARN_ON(!dev)) return; - rdma_restrack_put(res); + rt = &dev->res[res->type]; - wait_for_completion(&res->comp); - - down_write(&dev->res.rwsem); - hash_del(&res->node); + old = xa_erase(&rt->xa, res->id); + WARN_ON(old != res); res->valid = false; - up_write(&dev->res.rwsem); + + rdma_restrack_put(res); + wait_for_completion(&res->comp); out: if (res->task) { diff --git a/drivers/infiniband/core/restrack.h b/drivers/infiniband/core/restrack.h new file mode 100644 index 000000000000..09a1fbdf578e --- /dev/null +++ b/drivers/infiniband/core/restrack.h @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ +/* + * Copyright (c) 2017-2019 Mellanox Technologies. All rights reserved. + */ + +#ifndef _RDMA_CORE_RESTRACK_H_ +#define _RDMA_CORE_RESTRACK_H_ + +#include <linux/mutex.h> + +/** + * struct rdma_restrack_root - main resource tracking management + * entity, per-device + */ +struct rdma_restrack_root { + /** + * @xa: Array of XArray structure to hold restrack entries. + */ + struct xarray xa; + /** + * @next_id: Next ID to support cyclic allocation + */ + u32 next_id; +}; + +int rdma_restrack_init(struct ib_device *dev); +void rdma_restrack_clean(struct ib_device *dev); +#endif /* _RDMA_CORE_RESTRACK_H_ */ diff --git a/drivers/infiniband/core/rw.c b/drivers/infiniband/core/rw.c index d22c4a2ebac6..89a5be3a2f97 100644 --- a/drivers/infiniband/core/rw.c +++ b/drivers/infiniband/core/rw.c @@ -179,7 +179,6 @@ static int rdma_rw_init_map_wrs(struct rdma_rw_ctx *ctx, struct ib_qp *qp, struct scatterlist *sg, u32 sg_cnt, u32 offset, u64 remote_addr, u32 rkey, enum dma_data_direction dir) { - struct ib_device *dev = qp->pd->device; u32 max_sge = dir == DMA_TO_DEVICE ? qp->max_write_sge : qp->max_read_sge; struct ib_sge *sge; @@ -209,8 +208,8 @@ static int rdma_rw_init_map_wrs(struct rdma_rw_ctx *ctx, struct ib_qp *qp, rdma_wr->wr.sg_list = sge; for (j = 0; j < nr_sge; j++, sg = sg_next(sg)) { - sge->addr = ib_sg_dma_address(dev, sg) + offset; - sge->length = ib_sg_dma_len(dev, sg) - offset; + sge->addr = sg_dma_address(sg) + offset; + sge->length = sg_dma_len(sg) - offset; sge->lkey = qp->pd->local_dma_lkey; total_len += sge->length; @@ -236,14 +235,13 @@ static int rdma_rw_init_single_wr(struct rdma_rw_ctx *ctx, struct ib_qp *qp, struct scatterlist *sg, u32 offset, u64 remote_addr, u32 rkey, enum dma_data_direction dir) { - struct ib_device *dev = qp->pd->device; struct ib_rdma_wr *rdma_wr = &ctx->single.wr; ctx->nr_ops = 1; ctx->single.sge.lkey = qp->pd->local_dma_lkey; - ctx->single.sge.addr = ib_sg_dma_address(dev, sg) + offset; - ctx->single.sge.length = ib_sg_dma_len(dev, sg) - offset; + ctx->single.sge.addr = sg_dma_address(sg) + offset; + ctx->single.sge.length = sg_dma_len(sg) - offset; memset(rdma_wr, 0, sizeof(*rdma_wr)); if (dir == DMA_TO_DEVICE) @@ -294,7 +292,7 @@ int rdma_rw_ctx_init(struct rdma_rw_ctx *ctx, struct ib_qp *qp, u8 port_num, * Skip to the S/G entry that sg_offset falls into: */ for (;;) { - u32 len = ib_sg_dma_len(dev, sg); + u32 len = sg_dma_len(sg); if (sg_offset < len) break; diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c index 97e6d7b69abf..7925e45ea88a 100644 --- a/drivers/infiniband/core/sa_query.c +++ b/drivers/infiniband/core/sa_query.c @@ -2342,9 +2342,7 @@ static void ib_sa_add_one(struct ib_device *device) s = rdma_start_port(device); e = rdma_end_port(device); - sa_dev = kzalloc(sizeof *sa_dev + - (e - s + 1) * sizeof (struct ib_sa_port), - GFP_KERNEL); + sa_dev = kzalloc(struct_size(sa_dev, port, e - s + 1), GFP_KERNEL); if (!sa_dev) return; diff --git a/drivers/infiniband/core/security.c b/drivers/infiniband/core/security.c index 1efadbccf394..1ab423b19f77 100644 --- a/drivers/infiniband/core/security.c +++ b/drivers/infiniband/core/security.c @@ -39,22 +39,25 @@ #include "core_priv.h" #include "mad_priv.h" +static LIST_HEAD(mad_agent_list); +/* Lock to protect mad_agent_list */ +static DEFINE_SPINLOCK(mad_agent_list_lock); + static struct pkey_index_qp_list *get_pkey_idx_qp_list(struct ib_port_pkey *pp) { struct pkey_index_qp_list *pkey = NULL; struct pkey_index_qp_list *tmp_pkey; struct ib_device *dev = pp->sec->dev; - spin_lock(&dev->port_pkey_list[pp->port_num].list_lock); - list_for_each_entry(tmp_pkey, - &dev->port_pkey_list[pp->port_num].pkey_list, - pkey_index_list) { + spin_lock(&dev->port_data[pp->port_num].pkey_list_lock); + list_for_each_entry (tmp_pkey, &dev->port_data[pp->port_num].pkey_list, + pkey_index_list) { if (tmp_pkey->pkey_index == pp->pkey_index) { pkey = tmp_pkey; break; } } - spin_unlock(&dev->port_pkey_list[pp->port_num].list_lock); + spin_unlock(&dev->port_data[pp->port_num].pkey_list_lock); return pkey; } @@ -259,12 +262,12 @@ static int port_pkey_list_insert(struct ib_port_pkey *pp) if (!pkey) return -ENOMEM; - spin_lock(&dev->port_pkey_list[port_num].list_lock); + spin_lock(&dev->port_data[port_num].pkey_list_lock); /* Check for the PKey again. A racing process may * have created it. */ list_for_each_entry(tmp_pkey, - &dev->port_pkey_list[port_num].pkey_list, + &dev->port_data[port_num].pkey_list, pkey_index_list) { if (tmp_pkey->pkey_index == pp->pkey_index) { kfree(pkey); @@ -279,9 +282,9 @@ static int port_pkey_list_insert(struct ib_port_pkey *pp) spin_lock_init(&pkey->qp_list_lock); INIT_LIST_HEAD(&pkey->qp_list); list_add(&pkey->pkey_index_list, - &dev->port_pkey_list[port_num].pkey_list); + &dev->port_data[port_num].pkey_list); } - spin_unlock(&dev->port_pkey_list[port_num].list_lock); + spin_unlock(&dev->port_data[port_num].pkey_list_lock); } spin_lock(&pkey->qp_list_lock); @@ -418,12 +421,15 @@ void ib_close_shared_qp_security(struct ib_qp_security *sec) int ib_create_qp_security(struct ib_qp *qp, struct ib_device *dev) { - u8 i = rdma_start_port(dev); + unsigned int i; bool is_ib = false; int ret; - while (i <= rdma_end_port(dev) && !is_ib) + rdma_for_each_port (dev, i) { is_ib = rdma_protocol_ib(dev, i++); + if (is_ib) + break; + } /* If this isn't an IB device don't create the security context */ if (!is_ib) @@ -544,9 +550,8 @@ void ib_security_cache_change(struct ib_device *device, { struct pkey_index_qp_list *pkey; - list_for_each_entry(pkey, - &device->port_pkey_list[port_num].pkey_list, - pkey_index_list) { + list_for_each_entry (pkey, &device->port_data[port_num].pkey_list, + pkey_index_list) { check_pkey_qps(pkey, device, port_num, @@ -554,21 +559,19 @@ void ib_security_cache_change(struct ib_device *device, } } -void ib_security_destroy_port_pkey_list(struct ib_device *device) +void ib_security_release_port_pkey_list(struct ib_device *device) { struct pkey_index_qp_list *pkey, *tmp_pkey; - int i; + unsigned int i; - for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) { - spin_lock(&device->port_pkey_list[i].list_lock); + rdma_for_each_port (device, i) { list_for_each_entry_safe(pkey, tmp_pkey, - &device->port_pkey_list[i].pkey_list, + &device->port_data[i].pkey_list, pkey_index_list) { list_del(&pkey->pkey_index_list); kfree(pkey); } - spin_unlock(&device->port_pkey_list[i].list_lock); } } @@ -676,19 +679,18 @@ static int ib_security_pkey_access(struct ib_device *dev, return security_ib_pkey_access(sec, subnet_prefix, pkey); } -static int ib_mad_agent_security_change(struct notifier_block *nb, - unsigned long event, - void *data) +void ib_mad_agent_security_change(void) { - struct ib_mad_agent *ag = container_of(nb, struct ib_mad_agent, lsm_nb); - - if (event != LSM_POLICY_CHANGE) - return NOTIFY_DONE; - - ag->smp_allowed = !security_ib_endport_manage_subnet( - ag->security, dev_name(&ag->device->dev), ag->port_num); - - return NOTIFY_OK; + struct ib_mad_agent *ag; + + spin_lock(&mad_agent_list_lock); + list_for_each_entry(ag, + &mad_agent_list, + mad_agent_sec_list) + WRITE_ONCE(ag->smp_allowed, + !security_ib_endport_manage_subnet(ag->security, + dev_name(&ag->device->dev), ag->port_num)); + spin_unlock(&mad_agent_list_lock); } int ib_mad_agent_security_setup(struct ib_mad_agent *agent, @@ -699,6 +701,8 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent, if (!rdma_protocol_ib(agent->device, agent->port_num)) return 0; + INIT_LIST_HEAD(&agent->mad_agent_sec_list); + ret = security_ib_alloc_security(&agent->security); if (ret) return ret; @@ -706,20 +710,22 @@ int ib_mad_agent_security_setup(struct ib_mad_agent *agent, if (qp_type != IB_QPT_SMI) return 0; + spin_lock(&mad_agent_list_lock); ret = security_ib_endport_manage_subnet(agent->security, dev_name(&agent->device->dev), agent->port_num); if (ret) - return ret; + goto free_security; - agent->lsm_nb.notifier_call = ib_mad_agent_security_change; - ret = register_lsm_notifier(&agent->lsm_nb); - if (ret) - return ret; - - agent->smp_allowed = true; - agent->lsm_nb_reg = true; + WRITE_ONCE(agent->smp_allowed, true); + list_add(&agent->mad_agent_sec_list, &mad_agent_list); + spin_unlock(&mad_agent_list_lock); return 0; + +free_security: + spin_unlock(&mad_agent_list_lock); + security_ib_free_security(agent->security); + return ret; } void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent) @@ -727,9 +733,13 @@ void ib_mad_agent_security_cleanup(struct ib_mad_agent *agent) if (!rdma_protocol_ib(agent->device, agent->port_num)) return; + if (agent->qp->qp_type == IB_QPT_SMI) { + spin_lock(&mad_agent_list_lock); + list_del(&agent->mad_agent_sec_list); + spin_unlock(&mad_agent_list_lock); + } + security_ib_free_security(agent->security); - if (agent->lsm_nb_reg) - unregister_lsm_notifier(&agent->lsm_nb); } int ib_mad_enforce_security(struct ib_mad_agent_private *map, u16 pkey_index) @@ -738,7 +748,7 @@ int ib_mad_enforce_security(struct ib_mad_agent_private *map, u16 pkey_index) return 0; if (map->agent.qp->qp_type == IB_QPT_SMI) { - if (!map->agent.smp_allowed) + if (!READ_ONCE(map->agent.smp_allowed)) return -EACCES; return 0; } diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c index 80f68eb0ba5c..9b6a065bdfa5 100644 --- a/drivers/infiniband/core/sysfs.c +++ b/drivers/infiniband/core/sysfs.c @@ -1015,9 +1015,7 @@ err_free_stats: return; } -static int add_port(struct ib_device *device, int port_num, - int (*port_callback)(struct ib_device *, - u8, struct kobject *)) +static int add_port(struct ib_device *device, int port_num) { struct ib_port *p; struct ib_port_attr attr; @@ -1113,8 +1111,8 @@ static int add_port(struct ib_device *device, int port_num, if (ret) goto err_free_pkey; - if (port_callback) { - ret = port_callback(device, port_num, &p->kobj); + if (device->ops.init_port) { + ret = device->ops.init_port(device, port_num, &p->kobj); if (ret) goto err_remove_pkey; } @@ -1189,7 +1187,7 @@ err_put: static ssize_t node_type_show(struct device *device, struct device_attribute *attr, char *buf) { - struct ib_device *dev = container_of(device, struct ib_device, dev); + struct ib_device *dev = rdma_device_to_ibdev(device); switch (dev->node_type) { case RDMA_NODE_IB_CA: return sprintf(buf, "%d: CA\n", dev->node_type); @@ -1206,7 +1204,7 @@ static DEVICE_ATTR_RO(node_type); static ssize_t sys_image_guid_show(struct device *device, struct device_attribute *dev_attr, char *buf) { - struct ib_device *dev = container_of(device, struct ib_device, dev); + struct ib_device *dev = rdma_device_to_ibdev(device); return sprintf(buf, "%04x:%04x:%04x:%04x\n", be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]), @@ -1219,7 +1217,7 @@ static DEVICE_ATTR_RO(sys_image_guid); static ssize_t node_guid_show(struct device *device, struct device_attribute *attr, char *buf) { - struct ib_device *dev = container_of(device, struct ib_device, dev); + struct ib_device *dev = rdma_device_to_ibdev(device); return sprintf(buf, "%04x:%04x:%04x:%04x\n", be16_to_cpu(((__be16 *) &dev->node_guid)[0]), @@ -1232,7 +1230,7 @@ static DEVICE_ATTR_RO(node_guid); static ssize_t node_desc_show(struct device *device, struct device_attribute *attr, char *buf) { - struct ib_device *dev = container_of(device, struct ib_device, dev); + struct ib_device *dev = rdma_device_to_ibdev(device); return sprintf(buf, "%.64s\n", dev->node_desc); } @@ -1241,7 +1239,7 @@ static ssize_t node_desc_store(struct device *device, struct device_attribute *attr, const char *buf, size_t count) { - struct ib_device *dev = container_of(device, struct ib_device, dev); + struct ib_device *dev = rdma_device_to_ibdev(device); struct ib_device_modify desc = {}; int ret; @@ -1260,7 +1258,7 @@ static DEVICE_ATTR_RW(node_desc); static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr, char *buf) { - struct ib_device *dev = container_of(device, struct ib_device, dev); + struct ib_device *dev = rdma_device_to_ibdev(device); ib_get_device_fw_str(dev, buf); strlcat(buf, "\n", IB_FW_VERSION_NAME_MAX); @@ -1277,21 +1275,21 @@ static struct attribute *ib_dev_attrs[] = { NULL, }; -static const struct attribute_group dev_attr_group = { +const struct attribute_group ib_dev_attr_group = { .attrs = ib_dev_attrs, }; -static void free_port_list_attributes(struct ib_device *device) +static void ib_free_port_attrs(struct ib_device *device) { struct kobject *p, *t; list_for_each_entry_safe(p, t, &device->port_list, entry) { struct ib_port *port = container_of(p, struct ib_port, kobj); + list_del(&p->entry); - if (port->hw_stats) { - kfree(port->hw_stats); + if (port->hw_stats_ag) free_hsag(&port->kobj, port->hw_stats_ag); - } + kfree(port->hw_stats); if (port->pma_table) sysfs_remove_group(p, port->pma_table); @@ -1308,62 +1306,47 @@ static void free_port_list_attributes(struct ib_device *device) kobject_put(device->ports_kobj); } -int ib_device_register_sysfs(struct ib_device *device, - int (*port_callback)(struct ib_device *, - u8, struct kobject *)) +static int ib_setup_port_attrs(struct ib_device *device) { - struct device *class_dev = &device->dev; + unsigned int port; int ret; - int i; - - device->groups[0] = &dev_attr_group; - class_dev->groups = device->groups; - ret = device_add(class_dev); - if (ret) - goto err; - - device->ports_kobj = kobject_create_and_add("ports", &class_dev->kobj); - if (!device->ports_kobj) { - ret = -ENOMEM; - goto err_put; - } + device->ports_kobj = kobject_create_and_add("ports", &device->dev.kobj); + if (!device->ports_kobj) + return -ENOMEM; - if (rdma_cap_ib_switch(device)) { - ret = add_port(device, 0, port_callback); + rdma_for_each_port (device, port) { + ret = add_port(device, port); if (ret) goto err_put; - } else { - for (i = 1; i <= device->phys_port_cnt; ++i) { - ret = add_port(device, i, port_callback); - if (ret) - goto err_put; - } } - if (device->ops.alloc_hw_stats) - setup_hw_stats(device, NULL, 0); - return 0; err_put: - free_port_list_attributes(device); - device_del(class_dev); -err: + ib_free_port_attrs(device); return ret; } -void ib_device_unregister_sysfs(struct ib_device *device) +int ib_device_register_sysfs(struct ib_device *device) { - /* Hold device until ib_dealloc_device() */ - get_device(&device->dev); + int ret; + + ret = ib_setup_port_attrs(device); + if (ret) + return ret; + + if (device->ops.alloc_hw_stats) + setup_hw_stats(device, NULL, 0); - free_port_list_attributes(device); + return 0; +} - if (device->hw_stats) { - kfree(device->hw_stats); +void ib_device_unregister_sysfs(struct ib_device *device) +{ + if (device->hw_stats_ag) free_hsag(&device->dev.kobj, device->hw_stats_ag); - } + kfree(device->hw_stats); - device_unregister(&device->dev); + ib_free_port_attrs(device); } diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 01d68ed46c1b..7468b26b8a01 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -1236,6 +1236,13 @@ static int ucma_set_option_id(struct ucma_context *ctx, int optname, } ret = rdma_set_afonly(ctx->cm_id, *((int *) optval) ? 1 : 0); break; + case RDMA_OPTION_ID_ACK_TIMEOUT: + if (optlen != sizeof(u8)) { + ret = -EINVAL; + break; + } + ret = rdma_set_ack_timeout(ctx->cm_id, *((u8 *)optval)); + break; default: ret = -ENOSYS; } diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c index c6144df47ea4..fe5551562dbc 100644 --- a/drivers/infiniband/core/umem.c +++ b/drivers/infiniband/core/umem.c @@ -72,15 +72,16 @@ static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int d * If access flags indicate ODP memory, avoid pinning. Instead, stores * the mm for future page fault handling in conjunction with MMU notifiers. * - * @context: userspace context to pin memory for + * @udata: userspace context to pin memory for * @addr: userspace virtual address to start at * @size: length of region to pin * @access: IB_ACCESS_xxx flags for memory being pinned * @dmasync: flush in-flight DMA when the memory region is written */ -struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, +struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr, size_t size, int access, int dmasync) { + struct ib_ucontext *context; struct ib_umem *umem; struct page **page_list; struct vm_area_struct **vma_list; @@ -95,6 +96,14 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, struct scatterlist *sg, *sg_list_start; unsigned int gup_flags = FOLL_WRITE; + if (!udata) + return ERR_PTR(-EIO); + + context = container_of(udata, struct uverbs_attr_bundle, driver_udata) + ->context; + if (!context) + return ERR_PTR(-EIO); + if (dmasync) dma_attrs |= DMA_ATTR_WRITE_BARRIER; @@ -160,15 +169,12 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; - down_write(&mm->mmap_sem); - if (check_add_overflow(mm->pinned_vm, npages, &new_pinned) || - (new_pinned > lock_limit && !capable(CAP_IPC_LOCK))) { - up_write(&mm->mmap_sem); + new_pinned = atomic64_add_return(npages, &mm->pinned_vm); + if (new_pinned > lock_limit && !capable(CAP_IPC_LOCK)) { + atomic64_sub(npages, &mm->pinned_vm); ret = -ENOMEM; goto out; } - mm->pinned_vm = new_pinned; - up_write(&mm->mmap_sem); cur_base = addr & PAGE_MASK; @@ -228,9 +234,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, umem_release: __ib_umem_release(context->device, umem, 0); vma: - down_write(&mm->mmap_sem); - mm->pinned_vm -= ib_umem_num_pages(umem); - up_write(&mm->mmap_sem); + atomic64_sub(ib_umem_num_pages(umem), &mm->pinned_vm); out: if (vma_list) free_page((unsigned long) vma_list); @@ -253,25 +257,12 @@ static void __ib_umem_release_tail(struct ib_umem *umem) kfree(umem); } -static void ib_umem_release_defer(struct work_struct *work) -{ - struct ib_umem *umem = container_of(work, struct ib_umem, work); - - down_write(&umem->owning_mm->mmap_sem); - umem->owning_mm->pinned_vm -= ib_umem_num_pages(umem); - up_write(&umem->owning_mm->mmap_sem); - - __ib_umem_release_tail(umem); -} - /** * ib_umem_release - release memory pinned with ib_umem_get * @umem: umem struct to release */ void ib_umem_release(struct ib_umem *umem) { - struct ib_ucontext *context = umem->context; - if (umem->is_odp) { ib_umem_odp_release(to_ib_umem_odp(umem)); __ib_umem_release_tail(umem); @@ -280,26 +271,7 @@ void ib_umem_release(struct ib_umem *umem) __ib_umem_release(umem->context->device, umem, 1); - /* - * We may be called with the mm's mmap_sem already held. This - * can happen when a userspace munmap() is the call that drops - * the last reference to our file and calls our release - * method. If there are memory regions to destroy, we'll end - * up here and not be able to take the mmap_sem. In that case - * we defer the vm_locked accounting a workqueue. - */ - if (context->closing) { - if (!down_write_trylock(&umem->owning_mm->mmap_sem)) { - INIT_WORK(&umem->work, ib_umem_release_defer); - queue_work(ib_wq, &umem->work); - return; - } - } else { - down_write(&umem->owning_mm->mmap_sem); - } - umem->owning_mm->pinned_vm -= ib_umem_num_pages(umem); - up_write(&umem->owning_mm->mmap_sem); - + atomic64_sub(ib_umem_num_pages(umem), &umem->owning_mm->pinned_vm); __ib_umem_release_tail(umem); } EXPORT_SYMBOL(ib_umem_release); diff --git a/drivers/infiniband/core/umem_odp.c b/drivers/infiniband/core/umem_odp.c index acb882f279cb..e6ec79ad9cc8 100644 --- a/drivers/infiniband/core/umem_odp.c +++ b/drivers/infiniband/core/umem_odp.c @@ -40,6 +40,7 @@ #include <linux/vmalloc.h> #include <linux/hugetlb.h> #include <linux/interval_tree_generic.h> +#include <linux/pagemap.h> #include <rdma/ib_verbs.h> #include <rdma/ib_umem.h> @@ -299,7 +300,7 @@ static void free_per_mm(struct rcu_head *rcu) kfree(container_of(rcu, struct ib_ucontext_per_mm, rcu)); } -void put_per_mm(struct ib_umem_odp *umem_odp) +static void put_per_mm(struct ib_umem_odp *umem_odp) { struct ib_ucontext_per_mm *per_mm = umem_odp->per_mm; struct ib_ucontext *ctx = umem_odp->umem.context; @@ -332,9 +333,10 @@ void put_per_mm(struct ib_umem_odp *umem_odp) mmu_notifier_call_srcu(&per_mm->rcu, free_per_mm); } -struct ib_umem_odp *ib_alloc_odp_umem(struct ib_ucontext_per_mm *per_mm, +struct ib_umem_odp *ib_alloc_odp_umem(struct ib_umem_odp *root, unsigned long addr, size_t size) { + struct ib_ucontext_per_mm *per_mm = root->per_mm; struct ib_ucontext *ctx = per_mm->context; struct ib_umem_odp *odp_data; struct ib_umem *umem; @@ -349,7 +351,7 @@ struct ib_umem_odp *ib_alloc_odp_umem(struct ib_ucontext_per_mm *per_mm, umem->length = size; umem->address = addr; umem->page_shift = PAGE_SHIFT; - umem->writable = 1; + umem->writable = root->umem.writable; umem->is_odp = 1; odp_data->per_mm = per_mm; umem->owning_mm = per_mm->mm; @@ -617,7 +619,7 @@ int ib_umem_odp_map_dma_pages(struct ib_umem_odp *umem_odp, u64 user_virt, * mmget_not_zero will fail in this case. */ owning_process = get_pid_task(umem_odp->per_mm->tgid, PIDTYPE_PID); - if (WARN_ON(!mmget_not_zero(umem_odp->umem.owning_mm))) { + if (!owning_process || !mmget_not_zero(owning_mm)) { ret = -EINVAL; goto out_put_task; } @@ -684,9 +686,14 @@ int ib_umem_odp_map_dma_pages(struct ib_umem_odp *umem_odp, u64 user_virt, mutex_unlock(&umem_odp->umem_mutex); if (ret < 0) { - /* Release left over pages when handling errors. */ - for (++j; j < npages; ++j) - put_page(local_page_list[j]); + /* + * Release pages, remembering that the first page + * to hit an error was already released by + * ib_umem_odp_map_dma_single_page(). + */ + if (npages - (j + 1) > 0) + release_pages(&local_page_list[j+1], + npages - (j + 1)); break; } } diff --git a/drivers/infiniband/core/user_mad.c b/drivers/infiniband/core/user_mad.c index de8d31ab8945..02b7947ab215 100644 --- a/drivers/infiniband/core/user_mad.c +++ b/drivers/infiniband/core/user_mad.c @@ -957,19 +957,22 @@ static int ib_umad_open(struct inode *inode, struct file *filp) { struct ib_umad_port *port; struct ib_umad_file *file; - int ret = -ENXIO; + int ret = 0; port = container_of(inode->i_cdev, struct ib_umad_port, cdev); mutex_lock(&port->file_mutex); - if (!port->ib_dev) + if (!port->ib_dev) { + ret = -ENXIO; goto out; + } - ret = -ENOMEM; - file = kzalloc(sizeof *file, GFP_KERNEL); - if (!file) + file = kzalloc(sizeof(*file), GFP_KERNEL); + if (!file) { + ret = -ENOMEM; goto out; + } mutex_init(&file->mutex); spin_lock_init(&file->send_lock); @@ -982,14 +985,7 @@ static int ib_umad_open(struct inode *inode, struct file *filp) list_add_tail(&file->port_list, &port->file_list); - ret = nonseekable_open(inode, filp); - if (ret) { - list_del(&file->port_list); - kfree(file); - goto out; - } - - ib_umad_dev_get(port->umad_dev); + nonseekable_open(inode, filp); out: mutex_unlock(&port->file_mutex); return ret; @@ -998,7 +994,6 @@ out: static int ib_umad_close(struct inode *inode, struct file *filp) { struct ib_umad_file *file = filp->private_data; - struct ib_umad_device *dev = file->port->umad_dev; struct ib_umad_packet *packet, *tmp; int already_dead; int i; @@ -1027,7 +1022,6 @@ static int ib_umad_close(struct inode *inode, struct file *filp) mutex_unlock(&file->port->file_mutex); kfree(file); - ib_umad_dev_put(dev); return 0; } @@ -1073,17 +1067,9 @@ static int ib_umad_sm_open(struct inode *inode, struct file *filp) filp->private_data = port; - ret = nonseekable_open(inode, filp); - if (ret) - goto err_clr_sm_cap; - - ib_umad_dev_get(port->umad_dev); + nonseekable_open(inode, filp); return 0; -err_clr_sm_cap: - swap(props.set_port_cap_mask, props.clr_port_cap_mask); - ib_modify_port(port->ib_dev, port->port_num, 0, &props); - err_up_sem: up(&port->sm_sem); @@ -1106,7 +1092,6 @@ static int ib_umad_sm_close(struct inode *inode, struct file *filp) up(&port->sm_sem); - ib_umad_dev_put(port->umad_dev); return ret; } @@ -1283,10 +1268,12 @@ static void ib_umad_kill_port(struct ib_umad_port *port) mutex_unlock(&port->file_mutex); cdev_device_del(&port->sm_cdev, &port->sm_dev); - put_device(&port->sm_dev); cdev_device_del(&port->cdev, &port->dev); - put_device(&port->dev); ida_free(&umad_ida, port->dev_num); + + /* balances device_initialize() */ + put_device(&port->sm_dev); + put_device(&port->dev); } static void ib_umad_add_one(struct ib_device *device) @@ -1329,21 +1316,24 @@ err: ib_umad_kill_port(&umad_dev->ports[i - s]); } free: + /* balances kref_init */ ib_umad_dev_put(umad_dev); } static void ib_umad_remove_one(struct ib_device *device, void *client_data) { struct ib_umad_device *umad_dev = client_data; - int i; + unsigned int i; if (!umad_dev) return; - for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) { - if (rdma_cap_ib_mad(device, i + rdma_start_port(device))) - ib_umad_kill_port(&umad_dev->ports[i]); + rdma_for_each_port (device, i) { + if (rdma_cap_ib_mad(device, i)) + ib_umad_kill_port( + &umad_dev->ports[i - rdma_start_port(device)]); } + /* balances kref_init() */ ib_umad_dev_put(umad_dev); } diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c index 3317300ab036..062a86c04123 100644 --- a/drivers/infiniband/core/uverbs_cmd.c +++ b/drivers/infiniband/core/uverbs_cmd.c @@ -224,12 +224,13 @@ static int ib_uverbs_get_context(struct uverbs_attr_bundle *attrs) if (ret) goto err; - ucontext = ib_dev->ops.alloc_ucontext(ib_dev, &attrs->driver_udata); - if (IS_ERR(ucontext)) { - ret = PTR_ERR(ucontext); + ucontext = rdma_zalloc_drv_obj(ib_dev, ib_ucontext); + if (!ucontext) { + ret = -ENOMEM; goto err_alloc; } + ucontext->res.type = RDMA_RESTRACK_CTX; ucontext->device = ib_dev; ucontext->cg_obj = cg_obj; /* ufile is required when some objects are released */ @@ -238,15 +239,8 @@ static int ib_uverbs_get_context(struct uverbs_attr_bundle *attrs) ucontext->closing = false; ucontext->cleanup_retryable = false; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING mutex_init(&ucontext->per_mm_list_lock); INIT_LIST_HEAD(&ucontext->per_mm_list); - if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING)) - ucontext->invalidate_range = NULL; - -#endif - - resp.num_comp_vectors = file->device->num_comp_vectors; ret = get_unused_fd_flags(O_CLOEXEC); if (ret < 0) @@ -259,15 +253,22 @@ static int ib_uverbs_get_context(struct uverbs_attr_bundle *attrs) goto err_fd; } + resp.num_comp_vectors = file->device->num_comp_vectors; + ret = uverbs_response(attrs, &resp, sizeof(resp)); if (ret) goto err_file; - fd_install(resp.async_fd, filp); + ret = ib_dev->ops.alloc_ucontext(ucontext, &attrs->driver_udata); + if (ret) + goto err_file; + if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING)) + ucontext->invalidate_range = NULL; - ucontext->res.type = RDMA_RESTRACK_CTX; rdma_restrack_uadd(&ucontext->res); + fd_install(resp.async_fd, filp); + /* * Make sure that ib_uverbs_get_ucontext() sees the pointer update * only after all writes to setup the ucontext have completed @@ -286,7 +287,7 @@ err_fd: put_unused_fd(resp.async_fd); err_free: - ib_dev->ops.dealloc_ucontext(ucontext); + kfree(ucontext); err_alloc: ib_rdmacg_uncharge(&cg_obj, ib_dev, RDMACG_RESOURCE_HCA_HANDLE); @@ -410,9 +411,9 @@ static int ib_uverbs_alloc_pd(struct uverbs_attr_bundle *attrs) if (IS_ERR(uobj)) return PTR_ERR(uobj); - pd = ib_dev->ops.alloc_pd(ib_dev, uobj->context, &attrs->driver_udata); - if (IS_ERR(pd)) { - ret = PTR_ERR(pd); + pd = rdma_zalloc_drv_obj(ib_dev, ib_pd); + if (!pd) { + ret = -ENOMEM; goto err; } @@ -420,11 +421,15 @@ static int ib_uverbs_alloc_pd(struct uverbs_attr_bundle *attrs) pd->uobject = uobj; pd->__internal_mr = NULL; atomic_set(&pd->usecnt, 0); + pd->res.type = RDMA_RESTRACK_PD; + + ret = ib_dev->ops.alloc_pd(pd, uobj->context, &attrs->driver_udata); + if (ret) + goto err_alloc; uobj->object = pd; memset(&resp, 0, sizeof resp); resp.pd_handle = uobj->id; - pd->res.type = RDMA_RESTRACK_PD; rdma_restrack_uadd(&pd->res); ret = uverbs_response(attrs, &resp, sizeof(resp)); @@ -435,7 +440,9 @@ static int ib_uverbs_alloc_pd(struct uverbs_attr_bundle *attrs) err_copy: ib_dealloc_pd(pd); - + pd = NULL; +err_alloc: + kfree(pd); err: uobj_alloc_abort(uobj); return ret; @@ -822,14 +829,13 @@ static int ib_uverbs_rereg_mr(struct uverbs_attr_bundle *attrs) cmd.length, cmd.hca_va, cmd.access_flags, pd, &attrs->driver_udata); - if (!ret) { - if (cmd.flags & IB_MR_REREG_PD) { - atomic_inc(&pd->usecnt); - mr->pd = pd; - atomic_dec(&old_pd->usecnt); - } - } else { + if (ret) goto put_uobj_pd; + + if (cmd.flags & IB_MR_REREG_PD) { + atomic_inc(&pd->usecnt); + mr->pd = pd; + atomic_dec(&old_pd->usecnt); } memset(&resp, 0, sizeof(resp)); @@ -884,6 +890,11 @@ static int ib_uverbs_alloc_mw(struct uverbs_attr_bundle *attrs) goto err_free; } + if (cmd.mw_type != IB_MW_TYPE_1 && cmd.mw_type != IB_MW_TYPE_2) { + ret = -EINVAL; + goto err_put; + } + mw = pd->device->ops.alloc_mw(pd, cmd.mw_type, &attrs->driver_udata); if (IS_ERR(mw)) { ret = PTR_ERR(mw); @@ -1184,12 +1195,11 @@ static int ib_uverbs_poll_cq(struct uverbs_attr_bundle *attrs) ret = -EFAULT; goto out_put; } + ret = 0; if (uverbs_attr_is_valid(attrs, UVERBS_ATTR_CORE_OUT)) ret = uverbs_output_written(attrs, UVERBS_ATTR_CORE_OUT); - ret = 0; - out_put: uobj_put_obj_read(cq); return ret; @@ -2632,7 +2642,7 @@ void flow_resources_add(struct ib_uflow_resources *uflow_res, } EXPORT_SYMBOL(flow_resources_add); -static int kern_spec_to_ib_spec_action(const struct uverbs_attr_bundle *attrs, +static int kern_spec_to_ib_spec_action(struct uverbs_attr_bundle *attrs, struct ib_uverbs_flow_spec *kern_spec, union ib_flow_spec *ib_spec, struct ib_uflow_resources *uflow_res) @@ -3618,7 +3628,6 @@ static int ib_uverbs_ex_query_device(struct uverbs_attr_bundle *attrs) copy_query_dev_fields(ucontext, &resp.base, &attr); -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING resp.odp_caps.general_caps = attr.odp_caps.general_caps; resp.odp_caps.per_transport_caps.rc_odp_caps = attr.odp_caps.per_transport_caps.rc_odp_caps; @@ -3626,7 +3635,7 @@ static int ib_uverbs_ex_query_device(struct uverbs_attr_bundle *attrs) attr.odp_caps.per_transport_caps.uc_odp_caps; resp.odp_caps.per_transport_caps.ud_odp_caps = attr.odp_caps.per_transport_caps.ud_odp_caps; -#endif + resp.xrc_odp_caps = attr.odp_caps.per_transport_caps.xrc_odp_caps; resp.timestamp_mask = attr.timestamp_mask; resp.hca_core_clock = attr.hca_core_clock; diff --git a/drivers/infiniband/core/uverbs_ioctl.c b/drivers/infiniband/core/uverbs_ioctl.c index 0ca04d224015..e1379949e663 100644 --- a/drivers/infiniband/core/uverbs_ioctl.c +++ b/drivers/infiniband/core/uverbs_ioctl.c @@ -213,6 +213,7 @@ static int uverbs_process_idrs_array(struct bundle_priv *pbundle, ret = PTR_ERR(attr->uobjects[i]); break; } + pbundle->bundle.context = attr->uobjects[i]->context; } attr->len = i; @@ -330,6 +331,7 @@ static int uverbs_process_attr(struct bundle_priv *pbundle, uattr->data_s64); if (IS_ERR(o_attr->uobject)) return PTR_ERR(o_attr->uobject); + pbundle->bundle.context = o_attr->uobject->context; __set_bit(attr_bkey, pbundle->uobj_finalize); if (spec->u.obj.access == UVERBS_ACCESS_NEW) { @@ -592,6 +594,7 @@ static int ib_uverbs_cmd_verbs(struct ib_uverbs_file *ufile, pbundle->method_elm = method_elm; pbundle->method_key = attrs_iter.index; pbundle->bundle.ufile = ufile; + pbundle->bundle.context = NULL; /* only valid if bundle has uobject */ pbundle->radix = &uapi->radix; pbundle->radix_slots = slot; pbundle->radix_slots_len = radix_tree_chunk_size(&attrs_iter); diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c index 5f366838b7ff..70b7d80431a9 100644 --- a/drivers/infiniband/core/uverbs_main.c +++ b/drivers/infiniband/core/uverbs_main.c @@ -695,6 +695,7 @@ static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf, memset(bundle.attr_present, 0, sizeof(bundle.attr_present)); bundle.ufile = file; + bundle.context = NULL; /* only valid if bundle has uobject */ if (!method_elm->is_ex) { size_t in_len = hdr.in_words * 4 - sizeof(hdr); size_t out_len = hdr.out_words * 4; @@ -1135,6 +1136,7 @@ static const struct file_operations uverbs_mmap_fops = { static struct ib_client uverbs_client = { .name = "uverbs", + .no_kverbs_req = true, .add = ib_uverbs_add_one, .remove = ib_uverbs_remove_one }; diff --git a/drivers/infiniband/core/uverbs_std_types.c b/drivers/infiniband/core/uverbs_std_types.c index cbc72312eb41..f224cb727224 100644 --- a/drivers/infiniband/core/uverbs_std_types.c +++ b/drivers/infiniband/core/uverbs_std_types.c @@ -188,7 +188,7 @@ static int uverbs_free_pd(struct ib_uobject *uobject, if (ret) return ret; - ib_dealloc_pd((struct ib_pd *)uobject->object); + ib_dealloc_pd(pd); return 0; } diff --git a/drivers/infiniband/core/uverbs_uapi.c b/drivers/infiniband/core/uverbs_uapi.c index 9ae08e4b78a3..7a987acf0c0b 100644 --- a/drivers/infiniband/core/uverbs_uapi.c +++ b/drivers/infiniband/core/uverbs_uapi.c @@ -188,13 +188,18 @@ static int uapi_merge_obj_tree(struct uverbs_api *uapi, obj_elm->type_attrs = obj->type_attrs; obj_elm->type_class = obj->type_attrs->type_class; /* - * Today drivers are only permitted to use idr_class - * types. They cannot use FD types because we currently have - * no way to revoke the fops pointer after device - * disassociation. + * Today drivers are only permitted to use idr_class and + * fd_class types. We can revoke the IDR types during + * disassociation, and the FD types require the driver to use + * struct file_operations.owner to prevent the driver module + * code from unloading while the file is open. This provides + * enough safety that uverbs_close_fd() will continue to work. + * Drivers using FD are responsible to handle disassociation of + * the device on their own. */ if (WARN_ON(is_driver && - obj->type_attrs->type_class != &uverbs_idr_class)) + obj->type_attrs->type_class != &uverbs_idr_class && + obj->type_attrs->type_class != &uverbs_fd_class)) return -EINVAL; } diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c index ac011836bb54..5a5e83f5f0fc 100644 --- a/drivers/infiniband/core/verbs.c +++ b/drivers/infiniband/core/verbs.c @@ -254,10 +254,11 @@ struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, { struct ib_pd *pd; int mr_access_flags = 0; + int ret; - pd = device->ops.alloc_pd(device, NULL, NULL); - if (IS_ERR(pd)) - return pd; + pd = rdma_zalloc_drv_obj(device, ib_pd); + if (!pd) + return ERR_PTR(-ENOMEM); pd->device = device; pd->uobject = NULL; @@ -265,6 +266,16 @@ struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, atomic_set(&pd->usecnt, 0); pd->flags = flags; + pd->res.type = RDMA_RESTRACK_PD; + rdma_restrack_set_task(&pd->res, caller); + + ret = device->ops.alloc_pd(pd, NULL, NULL); + if (ret) { + kfree(pd); + return ERR_PTR(ret); + } + rdma_restrack_kadd(&pd->res); + if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY) pd->local_dma_lkey = device->local_dma_lkey; else @@ -275,10 +286,6 @@ struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, mr_access_flags |= IB_ACCESS_REMOTE_READ | IB_ACCESS_REMOTE_WRITE; } - pd->res.type = RDMA_RESTRACK_PD; - rdma_restrack_set_task(&pd->res, caller); - rdma_restrack_kadd(&pd->res); - if (mr_access_flags) { struct ib_mr *mr; @@ -329,10 +336,8 @@ void ib_dealloc_pd(struct ib_pd *pd) WARN_ON(atomic_read(&pd->usecnt)); rdma_restrack_del(&pd->res); - /* Making delalloc_pd a void return is a WIP, no driver should return - an error here. */ - ret = pd->device->ops.dealloc_pd(pd); - WARN_ONCE(ret, "Infiniband HW driver failed dealloc_pd"); + pd->device->ops.dealloc_pd(pd); + kfree(pd); } EXPORT_SYMBOL(ib_dealloc_pd); @@ -1106,8 +1111,8 @@ struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, } EXPORT_SYMBOL(ib_open_qp); -static struct ib_qp *ib_create_xrc_qp(struct ib_qp *qp, - struct ib_qp_init_attr *qp_init_attr) +static struct ib_qp *create_xrc_qp(struct ib_qp *qp, + struct ib_qp_init_attr *qp_init_attr) { struct ib_qp *real_qp = qp; @@ -1122,10 +1127,10 @@ static struct ib_qp *ib_create_xrc_qp(struct ib_qp *qp, qp = __ib_open_qp(real_qp, qp_init_attr->event_handler, qp_init_attr->qp_context); - if (!IS_ERR(qp)) - __ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp); - else - real_qp->device->ops.destroy_qp(real_qp); + if (IS_ERR(qp)) + return qp; + + __ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp); return qp; } @@ -1156,10 +1161,8 @@ struct ib_qp *ib_create_qp(struct ib_pd *pd, return qp; ret = ib_create_qp_security(qp, device); - if (ret) { - ib_destroy_qp(qp); - return ERR_PTR(ret); - } + if (ret) + goto err; qp->real_qp = qp; qp->qp_type = qp_init_attr->qp_type; @@ -1172,8 +1175,15 @@ struct ib_qp *ib_create_qp(struct ib_pd *pd, INIT_LIST_HEAD(&qp->sig_mrs); qp->port = 0; - if (qp_init_attr->qp_type == IB_QPT_XRC_TGT) - return ib_create_xrc_qp(qp, qp_init_attr); + if (qp_init_attr->qp_type == IB_QPT_XRC_TGT) { + struct ib_qp *xrc_qp = create_xrc_qp(qp, qp_init_attr); + + if (IS_ERR(xrc_qp)) { + ret = PTR_ERR(xrc_qp); + goto err; + } + return xrc_qp; + } qp->event_handler = qp_init_attr->event_handler; qp->qp_context = qp_init_attr->qp_context; @@ -1200,11 +1210,8 @@ struct ib_qp *ib_create_qp(struct ib_pd *pd, if (qp_init_attr->cap.max_rdma_ctxs) { ret = rdma_rw_init_mrs(qp, qp_init_attr); - if (ret) { - pr_err("failed to init MR pool ret= %d\n", ret); - ib_destroy_qp(qp); - return ERR_PTR(ret); - } + if (ret) + goto err; } /* @@ -1217,6 +1224,11 @@ struct ib_qp *ib_create_qp(struct ib_pd *pd, device->attrs.max_sge_rd); return qp; + +err: + ib_destroy_qp(qp); + return ERR_PTR(ret); + } EXPORT_SYMBOL(ib_create_qp); @@ -1711,10 +1723,7 @@ int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width) if (rdma_port_get_link_layer(dev, port_num) != IB_LINK_LAYER_ETHERNET) return -EINVAL; - if (!dev->ops.get_netdev) - return -EOPNOTSUPP; - - netdev = dev->ops.get_netdev(dev, port_num); + netdev = ib_device_get_netdev(dev, port_num); if (!netdev) return -ENODEV; diff --git a/drivers/infiniband/hw/bnxt_re/Kconfig b/drivers/infiniband/hw/bnxt_re/Kconfig index 19982a4a9bba..d25439c305f7 100644 --- a/drivers/infiniband/hw/bnxt_re/Kconfig +++ b/drivers/infiniband/hw/bnxt_re/Kconfig @@ -1,5 +1,6 @@ config INFINIBAND_BNXT_RE tristate "Broadcom Netxtreme HCA support" + depends on 64BIT depends on ETHERNET && NETDEVICES && PCI && INET && DCB select NET_VENDOR_BROADCOM select BNXT diff --git a/drivers/infiniband/hw/bnxt_re/Makefile b/drivers/infiniband/hw/bnxt_re/Makefile index 6e3bc25cc140..ee9bb1be61ea 100644 --- a/drivers/infiniband/hw/bnxt_re/Makefile +++ b/drivers/infiniband/hw/bnxt_re/Makefile @@ -1,6 +1,6 @@ # SPDX-License-Identifier: GPL-2.0 -ccflags-y := -Idrivers/net/ethernet/broadcom/bnxt +ccflags-y := -I $(srctree)/drivers/net/ethernet/broadcom/bnxt obj-$(CONFIG_INFINIBAND_BNXT_RE) += bnxt_re.o bnxt_re-y := main.o ib_verbs.o \ qplib_res.o qplib_rcfw.o \ diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h index 31baa8939a4f..e55a1666c0cd 100644 --- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h +++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h @@ -124,6 +124,7 @@ struct bnxt_re_dev { #define BNXT_RE_FLAG_ISSUE_ROCE_STATS 29 struct net_device *netdev; unsigned int version, major, minor; + struct bnxt_qplib_chip_ctx chip_ctx; struct bnxt_en_dev *en_dev; struct bnxt_msix_entry msix_entries[BNXT_RE_MAX_MSIX]; int num_msix; diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index 1e2515e2eb62..071b2fc38b0b 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -48,6 +48,7 @@ #include <rdma/ib_addr.h> #include <rdma/ib_mad.h> #include <rdma/ib_cache.h> +#include <rdma/uverbs_ioctl.h> #include "bnxt_ulp.h" @@ -563,41 +564,29 @@ fail: } /* Protection Domains */ -int bnxt_re_dealloc_pd(struct ib_pd *ib_pd) +void bnxt_re_dealloc_pd(struct ib_pd *ib_pd) { struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd); struct bnxt_re_dev *rdev = pd->rdev; - int rc; bnxt_re_destroy_fence_mr(pd); - if (pd->qplib_pd.id) { - rc = bnxt_qplib_dealloc_pd(&rdev->qplib_res, - &rdev->qplib_res.pd_tbl, - &pd->qplib_pd); - if (rc) - dev_err(rdev_to_dev(rdev), "Failed to deallocate HW PD"); - } - - kfree(pd); - return 0; + if (pd->qplib_pd.id) + bnxt_qplib_dealloc_pd(&rdev->qplib_res, &rdev->qplib_res.pd_tbl, + &pd->qplib_pd); } -struct ib_pd *bnxt_re_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *ucontext, - struct ib_udata *udata) +int bnxt_re_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *ucontext, + struct ib_udata *udata) { + struct ib_device *ibdev = ibpd->device; struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev); struct bnxt_re_ucontext *ucntx = container_of(ucontext, struct bnxt_re_ucontext, ib_uctx); - struct bnxt_re_pd *pd; + struct bnxt_re_pd *pd = container_of(ibpd, struct bnxt_re_pd, ib_pd); int rc; - pd = kzalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - pd->rdev = rdev; if (bnxt_qplib_alloc_pd(&rdev->qplib_res.pd_tbl, &pd->qplib_pd)) { dev_err(rdev_to_dev(rdev), "Failed to allocate HW PD"); @@ -637,13 +626,12 @@ struct ib_pd *bnxt_re_alloc_pd(struct ib_device *ibdev, if (bnxt_re_create_fence_mr(pd)) dev_warn(rdev_to_dev(rdev), "Failed to create Fence-MR\n"); - return &pd->ib_pd; + return 0; dbfail: - (void)bnxt_qplib_dealloc_pd(&rdev->qplib_res, &rdev->qplib_res.pd_tbl, - &pd->qplib_pd); + bnxt_qplib_dealloc_pd(&rdev->qplib_res, &rdev->qplib_res.pd_tbl, + &pd->qplib_pd); fail: - kfree(pd); - return ERR_PTR(rc); + return rc; } /* Address Handles */ @@ -663,17 +651,36 @@ int bnxt_re_destroy_ah(struct ib_ah *ib_ah, u32 flags) return 0; } +static u8 bnxt_re_stack_to_dev_nw_type(enum rdma_network_type ntype) +{ + u8 nw_type; + + switch (ntype) { + case RDMA_NETWORK_IPV4: + nw_type = CMDQ_CREATE_AH_TYPE_V2IPV4; + break; + case RDMA_NETWORK_IPV6: + nw_type = CMDQ_CREATE_AH_TYPE_V2IPV6; + break; + default: + nw_type = CMDQ_CREATE_AH_TYPE_V1; + break; + } + return nw_type; +} + struct ib_ah *bnxt_re_create_ah(struct ib_pd *ib_pd, struct rdma_ah_attr *ah_attr, u32 flags, struct ib_udata *udata) { struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd); + const struct ib_global_route *grh = rdma_ah_read_grh(ah_attr); struct bnxt_re_dev *rdev = pd->rdev; + const struct ib_gid_attr *sgid_attr; struct bnxt_re_ah *ah; - const struct ib_global_route *grh = rdma_ah_read_grh(ah_attr); - int rc; u8 nw_type; + int rc; if (!(rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH)) { dev_err(rdev_to_dev(rdev), "Failed to alloc AH: GRH not set"); @@ -700,28 +707,11 @@ struct ib_ah *bnxt_re_create_ah(struct ib_pd *ib_pd, ah->qplib_ah.flow_label = grh->flow_label; ah->qplib_ah.hop_limit = grh->hop_limit; ah->qplib_ah.sl = rdma_ah_get_sl(ah_attr); - if (udata && - !rdma_is_multicast_addr((struct in6_addr *) - grh->dgid.raw) && - !rdma_link_local_addr((struct in6_addr *) - grh->dgid.raw)) { - const struct ib_gid_attr *sgid_attr; - sgid_attr = grh->sgid_attr; - /* Get network header type for this GID */ - nw_type = rdma_gid_attr_network_type(sgid_attr); - switch (nw_type) { - case RDMA_NETWORK_IPV4: - ah->qplib_ah.nw_type = CMDQ_CREATE_AH_TYPE_V2IPV4; - break; - case RDMA_NETWORK_IPV6: - ah->qplib_ah.nw_type = CMDQ_CREATE_AH_TYPE_V2IPV6; - break; - default: - ah->qplib_ah.nw_type = CMDQ_CREATE_AH_TYPE_V1; - break; - } - } + sgid_attr = grh->sgid_attr; + /* Get network header type for this GID */ + nw_type = rdma_gid_attr_network_type(sgid_attr); + ah->qplib_ah.nw_type = bnxt_re_stack_to_dev_nw_type(nw_type); memcpy(ah->qplib_ah.dmac, ah_attr->roce.dmac, ETH_ALEN); rc = bnxt_qplib_create_ah(&rdev->qplib_res, &ah->qplib_ah, @@ -733,12 +723,11 @@ struct ib_ah *bnxt_re_create_ah(struct ib_pd *ib_pd, /* Write AVID to shared page. */ if (udata) { - struct ib_ucontext *ib_uctx = ib_pd->uobject->context; - struct bnxt_re_ucontext *uctx; + struct bnxt_re_ucontext *uctx = rdma_udata_to_drv_context( + udata, struct bnxt_re_ucontext, ib_uctx); unsigned long flag; u32 *wrptr; - uctx = container_of(ib_uctx, struct bnxt_re_ucontext, ib_uctx); spin_lock_irqsave(&uctx->sh_lock, flag); wrptr = (u32 *)(uctx->shpg + BNXT_RE_AVID_OFFT); *wrptr = ah->qplib_ah.id; @@ -804,8 +793,8 @@ int bnxt_re_destroy_qp(struct ib_qp *ib_qp) { struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp); struct bnxt_re_dev *rdev = qp->rdev; - int rc; unsigned int flags; + int rc; bnxt_qplib_flush_cqn_wq(&qp->qplib_qp); rc = bnxt_qplib_destroy_qp(&rdev->qplib_res, &qp->qplib_qp); @@ -814,9 +803,12 @@ int bnxt_re_destroy_qp(struct ib_qp *ib_qp) return rc; } - flags = bnxt_re_lock_cqs(qp); - bnxt_qplib_clean_qp(&qp->qplib_qp); - bnxt_re_unlock_cqs(qp, flags); + if (rdma_is_kernel_res(&qp->ib_qp.res)) { + flags = bnxt_re_lock_cqs(qp); + bnxt_qplib_clean_qp(&qp->qplib_qp); + bnxt_re_unlock_cqs(qp, flags); + } + bnxt_qplib_free_qp_res(&rdev->qplib_res, &qp->qplib_qp); if (ib_qp->qp_type == IB_QPT_GSI && rdev->qp1_sqp) { @@ -882,21 +874,23 @@ static int bnxt_re_init_user_qp(struct bnxt_re_dev *rdev, struct bnxt_re_pd *pd, struct bnxt_re_qp_req ureq; struct bnxt_qplib_qp *qplib_qp = &qp->qplib_qp; struct ib_umem *umem; - int bytes = 0; - struct ib_ucontext *context = pd->ib_pd.uobject->context; - struct bnxt_re_ucontext *cntx = container_of(context, - struct bnxt_re_ucontext, - ib_uctx); + int bytes = 0, psn_sz; + struct bnxt_re_ucontext *cntx = rdma_udata_to_drv_context( + udata, struct bnxt_re_ucontext, ib_uctx); + if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) return -EFAULT; bytes = (qplib_qp->sq.max_wqe * BNXT_QPLIB_MAX_SQE_ENTRY_SIZE); /* Consider mapping PSN search memory only for RC QPs. */ - if (qplib_qp->type == CMDQ_CREATE_QP_TYPE_RC) - bytes += (qplib_qp->sq.max_wqe * sizeof(struct sq_psn_search)); + if (qplib_qp->type == CMDQ_CREATE_QP_TYPE_RC) { + psn_sz = bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx) ? + sizeof(struct sq_psn_search_ext) : + sizeof(struct sq_psn_search); + bytes += (qplib_qp->sq.max_wqe * psn_sz); + } bytes = PAGE_ALIGN(bytes); - umem = ib_umem_get(context, ureq.qpsva, bytes, - IB_ACCESS_LOCAL_WRITE, 1); + umem = ib_umem_get(udata, ureq.qpsva, bytes, IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(umem)) return PTR_ERR(umem); @@ -908,7 +902,7 @@ static int bnxt_re_init_user_qp(struct bnxt_re_dev *rdev, struct bnxt_re_pd *pd, if (!qp->qplib_qp.srq) { bytes = (qplib_qp->rq.max_wqe * BNXT_QPLIB_MAX_RQE_ENTRY_SIZE); bytes = PAGE_ALIGN(bytes); - umem = ib_umem_get(context, ureq.qprva, bytes, + umem = ib_umem_get(udata, ureq.qprva, bytes, IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(umem)) goto rqfail; @@ -1066,12 +1060,17 @@ struct ib_qp *bnxt_re_create_qp(struct ib_pd *ib_pd, qp->qplib_qp.pd = &pd->qplib_pd; qp->qplib_qp.qp_handle = (u64)(unsigned long)(&qp->qplib_qp); qp->qplib_qp.type = __from_ib_qp_type(qp_init_attr->qp_type); + + if (qp_init_attr->qp_type == IB_QPT_GSI && + bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx)) + qp->qplib_qp.type = CMDQ_CREATE_QP_TYPE_GSI; if (qp->qplib_qp.type == IB_QPT_MAX) { dev_err(rdev_to_dev(rdev), "QP type 0x%x not supported", qp->qplib_qp.type); rc = -EINVAL; goto fail; } + qp->qplib_qp.max_inline_data = qp_init_attr->cap.max_inline_data; qp->qplib_qp.sig_type = ((qp_init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? true : false); @@ -1132,7 +1131,8 @@ struct ib_qp *bnxt_re_create_qp(struct ib_pd *ib_pd, qp->qplib_qp.mtu = ib_mtu_enum_to_int(iboe_get_mtu(rdev->netdev->mtu)); - if (qp_init_attr->qp_type == IB_QPT_GSI) { + if (qp_init_attr->qp_type == IB_QPT_GSI && + !(bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx))) { /* Allocate 1 more than what's provided */ entries = roundup_pow_of_two(qp_init_attr->cap.max_send_wr + 1); qp->qplib_qp.sq.max_wqe = min_t(u32, entries, @@ -1361,17 +1361,15 @@ static int bnxt_re_init_user_srq(struct bnxt_re_dev *rdev, struct bnxt_qplib_srq *qplib_srq = &srq->qplib_srq; struct ib_umem *umem; int bytes = 0; - struct ib_ucontext *context = pd->ib_pd.uobject->context; - struct bnxt_re_ucontext *cntx = container_of(context, - struct bnxt_re_ucontext, - ib_uctx); + struct bnxt_re_ucontext *cntx = rdma_udata_to_drv_context( + udata, struct bnxt_re_ucontext, ib_uctx); + if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) return -EFAULT; bytes = (qplib_srq->max_wqe * BNXT_QPLIB_MAX_RQE_ENTRY_SIZE); bytes = PAGE_ALIGN(bytes); - umem = ib_umem_get(context, ureq.srqva, bytes, - IB_ACCESS_LOCAL_WRITE, 1); + umem = ib_umem_get(udata, ureq.srqva, bytes, IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(umem)) return PTR_ERR(umem); @@ -1646,6 +1644,9 @@ int bnxt_re_modify_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr, __from_ib_access_flags(qp_attr->qp_access_flags); /* LOCAL_WRITE access must be set to allow RC receive */ qp->qplib_qp.access |= BNXT_QPLIB_ACCESS_LOCAL_WRITE; + /* Temp: Set all params on QP as of now */ + qp->qplib_qp.access |= CMDQ_MODIFY_QP_ACCESS_REMOTE_WRITE; + qp->qplib_qp.access |= CMDQ_MODIFY_QP_ACCESS_REMOTE_READ; } if (qp_attr_mask & IB_QP_PKEY_INDEX) { qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_PKEY; @@ -2093,7 +2094,8 @@ static int bnxt_re_build_qp1_shadow_qp_recv(struct bnxt_re_qp *qp, static int is_ud_qp(struct bnxt_re_qp *qp) { - return qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD; + return (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD || + qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_GSI); } static int bnxt_re_build_send_wqe(struct bnxt_re_qp *qp, @@ -2397,7 +2399,7 @@ int bnxt_re_post_send(struct ib_qp *ib_qp, const struct ib_send_wr *wr, switch (wr->opcode) { case IB_WR_SEND: case IB_WR_SEND_WITH_IMM: - if (ib_qp->qp_type == IB_QPT_GSI) { + if (qp->qplib_qp.type == CMDQ_CREATE_QP1_TYPE_GSI) { rc = bnxt_re_build_qp1_send_v2(qp, wr, &wqe, payload_sz); if (rc) @@ -2527,7 +2529,8 @@ int bnxt_re_post_recv(struct ib_qp *ib_qp, const struct ib_recv_wr *wr, wqe.wr_id = wr->wr_id; wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV; - if (ib_qp->qp_type == IB_QPT_GSI) + if (ib_qp->qp_type == IB_QPT_GSI && + qp->qplib_qp.type != CMDQ_CREATE_QP_TYPE_GSI) rc = bnxt_re_build_qp1_shadow_qp_recv(qp, wr, &wqe, payload_sz); if (!rc) @@ -2622,7 +2625,7 @@ struct ib_cq *bnxt_re_create_cq(struct ib_device *ibdev, goto fail; } - cq->umem = ib_umem_get(context, req.cq_va, + cq->umem = ib_umem_get(udata, req.cq_va, entries * sizeof(struct cq_base), IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(cq->umem)) { @@ -3122,19 +3125,33 @@ static void bnxt_re_process_res_shadow_qp_wc(struct bnxt_re_qp *qp, } } -static void bnxt_re_process_res_ud_wc(struct ib_wc *wc, +static void bnxt_re_process_res_ud_wc(struct bnxt_re_qp *qp, + struct ib_wc *wc, struct bnxt_qplib_cqe *cqe) { + u8 nw_type; + wc->opcode = IB_WC_RECV; wc->status = __rc_to_ib_wc_status(cqe->status); - if (cqe->flags & CQ_RES_RC_FLAGS_IMM) + if (cqe->flags & CQ_RES_UD_FLAGS_IMM) wc->wc_flags |= IB_WC_WITH_IMM; - if (cqe->flags & CQ_RES_RC_FLAGS_INV) - wc->wc_flags |= IB_WC_WITH_INVALIDATE; - if ((cqe->flags & (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM)) == - (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM)) - wc->opcode = IB_WC_RECV_RDMA_WITH_IMM; + /* report only on GSI QP for Thor */ + if (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_GSI) { + wc->wc_flags |= IB_WC_GRH; + memcpy(wc->smac, cqe->smac, ETH_ALEN); + wc->wc_flags |= IB_WC_WITH_SMAC; + if (cqe->flags & CQ_RES_UD_FLAGS_META_FORMAT_VLAN) { + wc->vlan_id = (cqe->cfa_meta & 0xFFF); + if (wc->vlan_id < 0x1000) + wc->wc_flags |= IB_WC_WITH_VLAN; + } + nw_type = (cqe->flags & CQ_RES_UD_FLAGS_ROCE_IP_VER_MASK) >> + CQ_RES_UD_FLAGS_ROCE_IP_VER_SFT; + wc->network_hdr_type = bnxt_re_to_ib_nw_type(nw_type); + wc->wc_flags |= IB_WC_WITH_NETWORK_HDR_TYPE; + } + } static int send_phantom_wqe(struct bnxt_re_qp *qp) @@ -3226,7 +3243,7 @@ int bnxt_re_poll_cq(struct ib_cq *ib_cq, int num_entries, struct ib_wc *wc) switch (cqe->opcode) { case CQ_BASE_CQE_TYPE_REQ: - if (qp->qplib_qp.id == + if (qp->rdev->qp1_sqp && qp->qplib_qp.id == qp->rdev->qp1_sqp->qplib_qp.id) { /* Handle this completion with * the stored completion @@ -3261,7 +3278,7 @@ int bnxt_re_poll_cq(struct ib_cq *ib_cq, int num_entries, struct ib_wc *wc) bnxt_re_process_res_rc_wc(wc, cqe); break; case CQ_BASE_CQE_TYPE_RES_UD: - if (qp->qplib_qp.id == + if (qp->rdev->qp1_sqp && qp->qplib_qp.id == qp->rdev->qp1_sqp->qplib_qp.id) { /* Handle this completion with * the stored completion @@ -3274,7 +3291,7 @@ int bnxt_re_poll_cq(struct ib_cq *ib_cq, int num_entries, struct ib_wc *wc) break; } } - bnxt_re_process_res_ud_wc(wc, cqe); + bnxt_re_process_res_ud_wc(qp, wc, cqe); break; default: dev_err(rdev_to_dev(cq->rdev), @@ -3301,10 +3318,10 @@ int bnxt_re_req_notify_cq(struct ib_cq *ib_cq, spin_lock_irqsave(&cq->cq_lock, flags); /* Trigger on the very next completion */ if (ib_cqn_flags & IB_CQ_NEXT_COMP) - type = DBR_DBR_TYPE_CQ_ARMALL; + type = DBC_DBC_TYPE_CQ_ARMALL; /* Trigger on the next solicited completion */ else if (ib_cqn_flags & IB_CQ_SOLICITED) - type = DBR_DBR_TYPE_CQ_ARMSE; + type = DBC_DBC_TYPE_CQ_ARMSE; /* Poll to see if there are missed events */ if ((ib_cqn_flags & IB_CQ_REPORT_MISSED_EVENTS) && @@ -3537,19 +3554,14 @@ static int fill_umem_pbl_tbl(struct ib_umem *umem, u64 *pbl_tbl_orig, u64 *pbl_tbl = pbl_tbl_orig; u64 paddr; u64 page_mask = (1ULL << page_shift) - 1; - int i, pages; - struct scatterlist *sg; - int entry; - - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { - pages = sg_dma_len(sg) >> PAGE_SHIFT; - for (i = 0; i < pages; i++) { - paddr = sg_dma_address(sg) + (i << PAGE_SHIFT); - if (pbl_tbl == pbl_tbl_orig) - *pbl_tbl++ = paddr & ~page_mask; - else if ((paddr & page_mask) == 0) - *pbl_tbl++ = paddr; - } + struct sg_dma_page_iter sg_iter; + + for_each_sg_dma_page (umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + paddr = sg_page_iter_dma_address(&sg_iter); + if (pbl_tbl == pbl_tbl_orig) + *pbl_tbl++ = paddr & ~page_mask; + else if ((paddr & page_mask) == 0) + *pbl_tbl++ = paddr; } return pbl_tbl - pbl_tbl_orig; } @@ -3589,8 +3601,7 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length, /* The fixed portion of the rkey is the same as the lkey */ mr->ib_mr.rkey = mr->qplib_mr.rkey; - umem = ib_umem_get(ib_pd->uobject->context, start, length, - mr_access_flags, 0); + umem = ib_umem_get(udata, start, length, mr_access_flags, 0); if (IS_ERR(umem)) { dev_err(rdev_to_dev(rdev), "Failed to get umem"); rc = -EFAULT; @@ -3613,7 +3624,7 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length, goto free_umem; } - page_shift = umem->page_shift; + page_shift = PAGE_SHIFT; if (!bnxt_re_page_size_ok(page_shift)) { dev_err(rdev_to_dev(rdev), "umem page size unsupported!"); @@ -3660,13 +3671,15 @@ free_mr: return ERR_PTR(rc); } -struct ib_ucontext *bnxt_re_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +int bnxt_re_alloc_ucontext(struct ib_ucontext *ctx, struct ib_udata *udata) { + struct ib_device *ibdev = ctx->device; + struct bnxt_re_ucontext *uctx = + container_of(ctx, struct bnxt_re_ucontext, ib_uctx); struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev); - struct bnxt_re_uctx_resp resp; - struct bnxt_re_ucontext *uctx; struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr; + struct bnxt_re_uctx_resp resp; + u32 chip_met_rev_num = 0; int rc; dev_dbg(rdev_to_dev(rdev), "ABI version requested %d", @@ -3675,13 +3688,9 @@ struct ib_ucontext *bnxt_re_alloc_ucontext(struct ib_device *ibdev, if (ibdev->uverbs_abi_ver != BNXT_RE_ABI_VERSION) { dev_dbg(rdev_to_dev(rdev), " is different from the device %d ", BNXT_RE_ABI_VERSION); - return ERR_PTR(-EPERM); + return -EPERM; } - uctx = kzalloc(sizeof(*uctx), GFP_KERNEL); - if (!uctx) - return ERR_PTR(-ENOMEM); - uctx->rdev = rdev; uctx->shpg = (void *)__get_free_page(GFP_KERNEL); @@ -3691,37 +3700,45 @@ struct ib_ucontext *bnxt_re_alloc_ucontext(struct ib_device *ibdev, } spin_lock_init(&uctx->sh_lock); - resp.dev_id = rdev->en_dev->pdev->devfn; /*Temp, Use idr_alloc instead*/ + resp.comp_mask = BNXT_RE_UCNTX_CMASK_HAVE_CCTX; + chip_met_rev_num = rdev->chip_ctx.chip_num; + chip_met_rev_num |= ((u32)rdev->chip_ctx.chip_rev & 0xFF) << + BNXT_RE_CHIP_ID0_CHIP_REV_SFT; + chip_met_rev_num |= ((u32)rdev->chip_ctx.chip_metal & 0xFF) << + BNXT_RE_CHIP_ID0_CHIP_MET_SFT; + resp.chip_id0 = chip_met_rev_num; + /* Future extension of chip info */ + resp.chip_id1 = 0; + /*Temp, Use idr_alloc instead */ + resp.dev_id = rdev->en_dev->pdev->devfn; resp.max_qp = rdev->qplib_ctx.qpc_count; resp.pg_size = PAGE_SIZE; resp.cqe_sz = sizeof(struct cq_base); resp.max_cqd = dev_attr->max_cq_wqes; resp.rsvd = 0; - rc = ib_copy_to_udata(udata, &resp, sizeof(resp)); + rc = ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp))); if (rc) { dev_err(rdev_to_dev(rdev), "Failed to copy user context"); rc = -EFAULT; goto cfail; } - return &uctx->ib_uctx; + return 0; cfail: free_page((unsigned long)uctx->shpg); uctx->shpg = NULL; fail: - kfree(uctx); - return ERR_PTR(rc); + return rc; } -int bnxt_re_dealloc_ucontext(struct ib_ucontext *ib_uctx) +void bnxt_re_dealloc_ucontext(struct ib_ucontext *ib_uctx) { struct bnxt_re_ucontext *uctx = container_of(ib_uctx, struct bnxt_re_ucontext, ib_uctx); struct bnxt_re_dev *rdev = uctx->rdev; - int rc = 0; if (uctx->shpg) free_page((unsigned long)uctx->shpg); @@ -3730,17 +3747,10 @@ int bnxt_re_dealloc_ucontext(struct ib_ucontext *ib_uctx) /* Free DPI only if this is the first PD allocated by the * application and mark the context dpi as NULL */ - rc = bnxt_qplib_dealloc_dpi(&rdev->qplib_res, - &rdev->qplib_res.dpi_tbl, - &uctx->dpi); - if (rc) - dev_err(rdev_to_dev(rdev), "Deallocate HW DPI failed!"); - /* Don't fail, continue*/ + bnxt_qplib_dealloc_dpi(&rdev->qplib_res, + &rdev->qplib_res.dpi_tbl, &uctx->dpi); uctx->dpi.dbr = NULL; } - - kfree(uctx); - return 0; } /* Helper function to mmap the virtual memory from user app */ diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.h b/drivers/infiniband/hw/bnxt_re/ib_verbs.h index c4af72604b4f..e45465ed4eee 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.h +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.h @@ -56,8 +56,8 @@ struct bnxt_re_fence_data { }; struct bnxt_re_pd { + struct ib_pd ib_pd; struct bnxt_re_dev *rdev; - struct ib_pd ib_pd; struct bnxt_qplib_pd qplib_pd; struct bnxt_re_fence_data fence; }; @@ -135,8 +135,8 @@ struct bnxt_re_mw { }; struct bnxt_re_ucontext { + struct ib_ucontext ib_uctx; struct bnxt_re_dev *rdev; - struct ib_ucontext ib_uctx; struct bnxt_qplib_dpi dpi; void *shpg; spinlock_t sh_lock; /* protect shpg */ @@ -163,10 +163,9 @@ int bnxt_re_query_gid(struct ib_device *ibdev, u8 port_num, int index, union ib_gid *gid); enum rdma_link_layer bnxt_re_get_link_layer(struct ib_device *ibdev, u8 port_num); -struct ib_pd *bnxt_re_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata); -int bnxt_re_dealloc_pd(struct ib_pd *pd); +int bnxt_re_alloc_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata); +void bnxt_re_dealloc_pd(struct ib_pd *pd); struct ib_ah *bnxt_re_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, u32 flags, @@ -216,9 +215,8 @@ int bnxt_re_dealloc_mw(struct ib_mw *mw); struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int mr_access_flags, struct ib_udata *udata); -struct ib_ucontext *bnxt_re_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata); -int bnxt_re_dealloc_ucontext(struct ib_ucontext *context); +int bnxt_re_alloc_ucontext(struct ib_ucontext *ctx, struct ib_udata *udata); +void bnxt_re_dealloc_ucontext(struct ib_ucontext *context); int bnxt_re_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); unsigned long bnxt_re_lock_cqs(struct bnxt_re_qp *qp); diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c index e7a997f2a537..2bd24ac45ee4 100644 --- a/drivers/infiniband/hw/bnxt_re/main.c +++ b/drivers/infiniband/hw/bnxt_re/main.c @@ -80,6 +80,29 @@ static DEFINE_MUTEX(bnxt_re_dev_lock); static struct workqueue_struct *bnxt_re_wq; static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev); +static void bnxt_re_destroy_chip_ctx(struct bnxt_re_dev *rdev) +{ + rdev->rcfw.res = NULL; + rdev->qplib_res.cctx = NULL; +} + +static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev) +{ + struct bnxt_en_dev *en_dev; + struct bnxt *bp; + + en_dev = rdev->en_dev; + bp = netdev_priv(en_dev->net); + + rdev->chip_ctx.chip_num = bp->chip_num; + /* rest members to follow eventually */ + + rdev->qplib_res.cctx = &rdev->chip_ctx; + rdev->rcfw.res = &rdev->qplib_res; + + return 0; +} + /* SR-IOV helper functions */ static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev) @@ -278,6 +301,7 @@ static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev) rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP, &bnxt_re_ulp_ops, rdev); + rdev->qplib_res.pdev = rdev->en_dev->pdev; return rc; } @@ -345,7 +369,8 @@ static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg, fw_msg->timeout = timeout; } -static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id) +static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, + u16 fw_ring_id, int type) { struct bnxt_en_dev *en_dev = rdev->en_dev; struct hwrm_ring_free_input req = {0}; @@ -359,7 +384,7 @@ static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id) memset(&fw_msg, 0, sizeof(fw_msg)); bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1); - req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL; + req.ring_type = type; req.ring_id = cpu_to_le16(fw_ring_id); bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp, sizeof(resp), DFLT_HWRM_CMD_TIMEOUT); @@ -396,7 +421,7 @@ static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr, /* Association of ring index with doorbell index and MSIX number */ req.logical_id = cpu_to_le16(map_index); req.length = cpu_to_le32(ring_mask + 1); - req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL; + req.ring_type = type; req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX; bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp, sizeof(resp), DFLT_HWRM_CMD_TIMEOUT); @@ -538,7 +563,8 @@ static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev) static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { - struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev); + struct bnxt_re_dev *rdev = + rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev); return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor); } @@ -547,7 +573,8 @@ static DEVICE_ATTR_RO(hw_rev); static ssize_t hca_type_show(struct device *device, struct device_attribute *attr, char *buf) { - struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev); + struct bnxt_re_dev *rdev = + rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev); return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc); } @@ -610,6 +637,8 @@ static const struct ib_device_ops bnxt_re_dev_ops = { .query_srq = bnxt_re_query_srq, .reg_user_mr = bnxt_re_reg_user_mr, .req_notify_cq = bnxt_re_req_notify_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx), }; static int bnxt_re_register_ib(struct bnxt_re_dev *rdev) @@ -662,7 +691,7 @@ static int bnxt_re_register_ib(struct bnxt_re_dev *rdev) rdma_set_device_sysfs_group(ibdev, &bnxt_re_dev_attr_group); ibdev->driver_id = RDMA_DRIVER_BNXT_RE; ib_set_device_ops(ibdev, &bnxt_re_dev_ops); - return ib_register_device(ibdev, "bnxt_re%d", NULL); + return ib_register_device(ibdev, "bnxt_re%d"); } static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev) @@ -686,7 +715,7 @@ static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev, struct bnxt_re_dev *rdev; /* Allocate bnxt_re_dev instance here */ - rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev)); + rdev = ib_alloc_device(bnxt_re_dev, ibdev); if (!rdev) { dev_err(NULL, "%s: bnxt_re_dev allocation failure!", ROCE_DRV_MODULE_NAME); @@ -858,6 +887,12 @@ static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq, return 0; } +static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx) +{ + return bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx) ? + 0x10000 : rdev->msix_entries[indx].db_offset; +} + static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev) { int i; @@ -871,18 +906,18 @@ static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev) static int bnxt_re_init_res(struct bnxt_re_dev *rdev) { - int rc = 0, i; int num_vec_enabled = 0; + int rc = 0, i; + u32 db_offt; bnxt_qplib_init_res(&rdev->qplib_res); for (i = 1; i < rdev->num_msix ; i++) { + db_offt = bnxt_re_get_nqdb_offset(rdev, i); rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1], i - 1, rdev->msix_entries[i].vector, - rdev->msix_entries[i].db_offset, - &bnxt_re_cqn_handler, + db_offt, &bnxt_re_cqn_handler, &bnxt_re_srqn_handler); - if (rc) { dev_err(rdev_to_dev(rdev), "Failed to enable NQ with rc = 0x%x", rc); @@ -894,16 +929,18 @@ static int bnxt_re_init_res(struct bnxt_re_dev *rdev) fail: for (i = num_vec_enabled; i >= 0; i--) bnxt_qplib_disable_nq(&rdev->nq[i]); - return rc; } static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev) { + u8 type; int i; for (i = 0; i < rdev->num_msix - 1; i++) { - bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id); + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx); + bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type); + rdev->nq[i].res = NULL; bnxt_qplib_free_nq(&rdev->nq[i]); } } @@ -925,8 +962,11 @@ static void bnxt_re_free_res(struct bnxt_re_dev *rdev) static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev) { - int rc = 0, i; int num_vec_created = 0; + dma_addr_t *pg_map; + int rc = 0, i; + int pages; + u8 type; /* Configure and allocate resources for qplib */ rdev->qplib_res.rcfw = &rdev->rcfw; @@ -947,6 +987,7 @@ static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev) goto dealloc_res; for (i = 0; i < rdev->num_msix - 1; i++) { + rdev->nq[i].res = &rdev->qplib_res; rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT + BNXT_RE_MAX_SRQC_COUNT + 2; rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]); @@ -955,13 +996,13 @@ static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev) i, rc); goto free_nq; } - rc = bnxt_re_net_ring_alloc - (rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr, - rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count, - HWRM_RING_ALLOC_CMPL, - BNXT_QPLIB_NQE_MAX_CNT - 1, - rdev->msix_entries[i + 1].ring_idx, - &rdev->nq[i].ring_id); + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx); + pg_map = rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr; + pages = rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count; + rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type, + BNXT_QPLIB_NQE_MAX_CNT - 1, + rdev->msix_entries[i + 1].ring_idx, + &rdev->nq[i].ring_id); if (rc) { dev_err(rdev_to_dev(rdev), "Failed to allocate NQ fw id with rc = 0x%x", @@ -974,7 +1015,8 @@ static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev) return 0; free_nq: for (i = num_vec_created; i >= 0; i--) { - bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id); + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx); + bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type); bnxt_qplib_free_nq(&rdev->nq[i]); } bnxt_qplib_dealloc_dpi(&rdev->qplib_res, @@ -1228,6 +1270,7 @@ static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev) static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev) { + u8 type; int rc; if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) { @@ -1251,7 +1294,8 @@ static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev) bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id); bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx); bnxt_qplib_disable_rcfw_channel(&rdev->rcfw); - bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id); + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx); + bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type); bnxt_qplib_free_rcfw_channel(&rdev->rcfw); } if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) { @@ -1260,6 +1304,8 @@ static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev) dev_warn(rdev_to_dev(rdev), "Failed to free MSI-X vectors: %#x", rc); } + + bnxt_re_destroy_chip_ctx(rdev); if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) { rc = bnxt_re_unregister_netdev(rdev); if (rc) @@ -1280,9 +1326,12 @@ static void bnxt_re_worker(struct work_struct *work) static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev) { - int rc; - + dma_addr_t *pg_map; + u32 db_offt, ridx; + int pages, vid; bool locked; + u8 type; + int rc; /* Acquire rtnl lock through out this function */ rtnl_lock(); @@ -1297,6 +1346,12 @@ static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev) } set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags); + rc = bnxt_re_setup_chip_ctx(rdev); + if (rc) { + dev_err(rdev_to_dev(rdev), "Failed to get chip context\n"); + return -EINVAL; + } + /* Check whether VF or PF */ bnxt_re_get_sriov_func_type(rdev); @@ -1320,21 +1375,22 @@ static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev) pr_err("Failed to allocate RCFW Channel: %#x\n", rc); goto fail; } - rc = bnxt_re_net_ring_alloc - (rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr, - rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count, - HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1, - rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx, - &rdev->rcfw.creq_ring_id); + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx); + pg_map = rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr; + pages = rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count; + ridx = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx; + rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type, + BNXT_QPLIB_CREQE_MAX_CNT - 1, + ridx, &rdev->rcfw.creq_ring_id); if (rc) { pr_err("Failed to allocate CREQ: %#x\n", rc); goto free_rcfw; } - rc = bnxt_qplib_enable_rcfw_channel - (rdev->en_dev->pdev, &rdev->rcfw, - rdev->msix_entries[BNXT_RE_AEQ_IDX].vector, - rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset, - rdev->is_virtfn, &bnxt_re_aeq_handler); + db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX); + vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector; + rc = bnxt_qplib_enable_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw, + vid, db_offt, rdev->is_virtfn, + &bnxt_re_aeq_handler); if (rc) { pr_err("Failed to enable RCFW channel: %#x\n", rc); goto free_ring; @@ -1347,7 +1403,8 @@ static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev) if (!rdev->is_virtfn) bnxt_re_set_resource_limits(rdev); - rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0); + rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0, + bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx)); if (rc) { pr_err("Failed to allocate QPLIB context: %#x\n", rc); goto disable_rcfw; @@ -1418,7 +1475,8 @@ free_ctx: disable_rcfw: bnxt_qplib_disable_rcfw_channel(&rdev->rcfw); free_ring: - bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id); + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx); + bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type); free_rcfw: bnxt_qplib_free_rcfw_channel(&rdev->rcfw); fail: diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c index b98b054148cd..71c34d5b0ac0 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c @@ -44,6 +44,7 @@ #include <linux/slab.h> #include <linux/pci.h> #include <linux/prefetch.h> +#include <linux/if_ether.h> #include "roce_hsi.h" @@ -244,6 +245,7 @@ static void bnxt_qplib_service_nq(unsigned long data) u16 type; int budget = nq->budget; uintptr_t q_handle; + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx); /* Service the NQ until empty */ raw_cons = hwq->cons; @@ -290,7 +292,7 @@ static void bnxt_qplib_service_nq(unsigned long data) q_handle |= (u64)le32_to_cpu(nqsrqe->srq_handle_high) << 32; bnxt_qplib_arm_srq((struct bnxt_qplib_srq *)q_handle, - DBR_DBR_TYPE_SRQ_ARMENA); + DBC_DBC_TYPE_SRQ_ARMENA); if (!nq->srqn_handler(nq, (struct bnxt_qplib_srq *)q_handle, nqsrqe->event)) @@ -312,7 +314,9 @@ static void bnxt_qplib_service_nq(unsigned long data) } if (hwq->cons != raw_cons) { hwq->cons = raw_cons; - NQ_DB_REARM(nq->bar_reg_iomem, hwq->cons, hwq->max_elements); + bnxt_qplib_ring_nq_db_rearm(nq->bar_reg_iomem, hwq->cons, + hwq->max_elements, nq->ring_id, + gen_p5); } } @@ -336,9 +340,11 @@ static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance) void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill) { + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx); tasklet_disable(&nq->worker); /* Mask h/w interrupt */ - NQ_DB(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements); + bnxt_qplib_ring_nq_db(nq->bar_reg_iomem, nq->hwq.cons, + nq->hwq.max_elements, nq->ring_id, gen_p5); /* Sync with last running IRQ handler */ synchronize_irq(nq->vector); if (kill) @@ -373,6 +379,7 @@ void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq) int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx, int msix_vector, bool need_init) { + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx); int rc; if (nq->requested) @@ -399,7 +406,8 @@ int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx, nq->vector, nq_indx); } nq->requested = true; - NQ_DB_REARM(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements); + bnxt_qplib_ring_nq_db_rearm(nq->bar_reg_iomem, nq->hwq.cons, + nq->hwq.max_elements, nq->ring_id, gen_p5); return rc; } @@ -433,7 +441,8 @@ int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq, rc = -ENOMEM; goto fail; } - nq->bar_reg_iomem = ioremap_nocache(nq_base + nq->bar_reg_off, 4); + /* Unconditionally map 8 bytes to support 57500 series */ + nq->bar_reg_iomem = ioremap_nocache(nq_base + nq->bar_reg_off, 8); if (!nq->bar_reg_iomem) { rc = -ENOMEM; goto fail; @@ -462,15 +471,17 @@ void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq) int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq) { + u8 hwq_type; + nq->pdev = pdev; if (!nq->hwq.max_elements || nq->hwq.max_elements > BNXT_QPLIB_NQE_MAX_CNT) nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT; - + hwq_type = bnxt_qplib_get_hwq_type(nq->res); if (bnxt_qplib_alloc_init_hwq(nq->pdev, &nq->hwq, NULL, 0, &nq->hwq.max_elements, BNXT_QPLIB_MAX_NQE_ENTRY_SIZE, 0, - PAGE_SIZE, HWQ_TYPE_L2_CMPL)) + PAGE_SIZE, hwq_type)) return -ENOMEM; nq->budget = 8; @@ -481,21 +492,19 @@ int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq) static void bnxt_qplib_arm_srq(struct bnxt_qplib_srq *srq, u32 arm_type) { struct bnxt_qplib_hwq *srq_hwq = &srq->hwq; - struct dbr_dbr db_msg = { 0 }; void __iomem *db; - u32 sw_prod = 0; + u32 sw_prod; + u64 val = 0; /* Ring DB */ - sw_prod = (arm_type == DBR_DBR_TYPE_SRQ_ARM) ? srq->threshold : - HWQ_CMP(srq_hwq->prod, srq_hwq); - db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) & - DBR_DBR_INDEX_MASK); - db_msg.type_xid = cpu_to_le32(((srq->id << DBR_DBR_XID_SFT) & - DBR_DBR_XID_MASK) | arm_type); - db = (arm_type == DBR_DBR_TYPE_SRQ_ARMENA) ? - srq->dbr_base : srq->dpi->dbr; - wmb(); /* barrier before db ring */ - __iowrite64_copy(db, &db_msg, sizeof(db_msg) / sizeof(u64)); + sw_prod = (arm_type == DBC_DBC_TYPE_SRQ_ARM) ? + srq->threshold : HWQ_CMP(srq_hwq->prod, srq_hwq); + db = (arm_type == DBC_DBC_TYPE_SRQ_ARMENA) ? srq->dbr_base : + srq->dpi->dbr; + val = ((srq->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) | arm_type; + val <<= 32; + val |= (sw_prod << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK; + writeq(val, db); } int bnxt_qplib_destroy_srq(struct bnxt_qplib_res *res, @@ -590,7 +599,7 @@ int bnxt_qplib_create_srq(struct bnxt_qplib_res *res, srq->id = le32_to_cpu(resp.xid); srq->dbr_base = res->dpi_tbl.dbr_bar_reg_iomem; if (srq->threshold) - bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ_ARMENA); + bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ_ARMENA); srq->arm_req = false; return 0; @@ -614,7 +623,7 @@ int bnxt_qplib_modify_srq(struct bnxt_qplib_res *res, srq_hwq->max_elements - sw_cons + sw_prod; if (count > srq->threshold) { srq->arm_req = false; - bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ_ARM); + bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ_ARM); } else { /* Deferred arming */ srq->arm_req = true; @@ -702,10 +711,10 @@ int bnxt_qplib_post_srq_recv(struct bnxt_qplib_srq *srq, srq_hwq->max_elements - sw_cons + sw_prod; spin_unlock(&srq_hwq->lock); /* Ring DB */ - bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ); + bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ); if (srq->arm_req == true && count > srq->threshold) { srq->arm_req = false; - bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ_ARM); + bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ_ARM); } done: return rc; @@ -853,18 +862,19 @@ exit: int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp) { struct bnxt_qplib_rcfw *rcfw = res->rcfw; - struct sq_send *hw_sq_send_hdr, **hw_sq_send_ptr; - struct cmdq_create_qp req; - struct creq_create_qp_resp resp; - struct bnxt_qplib_pbl *pbl; - struct sq_psn_search **psn_search_ptr; unsigned long int psn_search, poff = 0; + struct sq_psn_search **psn_search_ptr; struct bnxt_qplib_q *sq = &qp->sq; struct bnxt_qplib_q *rq = &qp->rq; + int i, rc, req_size, psn_sz = 0; + struct sq_send **hw_sq_send_ptr; + struct creq_create_qp_resp resp; struct bnxt_qplib_hwq *xrrq; - int i, rc, req_size, psn_sz; u16 cmd_flags = 0, max_ssge; - u32 sw_prod, qp_flags = 0; + struct cmdq_create_qp req; + struct bnxt_qplib_pbl *pbl; + u32 qp_flags = 0; + u16 max_rsge; RCFW_CMD_PREP(req, CREATE_QP, cmd_flags); @@ -874,8 +884,11 @@ int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp) req.qp_handle = cpu_to_le64(qp->qp_handle); /* SQ */ - psn_sz = (qp->type == CMDQ_CREATE_QP_TYPE_RC) ? - sizeof(struct sq_psn_search) : 0; + if (qp->type == CMDQ_CREATE_QP_TYPE_RC) { + psn_sz = bnxt_qplib_is_chip_gen_p5(res->cctx) ? + sizeof(struct sq_psn_search_ext) : + sizeof(struct sq_psn_search); + } sq->hwq.max_elements = sq->max_wqe; rc = bnxt_qplib_alloc_init_hwq(res->pdev, &sq->hwq, sq->sglist, sq->nmap, &sq->hwq.max_elements, @@ -905,10 +918,16 @@ int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp) poff = (psn_search & ~PAGE_MASK) / BNXT_QPLIB_MAX_PSNE_ENTRY_SIZE; } - for (i = 0; i < sq->hwq.max_elements; i++) + for (i = 0; i < sq->hwq.max_elements; i++) { sq->swq[i].psn_search = &psn_search_ptr[get_psne_pg(i + poff)] [get_psne_idx(i + poff)]; + /*psns_ext will be used only for P5 chips. */ + sq->swq[i].psn_ext = + (struct sq_psn_search_ext *) + &psn_search_ptr[get_psne_pg(i + poff)] + [get_psne_idx(i + poff)]; + } } pbl = &sq->hwq.pbl[PBL_LVL_0]; req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]); @@ -929,14 +948,6 @@ int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp) CMDQ_CREATE_QP_SQ_PG_SIZE_PG_1G : CMDQ_CREATE_QP_SQ_PG_SIZE_PG_4K); - /* initialize all SQ WQEs to LOCAL_INVALID (sq prep for hw fetch) */ - hw_sq_send_ptr = (struct sq_send **)sq->hwq.pbl_ptr; - for (sw_prod = 0; sw_prod < sq->hwq.max_elements; sw_prod++) { - hw_sq_send_hdr = &hw_sq_send_ptr[get_sqe_pg(sw_prod)] - [get_sqe_idx(sw_prod)]; - hw_sq_send_hdr->wqe_type = SQ_BASE_WQE_TYPE_LOCAL_INVALID; - } - if (qp->scq) req.scq_cid = cpu_to_le32(qp->scq->id); @@ -1007,8 +1018,9 @@ int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp) req.sq_fwo_sq_sge = cpu_to_le16( ((max_ssge & CMDQ_CREATE_QP_SQ_SGE_MASK) << CMDQ_CREATE_QP_SQ_SGE_SFT) | 0); + max_rsge = bnxt_qplib_is_chip_gen_p5(res->cctx) ? 6 : rq->max_sge; req.rq_fwo_rq_sge = cpu_to_le16( - ((rq->max_sge & CMDQ_CREATE_QP_RQ_SGE_MASK) + ((max_rsge & CMDQ_CREATE_QP_RQ_SGE_MASK) << CMDQ_CREATE_QP_RQ_SGE_SFT) | 0); /* ORRQ and IRRQ */ if (psn_sz) { @@ -1053,6 +1065,7 @@ int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp) qp->id = le32_to_cpu(resp.xid); qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET; + qp->cctx = res->cctx; INIT_LIST_HEAD(&qp->sq_flush); INIT_LIST_HEAD(&qp->rq_flush); rcfw->qp_tbl[qp->id].qp_id = qp->id; @@ -1494,19 +1507,16 @@ void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp, void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp) { struct bnxt_qplib_q *sq = &qp->sq; - struct dbr_dbr db_msg = { 0 }; u32 sw_prod; + u64 val = 0; + val = (((qp->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) | + DBC_DBC_TYPE_SQ); + val <<= 32; sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq); - - db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) & - DBR_DBR_INDEX_MASK); - db_msg.type_xid = - cpu_to_le32(((qp->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) | - DBR_DBR_TYPE_SQ); + val |= (sw_prod << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK; /* Flush all the WQE writes to HW */ - wmb(); - __iowrite64_copy(qp->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64)); + writeq(val, qp->dpi->dbr); } int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp, @@ -1617,7 +1627,8 @@ int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp, ((offsetof(typeof(*sqe), data) + 15) >> 4); sqe->inv_key_or_imm_data = cpu_to_le32( wqe->send.inv_key); - if (qp->type == CMDQ_CREATE_QP_TYPE_UD) { + if (qp->type == CMDQ_CREATE_QP_TYPE_UD || + qp->type == CMDQ_CREATE_QP_TYPE_GSI) { sqe->q_key = cpu_to_le32(wqe->send.q_key); sqe->dst_qp = cpu_to_le32( wqe->send.dst_qp & SQ_SEND_DST_QP_MASK); @@ -1741,14 +1752,26 @@ int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp, } swq->next_psn = sq->psn & BTH_PSN_MASK; if (swq->psn_search) { - swq->psn_search->opcode_start_psn = cpu_to_le32( - ((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) & - SQ_PSN_SEARCH_START_PSN_MASK) | - ((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) & - SQ_PSN_SEARCH_OPCODE_MASK)); - swq->psn_search->flags_next_psn = cpu_to_le32( - ((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) & - SQ_PSN_SEARCH_NEXT_PSN_MASK)); + u32 opcd_spsn; + u32 flg_npsn; + + opcd_spsn = ((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) & + SQ_PSN_SEARCH_START_PSN_MASK); + opcd_spsn |= ((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) & + SQ_PSN_SEARCH_OPCODE_MASK); + flg_npsn = ((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) & + SQ_PSN_SEARCH_NEXT_PSN_MASK); + if (bnxt_qplib_is_chip_gen_p5(qp->cctx)) { + swq->psn_ext->opcode_start_psn = + cpu_to_le32(opcd_spsn); + swq->psn_ext->flags_next_psn = + cpu_to_le32(flg_npsn); + } else { + swq->psn_search->opcode_start_psn = + cpu_to_le32(opcd_spsn); + swq->psn_search->flags_next_psn = + cpu_to_le32(flg_npsn); + } } queue_err: if (sch_handler) { @@ -1785,19 +1808,16 @@ done: void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp) { struct bnxt_qplib_q *rq = &qp->rq; - struct dbr_dbr db_msg = { 0 }; u32 sw_prod; + u64 val = 0; + val = (((qp->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) | + DBC_DBC_TYPE_RQ); + val <<= 32; sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq); - db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) & - DBR_DBR_INDEX_MASK); - db_msg.type_xid = - cpu_to_le32(((qp->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) | - DBR_DBR_TYPE_RQ); - + val |= (sw_prod << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK; /* Flush the writes to HW Rx WQE before the ringing Rx DB */ - wmb(); - __iowrite64_copy(qp->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64)); + writeq(val, qp->dpi->dbr); } int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp, @@ -1881,32 +1901,28 @@ done: /* Spinlock must be held */ static void bnxt_qplib_arm_cq_enable(struct bnxt_qplib_cq *cq) { - struct dbr_dbr db_msg = { 0 }; + u64 val = 0; - db_msg.type_xid = - cpu_to_le32(((cq->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) | - DBR_DBR_TYPE_CQ_ARMENA); + val = ((cq->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) | + DBC_DBC_TYPE_CQ_ARMENA; + val <<= 32; /* Flush memory writes before enabling the CQ */ - wmb(); - __iowrite64_copy(cq->dbr_base, &db_msg, sizeof(db_msg) / sizeof(u64)); + writeq(val, cq->dbr_base); } static void bnxt_qplib_arm_cq(struct bnxt_qplib_cq *cq, u32 arm_type) { struct bnxt_qplib_hwq *cq_hwq = &cq->hwq; - struct dbr_dbr db_msg = { 0 }; u32 sw_cons; + u64 val = 0; /* Ring DB */ + val = ((cq->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) | arm_type; + val <<= 32; sw_cons = HWQ_CMP(cq_hwq->cons, cq_hwq); - db_msg.index = cpu_to_le32((sw_cons << DBR_DBR_INDEX_SFT) & - DBR_DBR_INDEX_MASK); - db_msg.type_xid = - cpu_to_le32(((cq->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) | - arm_type); + val |= (sw_cons << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK; /* flush memory writes before arming the CQ */ - wmb(); - __iowrite64_copy(cq->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64)); + writeq(val, cq->dpi->dbr); } int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq) @@ -2053,6 +2069,7 @@ static int __flush_rq(struct bnxt_qplib_q *rq, struct bnxt_qplib_qp *qp, opcode = CQ_BASE_CQE_TYPE_RES_RC; break; case CMDQ_CREATE_QP_TYPE_UD: + case CMDQ_CREATE_QP_TYPE_GSI: opcode = CQ_BASE_CQE_TYPE_RES_UD; break; } @@ -2125,7 +2142,7 @@ static int do_wa9060(struct bnxt_qplib_qp *qp, struct bnxt_qplib_cq *cq, sq->send_phantom = true; /* TODO: Only ARM if the previous SQE is ARMALL */ - bnxt_qplib_arm_cq(cq, DBR_DBR_TYPE_CQ_ARMALL); + bnxt_qplib_arm_cq(cq, DBC_DBC_TYPE_CQ_ARMALL); rc = -EAGAIN; goto out; @@ -2410,12 +2427,14 @@ static int bnxt_qplib_cq_process_res_ud(struct bnxt_qplib_cq *cq, } cqe = *pcqe; cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK; - cqe->length = le32_to_cpu(hwcqe->length); + cqe->length = (u32)le16_to_cpu(hwcqe->length); + cqe->cfa_meta = le16_to_cpu(hwcqe->cfa_metadata); cqe->invrkey = le32_to_cpu(hwcqe->imm_data); cqe->flags = le16_to_cpu(hwcqe->flags); cqe->status = hwcqe->status; cqe->qp_handle = (u64)(unsigned long)qp; - memcpy(cqe->smac, hwcqe->src_mac, 6); + /*FIXME: Endianness fix needed for smace */ + memcpy(cqe->smac, hwcqe->src_mac, ETH_ALEN); wr_id_idx = le32_to_cpu(hwcqe->src_qp_high_srq_or_rq_wr_id) & CQ_RES_UD_SRQ_OR_RQ_WR_ID_MASK; cqe->src_qp = le16_to_cpu(hwcqe->src_qp_low) | @@ -2794,7 +2813,7 @@ int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe, } if (cq->hwq.cons != raw_cons) { cq->hwq.cons = raw_cons; - bnxt_qplib_arm_cq(cq, DBR_DBR_TYPE_CQ); + bnxt_qplib_arm_cq(cq, DBC_DBC_TYPE_CQ); } exit: return num_cqes - budget; diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.h b/drivers/infiniband/hw/bnxt_re/qplib_fp.h index 72352ca80ace..3f618b5f1f06 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_fp.h +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.h @@ -106,6 +106,7 @@ struct bnxt_qplib_swq { u32 start_psn; u32 next_psn; struct sq_psn_search *psn_search; + struct sq_psn_search_ext *psn_ext; }; struct bnxt_qplib_swqe { @@ -254,6 +255,7 @@ struct bnxt_qplib_q { struct bnxt_qplib_qp { struct bnxt_qplib_pd *pd; struct bnxt_qplib_dpi *dpi; + struct bnxt_qplib_chip_ctx *cctx; u64 qp_handle; #define BNXT_QPLIB_QP_ID_INVALID 0xFFFFFFFF u32 id; @@ -347,6 +349,7 @@ struct bnxt_qplib_cqe { u8 type; u8 opcode; u32 length; + u16 cfa_meta; u64 wr_id; union { __be32 immdata; @@ -432,13 +435,47 @@ struct bnxt_qplib_cq { #define NQ_DB_CP_FLAGS (NQ_DB_KEY_CP | \ NQ_DB_IDX_VALID | \ NQ_DB_IRQ_DIS) -#define NQ_DB_REARM(db, raw_cons, cp_bit) \ - writel(NQ_DB_CP_FLAGS_REARM | ((raw_cons) & ((cp_bit) - 1)), db) -#define NQ_DB(db, raw_cons, cp_bit) \ - writel(NQ_DB_CP_FLAGS | ((raw_cons) & ((cp_bit) - 1)), db) + +static inline void bnxt_qplib_ring_nq_db64(void __iomem *db, u32 index, + u32 xid, bool arm) +{ + u64 val; + + val = xid & DBC_DBC_XID_MASK; + val |= DBC_DBC_PATH_ROCE; + val |= arm ? DBC_DBC_TYPE_NQ_ARM : DBC_DBC_TYPE_NQ; + val <<= 32; + val |= index & DBC_DBC_INDEX_MASK; + writeq(val, db); +} + +static inline void bnxt_qplib_ring_nq_db_rearm(void __iomem *db, u32 raw_cons, + u32 max_elements, u32 xid, + bool gen_p5) +{ + u32 index = raw_cons & (max_elements - 1); + + if (gen_p5) + bnxt_qplib_ring_nq_db64(db, index, xid, true); + else + writel(NQ_DB_CP_FLAGS_REARM | (index & DBC_DBC32_XID_MASK), db); +} + +static inline void bnxt_qplib_ring_nq_db(void __iomem *db, u32 raw_cons, + u32 max_elements, u32 xid, + bool gen_p5) +{ + u32 index = raw_cons & (max_elements - 1); + + if (gen_p5) + bnxt_qplib_ring_nq_db64(db, index, xid, false); + else + writel(NQ_DB_CP_FLAGS | (index & DBC_DBC32_XID_MASK), db); +} struct bnxt_qplib_nq { struct pci_dev *pdev; + struct bnxt_qplib_res *res; int vector; cpumask_t mask; @@ -448,7 +485,7 @@ struct bnxt_qplib_nq { struct bnxt_qplib_hwq hwq; u16 bar_reg; - u16 bar_reg_off; + u32 bar_reg_off; u16 ring_id; void __iomem *bar_reg_iomem; diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c index 19551aa43850..c6461e957078 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c @@ -359,11 +359,12 @@ static int bnxt_qplib_process_qp_event(struct bnxt_qplib_rcfw *rcfw, static void bnxt_qplib_service_creq(unsigned long data) { struct bnxt_qplib_rcfw *rcfw = (struct bnxt_qplib_rcfw *)data; + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(rcfw->res->cctx); struct bnxt_qplib_hwq *creq = &rcfw->creq; + u32 type, budget = CREQ_ENTRY_POLL_BUDGET; struct creq_base *creqe, **creq_ptr; u32 sw_cons, raw_cons; unsigned long flags; - u32 type, budget = CREQ_ENTRY_POLL_BUDGET; /* Service the CREQ until budget is over */ spin_lock_irqsave(&creq->lock, flags); @@ -407,8 +408,9 @@ static void bnxt_qplib_service_creq(unsigned long data) if (creq->cons != raw_cons) { creq->cons = raw_cons; - CREQ_DB_REARM(rcfw->creq_bar_reg_iomem, raw_cons, - creq->max_elements); + bnxt_qplib_ring_creq_db_rearm(rcfw->creq_bar_reg_iomem, + raw_cons, creq->max_elements, + rcfw->creq_ring_id, gen_p5); } spin_unlock_irqrestore(&creq->lock, flags); } @@ -480,11 +482,13 @@ int bnxt_qplib_init_rcfw(struct bnxt_qplib_rcfw *rcfw, req.log2_dbr_pg_size = cpu_to_le16(PAGE_SHIFT - RCFW_DBR_BASE_PAGE_SHIFT); /* - * VFs need not setup the HW context area, PF + * Gen P5 devices doesn't require this allocation + * as the L2 driver does the same for RoCE also. + * Also, VFs need not setup the HW context area, PF * shall setup this area for VF. Skipping the * HW programming */ - if (is_virtfn) + if (is_virtfn || bnxt_qplib_is_chip_gen_p5(rcfw->res->cctx)) goto skip_ctx_setup; level = ctx->qpc_tbl.level; @@ -560,12 +564,15 @@ int bnxt_qplib_alloc_rcfw_channel(struct pci_dev *pdev, struct bnxt_qplib_ctx *ctx, int qp_tbl_sz) { + u8 hwq_type; + rcfw->pdev = pdev; rcfw->creq.max_elements = BNXT_QPLIB_CREQE_MAX_CNT; + hwq_type = bnxt_qplib_get_hwq_type(rcfw->res); if (bnxt_qplib_alloc_init_hwq(rcfw->pdev, &rcfw->creq, NULL, 0, &rcfw->creq.max_elements, - BNXT_QPLIB_CREQE_UNITS, 0, PAGE_SIZE, - HWQ_TYPE_L2_CMPL)) { + BNXT_QPLIB_CREQE_UNITS, + 0, PAGE_SIZE, hwq_type)) { dev_err(&rcfw->pdev->dev, "HW channel CREQ allocation failed\n"); goto fail; @@ -607,10 +614,13 @@ fail: void bnxt_qplib_rcfw_stop_irq(struct bnxt_qplib_rcfw *rcfw, bool kill) { + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(rcfw->res->cctx); + tasklet_disable(&rcfw->worker); /* Mask h/w interrupts */ - CREQ_DB(rcfw->creq_bar_reg_iomem, rcfw->creq.cons, - rcfw->creq.max_elements); + bnxt_qplib_ring_creq_db(rcfw->creq_bar_reg_iomem, rcfw->creq.cons, + rcfw->creq.max_elements, rcfw->creq_ring_id, + gen_p5); /* Sync with last running IRQ-handler */ synchronize_irq(rcfw->vector); if (kill) @@ -647,6 +657,7 @@ void bnxt_qplib_disable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw) int bnxt_qplib_rcfw_start_irq(struct bnxt_qplib_rcfw *rcfw, int msix_vector, bool need_init) { + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(rcfw->res->cctx); int rc; if (rcfw->requested) @@ -663,8 +674,9 @@ int bnxt_qplib_rcfw_start_irq(struct bnxt_qplib_rcfw *rcfw, int msix_vector, if (rc) return rc; rcfw->requested = true; - CREQ_DB_REARM(rcfw->creq_bar_reg_iomem, rcfw->creq.cons, - rcfw->creq.max_elements); + bnxt_qplib_ring_creq_db_rearm(rcfw->creq_bar_reg_iomem, + rcfw->creq.cons, rcfw->creq.max_elements, + rcfw->creq_ring_id, gen_p5); return 0; } @@ -684,8 +696,7 @@ int bnxt_qplib_enable_rcfw_channel(struct pci_dev *pdev, /* General */ rcfw->seq_num = 0; set_bit(FIRMWARE_FIRST_FLAG, &rcfw->flags); - bmap_size = BITS_TO_LONGS(rcfw->cmdq_depth * - sizeof(unsigned long)); + bmap_size = BITS_TO_LONGS(rcfw->cmdq_depth) * sizeof(unsigned long); rcfw->cmdq_bitmap = kzalloc(bmap_size, GFP_KERNEL); if (!rcfw->cmdq_bitmap) return -ENOMEM; @@ -718,8 +729,9 @@ int bnxt_qplib_enable_rcfw_channel(struct pci_dev *pdev, dev_err(&rcfw->pdev->dev, "CREQ BAR region %d resc start is 0!\n", rcfw->creq_bar_reg); + /* Unconditionally map 8 bytes to support 57500 series */ rcfw->creq_bar_reg_iomem = ioremap_nocache(res_base + cp_bar_reg_off, - 4); + 8); if (!rcfw->creq_bar_reg_iomem) { dev_err(&rcfw->pdev->dev, "CREQ BAR region %d mapping failed\n", rcfw->creq_bar_reg); diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h index be0ef0e8c53e..2138533bb642 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h +++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h @@ -157,10 +157,46 @@ static inline u32 get_creq_idx(u32 val) #define CREQ_DB_CP_FLAGS (CREQ_DB_KEY_CP | \ CREQ_DB_IDX_VALID | \ CREQ_DB_IRQ_DIS) -#define CREQ_DB_REARM(db, raw_cons, cp_bit) \ - writel(CREQ_DB_CP_FLAGS_REARM | ((raw_cons) & ((cp_bit) - 1)), db) -#define CREQ_DB(db, raw_cons, cp_bit) \ - writel(CREQ_DB_CP_FLAGS | ((raw_cons) & ((cp_bit) - 1)), db) + +static inline void bnxt_qplib_ring_creq_db64(void __iomem *db, u32 index, + u32 xid, bool arm) +{ + u64 val = 0; + + val = xid & DBC_DBC_XID_MASK; + val |= DBC_DBC_PATH_ROCE; + val |= arm ? DBC_DBC_TYPE_NQ_ARM : DBC_DBC_TYPE_NQ; + val <<= 32; + val |= index & DBC_DBC_INDEX_MASK; + + writeq(val, db); +} + +static inline void bnxt_qplib_ring_creq_db_rearm(void __iomem *db, u32 raw_cons, + u32 max_elements, u32 xid, + bool gen_p5) +{ + u32 index = raw_cons & (max_elements - 1); + + if (gen_p5) + bnxt_qplib_ring_creq_db64(db, index, xid, true); + else + writel(CREQ_DB_CP_FLAGS_REARM | (index & DBC_DBC32_XID_MASK), + db); +} + +static inline void bnxt_qplib_ring_creq_db(void __iomem *db, u32 raw_cons, + u32 max_elements, u32 xid, + bool gen_p5) +{ + u32 index = raw_cons & (max_elements - 1); + + if (gen_p5) + bnxt_qplib_ring_creq_db64(db, index, xid, true); + else + writel(CREQ_DB_CP_FLAGS | (index & DBC_DBC32_XID_MASK), + db); +} #define CREQ_ENTRY_POLL_BUDGET 0x100 @@ -187,6 +223,7 @@ struct bnxt_qplib_qp_node { /* RCFW Communication Channels */ struct bnxt_qplib_rcfw { struct pci_dev *pdev; + struct bnxt_qplib_res *res; int vector; struct tasklet_struct worker; bool requested; diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.c b/drivers/infiniband/hw/bnxt_re/qplib_res.c index 57d4951679cb..0bc24f934829 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_res.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_res.c @@ -85,7 +85,7 @@ static void __free_pbl(struct pci_dev *pdev, struct bnxt_qplib_pbl *pbl, static int __alloc_pbl(struct pci_dev *pdev, struct bnxt_qplib_pbl *pbl, struct scatterlist *sghead, u32 pages, u32 pg_size) { - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; bool is_umem = false; int i; @@ -116,13 +116,11 @@ static int __alloc_pbl(struct pci_dev *pdev, struct bnxt_qplib_pbl *pbl, } else { i = 0; is_umem = true; - for_each_sg(sghead, sg, pages, i) { - pbl->pg_map_arr[i] = sg_dma_address(sg); - pbl->pg_arr[i] = sg_virt(sg); - if (!pbl->pg_arr[i]) - goto fail; - + for_each_sg_dma_page (sghead, &sg_iter, pages, 0) { + pbl->pg_map_arr[i] = sg_page_iter_dma_address(&sg_iter); + pbl->pg_arr[i] = NULL; pbl->pg_count++; + i++; } } @@ -330,13 +328,13 @@ void bnxt_qplib_free_ctx(struct pci_dev *pdev, */ int bnxt_qplib_alloc_ctx(struct pci_dev *pdev, struct bnxt_qplib_ctx *ctx, - bool virt_fn) + bool virt_fn, bool is_p5) { int i, j, k, rc = 0; int fnz_idx = -1; __le64 **pbl_ptr; - if (virt_fn) + if (virt_fn || is_p5) goto stats_alloc; /* QPC Tables */ @@ -762,7 +760,11 @@ static int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev, { memset(stats, 0, sizeof(*stats)); stats->fw_id = -1; - stats->size = sizeof(struct ctx_hw_stats); + /* 128 byte aligned context memory is required only for 57500. + * However making this unconditional, it does not harm previous + * generation. + */ + stats->size = ALIGN(sizeof(struct ctx_hw_stats), 128); stats->dma = dma_alloc_coherent(&pdev->dev, stats->size, &stats->dma_map, GFP_KERNEL); if (!stats->dma) { diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.h b/drivers/infiniband/hw/bnxt_re/qplib_res.h index 1e80aa7bbcce..32cebd0f1436 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_res.h +++ b/drivers/infiniband/hw/bnxt_re/qplib_res.h @@ -180,12 +180,20 @@ struct bnxt_qplib_ctx { u64 hwrm_intf_ver; }; +struct bnxt_qplib_chip_ctx { + u16 chip_num; + u8 chip_rev; + u8 chip_metal; +}; + +#define CHIP_NUM_57500 0x1750 + struct bnxt_qplib_res { struct pci_dev *pdev; + struct bnxt_qplib_chip_ctx *cctx; struct net_device *netdev; struct bnxt_qplib_rcfw *rcfw; - struct bnxt_qplib_pd_tbl pd_tbl; struct bnxt_qplib_sgid_tbl sgid_tbl; struct bnxt_qplib_pkey_tbl pkey_tbl; @@ -193,6 +201,24 @@ struct bnxt_qplib_res { bool prio; }; +static inline bool bnxt_qplib_is_chip_gen_p5(struct bnxt_qplib_chip_ctx *cctx) +{ + return (cctx->chip_num == CHIP_NUM_57500); +} + +static inline u8 bnxt_qplib_get_hwq_type(struct bnxt_qplib_res *res) +{ + return bnxt_qplib_is_chip_gen_p5(res->cctx) ? + HWQ_TYPE_QUEUE : HWQ_TYPE_L2_CMPL; +} + +static inline u8 bnxt_qplib_get_ring_type(struct bnxt_qplib_chip_ctx *cctx) +{ + return bnxt_qplib_is_chip_gen_p5(cctx) ? + RING_ALLOC_REQ_RING_TYPE_NQ : + RING_ALLOC_REQ_RING_TYPE_ROCE_CMPL; +} + #define to_bnxt_qplib(ptr, type, member) \ container_of(ptr, type, member) @@ -226,5 +252,5 @@ void bnxt_qplib_free_ctx(struct pci_dev *pdev, struct bnxt_qplib_ctx *ctx); int bnxt_qplib_alloc_ctx(struct pci_dev *pdev, struct bnxt_qplib_ctx *ctx, - bool virt_fn); + bool virt_fn, bool is_p5); #endif /* __BNXT_QPLIB_RES_H__ */ diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c index efa0f2949dc7..e9c53e406404 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c @@ -119,7 +119,8 @@ int bnxt_qplib_get_dev_attr(struct bnxt_qplib_rcfw *rcfw, * reporting the max number */ attr->max_qp_wqes -= BNXT_QPLIB_RESERVED_QP_WRS; - attr->max_qp_sges = sb->max_sge; + attr->max_qp_sges = bnxt_qplib_is_chip_gen_p5(rcfw->res->cctx) ? + 6 : sb->max_sge; attr->max_cq = le32_to_cpu(sb->max_cq); attr->max_cq_wqes = le32_to_cpu(sb->max_cqe); attr->max_cq_sges = attr->max_qp_sges; diff --git a/drivers/infiniband/hw/bnxt_re/roce_hsi.h b/drivers/infiniband/hw/bnxt_re/roce_hsi.h index 8a9ead419ac2..e4b09e7c2175 100644 --- a/drivers/infiniband/hw/bnxt_re/roce_hsi.h +++ b/drivers/infiniband/hw/bnxt_re/roce_hsi.h @@ -49,11 +49,11 @@ struct cmpl_doorbell { #define CMPL_DOORBELL_IDX_SFT 0 #define CMPL_DOORBELL_RESERVED_MASK 0x3000000UL #define CMPL_DOORBELL_RESERVED_SFT 24 - #define CMPL_DOORBELL_IDX_VALID 0x4000000UL + #define CMPL_DOORBELL_IDX_VALID 0x4000000UL #define CMPL_DOORBELL_MASK 0x8000000UL #define CMPL_DOORBELL_KEY_MASK 0xf0000000UL #define CMPL_DOORBELL_KEY_SFT 28 - #define CMPL_DOORBELL_KEY_CMPL (0x2UL << 28) + #define CMPL_DOORBELL_KEY_CMPL (0x2UL << 28) }; /* Status Door Bell Format (4 bytes) */ @@ -71,46 +71,56 @@ struct status_doorbell { /* RoCE Host Structures */ /* Doorbell Structures */ -/* 64b Doorbell Format (8 bytes) */ -struct dbr_dbr { - __le32 index; - #define DBR_DBR_INDEX_MASK 0xfffffUL - #define DBR_DBR_INDEX_SFT 0 - #define DBR_DBR_RESERVED12_MASK 0xfff00000UL - #define DBR_DBR_RESERVED12_SFT 20 - __le32 type_xid; - #define DBR_DBR_XID_MASK 0xfffffUL - #define DBR_DBR_XID_SFT 0 - #define DBR_DBR_RESERVED8_MASK 0xff00000UL - #define DBR_DBR_RESERVED8_SFT 20 - #define DBR_DBR_TYPE_MASK 0xf0000000UL - #define DBR_DBR_TYPE_SFT 28 - #define DBR_DBR_TYPE_SQ (0x0UL << 28) - #define DBR_DBR_TYPE_RQ (0x1UL << 28) - #define DBR_DBR_TYPE_SRQ (0x2UL << 28) - #define DBR_DBR_TYPE_SRQ_ARM (0x3UL << 28) - #define DBR_DBR_TYPE_CQ (0x4UL << 28) - #define DBR_DBR_TYPE_CQ_ARMSE (0x5UL << 28) - #define DBR_DBR_TYPE_CQ_ARMALL (0x6UL << 28) - #define DBR_DBR_TYPE_CQ_ARMENA (0x7UL << 28) - #define DBR_DBR_TYPE_SRQ_ARMENA (0x8UL << 28) - #define DBR_DBR_TYPE_CQ_CUTOFF_ACK (0x9UL << 28) - #define DBR_DBR_TYPE_NULL (0xfUL << 28) -}; - -/* 32b Doorbell Format (4 bytes) */ -struct dbr_dbr32 { - __le32 type_abs_incr_xid; - #define DBR_DBR32_XID_MASK 0xfffffUL - #define DBR_DBR32_XID_SFT 0 - #define DBR_DBR32_RESERVED4_MASK 0xf00000UL - #define DBR_DBR32_RESERVED4_SFT 20 - #define DBR_DBR32_INCR_MASK 0xf000000UL - #define DBR_DBR32_INCR_SFT 24 - #define DBR_DBR32_ABS 0x10000000UL - #define DBR_DBR32_TYPE_MASK 0xe0000000UL - #define DBR_DBR32_TYPE_SFT 29 - #define DBR_DBR32_TYPE_SQ (0x0UL << 29) +/* dbc_dbc (size:64b/8B) */ +struct dbc_dbc { + __le32 index; + #define DBC_DBC_INDEX_MASK 0xffffffUL + #define DBC_DBC_INDEX_SFT 0 + __le32 type_path_xid; + #define DBC_DBC_XID_MASK 0xfffffUL + #define DBC_DBC_XID_SFT 0 + #define DBC_DBC_PATH_MASK 0x3000000UL + #define DBC_DBC_PATH_SFT 24 + #define DBC_DBC_PATH_ROCE (0x0UL << 24) + #define DBC_DBC_PATH_L2 (0x1UL << 24) + #define DBC_DBC_PATH_ENGINE (0x2UL << 24) + #define DBC_DBC_PATH_LAST DBC_DBC_PATH_ENGINE + #define DBC_DBC_DEBUG_TRACE 0x8000000UL + #define DBC_DBC_TYPE_MASK 0xf0000000UL + #define DBC_DBC_TYPE_SFT 28 + #define DBC_DBC_TYPE_SQ (0x0UL << 28) + #define DBC_DBC_TYPE_RQ (0x1UL << 28) + #define DBC_DBC_TYPE_SRQ (0x2UL << 28) + #define DBC_DBC_TYPE_SRQ_ARM (0x3UL << 28) + #define DBC_DBC_TYPE_CQ (0x4UL << 28) + #define DBC_DBC_TYPE_CQ_ARMSE (0x5UL << 28) + #define DBC_DBC_TYPE_CQ_ARMALL (0x6UL << 28) + #define DBC_DBC_TYPE_CQ_ARMENA (0x7UL << 28) + #define DBC_DBC_TYPE_SRQ_ARMENA (0x8UL << 28) + #define DBC_DBC_TYPE_CQ_CUTOFF_ACK (0x9UL << 28) + #define DBC_DBC_TYPE_NQ (0xaUL << 28) + #define DBC_DBC_TYPE_NQ_ARM (0xbUL << 28) + #define DBC_DBC_TYPE_NULL (0xfUL << 28) + #define DBC_DBC_TYPE_LAST DBC_DBC_TYPE_NULL +}; + +/* dbc_dbc32 (size:32b/4B) */ +struct dbc_dbc32 { + __le32 type_abs_incr_xid; + #define DBC_DBC32_XID_MASK 0xfffffUL + #define DBC_DBC32_XID_SFT 0 + #define DBC_DBC32_PATH_MASK 0xc00000UL + #define DBC_DBC32_PATH_SFT 22 + #define DBC_DBC32_PATH_ROCE (0x0UL << 22) + #define DBC_DBC32_PATH_L2 (0x1UL << 22) + #define DBC_DBC32_PATH_LAST DBC_DBC32_PATH_L2 + #define DBC_DBC32_INCR_MASK 0xf000000UL + #define DBC_DBC32_INCR_SFT 24 + #define DBC_DBC32_ABS 0x10000000UL + #define DBC_DBC32_TYPE_MASK 0xe0000000UL + #define DBC_DBC32_TYPE_SFT 29 + #define DBC_DBC32_TYPE_SQ (0x0UL << 29) + #define DBC_DBC32_TYPE_LAST DBC_DBC32_TYPE_SQ }; /* SQ WQE Structures */ @@ -149,7 +159,24 @@ struct sq_psn_search { #define SQ_PSN_SEARCH_NEXT_PSN_MASK 0xffffffUL #define SQ_PSN_SEARCH_NEXT_PSN_SFT 0 #define SQ_PSN_SEARCH_FLAGS_MASK 0xff000000UL - #define SQ_PSN_SEARCH_FLAGS_SFT 24 + #define SQ_PSN_SEARCH_FLAGS_SFT 24 +}; + +/* sq_psn_search_ext (size:128b/16B) */ +struct sq_psn_search_ext { + __le32 opcode_start_psn; + #define SQ_PSN_SEARCH_EXT_START_PSN_MASK 0xffffffUL + #define SQ_PSN_SEARCH_EXT_START_PSN_SFT 0 + #define SQ_PSN_SEARCH_EXT_OPCODE_MASK 0xff000000UL + #define SQ_PSN_SEARCH_EXT_OPCODE_SFT 24 + __le32 flags_next_psn; + #define SQ_PSN_SEARCH_EXT_NEXT_PSN_MASK 0xffffffUL + #define SQ_PSN_SEARCH_EXT_NEXT_PSN_SFT 0 + #define SQ_PSN_SEARCH_EXT_FLAGS_MASK 0xff000000UL + #define SQ_PSN_SEARCH_EXT_FLAGS_SFT 24 + __le16 start_slot_idx; + __le16 reserved16; + __le32 reserved32; }; /* Send SQ WQE (40 bytes) */ @@ -505,22 +532,24 @@ struct cq_res_rc { /* Responder UD CQE (32 bytes) */ struct cq_res_ud { - __le32 length; + __le16 length; #define CQ_RES_UD_LENGTH_MASK 0x3fffUL #define CQ_RES_UD_LENGTH_SFT 0 - #define CQ_RES_UD_RESERVED18_MASK 0xffffc000UL - #define CQ_RES_UD_RESERVED18_SFT 14 + __le16 cfa_metadata; + #define CQ_RES_UD_CFA_METADATA_VID_MASK 0xfffUL + #define CQ_RES_UD_CFA_METADATA_VID_SFT 0 + #define CQ_RES_UD_CFA_METADATA_DE 0x1000UL + #define CQ_RES_UD_CFA_METADATA_PRI_MASK 0xe000UL + #define CQ_RES_UD_CFA_METADATA_PRI_SFT 13 __le32 imm_data; __le64 qp_handle; __le16 src_mac[3]; __le16 src_qp_low; u8 cqe_type_toggle; - #define CQ_RES_UD_TOGGLE 0x1UL - #define CQ_RES_UD_CQE_TYPE_MASK 0x1eUL - #define CQ_RES_UD_CQE_TYPE_SFT 1 + #define CQ_RES_UD_TOGGLE 0x1UL + #define CQ_RES_UD_CQE_TYPE_MASK 0x1eUL + #define CQ_RES_UD_CQE_TYPE_SFT 1 #define CQ_RES_UD_CQE_TYPE_RES_UD (0x2UL << 1) - #define CQ_RES_UD_RESERVED3_MASK 0xe0UL - #define CQ_RES_UD_RESERVED3_SFT 5 u8 status; #define CQ_RES_UD_STATUS_OK 0x0UL #define CQ_RES_UD_STATUS_LOCAL_ACCESS_ERROR 0x1UL @@ -536,18 +565,30 @@ struct cq_res_ud { #define CQ_RES_UD_FLAGS_SRQ_SRQ (0x1UL << 0) #define CQ_RES_UD_FLAGS_SRQ_LAST CQ_RES_UD_FLAGS_SRQ_SRQ #define CQ_RES_UD_FLAGS_IMM 0x2UL - #define CQ_RES_UD_FLAGS_ROCE_IP_VER_MASK 0xcUL - #define CQ_RES_UD_FLAGS_ROCE_IP_VER_SFT 2 - #define CQ_RES_UD_FLAGS_ROCE_IP_VER_V1 (0x0UL << 2) - #define CQ_RES_UD_FLAGS_ROCE_IP_VER_V2IPV4 (0x2UL << 2) - #define CQ_RES_UD_FLAGS_ROCE_IP_VER_V2IPV6 (0x3UL << 2) + #define CQ_RES_UD_FLAGS_UNUSED_MASK 0xcUL + #define CQ_RES_UD_FLAGS_UNUSED_SFT 2 + #define CQ_RES_UD_FLAGS_ROCE_IP_VER_MASK 0x30UL + #define CQ_RES_UD_FLAGS_ROCE_IP_VER_SFT 4 + #define CQ_RES_UD_FLAGS_ROCE_IP_VER_V1 (0x0UL << 4) + #define CQ_RES_UD_FLAGS_ROCE_IP_VER_V2IPV4 (0x2UL << 4) + #define CQ_RES_UD_FLAGS_ROCE_IP_VER_V2IPV6 (0x3UL << 4) #define CQ_RES_UD_FLAGS_ROCE_IP_VER_LAST \ CQ_RES_UD_FLAGS_ROCE_IP_VER_V2IPV6 + #define CQ_RES_UD_FLAGS_META_FORMAT_MASK 0x3c0UL + #define CQ_RES_UD_FLAGS_META_FORMAT_SFT 6 + #define CQ_RES_UD_FLAGS_META_FORMAT_NONE (0x0UL << 6) + #define CQ_RES_UD_FLAGS_META_FORMAT_VLAN (0x1UL << 6) + #define CQ_RES_UD_FLAGS_META_FORMAT_TUNNEL_ID (0x2UL << 6) + #define CQ_RES_UD_FLAGS_META_FORMAT_CHDR_DATA (0x3UL << 6) + #define CQ_RES_UD_FLAGS_META_FORMAT_HDR_OFFSET (0x4UL << 6) + #define CQ_RES_UD_FLAGS_META_FORMAT_LAST \ + CQ_RES_UD_FLAGS_META_FORMAT_HDR_OFFSET + #define CQ_RES_UD_FLAGS_EXT_META_FORMAT_MASK 0xc00UL + #define CQ_RES_UD_FLAGS_EXT_META_FORMAT_SFT 10 + __le32 src_qp_high_srq_or_rq_wr_id; #define CQ_RES_UD_SRQ_OR_RQ_WR_ID_MASK 0xfffffUL #define CQ_RES_UD_SRQ_OR_RQ_WR_ID_SFT 0 - #define CQ_RES_UD_RESERVED4_MASK 0xf00000UL - #define CQ_RES_UD_RESERVED4_SFT 20 #define CQ_RES_UD_SRC_QP_HIGH_MASK 0xff000000UL #define CQ_RES_UD_SRC_QP_HIGH_SFT 24 }; @@ -983,6 +1024,7 @@ struct cmdq_create_qp { #define CMDQ_CREATE_QP_TYPE_RC 0x2UL #define CMDQ_CREATE_QP_TYPE_UD 0x4UL #define CMDQ_CREATE_QP_TYPE_RAW_ETHERTYPE 0x6UL + #define CMDQ_CREATE_QP_TYPE_GSI 0x7UL u8 sq_pg_size_sq_lvl; #define CMDQ_CREATE_QP_SQ_LVL_MASK 0xfUL #define CMDQ_CREATE_QP_SQ_LVL_SFT 0 @@ -2719,6 +2761,8 @@ struct creq_query_func_resp_sb { __le16 max_srq; __le32 max_gid; __le32 tqm_alloc_reqs[12]; + __le32 max_dpi; + __le32 reserved_32; }; /* Set resources command response (16 bytes) */ diff --git a/drivers/infiniband/hw/cxgb3/Makefile b/drivers/infiniband/hw/cxgb3/Makefile index 66fe0917aba0..34bb86a6ae3a 100644 --- a/drivers/infiniband/hw/cxgb3/Makefile +++ b/drivers/infiniband/hw/cxgb3/Makefile @@ -1,5 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 -ccflags-y := -Idrivers/net/ethernet/chelsio/cxgb3 +ccflags-y := -I $(srctree)/drivers/net/ethernet/chelsio/cxgb3 obj-$(CONFIG_INFINIBAND_CXGB3) += iw_cxgb3.o diff --git a/drivers/infiniband/hw/cxgb3/iwch.c b/drivers/infiniband/hw/cxgb3/iwch.c index 591de319c178..fb03bc492ef7 100644 --- a/drivers/infiniband/hw/cxgb3/iwch.c +++ b/drivers/infiniband/hw/cxgb3/iwch.c @@ -146,7 +146,7 @@ static void open_rnic_dev(struct t3cdev *tdev) pr_debug("%s t3cdev %p\n", __func__, tdev); pr_info_once("Chelsio T3 RDMA Driver - version %s\n", DRV_VERSION); - rnicp = (struct iwch_dev *)ib_alloc_device(sizeof(*rnicp)); + rnicp = ib_alloc_device(iwch_dev, ibdev); if (!rnicp) { pr_err("Cannot allocate ib device\n"); return; diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c index b34b1a1bd94b..4accf7b3dcf2 100644 --- a/drivers/infiniband/hw/cxgb3/iwch_provider.c +++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c @@ -53,6 +53,7 @@ #include <rdma/ib_smi.h> #include <rdma/ib_umem.h> #include <rdma/ib_user_verbs.h> +#include <rdma/uverbs_ioctl.h> #include "cxio_hal.h" #include "iwch.h" @@ -61,7 +62,7 @@ #include <rdma/cxgb3-abi.h> #include "common.h" -static int iwch_dealloc_ucontext(struct ib_ucontext *context) +static void iwch_dealloc_ucontext(struct ib_ucontext *context) { struct iwch_dev *rhp = to_iwch_dev(context->device); struct iwch_ucontext *ucontext = to_iwch_ucontext(context); @@ -71,24 +72,20 @@ static int iwch_dealloc_ucontext(struct ib_ucontext *context) list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry) kfree(mm); cxio_release_ucontext(&rhp->rdev, &ucontext->uctx); - kfree(ucontext); - return 0; } -static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +static int iwch_alloc_ucontext(struct ib_ucontext *ucontext, + struct ib_udata *udata) { - struct iwch_ucontext *context; + struct ib_device *ibdev = ucontext->device; + struct iwch_ucontext *context = to_iwch_ucontext(ucontext); struct iwch_dev *rhp = to_iwch_dev(ibdev); pr_debug("%s ibdev %p\n", __func__, ibdev); - context = kzalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); cxio_init_ucontext(&rhp->rdev, &context->uctx); INIT_LIST_HEAD(&context->mmaps); spin_lock_init(&context->mmap_lock); - return &context->ibucontext; + return 0; } static int iwch_destroy_cq(struct ib_cq *ib_cq) @@ -370,7 +367,7 @@ static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) return ret; } -static int iwch_deallocate_pd(struct ib_pd *pd) +static void iwch_deallocate_pd(struct ib_pd *pd) { struct iwch_dev *rhp; struct iwch_pd *php; @@ -379,15 +376,13 @@ static int iwch_deallocate_pd(struct ib_pd *pd) rhp = php->rhp; pr_debug("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid); cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid); - kfree(php); - return 0; } -static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +static int iwch_allocate_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct iwch_pd *php; + struct iwch_pd *php = to_iwch_pd(pd); + struct ib_device *ibdev = pd->device; u32 pdid; struct iwch_dev *rhp; @@ -395,12 +390,8 @@ static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev, rhp = (struct iwch_dev *) ibdev; pdid = cxio_hal_get_pdid(rhp->rdev.rscp); if (!pdid) - return ERR_PTR(-EINVAL); - php = kzalloc(sizeof(*php), GFP_KERNEL); - if (!php) { - cxio_hal_put_pdid(rhp->rdev.rscp, pdid); - return ERR_PTR(-ENOMEM); - } + return -EINVAL; + php->pdid = pdid; php->rhp = rhp; if (context) { @@ -408,11 +399,11 @@ static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev, if (ib_copy_to_udata(udata, &resp, sizeof(resp))) { iwch_deallocate_pd(&php->ibpd); - return ERR_PTR(-EFAULT); + return -EFAULT; } } pr_debug("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php); - return &php->ibpd; + return 0; } static int iwch_dereg_mr(struct ib_mr *ib_mr) @@ -522,14 +513,13 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt, int acc, struct ib_udata *udata) { __be64 *pages; - int shift, n, len; - int i, k, entry; + int shift, n, i; int err = 0; struct iwch_dev *rhp; struct iwch_pd *php; struct iwch_mr *mhp; struct iwch_reg_user_mr_resp uresp; - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; pr_debug("%s ib_pd %p\n", __func__, pd); php = to_iwch_pd(pd); @@ -540,14 +530,14 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mhp->rhp = rhp; - mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0); + mhp->umem = ib_umem_get(udata, start, length, acc, 0); if (IS_ERR(mhp->umem)) { err = PTR_ERR(mhp->umem); kfree(mhp); return ERR_PTR(err); } - shift = mhp->umem->page_shift; + shift = PAGE_SHIFT; n = mhp->umem->nmap; @@ -563,19 +553,15 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, i = n = 0; - for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) { - len = sg_dma_len(sg) >> shift; - for (k = 0; k < len; ++k) { - pages[i++] = cpu_to_be64(sg_dma_address(sg) + - (k << shift)); - if (i == PAGE_SIZE / sizeof *pages) { - err = iwch_write_pbl(mhp, pages, i, n); - if (err) - goto pbl_done; - n += i; - i = 0; - } - } + for_each_sg_dma_page(mhp->umem->sg_head.sgl, &sg_iter, mhp->umem->nmap, 0) { + pages[i++] = cpu_to_be64(sg_page_iter_dma_address(&sg_iter)); + if (i == PAGE_SIZE / sizeof *pages) { + err = iwch_write_pbl(mhp, pages, i, n); + if (err) + goto pbl_done; + n += i; + i = 0; + } } if (i) @@ -836,7 +822,8 @@ static struct ib_qp *iwch_create_qp(struct ib_pd *pd, * Kernel users need more wq space for fastreg WRs which can take * 2 WR fragments. */ - ucontext = udata ? to_iwch_ucontext(pd->uobject->context) : NULL; + ucontext = rdma_udata_to_drv_context(udata, struct iwch_ucontext, + ibucontext); if (!ucontext && wqsize < (rqsize + (2 * sqsize))) wqsize = roundup_pow_of_two(rqsize + roundup_pow_of_two(attrs->cap.max_send_wr * 2)); @@ -1130,8 +1117,9 @@ static int iwch_query_port(struct ib_device *ibdev, static ssize_t hw_rev_show(struct device *dev, struct device_attribute *attr, char *buf) { - struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev, - ibdev.dev); + struct iwch_dev *iwch_dev = + rdma_device_to_drv_device(dev, struct iwch_dev, ibdev); + pr_debug("%s dev 0x%p\n", __func__, dev); return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type); } @@ -1140,8 +1128,8 @@ static DEVICE_ATTR_RO(hw_rev); static ssize_t hca_type_show(struct device *dev, struct device_attribute *attr, char *buf) { - struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev, - ibdev.dev); + struct iwch_dev *iwch_dev = + rdma_device_to_drv_device(dev, struct iwch_dev, ibdev); struct ethtool_drvinfo info; struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev; @@ -1154,8 +1142,9 @@ static DEVICE_ATTR_RO(hca_type); static ssize_t board_id_show(struct device *dev, struct device_attribute *attr, char *buf) { - struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev, - ibdev.dev); + struct iwch_dev *iwch_dev = + rdma_device_to_drv_device(dev, struct iwch_dev, ibdev); + pr_debug("%s dev 0x%p\n", __func__, dev); return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor, iwch_dev->rdev.rnic_info.pdev->device); @@ -1348,6 +1337,8 @@ static const struct ib_device_ops iwch_dev_ops = { .reg_user_mr = iwch_reg_user_mr, .req_notify_cq = iwch_arm_cq, .resize_cq = iwch_resize_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, iwch_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, iwch_ucontext, ibucontext), }; int iwch_register_device(struct iwch_dev *dev) @@ -1391,7 +1382,7 @@ int iwch_register_device(struct iwch_dev *dev) dev->ibdev.dev.parent = &dev->rdev.rnic_info.pdev->dev; dev->ibdev.uverbs_abi_ver = IWCH_UVERBS_ABI_VERSION; - dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL); + dev->ibdev.iwcm = kzalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL); if (!dev->ibdev.iwcm) return -ENOMEM; @@ -1409,7 +1400,7 @@ int iwch_register_device(struct iwch_dev *dev) dev->ibdev.driver_id = RDMA_DRIVER_CXGB3; rdma_set_device_sysfs_group(&dev->ibdev, &iwch_attr_group); ib_set_device_ops(&dev->ibdev, &iwch_dev_ops); - ret = ib_register_device(&dev->ibdev, "cxgb3_%d", NULL); + ret = ib_register_device(&dev->ibdev, "cxgb3_%d"); if (ret) kfree(dev->ibdev.iwcm); return ret; diff --git a/drivers/infiniband/hw/cxgb4/Makefile b/drivers/infiniband/hw/cxgb4/Makefile index 9edd92023e18..31a87d90a40b 100644 --- a/drivers/infiniband/hw/cxgb4/Makefile +++ b/drivers/infiniband/hw/cxgb4/Makefile @@ -1,5 +1,5 @@ -ccflags-y := -Idrivers/net/ethernet/chelsio/cxgb4 -ccflags-y += -Idrivers/net/ethernet/chelsio/libcxgb +ccflags-y := -I $(srctree)/drivers/net/ethernet/chelsio/cxgb4 +ccflags-y += -I $(srctree)/drivers/net/ethernet/chelsio/libcxgb obj-$(CONFIG_INFINIBAND_CXGB4) += iw_cxgb4.o diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c index 8221813219e5..4d232bdf9e97 100644 --- a/drivers/infiniband/hw/cxgb4/cm.c +++ b/drivers/infiniband/hw/cxgb4/cm.c @@ -655,7 +655,33 @@ static int send_halfclose(struct c4iw_ep *ep) return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); } -static int send_abort(struct c4iw_ep *ep) +static void read_tcb(struct c4iw_ep *ep) +{ + struct sk_buff *skb; + struct cpl_get_tcb *req; + int wrlen = roundup(sizeof(*req), 16); + + skb = get_skb(NULL, sizeof(*req), GFP_KERNEL); + if (WARN_ON(!skb)) + return; + + set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx); + req = (struct cpl_get_tcb *) skb_put(skb, wrlen); + memset(req, 0, wrlen); + INIT_TP_WR(req, ep->hwtid); + OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid)); + req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid)); + + /* + * keep a ref on the ep so the tcb is not unlocked before this + * cpl completes. The ref is released in read_tcb_rpl(). + */ + c4iw_get_ep(&ep->com); + if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb))) + c4iw_put_ep(&ep->com); +} + +static int send_abort_req(struct c4iw_ep *ep) { u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16); struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list); @@ -670,6 +696,17 @@ static int send_abort(struct c4iw_ep *ep) return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t); } +static int send_abort(struct c4iw_ep *ep) +{ + if (!ep->com.qp || !ep->com.qp->srq) { + send_abort_req(ep); + return 0; + } + set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags); + read_tcb(ep); + return 0; +} + static int send_connect(struct c4iw_ep *ep) { struct cpl_act_open_req *req = NULL; @@ -1851,14 +1888,11 @@ static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb) return 0; } -static void complete_cached_srq_buffers(struct c4iw_ep *ep, - __be32 srqidx_status) +static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx) { enum chip_type adapter_type; - u32 srqidx; adapter_type = ep->com.dev->rdev.lldi.adapter_type; - srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(srqidx_status)); /* * If this TCB had a srq buffer cached, then we must complete @@ -1876,6 +1910,7 @@ static void complete_cached_srq_buffers(struct c4iw_ep *ep, static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) { + u32 srqidx; struct c4iw_ep *ep; struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb); int release = 0; @@ -1887,7 +1922,10 @@ static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) return 0; } - complete_cached_srq_buffers(ep, rpl->srqidx_status); + if (ep->com.qp && ep->com.qp->srq) { + srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status)); + complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx); + } pr_debug("ep %p tid %u\n", ep, ep->hwtid); mutex_lock(&ep->com.mutex); @@ -1903,8 +1941,10 @@ static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) } mutex_unlock(&ep->com.mutex); - if (release) + if (release) { + close_complete_upcall(ep, -ECONNRESET); release_ep_resources(ep); + } c4iw_put_ep(&ep->com); return 0; } @@ -2072,7 +2112,7 @@ static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip, } else { pdev = get_real_dev(n->dev); ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, - n, pdev, 0); + n, pdev, rt_tos2priority(tos)); if (!ep->l2t) goto out; ep->mtu = dst_mtu(dst); @@ -2161,7 +2201,8 @@ static int c4iw_reconnect(struct c4iw_ep *ep) laddr6->sin6_addr.s6_addr, raddr6->sin6_addr.s6_addr, laddr6->sin6_port, - raddr6->sin6_port, 0, + raddr6->sin6_port, + ep->com.cm_id->tos, raddr6->sin6_scope_id); iptype = 6; ra = (__u8 *)&raddr6->sin6_addr; @@ -2476,7 +2517,7 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) u16 peer_mss = ntohs(req->tcpopt.mss); int iptype; unsigned short hdrs; - u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid)); + u8 tos; parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid); if (!parent_ep) { @@ -2490,6 +2531,11 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) goto reject; } + if (parent_ep->com.cm_id->tos_set) + tos = parent_ep->com.cm_id->tos; + else + tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid)); + cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, &iptype, local_ip, peer_ip, &local_port, &peer_port); @@ -2509,7 +2555,7 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) ntohs(peer_port), peer_mss); dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev, local_ip, peer_ip, local_port, peer_port, - PASS_OPEN_TOS_G(ntohl(req->tos_stid)), + tos, ((struct sockaddr_in6 *) &parent_ep->com.local_addr)->sin6_scope_id); } @@ -2740,6 +2786,21 @@ static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb) return 0; } +static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep) +{ + complete_cached_srq_buffers(ep, ep->srqe_idx); + if (ep->com.cm_id && ep->com.qp) { + struct c4iw_qp_attributes attrs; + + attrs.next_state = C4IW_QP_STATE_ERROR; + c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, + C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); + } + peer_abort_upcall(ep); + release_ep_resources(ep); + c4iw_put_ep(&ep->com); +} + static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) { struct cpl_abort_req_rss6 *req = cplhdr(skb); @@ -2750,6 +2811,7 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) int release = 0; unsigned int tid = GET_TID(req); u8 status; + u32 srqidx; u32 len = roundup(sizeof(struct cpl_abort_rpl), 16); @@ -2769,8 +2831,6 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) goto deref_ep; } - complete_cached_srq_buffers(ep, req->srqidx_status); - pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state); set_bit(PEER_ABORT, &ep->com.history); @@ -2819,6 +2879,23 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) stop_ep_timer(ep); /*FALLTHROUGH*/ case FPDU_MODE: + if (ep->com.qp && ep->com.qp->srq) { + srqidx = ABORT_RSS_SRQIDX_G( + be32_to_cpu(req->srqidx_status)); + if (srqidx) { + complete_cached_srq_buffers(ep, + req->srqidx_status); + } else { + /* Hold ep ref until finish_peer_abort() */ + c4iw_get_ep(&ep->com); + __state_set(&ep->com, ABORTING); + set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags); + read_tcb(ep); + break; + + } + } + if (ep->com.cm_id && ep->com.qp) { attrs.next_state = C4IW_QP_STATE_ERROR; ret = c4iw_modify_qp(ep->com.qp->rhp, @@ -2942,15 +3019,18 @@ static int terminate(struct c4iw_dev *dev, struct sk_buff *skb) ep = get_ep_from_tid(dev, tid); - if (ep && ep->com.qp) { - pr_warn("TERM received tid %u qpid %u\n", - tid, ep->com.qp->wq.sq.qid); - attrs.next_state = C4IW_QP_STATE_TERMINATE; - c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, - C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); + if (ep) { + if (ep->com.qp) { + pr_warn("TERM received tid %u qpid %u\n", tid, + ep->com.qp->wq.sq.qid); + attrs.next_state = C4IW_QP_STATE_TERMINATE; + c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, + C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); + } + + c4iw_put_ep(&ep->com); } else pr_warn("TERM received tid %u no ep/qp\n", tid); - c4iw_put_ep(&ep->com); return 0; } @@ -3318,7 +3398,7 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) laddr6->sin6_addr.s6_addr, raddr6->sin6_addr.s6_addr, laddr6->sin6_port, - raddr6->sin6_port, 0, + raddr6->sin6_port, cm_id->tos, raddr6->sin6_scope_id); } if (!ep->dst) { @@ -3606,7 +3686,6 @@ int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) if (close) { if (abrupt) { set_bit(EP_DISC_ABORT, &ep->com.history); - close_complete_upcall(ep, -ECONNRESET); ret = send_abort(ep); } else { set_bit(EP_DISC_CLOSE, &ep->com.history); @@ -3717,6 +3796,80 @@ static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, return; } +static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word) +{ + u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]); + u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]); + u64 t; + u32 shift = 32; + + t = (thi << shift) | (tlo >> shift); + + return t; +} + +static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift) +{ + u32 v; + u64 t = be64_to_cpu(tcb[(31 - word) / 2]); + + if (word & 0x1) + shift += 32; + v = (t >> shift) & mask; + return v; +} + +static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb) +{ + struct cpl_get_tcb_rpl *rpl = cplhdr(skb); + __be64 *tcb = (__be64 *)(rpl + 1); + unsigned int tid = GET_TID(rpl); + struct c4iw_ep *ep; + u64 t_flags_64; + u32 rx_pdu_out; + + ep = get_ep_from_tid(dev, tid); + if (!ep) + return 0; + /* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to + * determine if there's a rx PDU feedback event pending. + * + * If that bit is set, it means we'll need to re-read the TCB's + * rq_start value. The final value is the one present in a TCB + * with the TF_RX_PDU_OUT bit cleared. + */ + + t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W); + rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S; + + c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */ + c4iw_put_ep(&ep->com); /* from read_tcb() */ + + /* If TF_RX_PDU_OUT bit is set, re-read the TCB */ + if (rx_pdu_out) { + if (++ep->rx_pdu_out_cnt >= 2) { + WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n"); + goto cleanup; + } + read_tcb(ep); + return 0; + } + + ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_W, + TCB_RQ_START_S); +cleanup: + pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx); + + if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) + finish_peer_abort(dev, ep); + else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) + send_abort_req(ep); + else + WARN_ONCE(1, "unexpected state!"); + + return 0; +} + static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) { struct cpl_fw6_msg *rpl = cplhdr(skb); @@ -4037,6 +4190,7 @@ static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = { [CPL_CLOSE_CON_RPL] = close_con_rpl, [CPL_RDMA_TERMINATE] = terminate, [CPL_FW4_ACK] = fw4_ack, + [CPL_GET_TCB_RPL] = read_tcb_rpl, [CPL_FW6_MSG] = deferred_fw6_msg, [CPL_RX_PKT] = rx_pkt, [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe, @@ -4268,6 +4422,7 @@ c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = { [CPL_RDMA_TERMINATE] = sched, [CPL_FW4_ACK] = sched, [CPL_SET_TCB_RPL] = set_tcb_rpl, + [CPL_GET_TCB_RPL] = sched, [CPL_FW6_MSG] = fw6_msg, [CPL_RX_PKT] = sched }; diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c index d499cd61c0e8..c79cf63fb0bb 100644 --- a/drivers/infiniband/hw/cxgb4/device.c +++ b/drivers/infiniband/hw/cxgb4/device.c @@ -720,11 +720,8 @@ static const struct file_operations ep_debugfs_fops = { .read = debugfs_read, }; -static int setup_debugfs(struct c4iw_dev *devp) +static void setup_debugfs(struct c4iw_dev *devp) { - if (!devp->debugfs_root) - return -1; - debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root, (void *)devp, &qp_debugfs_fops, 4096); @@ -740,7 +737,6 @@ static int setup_debugfs(struct c4iw_dev *devp) if (c4iw_wr_log) debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root, (void *)devp, &wr_log_debugfs_fops, 4096); - return 0; } void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev, @@ -981,7 +977,7 @@ static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop) pr_info("%s: On-Chip Queues not supported on this device\n", pci_name(infop->pdev)); - devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp)); + devp = ib_alloc_device(c4iw_dev, ibdev); if (!devp) { pr_err("Cannot allocate ib device\n"); return ERR_PTR(-ENOMEM); @@ -1564,8 +1560,6 @@ static int __init c4iw_init_module(void) return err; c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL); - if (!c4iw_debugfs_root) - pr_warn("could not create debugfs entry, continuing\n"); reg_workq = create_singlethread_workqueue("Register_iWARP_device"); if (!reg_workq) { diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h index f0fceadd0d12..5a5da41faef6 100644 --- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h +++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h @@ -589,7 +589,6 @@ struct c4iw_ucontext { u32 key; spinlock_t mmap_lock; struct list_head mmaps; - struct kref kref; bool is_32b_cqe; }; @@ -598,18 +597,6 @@ static inline struct c4iw_ucontext *to_c4iw_ucontext(struct ib_ucontext *c) return container_of(c, struct c4iw_ucontext, ibucontext); } -void _c4iw_free_ucontext(struct kref *kref); - -static inline void c4iw_put_ucontext(struct c4iw_ucontext *ucontext) -{ - kref_put(&ucontext->kref, _c4iw_free_ucontext); -} - -static inline void c4iw_get_ucontext(struct c4iw_ucontext *ucontext) -{ - kref_get(&ucontext->kref); -} - struct c4iw_mm_entry { struct list_head entry; u64 addr; @@ -982,6 +969,9 @@ struct c4iw_ep { int rcv_win; u32 snd_wscale; struct c4iw_ep_stats stats; + u32 srqe_idx; + u32 rx_pdu_out_cnt; + struct sk_buff *peer_abort_skb; }; static inline struct c4iw_ep *to_ep(struct iw_cm_id *cm_id) diff --git a/drivers/infiniband/hw/cxgb4/mem.c b/drivers/infiniband/hw/cxgb4/mem.c index 7b76e6f81aeb..5baa31ab6366 100644 --- a/drivers/infiniband/hw/cxgb4/mem.c +++ b/drivers/infiniband/hw/cxgb4/mem.c @@ -502,10 +502,9 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt, int acc, struct ib_udata *udata) { __be64 *pages; - int shift, n, len; - int i, k, entry; + int shift, n, i; int err = -ENOMEM; - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; struct c4iw_dev *rhp; struct c4iw_pd *php; struct c4iw_mr *mhp; @@ -537,11 +536,11 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mhp->rhp = rhp; - mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0); + mhp->umem = ib_umem_get(udata, start, length, acc, 0); if (IS_ERR(mhp->umem)) goto err_free_skb; - shift = mhp->umem->page_shift; + shift = PAGE_SHIFT; n = mhp->umem->nmap; err = alloc_pbl(mhp, n); @@ -556,21 +555,16 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, i = n = 0; - for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) { - len = sg_dma_len(sg) >> shift; - for (k = 0; k < len; ++k) { - pages[i++] = cpu_to_be64(sg_dma_address(sg) + - (k << shift)); - if (i == PAGE_SIZE / sizeof *pages) { - err = write_pbl(&mhp->rhp->rdev, - pages, - mhp->attr.pbl_addr + (n << 3), i, - mhp->wr_waitp); - if (err) - goto pbl_done; - n += i; - i = 0; - } + for_each_sg_dma_page(mhp->umem->sg_head.sgl, &sg_iter, mhp->umem->nmap, 0) { + pages[i++] = cpu_to_be64(sg_page_iter_dma_address(&sg_iter)); + if (i == PAGE_SIZE / sizeof(*pages)) { + err = write_pbl(&mhp->rhp->rdev, pages, + mhp->attr.pbl_addr + (n << 3), i, + mhp->wr_waitp); + if (err) + goto pbl_done; + n += i; + i = 0; } } @@ -684,8 +678,8 @@ int c4iw_dealloc_mw(struct ib_mw *mw) mhp->wr_waitp); kfree_skb(mhp->dereg_skb); c4iw_put_wr_wait(mhp->wr_waitp); - kfree(mhp); pr_debug("ib_mw %p mmid 0x%x ptr %p\n", mw, mmid, mhp); + kfree(mhp); return 0; } diff --git a/drivers/infiniband/hw/cxgb4/provider.c b/drivers/infiniband/hw/cxgb4/provider.c index 586b0c37481f..507c54572cc9 100644 --- a/drivers/infiniband/hw/cxgb4/provider.c +++ b/drivers/infiniband/hw/cxgb4/provider.c @@ -58,51 +58,34 @@ static int fastreg_support = 1; module_param(fastreg_support, int, 0644); MODULE_PARM_DESC(fastreg_support, "Advertise fastreg support (default=1)"); -void _c4iw_free_ucontext(struct kref *kref) +static void c4iw_dealloc_ucontext(struct ib_ucontext *context) { - struct c4iw_ucontext *ucontext; + struct c4iw_ucontext *ucontext = to_c4iw_ucontext(context); struct c4iw_dev *rhp; struct c4iw_mm_entry *mm, *tmp; - ucontext = container_of(kref, struct c4iw_ucontext, kref); + pr_debug("context %p\n", context); rhp = to_c4iw_dev(ucontext->ibucontext.device); - pr_debug("ucontext %p\n", ucontext); list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry) kfree(mm); c4iw_release_dev_ucontext(&rhp->rdev, &ucontext->uctx); - kfree(ucontext); } -static int c4iw_dealloc_ucontext(struct ib_ucontext *context) +static int c4iw_alloc_ucontext(struct ib_ucontext *ucontext, + struct ib_udata *udata) { - struct c4iw_ucontext *ucontext = to_c4iw_ucontext(context); - - pr_debug("context %p\n", context); - c4iw_put_ucontext(ucontext); - return 0; -} - -static struct ib_ucontext *c4iw_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) -{ - struct c4iw_ucontext *context; + struct ib_device *ibdev = ucontext->device; + struct c4iw_ucontext *context = to_c4iw_ucontext(ucontext); struct c4iw_dev *rhp = to_c4iw_dev(ibdev); struct c4iw_alloc_ucontext_resp uresp; int ret = 0; struct c4iw_mm_entry *mm = NULL; pr_debug("ibdev %p\n", ibdev); - context = kzalloc(sizeof(*context), GFP_KERNEL); - if (!context) { - ret = -ENOMEM; - goto err; - } - c4iw_init_dev_ucontext(&rhp->rdev, &context->uctx); INIT_LIST_HEAD(&context->mmaps); spin_lock_init(&context->mmap_lock); - kref_init(&context->kref); if (udata->outlen < sizeof(uresp) - sizeof(uresp.reserved)) { pr_err_once("Warning - downlevel libcxgb4 (non-fatal), device status page disabled\n"); @@ -111,7 +94,7 @@ static struct ib_ucontext *c4iw_alloc_ucontext(struct ib_device *ibdev, mm = kmalloc(sizeof(*mm), GFP_KERNEL); if (!mm) { ret = -ENOMEM; - goto err_free; + goto err; } uresp.status_page_size = PAGE_SIZE; @@ -131,13 +114,11 @@ static struct ib_ucontext *c4iw_alloc_ucontext(struct ib_device *ibdev, mm->len = PAGE_SIZE; insert_mmap(context, mm); } - return &context->ibucontext; + return 0; err_mm: kfree(mm); -err_free: - kfree(context); err: - return ERR_PTR(ret); + return ret; } static int c4iw_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) @@ -209,7 +190,7 @@ static int c4iw_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) return ret; } -static int c4iw_deallocate_pd(struct ib_pd *pd) +static void c4iw_deallocate_pd(struct ib_pd *pd) { struct c4iw_dev *rhp; struct c4iw_pd *php; @@ -221,15 +202,13 @@ static int c4iw_deallocate_pd(struct ib_pd *pd) mutex_lock(&rhp->rdev.stats.lock); rhp->rdev.stats.pd.cur--; mutex_unlock(&rhp->rdev.stats.lock); - kfree(php); - return 0; } -static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +static int c4iw_allocate_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct c4iw_pd *php; + struct c4iw_pd *php = to_c4iw_pd(pd); + struct ib_device *ibdev = pd->device; u32 pdid; struct c4iw_dev *rhp; @@ -237,12 +216,8 @@ static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev, rhp = (struct c4iw_dev *) ibdev; pdid = c4iw_get_resource(&rhp->rdev.resource.pdid_table); if (!pdid) - return ERR_PTR(-EINVAL); - php = kzalloc(sizeof(*php), GFP_KERNEL); - if (!php) { - c4iw_put_resource(&rhp->rdev.resource.pdid_table, pdid); - return ERR_PTR(-ENOMEM); - } + return -EINVAL; + php->pdid = pdid; php->rhp = rhp; if (context) { @@ -250,7 +225,7 @@ static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev, if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) { c4iw_deallocate_pd(&php->ibpd); - return ERR_PTR(-EFAULT); + return -EFAULT; } } mutex_lock(&rhp->rdev.stats.lock); @@ -259,7 +234,7 @@ static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev, rhp->rdev.stats.pd.max = rhp->rdev.stats.pd.cur; mutex_unlock(&rhp->rdev.stats.lock); pr_debug("pdid 0x%0x ptr 0x%p\n", pdid, php); - return &php->ibpd; + return 0; } static int c4iw_query_pkey(struct ib_device *ibdev, u8 port, u16 index, @@ -376,8 +351,9 @@ static int c4iw_query_port(struct ib_device *ibdev, u8 port, static ssize_t hw_rev_show(struct device *dev, struct device_attribute *attr, char *buf) { - struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev, - ibdev.dev); + struct c4iw_dev *c4iw_dev = + rdma_device_to_drv_device(dev, struct c4iw_dev, ibdev); + pr_debug("dev 0x%p\n", dev); return sprintf(buf, "%d\n", CHELSIO_CHIP_RELEASE(c4iw_dev->rdev.lldi.adapter_type)); @@ -387,8 +363,8 @@ static DEVICE_ATTR_RO(hw_rev); static ssize_t hca_type_show(struct device *dev, struct device_attribute *attr, char *buf) { - struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev, - ibdev.dev); + struct c4iw_dev *c4iw_dev = + rdma_device_to_drv_device(dev, struct c4iw_dev, ibdev); struct ethtool_drvinfo info; struct net_device *lldev = c4iw_dev->rdev.lldi.ports[0]; @@ -401,8 +377,9 @@ static DEVICE_ATTR_RO(hca_type); static ssize_t board_id_show(struct device *dev, struct device_attribute *attr, char *buf) { - struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev, - ibdev.dev); + struct c4iw_dev *c4iw_dev = + rdma_device_to_drv_device(dev, struct c4iw_dev, ibdev); + pr_debug("dev 0x%p\n", dev); return sprintf(buf, "%x.%x\n", c4iw_dev->rdev.lldi.pdev->vendor, c4iw_dev->rdev.lldi.pdev->device); @@ -547,6 +524,7 @@ static const struct ib_device_ops c4iw_dev_ops = { .destroy_cq = c4iw_destroy_cq, .destroy_qp = c4iw_destroy_qp, .destroy_srq = c4iw_destroy_srq, + .fill_res_entry = fill_res_entry, .get_dev_fw_str = get_dev_fw_str, .get_dma_mr = c4iw_get_dma_mr, .get_hw_stats = c4iw_get_mib, @@ -567,6 +545,8 @@ static const struct ib_device_ops c4iw_dev_ops = { .query_qp = c4iw_ib_query_qp, .reg_user_mr = c4iw_reg_user_mr, .req_notify_cq = c4iw_arm_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, c4iw_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, c4iw_ucontext, ibucontext), }; void c4iw_register_device(struct work_struct *work) @@ -613,7 +593,7 @@ void c4iw_register_device(struct work_struct *work) dev->ibdev.dev.parent = &dev->rdev.lldi.pdev->dev; dev->ibdev.uverbs_abi_ver = C4IW_UVERBS_ABI_VERSION; - dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL); + dev->ibdev.iwcm = kzalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL); if (!dev->ibdev.iwcm) { ret = -ENOMEM; goto err_dealloc_ctx; @@ -627,14 +607,13 @@ void c4iw_register_device(struct work_struct *work) dev->ibdev.iwcm->add_ref = c4iw_qp_add_ref; dev->ibdev.iwcm->rem_ref = c4iw_qp_rem_ref; dev->ibdev.iwcm->get_qp = c4iw_get_qp; - dev->ibdev.res.fill_res_entry = fill_res_entry; memcpy(dev->ibdev.iwcm->ifname, dev->rdev.lldi.ports[0]->name, sizeof(dev->ibdev.iwcm->ifname)); rdma_set_device_sysfs_group(&dev->ibdev, &c4iw_attr_group); dev->ibdev.driver_id = RDMA_DRIVER_CXGB4; ib_set_device_ops(&dev->ibdev, &c4iw_dev_ops); - ret = ib_register_device(&dev->ibdev, "cxgb4_%d", NULL); + ret = ib_register_device(&dev->ibdev, "cxgb4_%d"); if (ret) goto err_kfree_iwcm; return; diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c index 504cf525508f..d3a82839f5ea 100644 --- a/drivers/infiniband/hw/cxgb4/qp.c +++ b/drivers/infiniband/hw/cxgb4/qp.c @@ -31,6 +31,7 @@ */ #include <linux/module.h> +#include <rdma/uverbs_ioctl.h> #include "iw_cxgb4.h" @@ -632,7 +633,10 @@ static void build_rdma_write_cmpl(struct t4_sq *sq, wcwr->stag_sink = cpu_to_be32(rdma_wr(wr)->rkey); wcwr->to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr); - wcwr->stag_inv = cpu_to_be32(wr->next->ex.invalidate_rkey); + if (wr->next->opcode == IB_WR_SEND) + wcwr->stag_inv = 0; + else + wcwr->stag_inv = cpu_to_be32(wr->next->ex.invalidate_rkey); wcwr->r2 = 0; wcwr->r3 = 0; @@ -726,7 +730,10 @@ static void post_write_cmpl(struct c4iw_qp *qhp, const struct ib_send_wr *wr) /* SEND_WITH_INV swsqe */ swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx]; - swsqe->opcode = FW_RI_SEND_WITH_INV; + if (wr->next->opcode == IB_WR_SEND) + swsqe->opcode = FW_RI_SEND; + else + swsqe->opcode = FW_RI_SEND_WITH_INV; swsqe->idx = qhp->wq.sq.pidx; swsqe->complete = 0; swsqe->signaled = send_signaled; @@ -897,8 +904,6 @@ static void free_qp_work(struct work_struct *work) destroy_qp(&rhp->rdev, &qhp->wq, ucontext ? &ucontext->uctx : &rhp->rdev.uctx, !qhp->srq); - if (ucontext) - c4iw_put_ucontext(ucontext); c4iw_put_wr_wait(qhp->wr_waitp); kfree(qhp); } @@ -1133,9 +1138,9 @@ int c4iw_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, /* * Fastpath for NVMe-oF target WRITE + SEND_WITH_INV wr chain which is * the response for small NVMEe-oF READ requests. If the chain is - * exactly a WRITE->SEND_WITH_INV and the sgl depths and lengths - * meet the requirements of the fw_ri_write_cmpl_wr work request, - * then build and post the write_cmpl WR. If any of the tests + * exactly a WRITE->SEND_WITH_INV or a WRITE->SEND and the sgl depths + * and lengths meet the requirements of the fw_ri_write_cmpl_wr work + * request, then build and post the write_cmpl WR. If any of the tests * below are not true, then we continue on with the tradtional WRITE * and SEND WRs. */ @@ -1145,7 +1150,8 @@ int c4iw_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, wr && wr->next && !wr->next->next && wr->opcode == IB_WR_RDMA_WRITE && wr->sg_list[0].length && wr->num_sge <= T4_WRITE_CMPL_MAX_SGL && - wr->next->opcode == IB_WR_SEND_WITH_INV && + (wr->next->opcode == IB_WR_SEND || + wr->next->opcode == IB_WR_SEND_WITH_INV) && wr->next->sg_list[0].length == T4_WRITE_CMPL_MAX_CQE && wr->next->num_sge == 1 && num_wrs >= 2) { post_write_cmpl(qhp, wr); @@ -2129,7 +2135,8 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs, struct c4iw_cq *rchp; struct c4iw_create_qp_resp uresp; unsigned int sqsize, rqsize = 0; - struct c4iw_ucontext *ucontext; + struct c4iw_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct c4iw_ucontext, ibucontext); int ret; struct c4iw_mm_entry *sq_key_mm, *rq_key_mm = NULL, *sq_db_key_mm; struct c4iw_mm_entry *rq_db_key_mm = NULL, *ma_sync_key_mm = NULL; @@ -2163,8 +2170,6 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs, if (sqsize < 8) sqsize = 8; - ucontext = udata ? to_c4iw_ucontext(pd->uobject->context) : NULL; - qhp = kzalloc(sizeof(*qhp), GFP_KERNEL); if (!qhp) return ERR_PTR(-ENOMEM); @@ -2331,7 +2336,6 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs, insert_mmap(ucontext, ma_sync_key_mm); } - c4iw_get_ucontext(ucontext); qhp->ucontext = ucontext; } if (!attrs->srq) { @@ -2589,7 +2593,7 @@ static int alloc_srq_queue(struct c4iw_srq *srq, struct c4iw_dev_ucontext *uctx, /* build fw_ri_res_wr */ wr_len = sizeof(*res_wr) + sizeof(*res); - skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL); + skb = alloc_skb(wr_len, GFP_KERNEL); if (!skb) goto err_free_queue; set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0); @@ -2711,7 +2715,8 @@ struct ib_srq *c4iw_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *attrs, rqsize = attrs->attr.max_wr + 1; rqsize = roundup_pow_of_two(max_t(u16, rqsize, 16)); - ucontext = udata ? to_c4iw_ucontext(pd->uobject->context) : NULL; + ucontext = rdma_udata_to_drv_context(udata, struct c4iw_ucontext, + ibucontext); srq = kzalloc(sizeof(*srq), GFP_KERNEL); if (!srq) diff --git a/drivers/infiniband/hw/cxgb4/t4.h b/drivers/infiniband/hw/cxgb4/t4.h index fff6d48d262f..b170817b2741 100644 --- a/drivers/infiniband/hw/cxgb4/t4.h +++ b/drivers/infiniband/hw/cxgb4/t4.h @@ -35,6 +35,7 @@ #include "t4_regs.h" #include "t4_values.h" #include "t4_msg.h" +#include "t4_tcb.h" #include "t4fw_ri_api.h" #define T4_MAX_NUM_PD 65536 diff --git a/drivers/infiniband/hw/hfi1/Makefile b/drivers/infiniband/hw/hfi1/Makefile index 3ce9dc8c3463..4044a8c8dbf4 100644 --- a/drivers/infiniband/hw/hfi1/Makefile +++ b/drivers/infiniband/hw/hfi1/Makefile @@ -24,6 +24,7 @@ hfi1-y := \ mad.o \ mmu_rb.o \ msix.o \ + opfn.o \ pcie.o \ pio.o \ pio_copy.o \ diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c index b443642eac02..612f04190ed8 100644 --- a/drivers/infiniband/hw/hfi1/chip.c +++ b/drivers/infiniband/hw/hfi1/chip.c @@ -4253,6 +4253,8 @@ static struct cntr_entry dev_cntrs[DEV_CNTR_LAST] = { access_sw_pio_drain), [C_SW_KMEM_WAIT] = CNTR_ELEM("KmemWait", 0, 0, CNTR_NORMAL, access_sw_kmem_wait), +[C_SW_TID_WAIT] = CNTR_ELEM("TidWait", 0, 0, CNTR_NORMAL, + hfi1_access_sw_tid_wait), [C_SW_SEND_SCHED] = CNTR_ELEM("SendSched", 0, 0, CNTR_NORMAL, access_sw_send_schedule), [C_SDMA_DESC_FETCHED_CNT] = CNTR_ELEM("SDEDscFdCn", @@ -5222,6 +5224,17 @@ int is_bx(struct hfi1_devdata *dd) return (chip_rev_minor & 0xF0) == 0x10; } +/* return true is kernel urg disabled for rcd */ +bool is_urg_masked(struct hfi1_ctxtdata *rcd) +{ + u64 mask; + u32 is = IS_RCVURGENT_START + rcd->ctxt; + u8 bit = is % 64; + + mask = read_csr(rcd->dd, CCE_INT_MASK + (8 * (is / 64))); + return !(mask & BIT_ULL(bit)); +} + /* * Append string s to buffer buf. Arguments curp and len are the current * position and remaining length, respectively. diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h index 6b9c8f12dff8..6c27c1c6a868 100644 --- a/drivers/infiniband/hw/hfi1/chip.h +++ b/drivers/infiniband/hw/hfi1/chip.h @@ -1,7 +1,7 @@ #ifndef _CHIP_H #define _CHIP_H /* - * Copyright(c) 2015 - 2017 Intel Corporation. + * Copyright(c) 2015 - 2018 Intel Corporation. * * This file is provided under a dual BSD/GPLv2 license. When using or * redistributing this file, you may do so under either license. @@ -804,6 +804,7 @@ void clear_linkup_counters(struct hfi1_devdata *dd); u32 hdrqempty(struct hfi1_ctxtdata *rcd); int is_ax(struct hfi1_devdata *dd); int is_bx(struct hfi1_devdata *dd); +bool is_urg_masked(struct hfi1_ctxtdata *rcd); u32 read_physical_state(struct hfi1_devdata *dd); u32 chip_to_opa_pstate(struct hfi1_devdata *dd, u32 chip_pstate); const char *opa_lstate_name(u32 lstate); @@ -926,6 +927,7 @@ enum { C_SW_PIO_WAIT, C_SW_PIO_DRAIN, C_SW_KMEM_WAIT, + C_SW_TID_WAIT, C_SW_SEND_SCHED, C_SDMA_DESC_FETCHED_CNT, C_SDMA_INT_CNT, diff --git a/drivers/infiniband/hw/hfi1/common.h b/drivers/infiniband/hw/hfi1/common.h index 40d3cfb58bd1..7310a5dba420 100644 --- a/drivers/infiniband/hw/hfi1/common.h +++ b/drivers/infiniband/hw/hfi1/common.h @@ -340,6 +340,10 @@ struct diag_pkt { #define HFI1_PSM_IOC_BASE_SEQ 0x0 +/* Number of BTH.PSN bits used for sequence number in expected rcvs */ +#define HFI1_KDETH_BTH_SEQ_SHIFT 11 +#define HFI1_KDETH_BTH_SEQ_MASK (BIT(HFI1_KDETH_BTH_SEQ_SHIFT) - 1) + static inline __u64 rhf_to_cpu(const __le32 *rbuf) { return __le64_to_cpu(*((__le64 *)rbuf)); diff --git a/drivers/infiniband/hw/hfi1/debugfs.c b/drivers/infiniband/hw/hfi1/debugfs.c index 0a557795563c..427ba0ce74a5 100644 --- a/drivers/infiniband/hw/hfi1/debugfs.c +++ b/drivers/infiniband/hw/hfi1/debugfs.c @@ -1167,6 +1167,7 @@ void hfi1_dbg_ibdev_init(struct hfi1_ibdev *ibd) char link[10]; struct hfi1_devdata *dd = dd_from_dev(ibd); struct hfi1_pportdata *ppd; + struct dentry *root; int unit = dd->unit; int i, j; @@ -1174,31 +1175,29 @@ void hfi1_dbg_ibdev_init(struct hfi1_ibdev *ibd) return; snprintf(name, sizeof(name), "%s_%d", class_name(), unit); snprintf(link, sizeof(link), "%d", unit); - ibd->hfi1_ibdev_dbg = debugfs_create_dir(name, hfi1_dbg_root); - if (!ibd->hfi1_ibdev_dbg) { - pr_warn("create of %s failed\n", name); - return; - } + root = debugfs_create_dir(name, hfi1_dbg_root); + ibd->hfi1_ibdev_dbg = root; + ibd->hfi1_ibdev_link = debugfs_create_symlink(link, hfi1_dbg_root, name); - if (!ibd->hfi1_ibdev_link) { - pr_warn("create of %s symlink failed\n", name); - return; - } - DEBUGFS_SEQ_FILE_CREATE(opcode_stats, ibd->hfi1_ibdev_dbg, ibd); - DEBUGFS_SEQ_FILE_CREATE(tx_opcode_stats, ibd->hfi1_ibdev_dbg, ibd); - DEBUGFS_SEQ_FILE_CREATE(ctx_stats, ibd->hfi1_ibdev_dbg, ibd); - DEBUGFS_SEQ_FILE_CREATE(qp_stats, ibd->hfi1_ibdev_dbg, ibd); - DEBUGFS_SEQ_FILE_CREATE(sdes, ibd->hfi1_ibdev_dbg, ibd); - DEBUGFS_SEQ_FILE_CREATE(rcds, ibd->hfi1_ibdev_dbg, ibd); - DEBUGFS_SEQ_FILE_CREATE(pios, ibd->hfi1_ibdev_dbg, ibd); - DEBUGFS_SEQ_FILE_CREATE(sdma_cpu_list, ibd->hfi1_ibdev_dbg, ibd); + + debugfs_create_file("opcode_stats", 0444, root, ibd, + &_opcode_stats_file_ops); + debugfs_create_file("tx_opcode_stats", 0444, root, ibd, + &_tx_opcode_stats_file_ops); + debugfs_create_file("ctx_stats", 0444, root, ibd, &_ctx_stats_file_ops); + debugfs_create_file("qp_stats", 0444, root, ibd, &_qp_stats_file_ops); + debugfs_create_file("sdes", 0444, root, ibd, &_sdes_file_ops); + debugfs_create_file("rcds", 0444, root, ibd, &_rcds_file_ops); + debugfs_create_file("pios", 0444, root, ibd, &_pios_file_ops); + debugfs_create_file("sdma_cpu_list", 0444, root, ibd, + &_sdma_cpu_list_file_ops); + /* dev counter files */ for (i = 0; i < ARRAY_SIZE(cntr_ops); i++) - DEBUGFS_FILE_CREATE(cntr_ops[i].name, - ibd->hfi1_ibdev_dbg, - dd, - &cntr_ops[i].ops, S_IRUGO); + debugfs_create_file(cntr_ops[i].name, 0444, root, dd, + &cntr_ops[i].ops); + /* per port files */ for (ppd = dd->pport, j = 0; j < dd->num_pports; j++, ppd++) for (i = 0; i < ARRAY_SIZE(port_cntr_ops); i++) { @@ -1206,12 +1205,11 @@ void hfi1_dbg_ibdev_init(struct hfi1_ibdev *ibd) sizeof(name), port_cntr_ops[i].name, j + 1); - DEBUGFS_FILE_CREATE(name, - ibd->hfi1_ibdev_dbg, - ppd, - &port_cntr_ops[i].ops, + debugfs_create_file(name, !port_cntr_ops[i].ops.write ? - S_IRUGO : S_IRUGO | S_IWUSR); + S_IRUGO : + S_IRUGO | S_IWUSR, + root, ppd, &port_cntr_ops[i].ops); } hfi1_fault_init_debugfs(ibd); @@ -1341,10 +1339,10 @@ DEBUGFS_FILE_OPS(driver_stats); void hfi1_dbg_init(void) { hfi1_dbg_root = debugfs_create_dir(DRIVER_NAME, NULL); - if (!hfi1_dbg_root) - pr_warn("init of debugfs failed\n"); - DEBUGFS_SEQ_FILE_CREATE(driver_stats_names, hfi1_dbg_root, NULL); - DEBUGFS_SEQ_FILE_CREATE(driver_stats, hfi1_dbg_root, NULL); + debugfs_create_file("driver_stats_names", 0444, hfi1_dbg_root, NULL, + &_driver_stats_names_file_ops); + debugfs_create_file("driver_stats", 0444, hfi1_dbg_root, NULL, + &_driver_stats_file_ops); } void hfi1_dbg_exit(void) diff --git a/drivers/infiniband/hw/hfi1/debugfs.h b/drivers/infiniband/hw/hfi1/debugfs.h index d5d824459fcc..57e582caa5eb 100644 --- a/drivers/infiniband/hw/hfi1/debugfs.h +++ b/drivers/infiniband/hw/hfi1/debugfs.h @@ -49,16 +49,6 @@ struct hfi1_ibdev; -#define DEBUGFS_FILE_CREATE(name, parent, data, ops, mode) \ -do { \ - struct dentry *ent; \ - const char *__name = name; \ - ent = debugfs_create_file(__name, mode, parent, \ - data, ops); \ - if (!ent) \ - pr_warn("create of %s failed\n", __name); \ -} while (0) - #define DEBUGFS_SEQ_FILE_OPS(name) \ static const struct seq_operations _##name##_seq_ops = { \ .start = _##name##_seq_start, \ @@ -89,8 +79,6 @@ static const struct file_operations _##name##_file_ops = { \ .release = seq_release \ } -#define DEBUGFS_SEQ_FILE_CREATE(name, parent, data) \ - DEBUGFS_FILE_CREATE(#name, parent, data, &_##name##_file_ops, 0444) ssize_t hfi1_seq_read(struct file *file, char __user *buf, size_t size, loff_t *ppos); diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c index a8ad70730203..2a9d2912f5db 100644 --- a/drivers/infiniband/hw/hfi1/driver.c +++ b/drivers/infiniband/hw/hfi1/driver.c @@ -1575,25 +1575,32 @@ drop: return -EINVAL; } -void handle_eflags(struct hfi1_packet *packet) +static void show_eflags_errs(struct hfi1_packet *packet) { struct hfi1_ctxtdata *rcd = packet->rcd; u32 rte = rhf_rcv_type_err(packet->rhf); + dd_dev_err(rcd->dd, + "receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s%s] rte 0x%x\n", + rcd->ctxt, packet->rhf, + packet->rhf & RHF_K_HDR_LEN_ERR ? "k_hdr_len " : "", + packet->rhf & RHF_DC_UNC_ERR ? "dc_unc " : "", + packet->rhf & RHF_DC_ERR ? "dc " : "", + packet->rhf & RHF_TID_ERR ? "tid " : "", + packet->rhf & RHF_LEN_ERR ? "len " : "", + packet->rhf & RHF_ECC_ERR ? "ecc " : "", + packet->rhf & RHF_VCRC_ERR ? "vcrc " : "", + packet->rhf & RHF_ICRC_ERR ? "icrc " : "", + rte); +} + +void handle_eflags(struct hfi1_packet *packet) +{ + struct hfi1_ctxtdata *rcd = packet->rcd; + rcv_hdrerr(rcd, rcd->ppd, packet); if (rhf_err_flags(packet->rhf)) - dd_dev_err(rcd->dd, - "receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s%s] rte 0x%x\n", - rcd->ctxt, packet->rhf, - packet->rhf & RHF_K_HDR_LEN_ERR ? "k_hdr_len " : "", - packet->rhf & RHF_DC_UNC_ERR ? "dc_unc " : "", - packet->rhf & RHF_DC_ERR ? "dc " : "", - packet->rhf & RHF_TID_ERR ? "tid " : "", - packet->rhf & RHF_LEN_ERR ? "len " : "", - packet->rhf & RHF_ECC_ERR ? "ecc " : "", - packet->rhf & RHF_VCRC_ERR ? "vcrc " : "", - packet->rhf & RHF_ICRC_ERR ? "icrc " : "", - rte); + show_eflags_errs(packet); } /* @@ -1699,11 +1706,14 @@ static int kdeth_process_expected(struct hfi1_packet *packet) if (unlikely(hfi1_dbg_should_fault_rx(packet))) return RHF_RCV_CONTINUE; - if (unlikely(rhf_err_flags(packet->rhf))) - handle_eflags(packet); + if (unlikely(rhf_err_flags(packet->rhf))) { + struct hfi1_ctxtdata *rcd = packet->rcd; - dd_dev_err(packet->rcd->dd, - "Unhandled expected packet received. Dropping.\n"); + if (hfi1_handle_kdeth_eflags(rcd, rcd->ppd, packet)) + return RHF_RCV_CONTINUE; + } + + hfi1_kdeth_expected_rcv(packet); return RHF_RCV_CONTINUE; } @@ -1712,11 +1722,17 @@ static int kdeth_process_eager(struct hfi1_packet *packet) hfi1_setup_9B_packet(packet); if (unlikely(hfi1_dbg_should_fault_rx(packet))) return RHF_RCV_CONTINUE; - if (unlikely(rhf_err_flags(packet->rhf))) - handle_eflags(packet); - dd_dev_err(packet->rcd->dd, - "Unhandled eager packet received. Dropping.\n"); + trace_hfi1_rcvhdr(packet); + if (unlikely(rhf_err_flags(packet->rhf))) { + struct hfi1_ctxtdata *rcd = packet->rcd; + + show_eflags_errs(packet); + if (hfi1_handle_kdeth_eflags(rcd, rcd->ppd, packet)) + return RHF_RCV_CONTINUE; + } + + hfi1_kdeth_eager_rcv(packet); return RHF_RCV_CONTINUE; } diff --git a/drivers/infiniband/hw/hfi1/fault.c b/drivers/infiniband/hw/hfi1/fault.c index e2290f32c8d9..3fd3315d0fb0 100644 --- a/drivers/infiniband/hw/hfi1/fault.c +++ b/drivers/infiniband/hw/hfi1/fault.c @@ -250,6 +250,7 @@ void hfi1_fault_exit_debugfs(struct hfi1_ibdev *ibd) int hfi1_fault_init_debugfs(struct hfi1_ibdev *ibd) { struct dentry *parent = ibd->hfi1_ibdev_dbg; + struct dentry *fault_dir; ibd->fault = kzalloc(sizeof(*ibd->fault), GFP_KERNEL); if (!ibd->fault) @@ -269,45 +270,31 @@ int hfi1_fault_init_debugfs(struct hfi1_ibdev *ibd) bitmap_zero(ibd->fault->opcodes, sizeof(ibd->fault->opcodes) * BITS_PER_BYTE); - ibd->fault->dir = - fault_create_debugfs_attr("fault", parent, - &ibd->fault->attr); - if (IS_ERR(ibd->fault->dir)) { + fault_dir = + fault_create_debugfs_attr("fault", parent, &ibd->fault->attr); + if (IS_ERR(fault_dir)) { kfree(ibd->fault); ibd->fault = NULL; return -ENOENT; } - - DEBUGFS_SEQ_FILE_CREATE(fault_stats, ibd->fault->dir, ibd); - if (!debugfs_create_bool("enable", 0600, ibd->fault->dir, - &ibd->fault->enable)) - goto fail; - if (!debugfs_create_bool("suppress_err", 0600, - ibd->fault->dir, - &ibd->fault->suppress_err)) - goto fail; - if (!debugfs_create_bool("opcode_mode", 0600, ibd->fault->dir, - &ibd->fault->opcode)) - goto fail; - if (!debugfs_create_file("opcodes", 0600, ibd->fault->dir, - ibd->fault, &__fault_opcodes_fops)) - goto fail; - if (!debugfs_create_u64("skip_pkts", 0600, - ibd->fault->dir, - &ibd->fault->fault_skip)) - goto fail; - if (!debugfs_create_u64("skip_usec", 0600, - ibd->fault->dir, - &ibd->fault->fault_skip_usec)) - goto fail; - if (!debugfs_create_u8("direction", 0600, ibd->fault->dir, - &ibd->fault->direction)) - goto fail; + ibd->fault->dir = fault_dir; + + debugfs_create_file("fault_stats", 0444, fault_dir, ibd, + &_fault_stats_file_ops); + debugfs_create_bool("enable", 0600, fault_dir, &ibd->fault->enable); + debugfs_create_bool("suppress_err", 0600, fault_dir, + &ibd->fault->suppress_err); + debugfs_create_bool("opcode_mode", 0600, fault_dir, + &ibd->fault->opcode); + debugfs_create_file("opcodes", 0600, fault_dir, ibd->fault, + &__fault_opcodes_fops); + debugfs_create_u64("skip_pkts", 0600, fault_dir, + &ibd->fault->fault_skip); + debugfs_create_u64("skip_usec", 0600, fault_dir, + &ibd->fault->fault_skip_usec); + debugfs_create_u8("direction", 0600, fault_dir, &ibd->fault->direction); return 0; -fail: - hfi1_fault_exit_debugfs(ibd); - return -ENOMEM; } bool hfi1_dbg_fault_suppress_err(struct hfi1_ibdev *ibd) diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h index 6db2276f5c13..048b5d73ba39 100644 --- a/drivers/infiniband/hw/hfi1/hfi.h +++ b/drivers/infiniband/hw/hfi1/hfi.h @@ -73,6 +73,7 @@ #include "chip_registers.h" #include "common.h" +#include "opfn.h" #include "verbs.h" #include "pio.h" #include "chip.h" @@ -98,6 +99,8 @@ #define NEIGHBOR_TYPE_HFI 0 #define NEIGHBOR_TYPE_SWITCH 1 +#define HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES 5 + extern unsigned long hfi1_cap_mask; #define HFI1_CAP_KGET_MASK(mask, cap) ((mask) & HFI1_CAP_##cap) #define HFI1_CAP_UGET_MASK(mask, cap) \ @@ -195,6 +198,14 @@ struct exp_tid_set { }; typedef int (*rhf_rcv_function_ptr)(struct hfi1_packet *packet); + +struct tid_queue { + struct list_head queue_head; + /* queue head for QP TID resource waiters */ + u32 enqueue; /* count of tid enqueues */ + u32 dequeue; /* count of tid dequeues */ +}; + struct hfi1_ctxtdata { /* rcvhdrq base, needs mmap before useful */ void *rcvhdrq; @@ -288,6 +299,12 @@ struct hfi1_ctxtdata { /* PSM Specific fields */ /* lock protecting all Expected TID data */ struct mutex exp_mutex; + /* lock protecting all Expected TID data of kernel contexts */ + spinlock_t exp_lock; + /* Queue for QP's waiting for HW TID flows */ + struct tid_queue flow_queue; + /* Queue for QP's waiting for HW receive array entries */ + struct tid_queue rarr_queue; /* when waiting for rcv or pioavail */ wait_queue_head_t wait; /* uuid from PSM */ @@ -320,6 +337,9 @@ struct hfi1_ctxtdata { */ u8 subctxt_cnt; + /* Bit mask to track free TID RDMA HW flows */ + unsigned long flow_mask; + struct tid_flow_state flows[RXE_NUM_TID_FLOWS]; }; /** @@ -1435,7 +1455,7 @@ void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd, struct hfi1_devdata *dd, u8 hw_pidx, u8 port); void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd); int hfi1_rcd_put(struct hfi1_ctxtdata *rcd); -void hfi1_rcd_get(struct hfi1_ctxtdata *rcd); +int hfi1_rcd_get(struct hfi1_ctxtdata *rcd); struct hfi1_ctxtdata *hfi1_rcd_get_by_index_safe(struct hfi1_devdata *dd, u16 ctxt); struct hfi1_ctxtdata *hfi1_rcd_get_by_index(struct hfi1_devdata *dd, u16 ctxt); @@ -2100,7 +2120,7 @@ static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd, SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK | #endif HFI1_PKT_USER_SC_INTEGRITY; - else + else if (ctxt_type != SC_KERNEL) base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY; /* turn on send-side job key checks if !A0 */ diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c index 441b06e2a154..faaaac8fbc55 100644 --- a/drivers/infiniband/hw/hfi1/init.c +++ b/drivers/infiniband/hw/hfi1/init.c @@ -73,7 +73,6 @@ #undef pr_fmt #define pr_fmt(fmt) DRIVER_NAME ": " fmt -#define HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES 5 /* * min buffers we want to have per context, after driver */ @@ -216,12 +215,12 @@ static void hfi1_rcd_free(struct kref *kref) struct hfi1_ctxtdata *rcd = container_of(kref, struct hfi1_ctxtdata, kref); - hfi1_free_ctxtdata(rcd->dd, rcd); - spin_lock_irqsave(&rcd->dd->uctxt_lock, flags); rcd->dd->rcd[rcd->ctxt] = NULL; spin_unlock_irqrestore(&rcd->dd->uctxt_lock, flags); + hfi1_free_ctxtdata(rcd->dd, rcd); + kfree(rcd); } @@ -244,10 +243,13 @@ int hfi1_rcd_put(struct hfi1_ctxtdata *rcd) * @rcd: pointer to an initialized rcd data structure * * Use this to get a reference after the init. + * + * Return : reflect kref_get_unless_zero(), which returns non-zero on + * increment, otherwise 0. */ -void hfi1_rcd_get(struct hfi1_ctxtdata *rcd) +int hfi1_rcd_get(struct hfi1_ctxtdata *rcd) { - kref_get(&rcd->kref); + return kref_get_unless_zero(&rcd->kref); } /** @@ -327,7 +329,8 @@ struct hfi1_ctxtdata *hfi1_rcd_get_by_index(struct hfi1_devdata *dd, u16 ctxt) spin_lock_irqsave(&dd->uctxt_lock, flags); if (dd->rcd[ctxt]) { rcd = dd->rcd[ctxt]; - hfi1_rcd_get(rcd); + if (!hfi1_rcd_get(rcd)) + rcd = NULL; } spin_unlock_irqrestore(&dd->uctxt_lock, flags); @@ -372,6 +375,9 @@ int hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, int numa, rcd->rhf_rcv_function_map = normal_rhf_rcv_functions; mutex_init(&rcd->exp_mutex); + spin_lock_init(&rcd->exp_lock); + INIT_LIST_HEAD(&rcd->flow_queue.queue_head); + INIT_LIST_HEAD(&rcd->rarr_queue.queue_head); hfi1_cdbg(PROC, "setting up context %u\n", rcd->ctxt); @@ -474,6 +480,9 @@ int hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, int numa, GFP_KERNEL, numa); if (!rcd->opstats) goto bail; + + /* Initialize TID flow generations for the context */ + hfi1_kern_init_ctxt_generations(rcd); } *context = rcd; @@ -773,6 +782,8 @@ static void enable_chip(struct hfi1_devdata *dd) rcvmask |= HFI1_RCVCTRL_NO_RHQ_DROP_ENB; if (HFI1_CAP_KGET_MASK(rcd->flags, NODROP_EGR_FULL)) rcvmask |= HFI1_RCVCTRL_NO_EGR_DROP_ENB; + if (HFI1_CAP_IS_KSET(TID_RDMA)) + rcvmask |= HFI1_RCVCTRL_TIDFLOW_ENB; hfi1_rcvctrl(dd, rcvmask, rcd); sc_enable(rcd->sc); hfi1_rcd_put(rcd); @@ -928,6 +939,8 @@ int hfi1_init(struct hfi1_devdata *dd, int reinit) lastfail = hfi1_create_rcvhdrq(dd, rcd); if (!lastfail) lastfail = hfi1_setup_eagerbufs(rcd); + if (!lastfail) + lastfail = hfi1_kern_exp_rcv_init(rcd, reinit); if (lastfail) { dd_dev_err(dd, "failed to allocate kernel ctxt's rcvhdrq and/or egr bufs\n"); @@ -1498,6 +1511,13 @@ static int __init hfi1_mod_init(void) /* sanitize link CRC options */ link_crc_mask &= SUPPORTED_CRCS; + ret = opfn_init(); + if (ret < 0) { + pr_err("Failed to allocate opfn_wq"); + goto bail_dev; + } + + hfi1_compute_tid_rdma_flow_wt(); /* * These must be called before the driver is registered with * the PCI subsystem. @@ -1528,6 +1548,7 @@ module_init(hfi1_mod_init); static void __exit hfi1_mod_cleanup(void) { pci_unregister_driver(&hfi1_pci_driver); + opfn_exit(); node_affinity_destroy_all(); hfi1_dbg_exit(); @@ -1582,7 +1603,7 @@ static void cleanup_device_data(struct hfi1_devdata *dd) struct hfi1_ctxtdata *rcd = dd->rcd[ctxt]; if (rcd) { - hfi1_clear_tids(rcd); + hfi1_free_ctxt_rcv_groups(rcd); hfi1_free_ctxt(rcd); } } diff --git a/drivers/infiniband/hw/hfi1/iowait.c b/drivers/infiniband/hw/hfi1/iowait.c index 582f1ba136ff..adb4a1ba921b 100644 --- a/drivers/infiniband/hw/hfi1/iowait.c +++ b/drivers/infiniband/hw/hfi1/iowait.c @@ -6,6 +6,9 @@ #include "iowait.h" #include "trace_iowait.h" +/* 1 priority == 16 starve_cnt */ +#define IOWAIT_PRIORITY_STARVE_SHIFT 4 + void iowait_set_flag(struct iowait *wait, u32 flag) { trace_hfi1_iowait_set(wait, flag); @@ -44,7 +47,8 @@ void iowait_init(struct iowait *wait, u32 tx_limit, uint seq, bool pkts_sent), void (*wakeup)(struct iowait *wait, int reason), - void (*sdma_drained)(struct iowait *wait)) + void (*sdma_drained)(struct iowait *wait), + void (*init_priority)(struct iowait *wait)) { int i; @@ -58,6 +62,7 @@ void iowait_init(struct iowait *wait, u32 tx_limit, wait->sleep = sleep; wait->wakeup = wakeup; wait->sdma_drained = sdma_drained; + wait->init_priority = init_priority; wait->flags = 0; for (i = 0; i < IOWAIT_SES; i++) { wait->wait[i].iow = wait; @@ -92,3 +97,30 @@ int iowait_set_work_flag(struct iowait_work *w) iowait_set_flag(w->iow, IOWAIT_PENDING_TID); return IOWAIT_TID_SE; } + +/** + * iowait_priority_update_top - update the top priority entry + * @w: the iowait struct + * @top: a pointer to the top priority entry + * @idx: the index of the current iowait in an array + * @top_idx: the array index for the iowait entry that has the top priority + * + * This function is called to compare the priority of a given + * iowait with the given top priority entry. The top index will + * be returned. + */ +uint iowait_priority_update_top(struct iowait *w, + struct iowait *top, + uint idx, uint top_idx) +{ + u8 cnt, tcnt; + + /* Convert priority into starve_cnt and compare the total.*/ + cnt = (w->priority << IOWAIT_PRIORITY_STARVE_SHIFT) + w->starved_cnt; + tcnt = (top->priority << IOWAIT_PRIORITY_STARVE_SHIFT) + + top->starved_cnt; + if (cnt > tcnt) + return idx; + else + return top_idx; +} diff --git a/drivers/infiniband/hw/hfi1/iowait.h b/drivers/infiniband/hw/hfi1/iowait.h index 23a58ac0d47c..07847cb72169 100644 --- a/drivers/infiniband/hw/hfi1/iowait.h +++ b/drivers/infiniband/hw/hfi1/iowait.h @@ -100,6 +100,7 @@ struct iowait_work { * @sleep: no space callback * @wakeup: space callback wakeup * @sdma_drained: sdma count drained + * @init_priority: callback to manipulate priority * @lock: lock protected head of wait queue * @iowork: workqueue overhead * @wait_dma: wait for sdma_busy == 0 @@ -109,7 +110,7 @@ struct iowait_work { * @tx_limit: limit for overflow queuing * @tx_count: number of tx entry's in tx_head'ed list * @flags: wait flags (one per QP) - * @wait: SE array + * @wait: SE array for multiple legs * * This is to be embedded in user's state structure * (QP or PQ). @@ -120,10 +121,13 @@ struct iowait_work { * are callbacks for the ULP to implement * what ever queuing/dequeuing of * the embedded iowait and its containing struct - * when a resource shortage like SDMA ring space is seen. + * when a resource shortage like SDMA ring space + * or PIO credit space is seen. * * Both potentially have locks help - * so sleeping is not allowed. + * so sleeping is not allowed and it is not + * supported to submit txreqs from the wakeup + * call directly because of lock conflicts. * * The wait_dma member along with the iow * @@ -143,6 +147,7 @@ struct iowait { ); void (*wakeup)(struct iowait *wait, int reason); void (*sdma_drained)(struct iowait *wait); + void (*init_priority)(struct iowait *wait); seqlock_t *lock; wait_queue_head_t wait_dma; wait_queue_head_t wait_pio; @@ -152,6 +157,7 @@ struct iowait { u32 tx_limit; u32 tx_count; u8 starved_cnt; + u8 priority; unsigned long flags; struct iowait_work wait[IOWAIT_SES]; }; @@ -171,7 +177,8 @@ void iowait_init(struct iowait *wait, u32 tx_limit, uint seq, bool pkts_sent), void (*wakeup)(struct iowait *wait, int reason), - void (*sdma_drained)(struct iowait *wait)); + void (*sdma_drained)(struct iowait *wait), + void (*init_priority)(struct iowait *wait)); /** * iowait_schedule() - schedule the default send engine work @@ -186,6 +193,18 @@ static inline bool iowait_schedule(struct iowait *wait, } /** + * iowait_tid_schedule - schedule the tid SE + * @wait: the iowait structure + * @wq: the work queue + * @cpu: the cpu + */ +static inline bool iowait_tid_schedule(struct iowait *wait, + struct workqueue_struct *wq, int cpu) +{ + return !!queue_work_on(cpu, wq, &wait->wait[IOWAIT_TID_SE].iowork); +} + +/** * iowait_sdma_drain() - wait for DMAs to drain * * @wait: iowait structure @@ -327,6 +346,8 @@ static inline u16 iowait_get_desc(struct iowait_work *w) tx = list_first_entry(&w->tx_head, struct sdma_txreq, list); num_desc = tx->num_desc; + if (tx->flags & SDMA_TXREQ_F_VIP) + w->iow->priority++; } return num_desc; } @@ -340,6 +361,37 @@ static inline u32 iowait_get_all_desc(struct iowait *w) return num_desc; } +static inline void iowait_update_priority(struct iowait_work *w) +{ + struct sdma_txreq *tx = NULL; + + if (!list_empty(&w->tx_head)) { + tx = list_first_entry(&w->tx_head, struct sdma_txreq, + list); + if (tx->flags & SDMA_TXREQ_F_VIP) + w->iow->priority++; + } +} + +static inline void iowait_update_all_priority(struct iowait *w) +{ + iowait_update_priority(&w->wait[IOWAIT_IB_SE]); + iowait_update_priority(&w->wait[IOWAIT_TID_SE]); +} + +static inline void iowait_init_priority(struct iowait *w) +{ + w->priority = 0; + if (w->init_priority) + w->init_priority(w); +} + +static inline void iowait_get_priority(struct iowait *w) +{ + iowait_init_priority(w); + iowait_update_all_priority(w); +} + /** * iowait_queue - Put the iowait on a wait queue * @pkts_sent: have some packets been sent before queuing? @@ -356,14 +408,18 @@ static inline void iowait_queue(bool pkts_sent, struct iowait *w, /* * To play fair, insert the iowait at the tail of the wait queue if it * has already sent some packets; Otherwise, put it at the head. + * However, if it has priority packets to send, also put it at the + * head. */ - if (pkts_sent) { - list_add_tail(&w->list, wait_head); + if (pkts_sent) w->starved_cnt = 0; - } else { - list_add(&w->list, wait_head); + else w->starved_cnt++; - } + + if (w->priority > 0 || !pkts_sent) + list_add(&w->list, wait_head); + else + list_add_tail(&w->list, wait_head); } /** @@ -380,27 +436,10 @@ static inline void iowait_starve_clear(bool pkts_sent, struct iowait *w) w->starved_cnt = 0; } -/** - * iowait_starve_find_max - Find the maximum of the starve count - * @w: the iowait struct - * @max: a variable containing the max starve count - * @idx: the index of the current iowait in an array - * @max_idx: a variable containing the array index for the - * iowait entry that has the max starve count - * - * This function is called to compare the starve count of a - * given iowait with the given max starve count. The max starve - * count and the index will be updated if the iowait's start - * count is larger. - */ -static inline void iowait_starve_find_max(struct iowait *w, u8 *max, - uint idx, uint *max_idx) -{ - if (w->starved_cnt > *max) { - *max = w->starved_cnt; - *max_idx = idx; - } -} +/* Update the top priority index */ +uint iowait_priority_update_top(struct iowait *w, + struct iowait *top, + uint idx, uint top_idx); /** * iowait_packet_queued() - determine if a packet is queued diff --git a/drivers/infiniband/hw/hfi1/opfn.c b/drivers/infiniband/hw/hfi1/opfn.c new file mode 100644 index 000000000000..370a5a8eaa71 --- /dev/null +++ b/drivers/infiniband/hw/hfi1/opfn.c @@ -0,0 +1,323 @@ +// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) +/* + * Copyright(c) 2018 Intel Corporation. + * + */ +#include "hfi.h" +#include "trace.h" +#include "qp.h" +#include "opfn.h" + +#define IB_BTHE_E BIT(IB_BTHE_E_SHIFT) + +#define OPFN_CODE(code) BIT((code) - 1) +#define OPFN_MASK(code) OPFN_CODE(STL_VERBS_EXTD_##code) + +struct hfi1_opfn_type { + bool (*request)(struct rvt_qp *qp, u64 *data); + bool (*response)(struct rvt_qp *qp, u64 *data); + bool (*reply)(struct rvt_qp *qp, u64 data); + void (*error)(struct rvt_qp *qp); +}; + +static struct hfi1_opfn_type hfi1_opfn_handlers[STL_VERBS_EXTD_MAX] = { + [STL_VERBS_EXTD_TID_RDMA] = { + .request = tid_rdma_conn_req, + .response = tid_rdma_conn_resp, + .reply = tid_rdma_conn_reply, + .error = tid_rdma_conn_error, + }, +}; + +static struct workqueue_struct *opfn_wq; + +static void opfn_schedule_conn_request(struct rvt_qp *qp); + +static bool hfi1_opfn_extended(u32 bth1) +{ + return !!(bth1 & IB_BTHE_E); +} + +static void opfn_conn_request(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct ib_atomic_wr wr; + u16 mask, capcode; + struct hfi1_opfn_type *extd; + u64 data; + unsigned long flags; + int ret = 0; + + trace_hfi1_opfn_state_conn_request(qp); + spin_lock_irqsave(&priv->opfn.lock, flags); + /* + * Exit if the extended bit is not set, or if nothing is requested, or + * if we have completed all requests, or if a previous request is in + * progress + */ + if (!priv->opfn.extended || !priv->opfn.requested || + priv->opfn.requested == priv->opfn.completed || priv->opfn.curr) + goto done; + + mask = priv->opfn.requested & ~priv->opfn.completed; + capcode = ilog2(mask & ~(mask - 1)) + 1; + if (capcode >= STL_VERBS_EXTD_MAX) { + priv->opfn.completed |= OPFN_CODE(capcode); + goto done; + } + + extd = &hfi1_opfn_handlers[capcode]; + if (!extd || !extd->request || !extd->request(qp, &data)) { + /* + * Either there is no handler for this capability or the request + * packet could not be generated. Either way, mark it as done so + * we don't keep attempting to complete it. + */ + priv->opfn.completed |= OPFN_CODE(capcode); + goto done; + } + + trace_hfi1_opfn_data_conn_request(qp, capcode, data); + data = (data & ~0xf) | capcode; + + memset(&wr, 0, sizeof(wr)); + wr.wr.opcode = IB_WR_OPFN; + wr.remote_addr = HFI1_VERBS_E_ATOMIC_VADDR; + wr.compare_add = data; + + priv->opfn.curr = capcode; /* A new request is now in progress */ + /* Drop opfn.lock before calling ib_post_send() */ + spin_unlock_irqrestore(&priv->opfn.lock, flags); + + ret = ib_post_send(&qp->ibqp, &wr.wr, NULL); + if (ret) + goto err; + trace_hfi1_opfn_state_conn_request(qp); + return; +err: + trace_hfi1_msg_opfn_conn_request(qp, "ib_ost_send failed: ret = ", + (u64)ret); + spin_lock_irqsave(&priv->opfn.lock, flags); + /* + * In case of an unexpected error return from ib_post_send + * clear opfn.curr and reschedule to try again + */ + priv->opfn.curr = STL_VERBS_EXTD_NONE; + opfn_schedule_conn_request(qp); +done: + spin_unlock_irqrestore(&priv->opfn.lock, flags); +} + +void opfn_send_conn_request(struct work_struct *work) +{ + struct hfi1_opfn_data *od; + struct hfi1_qp_priv *qpriv; + + od = container_of(work, struct hfi1_opfn_data, opfn_work); + qpriv = container_of(od, struct hfi1_qp_priv, opfn); + + opfn_conn_request(qpriv->owner); +} + +/* + * When QP s_lock is held in the caller, the OPFN request must be scheduled + * to a different workqueue to avoid double locking QP s_lock in call to + * ib_post_send in opfn_conn_request + */ +static void opfn_schedule_conn_request(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + + trace_hfi1_opfn_state_sched_conn_request(qp); + queue_work(opfn_wq, &priv->opfn.opfn_work); +} + +void opfn_conn_response(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_atomic_eth *ateth) +{ + struct hfi1_qp_priv *priv = qp->priv; + u64 data = be64_to_cpu(ateth->compare_data); + struct hfi1_opfn_type *extd; + u8 capcode; + unsigned long flags; + + trace_hfi1_opfn_state_conn_response(qp); + capcode = data & 0xf; + trace_hfi1_opfn_data_conn_response(qp, capcode, data); + if (!capcode || capcode >= STL_VERBS_EXTD_MAX) + return; + + extd = &hfi1_opfn_handlers[capcode]; + + if (!extd || !extd->response) { + e->atomic_data = capcode; + return; + } + + spin_lock_irqsave(&priv->opfn.lock, flags); + if (priv->opfn.completed & OPFN_CODE(capcode)) { + /* + * We are receiving a request for a feature that has already + * been negotiated. This may mean that the other side has reset + */ + priv->opfn.completed &= ~OPFN_CODE(capcode); + if (extd->error) + extd->error(qp); + } + + if (extd->response(qp, &data)) + priv->opfn.completed |= OPFN_CODE(capcode); + e->atomic_data = (data & ~0xf) | capcode; + trace_hfi1_opfn_state_conn_response(qp); + spin_unlock_irqrestore(&priv->opfn.lock, flags); +} + +void opfn_conn_reply(struct rvt_qp *qp, u64 data) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct hfi1_opfn_type *extd; + u8 capcode; + unsigned long flags; + + trace_hfi1_opfn_state_conn_reply(qp); + capcode = data & 0xf; + trace_hfi1_opfn_data_conn_reply(qp, capcode, data); + if (!capcode || capcode >= STL_VERBS_EXTD_MAX) + return; + + spin_lock_irqsave(&priv->opfn.lock, flags); + /* + * Either there is no previous request or the reply is not for the + * current request + */ + if (!priv->opfn.curr || capcode != priv->opfn.curr) + goto done; + + extd = &hfi1_opfn_handlers[capcode]; + + if (!extd || !extd->reply) + goto clear; + + if (extd->reply(qp, data)) + priv->opfn.completed |= OPFN_CODE(capcode); +clear: + /* + * Clear opfn.curr to indicate that the previous request is no longer in + * progress + */ + priv->opfn.curr = STL_VERBS_EXTD_NONE; + trace_hfi1_opfn_state_conn_reply(qp); +done: + spin_unlock_irqrestore(&priv->opfn.lock, flags); +} + +void opfn_conn_error(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct hfi1_opfn_type *extd = NULL; + unsigned long flags; + u16 capcode; + + trace_hfi1_opfn_state_conn_error(qp); + trace_hfi1_msg_opfn_conn_error(qp, "error. qp state ", (u64)qp->state); + /* + * The QP has gone into the Error state. We have to invalidate all + * negotiated feature, including the one in progress (if any). The RC + * QP handling will clean the WQE for the connection request. + */ + spin_lock_irqsave(&priv->opfn.lock, flags); + while (priv->opfn.completed) { + capcode = priv->opfn.completed & ~(priv->opfn.completed - 1); + extd = &hfi1_opfn_handlers[ilog2(capcode) + 1]; + if (extd->error) + extd->error(qp); + priv->opfn.completed &= ~OPFN_CODE(capcode); + } + priv->opfn.extended = 0; + priv->opfn.requested = 0; + priv->opfn.curr = STL_VERBS_EXTD_NONE; + spin_unlock_irqrestore(&priv->opfn.lock, flags); +} + +void opfn_qp_init(struct rvt_qp *qp, struct ib_qp_attr *attr, int attr_mask) +{ + struct ib_qp *ibqp = &qp->ibqp; + struct hfi1_qp_priv *priv = qp->priv; + unsigned long flags; + + if (attr_mask & IB_QP_RETRY_CNT) + priv->s_retry = attr->retry_cnt; + + spin_lock_irqsave(&priv->opfn.lock, flags); + if (ibqp->qp_type == IB_QPT_RC && HFI1_CAP_IS_KSET(TID_RDMA)) { + struct tid_rdma_params *local = &priv->tid_rdma.local; + + if (attr_mask & IB_QP_TIMEOUT) + priv->tid_retry_timeout_jiffies = qp->timeout_jiffies; + if (qp->pmtu == enum_to_mtu(OPA_MTU_4096) || + qp->pmtu == enum_to_mtu(OPA_MTU_8192)) { + tid_rdma_opfn_init(qp, local); + /* + * We only want to set the OPFN requested bit when the + * QP transitions to RTS. + */ + if (attr_mask & IB_QP_STATE && + attr->qp_state == IB_QPS_RTS) { + priv->opfn.requested |= OPFN_MASK(TID_RDMA); + /* + * If the QP is transitioning to RTS and the + * opfn.completed for TID RDMA has already been + * set, the QP is being moved *back* into RTS. + * We can now renegotiate the TID RDMA + * parameters. + */ + if (priv->opfn.completed & + OPFN_MASK(TID_RDMA)) { + priv->opfn.completed &= + ~OPFN_MASK(TID_RDMA); + /* + * Since the opfn.completed bit was + * already set, it is safe to assume + * that the opfn.extended is also set. + */ + opfn_schedule_conn_request(qp); + } + } + } else { + memset(local, 0, sizeof(*local)); + } + } + spin_unlock_irqrestore(&priv->opfn.lock, flags); +} + +void opfn_trigger_conn_request(struct rvt_qp *qp, u32 bth1) +{ + struct hfi1_qp_priv *priv = qp->priv; + + if (!priv->opfn.extended && hfi1_opfn_extended(bth1) && + HFI1_CAP_IS_KSET(OPFN)) { + priv->opfn.extended = 1; + if (qp->state == IB_QPS_RTS) + opfn_conn_request(qp); + } +} + +int opfn_init(void) +{ + opfn_wq = alloc_workqueue("hfi_opfn", + WQ_SYSFS | WQ_HIGHPRI | WQ_CPU_INTENSIVE | + WQ_MEM_RECLAIM, + HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES); + if (!opfn_wq) + return -ENOMEM; + + return 0; +} + +void opfn_exit(void) +{ + if (opfn_wq) { + destroy_workqueue(opfn_wq); + opfn_wq = NULL; + } +} diff --git a/drivers/infiniband/hw/hfi1/opfn.h b/drivers/infiniband/hw/hfi1/opfn.h new file mode 100644 index 000000000000..5f2011cabc25 --- /dev/null +++ b/drivers/infiniband/hw/hfi1/opfn.h @@ -0,0 +1,85 @@ +/* SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) */ +/* + * Copyright(c) 2018 Intel Corporation. + * + */ +#ifndef _HFI1_OPFN_H +#define _HFI1_OPFN_H + +/** + * DOC: Omni Path Feature Negotion (OPFN) + * + * OPFN is a discovery protocol for Intel Omni-Path fabric that + * allows two RC QPs to negotiate a common feature that both QPs + * can support. Currently, the only OPA feature that OPFN + * supports is TID RDMA. + * + * Architecture + * + * OPFN involves the communication between two QPs on the HFI + * level on an Omni-Path fabric, and ULPs have no knowledge of + * OPFN at all. + * + * Implementation + * + * OPFN extends the existing IB RC protocol with the following + * changes: + * -- Uses Bit 24 (reserved) of DWORD 1 of Base Transport + * Header (BTH1) to indicate that the RC QP supports OPFN; + * -- Uses a combination of RC COMPARE_SWAP opcode (0x13) and + * the address U64_MAX (0xFFFFFFFFFFFFFFFF) as an OPFN + * request; The 64-bit data carried with the request/response + * contains the parameters for negotiation and will be + * defined in tid_rdma.c file; + * -- Defines IB_WR_RESERVED3 as IB_WR_OPFN. + * + * The OPFN communication will be triggered when an RC QP + * receives a request with Bit 24 of BTH1 set. The responder QP + * will then post send an OPFN request with its local + * parameters, which will be sent to the requester QP once all + * existing requests on the responder QP side have been sent. + * Once the requester QP receives the OPFN request, it will + * keep a copy of the responder QP's parameters, and return a + * response packet with its own local parameters. The responder + * QP receives the response packet and keeps a copy of the requester + * QP's parameters. After this exchange, each side has the parameters + * for both sides and therefore can select the right parameters + * for future transactions + */ + +/* STL Verbs Extended */ +#define IB_BTHE_E_SHIFT 24 +#define HFI1_VERBS_E_ATOMIC_VADDR U64_MAX + +struct ib_atomic_eth; + +enum hfi1_opfn_codes { + STL_VERBS_EXTD_NONE = 0, + STL_VERBS_EXTD_TID_RDMA, + STL_VERBS_EXTD_MAX +}; + +struct hfi1_opfn_data { + u8 extended; + u16 requested; + u16 completed; + enum hfi1_opfn_codes curr; + /* serialize opfn function calls */ + spinlock_t lock; + struct work_struct opfn_work; +}; + +/* WR opcode for OPFN */ +#define IB_WR_OPFN IB_WR_RESERVED3 + +void opfn_send_conn_request(struct work_struct *work); +void opfn_conn_response(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_atomic_eth *ateth); +void opfn_conn_reply(struct rvt_qp *qp, u64 data); +void opfn_conn_error(struct rvt_qp *qp); +void opfn_qp_init(struct rvt_qp *qp, struct ib_qp_attr *attr, int attr_mask); +void opfn_trigger_conn_request(struct rvt_qp *qp, u32 bth1); +int opfn_init(void); +void opfn_exit(void); + +#endif /* _HFI1_OPFN_H */ diff --git a/drivers/infiniband/hw/hfi1/pio.c b/drivers/infiniband/hw/hfi1/pio.c index 04126d7e318d..a1de566fe95e 100644 --- a/drivers/infiniband/hw/hfi1/pio.c +++ b/drivers/infiniband/hw/hfi1/pio.c @@ -1599,8 +1599,7 @@ static void sc_piobufavail(struct send_context *sc) struct rvt_qp *qp; struct hfi1_qp_priv *priv; unsigned long flags; - uint i, n = 0, max_idx = 0; - u8 max_starved_cnt = 0; + uint i, n = 0, top_idx = 0; if (dd->send_contexts[sc->sw_index].type != SC_KERNEL && dd->send_contexts[sc->sw_index].type != SC_VL15) @@ -1619,11 +1618,18 @@ static void sc_piobufavail(struct send_context *sc) if (n == ARRAY_SIZE(qps)) break; wait = list_first_entry(list, struct iowait, list); + iowait_get_priority(wait); qp = iowait_to_qp(wait); priv = qp->priv; list_del_init(&priv->s_iowait.list); priv->s_iowait.lock = NULL; - iowait_starve_find_max(wait, &max_starved_cnt, n, &max_idx); + if (n) { + priv = qps[top_idx]->priv; + top_idx = iowait_priority_update_top(wait, + &priv->s_iowait, + n, top_idx); + } + /* refcount held until actual wake up */ qps[n++] = qp; } @@ -1638,12 +1644,12 @@ static void sc_piobufavail(struct send_context *sc) } write_sequnlock_irqrestore(&sc->waitlock, flags); - /* Wake up the most starved one first */ + /* Wake up the top-priority one first */ if (n) - hfi1_qp_wakeup(qps[max_idx], + hfi1_qp_wakeup(qps[top_idx], RVT_S_WAIT_PIO | HFI1_S_WAIT_PIO_DRAIN); for (i = 0; i < n; i++) - if (i != max_idx) + if (i != top_idx) hfi1_qp_wakeup(qps[i], RVT_S_WAIT_PIO | HFI1_S_WAIT_PIO_DRAIN); } diff --git a/drivers/infiniband/hw/hfi1/qp.c b/drivers/infiniband/hw/hfi1/qp.c index 5344e8993b28..9b643c2409cf 100644 --- a/drivers/infiniband/hw/hfi1/qp.c +++ b/drivers/infiniband/hw/hfi1/qp.c @@ -132,6 +132,18 @@ const struct rvt_operation_params hfi1_post_parms[RVT_OPERATION_MAX] = { .qpt_support = BIT(IB_QPT_RC), }, +[IB_WR_OPFN] = { + .length = sizeof(struct ib_atomic_wr), + .qpt_support = BIT(IB_QPT_RC), + .flags = RVT_OPERATION_USE_RESERVE, +}, + +[IB_WR_TID_RDMA_WRITE] = { + .length = sizeof(struct ib_rdma_wr), + .qpt_support = BIT(IB_QPT_RC), + .flags = RVT_OPERATION_IGN_RNR_CNT, +}, + }; static void flush_list_head(struct list_head *l) @@ -285,6 +297,8 @@ void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr, priv->s_sendcontext = qp_to_send_context(qp, priv->s_sc); qp_set_16b(qp); } + + opfn_qp_init(qp, attr, attr_mask); } /** @@ -311,6 +325,8 @@ int hfi1_setup_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe, bool *call_send) switch (qp->ibqp.qp_type) { case IB_QPT_RC: + hfi1_setup_tid_rdma_wqe(qp, wqe); + /* fall through */ case IB_QPT_UC: if (wqe->length > 0x80000000U) return -EINVAL; @@ -422,6 +438,11 @@ static void hfi1_qp_schedule(struct rvt_qp *qp) if (ret) iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_IB); } + if (iowait_flag_set(&priv->s_iowait, IOWAIT_PENDING_TID)) { + ret = hfi1_schedule_tid_send(qp); + if (ret) + iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_TID); + } } void hfi1_qp_wakeup(struct rvt_qp *qp, u32 flag) @@ -441,8 +462,27 @@ void hfi1_qp_wakeup(struct rvt_qp *qp, u32 flag) void hfi1_qp_unbusy(struct rvt_qp *qp, struct iowait_work *wait) { - if (iowait_set_work_flag(wait) == IOWAIT_IB_SE) + struct hfi1_qp_priv *priv = qp->priv; + + if (iowait_set_work_flag(wait) == IOWAIT_IB_SE) { qp->s_flags &= ~RVT_S_BUSY; + /* + * If we are sending a first-leg packet from the second leg, + * we need to clear the busy flag from priv->s_flags to + * avoid a race condition when the qp wakes up before + * the call to hfi1_verbs_send() returns to the second + * leg. In that case, the second leg will terminate without + * being re-scheduled, resulting in failure to send TID RDMA + * WRITE DATA and TID RDMA ACK packets. + */ + if (priv->s_flags & HFI1_S_TID_BUSY_SET) { + priv->s_flags &= ~(HFI1_S_TID_BUSY_SET | + RVT_S_BUSY); + iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_TID); + } + } else { + priv->s_flags &= ~RVT_S_BUSY; + } } static int iowait_sleep( @@ -479,6 +519,7 @@ static int iowait_sleep( ibp->rvp.n_dmawait++; qp->s_flags |= RVT_S_WAIT_DMA_DESC; + iowait_get_priority(&priv->s_iowait); iowait_queue(pkts_sent, &priv->s_iowait, &sde->dmawait); priv->s_iowait.lock = &sde->waitlock; @@ -528,6 +569,17 @@ static void iowait_sdma_drained(struct iowait *wait) spin_unlock_irqrestore(&qp->s_lock, flags); } +static void hfi1_init_priority(struct iowait *w) +{ + struct rvt_qp *qp = iowait_to_qp(w); + struct hfi1_qp_priv *priv = qp->priv; + + if (qp->s_flags & RVT_S_ACK_PENDING) + w->priority++; + if (priv->s_flags & RVT_S_ACK_PENDING) + w->priority++; +} + /** * qp_to_sdma_engine - map a qp to a send engine * @qp: the QP @@ -685,10 +737,11 @@ void *qp_priv_alloc(struct rvt_dev_info *rdi, struct rvt_qp *qp) &priv->s_iowait, 1, _hfi1_do_send, - NULL, + _hfi1_do_tid_send, iowait_sleep, iowait_wakeup, - iowait_sdma_drained); + iowait_sdma_drained, + hfi1_init_priority); return priv; } @@ -696,6 +749,7 @@ void qp_priv_free(struct rvt_dev_info *rdi, struct rvt_qp *qp) { struct hfi1_qp_priv *priv = qp->priv; + hfi1_qp_priv_tid_free(rdi, qp); kfree(priv->s_ahg); kfree(priv); } @@ -729,6 +783,7 @@ void flush_qp_waiters(struct rvt_qp *qp) { lockdep_assert_held(&qp->s_lock); flush_iowait(qp); + hfi1_tid_rdma_flush_wait(qp); } void stop_send_queue(struct rvt_qp *qp) @@ -736,12 +791,16 @@ void stop_send_queue(struct rvt_qp *qp) struct hfi1_qp_priv *priv = qp->priv; iowait_cancel_work(&priv->s_iowait); + if (cancel_work_sync(&priv->tid_rdma.trigger_work)) + rvt_put_qp(qp); } void quiesce_qp(struct rvt_qp *qp) { struct hfi1_qp_priv *priv = qp->priv; + hfi1_del_tid_reap_timer(qp); + hfi1_del_tid_retry_timer(qp); iowait_sdma_drain(&priv->s_iowait); qp_pio_drain(qp); flush_tx_list(qp); @@ -749,8 +808,13 @@ void quiesce_qp(struct rvt_qp *qp) void notify_qp_reset(struct rvt_qp *qp) { + hfi1_qp_kern_exp_rcv_clear_all(qp); qp->r_adefered = 0; clear_ahg(qp); + + /* Clear any OPFN state */ + if (qp->ibqp.qp_type == IB_QPT_RC) + opfn_conn_error(qp); } /* @@ -832,7 +896,8 @@ void notify_error_qp(struct rvt_qp *qp) if (lock) { write_seqlock(lock); if (!list_empty(&priv->s_iowait.list) && - !(qp->s_flags & RVT_S_BUSY)) { + !(qp->s_flags & RVT_S_BUSY) && + !(priv->s_flags & RVT_S_BUSY)) { qp->s_flags &= ~RVT_S_ANY_WAIT_IO; list_del_init(&priv->s_iowait.list); priv->s_iowait.lock = NULL; @@ -841,7 +906,8 @@ void notify_error_qp(struct rvt_qp *qp) write_sequnlock(lock); } - if (!(qp->s_flags & RVT_S_BUSY)) { + if (!(qp->s_flags & RVT_S_BUSY) && !(priv->s_flags & RVT_S_BUSY)) { + qp->s_hdrwords = 0; if (qp->s_rdma_mr) { rvt_put_mr(qp->s_rdma_mr); qp->s_rdma_mr = NULL; diff --git a/drivers/infiniband/hw/hfi1/qp.h b/drivers/infiniband/hw/hfi1/qp.h index 7adb6dff6813..b670321365d3 100644 --- a/drivers/infiniband/hw/hfi1/qp.h +++ b/drivers/infiniband/hw/hfi1/qp.h @@ -63,11 +63,17 @@ extern const struct rvt_operation_params hfi1_post_parms[]; * HFI1_S_AHG_VALID - ahg header valid on chip * HFI1_S_AHG_CLEAR - have send engine clear ahg state * HFI1_S_WAIT_PIO_DRAIN - qp waiting for PIOs to drain + * HFI1_S_WAIT_TID_SPACE - a QP is waiting for TID resource + * HFI1_S_WAIT_TID_RESP - waiting for a TID RDMA WRITE response + * HFI1_S_WAIT_HALT - halt the first leg send engine * HFI1_S_MIN_BIT_MASK - the lowest bit that can be used by hfi1 */ #define HFI1_S_AHG_VALID 0x80000000 #define HFI1_S_AHG_CLEAR 0x40000000 #define HFI1_S_WAIT_PIO_DRAIN 0x20000000 +#define HFI1_S_WAIT_TID_SPACE 0x10000000 +#define HFI1_S_WAIT_TID_RESP 0x08000000 +#define HFI1_S_WAIT_HALT 0x04000000 #define HFI1_S_MIN_BIT_MASK 0x01000000 /* @@ -76,6 +82,7 @@ extern const struct rvt_operation_params hfi1_post_parms[]; #define HFI1_S_ANY_WAIT_IO (RVT_S_ANY_WAIT_IO | HFI1_S_WAIT_PIO_DRAIN) #define HFI1_S_ANY_WAIT (HFI1_S_ANY_WAIT_IO | RVT_S_ANY_WAIT_SEND) +#define HFI1_S_ANY_TID_WAIT_SEND (RVT_S_WAIT_SSN_CREDIT | RVT_S_WAIT_DMA) /* * Send if not busy or waiting for I/O and either diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c index be603f35d7e4..e6726c1ab866 100644 --- a/drivers/infiniband/hw/hfi1/rc.c +++ b/drivers/infiniband/hw/hfi1/rc.c @@ -51,24 +51,48 @@ #include "hfi.h" #include "qp.h" +#include "rc.h" #include "verbs_txreq.h" #include "trace.h" -/* cut down ridiculously long IB macro names */ -#define OP(x) RC_OP(x) - -static u32 restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe, - u32 psn, u32 pmtu) +struct rvt_ack_entry *find_prev_entry(struct rvt_qp *qp, u32 psn, u8 *prev, + u8 *prev_ack, bool *scheduled) + __must_hold(&qp->s_lock) { - u32 len; - - len = delta_psn(psn, wqe->psn) * pmtu; - ss->sge = wqe->sg_list[0]; - ss->sg_list = wqe->sg_list + 1; - ss->num_sge = wqe->wr.num_sge; - ss->total_len = wqe->length; - rvt_skip_sge(ss, len, false); - return wqe->length - len; + struct rvt_ack_entry *e = NULL; + u8 i, p; + bool s = true; + + for (i = qp->r_head_ack_queue; ; i = p) { + if (i == qp->s_tail_ack_queue) + s = false; + if (i) + p = i - 1; + else + p = rvt_size_atomic(ib_to_rvt(qp->ibqp.device)); + if (p == qp->r_head_ack_queue) { + e = NULL; + break; + } + e = &qp->s_ack_queue[p]; + if (!e->opcode) { + e = NULL; + break; + } + if (cmp_psn(psn, e->psn) >= 0) { + if (p == qp->s_tail_ack_queue && + cmp_psn(psn, e->lpsn) <= 0) + s = false; + break; + } + } + if (prev) + *prev = p; + if (prev_ack) + *prev_ack = i; + if (scheduled) + *scheduled = s; + return e; } /** @@ -87,20 +111,25 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp, struct hfi1_pkt_state *ps) { struct rvt_ack_entry *e; - u32 hwords; - u32 len; - u32 bth0; - u32 bth2; + u32 hwords, hdrlen; + u32 len = 0; + u32 bth0 = 0, bth2 = 0; + u32 bth1 = qp->remote_qpn | (HFI1_CAP_IS_KSET(OPFN) << IB_BTHE_E_SHIFT); int middle = 0; u32 pmtu = qp->pmtu; - struct hfi1_qp_priv *priv = qp->priv; + struct hfi1_qp_priv *qpriv = qp->priv; + bool last_pkt; + u32 delta; + u8 next = qp->s_tail_ack_queue; + struct tid_rdma_request *req; + trace_hfi1_rsp_make_rc_ack(qp, 0); lockdep_assert_held(&qp->s_lock); /* Don't send an ACK if we aren't supposed to. */ if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) goto bail; - if (priv->hdr_type == HFI1_PKT_TYPE_9B) + if (qpriv->hdr_type == HFI1_PKT_TYPE_9B) /* header size in 32-bit words LRH+BTH = (8+12)/4. */ hwords = 5; else @@ -122,8 +151,18 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp, * response has been sent instead of only being * constructed. */ - if (++qp->s_tail_ack_queue > HFI1_MAX_RDMA_ATOMIC) - qp->s_tail_ack_queue = 0; + if (++next > rvt_size_atomic(&dev->rdi)) + next = 0; + /* + * Only advance the s_acked_ack_queue pointer if there + * have been no TID RDMA requests. + */ + e = &qp->s_ack_queue[qp->s_tail_ack_queue]; + if (e->opcode != TID_OP(WRITE_REQ) && + qp->s_acked_ack_queue == qp->s_tail_ack_queue) + qp->s_acked_ack_queue = next; + qp->s_tail_ack_queue = next; + trace_hfi1_rsp_make_rc_ack(qp, e->psn); /* FALLTHROUGH */ case OP(SEND_ONLY): case OP(ACKNOWLEDGE): @@ -135,6 +174,12 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp, } e = &qp->s_ack_queue[qp->s_tail_ack_queue]; + /* Check for tid write fence */ + if ((qpriv->s_flags & HFI1_R_TID_WAIT_INTERLCK) || + hfi1_tid_rdma_ack_interlock(qp, e)) { + iowait_set_flag(&qpriv->s_iowait, IOWAIT_PENDING_IB); + goto bail; + } if (e->opcode == OP(RDMA_READ_REQUEST)) { /* * If a RDMA read response is being resent and @@ -144,6 +189,10 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp, */ len = e->rdma_sge.sge_length; if (len && !e->rdma_sge.mr) { + if (qp->s_acked_ack_queue == + qp->s_tail_ack_queue) + qp->s_acked_ack_queue = + qp->r_head_ack_queue; qp->s_tail_ack_queue = qp->r_head_ack_queue; goto bail; } @@ -165,6 +214,45 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp, hwords++; qp->s_ack_rdma_psn = e->psn; bth2 = mask_psn(qp->s_ack_rdma_psn++); + } else if (e->opcode == TID_OP(WRITE_REQ)) { + /* + * If a TID RDMA WRITE RESP is being resent, we have to + * wait for the actual request. All requests that are to + * be resent will have their state set to + * TID_REQUEST_RESEND. When the new request arrives, the + * state will be changed to TID_REQUEST_RESEND_ACTIVE. + */ + req = ack_to_tid_req(e); + if (req->state == TID_REQUEST_RESEND || + req->state == TID_REQUEST_INIT_RESEND) + goto bail; + qp->s_ack_state = TID_OP(WRITE_RESP); + qp->s_ack_rdma_psn = mask_psn(e->psn + req->cur_seg); + goto write_resp; + } else if (e->opcode == TID_OP(READ_REQ)) { + /* + * If a TID RDMA read response is being resent and + * we haven't seen the duplicate request yet, + * then stop sending the remaining responses the + * responder has seen until the requester re-sends it. + */ + len = e->rdma_sge.sge_length; + if (len && !e->rdma_sge.mr) { + if (qp->s_acked_ack_queue == + qp->s_tail_ack_queue) + qp->s_acked_ack_queue = + qp->r_head_ack_queue; + qp->s_tail_ack_queue = qp->r_head_ack_queue; + goto bail; + } + /* Copy SGE state in case we need to resend */ + ps->s_txreq->mr = e->rdma_sge.mr; + if (ps->s_txreq->mr) + rvt_get_mr(ps->s_txreq->mr); + qp->s_ack_rdma_sge.sge = e->rdma_sge; + qp->s_ack_rdma_sge.num_sge = 1; + qp->s_ack_state = TID_OP(READ_RESP); + goto read_resp; } else { /* COMPARE_SWAP or FETCH_ADD */ ps->s_txreq->ss = NULL; @@ -176,6 +264,7 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp, bth2 = mask_psn(e->psn); e->sent = 1; } + trace_hfi1_tid_write_rsp_make_rc_ack(qp); bth0 = qp->s_ack_state << 24; break; @@ -202,6 +291,83 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp, bth2 = mask_psn(qp->s_ack_rdma_psn++); break; + case TID_OP(WRITE_RESP): +write_resp: + /* + * 1. Check if RVT_S_ACK_PENDING is set. If yes, + * goto normal. + * 2. Attempt to allocate TID resources. + * 3. Remove RVT_S_RESP_PENDING flags from s_flags + * 4. If resources not available: + * 4.1 Set RVT_S_WAIT_TID_SPACE + * 4.2 Queue QP on RCD TID queue + * 4.3 Put QP on iowait list. + * 4.4 Build IB RNR NAK with appropriate timeout value + * 4.5 Return indication progress made. + * 5. If resources are available: + * 5.1 Program HW flow CSRs + * 5.2 Build TID RDMA WRITE RESP packet + * 5.3 If more resources needed, do 2.1 - 2.3. + * 5.4 Wake up next QP on RCD TID queue. + * 5.5 Return indication progress made. + */ + + e = &qp->s_ack_queue[qp->s_tail_ack_queue]; + req = ack_to_tid_req(e); + + /* + * Send scheduled RNR NAK's. RNR NAK's need to be sent at + * segment boundaries, not at request boundaries. Don't change + * s_ack_state because we are still in the middle of a request + */ + if (qpriv->rnr_nak_state == TID_RNR_NAK_SEND && + qp->s_tail_ack_queue == qpriv->r_tid_alloc && + req->cur_seg == req->alloc_seg) { + qpriv->rnr_nak_state = TID_RNR_NAK_SENT; + goto normal_no_state; + } + + bth2 = mask_psn(qp->s_ack_rdma_psn); + hdrlen = hfi1_build_tid_rdma_write_resp(qp, e, ohdr, &bth1, + bth2, &len, + &ps->s_txreq->ss); + if (!hdrlen) + return 0; + + hwords += hdrlen; + bth0 = qp->s_ack_state << 24; + qp->s_ack_rdma_psn++; + trace_hfi1_tid_req_make_rc_ack_write(qp, 0, e->opcode, e->psn, + e->lpsn, req); + if (req->cur_seg != req->total_segs) + break; + + e->sent = 1; + qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST); + break; + + case TID_OP(READ_RESP): +read_resp: + e = &qp->s_ack_queue[qp->s_tail_ack_queue]; + ps->s_txreq->ss = &qp->s_ack_rdma_sge; + delta = hfi1_build_tid_rdma_read_resp(qp, e, ohdr, &bth0, + &bth1, &bth2, &len, + &last_pkt); + if (delta == 0) + goto error_qp; + hwords += delta; + if (last_pkt) { + e->sent = 1; + /* + * Increment qp->s_tail_ack_queue through s_ack_state + * transition. + */ + qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST); + } + break; + case TID_OP(READ_REQ): + goto bail; + default: normal: /* @@ -211,8 +377,7 @@ normal: * (see above). */ qp->s_ack_state = OP(SEND_ONLY); - qp->s_flags &= ~RVT_S_ACK_PENDING; - ps->s_txreq->ss = NULL; +normal_no_state: if (qp->s_nak_state) ohdr->u.aeth = cpu_to_be32((qp->r_msn & IB_MSN_MASK) | @@ -224,14 +389,24 @@ normal: len = 0; bth0 = OP(ACKNOWLEDGE) << 24; bth2 = mask_psn(qp->s_ack_psn); + qp->s_flags &= ~RVT_S_ACK_PENDING; + ps->s_txreq->txreq.flags |= SDMA_TXREQ_F_VIP; + ps->s_txreq->ss = NULL; } qp->s_rdma_ack_cnt++; - ps->s_txreq->sde = priv->s_sde; + ps->s_txreq->sde = qpriv->s_sde; ps->s_txreq->s_cur_size = len; ps->s_txreq->hdr_dwords = hwords; - hfi1_make_ruc_header(qp, ohdr, bth0, bth2, middle, ps); + hfi1_make_ruc_header(qp, ohdr, bth0, bth1, bth2, middle, ps); return 1; - +error_qp: + spin_unlock_irqrestore(&qp->s_lock, ps->flags); + spin_lock_irqsave(&qp->r_lock, ps->flags); + spin_lock(&qp->s_lock); + rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR); + spin_unlock(&qp->s_lock); + spin_unlock_irqrestore(&qp->r_lock, ps->flags); + spin_lock_irqsave(&qp->s_lock, ps->flags); bail: qp->s_ack_state = OP(ACKNOWLEDGE); /* @@ -258,17 +433,23 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) struct hfi1_qp_priv *priv = qp->priv; struct hfi1_ibdev *dev = to_idev(qp->ibqp.device); struct ib_other_headers *ohdr; - struct rvt_sge_state *ss; + struct rvt_sge_state *ss = NULL; struct rvt_swqe *wqe; - u32 hwords; - u32 len; - u32 bth0 = 0; - u32 bth2; + struct hfi1_swqe_priv *wpriv; + struct tid_rdma_request *req = NULL; + /* header size in 32-bit words LRH+BTH = (8+12)/4. */ + u32 hwords = 5; + u32 len = 0; + u32 bth0 = 0, bth2 = 0; + u32 bth1 = qp->remote_qpn | (HFI1_CAP_IS_KSET(OPFN) << IB_BTHE_E_SHIFT); u32 pmtu = qp->pmtu; char newreq; int middle = 0; int delta; + struct tid_rdma_flow *flow = NULL; + struct tid_rdma_params *remote; + trace_hfi1_sender_make_rc_req(qp); lockdep_assert_held(&qp->s_lock); ps->s_txreq = get_txreq(ps->dev, qp); if (!ps->s_txreq) @@ -309,13 +490,13 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) } clear_ahg(qp); wqe = rvt_get_swqe_ptr(qp, qp->s_last); - rvt_send_complete(qp, wqe, qp->s_last != qp->s_acked ? - IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR); + hfi1_trdma_send_complete(qp, wqe, qp->s_last != qp->s_acked ? + IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR); /* will get called again */ goto done_free_tx; } - if (qp->s_flags & (RVT_S_WAIT_RNR | RVT_S_WAIT_ACK)) + if (qp->s_flags & (RVT_S_WAIT_RNR | RVT_S_WAIT_ACK | HFI1_S_WAIT_HALT)) goto bail; if (cmp_psn(qp->s_psn, qp->s_sending_hpsn) <= 0) { @@ -329,6 +510,7 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) /* Send a request. */ wqe = rvt_get_swqe_ptr(qp, qp->s_cur); +check_s_state: switch (qp->s_state) { default: if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_NEXT_SEND_OK)) @@ -350,9 +532,13 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) /* * If a fence is requested, wait for previous * RDMA read and atomic operations to finish. + * However, there is no need to guard against + * TID RDMA READ after TID RDMA READ. */ if ((wqe->wr.send_flags & IB_SEND_FENCE) && - qp->s_num_rd_atomic) { + qp->s_num_rd_atomic && + (wqe->wr.opcode != IB_WR_TID_RDMA_READ || + priv->pending_tid_r_segs < qp->s_num_rd_atomic)) { qp->s_flags |= RVT_S_WAIT_FENCE; goto bail; } @@ -397,6 +583,15 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) len = wqe->length; ss = &qp->s_sge; bth2 = mask_psn(qp->s_psn); + + /* + * Interlock between various IB requests and TID RDMA + * if necessary. + */ + if ((priv->s_flags & HFI1_S_TID_WAIT_INTERLCK) || + hfi1_tid_rdma_wqe_interlock(qp, wqe)) + goto bail; + switch (wqe->wr.opcode) { case IB_WR_SEND: case IB_WR_SEND_WITH_IMM: @@ -473,21 +668,126 @@ no_flow_control: qp->s_cur = 0; break; + case IB_WR_TID_RDMA_WRITE: + if (newreq) { + /* + * Limit the number of TID RDMA WRITE requests. + */ + if (atomic_read(&priv->n_tid_requests) >= + HFI1_TID_RDMA_WRITE_CNT) + goto bail; + + if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) + qp->s_lsn++; + } + + hwords += hfi1_build_tid_rdma_write_req(qp, wqe, ohdr, + &bth1, &bth2, + &len); + ss = NULL; + if (priv->s_tid_cur == HFI1_QP_WQE_INVALID) { + priv->s_tid_cur = qp->s_cur; + if (priv->s_tid_tail == HFI1_QP_WQE_INVALID) { + priv->s_tid_tail = qp->s_cur; + priv->s_state = TID_OP(WRITE_RESP); + } + } else if (priv->s_tid_cur == priv->s_tid_head) { + struct rvt_swqe *__w; + struct tid_rdma_request *__r; + + __w = rvt_get_swqe_ptr(qp, priv->s_tid_cur); + __r = wqe_to_tid_req(__w); + + /* + * The s_tid_cur pointer is advanced to s_cur if + * any of the following conditions about the WQE + * to which s_ti_cur currently points to are + * satisfied: + * 1. The request is not a TID RDMA WRITE + * request, + * 2. The request is in the INACTIVE or + * COMPLETE states (TID RDMA READ requests + * stay at INACTIVE and TID RDMA WRITE + * transition to COMPLETE when done), + * 3. The request is in the ACTIVE or SYNC + * state and the number of completed + * segments is equal to the total segment + * count. + * (If ACTIVE, the request is waiting for + * ACKs. If SYNC, the request has not + * received any responses because it's + * waiting on a sync point.) + */ + if (__w->wr.opcode != IB_WR_TID_RDMA_WRITE || + __r->state == TID_REQUEST_INACTIVE || + __r->state == TID_REQUEST_COMPLETE || + ((__r->state == TID_REQUEST_ACTIVE || + __r->state == TID_REQUEST_SYNC) && + __r->comp_seg == __r->total_segs)) { + if (priv->s_tid_tail == + priv->s_tid_cur && + priv->s_state == + TID_OP(WRITE_DATA_LAST)) { + priv->s_tid_tail = qp->s_cur; + priv->s_state = + TID_OP(WRITE_RESP); + } + priv->s_tid_cur = qp->s_cur; + } + /* + * A corner case: when the last TID RDMA WRITE + * request was completed, s_tid_head, + * s_tid_cur, and s_tid_tail all point to the + * same location. Other requests are posted and + * s_cur wraps around to the same location, + * where a new TID RDMA WRITE is posted. In + * this case, none of the indices need to be + * updated. However, the priv->s_state should. + */ + if (priv->s_tid_tail == qp->s_cur && + priv->s_state == TID_OP(WRITE_DATA_LAST)) + priv->s_state = TID_OP(WRITE_RESP); + } + req = wqe_to_tid_req(wqe); + if (newreq) { + priv->s_tid_head = qp->s_cur; + priv->pending_tid_w_resp += req->total_segs; + atomic_inc(&priv->n_tid_requests); + atomic_dec(&priv->n_requests); + } else { + req->state = TID_REQUEST_RESEND; + req->comp_seg = delta_psn(bth2, wqe->psn); + /* + * Pull back any segments since we are going + * to re-receive them. + */ + req->setup_head = req->clear_tail; + priv->pending_tid_w_resp += + delta_psn(wqe->lpsn, bth2) + 1; + } + + trace_hfi1_tid_write_sender_make_req(qp, newreq); + trace_hfi1_tid_req_make_req_write(qp, newreq, + wqe->wr.opcode, + wqe->psn, wqe->lpsn, + req); + if (++qp->s_cur == qp->s_size) + qp->s_cur = 0; + break; + case IB_WR_RDMA_READ: /* * Don't allow more operations to be started * than the QP limits allow. */ - if (newreq) { - if (qp->s_num_rd_atomic >= - qp->s_max_rd_atomic) { - qp->s_flags |= RVT_S_WAIT_RDMAR; - goto bail; - } - qp->s_num_rd_atomic++; - if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) - qp->s_lsn++; + if (qp->s_num_rd_atomic >= + qp->s_max_rd_atomic) { + qp->s_flags |= RVT_S_WAIT_RDMAR; + goto bail; } + qp->s_num_rd_atomic++; + if (newreq && !(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) + qp->s_lsn++; put_ib_reth_vaddr( wqe->rdma_wr.remote_addr, &ohdr->u.rc.reth); @@ -503,23 +803,99 @@ no_flow_control: qp->s_cur = 0; break; + case IB_WR_TID_RDMA_READ: + trace_hfi1_tid_read_sender_make_req(qp, newreq); + wpriv = wqe->priv; + req = wqe_to_tid_req(wqe); + trace_hfi1_tid_req_make_req_read(qp, newreq, + wqe->wr.opcode, + wqe->psn, wqe->lpsn, + req); + delta = cmp_psn(qp->s_psn, wqe->psn); + + /* + * Don't allow more operations to be started + * than the QP limits allow. We could get here under + * three conditions; (1) It's a new request; (2) We are + * sending the second or later segment of a request, + * but the qp->s_state is set to OP(RDMA_READ_REQUEST) + * when the last segment of a previous request is + * received just before this; (3) We are re-sending a + * request. + */ + if (qp->s_num_rd_atomic >= qp->s_max_rd_atomic) { + qp->s_flags |= RVT_S_WAIT_RDMAR; + goto bail; + } + if (newreq) { + struct tid_rdma_flow *flow = + &req->flows[req->setup_head]; + + /* + * Set up s_sge as it is needed for TID + * allocation. However, if the pages have been + * walked and mapped, skip it. An earlier try + * has failed to allocate the TID entries. + */ + if (!flow->npagesets) { + qp->s_sge.sge = wqe->sg_list[0]; + qp->s_sge.sg_list = wqe->sg_list + 1; + qp->s_sge.num_sge = wqe->wr.num_sge; + qp->s_sge.total_len = wqe->length; + qp->s_len = wqe->length; + req->isge = 0; + req->clear_tail = req->setup_head; + req->flow_idx = req->setup_head; + req->state = TID_REQUEST_ACTIVE; + } + } else if (delta == 0) { + /* Re-send a request */ + req->cur_seg = 0; + req->comp_seg = 0; + req->ack_pending = 0; + req->flow_idx = req->clear_tail; + req->state = TID_REQUEST_RESEND; + } + req->s_next_psn = qp->s_psn; + /* Read one segment at a time */ + len = min_t(u32, req->seg_len, + wqe->length - req->seg_len * req->cur_seg); + delta = hfi1_build_tid_rdma_read_req(qp, wqe, ohdr, + &bth1, &bth2, + &len); + if (delta <= 0) { + /* Wait for TID space */ + goto bail; + } + if (newreq && !(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) + qp->s_lsn++; + hwords += delta; + ss = &wpriv->ss; + /* Check if this is the last segment */ + if (req->cur_seg >= req->total_segs && + ++qp->s_cur == qp->s_size) + qp->s_cur = 0; + break; + case IB_WR_ATOMIC_CMP_AND_SWP: case IB_WR_ATOMIC_FETCH_AND_ADD: /* * Don't allow more operations to be started * than the QP limits allow. */ - if (newreq) { - if (qp->s_num_rd_atomic >= - qp->s_max_rd_atomic) { - qp->s_flags |= RVT_S_WAIT_RDMAR; - goto bail; - } - qp->s_num_rd_atomic++; - if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) - qp->s_lsn++; + if (qp->s_num_rd_atomic >= + qp->s_max_rd_atomic) { + qp->s_flags |= RVT_S_WAIT_RDMAR; + goto bail; } - if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) { + qp->s_num_rd_atomic++; + + /* FALLTHROUGH */ + case IB_WR_OPFN: + if (newreq && !(qp->s_flags & RVT_S_UNLIMITED_CREDIT)) + qp->s_lsn++; + if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP || + wqe->wr.opcode == IB_WR_OPFN) { qp->s_state = OP(COMPARE_SWAP); put_ib_ateth_swap(wqe->atomic_wr.swap, &ohdr->u.atomic_eth); @@ -546,18 +922,23 @@ no_flow_control: default: goto bail; } - qp->s_sge.sge = wqe->sg_list[0]; - qp->s_sge.sg_list = wqe->sg_list + 1; - qp->s_sge.num_sge = wqe->wr.num_sge; - qp->s_sge.total_len = wqe->length; - qp->s_len = wqe->length; + if (wqe->wr.opcode != IB_WR_TID_RDMA_READ) { + qp->s_sge.sge = wqe->sg_list[0]; + qp->s_sge.sg_list = wqe->sg_list + 1; + qp->s_sge.num_sge = wqe->wr.num_sge; + qp->s_sge.total_len = wqe->length; + qp->s_len = wqe->length; + } if (newreq) { qp->s_tail++; if (qp->s_tail >= qp->s_size) qp->s_tail = 0; } - if (wqe->wr.opcode == IB_WR_RDMA_READ) + if (wqe->wr.opcode == IB_WR_RDMA_READ || + wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) qp->s_psn = wqe->lpsn + 1; + else if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) + qp->s_psn = req->s_next_psn; else qp->s_psn++; break; @@ -674,10 +1055,137 @@ no_flow_control: if (qp->s_cur == qp->s_size) qp->s_cur = 0; break; + + case TID_OP(WRITE_RESP): + /* + * This value for s_state is used for restarting a TID RDMA + * WRITE request. See comment in OP(RDMA_READ_RESPONSE_MIDDLE + * for more). + */ + req = wqe_to_tid_req(wqe); + req->state = TID_REQUEST_RESEND; + rcu_read_lock(); + remote = rcu_dereference(priv->tid_rdma.remote); + req->comp_seg = delta_psn(qp->s_psn, wqe->psn); + len = wqe->length - (req->comp_seg * remote->max_len); + rcu_read_unlock(); + + bth2 = mask_psn(qp->s_psn); + hwords += hfi1_build_tid_rdma_write_req(qp, wqe, ohdr, &bth1, + &bth2, &len); + qp->s_psn = wqe->lpsn + 1; + ss = NULL; + qp->s_state = TID_OP(WRITE_REQ); + priv->pending_tid_w_resp += delta_psn(wqe->lpsn, bth2) + 1; + priv->s_tid_cur = qp->s_cur; + if (++qp->s_cur == qp->s_size) + qp->s_cur = 0; + trace_hfi1_tid_req_make_req_write(qp, 0, wqe->wr.opcode, + wqe->psn, wqe->lpsn, req); + break; + + case TID_OP(READ_RESP): + if (wqe->wr.opcode != IB_WR_TID_RDMA_READ) + goto bail; + /* This is used to restart a TID read request */ + req = wqe_to_tid_req(wqe); + wpriv = wqe->priv; + /* + * Back down. The field qp->s_psn has been set to the psn with + * which the request should be restart. It's OK to use division + * as this is on the retry path. + */ + req->cur_seg = delta_psn(qp->s_psn, wqe->psn) / priv->pkts_ps; + + /* + * The following function need to be redefined to return the + * status to make sure that we find the flow. At the same + * time, we can use the req->state change to check if the + * call succeeds or not. + */ + req->state = TID_REQUEST_RESEND; + hfi1_tid_rdma_restart_req(qp, wqe, &bth2); + if (req->state != TID_REQUEST_ACTIVE) { + /* + * Failed to find the flow. Release all allocated tid + * resources. + */ + hfi1_kern_exp_rcv_clear_all(req); + hfi1_kern_clear_hw_flow(priv->rcd, qp); + + hfi1_trdma_send_complete(qp, wqe, IB_WC_LOC_QP_OP_ERR); + goto bail; + } + req->state = TID_REQUEST_RESEND; + len = min_t(u32, req->seg_len, + wqe->length - req->seg_len * req->cur_seg); + flow = &req->flows[req->flow_idx]; + len -= flow->sent; + req->s_next_psn = flow->flow_state.ib_lpsn + 1; + delta = hfi1_build_tid_rdma_read_packet(wqe, ohdr, &bth1, + &bth2, &len); + if (delta <= 0) { + /* Wait for TID space */ + goto bail; + } + hwords += delta; + ss = &wpriv->ss; + /* Check if this is the last segment */ + if (req->cur_seg >= req->total_segs && + ++qp->s_cur == qp->s_size) + qp->s_cur = 0; + qp->s_psn = req->s_next_psn; + trace_hfi1_tid_req_make_req_read(qp, 0, wqe->wr.opcode, + wqe->psn, wqe->lpsn, req); + break; + case TID_OP(READ_REQ): + req = wqe_to_tid_req(wqe); + delta = cmp_psn(qp->s_psn, wqe->psn); + /* + * If the current WR is not TID RDMA READ, or this is the start + * of a new request, we need to change the qp->s_state so that + * the request can be set up properly. + */ + if (wqe->wr.opcode != IB_WR_TID_RDMA_READ || delta == 0 || + qp->s_cur == qp->s_tail) { + qp->s_state = OP(RDMA_READ_REQUEST); + if (delta == 0 || qp->s_cur == qp->s_tail) + goto check_s_state; + else + goto bail; + } + + /* Rate limiting */ + if (qp->s_num_rd_atomic >= qp->s_max_rd_atomic) { + qp->s_flags |= RVT_S_WAIT_RDMAR; + goto bail; + } + + wpriv = wqe->priv; + /* Read one segment at a time */ + len = min_t(u32, req->seg_len, + wqe->length - req->seg_len * req->cur_seg); + delta = hfi1_build_tid_rdma_read_req(qp, wqe, ohdr, &bth1, + &bth2, &len); + if (delta <= 0) { + /* Wait for TID space */ + goto bail; + } + hwords += delta; + ss = &wpriv->ss; + /* Check if this is the last segment */ + if (req->cur_seg >= req->total_segs && + ++qp->s_cur == qp->s_size) + qp->s_cur = 0; + qp->s_psn = req->s_next_psn; + trace_hfi1_tid_req_make_req_read(qp, 0, wqe->wr.opcode, + wqe->psn, wqe->lpsn, req); + break; } qp->s_sending_hpsn = bth2; delta = delta_psn(bth2, wqe->psn); - if (delta && delta % HFI1_PSN_CREDIT == 0) + if (delta && delta % HFI1_PSN_CREDIT == 0 && + wqe->wr.opcode != IB_WR_TID_RDMA_WRITE) bth2 |= IB_BTH_REQ_ACK; if (qp->s_flags & RVT_S_SEND_ONE) { qp->s_flags &= ~RVT_S_SEND_ONE; @@ -693,6 +1201,7 @@ no_flow_control: qp, ohdr, bth0 | (qp->s_state << 24), + bth1, bth2, middle, ps); @@ -709,6 +1218,12 @@ bail: bail_no_tx: ps->s_txreq = NULL; qp->s_flags &= ~RVT_S_BUSY; + /* + * If we didn't get a txreq, the QP will be woken up later to try + * again. Set the flags to indicate which work item to wake + * up. + */ + iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_IB); return 0; } @@ -796,6 +1311,11 @@ static inline void hfi1_make_rc_ack_9B(struct hfi1_packet *packet, if (qp->s_mig_state == IB_MIG_MIGRATED) bth0 |= IB_BTH_MIG_REQ; bth1 = (!!is_fecn) << IB_BECN_SHIFT; + /* + * Inline ACKs go out without the use of the Verbs send engine, so + * we need to set the STL Verbs Extended bit here + */ + bth1 |= HFI1_CAP_IS_KSET(OPFN) << IB_BTHE_E_SHIFT; hfi1_make_bth_aeth(qp, ohdr, bth0, bth1); } @@ -936,6 +1456,43 @@ void hfi1_send_rc_ack(struct hfi1_packet *packet, bool is_fecn) } /** + * update_num_rd_atomic - update the qp->s_num_rd_atomic + * @qp: the QP + * @psn: the packet sequence number to restart at + * @wqe: the wqe + * + * This is called from reset_psn() to update qp->s_num_rd_atomic + * for the current wqe. + * Called at interrupt level with the QP s_lock held. + */ +static void update_num_rd_atomic(struct rvt_qp *qp, u32 psn, + struct rvt_swqe *wqe) +{ + u32 opcode = wqe->wr.opcode; + + if (opcode == IB_WR_RDMA_READ || + opcode == IB_WR_ATOMIC_CMP_AND_SWP || + opcode == IB_WR_ATOMIC_FETCH_AND_ADD) { + qp->s_num_rd_atomic++; + } else if (opcode == IB_WR_TID_RDMA_READ) { + struct tid_rdma_request *req = wqe_to_tid_req(wqe); + struct hfi1_qp_priv *priv = qp->priv; + + if (cmp_psn(psn, wqe->lpsn) <= 0) { + u32 cur_seg; + + cur_seg = (psn - wqe->psn) / priv->pkts_ps; + req->ack_pending = cur_seg - req->comp_seg; + priv->pending_tid_r_segs += req->ack_pending; + qp->s_num_rd_atomic += req->ack_pending; + } else { + priv->pending_tid_r_segs += req->total_segs; + qp->s_num_rd_atomic += req->total_segs; + } + } +} + +/** * reset_psn - reset the QP state to send starting from PSN * @qp: the QP * @psn: the packet sequence number to restart at @@ -949,9 +1506,13 @@ static void reset_psn(struct rvt_qp *qp, u32 psn) u32 n = qp->s_acked; struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, n); u32 opcode; + struct hfi1_qp_priv *priv = qp->priv; lockdep_assert_held(&qp->s_lock); qp->s_cur = n; + priv->pending_tid_r_segs = 0; + priv->pending_tid_w_resp = 0; + qp->s_num_rd_atomic = 0; /* * If we are starting the request from the beginning, @@ -961,9 +1522,9 @@ static void reset_psn(struct rvt_qp *qp, u32 psn) qp->s_state = OP(SEND_LAST); goto done; } + update_num_rd_atomic(qp, psn, wqe); /* Find the work request opcode corresponding to the given PSN. */ - opcode = wqe->wr.opcode; for (;;) { int diff; @@ -973,8 +1534,11 @@ static void reset_psn(struct rvt_qp *qp, u32 psn) break; wqe = rvt_get_swqe_ptr(qp, n); diff = cmp_psn(psn, wqe->psn); - if (diff < 0) + if (diff < 0) { + /* Point wqe back to the previous one*/ + wqe = rvt_get_swqe_ptr(qp, qp->s_cur); break; + } qp->s_cur = n; /* * If we are starting the request from the beginning, @@ -984,8 +1548,10 @@ static void reset_psn(struct rvt_qp *qp, u32 psn) qp->s_state = OP(SEND_LAST); goto done; } - opcode = wqe->wr.opcode; + + update_num_rd_atomic(qp, psn, wqe); } + opcode = wqe->wr.opcode; /* * Set the state to restart in the middle of a request. @@ -1003,10 +1569,18 @@ static void reset_psn(struct rvt_qp *qp, u32 psn) qp->s_state = OP(RDMA_READ_RESPONSE_LAST); break; + case IB_WR_TID_RDMA_WRITE: + qp->s_state = TID_OP(WRITE_RESP); + break; + case IB_WR_RDMA_READ: qp->s_state = OP(RDMA_READ_RESPONSE_MIDDLE); break; + case IB_WR_TID_RDMA_READ: + qp->s_state = TID_OP(READ_RESP); + break; + default: /* * This case shouldn't happen since its only @@ -1015,6 +1589,7 @@ static void reset_psn(struct rvt_qp *qp, u32 psn) qp->s_state = OP(SEND_LAST); } done: + priv->s_flags &= ~HFI1_S_TID_WAIT_INTERLCK; qp->s_psn = psn; /* * Set RVT_S_WAIT_PSN as rc_complete() may start the timer @@ -1025,6 +1600,7 @@ done: (cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)) qp->s_flags |= RVT_S_WAIT_PSN; qp->s_flags &= ~HFI1_S_AHG_VALID; + trace_hfi1_sender_reset_psn(qp); } /* @@ -1033,18 +1609,47 @@ done: */ void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait) { + struct hfi1_qp_priv *priv = qp->priv; struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, qp->s_acked); struct hfi1_ibport *ibp; lockdep_assert_held(&qp->r_lock); lockdep_assert_held(&qp->s_lock); + trace_hfi1_sender_restart_rc(qp); if (qp->s_retry == 0) { if (qp->s_mig_state == IB_MIG_ARMED) { hfi1_migrate_qp(qp); qp->s_retry = qp->s_retry_cnt; } else if (qp->s_last == qp->s_acked) { - rvt_send_complete(qp, wqe, IB_WC_RETRY_EXC_ERR); - rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR); + /* + * We need special handling for the OPFN request WQEs as + * they are not allowed to generate real user errors + */ + if (wqe->wr.opcode == IB_WR_OPFN) { + struct hfi1_ibport *ibp = + to_iport(qp->ibqp.device, qp->port_num); + /* + * Call opfn_conn_reply() with capcode and + * remaining data as 0 to close out the + * current request + */ + opfn_conn_reply(qp, priv->opfn.curr); + wqe = do_rc_completion(qp, wqe, ibp); + qp->s_flags &= ~RVT_S_WAIT_ACK; + } else { + trace_hfi1_tid_write_sender_restart_rc(qp, 0); + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) { + struct tid_rdma_request *req; + + req = wqe_to_tid_req(wqe); + hfi1_kern_exp_rcv_clear_all(req); + hfi1_kern_clear_hw_flow(priv->rcd, qp); + } + + hfi1_trdma_send_complete(qp, wqe, + IB_WC_RETRY_EXC_ERR); + rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR); + } return; } else { /* need to handle delayed completion */ return; @@ -1054,14 +1659,15 @@ void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait) } ibp = to_iport(qp->ibqp.device, qp->port_num); - if (wqe->wr.opcode == IB_WR_RDMA_READ) + if (wqe->wr.opcode == IB_WR_RDMA_READ || + wqe->wr.opcode == IB_WR_TID_RDMA_READ) ibp->rvp.n_rc_resends++; else ibp->rvp.n_rc_resends += delta_psn(qp->s_psn, psn); qp->s_flags &= ~(RVT_S_WAIT_FENCE | RVT_S_WAIT_RDMAR | RVT_S_WAIT_SSN_CREDIT | RVT_S_WAIT_PSN | - RVT_S_WAIT_ACK); + RVT_S_WAIT_ACK | HFI1_S_WAIT_TID_RESP); if (wait) qp->s_flags |= RVT_S_SEND_ONE; reset_psn(qp, psn); @@ -1069,7 +1675,8 @@ void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait) /* * Set qp->s_sending_psn to the next PSN after the given one. - * This would be psn+1 except when RDMA reads are present. + * This would be psn+1 except when RDMA reads or TID RDMA ops + * are present. */ static void reset_sending_psn(struct rvt_qp *qp, u32 psn) { @@ -1081,7 +1688,9 @@ static void reset_sending_psn(struct rvt_qp *qp, u32 psn) for (;;) { wqe = rvt_get_swqe_ptr(qp, n); if (cmp_psn(psn, wqe->lpsn) <= 0) { - if (wqe->wr.opcode == IB_WR_RDMA_READ) + if (wqe->wr.opcode == IB_WR_RDMA_READ || + wqe->wr.opcode == IB_WR_TID_RDMA_READ || + wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) qp->s_sending_psn = wqe->lpsn + 1; else qp->s_sending_psn = psn + 1; @@ -1104,8 +1713,9 @@ void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah) struct rvt_swqe *wqe; struct ib_header *hdr = NULL; struct hfi1_16b_header *hdr_16b = NULL; - u32 opcode; + u32 opcode, head, tail; u32 psn; + struct tid_rdma_request *req; lockdep_assert_held(&qp->s_lock); if (!(ib_rvt_state_ops[qp->state] & RVT_SEND_OR_FLUSH_OR_RECV_OK)) @@ -1130,25 +1740,85 @@ void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah) } opcode = ib_bth_get_opcode(ohdr); - if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) && - opcode <= OP(ATOMIC_ACKNOWLEDGE)) { + if ((opcode >= OP(RDMA_READ_RESPONSE_FIRST) && + opcode <= OP(ATOMIC_ACKNOWLEDGE)) || + opcode == TID_OP(READ_RESP) || + opcode == TID_OP(WRITE_RESP)) { WARN_ON(!qp->s_rdma_ack_cnt); qp->s_rdma_ack_cnt--; return; } psn = ib_bth_get_psn(ohdr); - reset_sending_psn(qp, psn); + /* + * Don't attempt to reset the sending PSN for packets in the + * KDETH PSN space since the PSN does not match anything. + */ + if (opcode != TID_OP(WRITE_DATA) && + opcode != TID_OP(WRITE_DATA_LAST) && + opcode != TID_OP(ACK) && opcode != TID_OP(RESYNC)) + reset_sending_psn(qp, psn); + + /* Handle TID RDMA WRITE packets differently */ + if (opcode >= TID_OP(WRITE_REQ) && + opcode <= TID_OP(WRITE_DATA_LAST)) { + head = priv->s_tid_head; + tail = priv->s_tid_cur; + /* + * s_tid_cur is set to s_tid_head in the case, where + * a new TID RDMA request is being started and all + * previous ones have been completed. + * Therefore, we need to do a secondary check in order + * to properly determine whether we should start the + * RC timer. + */ + wqe = rvt_get_swqe_ptr(qp, tail); + req = wqe_to_tid_req(wqe); + if (head == tail && req->comp_seg < req->total_segs) { + if (tail == 0) + tail = qp->s_size - 1; + else + tail -= 1; + } + } else { + head = qp->s_tail; + tail = qp->s_acked; + } /* * Start timer after a packet requesting an ACK has been sent and * there are still requests that haven't been acked. */ - if ((psn & IB_BTH_REQ_ACK) && qp->s_acked != qp->s_tail && + if ((psn & IB_BTH_REQ_ACK) && tail != head && + opcode != TID_OP(WRITE_DATA) && opcode != TID_OP(WRITE_DATA_LAST) && + opcode != TID_OP(RESYNC) && !(qp->s_flags & - (RVT_S_TIMER | RVT_S_WAIT_RNR | RVT_S_WAIT_PSN)) && - (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) - rvt_add_retry_timer(qp); + (RVT_S_TIMER | RVT_S_WAIT_RNR | RVT_S_WAIT_PSN)) && + (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) { + if (opcode == TID_OP(READ_REQ)) + rvt_add_retry_timer_ext(qp, priv->timeout_shift); + else + rvt_add_retry_timer(qp); + } + + /* Start TID RDMA ACK timer */ + if ((opcode == TID_OP(WRITE_DATA) || + opcode == TID_OP(WRITE_DATA_LAST) || + opcode == TID_OP(RESYNC)) && + (psn & IB_BTH_REQ_ACK) && + !(priv->s_flags & HFI1_S_TID_RETRY_TIMER) && + (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) { + /* + * The TID RDMA ACK packet could be received before this + * function is called. Therefore, add the timer only if TID + * RDMA ACK packets are actually pending. + */ + wqe = rvt_get_swqe_ptr(qp, qp->s_acked); + req = wqe_to_tid_req(wqe); + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE && + req->ack_seg < req->cur_seg) + hfi1_add_tid_retry_timer(qp); + } while (qp->s_last != qp->s_acked) { u32 s_last; @@ -1157,6 +1827,7 @@ void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_opa_header *opah) if (cmp_psn(wqe->lpsn, qp->s_sending_psn) >= 0 && cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) <= 0) break; + trdma_clean_swqe(qp, wqe); rvt_qp_wqe_unreserve(qp, wqe); s_last = qp->s_last; trace_hfi1_qp_send_completion(qp, wqe, s_last); @@ -1195,20 +1866,24 @@ static inline void update_last_psn(struct rvt_qp *qp, u32 psn) * This is similar to hfi1_send_complete but has to check to be sure * that the SGEs are not being referenced if the SWQE is being resent. */ -static struct rvt_swqe *do_rc_completion(struct rvt_qp *qp, - struct rvt_swqe *wqe, - struct hfi1_ibport *ibp) +struct rvt_swqe *do_rc_completion(struct rvt_qp *qp, + struct rvt_swqe *wqe, + struct hfi1_ibport *ibp) { + struct hfi1_qp_priv *priv = qp->priv; + lockdep_assert_held(&qp->s_lock); /* * Don't decrement refcount and don't generate a * completion if the SWQE is being resent until the send * is finished. */ + trace_hfi1_rc_completion(qp, wqe->lpsn); if (cmp_psn(wqe->lpsn, qp->s_sending_psn) < 0 || cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) > 0) { u32 s_last; + trdma_clean_swqe(qp, wqe); rvt_put_swqe(wqe); rvt_qp_wqe_unreserve(qp, wqe); s_last = qp->s_last; @@ -1243,7 +1918,16 @@ static struct rvt_swqe *do_rc_completion(struct rvt_qp *qp, } qp->s_retry = qp->s_retry_cnt; - update_last_psn(qp, wqe->lpsn); + /* + * Don't update the last PSN if the request being completed is + * a TID RDMA WRITE request. + * Completion of the TID RDMA WRITE requests are done by the + * TID RDMA ACKs and as such could be for a request that has + * already been ACKed as far as the IB state machine is + * concerned. + */ + if (wqe->wr.opcode != IB_WR_TID_RDMA_WRITE) + update_last_psn(qp, wqe->lpsn); /* * If we are completing a request which is in the process of @@ -1266,9 +1950,61 @@ static struct rvt_swqe *do_rc_completion(struct rvt_qp *qp, qp->s_draining = 0; wqe = rvt_get_swqe_ptr(qp, qp->s_acked); } + if (priv->s_flags & HFI1_S_TID_WAIT_INTERLCK) { + priv->s_flags &= ~HFI1_S_TID_WAIT_INTERLCK; + hfi1_schedule_send(qp); + } return wqe; } +static void set_restart_qp(struct rvt_qp *qp, struct hfi1_ctxtdata *rcd) +{ + /* Retry this request. */ + if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) { + qp->r_flags |= RVT_R_RDMAR_SEQ; + hfi1_restart_rc(qp, qp->s_last_psn + 1, 0); + if (list_empty(&qp->rspwait)) { + qp->r_flags |= RVT_R_RSP_SEND; + rvt_get_qp(qp); + list_add_tail(&qp->rspwait, &rcd->qp_wait_list); + } + } +} + +/** + * update_qp_retry_state - Update qp retry state. + * @qp: the QP + * @psn: the packet sequence number of the TID RDMA WRITE RESP. + * @spsn: The start psn for the given TID RDMA WRITE swqe. + * @lpsn: The last psn for the given TID RDMA WRITE swqe. + * + * This function is called to update the qp retry state upon + * receiving a TID WRITE RESP after the qp is scheduled to retry + * a request. + */ +static void update_qp_retry_state(struct rvt_qp *qp, u32 psn, u32 spsn, + u32 lpsn) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + + qp->s_psn = psn + 1; + /* + * If this is the first TID RDMA WRITE RESP packet for the current + * request, change the s_state so that the retry will be processed + * correctly. Similarly, if this is the last TID RDMA WRITE RESP + * packet, change the s_state and advance the s_cur. + */ + if (cmp_psn(psn, lpsn) >= 0) { + qp->s_cur = qpriv->s_tid_cur + 1; + if (qp->s_cur >= qp->s_size) + qp->s_cur = 0; + qp->s_state = TID_OP(WRITE_REQ); + } else if (!cmp_psn(psn, spsn)) { + qp->s_cur = qpriv->s_tid_cur; + qp->s_state = TID_OP(WRITE_RESP); + } +} + /** * do_rc_ack - process an incoming RC ACK * @qp: the QP the ACK came in on @@ -1280,15 +2016,17 @@ static struct rvt_swqe *do_rc_completion(struct rvt_qp *qp, * May be called at interrupt level, with the QP s_lock held. * Returns 1 if OK, 0 if current operation should be aborted (NAK). */ -static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, - u64 val, struct hfi1_ctxtdata *rcd) +int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, + u64 val, struct hfi1_ctxtdata *rcd) { struct hfi1_ibport *ibp; enum ib_wc_status status; + struct hfi1_qp_priv *qpriv = qp->priv; struct rvt_swqe *wqe; int ret = 0; u32 ack_psn; int diff; + struct rvt_dev_info *rdi; lockdep_assert_held(&qp->s_lock); /* @@ -1331,20 +2069,14 @@ static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, */ if ((wqe->wr.opcode == IB_WR_RDMA_READ && (opcode != OP(RDMA_READ_RESPONSE_LAST) || diff != 0)) || + (wqe->wr.opcode == IB_WR_TID_RDMA_READ && + (opcode != TID_OP(READ_RESP) || diff != 0)) || ((wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP || wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) && - (opcode != OP(ATOMIC_ACKNOWLEDGE) || diff != 0))) { - /* Retry this request. */ - if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) { - qp->r_flags |= RVT_R_RDMAR_SEQ; - hfi1_restart_rc(qp, qp->s_last_psn + 1, 0); - if (list_empty(&qp->rspwait)) { - qp->r_flags |= RVT_R_RSP_SEND; - rvt_get_qp(qp); - list_add_tail(&qp->rspwait, - &rcd->qp_wait_list); - } - } + (opcode != OP(ATOMIC_ACKNOWLEDGE) || diff != 0)) || + (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE && + (delta_psn(psn, qp->s_last_psn) != 1))) { + set_restart_qp(qp, rcd); /* * No need to process the ACK/NAK since we are * restarting an earlier request. @@ -1356,6 +2088,9 @@ static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, u64 *vaddr = wqe->sg_list[0].vaddr; *vaddr = val; } + if (wqe->wr.opcode == IB_WR_OPFN) + opfn_conn_reply(qp, val); + if (qp->s_num_rd_atomic && (wqe->wr.opcode == IB_WR_RDMA_READ || wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP || @@ -1373,26 +2108,85 @@ static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, hfi1_schedule_send(qp); } } + + /* + * TID RDMA WRITE requests will be completed by the TID RDMA + * ACK packet handler (see tid_rdma.c). + */ + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) + break; + wqe = do_rc_completion(qp, wqe, ibp); if (qp->s_acked == qp->s_tail) break; } + trace_hfi1_rc_ack_do(qp, aeth, psn, wqe); + trace_hfi1_sender_do_rc_ack(qp); switch (aeth >> IB_AETH_NAK_SHIFT) { case 0: /* ACK */ this_cpu_inc(*ibp->rvp.rc_acks); - if (qp->s_acked != qp->s_tail) { - /* - * We are expecting more ACKs so - * mod the retry timer. - */ - rvt_mod_retry_timer(qp); + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) { + if (wqe_to_tid_req(wqe)->ack_pending) + rvt_mod_retry_timer_ext(qp, + qpriv->timeout_shift); + else + rvt_stop_rc_timers(qp); + } else if (qp->s_acked != qp->s_tail) { + struct rvt_swqe *__w = NULL; + + if (qpriv->s_tid_cur != HFI1_QP_WQE_INVALID) + __w = rvt_get_swqe_ptr(qp, qpriv->s_tid_cur); + /* - * We can stop re-sending the earlier packets and - * continue with the next packet the receiver wants. + * Stop timers if we've received all of the TID RDMA + * WRITE * responses. */ - if (cmp_psn(qp->s_psn, psn) <= 0) - reset_psn(qp, psn + 1); + if (__w && __w->wr.opcode == IB_WR_TID_RDMA_WRITE && + opcode == TID_OP(WRITE_RESP)) { + /* + * Normally, the loop above would correctly + * process all WQEs from s_acked onward and + * either complete them or check for correct + * PSN sequencing. + * However, for TID RDMA, due to pipelining, + * the response may not be for the request at + * s_acked so the above look would just be + * skipped. This does not allow for checking + * the PSN sequencing. It has to be done + * separately. + */ + if (cmp_psn(psn, qp->s_last_psn + 1)) { + set_restart_qp(qp, rcd); + goto bail_stop; + } + /* + * If the psn is being resent, stop the + * resending. + */ + if (qp->s_cur != qp->s_tail && + cmp_psn(qp->s_psn, psn) <= 0) + update_qp_retry_state(qp, psn, + __w->psn, + __w->lpsn); + else if (--qpriv->pending_tid_w_resp) + rvt_mod_retry_timer(qp); + else + rvt_stop_rc_timers(qp); + } else { + /* + * We are expecting more ACKs so + * mod the retry timer. + */ + rvt_mod_retry_timer(qp); + /* + * We can stop re-sending the earlier packets + * and continue with the next packet the + * receiver wants. + */ + if (cmp_psn(qp->s_psn, psn) <= 0) + reset_psn(qp, psn + 1); + } } else { /* No more acks - kill all timers */ rvt_stop_rc_timers(qp); @@ -1408,6 +2202,15 @@ static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, rvt_get_credit(qp, aeth); qp->s_rnr_retry = qp->s_rnr_retry_cnt; qp->s_retry = qp->s_retry_cnt; + /* + * If the current request is a TID RDMA WRITE request and the + * response is not a TID RDMA WRITE RESP packet, s_last_psn + * can't be advanced. + */ + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE && + opcode != TID_OP(WRITE_RESP) && + cmp_psn(psn, wqe->psn) >= 0) + return 1; update_last_psn(qp, psn); return 1; @@ -1417,20 +2220,31 @@ static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, goto bail_stop; if (qp->s_flags & RVT_S_WAIT_RNR) goto bail_stop; - if (qp->s_rnr_retry == 0) { + rdi = ib_to_rvt(qp->ibqp.device); + if (qp->s_rnr_retry == 0 && + !((rdi->post_parms[wqe->wr.opcode].flags & + RVT_OPERATION_IGN_RNR_CNT) && + qp->s_rnr_retry_cnt == 0)) { status = IB_WC_RNR_RETRY_EXC_ERR; goto class_b; } - if (qp->s_rnr_retry_cnt < 7) + if (qp->s_rnr_retry_cnt < 7 && qp->s_rnr_retry_cnt > 0) qp->s_rnr_retry--; - /* The last valid PSN is the previous PSN. */ - update_last_psn(qp, psn - 1); + /* + * The last valid PSN is the previous PSN. For TID RDMA WRITE + * request, s_last_psn should be incremented only when a TID + * RDMA WRITE RESP is received to avoid skipping lost TID RDMA + * WRITE RESP packets. + */ + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) { + reset_psn(qp, qp->s_last_psn + 1); + } else { + update_last_psn(qp, psn - 1); + reset_psn(qp, psn); + } ibp->rvp.n_rc_resends += delta_psn(qp->s_psn, psn); - - reset_psn(qp, psn); - qp->s_flags &= ~(RVT_S_WAIT_SSN_CREDIT | RVT_S_WAIT_ACK); rvt_stop_rc_timers(qp); rvt_add_rnr_timer(qp, aeth); @@ -1470,7 +2284,10 @@ static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, ibp->rvp.n_other_naks++; class_b: if (qp->s_last == qp->s_acked) { - rvt_send_complete(qp, wqe, status); + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) + hfi1_kern_read_tid_flow_free(qp); + + hfi1_trdma_send_complete(qp, wqe, status); rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR); } break; @@ -1511,6 +2328,8 @@ static void rdma_seq_err(struct rvt_qp *qp, struct hfi1_ibport *ibp, u32 psn, while (cmp_psn(psn, wqe->lpsn) > 0) { if (wqe->wr.opcode == IB_WR_RDMA_READ || + wqe->wr.opcode == IB_WR_TID_RDMA_READ || + wqe->wr.opcode == IB_WR_TID_RDMA_WRITE || wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP || wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) break; @@ -1717,16 +2536,6 @@ bail: return; } -static inline void rc_defered_ack(struct hfi1_ctxtdata *rcd, - struct rvt_qp *qp) -{ - if (list_empty(&qp->rspwait)) { - qp->r_flags |= RVT_R_RSP_NAK; - rvt_get_qp(qp); - list_add_tail(&qp->rspwait, &rcd->qp_wait_list); - } -} - static inline void rc_cancel_ack(struct rvt_qp *qp) { qp->r_adefered = 0; @@ -1759,8 +2568,9 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data, struct hfi1_ibport *ibp = rcd_to_iport(rcd); struct rvt_ack_entry *e; unsigned long flags; - u8 i, prev; - int old_req; + u8 prev; + u8 mra; /* most recent ACK */ + bool old_req; trace_hfi1_rcv_error(qp, psn); if (diff > 0) { @@ -1806,29 +2616,8 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data, spin_lock_irqsave(&qp->s_lock, flags); - for (i = qp->r_head_ack_queue; ; i = prev) { - if (i == qp->s_tail_ack_queue) - old_req = 0; - if (i) - prev = i - 1; - else - prev = HFI1_MAX_RDMA_ATOMIC; - if (prev == qp->r_head_ack_queue) { - e = NULL; - break; - } - e = &qp->s_ack_queue[prev]; - if (!e->opcode) { - e = NULL; - break; - } - if (cmp_psn(psn, e->psn) >= 0) { - if (prev == qp->s_tail_ack_queue && - cmp_psn(psn, e->lpsn) <= 0) - old_req = 0; - break; - } - } + e = find_prev_entry(qp, psn, &prev, &mra, &old_req); + switch (opcode) { case OP(RDMA_READ_REQUEST): { struct ib_reth *reth; @@ -1875,6 +2664,8 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data, e->psn = psn; if (old_req) goto unlock_done; + if (qp->s_acked_ack_queue == qp->s_tail_ack_queue) + qp->s_acked_ack_queue = prev; qp->s_tail_ack_queue = prev; break; } @@ -1888,6 +2679,8 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data, */ if (!e || e->opcode != (u8)opcode || old_req) goto unlock_done; + if (qp->s_tail_ack_queue == qp->s_acked_ack_queue) + qp->s_acked_ack_queue = prev; qp->s_tail_ack_queue = prev; break; } @@ -1903,7 +2696,7 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data, * Resend the most recent ACK if this request is * after all the previous RDMA reads and atomics. */ - if (i == qp->r_head_ack_queue) { + if (mra == qp->r_head_ack_queue) { spin_unlock_irqrestore(&qp->s_lock, flags); qp->r_nak_state = 0; qp->r_ack_psn = qp->r_psn - 1; @@ -1914,7 +2707,9 @@ static noinline int rc_rcv_error(struct ib_other_headers *ohdr, void *data, * Resend the RDMA read or atomic op which * ACKs this duplicate request. */ - qp->s_tail_ack_queue = i; + if (qp->s_tail_ack_queue == qp->s_acked_ack_queue) + qp->s_acked_ack_queue = mra; + qp->s_tail_ack_queue = mra; break; } qp->s_ack_state = OP(ACKNOWLEDGE); @@ -1931,17 +2726,6 @@ send_ack: return 0; } -static inline void update_ack_queue(struct rvt_qp *qp, unsigned n) -{ - unsigned next; - - next = n + 1; - if (next > HFI1_MAX_RDMA_ATOMIC) - next = 0; - qp->s_tail_ack_queue = next; - qp->s_ack_state = OP(ACKNOWLEDGE); -} - static void log_cca_event(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn, u32 rqpn, u8 svc_type) { @@ -2039,6 +2823,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet) void *data = packet->payload; u32 tlen = packet->tlen; struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *qpriv = qp->priv; struct hfi1_ibport *ibp = rcd_to_iport(rcd); struct ib_other_headers *ohdr = packet->ohdr; u32 opcode = packet->opcode; @@ -2061,6 +2846,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet) return; fecn = process_ecn(qp, packet); + opfn_trigger_conn_request(qp, be32_to_cpu(ohdr->bth[1])); /* * Process responses (ACKs) before anything else. Note that the @@ -2292,11 +3078,11 @@ send_last: if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ))) goto nack_inv; next = qp->r_head_ack_queue + 1; - /* s_ack_queue is size HFI1_MAX_RDMA_ATOMIC+1 so use > not >= */ - if (next > HFI1_MAX_RDMA_ATOMIC) + /* s_ack_queue is size rvt_size_atomic()+1 so use > not >= */ + if (next > rvt_size_atomic(ib_to_rvt(qp->ibqp.device))) next = 0; spin_lock_irqsave(&qp->s_lock, flags); - if (unlikely(next == qp->s_tail_ack_queue)) { + if (unlikely(next == qp->s_acked_ack_queue)) { if (!qp->s_ack_queue[next].sent) goto nack_inv_unlck; update_ack_queue(qp, next); @@ -2343,6 +3129,7 @@ send_last: qp->r_state = opcode; qp->r_nak_state = 0; qp->r_head_ack_queue = next; + qpriv->r_tid_alloc = qp->r_head_ack_queue; /* Schedule the send engine. */ qp->s_flags |= RVT_S_RESP_PENDING; @@ -2356,21 +3143,24 @@ send_last: case OP(COMPARE_SWAP): case OP(FETCH_ADD): { - struct ib_atomic_eth *ateth; + struct ib_atomic_eth *ateth = &ohdr->u.atomic_eth; + u64 vaddr = get_ib_ateth_vaddr(ateth); + bool opfn = opcode == OP(COMPARE_SWAP) && + vaddr == HFI1_VERBS_E_ATOMIC_VADDR; struct rvt_ack_entry *e; - u64 vaddr; atomic64_t *maddr; u64 sdata; u32 rkey; u8 next; - if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC))) + if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC) && + !opfn)) goto nack_inv; next = qp->r_head_ack_queue + 1; - if (next > HFI1_MAX_RDMA_ATOMIC) + if (next > rvt_size_atomic(ib_to_rvt(qp->ibqp.device))) next = 0; spin_lock_irqsave(&qp->s_lock, flags); - if (unlikely(next == qp->s_tail_ack_queue)) { + if (unlikely(next == qp->s_acked_ack_queue)) { if (!qp->s_ack_queue[next].sent) goto nack_inv_unlck; update_ack_queue(qp, next); @@ -2380,8 +3170,11 @@ send_last: rvt_put_mr(e->rdma_sge.mr); e->rdma_sge.mr = NULL; } - ateth = &ohdr->u.atomic_eth; - vaddr = get_ib_ateth_vaddr(ateth); + /* Process OPFN special virtual address */ + if (opfn) { + opfn_conn_response(qp, e, ateth); + goto ack; + } if (unlikely(vaddr & (sizeof(u64) - 1))) goto nack_inv_unlck; rkey = be32_to_cpu(ateth->rkey); @@ -2400,6 +3193,7 @@ send_last: sdata); rvt_put_mr(qp->r_sge.sge.mr); qp->r_sge.num_sge = 0; +ack: e->opcode = opcode; e->sent = 0; e->psn = psn; @@ -2409,6 +3203,7 @@ send_last: qp->r_state = opcode; qp->r_nak_state = 0; qp->r_head_ack_queue = next; + qpriv->r_tid_alloc = qp->r_head_ack_queue; /* Schedule the send engine. */ qp->s_flags |= RVT_S_RESP_PENDING; diff --git a/drivers/infiniband/hw/hfi1/rc.h b/drivers/infiniband/hw/hfi1/rc.h new file mode 100644 index 000000000000..8e0935b9bf2a --- /dev/null +++ b/drivers/infiniband/hw/hfi1/rc.h @@ -0,0 +1,51 @@ +/* SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) */ +/* + * Copyright(c) 2018 Intel Corporation. + * + */ + +#ifndef HFI1_RC_H +#define HFI1_RC_H + +/* cut down ridiculously long IB macro names */ +#define OP(x) IB_OPCODE_RC_##x + +static inline void update_ack_queue(struct rvt_qp *qp, unsigned int n) +{ + unsigned int next; + + next = n + 1; + if (next > rvt_size_atomic(ib_to_rvt(qp->ibqp.device))) + next = 0; + qp->s_tail_ack_queue = next; + qp->s_acked_ack_queue = next; + qp->s_ack_state = OP(ACKNOWLEDGE); +} + +static inline void rc_defered_ack(struct hfi1_ctxtdata *rcd, + struct rvt_qp *qp) +{ + if (list_empty(&qp->rspwait)) { + qp->r_flags |= RVT_R_RSP_NAK; + rvt_get_qp(qp); + list_add_tail(&qp->rspwait, &rcd->qp_wait_list); + } +} + +static inline u32 restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe, + u32 psn, u32 pmtu) +{ + u32 len; + + len = delta_psn(psn, wqe->psn) * pmtu; + return rvt_restart_sge(ss, wqe, len); +} + +struct rvt_ack_entry *find_prev_entry(struct rvt_qp *qp, u32 psn, u8 *prev, + u8 *prev_ack, bool *scheduled); +int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, u64 val, + struct hfi1_ctxtdata *rcd); +struct rvt_swqe *do_rc_completion(struct rvt_qp *qp, struct rvt_swqe *wqe, + struct hfi1_ibport *ibp); + +#endif /* HFI1_RC_H */ diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c index 7fb317c711df..124a3ec1e15c 100644 --- a/drivers/infiniband/hw/hfi1/ruc.c +++ b/drivers/infiniband/hw/hfi1/ruc.c @@ -250,7 +250,6 @@ static inline void hfi1_make_ruc_bth(struct rvt_qp *qp, struct ib_other_headers *ohdr, u32 bth0, u32 bth1, u32 bth2) { - bth1 |= qp->remote_qpn; ohdr->bth[0] = cpu_to_be32(bth0); ohdr->bth[1] = cpu_to_be32(bth1); ohdr->bth[2] = cpu_to_be32(bth2); @@ -272,13 +271,13 @@ static inline void hfi1_make_ruc_bth(struct rvt_qp *qp, */ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp, struct ib_other_headers *ohdr, - u32 bth0, u32 bth2, int middle, + u32 bth0, u32 bth1, u32 bth2, + int middle, struct hfi1_pkt_state *ps) { struct hfi1_qp_priv *priv = qp->priv; struct hfi1_ibport *ibp = ps->ibp; struct hfi1_pportdata *ppd = ppd_from_ibp(ibp); - u32 bth1 = 0; u32 slid; u16 pkey = hfi1_get_pkey(ibp, qp->s_pkey_index); u8 l4 = OPA_16B_L4_IB_LOCAL; @@ -360,12 +359,12 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp, */ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp, struct ib_other_headers *ohdr, - u32 bth0, u32 bth2, int middle, + u32 bth0, u32 bth1, u32 bth2, + int middle, struct hfi1_pkt_state *ps) { struct hfi1_qp_priv *priv = qp->priv; struct hfi1_ibport *ibp = ps->ibp; - u32 bth1 = 0; u16 pkey = hfi1_get_pkey(ibp, qp->s_pkey_index); u16 lrh0 = HFI1_LRH_BTH; u8 extra_bytes = -ps->s_txreq->s_cur_size & 3; @@ -415,7 +414,7 @@ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp, typedef void (*hfi1_make_ruc_hdr)(struct rvt_qp *qp, struct ib_other_headers *ohdr, - u32 bth0, u32 bth2, int middle, + u32 bth0, u32 bth1, u32 bth2, int middle, struct hfi1_pkt_state *ps); /* We support only two types - 9B and 16B for now */ @@ -425,7 +424,7 @@ static const hfi1_make_ruc_hdr hfi1_ruc_header_tbl[2] = { }; void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr, - u32 bth0, u32 bth2, int middle, + u32 bth0, u32 bth1, u32 bth2, int middle, struct hfi1_pkt_state *ps) { struct hfi1_qp_priv *priv = qp->priv; @@ -446,18 +445,21 @@ void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr, priv->s_ahg->ahgidx = 0; /* Make the appropriate header */ - hfi1_ruc_header_tbl[priv->hdr_type](qp, ohdr, bth0, bth2, middle, ps); + hfi1_ruc_header_tbl[priv->hdr_type](qp, ohdr, bth0, bth1, bth2, middle, + ps); } /* when sending, force a reschedule every one of these periods */ #define SEND_RESCHED_TIMEOUT (5 * HZ) /* 5s in jiffies */ /** - * schedule_send_yield - test for a yield required for QP send engine + * hfi1_schedule_send_yield - test for a yield required for QP + * send engine * @timeout: Final time for timeout slice for jiffies * @qp: a pointer to QP * @ps: a pointer to a structure with commonly lookup values for * the the send engine progress + * @tid - true if it is the tid leg * * This routine checks if the time slice for the QP has expired * for RC QPs, if so an additional work entry is queued. At this @@ -465,8 +467,8 @@ void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr, * returns true if a yield is required, otherwise, false * is returned. */ -static bool schedule_send_yield(struct rvt_qp *qp, - struct hfi1_pkt_state *ps) +bool hfi1_schedule_send_yield(struct rvt_qp *qp, struct hfi1_pkt_state *ps, + bool tid) { ps->pkts_sent = true; @@ -474,8 +476,24 @@ static bool schedule_send_yield(struct rvt_qp *qp, if (!ps->in_thread || workqueue_congested(ps->cpu, ps->ppd->hfi1_wq)) { spin_lock_irqsave(&qp->s_lock, ps->flags); - qp->s_flags &= ~RVT_S_BUSY; - hfi1_schedule_send(qp); + if (!tid) { + qp->s_flags &= ~RVT_S_BUSY; + hfi1_schedule_send(qp); + } else { + struct hfi1_qp_priv *priv = qp->priv; + + if (priv->s_flags & + HFI1_S_TID_BUSY_SET) { + qp->s_flags &= ~RVT_S_BUSY; + priv->s_flags &= + ~(HFI1_S_TID_BUSY_SET | + RVT_S_BUSY); + } else { + priv->s_flags &= ~RVT_S_BUSY; + } + hfi1_schedule_tid_send(qp); + } + spin_unlock_irqrestore(&qp->s_lock, ps->flags); this_cpu_inc(*ps->ppd->dd->send_schedule); trace_hfi1_rc_expired_time_slice(qp, true); @@ -576,6 +594,8 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread) do { /* Check for a constructed packet to be sent. */ if (ps.s_txreq) { + if (priv->s_flags & HFI1_S_TID_BUSY_SET) + qp->s_flags |= RVT_S_BUSY; spin_unlock_irqrestore(&qp->s_lock, ps.flags); /* * If the packet cannot be sent now, return and @@ -585,7 +605,7 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread) return; /* allow other tasks to run */ - if (schedule_send_yield(qp, &ps)) + if (hfi1_schedule_send_yield(qp, &ps, false)) return; spin_lock_irqsave(&qp->s_lock, ps.flags); diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c index 96897a91fb0a..b0110728f541 100644 --- a/drivers/infiniband/hw/hfi1/sdma.c +++ b/drivers/infiniband/hw/hfi1/sdma.c @@ -1747,10 +1747,9 @@ retry: */ static void sdma_desc_avail(struct sdma_engine *sde, uint avail) { - struct iowait *wait, *nw; + struct iowait *wait, *nw, *twait; struct iowait *waits[SDMA_WAIT_BATCH_SIZE]; - uint i, n = 0, seq, max_idx = 0; - u8 max_starved_cnt = 0; + uint i, n = 0, seq, tidx = 0; #ifdef CONFIG_SDMA_VERBOSITY dd_dev_err(sde->dd, "CONFIG SDMA(%u) %s:%d %s()\n", sde->this_idx, @@ -1775,13 +1774,20 @@ static void sdma_desc_avail(struct sdma_engine *sde, uint avail) continue; if (n == ARRAY_SIZE(waits)) break; + iowait_init_priority(wait); num_desc = iowait_get_all_desc(wait); if (num_desc > avail) break; avail -= num_desc; - /* Find the most starved wait memeber */ - iowait_starve_find_max(wait, &max_starved_cnt, - n, &max_idx); + /* Find the top-priority wait memeber */ + if (n) { + twait = waits[tidx]; + tidx = + iowait_priority_update_top(wait, + twait, + n, + tidx); + } list_del_init(&wait->list); waits[n++] = wait; } @@ -1790,12 +1796,12 @@ static void sdma_desc_avail(struct sdma_engine *sde, uint avail) } } while (read_seqretry(&sde->waitlock, seq)); - /* Schedule the most starved one first */ + /* Schedule the top-priority entry first */ if (n) - waits[max_idx]->wakeup(waits[max_idx], SDMA_AVAIL_REASON); + waits[tidx]->wakeup(waits[tidx], SDMA_AVAIL_REASON); for (i = 0; i < n; i++) - if (i != max_idx) + if (i != tidx) waits[i]->wakeup(waits[i], SDMA_AVAIL_REASON); } diff --git a/drivers/infiniband/hw/hfi1/sdma_txreq.h b/drivers/infiniband/hw/hfi1/sdma_txreq.h index bf7d777d756e..514a4784566b 100644 --- a/drivers/infiniband/hw/hfi1/sdma_txreq.h +++ b/drivers/infiniband/hw/hfi1/sdma_txreq.h @@ -91,6 +91,7 @@ struct sdma_desc { #define SDMA_TXREQ_F_URGENT 0x0001 #define SDMA_TXREQ_F_AHG_COPY 0x0002 #define SDMA_TXREQ_F_USE_AHG 0x0004 +#define SDMA_TXREQ_F_VIP 0x0010 struct sdma_txreq; typedef void (*callback_t)(struct sdma_txreq *, int); diff --git a/drivers/infiniband/hw/hfi1/sysfs.c b/drivers/infiniband/hw/hfi1/sysfs.c index 2be513d4c9da..90f62c4bddba 100644 --- a/drivers/infiniband/hw/hfi1/sysfs.c +++ b/drivers/infiniband/hw/hfi1/sysfs.c @@ -498,7 +498,7 @@ static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); return sprintf(buf, "%x\n", dd_from_dev(dev)->minrev); } @@ -508,7 +508,7 @@ static ssize_t board_id_show(struct device *device, struct device_attribute *attr, char *buf) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); struct hfi1_devdata *dd = dd_from_dev(dev); int ret; @@ -524,7 +524,7 @@ static ssize_t boardversion_show(struct device *device, struct device_attribute *attr, char *buf) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); struct hfi1_devdata *dd = dd_from_dev(dev); /* The string printed here is already newline-terminated. */ @@ -536,7 +536,7 @@ static ssize_t nctxts_show(struct device *device, struct device_attribute *attr, char *buf) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); struct hfi1_devdata *dd = dd_from_dev(dev); /* @@ -555,7 +555,7 @@ static ssize_t nfreectxts_show(struct device *device, struct device_attribute *attr, char *buf) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); struct hfi1_devdata *dd = dd_from_dev(dev); /* Return the number of free user ports (contexts) available. */ @@ -567,7 +567,7 @@ static ssize_t serial_show(struct device *device, struct device_attribute *attr, char *buf) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); struct hfi1_devdata *dd = dd_from_dev(dev); return scnprintf(buf, PAGE_SIZE, "%s", dd->serial); @@ -579,7 +579,7 @@ static ssize_t chip_reset_store(struct device *device, size_t count) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); struct hfi1_devdata *dd = dd_from_dev(dev); int ret; @@ -609,7 +609,7 @@ static ssize_t tempsense_show(struct device *device, struct device_attribute *attr, char *buf) { struct hfi1_ibdev *dev = - container_of(device, struct hfi1_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct hfi1_ibdev, rdi.ibdev); struct hfi1_devdata *dd = dd_from_dev(dev); struct hfi1_temp temp; int ret; diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.c b/drivers/infiniband/hw/hfi1/tid_rdma.c index da1ecb68a928..fdda33aca77f 100644 --- a/drivers/infiniband/hw/hfi1/tid_rdma.c +++ b/drivers/infiniband/hw/hfi1/tid_rdma.c @@ -5,8 +5,282 @@ */ #include "hfi.h" +#include "qp.h" +#include "rc.h" #include "verbs.h" #include "tid_rdma.h" +#include "exp_rcv.h" +#include "trace.h" + +/** + * DOC: TID RDMA READ protocol + * + * This is an end-to-end protocol at the hfi1 level between two nodes that + * improves performance by avoiding data copy on the requester side. It + * converts a qualified RDMA READ request into a TID RDMA READ request on + * the requester side and thereafter handles the request and response + * differently. To be qualified, the RDMA READ request should meet the + * following: + * -- The total data length should be greater than 256K; + * -- The total data length should be a multiple of 4K page size; + * -- Each local scatter-gather entry should be 4K page aligned; + * -- Each local scatter-gather entry should be a multiple of 4K page size; + */ + +#define RCV_TID_FLOW_TABLE_CTRL_FLOW_VALID_SMASK BIT_ULL(32) +#define RCV_TID_FLOW_TABLE_CTRL_HDR_SUPP_EN_SMASK BIT_ULL(33) +#define RCV_TID_FLOW_TABLE_CTRL_KEEP_AFTER_SEQ_ERR_SMASK BIT_ULL(34) +#define RCV_TID_FLOW_TABLE_CTRL_KEEP_ON_GEN_ERR_SMASK BIT_ULL(35) +#define RCV_TID_FLOW_TABLE_STATUS_SEQ_MISMATCH_SMASK BIT_ULL(37) +#define RCV_TID_FLOW_TABLE_STATUS_GEN_MISMATCH_SMASK BIT_ULL(38) + +/* Maximum number of packets within a flow generation. */ +#define MAX_TID_FLOW_PSN BIT(HFI1_KDETH_BTH_SEQ_SHIFT) + +#define GENERATION_MASK 0xFFFFF + +static u32 mask_generation(u32 a) +{ + return a & GENERATION_MASK; +} + +/* Reserved generation value to set to unused flows for kernel contexts */ +#define KERN_GENERATION_RESERVED mask_generation(U32_MAX) + +/* + * J_KEY for kernel contexts when TID RDMA is used. + * See generate_jkey() in hfi.h for more information. + */ +#define TID_RDMA_JKEY 32 +#define HFI1_KERNEL_MIN_JKEY HFI1_ADMIN_JKEY_RANGE +#define HFI1_KERNEL_MAX_JKEY (2 * HFI1_ADMIN_JKEY_RANGE - 1) + +/* Maximum number of segments in flight per QP request. */ +#define TID_RDMA_MAX_READ_SEGS_PER_REQ 6 +#define TID_RDMA_MAX_WRITE_SEGS_PER_REQ 4 +#define MAX_REQ max_t(u16, TID_RDMA_MAX_READ_SEGS_PER_REQ, \ + TID_RDMA_MAX_WRITE_SEGS_PER_REQ) +#define MAX_FLOWS roundup_pow_of_two(MAX_REQ + 1) + +#define MAX_EXPECTED_PAGES (MAX_EXPECTED_BUFFER / PAGE_SIZE) + +#define TID_RDMA_DESTQP_FLOW_SHIFT 11 +#define TID_RDMA_DESTQP_FLOW_MASK 0x1f + +#define TID_FLOW_SW_PSN BIT(0) + +#define TID_OPFN_QP_CTXT_MASK 0xff +#define TID_OPFN_QP_CTXT_SHIFT 56 +#define TID_OPFN_QP_KDETH_MASK 0xff +#define TID_OPFN_QP_KDETH_SHIFT 48 +#define TID_OPFN_MAX_LEN_MASK 0x7ff +#define TID_OPFN_MAX_LEN_SHIFT 37 +#define TID_OPFN_TIMEOUT_MASK 0x1f +#define TID_OPFN_TIMEOUT_SHIFT 32 +#define TID_OPFN_RESERVED_MASK 0x3f +#define TID_OPFN_RESERVED_SHIFT 26 +#define TID_OPFN_URG_MASK 0x1 +#define TID_OPFN_URG_SHIFT 25 +#define TID_OPFN_VER_MASK 0x7 +#define TID_OPFN_VER_SHIFT 22 +#define TID_OPFN_JKEY_MASK 0x3f +#define TID_OPFN_JKEY_SHIFT 16 +#define TID_OPFN_MAX_READ_MASK 0x3f +#define TID_OPFN_MAX_READ_SHIFT 10 +#define TID_OPFN_MAX_WRITE_MASK 0x3f +#define TID_OPFN_MAX_WRITE_SHIFT 4 + +/* + * OPFN TID layout + * + * 63 47 31 15 + * NNNNNNNNKKKKKKKK MMMMMMMMMMMTTTTT DDDDDDUVVVJJJJJJ RRRRRRWWWWWWCCCC + * 3210987654321098 7654321098765432 1098765432109876 5432109876543210 + * N - the context Number + * K - the Kdeth_qp + * M - Max_len + * T - Timeout + * D - reserveD + * V - version + * U - Urg capable + * J - Jkey + * R - max_Read + * W - max_Write + * C - Capcode + */ + +static u32 tid_rdma_flow_wt; + +static void tid_rdma_trigger_resume(struct work_struct *work); +static void hfi1_kern_exp_rcv_free_flows(struct tid_rdma_request *req); +static int hfi1_kern_exp_rcv_alloc_flows(struct tid_rdma_request *req, + gfp_t gfp); +static void hfi1_init_trdma_req(struct rvt_qp *qp, + struct tid_rdma_request *req); +static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx); +static void hfi1_tid_timeout(struct timer_list *t); +static void hfi1_add_tid_reap_timer(struct rvt_qp *qp); +static void hfi1_mod_tid_reap_timer(struct rvt_qp *qp); +static void hfi1_mod_tid_retry_timer(struct rvt_qp *qp); +static int hfi1_stop_tid_retry_timer(struct rvt_qp *qp); +static void hfi1_tid_retry_timeout(struct timer_list *t); +static int make_tid_rdma_ack(struct rvt_qp *qp, + struct ib_other_headers *ohdr, + struct hfi1_pkt_state *ps); +static void hfi1_do_tid_send(struct rvt_qp *qp); + +static u64 tid_rdma_opfn_encode(struct tid_rdma_params *p) +{ + return + (((u64)p->qp & TID_OPFN_QP_CTXT_MASK) << + TID_OPFN_QP_CTXT_SHIFT) | + ((((u64)p->qp >> 16) & TID_OPFN_QP_KDETH_MASK) << + TID_OPFN_QP_KDETH_SHIFT) | + (((u64)((p->max_len >> PAGE_SHIFT) - 1) & + TID_OPFN_MAX_LEN_MASK) << TID_OPFN_MAX_LEN_SHIFT) | + (((u64)p->timeout & TID_OPFN_TIMEOUT_MASK) << + TID_OPFN_TIMEOUT_SHIFT) | + (((u64)p->urg & TID_OPFN_URG_MASK) << TID_OPFN_URG_SHIFT) | + (((u64)p->jkey & TID_OPFN_JKEY_MASK) << TID_OPFN_JKEY_SHIFT) | + (((u64)p->max_read & TID_OPFN_MAX_READ_MASK) << + TID_OPFN_MAX_READ_SHIFT) | + (((u64)p->max_write & TID_OPFN_MAX_WRITE_MASK) << + TID_OPFN_MAX_WRITE_SHIFT); +} + +static void tid_rdma_opfn_decode(struct tid_rdma_params *p, u64 data) +{ + p->max_len = (((data >> TID_OPFN_MAX_LEN_SHIFT) & + TID_OPFN_MAX_LEN_MASK) + 1) << PAGE_SHIFT; + p->jkey = (data >> TID_OPFN_JKEY_SHIFT) & TID_OPFN_JKEY_MASK; + p->max_write = (data >> TID_OPFN_MAX_WRITE_SHIFT) & + TID_OPFN_MAX_WRITE_MASK; + p->max_read = (data >> TID_OPFN_MAX_READ_SHIFT) & + TID_OPFN_MAX_READ_MASK; + p->qp = + ((((data >> TID_OPFN_QP_KDETH_SHIFT) & TID_OPFN_QP_KDETH_MASK) + << 16) | + ((data >> TID_OPFN_QP_CTXT_SHIFT) & TID_OPFN_QP_CTXT_MASK)); + p->urg = (data >> TID_OPFN_URG_SHIFT) & TID_OPFN_URG_MASK; + p->timeout = (data >> TID_OPFN_TIMEOUT_SHIFT) & TID_OPFN_TIMEOUT_MASK; +} + +void tid_rdma_opfn_init(struct rvt_qp *qp, struct tid_rdma_params *p) +{ + struct hfi1_qp_priv *priv = qp->priv; + + p->qp = (kdeth_qp << 16) | priv->rcd->ctxt; + p->max_len = TID_RDMA_MAX_SEGMENT_SIZE; + p->jkey = priv->rcd->jkey; + p->max_read = TID_RDMA_MAX_READ_SEGS_PER_REQ; + p->max_write = TID_RDMA_MAX_WRITE_SEGS_PER_REQ; + p->timeout = qp->timeout; + p->urg = is_urg_masked(priv->rcd); +} + +bool tid_rdma_conn_req(struct rvt_qp *qp, u64 *data) +{ + struct hfi1_qp_priv *priv = qp->priv; + + *data = tid_rdma_opfn_encode(&priv->tid_rdma.local); + return true; +} + +bool tid_rdma_conn_reply(struct rvt_qp *qp, u64 data) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct tid_rdma_params *remote, *old; + bool ret = true; + + old = rcu_dereference_protected(priv->tid_rdma.remote, + lockdep_is_held(&priv->opfn.lock)); + data &= ~0xfULL; + /* + * If data passed in is zero, return true so as not to continue the + * negotiation process + */ + if (!data || !HFI1_CAP_IS_KSET(TID_RDMA)) + goto null; + /* + * If kzalloc fails, return false. This will result in: + * * at the requester a new OPFN request being generated to retry + * the negotiation + * * at the responder, 0 being returned to the requester so as to + * disable TID RDMA at both the requester and the responder + */ + remote = kzalloc(sizeof(*remote), GFP_ATOMIC); + if (!remote) { + ret = false; + goto null; + } + + tid_rdma_opfn_decode(remote, data); + priv->tid_timer_timeout_jiffies = + usecs_to_jiffies((((4096UL * (1UL << remote->timeout)) / + 1000UL) << 3) * 7); + trace_hfi1_opfn_param(qp, 0, &priv->tid_rdma.local); + trace_hfi1_opfn_param(qp, 1, remote); + rcu_assign_pointer(priv->tid_rdma.remote, remote); + /* + * A TID RDMA READ request's segment size is not equal to + * remote->max_len only when the request's data length is smaller + * than remote->max_len. In that case, there will be only one segment. + * Therefore, when priv->pkts_ps is used to calculate req->cur_seg + * during retry, it will lead to req->cur_seg = 0, which is exactly + * what is expected. + */ + priv->pkts_ps = (u16)rvt_div_mtu(qp, remote->max_len); + priv->timeout_shift = ilog2(priv->pkts_ps - 1) + 1; + goto free; +null: + RCU_INIT_POINTER(priv->tid_rdma.remote, NULL); + priv->timeout_shift = 0; +free: + if (old) + kfree_rcu(old, rcu_head); + return ret; +} + +bool tid_rdma_conn_resp(struct rvt_qp *qp, u64 *data) +{ + bool ret; + + ret = tid_rdma_conn_reply(qp, *data); + *data = 0; + /* + * If tid_rdma_conn_reply() returns error, set *data as 0 to indicate + * TID RDMA could not be enabled. This will result in TID RDMA being + * disabled at the requester too. + */ + if (ret) + (void)tid_rdma_conn_req(qp, data); + return ret; +} + +void tid_rdma_conn_error(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct tid_rdma_params *old; + + old = rcu_dereference_protected(priv->tid_rdma.remote, + lockdep_is_held(&priv->opfn.lock)); + RCU_INIT_POINTER(priv->tid_rdma.remote, NULL); + if (old) + kfree_rcu(old, rcu_head); +} + +/* This is called at context initialization time */ +int hfi1_kern_exp_rcv_init(struct hfi1_ctxtdata *rcd, int reinit) +{ + if (reinit) + return 0; + + BUILD_BUG_ON(TID_RDMA_JKEY < HFI1_KERNEL_MIN_JKEY); + BUILD_BUG_ON(TID_RDMA_JKEY > HFI1_KERNEL_MAX_JKEY); + rcd->jkey = TID_RDMA_JKEY; + hfi1_set_ctxt_jkey(rcd->dd, rcd, rcd->jkey); + return hfi1_alloc_ctxt_rcv_groups(rcd); +} /** * qp_to_rcd - determine the receive context used by a qp @@ -41,8 +315,5152 @@ int hfi1_qp_priv_init(struct rvt_dev_info *rdi, struct rvt_qp *qp, struct ib_qp_init_attr *init_attr) { struct hfi1_qp_priv *qpriv = qp->priv; + int i, ret; qpriv->rcd = qp_to_rcd(rdi, qp); + spin_lock_init(&qpriv->opfn.lock); + INIT_WORK(&qpriv->opfn.opfn_work, opfn_send_conn_request); + INIT_WORK(&qpriv->tid_rdma.trigger_work, tid_rdma_trigger_resume); + qpriv->flow_state.psn = 0; + qpriv->flow_state.index = RXE_NUM_TID_FLOWS; + qpriv->flow_state.last_index = RXE_NUM_TID_FLOWS; + qpriv->flow_state.generation = KERN_GENERATION_RESERVED; + qpriv->s_state = TID_OP(WRITE_RESP); + qpriv->s_tid_cur = HFI1_QP_WQE_INVALID; + qpriv->s_tid_head = HFI1_QP_WQE_INVALID; + qpriv->s_tid_tail = HFI1_QP_WQE_INVALID; + qpriv->rnr_nak_state = TID_RNR_NAK_INIT; + qpriv->r_tid_head = HFI1_QP_WQE_INVALID; + qpriv->r_tid_tail = HFI1_QP_WQE_INVALID; + qpriv->r_tid_ack = HFI1_QP_WQE_INVALID; + qpriv->r_tid_alloc = HFI1_QP_WQE_INVALID; + atomic_set(&qpriv->n_requests, 0); + atomic_set(&qpriv->n_tid_requests, 0); + timer_setup(&qpriv->s_tid_timer, hfi1_tid_timeout, 0); + timer_setup(&qpriv->s_tid_retry_timer, hfi1_tid_retry_timeout, 0); + INIT_LIST_HEAD(&qpriv->tid_wait); + + if (init_attr->qp_type == IB_QPT_RC && HFI1_CAP_IS_KSET(TID_RDMA)) { + struct hfi1_devdata *dd = qpriv->rcd->dd; + + qpriv->pages = kzalloc_node(TID_RDMA_MAX_PAGES * + sizeof(*qpriv->pages), + GFP_KERNEL, dd->node); + if (!qpriv->pages) + return -ENOMEM; + for (i = 0; i < qp->s_size; i++) { + struct hfi1_swqe_priv *priv; + struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, i); + + priv = kzalloc_node(sizeof(*priv), GFP_KERNEL, + dd->node); + if (!priv) + return -ENOMEM; + + hfi1_init_trdma_req(qp, &priv->tid_req); + priv->tid_req.e.swqe = wqe; + wqe->priv = priv; + } + for (i = 0; i < rvt_max_atomic(rdi); i++) { + struct hfi1_ack_priv *priv; + + priv = kzalloc_node(sizeof(*priv), GFP_KERNEL, + dd->node); + if (!priv) + return -ENOMEM; + + hfi1_init_trdma_req(qp, &priv->tid_req); + priv->tid_req.e.ack = &qp->s_ack_queue[i]; + + ret = hfi1_kern_exp_rcv_alloc_flows(&priv->tid_req, + GFP_KERNEL); + if (ret) { + kfree(priv); + return ret; + } + qp->s_ack_queue[i].priv = priv; + } + } + return 0; } + +void hfi1_qp_priv_tid_free(struct rvt_dev_info *rdi, struct rvt_qp *qp) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + struct rvt_swqe *wqe; + u32 i; + + if (qp->ibqp.qp_type == IB_QPT_RC && HFI1_CAP_IS_KSET(TID_RDMA)) { + for (i = 0; i < qp->s_size; i++) { + wqe = rvt_get_swqe_ptr(qp, i); + kfree(wqe->priv); + wqe->priv = NULL; + } + for (i = 0; i < rvt_max_atomic(rdi); i++) { + struct hfi1_ack_priv *priv = qp->s_ack_queue[i].priv; + + if (priv) + hfi1_kern_exp_rcv_free_flows(&priv->tid_req); + kfree(priv); + qp->s_ack_queue[i].priv = NULL; + } + cancel_work_sync(&qpriv->opfn.opfn_work); + kfree(qpriv->pages); + qpriv->pages = NULL; + } +} + +/* Flow and tid waiter functions */ +/** + * DOC: lock ordering + * + * There are two locks involved with the queuing + * routines: the qp s_lock and the exp_lock. + * + * Since the tid space allocation is called from + * the send engine, the qp s_lock is already held. + * + * The allocation routines will get the exp_lock. + * + * The first_qp() call is provided to allow the head of + * the rcd wait queue to be fetched under the exp_lock and + * followed by a drop of the exp_lock. + * + * Any qp in the wait list will have the qp reference count held + * to hold the qp in memory. + */ + +/* + * return head of rcd wait list + * + * Must hold the exp_lock. + * + * Get a reference to the QP to hold the QP in memory. + * + * The caller must release the reference when the local + * is no longer being used. + */ +static struct rvt_qp *first_qp(struct hfi1_ctxtdata *rcd, + struct tid_queue *queue) + __must_hold(&rcd->exp_lock) +{ + struct hfi1_qp_priv *priv; + + lockdep_assert_held(&rcd->exp_lock); + priv = list_first_entry_or_null(&queue->queue_head, + struct hfi1_qp_priv, + tid_wait); + if (!priv) + return NULL; + rvt_get_qp(priv->owner); + return priv->owner; +} + +/** + * kernel_tid_waiters - determine rcd wait + * @rcd: the receive context + * @qp: the head of the qp being processed + * + * This routine will return false IFF + * the list is NULL or the head of the + * list is the indicated qp. + * + * Must hold the qp s_lock and the exp_lock. + * + * Return: + * false if either of the conditions below are statisfied: + * 1. The list is empty or + * 2. The indicated qp is at the head of the list and the + * HFI1_S_WAIT_TID_SPACE bit is set in qp->s_flags. + * true is returned otherwise. + */ +static bool kernel_tid_waiters(struct hfi1_ctxtdata *rcd, + struct tid_queue *queue, struct rvt_qp *qp) + __must_hold(&rcd->exp_lock) __must_hold(&qp->s_lock) +{ + struct rvt_qp *fqp; + bool ret = true; + + lockdep_assert_held(&qp->s_lock); + lockdep_assert_held(&rcd->exp_lock); + fqp = first_qp(rcd, queue); + if (!fqp || (fqp == qp && (qp->s_flags & HFI1_S_WAIT_TID_SPACE))) + ret = false; + rvt_put_qp(fqp); + return ret; +} + +/** + * dequeue_tid_waiter - dequeue the qp from the list + * @qp - the qp to remove the wait list + * + * This routine removes the indicated qp from the + * wait list if it is there. + * + * This should be done after the hardware flow and + * tid array resources have been allocated. + * + * Must hold the qp s_lock and the rcd exp_lock. + * + * It assumes the s_lock to protect the s_flags + * field and to reliably test the HFI1_S_WAIT_TID_SPACE flag. + */ +static void dequeue_tid_waiter(struct hfi1_ctxtdata *rcd, + struct tid_queue *queue, struct rvt_qp *qp) + __must_hold(&rcd->exp_lock) __must_hold(&qp->s_lock) +{ + struct hfi1_qp_priv *priv = qp->priv; + + lockdep_assert_held(&qp->s_lock); + lockdep_assert_held(&rcd->exp_lock); + if (list_empty(&priv->tid_wait)) + return; + list_del_init(&priv->tid_wait); + qp->s_flags &= ~HFI1_S_WAIT_TID_SPACE; + queue->dequeue++; + rvt_put_qp(qp); +} + +/** + * queue_qp_for_tid_wait - suspend QP on tid space + * @rcd: the receive context + * @qp: the qp + * + * The qp is inserted at the tail of the rcd + * wait queue and the HFI1_S_WAIT_TID_SPACE s_flag is set. + * + * Must hold the qp s_lock and the exp_lock. + */ +static void queue_qp_for_tid_wait(struct hfi1_ctxtdata *rcd, + struct tid_queue *queue, struct rvt_qp *qp) + __must_hold(&rcd->exp_lock) __must_hold(&qp->s_lock) +{ + struct hfi1_qp_priv *priv = qp->priv; + + lockdep_assert_held(&qp->s_lock); + lockdep_assert_held(&rcd->exp_lock); + if (list_empty(&priv->tid_wait)) { + qp->s_flags |= HFI1_S_WAIT_TID_SPACE; + list_add_tail(&priv->tid_wait, &queue->queue_head); + priv->tid_enqueue = ++queue->enqueue; + rcd->dd->verbs_dev.n_tidwait++; + trace_hfi1_qpsleep(qp, HFI1_S_WAIT_TID_SPACE); + rvt_get_qp(qp); + } +} + +/** + * __trigger_tid_waiter - trigger tid waiter + * @qp: the qp + * + * This is a private entrance to schedule the qp + * assuming the caller is holding the qp->s_lock. + */ +static void __trigger_tid_waiter(struct rvt_qp *qp) + __must_hold(&qp->s_lock) +{ + lockdep_assert_held(&qp->s_lock); + if (!(qp->s_flags & HFI1_S_WAIT_TID_SPACE)) + return; + trace_hfi1_qpwakeup(qp, HFI1_S_WAIT_TID_SPACE); + hfi1_schedule_send(qp); +} + +/** + * tid_rdma_schedule_tid_wakeup - schedule wakeup for a qp + * @qp - the qp + * + * trigger a schedule or a waiting qp in a deadlock + * safe manner. The qp reference is held prior + * to this call via first_qp(). + * + * If the qp trigger was already scheduled (!rval) + * the the reference is dropped, otherwise the resume + * or the destroy cancel will dispatch the reference. + */ +static void tid_rdma_schedule_tid_wakeup(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv; + struct hfi1_ibport *ibp; + struct hfi1_pportdata *ppd; + struct hfi1_devdata *dd; + bool rval; + + if (!qp) + return; + + priv = qp->priv; + ibp = to_iport(qp->ibqp.device, qp->port_num); + ppd = ppd_from_ibp(ibp); + dd = dd_from_ibdev(qp->ibqp.device); + + rval = queue_work_on(priv->s_sde ? + priv->s_sde->cpu : + cpumask_first(cpumask_of_node(dd->node)), + ppd->hfi1_wq, + &priv->tid_rdma.trigger_work); + if (!rval) + rvt_put_qp(qp); +} + +/** + * tid_rdma_trigger_resume - field a trigger work request + * @work - the work item + * + * Complete the off qp trigger processing by directly + * calling the progress routine. + */ +static void tid_rdma_trigger_resume(struct work_struct *work) +{ + struct tid_rdma_qp_params *tr; + struct hfi1_qp_priv *priv; + struct rvt_qp *qp; + + tr = container_of(work, struct tid_rdma_qp_params, trigger_work); + priv = container_of(tr, struct hfi1_qp_priv, tid_rdma); + qp = priv->owner; + spin_lock_irq(&qp->s_lock); + if (qp->s_flags & HFI1_S_WAIT_TID_SPACE) { + spin_unlock_irq(&qp->s_lock); + hfi1_do_send(priv->owner, true); + } else { + spin_unlock_irq(&qp->s_lock); + } + rvt_put_qp(qp); +} + +/** + * tid_rdma_flush_wait - unwind any tid space wait + * + * This is called when resetting a qp to + * allow a destroy or reset to get rid + * of any tid space linkage and reference counts. + */ +static void _tid_rdma_flush_wait(struct rvt_qp *qp, struct tid_queue *queue) + __must_hold(&qp->s_lock) +{ + struct hfi1_qp_priv *priv; + + if (!qp) + return; + lockdep_assert_held(&qp->s_lock); + priv = qp->priv; + qp->s_flags &= ~HFI1_S_WAIT_TID_SPACE; + spin_lock(&priv->rcd->exp_lock); + if (!list_empty(&priv->tid_wait)) { + list_del_init(&priv->tid_wait); + qp->s_flags &= ~HFI1_S_WAIT_TID_SPACE; + queue->dequeue++; + rvt_put_qp(qp); + } + spin_unlock(&priv->rcd->exp_lock); +} + +void hfi1_tid_rdma_flush_wait(struct rvt_qp *qp) + __must_hold(&qp->s_lock) +{ + struct hfi1_qp_priv *priv = qp->priv; + + _tid_rdma_flush_wait(qp, &priv->rcd->flow_queue); + _tid_rdma_flush_wait(qp, &priv->rcd->rarr_queue); +} + +/* Flow functions */ +/** + * kern_reserve_flow - allocate a hardware flow + * @rcd - the context to use for allocation + * @last - the index of the preferred flow. Use RXE_NUM_TID_FLOWS to + * signify "don't care". + * + * Use a bit mask based allocation to reserve a hardware + * flow for use in receiving KDETH data packets. If a preferred flow is + * specified the function will attempt to reserve that flow again, if + * available. + * + * The exp_lock must be held. + * + * Return: + * On success: a value postive value between 0 and RXE_NUM_TID_FLOWS - 1 + * On failure: -EAGAIN + */ +static int kern_reserve_flow(struct hfi1_ctxtdata *rcd, int last) + __must_hold(&rcd->exp_lock) +{ + int nr; + + /* Attempt to reserve the preferred flow index */ + if (last >= 0 && last < RXE_NUM_TID_FLOWS && + !test_and_set_bit(last, &rcd->flow_mask)) + return last; + + nr = ffz(rcd->flow_mask); + BUILD_BUG_ON(RXE_NUM_TID_FLOWS >= + (sizeof(rcd->flow_mask) * BITS_PER_BYTE)); + if (nr > (RXE_NUM_TID_FLOWS - 1)) + return -EAGAIN; + set_bit(nr, &rcd->flow_mask); + return nr; +} + +static void kern_set_hw_flow(struct hfi1_ctxtdata *rcd, u32 generation, + u32 flow_idx) +{ + u64 reg; + + reg = ((u64)generation << HFI1_KDETH_BTH_SEQ_SHIFT) | + RCV_TID_FLOW_TABLE_CTRL_FLOW_VALID_SMASK | + RCV_TID_FLOW_TABLE_CTRL_KEEP_AFTER_SEQ_ERR_SMASK | + RCV_TID_FLOW_TABLE_CTRL_KEEP_ON_GEN_ERR_SMASK | + RCV_TID_FLOW_TABLE_STATUS_SEQ_MISMATCH_SMASK | + RCV_TID_FLOW_TABLE_STATUS_GEN_MISMATCH_SMASK; + + if (generation != KERN_GENERATION_RESERVED) + reg |= RCV_TID_FLOW_TABLE_CTRL_HDR_SUPP_EN_SMASK; + + write_uctxt_csr(rcd->dd, rcd->ctxt, + RCV_TID_FLOW_TABLE + 8 * flow_idx, reg); +} + +static u32 kern_setup_hw_flow(struct hfi1_ctxtdata *rcd, u32 flow_idx) + __must_hold(&rcd->exp_lock) +{ + u32 generation = rcd->flows[flow_idx].generation; + + kern_set_hw_flow(rcd, generation, flow_idx); + return generation; +} + +static u32 kern_flow_generation_next(u32 gen) +{ + u32 generation = mask_generation(gen + 1); + + if (generation == KERN_GENERATION_RESERVED) + generation = mask_generation(generation + 1); + return generation; +} + +static void kern_clear_hw_flow(struct hfi1_ctxtdata *rcd, u32 flow_idx) + __must_hold(&rcd->exp_lock) +{ + rcd->flows[flow_idx].generation = + kern_flow_generation_next(rcd->flows[flow_idx].generation); + kern_set_hw_flow(rcd, KERN_GENERATION_RESERVED, flow_idx); +} + +int hfi1_kern_setup_hw_flow(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp) +{ + struct hfi1_qp_priv *qpriv = (struct hfi1_qp_priv *)qp->priv; + struct tid_flow_state *fs = &qpriv->flow_state; + struct rvt_qp *fqp; + unsigned long flags; + int ret = 0; + + /* The QP already has an allocated flow */ + if (fs->index != RXE_NUM_TID_FLOWS) + return ret; + + spin_lock_irqsave(&rcd->exp_lock, flags); + if (kernel_tid_waiters(rcd, &rcd->flow_queue, qp)) + goto queue; + + ret = kern_reserve_flow(rcd, fs->last_index); + if (ret < 0) + goto queue; + fs->index = ret; + fs->last_index = fs->index; + + /* Generation received in a RESYNC overrides default flow generation */ + if (fs->generation != KERN_GENERATION_RESERVED) + rcd->flows[fs->index].generation = fs->generation; + fs->generation = kern_setup_hw_flow(rcd, fs->index); + fs->psn = 0; + fs->flags = 0; + dequeue_tid_waiter(rcd, &rcd->flow_queue, qp); + /* get head before dropping lock */ + fqp = first_qp(rcd, &rcd->flow_queue); + spin_unlock_irqrestore(&rcd->exp_lock, flags); + + tid_rdma_schedule_tid_wakeup(fqp); + return 0; +queue: + queue_qp_for_tid_wait(rcd, &rcd->flow_queue, qp); + spin_unlock_irqrestore(&rcd->exp_lock, flags); + return -EAGAIN; +} + +void hfi1_kern_clear_hw_flow(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp) +{ + struct hfi1_qp_priv *qpriv = (struct hfi1_qp_priv *)qp->priv; + struct tid_flow_state *fs = &qpriv->flow_state; + struct rvt_qp *fqp; + unsigned long flags; + + if (fs->index >= RXE_NUM_TID_FLOWS) + return; + spin_lock_irqsave(&rcd->exp_lock, flags); + kern_clear_hw_flow(rcd, fs->index); + clear_bit(fs->index, &rcd->flow_mask); + fs->index = RXE_NUM_TID_FLOWS; + fs->psn = 0; + fs->generation = KERN_GENERATION_RESERVED; + + /* get head before dropping lock */ + fqp = first_qp(rcd, &rcd->flow_queue); + spin_unlock_irqrestore(&rcd->exp_lock, flags); + + if (fqp == qp) { + __trigger_tid_waiter(fqp); + rvt_put_qp(fqp); + } else { + tid_rdma_schedule_tid_wakeup(fqp); + } +} + +void hfi1_kern_init_ctxt_generations(struct hfi1_ctxtdata *rcd) +{ + int i; + + for (i = 0; i < RXE_NUM_TID_FLOWS; i++) { + rcd->flows[i].generation = mask_generation(prandom_u32()); + kern_set_hw_flow(rcd, KERN_GENERATION_RESERVED, i); + } +} + +/* TID allocation functions */ +static u8 trdma_pset_order(struct tid_rdma_pageset *s) +{ + u8 count = s->count; + + return ilog2(count) + 1; +} + +/** + * tid_rdma_find_phys_blocks_4k - get groups base on mr info + * @npages - number of pages + * @pages - pointer to an array of page structs + * @list - page set array to return + * + * This routine returns the number of groups associated with + * the current sge information. This implementation is based + * on the expected receive find_phys_blocks() adjusted to + * use the MR information vs. the pfn. + * + * Return: + * the number of RcvArray entries + */ +static u32 tid_rdma_find_phys_blocks_4k(struct tid_rdma_flow *flow, + struct page **pages, + u32 npages, + struct tid_rdma_pageset *list) +{ + u32 pagecount, pageidx, setcount = 0, i; + void *vaddr, *this_vaddr; + + if (!npages) + return 0; + + /* + * Look for sets of physically contiguous pages in the user buffer. + * This will allow us to optimize Expected RcvArray entry usage by + * using the bigger supported sizes. + */ + vaddr = page_address(pages[0]); + trace_hfi1_tid_flow_page(flow->req->qp, flow, 0, 0, 0, vaddr); + for (pageidx = 0, pagecount = 1, i = 1; i <= npages; i++) { + this_vaddr = i < npages ? page_address(pages[i]) : NULL; + trace_hfi1_tid_flow_page(flow->req->qp, flow, i, 0, 0, + this_vaddr); + /* + * If the vaddr's are not sequential, pages are not physically + * contiguous. + */ + if (this_vaddr != (vaddr + PAGE_SIZE)) { + /* + * At this point we have to loop over the set of + * physically contiguous pages and break them down it + * sizes supported by the HW. + * There are two main constraints: + * 1. The max buffer size is MAX_EXPECTED_BUFFER. + * If the total set size is bigger than that + * program only a MAX_EXPECTED_BUFFER chunk. + * 2. The buffer size has to be a power of two. If + * it is not, round down to the closes power of + * 2 and program that size. + */ + while (pagecount) { + int maxpages = pagecount; + u32 bufsize = pagecount * PAGE_SIZE; + + if (bufsize > MAX_EXPECTED_BUFFER) + maxpages = + MAX_EXPECTED_BUFFER >> + PAGE_SHIFT; + else if (!is_power_of_2(bufsize)) + maxpages = + rounddown_pow_of_two(bufsize) >> + PAGE_SHIFT; + + list[setcount].idx = pageidx; + list[setcount].count = maxpages; + trace_hfi1_tid_pageset(flow->req->qp, setcount, + list[setcount].idx, + list[setcount].count); + pagecount -= maxpages; + pageidx += maxpages; + setcount++; + } + pageidx = i; + pagecount = 1; + vaddr = this_vaddr; + } else { + vaddr += PAGE_SIZE; + pagecount++; + } + } + /* insure we always return an even number of sets */ + if (setcount & 1) + list[setcount++].count = 0; + return setcount; +} + +/** + * tid_flush_pages - dump out pages into pagesets + * @list - list of pagesets + * @idx - pointer to current page index + * @pages - number of pages to dump + * @sets - current number of pagesset + * + * This routine flushes out accumuated pages. + * + * To insure an even number of sets the + * code may add a filler. + * + * This can happen with when pages is not + * a power of 2 or pages is a power of 2 + * less than the maximum pages. + * + * Return: + * The new number of sets + */ + +static u32 tid_flush_pages(struct tid_rdma_pageset *list, + u32 *idx, u32 pages, u32 sets) +{ + while (pages) { + u32 maxpages = pages; + + if (maxpages > MAX_EXPECTED_PAGES) + maxpages = MAX_EXPECTED_PAGES; + else if (!is_power_of_2(maxpages)) + maxpages = rounddown_pow_of_two(maxpages); + list[sets].idx = *idx; + list[sets++].count = maxpages; + *idx += maxpages; + pages -= maxpages; + } + /* might need a filler */ + if (sets & 1) + list[sets++].count = 0; + return sets; +} + +/** + * tid_rdma_find_phys_blocks_8k - get groups base on mr info + * @pages - pointer to an array of page structs + * @npages - number of pages + * @list - page set array to return + * + * This routine parses an array of pages to compute pagesets + * in an 8k compatible way. + * + * pages are tested two at a time, i, i + 1 for contiguous + * pages and i - 1 and i contiguous pages. + * + * If any condition is false, any accumlated pages are flushed and + * v0,v1 are emitted as separate PAGE_SIZE pagesets + * + * Otherwise, the current 8k is totaled for a future flush. + * + * Return: + * The number of pagesets + * list set with the returned number of pagesets + * + */ +static u32 tid_rdma_find_phys_blocks_8k(struct tid_rdma_flow *flow, + struct page **pages, + u32 npages, + struct tid_rdma_pageset *list) +{ + u32 idx, sets = 0, i; + u32 pagecnt = 0; + void *v0, *v1, *vm1; + + if (!npages) + return 0; + for (idx = 0, i = 0, vm1 = NULL; i < npages; i += 2) { + /* get a new v0 */ + v0 = page_address(pages[i]); + trace_hfi1_tid_flow_page(flow->req->qp, flow, i, 1, 0, v0); + v1 = i + 1 < npages ? + page_address(pages[i + 1]) : NULL; + trace_hfi1_tid_flow_page(flow->req->qp, flow, i, 1, 1, v1); + /* compare i, i + 1 vaddr */ + if (v1 != (v0 + PAGE_SIZE)) { + /* flush out pages */ + sets = tid_flush_pages(list, &idx, pagecnt, sets); + /* output v0,v1 as two pagesets */ + list[sets].idx = idx++; + list[sets++].count = 1; + if (v1) { + list[sets].count = 1; + list[sets++].idx = idx++; + } else { + list[sets++].count = 0; + } + vm1 = NULL; + pagecnt = 0; + continue; + } + /* i,i+1 consecutive, look at i-1,i */ + if (vm1 && v0 != (vm1 + PAGE_SIZE)) { + /* flush out pages */ + sets = tid_flush_pages(list, &idx, pagecnt, sets); + pagecnt = 0; + } + /* pages will always be a multiple of 8k */ + pagecnt += 2; + /* save i-1 */ + vm1 = v1; + /* move to next pair */ + } + /* dump residual pages at end */ + sets = tid_flush_pages(list, &idx, npages - idx, sets); + /* by design cannot be odd sets */ + WARN_ON(sets & 1); + return sets; +} + +/** + * Find pages for one segment of a sge array represented by @ss. The function + * does not check the sge, the sge must have been checked for alignment with a + * prior call to hfi1_kern_trdma_ok. Other sge checking is done as part of + * rvt_lkey_ok and rvt_rkey_ok. Also, the function only modifies the local sge + * copy maintained in @ss->sge, the original sge is not modified. + * + * Unlike IB RDMA WRITE, we can't decrement ss->num_sge here because we are not + * releasing the MR reference count at the same time. Otherwise, we'll "leak" + * references to the MR. This difference requires that we keep track of progress + * into the sg_list. This is done by the cur_seg cursor in the tid_rdma_request + * structure. + */ +static u32 kern_find_pages(struct tid_rdma_flow *flow, + struct page **pages, + struct rvt_sge_state *ss, bool *last) +{ + struct tid_rdma_request *req = flow->req; + struct rvt_sge *sge = &ss->sge; + u32 length = flow->req->seg_len; + u32 len = PAGE_SIZE; + u32 i = 0; + + while (length && req->isge < ss->num_sge) { + pages[i++] = virt_to_page(sge->vaddr); + + sge->vaddr += len; + sge->length -= len; + sge->sge_length -= len; + if (!sge->sge_length) { + if (++req->isge < ss->num_sge) + *sge = ss->sg_list[req->isge - 1]; + } else if (sge->length == 0 && sge->mr->lkey) { + if (++sge->n >= RVT_SEGSZ) { + ++sge->m; + sge->n = 0; + } + sge->vaddr = sge->mr->map[sge->m]->segs[sge->n].vaddr; + sge->length = sge->mr->map[sge->m]->segs[sge->n].length; + } + length -= len; + } + + flow->length = flow->req->seg_len - length; + *last = req->isge == ss->num_sge ? false : true; + return i; +} + +static void dma_unmap_flow(struct tid_rdma_flow *flow) +{ + struct hfi1_devdata *dd; + int i; + struct tid_rdma_pageset *pset; + + dd = flow->req->rcd->dd; + for (i = 0, pset = &flow->pagesets[0]; i < flow->npagesets; + i++, pset++) { + if (pset->count && pset->addr) { + dma_unmap_page(&dd->pcidev->dev, + pset->addr, + PAGE_SIZE * pset->count, + DMA_FROM_DEVICE); + pset->mapped = 0; + } + } +} + +static int dma_map_flow(struct tid_rdma_flow *flow, struct page **pages) +{ + int i; + struct hfi1_devdata *dd = flow->req->rcd->dd; + struct tid_rdma_pageset *pset; + + for (i = 0, pset = &flow->pagesets[0]; i < flow->npagesets; + i++, pset++) { + if (pset->count) { + pset->addr = dma_map_page(&dd->pcidev->dev, + pages[pset->idx], + 0, + PAGE_SIZE * pset->count, + DMA_FROM_DEVICE); + + if (dma_mapping_error(&dd->pcidev->dev, pset->addr)) { + dma_unmap_flow(flow); + return -ENOMEM; + } + pset->mapped = 1; + } + } + return 0; +} + +static inline bool dma_mapped(struct tid_rdma_flow *flow) +{ + return !!flow->pagesets[0].mapped; +} + +/* + * Get pages pointers and identify contiguous physical memory chunks for a + * segment. All segments are of length flow->req->seg_len. + */ +static int kern_get_phys_blocks(struct tid_rdma_flow *flow, + struct page **pages, + struct rvt_sge_state *ss, bool *last) +{ + u8 npages; + + /* Reuse previously computed pagesets, if any */ + if (flow->npagesets) { + trace_hfi1_tid_flow_alloc(flow->req->qp, flow->req->setup_head, + flow); + if (!dma_mapped(flow)) + return dma_map_flow(flow, pages); + return 0; + } + + npages = kern_find_pages(flow, pages, ss, last); + + if (flow->req->qp->pmtu == enum_to_mtu(OPA_MTU_4096)) + flow->npagesets = + tid_rdma_find_phys_blocks_4k(flow, pages, npages, + flow->pagesets); + else + flow->npagesets = + tid_rdma_find_phys_blocks_8k(flow, pages, npages, + flow->pagesets); + + return dma_map_flow(flow, pages); +} + +static inline void kern_add_tid_node(struct tid_rdma_flow *flow, + struct hfi1_ctxtdata *rcd, char *s, + struct tid_group *grp, u8 cnt) +{ + struct kern_tid_node *node = &flow->tnode[flow->tnode_cnt++]; + + WARN_ON_ONCE(flow->tnode_cnt >= + (TID_RDMA_MAX_SEGMENT_SIZE >> PAGE_SHIFT)); + if (WARN_ON_ONCE(cnt & 1)) + dd_dev_err(rcd->dd, + "unexpected odd allocation cnt %u map 0x%x used %u", + cnt, grp->map, grp->used); + + node->grp = grp; + node->map = grp->map; + node->cnt = cnt; + trace_hfi1_tid_node_add(flow->req->qp, s, flow->tnode_cnt - 1, + grp->base, grp->map, grp->used, cnt); +} + +/* + * Try to allocate pageset_count TID's from TID groups for a context + * + * This function allocates TID's without moving groups between lists or + * modifying grp->map. This is done as follows, being cogizant of the lists + * between which the TID groups will move: + * 1. First allocate complete groups of 8 TID's since this is more efficient, + * these groups will move from group->full without affecting used + * 2. If more TID's are needed allocate from used (will move from used->full or + * stay in used) + * 3. If we still don't have the required number of TID's go back and look again + * at a complete group (will move from group->used) + */ +static int kern_alloc_tids(struct tid_rdma_flow *flow) +{ + struct hfi1_ctxtdata *rcd = flow->req->rcd; + struct hfi1_devdata *dd = rcd->dd; + u32 ngroups, pageidx = 0; + struct tid_group *group = NULL, *used; + u8 use; + + flow->tnode_cnt = 0; + ngroups = flow->npagesets / dd->rcv_entries.group_size; + if (!ngroups) + goto used_list; + + /* First look at complete groups */ + list_for_each_entry(group, &rcd->tid_group_list.list, list) { + kern_add_tid_node(flow, rcd, "complete groups", group, + group->size); + + pageidx += group->size; + if (!--ngroups) + break; + } + + if (pageidx >= flow->npagesets) + goto ok; + +used_list: + /* Now look at partially used groups */ + list_for_each_entry(used, &rcd->tid_used_list.list, list) { + use = min_t(u32, flow->npagesets - pageidx, + used->size - used->used); + kern_add_tid_node(flow, rcd, "used groups", used, use); + + pageidx += use; + if (pageidx >= flow->npagesets) + goto ok; + } + + /* + * Look again at a complete group, continuing from where we left. + * However, if we are at the head, we have reached the end of the + * complete groups list from the first loop above + */ + if (group && &group->list == &rcd->tid_group_list.list) + goto bail_eagain; + group = list_prepare_entry(group, &rcd->tid_group_list.list, + list); + if (list_is_last(&group->list, &rcd->tid_group_list.list)) + goto bail_eagain; + group = list_next_entry(group, list); + use = min_t(u32, flow->npagesets - pageidx, group->size); + kern_add_tid_node(flow, rcd, "complete continue", group, use); + pageidx += use; + if (pageidx >= flow->npagesets) + goto ok; +bail_eagain: + trace_hfi1_msg_alloc_tids(flow->req->qp, " insufficient tids: needed ", + (u64)flow->npagesets); + return -EAGAIN; +ok: + return 0; +} + +static void kern_program_rcv_group(struct tid_rdma_flow *flow, int grp_num, + u32 *pset_idx) +{ + struct hfi1_ctxtdata *rcd = flow->req->rcd; + struct hfi1_devdata *dd = rcd->dd; + struct kern_tid_node *node = &flow->tnode[grp_num]; + struct tid_group *grp = node->grp; + struct tid_rdma_pageset *pset; + u32 pmtu_pg = flow->req->qp->pmtu >> PAGE_SHIFT; + u32 rcventry, npages = 0, pair = 0, tidctrl; + u8 i, cnt = 0; + + for (i = 0; i < grp->size; i++) { + rcventry = grp->base + i; + + if (node->map & BIT(i) || cnt >= node->cnt) { + rcv_array_wc_fill(dd, rcventry); + continue; + } + pset = &flow->pagesets[(*pset_idx)++]; + if (pset->count) { + hfi1_put_tid(dd, rcventry, PT_EXPECTED, + pset->addr, trdma_pset_order(pset)); + } else { + hfi1_put_tid(dd, rcventry, PT_INVALID, 0, 0); + } + npages += pset->count; + + rcventry -= rcd->expected_base; + tidctrl = pair ? 0x3 : rcventry & 0x1 ? 0x2 : 0x1; + /* + * A single TID entry will be used to use a rcvarr pair (with + * tidctrl 0x3), if ALL these are true (a) the bit pos is even + * (b) the group map shows current and the next bits as free + * indicating two consecutive rcvarry entries are available (c) + * we actually need 2 more entries + */ + pair = !(i & 0x1) && !((node->map >> i) & 0x3) && + node->cnt >= cnt + 2; + if (!pair) { + if (!pset->count) + tidctrl = 0x1; + flow->tid_entry[flow->tidcnt++] = + EXP_TID_SET(IDX, rcventry >> 1) | + EXP_TID_SET(CTRL, tidctrl) | + EXP_TID_SET(LEN, npages); + trace_hfi1_tid_entry_alloc(/* entry */ + flow->req->qp, flow->tidcnt - 1, + flow->tid_entry[flow->tidcnt - 1]); + + /* Efficient DIV_ROUND_UP(npages, pmtu_pg) */ + flow->npkts += (npages + pmtu_pg - 1) >> ilog2(pmtu_pg); + npages = 0; + } + + if (grp->used == grp->size - 1) + tid_group_move(grp, &rcd->tid_used_list, + &rcd->tid_full_list); + else if (!grp->used) + tid_group_move(grp, &rcd->tid_group_list, + &rcd->tid_used_list); + + grp->used++; + grp->map |= BIT(i); + cnt++; + } +} + +static void kern_unprogram_rcv_group(struct tid_rdma_flow *flow, int grp_num) +{ + struct hfi1_ctxtdata *rcd = flow->req->rcd; + struct hfi1_devdata *dd = rcd->dd; + struct kern_tid_node *node = &flow->tnode[grp_num]; + struct tid_group *grp = node->grp; + u32 rcventry; + u8 i, cnt = 0; + + for (i = 0; i < grp->size; i++) { + rcventry = grp->base + i; + + if (node->map & BIT(i) || cnt >= node->cnt) { + rcv_array_wc_fill(dd, rcventry); + continue; + } + + hfi1_put_tid(dd, rcventry, PT_INVALID, 0, 0); + + grp->used--; + grp->map &= ~BIT(i); + cnt++; + + if (grp->used == grp->size - 1) + tid_group_move(grp, &rcd->tid_full_list, + &rcd->tid_used_list); + else if (!grp->used) + tid_group_move(grp, &rcd->tid_used_list, + &rcd->tid_group_list); + } + if (WARN_ON_ONCE(cnt & 1)) { + struct hfi1_ctxtdata *rcd = flow->req->rcd; + struct hfi1_devdata *dd = rcd->dd; + + dd_dev_err(dd, "unexpected odd free cnt %u map 0x%x used %u", + cnt, grp->map, grp->used); + } +} + +static void kern_program_rcvarray(struct tid_rdma_flow *flow) +{ + u32 pset_idx = 0; + int i; + + flow->npkts = 0; + flow->tidcnt = 0; + for (i = 0; i < flow->tnode_cnt; i++) + kern_program_rcv_group(flow, i, &pset_idx); + trace_hfi1_tid_flow_alloc(flow->req->qp, flow->req->setup_head, flow); +} + +/** + * hfi1_kern_exp_rcv_setup() - setup TID's and flow for one segment of a + * TID RDMA request + * + * @req: TID RDMA request for which the segment/flow is being set up + * @ss: sge state, maintains state across successive segments of a sge + * @last: set to true after the last sge segment has been processed + * + * This function + * (1) finds a free flow entry in the flow circular buffer + * (2) finds pages and continuous physical chunks constituing one segment + * of an sge + * (3) allocates TID group entries for those chunks + * (4) programs rcvarray entries in the hardware corresponding to those + * TID's + * (5) computes a tidarray with formatted TID entries which can be sent + * to the sender + * (6) Reserves and programs HW flows. + * (7) It also manages queing the QP when TID/flow resources are not + * available. + * + * @req points to struct tid_rdma_request of which the segments are a part. The + * function uses qp, rcd and seg_len members of @req. In the absence of errors, + * req->flow_idx is the index of the flow which has been prepared in this + * invocation of function call. With flow = &req->flows[req->flow_idx], + * flow->tid_entry contains the TID array which the sender can use for TID RDMA + * sends and flow->npkts contains number of packets required to send the + * segment. + * + * hfi1_check_sge_align should be called prior to calling this function and if + * it signals error TID RDMA cannot be used for this sge and this function + * should not be called. + * + * For the queuing, caller must hold the flow->req->qp s_lock from the send + * engine and the function will procure the exp_lock. + * + * Return: + * The function returns -EAGAIN if sufficient number of TID/flow resources to + * map the segment could not be allocated. In this case the function should be + * called again with previous arguments to retry the TID allocation. There are + * no other error returns. The function returns 0 on success. + */ +int hfi1_kern_exp_rcv_setup(struct tid_rdma_request *req, + struct rvt_sge_state *ss, bool *last) + __must_hold(&req->qp->s_lock) +{ + struct tid_rdma_flow *flow = &req->flows[req->setup_head]; + struct hfi1_ctxtdata *rcd = req->rcd; + struct hfi1_qp_priv *qpriv = req->qp->priv; + unsigned long flags; + struct rvt_qp *fqp; + u16 clear_tail = req->clear_tail; + + lockdep_assert_held(&req->qp->s_lock); + /* + * We return error if either (a) we don't have space in the flow + * circular buffer, or (b) we already have max entries in the buffer. + * Max entries depend on the type of request we are processing and the + * negotiated TID RDMA parameters. + */ + if (!CIRC_SPACE(req->setup_head, clear_tail, MAX_FLOWS) || + CIRC_CNT(req->setup_head, clear_tail, MAX_FLOWS) >= + req->n_flows) + return -EINVAL; + + /* + * Get pages, identify contiguous physical memory chunks for the segment + * If we can not determine a DMA address mapping we will treat it just + * like if we ran out of space above. + */ + if (kern_get_phys_blocks(flow, qpriv->pages, ss, last)) { + hfi1_wait_kmem(flow->req->qp); + return -ENOMEM; + } + + spin_lock_irqsave(&rcd->exp_lock, flags); + if (kernel_tid_waiters(rcd, &rcd->rarr_queue, flow->req->qp)) + goto queue; + + /* + * At this point we know the number of pagesets and hence the number of + * TID's to map the segment. Allocate the TID's from the TID groups. If + * we cannot allocate the required number we exit and try again later + */ + if (kern_alloc_tids(flow)) + goto queue; + /* + * Finally program the TID entries with the pagesets, compute the + * tidarray and enable the HW flow + */ + kern_program_rcvarray(flow); + + /* + * Setup the flow state with relevant information. + * This information is used for tracking the sequence of data packets + * for the segment. + * The flow is setup here as this is the most accurate time and place + * to do so. Doing at a later time runs the risk of the flow data in + * qpriv getting out of sync. + */ + memset(&flow->flow_state, 0x0, sizeof(flow->flow_state)); + flow->idx = qpriv->flow_state.index; + flow->flow_state.generation = qpriv->flow_state.generation; + flow->flow_state.spsn = qpriv->flow_state.psn; + flow->flow_state.lpsn = flow->flow_state.spsn + flow->npkts - 1; + flow->flow_state.r_next_psn = + full_flow_psn(flow, flow->flow_state.spsn); + qpriv->flow_state.psn += flow->npkts; + + dequeue_tid_waiter(rcd, &rcd->rarr_queue, flow->req->qp); + /* get head before dropping lock */ + fqp = first_qp(rcd, &rcd->rarr_queue); + spin_unlock_irqrestore(&rcd->exp_lock, flags); + tid_rdma_schedule_tid_wakeup(fqp); + + req->setup_head = (req->setup_head + 1) & (MAX_FLOWS - 1); + return 0; +queue: + queue_qp_for_tid_wait(rcd, &rcd->rarr_queue, flow->req->qp); + spin_unlock_irqrestore(&rcd->exp_lock, flags); + return -EAGAIN; +} + +static void hfi1_tid_rdma_reset_flow(struct tid_rdma_flow *flow) +{ + flow->npagesets = 0; +} + +/* + * This function is called after one segment has been successfully sent to + * release the flow and TID HW/SW resources for that segment. The segments for a + * TID RDMA request are setup and cleared in FIFO order which is managed using a + * circular buffer. + */ +int hfi1_kern_exp_rcv_clear(struct tid_rdma_request *req) + __must_hold(&req->qp->s_lock) +{ + struct tid_rdma_flow *flow = &req->flows[req->clear_tail]; + struct hfi1_ctxtdata *rcd = req->rcd; + unsigned long flags; + int i; + struct rvt_qp *fqp; + + lockdep_assert_held(&req->qp->s_lock); + /* Exit if we have nothing in the flow circular buffer */ + if (!CIRC_CNT(req->setup_head, req->clear_tail, MAX_FLOWS)) + return -EINVAL; + + spin_lock_irqsave(&rcd->exp_lock, flags); + + for (i = 0; i < flow->tnode_cnt; i++) + kern_unprogram_rcv_group(flow, i); + /* To prevent double unprogramming */ + flow->tnode_cnt = 0; + /* get head before dropping lock */ + fqp = first_qp(rcd, &rcd->rarr_queue); + spin_unlock_irqrestore(&rcd->exp_lock, flags); + + dma_unmap_flow(flow); + + hfi1_tid_rdma_reset_flow(flow); + req->clear_tail = (req->clear_tail + 1) & (MAX_FLOWS - 1); + + if (fqp == req->qp) { + __trigger_tid_waiter(fqp); + rvt_put_qp(fqp); + } else { + tid_rdma_schedule_tid_wakeup(fqp); + } + + return 0; +} + +/* + * This function is called to release all the tid entries for + * a request. + */ +void hfi1_kern_exp_rcv_clear_all(struct tid_rdma_request *req) + __must_hold(&req->qp->s_lock) +{ + /* Use memory barrier for proper ordering */ + while (CIRC_CNT(req->setup_head, req->clear_tail, MAX_FLOWS)) { + if (hfi1_kern_exp_rcv_clear(req)) + break; + } +} + +/** + * hfi1_kern_exp_rcv_free_flows - free priviously allocated flow information + * @req - the tid rdma request to be cleaned + */ +static void hfi1_kern_exp_rcv_free_flows(struct tid_rdma_request *req) +{ + kfree(req->flows); + req->flows = NULL; +} + +/** + * __trdma_clean_swqe - clean up for large sized QPs + * @qp: the queue patch + * @wqe: the send wqe + */ +void __trdma_clean_swqe(struct rvt_qp *qp, struct rvt_swqe *wqe) +{ + struct hfi1_swqe_priv *p = wqe->priv; + + hfi1_kern_exp_rcv_free_flows(&p->tid_req); +} + +/* + * This can be called at QP create time or in the data path. + */ +static int hfi1_kern_exp_rcv_alloc_flows(struct tid_rdma_request *req, + gfp_t gfp) +{ + struct tid_rdma_flow *flows; + int i; + + if (likely(req->flows)) + return 0; + flows = kmalloc_node(MAX_FLOWS * sizeof(*flows), gfp, + req->rcd->numa_id); + if (!flows) + return -ENOMEM; + /* mini init */ + for (i = 0; i < MAX_FLOWS; i++) { + flows[i].req = req; + flows[i].npagesets = 0; + flows[i].pagesets[0].mapped = 0; + } + req->flows = flows; + return 0; +} + +static void hfi1_init_trdma_req(struct rvt_qp *qp, + struct tid_rdma_request *req) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + + /* + * Initialize various TID RDMA request variables. + * These variables are "static", which is why they + * can be pre-initialized here before the WRs has + * even been submitted. + * However, non-NULL values for these variables do not + * imply that this WQE has been enabled for TID RDMA. + * Drivers should check the WQE's opcode to determine + * if a request is a TID RDMA one or not. + */ + req->qp = qp; + req->rcd = qpriv->rcd; +} + +u64 hfi1_access_sw_tid_wait(const struct cntr_entry *entry, + void *context, int vl, int mode, u64 data) +{ + struct hfi1_devdata *dd = context; + + return dd->verbs_dev.n_tidwait; +} + +static struct tid_rdma_flow *find_flow_ib(struct tid_rdma_request *req, + u32 psn, u16 *fidx) +{ + u16 head, tail; + struct tid_rdma_flow *flow; + + head = req->setup_head; + tail = req->clear_tail; + for ( ; CIRC_CNT(head, tail, MAX_FLOWS); + tail = CIRC_NEXT(tail, MAX_FLOWS)) { + flow = &req->flows[tail]; + if (cmp_psn(psn, flow->flow_state.ib_spsn) >= 0 && + cmp_psn(psn, flow->flow_state.ib_lpsn) <= 0) { + if (fidx) + *fidx = tail; + return flow; + } + } + return NULL; +} + +static struct tid_rdma_flow * +__find_flow_ranged(struct tid_rdma_request *req, u16 head, u16 tail, + u32 psn, u16 *fidx) +{ + for ( ; CIRC_CNT(head, tail, MAX_FLOWS); + tail = CIRC_NEXT(tail, MAX_FLOWS)) { + struct tid_rdma_flow *flow = &req->flows[tail]; + u32 spsn, lpsn; + + spsn = full_flow_psn(flow, flow->flow_state.spsn); + lpsn = full_flow_psn(flow, flow->flow_state.lpsn); + + if (cmp_psn(psn, spsn) >= 0 && cmp_psn(psn, lpsn) <= 0) { + if (fidx) + *fidx = tail; + return flow; + } + } + return NULL; +} + +static struct tid_rdma_flow *find_flow(struct tid_rdma_request *req, + u32 psn, u16 *fidx) +{ + return __find_flow_ranged(req, req->setup_head, req->clear_tail, psn, + fidx); +} + +/* TID RDMA READ functions */ +u32 hfi1_build_tid_rdma_read_packet(struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, u32 *bth1, + u32 *bth2, u32 *len) +{ + struct tid_rdma_request *req = wqe_to_tid_req(wqe); + struct tid_rdma_flow *flow = &req->flows[req->flow_idx]; + struct rvt_qp *qp = req->qp; + struct hfi1_qp_priv *qpriv = qp->priv; + struct hfi1_swqe_priv *wpriv = wqe->priv; + struct tid_rdma_read_req *rreq = &ohdr->u.tid_rdma.r_req; + struct tid_rdma_params *remote; + u32 req_len = 0; + void *req_addr = NULL; + + /* This is the IB psn used to send the request */ + *bth2 = mask_psn(flow->flow_state.ib_spsn + flow->pkt); + trace_hfi1_tid_flow_build_read_pkt(qp, req->flow_idx, flow); + + /* TID Entries for TID RDMA READ payload */ + req_addr = &flow->tid_entry[flow->tid_idx]; + req_len = sizeof(*flow->tid_entry) * + (flow->tidcnt - flow->tid_idx); + + memset(&ohdr->u.tid_rdma.r_req, 0, sizeof(ohdr->u.tid_rdma.r_req)); + wpriv->ss.sge.vaddr = req_addr; + wpriv->ss.sge.sge_length = req_len; + wpriv->ss.sge.length = wpriv->ss.sge.sge_length; + /* + * We can safely zero these out. Since the first SGE covers the + * entire packet, nothing else should even look at the MR. + */ + wpriv->ss.sge.mr = NULL; + wpriv->ss.sge.m = 0; + wpriv->ss.sge.n = 0; + + wpriv->ss.sg_list = NULL; + wpriv->ss.total_len = wpriv->ss.sge.sge_length; + wpriv->ss.num_sge = 1; + + /* Construct the TID RDMA READ REQ packet header */ + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + + KDETH_RESET(rreq->kdeth0, KVER, 0x1); + KDETH_RESET(rreq->kdeth1, JKEY, remote->jkey); + rreq->reth.vaddr = cpu_to_be64(wqe->rdma_wr.remote_addr + + req->cur_seg * req->seg_len + flow->sent); + rreq->reth.rkey = cpu_to_be32(wqe->rdma_wr.rkey); + rreq->reth.length = cpu_to_be32(*len); + rreq->tid_flow_psn = + cpu_to_be32((flow->flow_state.generation << + HFI1_KDETH_BTH_SEQ_SHIFT) | + ((flow->flow_state.spsn + flow->pkt) & + HFI1_KDETH_BTH_SEQ_MASK)); + rreq->tid_flow_qp = + cpu_to_be32(qpriv->tid_rdma.local.qp | + ((flow->idx & TID_RDMA_DESTQP_FLOW_MASK) << + TID_RDMA_DESTQP_FLOW_SHIFT) | + qpriv->rcd->ctxt); + rreq->verbs_qp = cpu_to_be32(qp->remote_qpn); + *bth1 &= ~RVT_QPN_MASK; + *bth1 |= remote->qp; + *bth2 |= IB_BTH_REQ_ACK; + rcu_read_unlock(); + + /* We are done with this segment */ + flow->sent += *len; + req->cur_seg++; + qp->s_state = TID_OP(READ_REQ); + req->ack_pending++; + req->flow_idx = (req->flow_idx + 1) & (MAX_FLOWS - 1); + qpriv->pending_tid_r_segs++; + qp->s_num_rd_atomic++; + + /* Set the TID RDMA READ request payload size */ + *len = req_len; + + return sizeof(ohdr->u.tid_rdma.r_req) / sizeof(u32); +} + +/* + * @len: contains the data length to read upon entry and the read request + * payload length upon exit. + */ +u32 hfi1_build_tid_rdma_read_req(struct rvt_qp *qp, struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, u32 *bth1, + u32 *bth2, u32 *len) + __must_hold(&qp->s_lock) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_rdma_request *req = wqe_to_tid_req(wqe); + struct tid_rdma_flow *flow = NULL; + u32 hdwords = 0; + bool last; + bool retry = true; + u32 npkts = rvt_div_round_up_mtu(qp, *len); + + trace_hfi1_tid_req_build_read_req(qp, 0, wqe->wr.opcode, wqe->psn, + wqe->lpsn, req); + /* + * Check sync conditions. Make sure that there are no pending + * segments before freeing the flow. + */ +sync_check: + if (req->state == TID_REQUEST_SYNC) { + if (qpriv->pending_tid_r_segs) + goto done; + + hfi1_kern_clear_hw_flow(req->rcd, qp); + req->state = TID_REQUEST_ACTIVE; + } + + /* + * If the request for this segment is resent, the tid resources should + * have been allocated before. In this case, req->flow_idx should + * fall behind req->setup_head. + */ + if (req->flow_idx == req->setup_head) { + retry = false; + if (req->state == TID_REQUEST_RESEND) { + /* + * This is the first new segment for a request whose + * earlier segments have been re-sent. We need to + * set up the sge pointer correctly. + */ + restart_sge(&qp->s_sge, wqe, req->s_next_psn, + qp->pmtu); + req->isge = 0; + req->state = TID_REQUEST_ACTIVE; + } + + /* + * Check sync. The last PSN of each generation is reserved for + * RESYNC. + */ + if ((qpriv->flow_state.psn + npkts) > MAX_TID_FLOW_PSN - 1) { + req->state = TID_REQUEST_SYNC; + goto sync_check; + } + + /* Allocate the flow if not yet */ + if (hfi1_kern_setup_hw_flow(qpriv->rcd, qp)) + goto done; + + /* + * The following call will advance req->setup_head after + * allocating the tid entries. + */ + if (hfi1_kern_exp_rcv_setup(req, &qp->s_sge, &last)) { + req->state = TID_REQUEST_QUEUED; + + /* + * We don't have resources for this segment. The QP has + * already been queued. + */ + goto done; + } + } + + /* req->flow_idx should only be one slot behind req->setup_head */ + flow = &req->flows[req->flow_idx]; + flow->pkt = 0; + flow->tid_idx = 0; + flow->sent = 0; + if (!retry) { + /* Set the first and last IB PSN for the flow in use.*/ + flow->flow_state.ib_spsn = req->s_next_psn; + flow->flow_state.ib_lpsn = + flow->flow_state.ib_spsn + flow->npkts - 1; + } + + /* Calculate the next segment start psn.*/ + req->s_next_psn += flow->npkts; + + /* Build the packet header */ + hdwords = hfi1_build_tid_rdma_read_packet(wqe, ohdr, bth1, bth2, len); +done: + return hdwords; +} + +/* + * Validate and accept the TID RDMA READ request parameters. + * Return 0 if the request is accepted successfully; + * Return 1 otherwise. + */ +static int tid_rdma_rcv_read_request(struct rvt_qp *qp, + struct rvt_ack_entry *e, + struct hfi1_packet *packet, + struct ib_other_headers *ohdr, + u32 bth0, u32 psn, u64 vaddr, u32 len) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + u32 flow_psn, i, tidlen = 0, pktlen, tlen; + + req = ack_to_tid_req(e); + + /* Validate the payload first */ + flow = &req->flows[req->setup_head]; + + /* payload length = packet length - (header length + ICRC length) */ + pktlen = packet->tlen - (packet->hlen + 4); + if (pktlen > sizeof(flow->tid_entry)) + return 1; + memcpy(flow->tid_entry, packet->ebuf, pktlen); + flow->tidcnt = pktlen / sizeof(*flow->tid_entry); + + /* + * Walk the TID_ENTRY list to make sure we have enough space for a + * complete segment. Also calculate the number of required packets. + */ + flow->npkts = rvt_div_round_up_mtu(qp, len); + for (i = 0; i < flow->tidcnt; i++) { + trace_hfi1_tid_entry_rcv_read_req(qp, i, + flow->tid_entry[i]); + tlen = EXP_TID_GET(flow->tid_entry[i], LEN); + if (!tlen) + return 1; + + /* + * For tid pair (tidctr == 3), the buffer size of the pair + * should be the sum of the buffer size described by each + * tid entry. However, only the first entry needs to be + * specified in the request (see WFR HAS Section 8.5.7.1). + */ + tidlen += tlen; + } + if (tidlen * PAGE_SIZE < len) + return 1; + + /* Empty the flow array */ + req->clear_tail = req->setup_head; + flow->pkt = 0; + flow->tid_idx = 0; + flow->tid_offset = 0; + flow->sent = 0; + flow->tid_qpn = be32_to_cpu(ohdr->u.tid_rdma.r_req.tid_flow_qp); + flow->idx = (flow->tid_qpn >> TID_RDMA_DESTQP_FLOW_SHIFT) & + TID_RDMA_DESTQP_FLOW_MASK; + flow_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.r_req.tid_flow_psn)); + flow->flow_state.generation = flow_psn >> HFI1_KDETH_BTH_SEQ_SHIFT; + flow->flow_state.spsn = flow_psn & HFI1_KDETH_BTH_SEQ_MASK; + flow->length = len; + + flow->flow_state.lpsn = flow->flow_state.spsn + + flow->npkts - 1; + flow->flow_state.ib_spsn = psn; + flow->flow_state.ib_lpsn = flow->flow_state.ib_spsn + flow->npkts - 1; + + trace_hfi1_tid_flow_rcv_read_req(qp, req->setup_head, flow); + /* Set the initial flow index to the current flow. */ + req->flow_idx = req->setup_head; + + /* advance circular buffer head */ + req->setup_head = (req->setup_head + 1) & (MAX_FLOWS - 1); + + /* + * Compute last PSN for request. + */ + e->opcode = (bth0 >> 24) & 0xff; + e->psn = psn; + e->lpsn = psn + flow->npkts - 1; + e->sent = 0; + + req->n_flows = qpriv->tid_rdma.local.max_read; + req->state = TID_REQUEST_ACTIVE; + req->cur_seg = 0; + req->comp_seg = 0; + req->ack_seg = 0; + req->isge = 0; + req->seg_len = qpriv->tid_rdma.local.max_len; + req->total_len = len; + req->total_segs = 1; + req->r_flow_psn = e->psn; + + trace_hfi1_tid_req_rcv_read_req(qp, 0, e->opcode, e->psn, e->lpsn, + req); + return 0; +} + +static int tid_rdma_rcv_error(struct hfi1_packet *packet, + struct ib_other_headers *ohdr, + struct rvt_qp *qp, u32 psn, int diff) +{ + struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); + struct hfi1_ctxtdata *rcd = ((struct hfi1_qp_priv *)qp->priv)->rcd; + struct hfi1_ibdev *dev = to_idev(qp->ibqp.device); + struct hfi1_qp_priv *qpriv = qp->priv; + struct rvt_ack_entry *e; + struct tid_rdma_request *req; + unsigned long flags; + u8 prev; + bool old_req; + + trace_hfi1_rsp_tid_rcv_error(qp, psn); + trace_hfi1_tid_rdma_rcv_err(qp, 0, psn, diff); + if (diff > 0) { + /* sequence error */ + if (!qp->r_nak_state) { + ibp->rvp.n_rc_seqnak++; + qp->r_nak_state = IB_NAK_PSN_ERROR; + qp->r_ack_psn = qp->r_psn; + rc_defered_ack(rcd, qp); + } + goto done; + } + + ibp->rvp.n_rc_dupreq++; + + spin_lock_irqsave(&qp->s_lock, flags); + e = find_prev_entry(qp, psn, &prev, NULL, &old_req); + if (!e || (e->opcode != TID_OP(READ_REQ) && + e->opcode != TID_OP(WRITE_REQ))) + goto unlock; + + req = ack_to_tid_req(e); + req->r_flow_psn = psn; + trace_hfi1_tid_req_rcv_err(qp, 0, e->opcode, e->psn, e->lpsn, req); + if (e->opcode == TID_OP(READ_REQ)) { + struct ib_reth *reth; + u32 offset; + u32 len; + u32 rkey; + u64 vaddr; + int ok; + u32 bth0; + + reth = &ohdr->u.tid_rdma.r_req.reth; + /* + * The requester always restarts from the start of the original + * request. + */ + offset = delta_psn(psn, e->psn) * qp->pmtu; + len = be32_to_cpu(reth->length); + if (psn != e->psn || len != req->total_len) + goto unlock; + + if (e->rdma_sge.mr) { + rvt_put_mr(e->rdma_sge.mr); + e->rdma_sge.mr = NULL; + } + + rkey = be32_to_cpu(reth->rkey); + vaddr = get_ib_reth_vaddr(reth); + + qp->r_len = len; + ok = rvt_rkey_ok(qp, &e->rdma_sge, len, vaddr, rkey, + IB_ACCESS_REMOTE_READ); + if (unlikely(!ok)) + goto unlock; + + /* + * If all the response packets for the current request have + * been sent out and this request is complete (old_request + * == false) and the TID flow may be unusable (the + * req->clear_tail is advanced). However, when an earlier + * request is received, this request will not be complete any + * more (qp->s_tail_ack_queue is moved back, see below). + * Consequently, we need to update the TID flow info everytime + * a duplicate request is received. + */ + bth0 = be32_to_cpu(ohdr->bth[0]); + if (tid_rdma_rcv_read_request(qp, e, packet, ohdr, bth0, psn, + vaddr, len)) + goto unlock; + + /* + * True if the request is already scheduled (between + * qp->s_tail_ack_queue and qp->r_head_ack_queue); + */ + if (old_req) + goto unlock; + } else { + struct flow_state *fstate; + bool schedule = false; + u8 i; + + if (req->state == TID_REQUEST_RESEND) { + req->state = TID_REQUEST_RESEND_ACTIVE; + } else if (req->state == TID_REQUEST_INIT_RESEND) { + req->state = TID_REQUEST_INIT; + schedule = true; + } + + /* + * True if the request is already scheduled (between + * qp->s_tail_ack_queue and qp->r_head_ack_queue). + * Also, don't change requests, which are at the SYNC + * point and haven't generated any responses yet. + * There is nothing to retransmit for them yet. + */ + if (old_req || req->state == TID_REQUEST_INIT || + (req->state == TID_REQUEST_SYNC && !req->cur_seg)) { + for (i = prev + 1; ; i++) { + if (i > rvt_size_atomic(&dev->rdi)) + i = 0; + if (i == qp->r_head_ack_queue) + break; + e = &qp->s_ack_queue[i]; + req = ack_to_tid_req(e); + if (e->opcode == TID_OP(WRITE_REQ) && + req->state == TID_REQUEST_INIT) + req->state = TID_REQUEST_INIT_RESEND; + } + /* + * If the state of the request has been changed, + * the first leg needs to get scheduled in order to + * pick up the change. Otherwise, normal response + * processing should take care of it. + */ + if (!schedule) + goto unlock; + } + + /* + * If there is no more allocated segment, just schedule the qp + * without changing any state. + */ + if (req->clear_tail == req->setup_head) + goto schedule; + /* + * If this request has sent responses for segments, which have + * not received data yet (flow_idx != clear_tail), the flow_idx + * pointer needs to be adjusted so the same responses can be + * re-sent. + */ + if (CIRC_CNT(req->flow_idx, req->clear_tail, MAX_FLOWS)) { + fstate = &req->flows[req->clear_tail].flow_state; + qpriv->pending_tid_w_segs -= + CIRC_CNT(req->flow_idx, req->clear_tail, + MAX_FLOWS); + req->flow_idx = + CIRC_ADD(req->clear_tail, + delta_psn(psn, fstate->resp_ib_psn), + MAX_FLOWS); + qpriv->pending_tid_w_segs += + delta_psn(psn, fstate->resp_ib_psn); + /* + * When flow_idx == setup_head, we've gotten a duplicate + * request for a segment, which has not been allocated + * yet. In that case, don't adjust this request. + * However, we still want to go through the loop below + * to adjust all subsequent requests. + */ + if (CIRC_CNT(req->setup_head, req->flow_idx, + MAX_FLOWS)) { + req->cur_seg = delta_psn(psn, e->psn); + req->state = TID_REQUEST_RESEND_ACTIVE; + } + } + + for (i = prev + 1; ; i++) { + /* + * Look at everything up to and including + * s_tail_ack_queue + */ + if (i > rvt_size_atomic(&dev->rdi)) + i = 0; + if (i == qp->r_head_ack_queue) + break; + e = &qp->s_ack_queue[i]; + req = ack_to_tid_req(e); + trace_hfi1_tid_req_rcv_err(qp, 0, e->opcode, e->psn, + e->lpsn, req); + if (e->opcode != TID_OP(WRITE_REQ) || + req->cur_seg == req->comp_seg || + req->state == TID_REQUEST_INIT || + req->state == TID_REQUEST_INIT_RESEND) { + if (req->state == TID_REQUEST_INIT) + req->state = TID_REQUEST_INIT_RESEND; + continue; + } + qpriv->pending_tid_w_segs -= + CIRC_CNT(req->flow_idx, + req->clear_tail, + MAX_FLOWS); + req->flow_idx = req->clear_tail; + req->state = TID_REQUEST_RESEND; + req->cur_seg = req->comp_seg; + } + qpriv->s_flags &= ~HFI1_R_TID_WAIT_INTERLCK; + } + /* Re-process old requests.*/ + if (qp->s_acked_ack_queue == qp->s_tail_ack_queue) + qp->s_acked_ack_queue = prev; + qp->s_tail_ack_queue = prev; + /* + * Since the qp->s_tail_ack_queue is modified, the + * qp->s_ack_state must be changed to re-initialize + * qp->s_ack_rdma_sge; Otherwise, we will end up in + * wrong memory region. + */ + qp->s_ack_state = OP(ACKNOWLEDGE); +schedule: + /* + * It's possible to receive a retry psn that is earlier than an RNRNAK + * psn. In this case, the rnrnak state should be cleared. + */ + if (qpriv->rnr_nak_state) { + qp->s_nak_state = 0; + qpriv->rnr_nak_state = TID_RNR_NAK_INIT; + qp->r_psn = e->lpsn + 1; + hfi1_tid_write_alloc_resources(qp, true); + } + + qp->r_state = e->opcode; + qp->r_nak_state = 0; + qp->s_flags |= RVT_S_RESP_PENDING; + hfi1_schedule_send(qp); +unlock: + spin_unlock_irqrestore(&qp->s_lock, flags); +done: + return 1; +} + +void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet) +{ + /* HANDLER FOR TID RDMA READ REQUEST packet (Responder side)*/ + + /* + * 1. Verify TID RDMA READ REQ as per IB_OPCODE_RC_RDMA_READ + * (see hfi1_rc_rcv()) + * 2. Put TID RDMA READ REQ into the response queueu (s_ack_queue) + * - Setup struct tid_rdma_req with request info + * - Initialize struct tid_rdma_flow info; + * - Copy TID entries; + * 3. Set the qp->s_ack_state. + * 4. Set RVT_S_RESP_PENDING in s_flags. + * 5. Kick the send engine (hfi1_schedule_send()) + */ + struct hfi1_ctxtdata *rcd = packet->rcd; + struct rvt_qp *qp = packet->qp; + struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_ack_entry *e; + unsigned long flags; + struct ib_reth *reth; + struct hfi1_qp_priv *qpriv = qp->priv; + u32 bth0, psn, len, rkey; + bool is_fecn; + u8 next; + u64 vaddr; + int diff; + u8 nack_state = IB_NAK_INVALID_REQUEST; + + bth0 = be32_to_cpu(ohdr->bth[0]); + if (hfi1_ruc_check_hdr(ibp, packet)) + return; + + is_fecn = process_ecn(qp, packet); + psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + trace_hfi1_rsp_rcv_tid_read_req(qp, psn); + + if (qp->state == IB_QPS_RTR && !(qp->r_flags & RVT_R_COMM_EST)) + rvt_comm_est(qp); + + if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ))) + goto nack_inv; + + reth = &ohdr->u.tid_rdma.r_req.reth; + vaddr = be64_to_cpu(reth->vaddr); + len = be32_to_cpu(reth->length); + /* The length needs to be in multiples of PAGE_SIZE */ + if (!len || len & ~PAGE_MASK || len > qpriv->tid_rdma.local.max_len) + goto nack_inv; + + diff = delta_psn(psn, qp->r_psn); + if (unlikely(diff)) { + if (tid_rdma_rcv_error(packet, ohdr, qp, psn, diff)) + return; + goto send_ack; + } + + /* We've verified the request, insert it into the ack queue. */ + next = qp->r_head_ack_queue + 1; + if (next > rvt_size_atomic(ib_to_rvt(qp->ibqp.device))) + next = 0; + spin_lock_irqsave(&qp->s_lock, flags); + if (unlikely(next == qp->s_tail_ack_queue)) { + if (!qp->s_ack_queue[next].sent) { + nack_state = IB_NAK_REMOTE_OPERATIONAL_ERROR; + goto nack_inv_unlock; + } + update_ack_queue(qp, next); + } + e = &qp->s_ack_queue[qp->r_head_ack_queue]; + if (e->rdma_sge.mr) { + rvt_put_mr(e->rdma_sge.mr); + e->rdma_sge.mr = NULL; + } + + rkey = be32_to_cpu(reth->rkey); + qp->r_len = len; + + if (unlikely(!rvt_rkey_ok(qp, &e->rdma_sge, qp->r_len, vaddr, + rkey, IB_ACCESS_REMOTE_READ))) + goto nack_acc; + + /* Accept the request parameters */ + if (tid_rdma_rcv_read_request(qp, e, packet, ohdr, bth0, psn, vaddr, + len)) + goto nack_inv_unlock; + + qp->r_state = e->opcode; + qp->r_nak_state = 0; + /* + * We need to increment the MSN here instead of when we + * finish sending the result since a duplicate request would + * increment it more than once. + */ + qp->r_msn++; + qp->r_psn += e->lpsn - e->psn + 1; + + qp->r_head_ack_queue = next; + + /* + * For all requests other than TID WRITE which are added to the ack + * queue, qpriv->r_tid_alloc follows qp->r_head_ack_queue. It is ok to + * do this because of interlocks between these and TID WRITE + * requests. The same change has also been made in hfi1_rc_rcv(). + */ + qpriv->r_tid_alloc = qp->r_head_ack_queue; + + /* Schedule the send tasklet. */ + qp->s_flags |= RVT_S_RESP_PENDING; + hfi1_schedule_send(qp); + + spin_unlock_irqrestore(&qp->s_lock, flags); + if (is_fecn) + goto send_ack; + return; + +nack_inv_unlock: + spin_unlock_irqrestore(&qp->s_lock, flags); +nack_inv: + rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR); + qp->r_nak_state = nack_state; + qp->r_ack_psn = qp->r_psn; + /* Queue NAK for later */ + rc_defered_ack(rcd, qp); + return; +nack_acc: + spin_unlock_irqrestore(&qp->s_lock, flags); + rvt_rc_error(qp, IB_WC_LOC_PROT_ERR); + qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR; + qp->r_ack_psn = qp->r_psn; +send_ack: + hfi1_send_rc_ack(packet, is_fecn); +} + +u32 hfi1_build_tid_rdma_read_resp(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_other_headers *ohdr, u32 *bth0, + u32 *bth1, u32 *bth2, u32 *len, bool *last) +{ + struct hfi1_ack_priv *epriv = e->priv; + struct tid_rdma_request *req = &epriv->tid_req; + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_rdma_flow *flow = &req->flows[req->clear_tail]; + u32 tidentry = flow->tid_entry[flow->tid_idx]; + u32 tidlen = EXP_TID_GET(tidentry, LEN) << PAGE_SHIFT; + struct tid_rdma_read_resp *resp = &ohdr->u.tid_rdma.r_rsp; + u32 next_offset, om = KDETH_OM_LARGE; + bool last_pkt; + u32 hdwords = 0; + struct tid_rdma_params *remote; + + *len = min_t(u32, qp->pmtu, tidlen - flow->tid_offset); + flow->sent += *len; + next_offset = flow->tid_offset + *len; + last_pkt = (flow->sent >= flow->length); + + trace_hfi1_tid_entry_build_read_resp(qp, flow->tid_idx, tidentry); + trace_hfi1_tid_flow_build_read_resp(qp, req->clear_tail, flow); + + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + if (!remote) { + rcu_read_unlock(); + goto done; + } + KDETH_RESET(resp->kdeth0, KVER, 0x1); + KDETH_SET(resp->kdeth0, SH, !last_pkt); + KDETH_SET(resp->kdeth0, INTR, !!(!last_pkt && remote->urg)); + KDETH_SET(resp->kdeth0, TIDCTRL, EXP_TID_GET(tidentry, CTRL)); + KDETH_SET(resp->kdeth0, TID, EXP_TID_GET(tidentry, IDX)); + KDETH_SET(resp->kdeth0, OM, om == KDETH_OM_LARGE); + KDETH_SET(resp->kdeth0, OFFSET, flow->tid_offset / om); + KDETH_RESET(resp->kdeth1, JKEY, remote->jkey); + resp->verbs_qp = cpu_to_be32(qp->remote_qpn); + rcu_read_unlock(); + + resp->aeth = rvt_compute_aeth(qp); + resp->verbs_psn = cpu_to_be32(mask_psn(flow->flow_state.ib_spsn + + flow->pkt)); + + *bth0 = TID_OP(READ_RESP) << 24; + *bth1 = flow->tid_qpn; + *bth2 = mask_psn(((flow->flow_state.spsn + flow->pkt++) & + HFI1_KDETH_BTH_SEQ_MASK) | + (flow->flow_state.generation << + HFI1_KDETH_BTH_SEQ_SHIFT)); + *last = last_pkt; + if (last_pkt) + /* Advance to next flow */ + req->clear_tail = (req->clear_tail + 1) & + (MAX_FLOWS - 1); + + if (next_offset >= tidlen) { + flow->tid_offset = 0; + flow->tid_idx++; + } else { + flow->tid_offset = next_offset; + } + + hdwords = sizeof(ohdr->u.tid_rdma.r_rsp) / sizeof(u32); + +done: + return hdwords; +} + +static inline struct tid_rdma_request * +find_tid_request(struct rvt_qp *qp, u32 psn, enum ib_wr_opcode opcode) + __must_hold(&qp->s_lock) +{ + struct rvt_swqe *wqe; + struct tid_rdma_request *req = NULL; + u32 i, end; + + end = qp->s_cur + 1; + if (end == qp->s_size) + end = 0; + for (i = qp->s_acked; i != end;) { + wqe = rvt_get_swqe_ptr(qp, i); + if (cmp_psn(psn, wqe->psn) >= 0 && + cmp_psn(psn, wqe->lpsn) <= 0) { + if (wqe->wr.opcode == opcode) + req = wqe_to_tid_req(wqe); + break; + } + if (++i == qp->s_size) + i = 0; + } + + return req; +} + +void hfi1_rc_rcv_tid_rdma_read_resp(struct hfi1_packet *packet) +{ + /* HANDLER FOR TID RDMA READ RESPONSE packet (Requestor side */ + + /* + * 1. Find matching SWQE + * 2. Check that the entire segment has been read. + * 3. Remove HFI1_S_WAIT_TID_RESP from s_flags. + * 4. Free the TID flow resources. + * 5. Kick the send engine (hfi1_schedule_send()) + */ + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *priv = qp->priv; + struct hfi1_ctxtdata *rcd = packet->rcd; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + u32 opcode, aeth; + bool is_fecn; + unsigned long flags; + u32 kpsn, ipsn; + + trace_hfi1_sender_rcv_tid_read_resp(qp); + is_fecn = process_ecn(qp, packet); + kpsn = mask_psn(be32_to_cpu(ohdr->bth[2])); + aeth = be32_to_cpu(ohdr->u.tid_rdma.r_rsp.aeth); + opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff; + + spin_lock_irqsave(&qp->s_lock, flags); + ipsn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_psn)); + req = find_tid_request(qp, ipsn, IB_WR_TID_RDMA_READ); + if (unlikely(!req)) + goto ack_op_err; + + flow = &req->flows[req->clear_tail]; + /* When header suppression is disabled */ + if (cmp_psn(ipsn, flow->flow_state.ib_lpsn)) + goto ack_done; + req->ack_pending--; + priv->pending_tid_r_segs--; + qp->s_num_rd_atomic--; + if ((qp->s_flags & RVT_S_WAIT_FENCE) && + !qp->s_num_rd_atomic) { + qp->s_flags &= ~(RVT_S_WAIT_FENCE | + RVT_S_WAIT_ACK); + hfi1_schedule_send(qp); + } + if (qp->s_flags & RVT_S_WAIT_RDMAR) { + qp->s_flags &= ~(RVT_S_WAIT_RDMAR | RVT_S_WAIT_ACK); + hfi1_schedule_send(qp); + } + + trace_hfi1_ack(qp, ipsn); + trace_hfi1_tid_req_rcv_read_resp(qp, 0, req->e.swqe->wr.opcode, + req->e.swqe->psn, req->e.swqe->lpsn, + req); + trace_hfi1_tid_flow_rcv_read_resp(qp, req->clear_tail, flow); + + /* Release the tid resources */ + hfi1_kern_exp_rcv_clear(req); + + if (!do_rc_ack(qp, aeth, ipsn, opcode, 0, rcd)) + goto ack_done; + + /* If not done yet, build next read request */ + if (++req->comp_seg >= req->total_segs) { + priv->tid_r_comp++; + req->state = TID_REQUEST_COMPLETE; + } + + /* + * Clear the hw flow under two conditions: + * 1. This request is a sync point and it is complete; + * 2. Current request is completed and there are no more requests. + */ + if ((req->state == TID_REQUEST_SYNC && + req->comp_seg == req->cur_seg) || + priv->tid_r_comp == priv->tid_r_reqs) { + hfi1_kern_clear_hw_flow(priv->rcd, qp); + if (req->state == TID_REQUEST_SYNC) + req->state = TID_REQUEST_ACTIVE; + } + + hfi1_schedule_send(qp); + goto ack_done; + +ack_op_err: + /* + * The test indicates that the send engine has finished its cleanup + * after sending the request and it's now safe to put the QP into error + * state. However, if the wqe queue is empty (qp->s_acked == qp->s_tail + * == qp->s_head), it would be unsafe to complete the wqe pointed by + * qp->s_acked here. Putting the qp into error state will safely flush + * all remaining requests. + */ + if (qp->s_last == qp->s_acked) + rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR); + +ack_done: + spin_unlock_irqrestore(&qp->s_lock, flags); + if (is_fecn) + hfi1_send_rc_ack(packet, is_fecn); +} + +void hfi1_kern_read_tid_flow_free(struct rvt_qp *qp) + __must_hold(&qp->s_lock) +{ + u32 n = qp->s_acked; + struct rvt_swqe *wqe; + struct tid_rdma_request *req; + struct hfi1_qp_priv *priv = qp->priv; + + lockdep_assert_held(&qp->s_lock); + /* Free any TID entries */ + while (n != qp->s_tail) { + wqe = rvt_get_swqe_ptr(qp, n); + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) { + req = wqe_to_tid_req(wqe); + hfi1_kern_exp_rcv_clear_all(req); + } + + if (++n == qp->s_size) + n = 0; + } + /* Free flow */ + hfi1_kern_clear_hw_flow(priv->rcd, qp); +} + +static bool tid_rdma_tid_err(struct hfi1_ctxtdata *rcd, + struct hfi1_packet *packet, u8 rcv_type, + u8 opcode) +{ + struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *qpriv = qp->priv; + u32 ipsn; + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_ack_entry *e; + struct tid_rdma_request *req; + struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device); + u32 i; + + if (rcv_type >= RHF_RCV_TYPE_IB) + goto done; + + spin_lock(&qp->s_lock); + + /* + * We've ran out of space in the eager buffer. + * Eagerly received KDETH packets which require space in the + * Eager buffer (packet that have payload) are TID RDMA WRITE + * response packets. In this case, we have to re-transmit the + * TID RDMA WRITE request. + */ + if (rcv_type == RHF_RCV_TYPE_EAGER) { + hfi1_restart_rc(qp, qp->s_last_psn + 1, 1); + hfi1_schedule_send(qp); + goto done_unlock; + } + + /* + * For TID READ response, error out QP after freeing the tid + * resources. + */ + if (opcode == TID_OP(READ_RESP)) { + ipsn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_psn)); + if (cmp_psn(ipsn, qp->s_last_psn) > 0 && + cmp_psn(ipsn, qp->s_psn) < 0) { + hfi1_kern_read_tid_flow_free(qp); + spin_unlock(&qp->s_lock); + rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR); + goto done; + } + goto done_unlock; + } + + /* + * Error out the qp for TID RDMA WRITE + */ + hfi1_kern_clear_hw_flow(qpriv->rcd, qp); + for (i = 0; i < rvt_max_atomic(rdi); i++) { + e = &qp->s_ack_queue[i]; + if (e->opcode == TID_OP(WRITE_REQ)) { + req = ack_to_tid_req(e); + hfi1_kern_exp_rcv_clear_all(req); + } + } + spin_unlock(&qp->s_lock); + rvt_rc_error(qp, IB_WC_LOC_LEN_ERR); + goto done; + +done_unlock: + spin_unlock(&qp->s_lock); +done: + return true; +} + +static void restart_tid_rdma_read_req(struct hfi1_ctxtdata *rcd, + struct rvt_qp *qp, struct rvt_swqe *wqe) +{ + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + + /* Start from the right segment */ + qp->r_flags |= RVT_R_RDMAR_SEQ; + req = wqe_to_tid_req(wqe); + flow = &req->flows[req->clear_tail]; + hfi1_restart_rc(qp, flow->flow_state.ib_spsn, 0); + if (list_empty(&qp->rspwait)) { + qp->r_flags |= RVT_R_RSP_SEND; + rvt_get_qp(qp); + list_add_tail(&qp->rspwait, &rcd->qp_wait_list); + } +} + +/* + * Handle the KDETH eflags for TID RDMA READ response. + * + * Return true if the last packet for a segment has been received and it is + * time to process the response normally; otherwise, return true. + * + * The caller must hold the packet->qp->r_lock and the rcu_read_lock. + */ +static bool handle_read_kdeth_eflags(struct hfi1_ctxtdata *rcd, + struct hfi1_packet *packet, u8 rcv_type, + u8 rte, u32 psn, u32 ibpsn) + __must_hold(&packet->qp->r_lock) __must_hold(RCU) +{ + struct hfi1_pportdata *ppd = rcd->ppd; + struct hfi1_devdata *dd = ppd->dd; + struct hfi1_ibport *ibp; + struct rvt_swqe *wqe; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + u32 ack_psn; + struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *priv = qp->priv; + bool ret = true; + int diff = 0; + u32 fpsn; + + lockdep_assert_held(&qp->r_lock); + /* If the psn is out of valid range, drop the packet */ + if (cmp_psn(ibpsn, qp->s_last_psn) < 0 || + cmp_psn(ibpsn, qp->s_psn) > 0) + return ret; + + spin_lock(&qp->s_lock); + /* + * Note that NAKs implicitly ACK outstanding SEND and RDMA write + * requests and implicitly NAK RDMA read and atomic requests issued + * before the NAK'ed request. + */ + ack_psn = ibpsn - 1; + wqe = rvt_get_swqe_ptr(qp, qp->s_acked); + ibp = to_iport(qp->ibqp.device, qp->port_num); + + /* Complete WQEs that the PSN finishes. */ + while ((int)delta_psn(ack_psn, wqe->lpsn) >= 0) { + /* + * If this request is a RDMA read or atomic, and the NACK is + * for a later operation, this NACK NAKs the RDMA read or + * atomic. + */ + if (wqe->wr.opcode == IB_WR_RDMA_READ || + wqe->wr.opcode == IB_WR_TID_RDMA_READ || + wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP || + wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) { + /* Retry this request. */ + if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) { + qp->r_flags |= RVT_R_RDMAR_SEQ; + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) { + restart_tid_rdma_read_req(rcd, qp, + wqe); + } else { + hfi1_restart_rc(qp, qp->s_last_psn + 1, + 0); + if (list_empty(&qp->rspwait)) { + qp->r_flags |= RVT_R_RSP_SEND; + rvt_get_qp(qp); + list_add_tail(/* wait */ + &qp->rspwait, + &rcd->qp_wait_list); + } + } + } + /* + * No need to process the NAK since we are + * restarting an earlier request. + */ + break; + } + + wqe = do_rc_completion(qp, wqe, ibp); + if (qp->s_acked == qp->s_tail) + break; + } + + /* Handle the eflags for the request */ + if (wqe->wr.opcode != IB_WR_TID_RDMA_READ) + goto s_unlock; + + req = wqe_to_tid_req(wqe); + switch (rcv_type) { + case RHF_RCV_TYPE_EXPECTED: + switch (rte) { + case RHF_RTE_EXPECTED_FLOW_SEQ_ERR: + /* + * On the first occurrence of a Flow Sequence error, + * the flag TID_FLOW_SW_PSN is set. + * + * After that, the flow is *not* reprogrammed and the + * protocol falls back to SW PSN checking. This is done + * to prevent continuous Flow Sequence errors for any + * packets that could be still in the fabric. + */ + flow = find_flow(req, psn, NULL); + if (!flow) { + /* + * We can't find the IB PSN matching the + * received KDETH PSN. The only thing we can + * do at this point is report the error to + * the QP. + */ + hfi1_kern_read_tid_flow_free(qp); + spin_unlock(&qp->s_lock); + rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR); + return ret; + } + if (priv->flow_state.flags & TID_FLOW_SW_PSN) { + diff = cmp_psn(psn, + priv->flow_state.r_next_psn); + if (diff > 0) { + if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) + restart_tid_rdma_read_req(rcd, + qp, + wqe); + + /* Drop the packet.*/ + goto s_unlock; + } else if (diff < 0) { + /* + * If a response packet for a restarted + * request has come back, reset the + * restart flag. + */ + if (qp->r_flags & RVT_R_RDMAR_SEQ) + qp->r_flags &= + ~RVT_R_RDMAR_SEQ; + + /* Drop the packet.*/ + goto s_unlock; + } + + /* + * If SW PSN verification is successful and + * this is the last packet in the segment, tell + * the caller to process it as a normal packet. + */ + fpsn = full_flow_psn(flow, + flow->flow_state.lpsn); + if (cmp_psn(fpsn, psn) == 0) { + ret = false; + if (qp->r_flags & RVT_R_RDMAR_SEQ) + qp->r_flags &= + ~RVT_R_RDMAR_SEQ; + } + priv->flow_state.r_next_psn++; + } else { + u64 reg; + u32 last_psn; + + /* + * The only sane way to get the amount of + * progress is to read the HW flow state. + */ + reg = read_uctxt_csr(dd, rcd->ctxt, + RCV_TID_FLOW_TABLE + + (8 * flow->idx)); + last_psn = mask_psn(reg); + + priv->flow_state.r_next_psn = last_psn; + priv->flow_state.flags |= TID_FLOW_SW_PSN; + /* + * If no request has been restarted yet, + * restart the current one. + */ + if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) + restart_tid_rdma_read_req(rcd, qp, + wqe); + } + + break; + + case RHF_RTE_EXPECTED_FLOW_GEN_ERR: + /* + * Since the TID flow is able to ride through + * generation mismatch, drop this stale packet. + */ + break; + + default: + break; + } + break; + + case RHF_RCV_TYPE_ERROR: + switch (rte) { + case RHF_RTE_ERROR_OP_CODE_ERR: + case RHF_RTE_ERROR_KHDR_MIN_LEN_ERR: + case RHF_RTE_ERROR_KHDR_HCRC_ERR: + case RHF_RTE_ERROR_KHDR_KVER_ERR: + case RHF_RTE_ERROR_CONTEXT_ERR: + case RHF_RTE_ERROR_KHDR_TID_ERR: + default: + break; + } + default: + break; + } +s_unlock: + spin_unlock(&qp->s_lock); + return ret; +} + +bool hfi1_handle_kdeth_eflags(struct hfi1_ctxtdata *rcd, + struct hfi1_pportdata *ppd, + struct hfi1_packet *packet) +{ + struct hfi1_ibport *ibp = &ppd->ibport_data; + struct hfi1_devdata *dd = ppd->dd; + struct rvt_dev_info *rdi = &dd->verbs_dev.rdi; + u8 rcv_type = rhf_rcv_type(packet->rhf); + u8 rte = rhf_rcv_type_err(packet->rhf); + struct ib_header *hdr = packet->hdr; + struct ib_other_headers *ohdr = NULL; + int lnh = be16_to_cpu(hdr->lrh[0]) & 3; + u16 lid = be16_to_cpu(hdr->lrh[1]); + u8 opcode; + u32 qp_num, psn, ibpsn; + struct rvt_qp *qp; + struct hfi1_qp_priv *qpriv; + unsigned long flags; + bool ret = true; + struct rvt_ack_entry *e; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + + trace_hfi1_msg_handle_kdeth_eflags(NULL, "Kdeth error: rhf ", + packet->rhf); + if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR)) + return ret; + + packet->ohdr = &hdr->u.oth; + ohdr = packet->ohdr; + trace_input_ibhdr(rcd->dd, packet, !!(rhf_dc_info(packet->rhf))); + + /* Get the destination QP number. */ + qp_num = be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_qp) & + RVT_QPN_MASK; + if (lid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) + goto drop; + + psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff; + + rcu_read_lock(); + qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num); + if (!qp) + goto rcu_unlock; + + packet->qp = qp; + + /* Check for valid receive state. */ + spin_lock_irqsave(&qp->r_lock, flags); + if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) { + ibp->rvp.n_pkt_drops++; + goto r_unlock; + } + + if (packet->rhf & RHF_TID_ERR) { + /* For TIDERR and RC QPs preemptively schedule a NAK */ + u32 tlen = rhf_pkt_len(packet->rhf); /* in bytes */ + + /* Sanity check packet */ + if (tlen < 24) + goto r_unlock; + + /* + * Check for GRH. We should never get packets with GRH in this + * path. + */ + if (lnh == HFI1_LRH_GRH) + goto r_unlock; + + if (tid_rdma_tid_err(rcd, packet, rcv_type, opcode)) + goto r_unlock; + } + + /* handle TID RDMA READ */ + if (opcode == TID_OP(READ_RESP)) { + ibpsn = be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_psn); + ibpsn = mask_psn(ibpsn); + ret = handle_read_kdeth_eflags(rcd, packet, rcv_type, rte, psn, + ibpsn); + goto r_unlock; + } + + /* + * qp->s_tail_ack_queue points to the rvt_ack_entry currently being + * processed. These a completed sequentially so we can be sure that + * the pointer will not change until the entire request has completed. + */ + spin_lock(&qp->s_lock); + qpriv = qp->priv; + e = &qp->s_ack_queue[qpriv->r_tid_tail]; + req = ack_to_tid_req(e); + flow = &req->flows[req->clear_tail]; + trace_hfi1_eflags_err_write(qp, rcv_type, rte, psn); + trace_hfi1_rsp_handle_kdeth_eflags(qp, psn); + trace_hfi1_tid_write_rsp_handle_kdeth_eflags(qp); + trace_hfi1_tid_req_handle_kdeth_eflags(qp, 0, e->opcode, e->psn, + e->lpsn, req); + trace_hfi1_tid_flow_handle_kdeth_eflags(qp, req->clear_tail, flow); + + switch (rcv_type) { + case RHF_RCV_TYPE_EXPECTED: + switch (rte) { + case RHF_RTE_EXPECTED_FLOW_SEQ_ERR: + if (!(qpriv->s_flags & HFI1_R_TID_SW_PSN)) { + u64 reg; + + qpriv->s_flags |= HFI1_R_TID_SW_PSN; + /* + * The only sane way to get the amount of + * progress is to read the HW flow state. + */ + reg = read_uctxt_csr(dd, rcd->ctxt, + RCV_TID_FLOW_TABLE + + (8 * flow->idx)); + flow->flow_state.r_next_psn = mask_psn(reg); + qpriv->r_next_psn_kdeth = + flow->flow_state.r_next_psn; + goto nak_psn; + } else { + /* + * If the received PSN does not match the next + * expected PSN, NAK the packet. + * However, only do that if we know that the a + * NAK has already been sent. Otherwise, this + * mismatch could be due to packets that were + * already in flight. + */ + if (psn != flow->flow_state.r_next_psn) { + psn = flow->flow_state.r_next_psn; + goto nak_psn; + } + + qpriv->s_nak_state = 0; + /* + * If SW PSN verification is successful and this + * is the last packet in the segment, tell the + * caller to process it as a normal packet. + */ + if (psn == full_flow_psn(flow, + flow->flow_state.lpsn)) + ret = false; + qpriv->r_next_psn_kdeth = + ++flow->flow_state.r_next_psn; + } + break; + + case RHF_RTE_EXPECTED_FLOW_GEN_ERR: + goto nak_psn; + + default: + break; + } + break; + + case RHF_RCV_TYPE_ERROR: + switch (rte) { + case RHF_RTE_ERROR_OP_CODE_ERR: + case RHF_RTE_ERROR_KHDR_MIN_LEN_ERR: + case RHF_RTE_ERROR_KHDR_HCRC_ERR: + case RHF_RTE_ERROR_KHDR_KVER_ERR: + case RHF_RTE_ERROR_CONTEXT_ERR: + case RHF_RTE_ERROR_KHDR_TID_ERR: + default: + break; + } + default: + break; + } + +unlock: + spin_unlock(&qp->s_lock); +r_unlock: + spin_unlock_irqrestore(&qp->r_lock, flags); +rcu_unlock: + rcu_read_unlock(); +drop: + return ret; +nak_psn: + ibp->rvp.n_rc_seqnak++; + if (!qpriv->s_nak_state) { + qpriv->s_nak_state = IB_NAK_PSN_ERROR; + /* We are NAK'ing the next expected PSN */ + qpriv->s_nak_psn = mask_psn(flow->flow_state.r_next_psn); + qpriv->s_flags |= RVT_S_ACK_PENDING; + if (qpriv->r_tid_ack == HFI1_QP_WQE_INVALID) + qpriv->r_tid_ack = qpriv->r_tid_tail; + hfi1_schedule_tid_send(qp); + } + goto unlock; +} + +/* + * "Rewind" the TID request information. + * This means that we reset the state back to ACTIVE, + * find the proper flow, set the flow index to that flow, + * and reset the flow information. + */ +void hfi1_tid_rdma_restart_req(struct rvt_qp *qp, struct rvt_swqe *wqe, + u32 *bth2) +{ + struct tid_rdma_request *req = wqe_to_tid_req(wqe); + struct tid_rdma_flow *flow; + struct hfi1_qp_priv *qpriv = qp->priv; + int diff, delta_pkts; + u32 tididx = 0, i; + u16 fidx; + + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) { + *bth2 = mask_psn(qp->s_psn); + flow = find_flow_ib(req, *bth2, &fidx); + if (!flow) { + trace_hfi1_msg_tid_restart_req(/* msg */ + qp, "!!!!!! Could not find flow to restart: bth2 ", + (u64)*bth2); + trace_hfi1_tid_req_restart_req(qp, 0, wqe->wr.opcode, + wqe->psn, wqe->lpsn, + req); + return; + } + } else { + fidx = req->acked_tail; + flow = &req->flows[fidx]; + *bth2 = mask_psn(req->r_ack_psn); + } + + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) + delta_pkts = delta_psn(*bth2, flow->flow_state.ib_spsn); + else + delta_pkts = delta_psn(*bth2, + full_flow_psn(flow, + flow->flow_state.spsn)); + + trace_hfi1_tid_flow_restart_req(qp, fidx, flow); + diff = delta_pkts + flow->resync_npkts; + + flow->sent = 0; + flow->pkt = 0; + flow->tid_idx = 0; + flow->tid_offset = 0; + if (diff) { + for (tididx = 0; tididx < flow->tidcnt; tididx++) { + u32 tidentry = flow->tid_entry[tididx], tidlen, + tidnpkts, npkts; + + flow->tid_offset = 0; + tidlen = EXP_TID_GET(tidentry, LEN) * PAGE_SIZE; + tidnpkts = rvt_div_round_up_mtu(qp, tidlen); + npkts = min_t(u32, diff, tidnpkts); + flow->pkt += npkts; + flow->sent += (npkts == tidnpkts ? tidlen : + npkts * qp->pmtu); + flow->tid_offset += npkts * qp->pmtu; + diff -= npkts; + if (!diff) + break; + } + } + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) { + rvt_skip_sge(&qpriv->tid_ss, (req->cur_seg * req->seg_len) + + flow->sent, 0); + /* + * Packet PSN is based on flow_state.spsn + flow->pkt. However, + * during a RESYNC, the generation is incremented and the + * sequence is reset to 0. Since we've adjusted the npkts in the + * flow and the SGE has been sufficiently advanced, we have to + * adjust flow->pkt in order to calculate the correct PSN. + */ + flow->pkt -= flow->resync_npkts; + } + + if (flow->tid_offset == + EXP_TID_GET(flow->tid_entry[tididx], LEN) * PAGE_SIZE) { + tididx++; + flow->tid_offset = 0; + } + flow->tid_idx = tididx; + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) + /* Move flow_idx to correct index */ + req->flow_idx = fidx; + else + req->clear_tail = fidx; + + trace_hfi1_tid_flow_restart_req(qp, fidx, flow); + trace_hfi1_tid_req_restart_req(qp, 0, wqe->wr.opcode, wqe->psn, + wqe->lpsn, req); + req->state = TID_REQUEST_ACTIVE; + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) { + /* Reset all the flows that we are going to resend */ + fidx = CIRC_NEXT(fidx, MAX_FLOWS); + i = qpriv->s_tid_tail; + do { + for (; CIRC_CNT(req->setup_head, fidx, MAX_FLOWS); + fidx = CIRC_NEXT(fidx, MAX_FLOWS)) { + req->flows[fidx].sent = 0; + req->flows[fidx].pkt = 0; + req->flows[fidx].tid_idx = 0; + req->flows[fidx].tid_offset = 0; + req->flows[fidx].resync_npkts = 0; + } + if (i == qpriv->s_tid_cur) + break; + do { + i = (++i == qp->s_size ? 0 : i); + wqe = rvt_get_swqe_ptr(qp, i); + } while (wqe->wr.opcode != IB_WR_TID_RDMA_WRITE); + req = wqe_to_tid_req(wqe); + req->cur_seg = req->ack_seg; + fidx = req->acked_tail; + /* Pull req->clear_tail back */ + req->clear_tail = fidx; + } while (1); + } +} + +void hfi1_qp_kern_exp_rcv_clear_all(struct rvt_qp *qp) +{ + int i, ret; + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_flow_state *fs; + + if (qp->ibqp.qp_type != IB_QPT_RC || !HFI1_CAP_IS_KSET(TID_RDMA)) + return; + + /* + * First, clear the flow to help prevent any delayed packets from + * being delivered. + */ + fs = &qpriv->flow_state; + if (fs->index != RXE_NUM_TID_FLOWS) + hfi1_kern_clear_hw_flow(qpriv->rcd, qp); + + for (i = qp->s_acked; i != qp->s_head;) { + struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, i); + + if (++i == qp->s_size) + i = 0; + /* Free only locally allocated TID entries */ + if (wqe->wr.opcode != IB_WR_TID_RDMA_READ) + continue; + do { + struct hfi1_swqe_priv *priv = wqe->priv; + + ret = hfi1_kern_exp_rcv_clear(&priv->tid_req); + } while (!ret); + } + for (i = qp->s_acked_ack_queue; i != qp->r_head_ack_queue;) { + struct rvt_ack_entry *e = &qp->s_ack_queue[i]; + + if (++i == rvt_max_atomic(ib_to_rvt(qp->ibqp.device))) + i = 0; + /* Free only locally allocated TID entries */ + if (e->opcode != TID_OP(WRITE_REQ)) + continue; + do { + struct hfi1_ack_priv *priv = e->priv; + + ret = hfi1_kern_exp_rcv_clear(&priv->tid_req); + } while (!ret); + } +} + +bool hfi1_tid_rdma_wqe_interlock(struct rvt_qp *qp, struct rvt_swqe *wqe) +{ + struct rvt_swqe *prev; + struct hfi1_qp_priv *priv = qp->priv; + u32 s_prev; + struct tid_rdma_request *req; + + s_prev = (qp->s_cur == 0 ? qp->s_size : qp->s_cur) - 1; + prev = rvt_get_swqe_ptr(qp, s_prev); + + switch (wqe->wr.opcode) { + case IB_WR_SEND: + case IB_WR_SEND_WITH_IMM: + case IB_WR_SEND_WITH_INV: + case IB_WR_ATOMIC_CMP_AND_SWP: + case IB_WR_ATOMIC_FETCH_AND_ADD: + case IB_WR_RDMA_WRITE: + switch (prev->wr.opcode) { + case IB_WR_TID_RDMA_WRITE: + req = wqe_to_tid_req(prev); + if (req->ack_seg != req->total_segs) + goto interlock; + default: + break; + } + break; + case IB_WR_RDMA_READ: + if (prev->wr.opcode != IB_WR_TID_RDMA_WRITE) + break; + /* fall through */ + case IB_WR_TID_RDMA_READ: + switch (prev->wr.opcode) { + case IB_WR_RDMA_READ: + if (qp->s_acked != qp->s_cur) + goto interlock; + break; + case IB_WR_TID_RDMA_WRITE: + req = wqe_to_tid_req(prev); + if (req->ack_seg != req->total_segs) + goto interlock; + default: + break; + } + default: + break; + } + return false; + +interlock: + priv->s_flags |= HFI1_S_TID_WAIT_INTERLCK; + return true; +} + +/* Does @sge meet the alignment requirements for tid rdma? */ +static inline bool hfi1_check_sge_align(struct rvt_qp *qp, + struct rvt_sge *sge, int num_sge) +{ + int i; + + for (i = 0; i < num_sge; i++, sge++) { + trace_hfi1_sge_check_align(qp, i, sge); + if ((u64)sge->vaddr & ~PAGE_MASK || + sge->sge_length & ~PAGE_MASK) + return false; + } + return true; +} + +void setup_tid_rdma_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe) +{ + struct hfi1_qp_priv *qpriv = (struct hfi1_qp_priv *)qp->priv; + struct hfi1_swqe_priv *priv = wqe->priv; + struct tid_rdma_params *remote; + enum ib_wr_opcode new_opcode; + bool do_tid_rdma = false; + struct hfi1_pportdata *ppd = qpriv->rcd->ppd; + + if ((rdma_ah_get_dlid(&qp->remote_ah_attr) & ~((1 << ppd->lmc) - 1)) == + ppd->lid) + return; + if (qpriv->hdr_type != HFI1_PKT_TYPE_9B) + return; + + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + /* + * If TID RDMA is disabled by the negotiation, don't + * use it. + */ + if (!remote) + goto exit; + + if (wqe->wr.opcode == IB_WR_RDMA_READ) { + if (hfi1_check_sge_align(qp, &wqe->sg_list[0], + wqe->wr.num_sge)) { + new_opcode = IB_WR_TID_RDMA_READ; + do_tid_rdma = true; + } + } else if (wqe->wr.opcode == IB_WR_RDMA_WRITE) { + /* + * TID RDMA is enabled for this RDMA WRITE request iff: + * 1. The remote address is page-aligned, + * 2. The length is larger than the minimum segment size, + * 3. The length is page-multiple. + */ + if (!(wqe->rdma_wr.remote_addr & ~PAGE_MASK) && + !(wqe->length & ~PAGE_MASK)) { + new_opcode = IB_WR_TID_RDMA_WRITE; + do_tid_rdma = true; + } + } + + if (do_tid_rdma) { + if (hfi1_kern_exp_rcv_alloc_flows(&priv->tid_req, GFP_ATOMIC)) + goto exit; + wqe->wr.opcode = new_opcode; + priv->tid_req.seg_len = + min_t(u32, remote->max_len, wqe->length); + priv->tid_req.total_segs = + DIV_ROUND_UP(wqe->length, priv->tid_req.seg_len); + /* Compute the last PSN of the request */ + wqe->lpsn = wqe->psn; + if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) { + priv->tid_req.n_flows = remote->max_read; + qpriv->tid_r_reqs++; + wqe->lpsn += rvt_div_round_up_mtu(qp, wqe->length) - 1; + } else { + wqe->lpsn += priv->tid_req.total_segs - 1; + atomic_inc(&qpriv->n_requests); + } + + priv->tid_req.cur_seg = 0; + priv->tid_req.comp_seg = 0; + priv->tid_req.ack_seg = 0; + priv->tid_req.state = TID_REQUEST_INACTIVE; + /* + * Reset acked_tail. + * TID RDMA READ does not have ACKs so it does not + * update the pointer. We have to reset it so TID RDMA + * WRITE does not get confused. + */ + priv->tid_req.acked_tail = priv->tid_req.setup_head; + trace_hfi1_tid_req_setup_tid_wqe(qp, 1, wqe->wr.opcode, + wqe->psn, wqe->lpsn, + &priv->tid_req); + } +exit: + rcu_read_unlock(); +} + +/* TID RDMA WRITE functions */ + +u32 hfi1_build_tid_rdma_write_req(struct rvt_qp *qp, struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, + u32 *bth1, u32 *bth2, u32 *len) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_rdma_request *req = wqe_to_tid_req(wqe); + struct tid_rdma_params *remote; + + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + /* + * Set the number of flow to be used based on negotiated + * parameters. + */ + req->n_flows = remote->max_write; + req->state = TID_REQUEST_ACTIVE; + + KDETH_RESET(ohdr->u.tid_rdma.w_req.kdeth0, KVER, 0x1); + KDETH_RESET(ohdr->u.tid_rdma.w_req.kdeth1, JKEY, remote->jkey); + ohdr->u.tid_rdma.w_req.reth.vaddr = + cpu_to_be64(wqe->rdma_wr.remote_addr + (wqe->length - *len)); + ohdr->u.tid_rdma.w_req.reth.rkey = + cpu_to_be32(wqe->rdma_wr.rkey); + ohdr->u.tid_rdma.w_req.reth.length = cpu_to_be32(*len); + ohdr->u.tid_rdma.w_req.verbs_qp = cpu_to_be32(qp->remote_qpn); + *bth1 &= ~RVT_QPN_MASK; + *bth1 |= remote->qp; + qp->s_state = TID_OP(WRITE_REQ); + qp->s_flags |= HFI1_S_WAIT_TID_RESP; + *bth2 |= IB_BTH_REQ_ACK; + *len = 0; + + rcu_read_unlock(); + return sizeof(ohdr->u.tid_rdma.w_req) / sizeof(u32); +} + +void hfi1_compute_tid_rdma_flow_wt(void) +{ + /* + * Heuristic for computing the RNR timeout when waiting on the flow + * queue. Rather than a computationaly expensive exact estimate of when + * a flow will be available, we assume that if a QP is at position N in + * the flow queue it has to wait approximately (N + 1) * (number of + * segments between two sync points), assuming PMTU of 4K. The rationale + * for this is that flows are released and recycled at each sync point. + */ + tid_rdma_flow_wt = MAX_TID_FLOW_PSN * enum_to_mtu(OPA_MTU_4096) / + TID_RDMA_MAX_SEGMENT_SIZE; +} + +static u32 position_in_queue(struct hfi1_qp_priv *qpriv, + struct tid_queue *queue) +{ + return qpriv->tid_enqueue - queue->dequeue; +} + +/* + * @qp: points to rvt_qp context. + * @to_seg: desired RNR timeout in segments. + * Return: index of the next highest timeout in the ib_hfi1_rnr_table[] + */ +static u32 hfi1_compute_tid_rnr_timeout(struct rvt_qp *qp, u32 to_seg) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + u64 timeout; + u32 bytes_per_us; + u8 i; + + bytes_per_us = active_egress_rate(qpriv->rcd->ppd) / 8; + timeout = (to_seg * TID_RDMA_MAX_SEGMENT_SIZE) / bytes_per_us; + /* + * Find the next highest value in the RNR table to the required + * timeout. This gives the responder some padding. + */ + for (i = 1; i <= IB_AETH_CREDIT_MASK; i++) + if (rvt_rnr_tbl_to_usec(i) >= timeout) + return i; + return 0; +} + +/** + * Central place for resource allocation at TID write responder, + * is called from write_req and write_data interrupt handlers as + * well as the send thread when a queued QP is scheduled for + * resource allocation. + * + * Iterates over (a) segments of a request and then (b) queued requests + * themselves to allocate resources for up to local->max_write + * segments across multiple requests. Stop allocating when we + * hit a sync point, resume allocating after data packets at + * sync point have been received. + * + * Resource allocation and sending of responses is decoupled. The + * request/segment which are being allocated and sent are as follows. + * Resources are allocated for: + * [request: qpriv->r_tid_alloc, segment: req->alloc_seg] + * The send thread sends: + * [request: qp->s_tail_ack_queue, segment:req->cur_seg] + */ +static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx) +{ + struct tid_rdma_request *req; + struct hfi1_qp_priv *qpriv = qp->priv; + struct hfi1_ctxtdata *rcd = qpriv->rcd; + struct tid_rdma_params *local = &qpriv->tid_rdma.local; + struct rvt_ack_entry *e; + u32 npkts, to_seg; + bool last; + int ret = 0; + + lockdep_assert_held(&qp->s_lock); + + while (1) { + trace_hfi1_rsp_tid_write_alloc_res(qp, 0); + trace_hfi1_tid_write_rsp_alloc_res(qp); + /* + * Don't allocate more segments if a RNR NAK has already been + * scheduled to avoid messing up qp->r_psn: the RNR NAK will + * be sent only when all allocated segments have been sent. + * However, if more segments are allocated before that, TID RDMA + * WRITE RESP packets will be sent out for these new segments + * before the RNR NAK packet. When the requester receives the + * RNR NAK packet, it will restart with qp->s_last_psn + 1, + * which does not match qp->r_psn and will be dropped. + * Consequently, the requester will exhaust its retries and + * put the qp into error state. + */ + if (qpriv->rnr_nak_state == TID_RNR_NAK_SEND) + break; + + /* No requests left to process */ + if (qpriv->r_tid_alloc == qpriv->r_tid_head) { + /* If all data has been received, clear the flow */ + if (qpriv->flow_state.index < RXE_NUM_TID_FLOWS && + !qpriv->alloc_w_segs) + hfi1_kern_clear_hw_flow(rcd, qp); + break; + } + + e = &qp->s_ack_queue[qpriv->r_tid_alloc]; + if (e->opcode != TID_OP(WRITE_REQ)) + goto next_req; + req = ack_to_tid_req(e); + trace_hfi1_tid_req_write_alloc_res(qp, 0, e->opcode, e->psn, + e->lpsn, req); + /* Finished allocating for all segments of this request */ + if (req->alloc_seg >= req->total_segs) + goto next_req; + + /* Can allocate only a maximum of local->max_write for a QP */ + if (qpriv->alloc_w_segs >= local->max_write) + break; + + /* Don't allocate at a sync point with data packets pending */ + if (qpriv->sync_pt && qpriv->alloc_w_segs) + break; + + /* All data received at the sync point, continue */ + if (qpriv->sync_pt && !qpriv->alloc_w_segs) { + hfi1_kern_clear_hw_flow(rcd, qp); + qpriv->sync_pt = false; + if (qpriv->s_flags & HFI1_R_TID_SW_PSN) + qpriv->s_flags &= ~HFI1_R_TID_SW_PSN; + } + + /* Allocate flow if we don't have one */ + if (qpriv->flow_state.index >= RXE_NUM_TID_FLOWS) { + ret = hfi1_kern_setup_hw_flow(qpriv->rcd, qp); + if (ret) { + to_seg = tid_rdma_flow_wt * + position_in_queue(qpriv, + &rcd->flow_queue); + break; + } + } + + npkts = rvt_div_round_up_mtu(qp, req->seg_len); + + /* + * We are at a sync point if we run out of KDETH PSN space. + * Last PSN of every generation is reserved for RESYNC. + */ + if (qpriv->flow_state.psn + npkts > MAX_TID_FLOW_PSN - 1) { + qpriv->sync_pt = true; + break; + } + + /* + * If overtaking req->acked_tail, send an RNR NAK. Because the + * QP is not queued in this case, and the issue can only be + * caused due a delay in scheduling the second leg which we + * cannot estimate, we use a rather arbitrary RNR timeout of + * (MAX_FLOWS / 2) segments + */ + if (!CIRC_SPACE(req->setup_head, req->acked_tail, + MAX_FLOWS)) { + ret = -EAGAIN; + to_seg = MAX_FLOWS >> 1; + qpriv->s_flags |= RVT_S_ACK_PENDING; + hfi1_schedule_tid_send(qp); + break; + } + + /* Try to allocate rcv array / TID entries */ + ret = hfi1_kern_exp_rcv_setup(req, &req->ss, &last); + if (ret == -EAGAIN) + to_seg = position_in_queue(qpriv, &rcd->rarr_queue); + if (ret) + break; + + qpriv->alloc_w_segs++; + req->alloc_seg++; + continue; +next_req: + /* Begin processing the next request */ + if (++qpriv->r_tid_alloc > + rvt_size_atomic(ib_to_rvt(qp->ibqp.device))) + qpriv->r_tid_alloc = 0; + } + + /* + * Schedule an RNR NAK to be sent if (a) flow or rcv array allocation + * has failed (b) we are called from the rcv handler interrupt context + * (c) an RNR NAK has not already been scheduled + */ + if (ret == -EAGAIN && intr_ctx && !qp->r_nak_state) + goto send_rnr_nak; + + return; + +send_rnr_nak: + lockdep_assert_held(&qp->r_lock); + + /* Set r_nak_state to prevent unrelated events from generating NAK's */ + qp->r_nak_state = hfi1_compute_tid_rnr_timeout(qp, to_seg) | IB_RNR_NAK; + + /* Pull back r_psn to the segment being RNR NAK'd */ + qp->r_psn = e->psn + req->alloc_seg; + qp->r_ack_psn = qp->r_psn; + /* + * Pull back r_head_ack_queue to the ack entry following the request + * being RNR NAK'd. This allows resources to be allocated to the request + * if the queued QP is scheduled. + */ + qp->r_head_ack_queue = qpriv->r_tid_alloc + 1; + if (qp->r_head_ack_queue > rvt_size_atomic(ib_to_rvt(qp->ibqp.device))) + qp->r_head_ack_queue = 0; + qpriv->r_tid_head = qp->r_head_ack_queue; + /* + * These send side fields are used in make_rc_ack(). They are set in + * hfi1_send_rc_ack() but must be set here before dropping qp->s_lock + * for consistency + */ + qp->s_nak_state = qp->r_nak_state; + qp->s_ack_psn = qp->r_ack_psn; + /* + * Clear the ACK PENDING flag to prevent unwanted ACK because we + * have modified qp->s_ack_psn here. + */ + qp->s_flags &= ~(RVT_S_ACK_PENDING); + + trace_hfi1_rsp_tid_write_alloc_res(qp, qp->r_psn); + /* + * qpriv->rnr_nak_state is used to determine when the scheduled RNR NAK + * has actually been sent. qp->s_flags RVT_S_ACK_PENDING bit cannot be + * used for this because qp->s_lock is dropped before calling + * hfi1_send_rc_ack() leading to inconsistency between the receive + * interrupt handlers and the send thread in make_rc_ack() + */ + qpriv->rnr_nak_state = TID_RNR_NAK_SEND; + + /* + * Schedule RNR NAK to be sent. RNR NAK's are scheduled from the receive + * interrupt handlers but will be sent from the send engine behind any + * previous responses that may have been scheduled + */ + rc_defered_ack(rcd, qp); +} + +void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet) +{ + /* HANDLER FOR TID RDMA WRITE REQUEST packet (Responder side)*/ + + /* + * 1. Verify TID RDMA WRITE REQ as per IB_OPCODE_RC_RDMA_WRITE_FIRST + * (see hfi1_rc_rcv()) + * - Don't allow 0-length requests. + * 2. Put TID RDMA WRITE REQ into the response queueu (s_ack_queue) + * - Setup struct tid_rdma_req with request info + * - Prepare struct tid_rdma_flow array? + * 3. Set the qp->s_ack_state as state diagram in design doc. + * 4. Set RVT_S_RESP_PENDING in s_flags. + * 5. Kick the send engine (hfi1_schedule_send()) + */ + struct hfi1_ctxtdata *rcd = packet->rcd; + struct rvt_qp *qp = packet->qp; + struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_ack_entry *e; + unsigned long flags; + struct ib_reth *reth; + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_rdma_request *req; + u32 bth0, psn, len, rkey, num_segs; + bool is_fecn; + u8 next; + u64 vaddr; + int diff; + + bth0 = be32_to_cpu(ohdr->bth[0]); + if (hfi1_ruc_check_hdr(ibp, packet)) + return; + + is_fecn = process_ecn(qp, packet); + psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + trace_hfi1_rsp_rcv_tid_write_req(qp, psn); + + if (qp->state == IB_QPS_RTR && !(qp->r_flags & RVT_R_COMM_EST)) + rvt_comm_est(qp); + + if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE))) + goto nack_inv; + + reth = &ohdr->u.tid_rdma.w_req.reth; + vaddr = be64_to_cpu(reth->vaddr); + len = be32_to_cpu(reth->length); + + num_segs = DIV_ROUND_UP(len, qpriv->tid_rdma.local.max_len); + diff = delta_psn(psn, qp->r_psn); + if (unlikely(diff)) { + if (tid_rdma_rcv_error(packet, ohdr, qp, psn, diff)) + return; + goto send_ack; + } + + /* + * The resent request which was previously RNR NAK'd is inserted at the + * location of the original request, which is one entry behind + * r_head_ack_queue + */ + if (qpriv->rnr_nak_state) + qp->r_head_ack_queue = qp->r_head_ack_queue ? + qp->r_head_ack_queue - 1 : + rvt_size_atomic(ib_to_rvt(qp->ibqp.device)); + + /* We've verified the request, insert it into the ack queue. */ + next = qp->r_head_ack_queue + 1; + if (next > rvt_size_atomic(ib_to_rvt(qp->ibqp.device))) + next = 0; + spin_lock_irqsave(&qp->s_lock, flags); + if (unlikely(next == qp->s_acked_ack_queue)) { + if (!qp->s_ack_queue[next].sent) + goto nack_inv_unlock; + update_ack_queue(qp, next); + } + e = &qp->s_ack_queue[qp->r_head_ack_queue]; + req = ack_to_tid_req(e); + + /* Bring previously RNR NAK'd request back to life */ + if (qpriv->rnr_nak_state) { + qp->r_nak_state = 0; + qp->s_nak_state = 0; + qpriv->rnr_nak_state = TID_RNR_NAK_INIT; + qp->r_psn = e->lpsn + 1; + req->state = TID_REQUEST_INIT; + goto update_head; + } + + if (e->rdma_sge.mr) { + rvt_put_mr(e->rdma_sge.mr); + e->rdma_sge.mr = NULL; + } + + /* The length needs to be in multiples of PAGE_SIZE */ + if (!len || len & ~PAGE_MASK) + goto nack_inv_unlock; + + rkey = be32_to_cpu(reth->rkey); + qp->r_len = len; + + if (e->opcode == TID_OP(WRITE_REQ) && + (req->setup_head != req->clear_tail || + req->clear_tail != req->acked_tail)) + goto nack_inv_unlock; + + if (unlikely(!rvt_rkey_ok(qp, &e->rdma_sge, qp->r_len, vaddr, + rkey, IB_ACCESS_REMOTE_WRITE))) + goto nack_acc; + + qp->r_psn += num_segs - 1; + + e->opcode = (bth0 >> 24) & 0xff; + e->psn = psn; + e->lpsn = qp->r_psn; + e->sent = 0; + + req->n_flows = min_t(u16, num_segs, qpriv->tid_rdma.local.max_write); + req->state = TID_REQUEST_INIT; + req->cur_seg = 0; + req->comp_seg = 0; + req->ack_seg = 0; + req->alloc_seg = 0; + req->isge = 0; + req->seg_len = qpriv->tid_rdma.local.max_len; + req->total_len = len; + req->total_segs = num_segs; + req->r_flow_psn = e->psn; + req->ss.sge = e->rdma_sge; + req->ss.num_sge = 1; + + req->flow_idx = req->setup_head; + req->clear_tail = req->setup_head; + req->acked_tail = req->setup_head; + + qp->r_state = e->opcode; + qp->r_nak_state = 0; + /* + * We need to increment the MSN here instead of when we + * finish sending the result since a duplicate request would + * increment it more than once. + */ + qp->r_msn++; + qp->r_psn++; + + trace_hfi1_tid_req_rcv_write_req(qp, 0, e->opcode, e->psn, e->lpsn, + req); + + if (qpriv->r_tid_tail == HFI1_QP_WQE_INVALID) { + qpriv->r_tid_tail = qp->r_head_ack_queue; + } else if (qpriv->r_tid_tail == qpriv->r_tid_head) { + struct tid_rdma_request *ptr; + + e = &qp->s_ack_queue[qpriv->r_tid_tail]; + ptr = ack_to_tid_req(e); + + if (e->opcode != TID_OP(WRITE_REQ) || + ptr->comp_seg == ptr->total_segs) { + if (qpriv->r_tid_tail == qpriv->r_tid_ack) + qpriv->r_tid_ack = qp->r_head_ack_queue; + qpriv->r_tid_tail = qp->r_head_ack_queue; + } + } +update_head: + qp->r_head_ack_queue = next; + qpriv->r_tid_head = qp->r_head_ack_queue; + + hfi1_tid_write_alloc_resources(qp, true); + trace_hfi1_tid_write_rsp_rcv_req(qp); + + /* Schedule the send tasklet. */ + qp->s_flags |= RVT_S_RESP_PENDING; + hfi1_schedule_send(qp); + + spin_unlock_irqrestore(&qp->s_lock, flags); + if (is_fecn) + goto send_ack; + return; + +nack_inv_unlock: + spin_unlock_irqrestore(&qp->s_lock, flags); +nack_inv: + rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR); + qp->r_nak_state = IB_NAK_INVALID_REQUEST; + qp->r_ack_psn = qp->r_psn; + /* Queue NAK for later */ + rc_defered_ack(rcd, qp); + return; +nack_acc: + spin_unlock_irqrestore(&qp->s_lock, flags); + rvt_rc_error(qp, IB_WC_LOC_PROT_ERR); + qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR; + qp->r_ack_psn = qp->r_psn; +send_ack: + hfi1_send_rc_ack(packet, is_fecn); +} + +u32 hfi1_build_tid_rdma_write_resp(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_other_headers *ohdr, u32 *bth1, + u32 bth2, u32 *len, + struct rvt_sge_state **ss) +{ + struct hfi1_ack_priv *epriv = e->priv; + struct tid_rdma_request *req = &epriv->tid_req; + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_rdma_flow *flow = NULL; + u32 resp_len = 0, hdwords = 0; + void *resp_addr = NULL; + struct tid_rdma_params *remote; + + trace_hfi1_tid_req_build_write_resp(qp, 0, e->opcode, e->psn, e->lpsn, + req); + trace_hfi1_tid_write_rsp_build_resp(qp); + trace_hfi1_rsp_build_tid_write_resp(qp, bth2); + flow = &req->flows[req->flow_idx]; + switch (req->state) { + default: + /* + * Try to allocate resources here in case QP was queued and was + * later scheduled when resources became available + */ + hfi1_tid_write_alloc_resources(qp, false); + + /* We've already sent everything which is ready */ + if (req->cur_seg >= req->alloc_seg) + goto done; + + /* + * Resources can be assigned but responses cannot be sent in + * rnr_nak state, till the resent request is received + */ + if (qpriv->rnr_nak_state == TID_RNR_NAK_SENT) + goto done; + + req->state = TID_REQUEST_ACTIVE; + trace_hfi1_tid_flow_build_write_resp(qp, req->flow_idx, flow); + req->flow_idx = CIRC_NEXT(req->flow_idx, MAX_FLOWS); + hfi1_add_tid_reap_timer(qp); + break; + + case TID_REQUEST_RESEND_ACTIVE: + case TID_REQUEST_RESEND: + trace_hfi1_tid_flow_build_write_resp(qp, req->flow_idx, flow); + req->flow_idx = CIRC_NEXT(req->flow_idx, MAX_FLOWS); + if (!CIRC_CNT(req->setup_head, req->flow_idx, MAX_FLOWS)) + req->state = TID_REQUEST_ACTIVE; + + hfi1_mod_tid_reap_timer(qp); + break; + } + flow->flow_state.resp_ib_psn = bth2; + resp_addr = (void *)flow->tid_entry; + resp_len = sizeof(*flow->tid_entry) * flow->tidcnt; + req->cur_seg++; + + memset(&ohdr->u.tid_rdma.w_rsp, 0, sizeof(ohdr->u.tid_rdma.w_rsp)); + epriv->ss.sge.vaddr = resp_addr; + epriv->ss.sge.sge_length = resp_len; + epriv->ss.sge.length = epriv->ss.sge.sge_length; + /* + * We can safely zero these out. Since the first SGE covers the + * entire packet, nothing else should even look at the MR. + */ + epriv->ss.sge.mr = NULL; + epriv->ss.sge.m = 0; + epriv->ss.sge.n = 0; + + epriv->ss.sg_list = NULL; + epriv->ss.total_len = epriv->ss.sge.sge_length; + epriv->ss.num_sge = 1; + + *ss = &epriv->ss; + *len = epriv->ss.total_len; + + /* Construct the TID RDMA WRITE RESP packet header */ + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + + KDETH_RESET(ohdr->u.tid_rdma.w_rsp.kdeth0, KVER, 0x1); + KDETH_RESET(ohdr->u.tid_rdma.w_rsp.kdeth1, JKEY, remote->jkey); + ohdr->u.tid_rdma.w_rsp.aeth = rvt_compute_aeth(qp); + ohdr->u.tid_rdma.w_rsp.tid_flow_psn = + cpu_to_be32((flow->flow_state.generation << + HFI1_KDETH_BTH_SEQ_SHIFT) | + (flow->flow_state.spsn & + HFI1_KDETH_BTH_SEQ_MASK)); + ohdr->u.tid_rdma.w_rsp.tid_flow_qp = + cpu_to_be32(qpriv->tid_rdma.local.qp | + ((flow->idx & TID_RDMA_DESTQP_FLOW_MASK) << + TID_RDMA_DESTQP_FLOW_SHIFT) | + qpriv->rcd->ctxt); + ohdr->u.tid_rdma.w_rsp.verbs_qp = cpu_to_be32(qp->remote_qpn); + *bth1 = remote->qp; + rcu_read_unlock(); + hdwords = sizeof(ohdr->u.tid_rdma.w_rsp) / sizeof(u32); + qpriv->pending_tid_w_segs++; +done: + return hdwords; +} + +static void hfi1_add_tid_reap_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + + lockdep_assert_held(&qp->s_lock); + if (!(qpriv->s_flags & HFI1_R_TID_RSC_TIMER)) { + qpriv->s_flags |= HFI1_R_TID_RSC_TIMER; + qpriv->s_tid_timer.expires = jiffies + + qpriv->tid_timer_timeout_jiffies; + add_timer(&qpriv->s_tid_timer); + } +} + +static void hfi1_mod_tid_reap_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + + lockdep_assert_held(&qp->s_lock); + qpriv->s_flags |= HFI1_R_TID_RSC_TIMER; + mod_timer(&qpriv->s_tid_timer, jiffies + + qpriv->tid_timer_timeout_jiffies); +} + +static int hfi1_stop_tid_reap_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + int rval = 0; + + lockdep_assert_held(&qp->s_lock); + if (qpriv->s_flags & HFI1_R_TID_RSC_TIMER) { + rval = del_timer(&qpriv->s_tid_timer); + qpriv->s_flags &= ~HFI1_R_TID_RSC_TIMER; + } + return rval; +} + +void hfi1_del_tid_reap_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + + del_timer_sync(&qpriv->s_tid_timer); + qpriv->s_flags &= ~HFI1_R_TID_RSC_TIMER; +} + +static void hfi1_tid_timeout(struct timer_list *t) +{ + struct hfi1_qp_priv *qpriv = from_timer(qpriv, t, s_tid_timer); + struct rvt_qp *qp = qpriv->owner; + struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device); + unsigned long flags; + u32 i; + + spin_lock_irqsave(&qp->r_lock, flags); + spin_lock(&qp->s_lock); + if (qpriv->s_flags & HFI1_R_TID_RSC_TIMER) { + dd_dev_warn(dd_from_ibdev(qp->ibqp.device), "[QP%u] %s %d\n", + qp->ibqp.qp_num, __func__, __LINE__); + trace_hfi1_msg_tid_timeout(/* msg */ + qp, "resource timeout = ", + (u64)qpriv->tid_timer_timeout_jiffies); + hfi1_stop_tid_reap_timer(qp); + /* + * Go though the entire ack queue and clear any outstanding + * HW flow and RcvArray resources. + */ + hfi1_kern_clear_hw_flow(qpriv->rcd, qp); + for (i = 0; i < rvt_max_atomic(rdi); i++) { + struct tid_rdma_request *req = + ack_to_tid_req(&qp->s_ack_queue[i]); + + hfi1_kern_exp_rcv_clear_all(req); + } + spin_unlock(&qp->s_lock); + if (qp->ibqp.event_handler) { + struct ib_event ev; + + ev.device = qp->ibqp.device; + ev.element.qp = &qp->ibqp; + ev.event = IB_EVENT_QP_FATAL; + qp->ibqp.event_handler(&ev, qp->ibqp.qp_context); + } + rvt_rc_error(qp, IB_WC_RESP_TIMEOUT_ERR); + goto unlock_r_lock; + } + spin_unlock(&qp->s_lock); +unlock_r_lock: + spin_unlock_irqrestore(&qp->r_lock, flags); +} + +void hfi1_rc_rcv_tid_rdma_write_resp(struct hfi1_packet *packet) +{ + /* HANDLER FOR TID RDMA WRITE RESPONSE packet (Requestor side */ + + /* + * 1. Find matching SWQE + * 2. Check that TIDENTRY array has enough space for a complete + * segment. If not, put QP in error state. + * 3. Save response data in struct tid_rdma_req and struct tid_rdma_flow + * 4. Remove HFI1_S_WAIT_TID_RESP from s_flags. + * 5. Set qp->s_state + * 6. Kick the send engine (hfi1_schedule_send()) + */ + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *qpriv = qp->priv; + struct hfi1_ctxtdata *rcd = packet->rcd; + struct rvt_swqe *wqe; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + enum ib_wc_status status; + u32 opcode, aeth, psn, flow_psn, i, tidlen = 0, pktlen; + bool is_fecn; + unsigned long flags; + + is_fecn = process_ecn(qp, packet); + psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + aeth = be32_to_cpu(ohdr->u.tid_rdma.w_rsp.aeth); + opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff; + + spin_lock_irqsave(&qp->s_lock, flags); + + /* Ignore invalid responses */ + if (cmp_psn(psn, qp->s_next_psn) >= 0) + goto ack_done; + + /* Ignore duplicate responses. */ + if (unlikely(cmp_psn(psn, qp->s_last_psn) <= 0)) + goto ack_done; + + if (unlikely(qp->s_acked == qp->s_tail)) + goto ack_done; + + /* + * If we are waiting for a particular packet sequence number + * due to a request being resent, check for it. Otherwise, + * ensure that we haven't missed anything. + */ + if (qp->r_flags & RVT_R_RDMAR_SEQ) { + if (cmp_psn(psn, qp->s_last_psn + 1) != 0) + goto ack_done; + qp->r_flags &= ~RVT_R_RDMAR_SEQ; + } + + wqe = rvt_get_swqe_ptr(qp, qpriv->s_tid_cur); + if (unlikely(wqe->wr.opcode != IB_WR_TID_RDMA_WRITE)) + goto ack_op_err; + + req = wqe_to_tid_req(wqe); + /* + * If we've lost ACKs and our acked_tail pointer is too far + * behind, don't overwrite segments. Just drop the packet and + * let the reliability protocol take care of it. + */ + if (!CIRC_SPACE(req->setup_head, req->acked_tail, MAX_FLOWS)) + goto ack_done; + + /* + * The call to do_rc_ack() should be last in the chain of + * packet checks because it will end up updating the QP state. + * Therefore, anything that would prevent the packet from + * being accepted as a successful response should be prior + * to it. + */ + if (!do_rc_ack(qp, aeth, psn, opcode, 0, rcd)) + goto ack_done; + + trace_hfi1_ack(qp, psn); + + flow = &req->flows[req->setup_head]; + flow->pkt = 0; + flow->tid_idx = 0; + flow->tid_offset = 0; + flow->sent = 0; + flow->resync_npkts = 0; + flow->tid_qpn = be32_to_cpu(ohdr->u.tid_rdma.w_rsp.tid_flow_qp); + flow->idx = (flow->tid_qpn >> TID_RDMA_DESTQP_FLOW_SHIFT) & + TID_RDMA_DESTQP_FLOW_MASK; + flow_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.w_rsp.tid_flow_psn)); + flow->flow_state.generation = flow_psn >> HFI1_KDETH_BTH_SEQ_SHIFT; + flow->flow_state.spsn = flow_psn & HFI1_KDETH_BTH_SEQ_MASK; + flow->flow_state.resp_ib_psn = psn; + flow->length = min_t(u32, req->seg_len, + (wqe->length - (req->comp_seg * req->seg_len))); + + flow->npkts = rvt_div_round_up_mtu(qp, flow->length); + flow->flow_state.lpsn = flow->flow_state.spsn + + flow->npkts - 1; + /* payload length = packet length - (header length + ICRC length) */ + pktlen = packet->tlen - (packet->hlen + 4); + if (pktlen > sizeof(flow->tid_entry)) { + status = IB_WC_LOC_LEN_ERR; + goto ack_err; + } + memcpy(flow->tid_entry, packet->ebuf, pktlen); + flow->tidcnt = pktlen / sizeof(*flow->tid_entry); + trace_hfi1_tid_flow_rcv_write_resp(qp, req->setup_head, flow); + + req->comp_seg++; + trace_hfi1_tid_write_sender_rcv_resp(qp, 0); + /* + * Walk the TID_ENTRY list to make sure we have enough space for a + * complete segment. + */ + for (i = 0; i < flow->tidcnt; i++) { + trace_hfi1_tid_entry_rcv_write_resp(/* entry */ + qp, i, flow->tid_entry[i]); + if (!EXP_TID_GET(flow->tid_entry[i], LEN)) { + status = IB_WC_LOC_LEN_ERR; + goto ack_err; + } + tidlen += EXP_TID_GET(flow->tid_entry[i], LEN); + } + if (tidlen * PAGE_SIZE < flow->length) { + status = IB_WC_LOC_LEN_ERR; + goto ack_err; + } + + trace_hfi1_tid_req_rcv_write_resp(qp, 0, wqe->wr.opcode, wqe->psn, + wqe->lpsn, req); + /* + * If this is the first response for this request, set the initial + * flow index to the current flow. + */ + if (!cmp_psn(psn, wqe->psn)) { + req->r_last_acked = mask_psn(wqe->psn - 1); + /* Set acked flow index to head index */ + req->acked_tail = req->setup_head; + } + + /* advance circular buffer head */ + req->setup_head = CIRC_NEXT(req->setup_head, MAX_FLOWS); + req->state = TID_REQUEST_ACTIVE; + + /* + * If all responses for this TID RDMA WRITE request have been received + * advance the pointer to the next one. + * Since TID RDMA requests could be mixed in with regular IB requests, + * they might not appear sequentially in the queue. Therefore, the + * next request needs to be "found". + */ + if (qpriv->s_tid_cur != qpriv->s_tid_head && + req->comp_seg == req->total_segs) { + for (i = qpriv->s_tid_cur + 1; ; i++) { + if (i == qp->s_size) + i = 0; + wqe = rvt_get_swqe_ptr(qp, i); + if (i == qpriv->s_tid_head) + break; + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) + break; + } + qpriv->s_tid_cur = i; + } + qp->s_flags &= ~HFI1_S_WAIT_TID_RESP; + + hfi1_schedule_tid_send(qp); + goto ack_done; + +ack_op_err: + status = IB_WC_LOC_QP_OP_ERR; +ack_err: + rvt_error_qp(qp, status); +ack_done: + spin_unlock_irqrestore(&qp->s_lock, flags); + if (is_fecn) + hfi1_send_rc_ack(packet, is_fecn); +} + +bool hfi1_build_tid_rdma_packet(struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, + u32 *bth1, u32 *bth2, u32 *len) +{ + struct tid_rdma_request *req = wqe_to_tid_req(wqe); + struct tid_rdma_flow *flow = &req->flows[req->clear_tail]; + struct tid_rdma_params *remote; + struct rvt_qp *qp = req->qp; + struct hfi1_qp_priv *qpriv = qp->priv; + u32 tidentry = flow->tid_entry[flow->tid_idx]; + u32 tidlen = EXP_TID_GET(tidentry, LEN) << PAGE_SHIFT; + struct tid_rdma_write_data *wd = &ohdr->u.tid_rdma.w_data; + u32 next_offset, om = KDETH_OM_LARGE; + bool last_pkt; + + if (!tidlen) { + hfi1_trdma_send_complete(qp, wqe, IB_WC_REM_INV_RD_REQ_ERR); + rvt_error_qp(qp, IB_WC_REM_INV_RD_REQ_ERR); + } + + *len = min_t(u32, qp->pmtu, tidlen - flow->tid_offset); + flow->sent += *len; + next_offset = flow->tid_offset + *len; + last_pkt = (flow->tid_idx == (flow->tidcnt - 1) && + next_offset >= tidlen) || (flow->sent >= flow->length); + trace_hfi1_tid_entry_build_write_data(qp, flow->tid_idx, tidentry); + trace_hfi1_tid_flow_build_write_data(qp, req->clear_tail, flow); + + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + KDETH_RESET(wd->kdeth0, KVER, 0x1); + KDETH_SET(wd->kdeth0, SH, !last_pkt); + KDETH_SET(wd->kdeth0, INTR, !!(!last_pkt && remote->urg)); + KDETH_SET(wd->kdeth0, TIDCTRL, EXP_TID_GET(tidentry, CTRL)); + KDETH_SET(wd->kdeth0, TID, EXP_TID_GET(tidentry, IDX)); + KDETH_SET(wd->kdeth0, OM, om == KDETH_OM_LARGE); + KDETH_SET(wd->kdeth0, OFFSET, flow->tid_offset / om); + KDETH_RESET(wd->kdeth1, JKEY, remote->jkey); + wd->verbs_qp = cpu_to_be32(qp->remote_qpn); + rcu_read_unlock(); + + *bth1 = flow->tid_qpn; + *bth2 = mask_psn(((flow->flow_state.spsn + flow->pkt++) & + HFI1_KDETH_BTH_SEQ_MASK) | + (flow->flow_state.generation << + HFI1_KDETH_BTH_SEQ_SHIFT)); + if (last_pkt) { + /* PSNs are zero-based, so +1 to count number of packets */ + if (flow->flow_state.lpsn + 1 + + rvt_div_round_up_mtu(qp, req->seg_len) > + MAX_TID_FLOW_PSN) + req->state = TID_REQUEST_SYNC; + *bth2 |= IB_BTH_REQ_ACK; + } + + if (next_offset >= tidlen) { + flow->tid_offset = 0; + flow->tid_idx++; + } else { + flow->tid_offset = next_offset; + } + return last_pkt; +} + +void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet) +{ + struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *priv = qp->priv; + struct hfi1_ctxtdata *rcd = priv->rcd; + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_ack_entry *e; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + struct hfi1_ibdev *dev = to_idev(qp->ibqp.device); + unsigned long flags; + u32 psn, next; + u8 opcode; + + psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff; + + /* + * All error handling should be done by now. If we are here, the packet + * is either good or been accepted by the error handler. + */ + spin_lock_irqsave(&qp->s_lock, flags); + e = &qp->s_ack_queue[priv->r_tid_tail]; + req = ack_to_tid_req(e); + flow = &req->flows[req->clear_tail]; + if (cmp_psn(psn, full_flow_psn(flow, flow->flow_state.lpsn))) { + if (cmp_psn(psn, flow->flow_state.r_next_psn)) + goto send_nak; + flow->flow_state.r_next_psn++; + goto exit; + } + flow->flow_state.r_next_psn = mask_psn(psn + 1); + hfi1_kern_exp_rcv_clear(req); + priv->alloc_w_segs--; + rcd->flows[flow->idx].psn = psn & HFI1_KDETH_BTH_SEQ_MASK; + req->comp_seg++; + priv->s_nak_state = 0; + + /* + * Release the flow if one of the following conditions has been met: + * - The request has reached a sync point AND all outstanding + * segments have been completed, or + * - The entire request is complete and there are no more requests + * (of any kind) in the queue. + */ + trace_hfi1_rsp_rcv_tid_write_data(qp, psn); + trace_hfi1_tid_req_rcv_write_data(qp, 0, e->opcode, e->psn, e->lpsn, + req); + trace_hfi1_tid_write_rsp_rcv_data(qp); + if (priv->r_tid_ack == HFI1_QP_WQE_INVALID) + priv->r_tid_ack = priv->r_tid_tail; + + if (opcode == TID_OP(WRITE_DATA_LAST)) { + for (next = priv->r_tid_tail + 1; ; next++) { + if (next > rvt_size_atomic(&dev->rdi)) + next = 0; + if (next == priv->r_tid_head) + break; + e = &qp->s_ack_queue[next]; + if (e->opcode == TID_OP(WRITE_REQ)) + break; + } + priv->r_tid_tail = next; + if (++qp->s_acked_ack_queue > rvt_size_atomic(&dev->rdi)) + qp->s_acked_ack_queue = 0; + } + + hfi1_tid_write_alloc_resources(qp, true); + + /* + * If we need to generate more responses, schedule the + * send engine. + */ + if (req->cur_seg < req->total_segs || + qp->s_tail_ack_queue != qp->r_head_ack_queue) { + qp->s_flags |= RVT_S_RESP_PENDING; + hfi1_schedule_send(qp); + } + + priv->pending_tid_w_segs--; + if (priv->s_flags & HFI1_R_TID_RSC_TIMER) { + if (priv->pending_tid_w_segs) + hfi1_mod_tid_reap_timer(req->qp); + else + hfi1_stop_tid_reap_timer(req->qp); + } + +done: + priv->s_flags |= RVT_S_ACK_PENDING; + hfi1_schedule_tid_send(qp); +exit: + priv->r_next_psn_kdeth = flow->flow_state.r_next_psn; + spin_unlock_irqrestore(&qp->s_lock, flags); + return; + +send_nak: + if (!priv->s_nak_state) { + priv->s_nak_state = IB_NAK_PSN_ERROR; + priv->s_nak_psn = flow->flow_state.r_next_psn; + priv->s_flags |= RVT_S_ACK_PENDING; + if (priv->r_tid_ack == HFI1_QP_WQE_INVALID) + priv->r_tid_ack = priv->r_tid_tail; + hfi1_schedule_tid_send(qp); + } + goto done; +} + +static bool hfi1_tid_rdma_is_resync_psn(u32 psn) +{ + return (bool)((psn & HFI1_KDETH_BTH_SEQ_MASK) == + HFI1_KDETH_BTH_SEQ_MASK); +} + +u32 hfi1_build_tid_rdma_write_ack(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_other_headers *ohdr, u16 iflow, + u32 *bth1, u32 *bth2) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_flow_state *fs = &qpriv->flow_state; + struct tid_rdma_request *req = ack_to_tid_req(e); + struct tid_rdma_flow *flow = &req->flows[iflow]; + struct tid_rdma_params *remote; + + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth1, JKEY, remote->jkey); + ohdr->u.tid_rdma.ack.verbs_qp = cpu_to_be32(qp->remote_qpn); + *bth1 = remote->qp; + rcu_read_unlock(); + + if (qpriv->resync) { + *bth2 = mask_psn((fs->generation << + HFI1_KDETH_BTH_SEQ_SHIFT) - 1); + ohdr->u.tid_rdma.ack.aeth = rvt_compute_aeth(qp); + } else if (qpriv->s_nak_state) { + *bth2 = mask_psn(qpriv->s_nak_psn); + ohdr->u.tid_rdma.ack.aeth = + cpu_to_be32((qp->r_msn & IB_MSN_MASK) | + (qpriv->s_nak_state << + IB_AETH_CREDIT_SHIFT)); + } else { + *bth2 = full_flow_psn(flow, flow->flow_state.lpsn); + ohdr->u.tid_rdma.ack.aeth = rvt_compute_aeth(qp); + } + KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth0, KVER, 0x1); + ohdr->u.tid_rdma.ack.tid_flow_qp = + cpu_to_be32(qpriv->tid_rdma.local.qp | + ((flow->idx & TID_RDMA_DESTQP_FLOW_MASK) << + TID_RDMA_DESTQP_FLOW_SHIFT) | + qpriv->rcd->ctxt); + + ohdr->u.tid_rdma.ack.tid_flow_psn = 0; + ohdr->u.tid_rdma.ack.verbs_psn = + cpu_to_be32(flow->flow_state.resp_ib_psn); + + if (qpriv->resync) { + /* + * If the PSN before the current expect KDETH PSN is the + * RESYNC PSN, then we never received a good TID RDMA WRITE + * DATA packet after a previous RESYNC. + * In this case, the next expected KDETH PSN stays the same. + */ + if (hfi1_tid_rdma_is_resync_psn(qpriv->r_next_psn_kdeth - 1)) { + ohdr->u.tid_rdma.ack.tid_flow_psn = + cpu_to_be32(qpriv->r_next_psn_kdeth_save); + } else { + /* + * Because the KDETH PSNs jump during a RESYNC, it's + * not possible to infer (or compute) the previous value + * of r_next_psn_kdeth in the case of back-to-back + * RESYNC packets. Therefore, we save it. + */ + qpriv->r_next_psn_kdeth_save = + qpriv->r_next_psn_kdeth - 1; + ohdr->u.tid_rdma.ack.tid_flow_psn = + cpu_to_be32(qpriv->r_next_psn_kdeth_save); + qpriv->r_next_psn_kdeth = mask_psn(*bth2 + 1); + } + qpriv->resync = false; + } + + return sizeof(ohdr->u.tid_rdma.ack) / sizeof(u32); +} + +void hfi1_rc_rcv_tid_rdma_ack(struct hfi1_packet *packet) +{ + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *qpriv = qp->priv; + struct rvt_swqe *wqe; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + u32 aeth, psn, req_psn, ack_psn, fspsn, resync_psn, ack_kpsn; + bool is_fecn; + unsigned long flags; + u16 fidx; + + trace_hfi1_tid_write_sender_rcv_tid_ack(qp, 0); + is_fecn = process_ecn(qp, packet); + psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + aeth = be32_to_cpu(ohdr->u.tid_rdma.ack.aeth); + req_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.ack.verbs_psn)); + resync_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.ack.tid_flow_psn)); + + spin_lock_irqsave(&qp->s_lock, flags); + trace_hfi1_rcv_tid_ack(qp, aeth, psn, req_psn, resync_psn); + + /* If we are waiting for an ACK to RESYNC, drop any other packets */ + if ((qp->s_flags & HFI1_S_WAIT_HALT) && + cmp_psn(psn, qpriv->s_resync_psn)) + goto ack_op_err; + + ack_psn = req_psn; + if (hfi1_tid_rdma_is_resync_psn(psn)) + ack_kpsn = resync_psn; + else + ack_kpsn = psn; + if (aeth >> 29) { + ack_psn--; + ack_kpsn--; + } + + wqe = rvt_get_swqe_ptr(qp, qp->s_acked); + + if (wqe->wr.opcode != IB_WR_TID_RDMA_WRITE) + goto ack_op_err; + + req = wqe_to_tid_req(wqe); + trace_hfi1_tid_req_rcv_tid_ack(qp, 0, wqe->wr.opcode, wqe->psn, + wqe->lpsn, req); + flow = &req->flows[req->acked_tail]; + trace_hfi1_tid_flow_rcv_tid_ack(qp, req->acked_tail, flow); + + /* Drop stale ACK/NAK */ + if (cmp_psn(psn, full_flow_psn(flow, flow->flow_state.spsn)) < 0) + goto ack_op_err; + + while (cmp_psn(ack_kpsn, + full_flow_psn(flow, flow->flow_state.lpsn)) >= 0 && + req->ack_seg < req->cur_seg) { + req->ack_seg++; + /* advance acked segment pointer */ + req->acked_tail = CIRC_NEXT(req->acked_tail, MAX_FLOWS); + req->r_last_acked = flow->flow_state.resp_ib_psn; + trace_hfi1_tid_req_rcv_tid_ack(qp, 0, wqe->wr.opcode, wqe->psn, + wqe->lpsn, req); + if (req->ack_seg == req->total_segs) { + req->state = TID_REQUEST_COMPLETE; + wqe = do_rc_completion(qp, wqe, + to_iport(qp->ibqp.device, + qp->port_num)); + trace_hfi1_sender_rcv_tid_ack(qp); + atomic_dec(&qpriv->n_tid_requests); + if (qp->s_acked == qp->s_tail) + break; + if (wqe->wr.opcode != IB_WR_TID_RDMA_WRITE) + break; + req = wqe_to_tid_req(wqe); + } + flow = &req->flows[req->acked_tail]; + trace_hfi1_tid_flow_rcv_tid_ack(qp, req->acked_tail, flow); + } + + trace_hfi1_tid_req_rcv_tid_ack(qp, 0, wqe->wr.opcode, wqe->psn, + wqe->lpsn, req); + switch (aeth >> 29) { + case 0: /* ACK */ + if (qpriv->s_flags & RVT_S_WAIT_ACK) + qpriv->s_flags &= ~RVT_S_WAIT_ACK; + if (!hfi1_tid_rdma_is_resync_psn(psn)) { + /* Check if there is any pending TID ACK */ + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE && + req->ack_seg < req->cur_seg) + hfi1_mod_tid_retry_timer(qp); + else + hfi1_stop_tid_retry_timer(qp); + hfi1_schedule_send(qp); + } else { + u32 spsn, fpsn, last_acked, generation; + struct tid_rdma_request *rptr; + + /* ACK(RESYNC) */ + hfi1_stop_tid_retry_timer(qp); + /* Allow new requests (see hfi1_make_tid_rdma_pkt) */ + qp->s_flags &= ~HFI1_S_WAIT_HALT; + /* + * Clear RVT_S_SEND_ONE flag in case that the TID RDMA + * ACK is received after the TID retry timer is fired + * again. In this case, do not send any more TID + * RESYNC request or wait for any more TID ACK packet. + */ + qpriv->s_flags &= ~RVT_S_SEND_ONE; + hfi1_schedule_send(qp); + + if ((qp->s_acked == qpriv->s_tid_tail && + req->ack_seg == req->total_segs) || + qp->s_acked == qp->s_tail) { + qpriv->s_state = TID_OP(WRITE_DATA_LAST); + goto done; + } + + if (req->ack_seg == req->comp_seg) { + qpriv->s_state = TID_OP(WRITE_DATA); + goto done; + } + + /* + * The PSN to start with is the next PSN after the + * RESYNC PSN. + */ + psn = mask_psn(psn + 1); + generation = psn >> HFI1_KDETH_BTH_SEQ_SHIFT; + spsn = 0; + + /* + * Update to the correct WQE when we get an ACK(RESYNC) + * in the middle of a request. + */ + if (delta_psn(ack_psn, wqe->lpsn)) + wqe = rvt_get_swqe_ptr(qp, qp->s_acked); + req = wqe_to_tid_req(wqe); + flow = &req->flows[req->acked_tail]; + /* + * RESYNC re-numbers the PSN ranges of all remaining + * segments. Also, PSN's start from 0 in the middle of a + * segment and the first segment size is less than the + * default number of packets. flow->resync_npkts is used + * to track the number of packets from the start of the + * real segment to the point of 0 PSN after the RESYNC + * in order to later correctly rewind the SGE. + */ + fpsn = full_flow_psn(flow, flow->flow_state.spsn); + req->r_ack_psn = psn; + flow->resync_npkts += + delta_psn(mask_psn(resync_psn + 1), fpsn); + /* + * Renumber all packet sequence number ranges + * based on the new generation. + */ + last_acked = qp->s_acked; + rptr = req; + while (1) { + /* start from last acked segment */ + for (fidx = rptr->acked_tail; + CIRC_CNT(rptr->setup_head, fidx, + MAX_FLOWS); + fidx = CIRC_NEXT(fidx, MAX_FLOWS)) { + u32 lpsn; + u32 gen; + + flow = &rptr->flows[fidx]; + gen = flow->flow_state.generation; + if (WARN_ON(gen == generation && + flow->flow_state.spsn != + spsn)) + continue; + lpsn = flow->flow_state.lpsn; + lpsn = full_flow_psn(flow, lpsn); + flow->npkts = + delta_psn(lpsn, + mask_psn(resync_psn) + ); + flow->flow_state.generation = + generation; + flow->flow_state.spsn = spsn; + flow->flow_state.lpsn = + flow->flow_state.spsn + + flow->npkts - 1; + flow->pkt = 0; + spsn += flow->npkts; + resync_psn += flow->npkts; + trace_hfi1_tid_flow_rcv_tid_ack(qp, + fidx, + flow); + } + if (++last_acked == qpriv->s_tid_cur + 1) + break; + if (last_acked == qp->s_size) + last_acked = 0; + wqe = rvt_get_swqe_ptr(qp, last_acked); + rptr = wqe_to_tid_req(wqe); + } + req->cur_seg = req->ack_seg; + qpriv->s_tid_tail = qp->s_acked; + qpriv->s_state = TID_OP(WRITE_REQ); + hfi1_schedule_tid_send(qp); + } +done: + qpriv->s_retry = qp->s_retry_cnt; + break; + + case 3: /* NAK */ + hfi1_stop_tid_retry_timer(qp); + switch ((aeth >> IB_AETH_CREDIT_SHIFT) & + IB_AETH_CREDIT_MASK) { + case 0: /* PSN sequence error */ + flow = &req->flows[req->acked_tail]; + fspsn = full_flow_psn(flow, flow->flow_state.spsn); + trace_hfi1_tid_flow_rcv_tid_ack(qp, req->acked_tail, + flow); + req->r_ack_psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + req->cur_seg = req->ack_seg; + qpriv->s_tid_tail = qp->s_acked; + qpriv->s_state = TID_OP(WRITE_REQ); + qpriv->s_retry = qp->s_retry_cnt; + hfi1_schedule_tid_send(qp); + break; + + default: + break; + } + break; + + default: + break; + } + +ack_op_err: + spin_unlock_irqrestore(&qp->s_lock, flags); +} + +void hfi1_add_tid_retry_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct ib_qp *ibqp = &qp->ibqp; + struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device); + + lockdep_assert_held(&qp->s_lock); + if (!(priv->s_flags & HFI1_S_TID_RETRY_TIMER)) { + priv->s_flags |= HFI1_S_TID_RETRY_TIMER; + priv->s_tid_retry_timer.expires = jiffies + + priv->tid_retry_timeout_jiffies + rdi->busy_jiffies; + add_timer(&priv->s_tid_retry_timer); + } +} + +static void hfi1_mod_tid_retry_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct ib_qp *ibqp = &qp->ibqp; + struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device); + + lockdep_assert_held(&qp->s_lock); + priv->s_flags |= HFI1_S_TID_RETRY_TIMER; + mod_timer(&priv->s_tid_retry_timer, jiffies + + priv->tid_retry_timeout_jiffies + rdi->busy_jiffies); +} + +static int hfi1_stop_tid_retry_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + int rval = 0; + + lockdep_assert_held(&qp->s_lock); + if (priv->s_flags & HFI1_S_TID_RETRY_TIMER) { + rval = del_timer(&priv->s_tid_retry_timer); + priv->s_flags &= ~HFI1_S_TID_RETRY_TIMER; + } + return rval; +} + +void hfi1_del_tid_retry_timer(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + + del_timer_sync(&priv->s_tid_retry_timer); + priv->s_flags &= ~HFI1_S_TID_RETRY_TIMER; +} + +static void hfi1_tid_retry_timeout(struct timer_list *t) +{ + struct hfi1_qp_priv *priv = from_timer(priv, t, s_tid_retry_timer); + struct rvt_qp *qp = priv->owner; + struct rvt_swqe *wqe; + unsigned long flags; + struct tid_rdma_request *req; + + spin_lock_irqsave(&qp->r_lock, flags); + spin_lock(&qp->s_lock); + trace_hfi1_tid_write_sender_retry_timeout(qp, 0); + if (priv->s_flags & HFI1_S_TID_RETRY_TIMER) { + hfi1_stop_tid_retry_timer(qp); + if (!priv->s_retry) { + trace_hfi1_msg_tid_retry_timeout(/* msg */ + qp, + "Exhausted retries. Tid retry timeout = ", + (u64)priv->tid_retry_timeout_jiffies); + + wqe = rvt_get_swqe_ptr(qp, qp->s_acked); + hfi1_trdma_send_complete(qp, wqe, IB_WC_RETRY_EXC_ERR); + rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR); + } else { + wqe = rvt_get_swqe_ptr(qp, qp->s_acked); + req = wqe_to_tid_req(wqe); + trace_hfi1_tid_req_tid_retry_timeout(/* req */ + qp, 0, wqe->wr.opcode, wqe->psn, wqe->lpsn, req); + + priv->s_flags &= ~RVT_S_WAIT_ACK; + /* Only send one packet (the RESYNC) */ + priv->s_flags |= RVT_S_SEND_ONE; + /* + * No additional request shall be made by this QP until + * the RESYNC has been complete. + */ + qp->s_flags |= HFI1_S_WAIT_HALT; + priv->s_state = TID_OP(RESYNC); + priv->s_retry--; + hfi1_schedule_tid_send(qp); + } + } + spin_unlock(&qp->s_lock); + spin_unlock_irqrestore(&qp->r_lock, flags); +} + +u32 hfi1_build_tid_rdma_resync(struct rvt_qp *qp, struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, u32 *bth1, + u32 *bth2, u16 fidx) +{ + struct hfi1_qp_priv *qpriv = qp->priv; + struct tid_rdma_params *remote; + struct tid_rdma_request *req = wqe_to_tid_req(wqe); + struct tid_rdma_flow *flow = &req->flows[fidx]; + u32 generation; + + rcu_read_lock(); + remote = rcu_dereference(qpriv->tid_rdma.remote); + KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth1, JKEY, remote->jkey); + ohdr->u.tid_rdma.ack.verbs_qp = cpu_to_be32(qp->remote_qpn); + *bth1 = remote->qp; + rcu_read_unlock(); + + generation = kern_flow_generation_next(flow->flow_state.generation); + *bth2 = mask_psn((generation << HFI1_KDETH_BTH_SEQ_SHIFT) - 1); + qpriv->s_resync_psn = *bth2; + *bth2 |= IB_BTH_REQ_ACK; + KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth0, KVER, 0x1); + + return sizeof(ohdr->u.tid_rdma.resync) / sizeof(u32); +} + +void hfi1_rc_rcv_tid_rdma_resync(struct hfi1_packet *packet) +{ + struct ib_other_headers *ohdr = packet->ohdr; + struct rvt_qp *qp = packet->qp; + struct hfi1_qp_priv *qpriv = qp->priv; + struct hfi1_ctxtdata *rcd = qpriv->rcd; + struct hfi1_ibdev *dev = to_idev(qp->ibqp.device); + struct rvt_ack_entry *e; + struct tid_rdma_request *req; + struct tid_rdma_flow *flow; + struct tid_flow_state *fs = &qpriv->flow_state; + u32 psn, generation, idx, gen_next; + bool is_fecn; + unsigned long flags; + + is_fecn = process_ecn(qp, packet); + psn = mask_psn(be32_to_cpu(ohdr->bth[2])); + + generation = mask_psn(psn + 1) >> HFI1_KDETH_BTH_SEQ_SHIFT; + spin_lock_irqsave(&qp->s_lock, flags); + + gen_next = (fs->generation == KERN_GENERATION_RESERVED) ? + generation : kern_flow_generation_next(fs->generation); + /* + * RESYNC packet contains the "next" generation and can only be + * from the current or previous generations + */ + if (generation != mask_generation(gen_next - 1) && + generation != gen_next) + goto bail; + /* Already processing a resync */ + if (qpriv->resync) + goto bail; + + spin_lock(&rcd->exp_lock); + if (fs->index >= RXE_NUM_TID_FLOWS) { + /* + * If we don't have a flow, save the generation so it can be + * applied when a new flow is allocated + */ + fs->generation = generation; + } else { + /* Reprogram the QP flow with new generation */ + rcd->flows[fs->index].generation = generation; + fs->generation = kern_setup_hw_flow(rcd, fs->index); + } + fs->psn = 0; + /* + * Disable SW PSN checking since a RESYNC is equivalent to a + * sync point and the flow has/will be reprogrammed + */ + qpriv->s_flags &= ~HFI1_R_TID_SW_PSN; + trace_hfi1_tid_write_rsp_rcv_resync(qp); + + /* + * Reset all TID flow information with the new generation. + * This is done for all requests and segments after the + * last received segment + */ + for (idx = qpriv->r_tid_tail; ; idx++) { + u16 flow_idx; + + if (idx > rvt_size_atomic(&dev->rdi)) + idx = 0; + e = &qp->s_ack_queue[idx]; + if (e->opcode == TID_OP(WRITE_REQ)) { + req = ack_to_tid_req(e); + trace_hfi1_tid_req_rcv_resync(qp, 0, e->opcode, e->psn, + e->lpsn, req); + + /* start from last unacked segment */ + for (flow_idx = req->clear_tail; + CIRC_CNT(req->setup_head, flow_idx, + MAX_FLOWS); + flow_idx = CIRC_NEXT(flow_idx, MAX_FLOWS)) { + u32 lpsn; + u32 next; + + flow = &req->flows[flow_idx]; + lpsn = full_flow_psn(flow, + flow->flow_state.lpsn); + next = flow->flow_state.r_next_psn; + flow->npkts = delta_psn(lpsn, next - 1); + flow->flow_state.generation = fs->generation; + flow->flow_state.spsn = fs->psn; + flow->flow_state.lpsn = + flow->flow_state.spsn + flow->npkts - 1; + flow->flow_state.r_next_psn = + full_flow_psn(flow, + flow->flow_state.spsn); + fs->psn += flow->npkts; + trace_hfi1_tid_flow_rcv_resync(qp, flow_idx, + flow); + } + } + if (idx == qp->s_tail_ack_queue) + break; + } + + spin_unlock(&rcd->exp_lock); + qpriv->resync = true; + /* RESYNC request always gets a TID RDMA ACK. */ + qpriv->s_nak_state = 0; + qpriv->s_flags |= RVT_S_ACK_PENDING; + hfi1_schedule_tid_send(qp); +bail: + spin_unlock_irqrestore(&qp->s_lock, flags); +} + +/* + * Call this function when the last TID RDMA WRITE DATA packet for a request + * is built. + */ +static void update_tid_tail(struct rvt_qp *qp) + __must_hold(&qp->s_lock) +{ + struct hfi1_qp_priv *priv = qp->priv; + u32 i; + struct rvt_swqe *wqe; + + lockdep_assert_held(&qp->s_lock); + /* Can't move beyond s_tid_cur */ + if (priv->s_tid_tail == priv->s_tid_cur) + return; + for (i = priv->s_tid_tail + 1; ; i++) { + if (i == qp->s_size) + i = 0; + + if (i == priv->s_tid_cur) + break; + wqe = rvt_get_swqe_ptr(qp, i); + if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) + break; + } + priv->s_tid_tail = i; + priv->s_state = TID_OP(WRITE_RESP); +} + +int hfi1_make_tid_rdma_pkt(struct rvt_qp *qp, struct hfi1_pkt_state *ps) + __must_hold(&qp->s_lock) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct rvt_swqe *wqe; + u32 bth1 = 0, bth2 = 0, hwords = 5, len, middle = 0; + struct ib_other_headers *ohdr; + struct rvt_sge_state *ss = &qp->s_sge; + struct rvt_ack_entry *e = &qp->s_ack_queue[qp->s_tail_ack_queue]; + struct tid_rdma_request *req = ack_to_tid_req(e); + bool last = false; + u8 opcode = TID_OP(WRITE_DATA); + + lockdep_assert_held(&qp->s_lock); + trace_hfi1_tid_write_sender_make_tid_pkt(qp, 0); + /* + * Prioritize the sending of the requests and responses over the + * sending of the TID RDMA data packets. + */ + if (((atomic_read(&priv->n_tid_requests) < HFI1_TID_RDMA_WRITE_CNT) && + atomic_read(&priv->n_requests) && + !(qp->s_flags & (RVT_S_BUSY | RVT_S_WAIT_ACK | + HFI1_S_ANY_WAIT_IO))) || + (e->opcode == TID_OP(WRITE_REQ) && req->cur_seg < req->alloc_seg && + !(qp->s_flags & (RVT_S_BUSY | HFI1_S_ANY_WAIT_IO)))) { + struct iowait_work *iowork; + + iowork = iowait_get_ib_work(&priv->s_iowait); + ps->s_txreq = get_waiting_verbs_txreq(iowork); + if (ps->s_txreq || hfi1_make_rc_req(qp, ps)) { + priv->s_flags |= HFI1_S_TID_BUSY_SET; + return 1; + } + } + + ps->s_txreq = get_txreq(ps->dev, qp); + if (!ps->s_txreq) + goto bail_no_tx; + + ohdr = &ps->s_txreq->phdr.hdr.ibh.u.oth; + + if ((priv->s_flags & RVT_S_ACK_PENDING) && + make_tid_rdma_ack(qp, ohdr, ps)) + return 1; + + if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_SEND_OK)) { + if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND)) + goto bail; + /* We are in the error state, flush the work request. */ + if (qp->s_last == READ_ONCE(qp->s_head)) + goto bail; + /* If DMAs are in progress, we can't flush immediately. */ + if (iowait_sdma_pending(&priv->s_iowait)) { + qp->s_flags |= RVT_S_WAIT_DMA; + goto bail; + } + clear_ahg(qp); + wqe = rvt_get_swqe_ptr(qp, qp->s_last); + hfi1_trdma_send_complete(qp, wqe, qp->s_last != qp->s_acked ? + IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR); + /* will get called again */ + goto done_free_tx; + } + + if (priv->s_flags & RVT_S_WAIT_ACK) + goto bail; + + /* Check whether there is anything to do. */ + if (priv->s_tid_tail == HFI1_QP_WQE_INVALID) + goto bail; + wqe = rvt_get_swqe_ptr(qp, priv->s_tid_tail); + req = wqe_to_tid_req(wqe); + trace_hfi1_tid_req_make_tid_pkt(qp, 0, wqe->wr.opcode, wqe->psn, + wqe->lpsn, req); + switch (priv->s_state) { + case TID_OP(WRITE_REQ): + case TID_OP(WRITE_RESP): + priv->tid_ss.sge = wqe->sg_list[0]; + priv->tid_ss.sg_list = wqe->sg_list + 1; + priv->tid_ss.num_sge = wqe->wr.num_sge; + priv->tid_ss.total_len = wqe->length; + + if (priv->s_state == TID_OP(WRITE_REQ)) + hfi1_tid_rdma_restart_req(qp, wqe, &bth2); + priv->s_state = TID_OP(WRITE_DATA); + /* fall through */ + + case TID_OP(WRITE_DATA): + /* + * 1. Check whether TID RDMA WRITE RESP available. + * 2. If no: + * 2.1 If have more segments and no TID RDMA WRITE RESP, + * set HFI1_S_WAIT_TID_RESP + * 2.2 Return indicating no progress made. + * 3. If yes: + * 3.1 Build TID RDMA WRITE DATA packet. + * 3.2 If last packet in segment: + * 3.2.1 Change KDETH header bits + * 3.2.2 Advance RESP pointers. + * 3.3 Return indicating progress made. + */ + trace_hfi1_sender_make_tid_pkt(qp); + trace_hfi1_tid_write_sender_make_tid_pkt(qp, 0); + wqe = rvt_get_swqe_ptr(qp, priv->s_tid_tail); + req = wqe_to_tid_req(wqe); + len = wqe->length; + + if (!req->comp_seg || req->cur_seg == req->comp_seg) + goto bail; + + trace_hfi1_tid_req_make_tid_pkt(qp, 0, wqe->wr.opcode, + wqe->psn, wqe->lpsn, req); + last = hfi1_build_tid_rdma_packet(wqe, ohdr, &bth1, &bth2, + &len); + + if (last) { + /* move pointer to next flow */ + req->clear_tail = CIRC_NEXT(req->clear_tail, + MAX_FLOWS); + if (++req->cur_seg < req->total_segs) { + if (!CIRC_CNT(req->setup_head, req->clear_tail, + MAX_FLOWS)) + qp->s_flags |= HFI1_S_WAIT_TID_RESP; + } else { + priv->s_state = TID_OP(WRITE_DATA_LAST); + opcode = TID_OP(WRITE_DATA_LAST); + + /* Advance the s_tid_tail now */ + update_tid_tail(qp); + } + } + hwords += sizeof(ohdr->u.tid_rdma.w_data) / sizeof(u32); + ss = &priv->tid_ss; + break; + + case TID_OP(RESYNC): + trace_hfi1_sender_make_tid_pkt(qp); + /* Use generation from the most recently received response */ + wqe = rvt_get_swqe_ptr(qp, priv->s_tid_cur); + req = wqe_to_tid_req(wqe); + /* If no responses for this WQE look at the previous one */ + if (!req->comp_seg) { + wqe = rvt_get_swqe_ptr(qp, + (!priv->s_tid_cur ? qp->s_size : + priv->s_tid_cur) - 1); + req = wqe_to_tid_req(wqe); + } + hwords += hfi1_build_tid_rdma_resync(qp, wqe, ohdr, &bth1, + &bth2, + CIRC_PREV(req->setup_head, + MAX_FLOWS)); + ss = NULL; + len = 0; + opcode = TID_OP(RESYNC); + break; + + default: + goto bail; + } + if (priv->s_flags & RVT_S_SEND_ONE) { + priv->s_flags &= ~RVT_S_SEND_ONE; + priv->s_flags |= RVT_S_WAIT_ACK; + bth2 |= IB_BTH_REQ_ACK; + } + qp->s_len -= len; + ps->s_txreq->hdr_dwords = hwords; + ps->s_txreq->sde = priv->s_sde; + ps->s_txreq->ss = ss; + ps->s_txreq->s_cur_size = len; + hfi1_make_ruc_header(qp, ohdr, (opcode << 24), bth1, bth2, + middle, ps); + return 1; +done_free_tx: + hfi1_put_txreq(ps->s_txreq); + ps->s_txreq = NULL; + return 1; + +bail: + hfi1_put_txreq(ps->s_txreq); +bail_no_tx: + ps->s_txreq = NULL; + priv->s_flags &= ~RVT_S_BUSY; + /* + * If we didn't get a txreq, the QP will be woken up later to try + * again, set the flags to the the wake up which work item to wake + * up. + * (A better algorithm should be found to do this and generalize the + * sleep/wakeup flags.) + */ + iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_TID); + return 0; +} + +static int make_tid_rdma_ack(struct rvt_qp *qp, + struct ib_other_headers *ohdr, + struct hfi1_pkt_state *ps) +{ + struct rvt_ack_entry *e; + struct hfi1_qp_priv *qpriv = qp->priv; + struct hfi1_ibdev *dev = to_idev(qp->ibqp.device); + u32 hwords, next; + u32 len = 0; + u32 bth1 = 0, bth2 = 0; + int middle = 0; + u16 flow; + struct tid_rdma_request *req, *nreq; + + trace_hfi1_tid_write_rsp_make_tid_ack(qp); + /* Don't send an ACK if we aren't supposed to. */ + if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) + goto bail; + + /* header size in 32-bit words LRH+BTH = (8+12)/4. */ + hwords = 5; + + e = &qp->s_ack_queue[qpriv->r_tid_ack]; + req = ack_to_tid_req(e); + /* + * In the RESYNC case, we are exactly one segment past the + * previously sent ack or at the previously sent NAK. So to send + * the resync ack, we go back one segment (which might be part of + * the previous request) and let the do-while loop execute again. + * The advantage of executing the do-while loop is that any data + * received after the previous ack is automatically acked in the + * RESYNC ack. It turns out that for the do-while loop we only need + * to pull back qpriv->r_tid_ack, not the segment + * indices/counters. The scheme works even if the previous request + * was not a TID WRITE request. + */ + if (qpriv->resync) { + if (!req->ack_seg || req->ack_seg == req->total_segs) + qpriv->r_tid_ack = !qpriv->r_tid_ack ? + rvt_size_atomic(&dev->rdi) : + qpriv->r_tid_ack - 1; + e = &qp->s_ack_queue[qpriv->r_tid_ack]; + req = ack_to_tid_req(e); + } + + trace_hfi1_rsp_make_tid_ack(qp, e->psn); + trace_hfi1_tid_req_make_tid_ack(qp, 0, e->opcode, e->psn, e->lpsn, + req); + /* + * If we've sent all the ACKs that we can, we are done + * until we get more segments... + */ + if (!qpriv->s_nak_state && !qpriv->resync && + req->ack_seg == req->comp_seg) + goto bail; + + do { + /* + * To deal with coalesced ACKs, the acked_tail pointer + * into the flow array is used. The distance between it + * and the clear_tail is the number of flows that are + * being ACK'ed. + */ + req->ack_seg += + /* Get up-to-date value */ + CIRC_CNT(req->clear_tail, req->acked_tail, + MAX_FLOWS); + /* Advance acked index */ + req->acked_tail = req->clear_tail; + + /* + * req->clear_tail points to the segment currently being + * received. So, when sending an ACK, the previous + * segment is being ACK'ed. + */ + flow = CIRC_PREV(req->acked_tail, MAX_FLOWS); + if (req->ack_seg != req->total_segs) + break; + req->state = TID_REQUEST_COMPLETE; + + next = qpriv->r_tid_ack + 1; + if (next > rvt_size_atomic(&dev->rdi)) + next = 0; + qpriv->r_tid_ack = next; + if (qp->s_ack_queue[next].opcode != TID_OP(WRITE_REQ)) + break; + nreq = ack_to_tid_req(&qp->s_ack_queue[next]); + if (!nreq->comp_seg || nreq->ack_seg == nreq->comp_seg) + break; + + /* Move to the next ack entry now */ + e = &qp->s_ack_queue[qpriv->r_tid_ack]; + req = ack_to_tid_req(e); + } while (1); + + /* + * At this point qpriv->r_tid_ack == qpriv->r_tid_tail but e and + * req could be pointing at the previous ack queue entry + */ + if (qpriv->s_nak_state || + (qpriv->resync && + !hfi1_tid_rdma_is_resync_psn(qpriv->r_next_psn_kdeth - 1) && + (cmp_psn(qpriv->r_next_psn_kdeth - 1, + full_flow_psn(&req->flows[flow], + req->flows[flow].flow_state.lpsn)) > 0))) { + /* + * A NAK will implicitly acknowledge all previous TID RDMA + * requests. Therefore, we NAK with the req->acked_tail + * segment for the request at qpriv->r_tid_ack (same at + * this point as the req->clear_tail segment for the + * qpriv->r_tid_tail request) + */ + e = &qp->s_ack_queue[qpriv->r_tid_ack]; + req = ack_to_tid_req(e); + flow = req->acked_tail; + } else if (req->ack_seg == req->total_segs && + qpriv->s_flags & HFI1_R_TID_WAIT_INTERLCK) + qpriv->s_flags &= ~HFI1_R_TID_WAIT_INTERLCK; + + trace_hfi1_tid_write_rsp_make_tid_ack(qp); + trace_hfi1_tid_req_make_tid_ack(qp, 0, e->opcode, e->psn, e->lpsn, + req); + hwords += hfi1_build_tid_rdma_write_ack(qp, e, ohdr, flow, &bth1, + &bth2); + len = 0; + qpriv->s_flags &= ~RVT_S_ACK_PENDING; + ps->s_txreq->hdr_dwords = hwords; + ps->s_txreq->sde = qpriv->s_sde; + ps->s_txreq->s_cur_size = len; + ps->s_txreq->ss = NULL; + hfi1_make_ruc_header(qp, ohdr, (TID_OP(ACK) << 24), bth1, bth2, middle, + ps); + ps->s_txreq->txreq.flags |= SDMA_TXREQ_F_VIP; + return 1; +bail: + /* + * Ensure s_rdma_ack_cnt changes are committed prior to resetting + * RVT_S_RESP_PENDING + */ + smp_wmb(); + qpriv->s_flags &= ~RVT_S_ACK_PENDING; + return 0; +} + +static int hfi1_send_tid_ok(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + + return !(priv->s_flags & RVT_S_BUSY || + qp->s_flags & HFI1_S_ANY_WAIT_IO) && + (verbs_txreq_queued(iowait_get_tid_work(&priv->s_iowait)) || + (priv->s_flags & RVT_S_RESP_PENDING) || + !(qp->s_flags & HFI1_S_ANY_TID_WAIT_SEND)); +} + +void _hfi1_do_tid_send(struct work_struct *work) +{ + struct iowait_work *w = container_of(work, struct iowait_work, iowork); + struct rvt_qp *qp = iowait_to_qp(w->iow); + + hfi1_do_tid_send(qp); +} + +static void hfi1_do_tid_send(struct rvt_qp *qp) +{ + struct hfi1_pkt_state ps; + struct hfi1_qp_priv *priv = qp->priv; + + ps.dev = to_idev(qp->ibqp.device); + ps.ibp = to_iport(qp->ibqp.device, qp->port_num); + ps.ppd = ppd_from_ibp(ps.ibp); + ps.wait = iowait_get_tid_work(&priv->s_iowait); + ps.in_thread = false; + ps.timeout_int = qp->timeout_jiffies / 8; + + trace_hfi1_rc_do_tid_send(qp, false); + spin_lock_irqsave(&qp->s_lock, ps.flags); + + /* Return if we are already busy processing a work request. */ + if (!hfi1_send_tid_ok(qp)) { + if (qp->s_flags & HFI1_S_ANY_WAIT_IO) + iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_TID); + spin_unlock_irqrestore(&qp->s_lock, ps.flags); + return; + } + + priv->s_flags |= RVT_S_BUSY; + + ps.timeout = jiffies + ps.timeout_int; + ps.cpu = priv->s_sde ? priv->s_sde->cpu : + cpumask_first(cpumask_of_node(ps.ppd->dd->node)); + ps.pkts_sent = false; + + /* insure a pre-built packet is handled */ + ps.s_txreq = get_waiting_verbs_txreq(ps.wait); + do { + /* Check for a constructed packet to be sent. */ + if (ps.s_txreq) { + if (priv->s_flags & HFI1_S_TID_BUSY_SET) { + qp->s_flags |= RVT_S_BUSY; + ps.wait = iowait_get_ib_work(&priv->s_iowait); + } + spin_unlock_irqrestore(&qp->s_lock, ps.flags); + + /* + * If the packet cannot be sent now, return and + * the send tasklet will be woken up later. + */ + if (hfi1_verbs_send(qp, &ps)) + return; + + /* allow other tasks to run */ + if (hfi1_schedule_send_yield(qp, &ps, true)) + return; + + spin_lock_irqsave(&qp->s_lock, ps.flags); + if (priv->s_flags & HFI1_S_TID_BUSY_SET) { + qp->s_flags &= ~RVT_S_BUSY; + priv->s_flags &= ~HFI1_S_TID_BUSY_SET; + ps.wait = iowait_get_tid_work(&priv->s_iowait); + if (iowait_flag_set(&priv->s_iowait, + IOWAIT_PENDING_IB)) + hfi1_schedule_send(qp); + } + } + } while (hfi1_make_tid_rdma_pkt(qp, &ps)); + iowait_starve_clear(ps.pkts_sent, &priv->s_iowait); + spin_unlock_irqrestore(&qp->s_lock, ps.flags); +} + +static bool _hfi1_schedule_tid_send(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct hfi1_ibport *ibp = + to_iport(qp->ibqp.device, qp->port_num); + struct hfi1_pportdata *ppd = ppd_from_ibp(ibp); + struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device); + + return iowait_tid_schedule(&priv->s_iowait, ppd->hfi1_wq, + priv->s_sde ? + priv->s_sde->cpu : + cpumask_first(cpumask_of_node(dd->node))); +} + +/** + * hfi1_schedule_tid_send - schedule progress on TID RDMA state machine + * @qp: the QP + * + * This schedules qp progress on the TID RDMA state machine. Caller + * should hold the s_lock. + * Unlike hfi1_schedule_send(), this cannot use hfi1_send_ok() because + * the two state machines can step on each other with respect to the + * RVT_S_BUSY flag. + * Therefore, a modified test is used. + * @return true if the second leg is scheduled; + * false if the second leg is not scheduled. + */ +bool hfi1_schedule_tid_send(struct rvt_qp *qp) +{ + lockdep_assert_held(&qp->s_lock); + if (hfi1_send_tid_ok(qp)) { + /* + * The following call returns true if the qp is not on the + * queue and false if the qp is already on the queue before + * this call. Either way, the qp will be on the queue when the + * call returns. + */ + _hfi1_schedule_tid_send(qp); + return true; + } + if (qp->s_flags & HFI1_S_ANY_WAIT_IO) + iowait_set_flag(&((struct hfi1_qp_priv *)qp->priv)->s_iowait, + IOWAIT_PENDING_TID); + return false; +} + +bool hfi1_tid_rdma_ack_interlock(struct rvt_qp *qp, struct rvt_ack_entry *e) +{ + struct rvt_ack_entry *prev; + struct tid_rdma_request *req; + struct hfi1_ibdev *dev = to_idev(qp->ibqp.device); + struct hfi1_qp_priv *priv = qp->priv; + u32 s_prev; + + s_prev = qp->s_tail_ack_queue == 0 ? rvt_size_atomic(&dev->rdi) : + (qp->s_tail_ack_queue - 1); + prev = &qp->s_ack_queue[s_prev]; + + if ((e->opcode == TID_OP(READ_REQ) || + e->opcode == OP(RDMA_READ_REQUEST)) && + prev->opcode == TID_OP(WRITE_REQ)) { + req = ack_to_tid_req(prev); + if (req->ack_seg != req->total_segs) { + priv->s_flags |= HFI1_R_TID_WAIT_INTERLCK; + return true; + } + } + return false; +} diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.h b/drivers/infiniband/hw/hfi1/tid_rdma.h index 6fcd3adcdcc3..53ab24ef4f02 100644 --- a/drivers/infiniband/hw/hfi1/tid_rdma.h +++ b/drivers/infiniband/hw/hfi1/tid_rdma.h @@ -6,8 +6,317 @@ #ifndef HFI1_TID_RDMA_H #define HFI1_TID_RDMA_H +#include <linux/circ_buf.h> +#include "common.h" + +/* Add a convenience helper */ +#define CIRC_ADD(val, add, size) (((val) + (add)) & ((size) - 1)) +#define CIRC_NEXT(val, size) CIRC_ADD(val, 1, size) +#define CIRC_PREV(val, size) CIRC_ADD(val, -1, size) + +#define TID_RDMA_MIN_SEGMENT_SIZE BIT(18) /* 256 KiB (for now) */ +#define TID_RDMA_MAX_SEGMENT_SIZE BIT(18) /* 256 KiB (for now) */ +#define TID_RDMA_MAX_PAGES (BIT(18) >> PAGE_SHIFT) + +/* + * Bit definitions for priv->s_flags. + * These bit flags overload the bit flags defined for the QP's s_flags. + * Due to the fact that these bit fields are used only for the QP priv + * s_flags, there are no collisions. + * + * HFI1_S_TID_WAIT_INTERLCK - QP is waiting for requester interlock + * HFI1_R_TID_WAIT_INTERLCK - QP is waiting for responder interlock + */ +#define HFI1_S_TID_BUSY_SET BIT(0) +/* BIT(1) reserved for RVT_S_BUSY. */ +#define HFI1_R_TID_RSC_TIMER BIT(2) +/* BIT(3) reserved for RVT_S_RESP_PENDING. */ +/* BIT(4) reserved for RVT_S_ACK_PENDING. */ +#define HFI1_S_TID_WAIT_INTERLCK BIT(5) +#define HFI1_R_TID_WAIT_INTERLCK BIT(6) +/* BIT(7) - BIT(15) reserved for RVT_S_WAIT_*. */ +/* BIT(16) reserved for RVT_S_SEND_ONE */ +#define HFI1_S_TID_RETRY_TIMER BIT(17) +/* BIT(18) reserved for RVT_S_ECN. */ +#define HFI1_R_TID_SW_PSN BIT(19) +/* BIT(26) reserved for HFI1_S_WAIT_HALT */ +/* BIT(27) reserved for HFI1_S_WAIT_TID_RESP */ +/* BIT(28) reserved for HFI1_S_WAIT_TID_SPACE */ + +/* + * Unlike regular IB RDMA VERBS, which do not require an entry + * in the s_ack_queue, TID RDMA WRITE requests do because they + * generate responses. + * Therefore, the s_ack_queue needs to be extended by a certain + * amount. The key point is that the queue needs to be extended + * without letting the "user" know so they user doesn't end up + * using these extra entries. + */ +#define HFI1_TID_RDMA_WRITE_CNT 8 + +struct tid_rdma_params { + struct rcu_head rcu_head; + u32 qp; + u32 max_len; + u16 jkey; + u8 max_read; + u8 max_write; + u8 timeout; + u8 urg; + u8 version; +}; + +struct tid_rdma_qp_params { + struct work_struct trigger_work; + struct tid_rdma_params local; + struct tid_rdma_params __rcu *remote; +}; + +/* Track state for each hardware flow */ +struct tid_flow_state { + u32 generation; + u32 psn; + u32 r_next_psn; /* next PSN to be received (in TID space) */ + u8 index; + u8 last_index; + u8 flags; +}; + +enum tid_rdma_req_state { + TID_REQUEST_INACTIVE = 0, + TID_REQUEST_INIT, + TID_REQUEST_INIT_RESEND, + TID_REQUEST_ACTIVE, + TID_REQUEST_RESEND, + TID_REQUEST_RESEND_ACTIVE, + TID_REQUEST_QUEUED, + TID_REQUEST_SYNC, + TID_REQUEST_RNR_NAK, + TID_REQUEST_COMPLETE, +}; + +struct tid_rdma_request { + struct rvt_qp *qp; + struct hfi1_ctxtdata *rcd; + union { + struct rvt_swqe *swqe; + struct rvt_ack_entry *ack; + } e; + + struct tid_rdma_flow *flows; /* array of tid flows */ + struct rvt_sge_state ss; /* SGE state for TID RDMA requests */ + u16 n_flows; /* size of the flow buffer window */ + u16 setup_head; /* flow index we are setting up */ + u16 clear_tail; /* flow index we are clearing */ + u16 flow_idx; /* flow index most recently set up */ + u16 acked_tail; + + u32 seg_len; + u32 total_len; + u32 r_ack_psn; /* next expected ack PSN */ + u32 r_flow_psn; /* IB PSN of next segment start */ + u32 r_last_acked; /* IB PSN of last ACK'ed packet */ + u32 s_next_psn; /* IB PSN of next segment start for read */ + + u32 total_segs; /* segments required to complete a request */ + u32 cur_seg; /* index of current segment */ + u32 comp_seg; /* index of last completed segment */ + u32 ack_seg; /* index of last ack'ed segment */ + u32 alloc_seg; /* index of next segment to be allocated */ + u32 isge; /* index of "current" sge */ + u32 ack_pending; /* num acks pending for this request */ + + enum tid_rdma_req_state state; +}; + +/* + * When header suppression is used, PSNs associated with a "flow" are + * relevant (and not the PSNs maintained by verbs). Track per-flow + * PSNs here for a TID RDMA segment. + * + */ +struct flow_state { + u32 flags; + u32 resp_ib_psn; /* The IB PSN of the response for this flow */ + u32 generation; /* generation of flow */ + u32 spsn; /* starting PSN in TID space */ + u32 lpsn; /* last PSN in TID space */ + u32 r_next_psn; /* next PSN to be received (in TID space) */ + + /* For tid rdma read */ + u32 ib_spsn; /* starting PSN in Verbs space */ + u32 ib_lpsn; /* last PSn in Verbs space */ +}; + +struct tid_rdma_pageset { + dma_addr_t addr : 48; /* Only needed for the first page */ + u8 idx: 8; + u8 count : 7; + u8 mapped: 1; +}; + +/** + * kern_tid_node - used for managing TID's in TID groups + * + * @grp_idx: rcd relative index to tid_group + * @map: grp->map captured prior to programming this TID group in HW + * @cnt: Only @cnt of available group entries are actually programmed + */ +struct kern_tid_node { + struct tid_group *grp; + u8 map; + u8 cnt; +}; + +/* Overall info for a TID RDMA segment */ +struct tid_rdma_flow { + /* + * While a TID RDMA segment is being transferred, it uses a QP number + * from the "KDETH section of QP numbers" (which is different from the + * QP number that originated the request). Bits 11-15 of these QP + * numbers identify the "TID flow" for the segment. + */ + struct flow_state flow_state; + struct tid_rdma_request *req; + u32 tid_qpn; + u32 tid_offset; + u32 length; + u32 sent; + u8 tnode_cnt; + u8 tidcnt; + u8 tid_idx; + u8 idx; + u8 npagesets; + u8 npkts; + u8 pkt; + u8 resync_npkts; + struct kern_tid_node tnode[TID_RDMA_MAX_PAGES]; + struct tid_rdma_pageset pagesets[TID_RDMA_MAX_PAGES]; + u32 tid_entry[TID_RDMA_MAX_PAGES]; +}; + +enum tid_rnr_nak_state { + TID_RNR_NAK_INIT = 0, + TID_RNR_NAK_SEND, + TID_RNR_NAK_SENT, +}; + +bool tid_rdma_conn_req(struct rvt_qp *qp, u64 *data); +bool tid_rdma_conn_reply(struct rvt_qp *qp, u64 data); +bool tid_rdma_conn_resp(struct rvt_qp *qp, u64 *data); +void tid_rdma_conn_error(struct rvt_qp *qp); +void tid_rdma_opfn_init(struct rvt_qp *qp, struct tid_rdma_params *p); + +int hfi1_kern_exp_rcv_init(struct hfi1_ctxtdata *rcd, int reinit); +int hfi1_kern_exp_rcv_setup(struct tid_rdma_request *req, + struct rvt_sge_state *ss, bool *last); +int hfi1_kern_exp_rcv_clear(struct tid_rdma_request *req); +void hfi1_kern_exp_rcv_clear_all(struct tid_rdma_request *req); +void __trdma_clean_swqe(struct rvt_qp *qp, struct rvt_swqe *wqe); + +/** + * trdma_clean_swqe - clean flows for swqe if large send queue + * @qp: the qp + * @wqe: the send wqe + */ +static inline void trdma_clean_swqe(struct rvt_qp *qp, struct rvt_swqe *wqe) +{ + if (!wqe->priv) + return; + __trdma_clean_swqe(qp, wqe); +} + +void hfi1_kern_read_tid_flow_free(struct rvt_qp *qp); + int hfi1_qp_priv_init(struct rvt_dev_info *rdi, struct rvt_qp *qp, struct ib_qp_init_attr *init_attr); +void hfi1_qp_priv_tid_free(struct rvt_dev_info *rdi, struct rvt_qp *qp); -#endif /* HFI1_TID_RDMA_H */ +void hfi1_tid_rdma_flush_wait(struct rvt_qp *qp); + +int hfi1_kern_setup_hw_flow(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp); +void hfi1_kern_clear_hw_flow(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp); +void hfi1_kern_init_ctxt_generations(struct hfi1_ctxtdata *rcd); + +struct cntr_entry; +u64 hfi1_access_sw_tid_wait(const struct cntr_entry *entry, + void *context, int vl, int mode, u64 data); + +u32 hfi1_build_tid_rdma_read_packet(struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, + u32 *bth1, u32 *bth2, u32 *len); +u32 hfi1_build_tid_rdma_read_req(struct rvt_qp *qp, struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, u32 *bth1, + u32 *bth2, u32 *len); +void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet); +u32 hfi1_build_tid_rdma_read_resp(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_other_headers *ohdr, u32 *bth0, + u32 *bth1, u32 *bth2, u32 *len, bool *last); +void hfi1_rc_rcv_tid_rdma_read_resp(struct hfi1_packet *packet); +bool hfi1_handle_kdeth_eflags(struct hfi1_ctxtdata *rcd, + struct hfi1_pportdata *ppd, + struct hfi1_packet *packet); +void hfi1_tid_rdma_restart_req(struct rvt_qp *qp, struct rvt_swqe *wqe, + u32 *bth2); +void hfi1_qp_kern_exp_rcv_clear_all(struct rvt_qp *qp); +bool hfi1_tid_rdma_wqe_interlock(struct rvt_qp *qp, struct rvt_swqe *wqe); + +void setup_tid_rdma_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe); +static inline void hfi1_setup_tid_rdma_wqe(struct rvt_qp *qp, + struct rvt_swqe *wqe) +{ + if (wqe->priv && + (wqe->wr.opcode == IB_WR_RDMA_READ || + wqe->wr.opcode == IB_WR_RDMA_WRITE) && + wqe->length >= TID_RDMA_MIN_SEGMENT_SIZE) + setup_tid_rdma_wqe(qp, wqe); +} + +u32 hfi1_build_tid_rdma_write_req(struct rvt_qp *qp, struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, + u32 *bth1, u32 *bth2, u32 *len); + +void hfi1_compute_tid_rdma_flow_wt(void); + +void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet); + +u32 hfi1_build_tid_rdma_write_resp(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_other_headers *ohdr, u32 *bth1, + u32 bth2, u32 *len, + struct rvt_sge_state **ss); +void hfi1_del_tid_reap_timer(struct rvt_qp *qp); + +void hfi1_rc_rcv_tid_rdma_write_resp(struct hfi1_packet *packet); + +bool hfi1_build_tid_rdma_packet(struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, + u32 *bth1, u32 *bth2, u32 *len); + +void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet); + +u32 hfi1_build_tid_rdma_write_ack(struct rvt_qp *qp, struct rvt_ack_entry *e, + struct ib_other_headers *ohdr, u16 iflow, + u32 *bth1, u32 *bth2); + +void hfi1_rc_rcv_tid_rdma_ack(struct hfi1_packet *packet); + +void hfi1_add_tid_retry_timer(struct rvt_qp *qp); +void hfi1_del_tid_retry_timer(struct rvt_qp *qp); + +u32 hfi1_build_tid_rdma_resync(struct rvt_qp *qp, struct rvt_swqe *wqe, + struct ib_other_headers *ohdr, u32 *bth1, + u32 *bth2, u16 fidx); + +void hfi1_rc_rcv_tid_rdma_resync(struct hfi1_packet *packet); + +struct hfi1_pkt_state; +int hfi1_make_tid_rdma_pkt(struct rvt_qp *qp, struct hfi1_pkt_state *ps); + +void _hfi1_do_tid_send(struct work_struct *work); + +bool hfi1_schedule_tid_send(struct rvt_qp *qp); + +bool hfi1_tid_rdma_ack_interlock(struct rvt_qp *qp, struct rvt_ack_entry *e); + +#endif /* HFI1_TID_RDMA_H */ diff --git a/drivers/infiniband/hw/hfi1/trace.c b/drivers/infiniband/hw/hfi1/trace.c index 7c8aed0ffc07..9a3d236bcc88 100644 --- a/drivers/infiniband/hw/hfi1/trace.c +++ b/drivers/infiniband/hw/hfi1/trace.c @@ -46,6 +46,7 @@ */ #define CREATE_TRACE_POINTS #include "trace.h" +#include "exp_rcv.h" static u8 __get_ib_hdr_len(struct ib_header *hdr) { @@ -128,6 +129,15 @@ const char *hfi1_trace_get_packet_l2_str(u8 l2) #define IETH_PRN "ieth rkey:0x%.8x" #define ATOMICACKETH_PRN "origdata:%llx" #define ATOMICETH_PRN "vaddr:0x%llx rkey:0x%.8x sdata:%llx cdata:%llx" +#define TID_RDMA_KDETH "kdeth0 0x%x kdeth1 0x%x" +#define TID_RDMA_KDETH_DATA "kdeth0 0x%x: kver %u sh %u intr %u tidctrl %u tid %x offset %x kdeth1 0x%x: jkey %x" +#define TID_READ_REQ_PRN "tid_flow_psn 0x%x tid_flow_qp 0x%x verbs_qp 0x%x" +#define TID_READ_RSP_PRN "verbs_qp 0x%x" +#define TID_WRITE_REQ_PRN "original_qp 0x%x" +#define TID_WRITE_RSP_PRN "tid_flow_psn 0x%x tid_flow_qp 0x%x verbs_qp 0x%x" +#define TID_WRITE_DATA_PRN "verbs_qp 0x%x" +#define TID_ACK_PRN "tid_flow_psn 0x%x verbs_psn 0x%x tid_flow_qp 0x%x verbs_qp 0x%x" +#define TID_RESYNC_PRN "verbs_qp 0x%x" #define OP(transport, op) IB_OPCODE_## transport ## _ ## op @@ -322,6 +332,99 @@ const char *parse_everbs_hdrs( parse_syndrome(be32_to_cpu(eh->aeth) >> 24), be32_to_cpu(eh->aeth) & IB_MSN_MASK); break; + case OP(TID_RDMA, WRITE_REQ): + trace_seq_printf(p, TID_RDMA_KDETH " " RETH_PRN " " + TID_WRITE_REQ_PRN, + le32_to_cpu(eh->tid_rdma.w_req.kdeth0), + le32_to_cpu(eh->tid_rdma.w_req.kdeth1), + ib_u64_get(&eh->tid_rdma.w_req.reth.vaddr), + be32_to_cpu(eh->tid_rdma.w_req.reth.rkey), + be32_to_cpu(eh->tid_rdma.w_req.reth.length), + be32_to_cpu(eh->tid_rdma.w_req.verbs_qp)); + break; + case OP(TID_RDMA, WRITE_RESP): + trace_seq_printf(p, TID_RDMA_KDETH " " AETH_PRN " " + TID_WRITE_RSP_PRN, + le32_to_cpu(eh->tid_rdma.w_rsp.kdeth0), + le32_to_cpu(eh->tid_rdma.w_rsp.kdeth1), + be32_to_cpu(eh->tid_rdma.w_rsp.aeth) >> 24, + parse_syndrome(/* aeth */ + be32_to_cpu(eh->tid_rdma.w_rsp.aeth) + >> 24), + (be32_to_cpu(eh->tid_rdma.w_rsp.aeth) & + IB_MSN_MASK), + be32_to_cpu(eh->tid_rdma.w_rsp.tid_flow_psn), + be32_to_cpu(eh->tid_rdma.w_rsp.tid_flow_qp), + be32_to_cpu(eh->tid_rdma.w_rsp.verbs_qp)); + break; + case OP(TID_RDMA, WRITE_DATA_LAST): + case OP(TID_RDMA, WRITE_DATA): + trace_seq_printf(p, TID_RDMA_KDETH_DATA " " TID_WRITE_DATA_PRN, + le32_to_cpu(eh->tid_rdma.w_data.kdeth0), + KDETH_GET(eh->tid_rdma.w_data.kdeth0, KVER), + KDETH_GET(eh->tid_rdma.w_data.kdeth0, SH), + KDETH_GET(eh->tid_rdma.w_data.kdeth0, INTR), + KDETH_GET(eh->tid_rdma.w_data.kdeth0, TIDCTRL), + KDETH_GET(eh->tid_rdma.w_data.kdeth0, TID), + KDETH_GET(eh->tid_rdma.w_data.kdeth0, OFFSET), + le32_to_cpu(eh->tid_rdma.w_data.kdeth1), + KDETH_GET(eh->tid_rdma.w_data.kdeth1, JKEY), + be32_to_cpu(eh->tid_rdma.w_data.verbs_qp)); + break; + case OP(TID_RDMA, READ_REQ): + trace_seq_printf(p, TID_RDMA_KDETH " " RETH_PRN " " + TID_READ_REQ_PRN, + le32_to_cpu(eh->tid_rdma.r_req.kdeth0), + le32_to_cpu(eh->tid_rdma.r_req.kdeth1), + ib_u64_get(&eh->tid_rdma.r_req.reth.vaddr), + be32_to_cpu(eh->tid_rdma.r_req.reth.rkey), + be32_to_cpu(eh->tid_rdma.r_req.reth.length), + be32_to_cpu(eh->tid_rdma.r_req.tid_flow_psn), + be32_to_cpu(eh->tid_rdma.r_req.tid_flow_qp), + be32_to_cpu(eh->tid_rdma.r_req.verbs_qp)); + break; + case OP(TID_RDMA, READ_RESP): + trace_seq_printf(p, TID_RDMA_KDETH_DATA " " AETH_PRN " " + TID_READ_RSP_PRN, + le32_to_cpu(eh->tid_rdma.r_rsp.kdeth0), + KDETH_GET(eh->tid_rdma.r_rsp.kdeth0, KVER), + KDETH_GET(eh->tid_rdma.r_rsp.kdeth0, SH), + KDETH_GET(eh->tid_rdma.r_rsp.kdeth0, INTR), + KDETH_GET(eh->tid_rdma.r_rsp.kdeth0, TIDCTRL), + KDETH_GET(eh->tid_rdma.r_rsp.kdeth0, TID), + KDETH_GET(eh->tid_rdma.r_rsp.kdeth0, OFFSET), + le32_to_cpu(eh->tid_rdma.r_rsp.kdeth1), + KDETH_GET(eh->tid_rdma.r_rsp.kdeth1, JKEY), + be32_to_cpu(eh->tid_rdma.r_rsp.aeth) >> 24, + parse_syndrome(/* aeth */ + be32_to_cpu(eh->tid_rdma.r_rsp.aeth) + >> 24), + (be32_to_cpu(eh->tid_rdma.r_rsp.aeth) & + IB_MSN_MASK), + be32_to_cpu(eh->tid_rdma.r_rsp.verbs_qp)); + break; + case OP(TID_RDMA, ACK): + trace_seq_printf(p, TID_RDMA_KDETH " " AETH_PRN " " + TID_ACK_PRN, + le32_to_cpu(eh->tid_rdma.ack.kdeth0), + le32_to_cpu(eh->tid_rdma.ack.kdeth1), + be32_to_cpu(eh->tid_rdma.ack.aeth) >> 24, + parse_syndrome(/* aeth */ + be32_to_cpu(eh->tid_rdma.ack.aeth) + >> 24), + (be32_to_cpu(eh->tid_rdma.ack.aeth) & + IB_MSN_MASK), + be32_to_cpu(eh->tid_rdma.ack.tid_flow_psn), + be32_to_cpu(eh->tid_rdma.ack.verbs_psn), + be32_to_cpu(eh->tid_rdma.ack.tid_flow_qp), + be32_to_cpu(eh->tid_rdma.ack.verbs_qp)); + break; + case OP(TID_RDMA, RESYNC): + trace_seq_printf(p, TID_RDMA_KDETH " " TID_RESYNC_PRN, + le32_to_cpu(eh->tid_rdma.resync.kdeth0), + le32_to_cpu(eh->tid_rdma.resync.kdeth1), + be32_to_cpu(eh->tid_rdma.resync.verbs_qp)); + break; /* aeth + atomicacketh */ case OP(RC, ATOMIC_ACKNOWLEDGE): trace_seq_printf(p, AETH_PRN " " ATOMICACKETH_PRN, @@ -394,6 +497,21 @@ const char *print_u32_array( return ret; } +u8 hfi1_trace_get_tid_ctrl(u32 ent) +{ + return EXP_TID_GET(ent, CTRL); +} + +u16 hfi1_trace_get_tid_len(u32 ent) +{ + return EXP_TID_GET(ent, LEN); +} + +u16 hfi1_trace_get_tid_idx(u32 ent) +{ + return EXP_TID_GET(ent, IDX); +} + __hfi1_trace_fn(AFFINITY); __hfi1_trace_fn(PKT); __hfi1_trace_fn(PROC); diff --git a/drivers/infiniband/hw/hfi1/trace.h b/drivers/infiniband/hw/hfi1/trace.h index 84458f1325e1..1ce551864118 100644 --- a/drivers/infiniband/hw/hfi1/trace.h +++ b/drivers/infiniband/hw/hfi1/trace.h @@ -63,3 +63,4 @@ __print_symbolic(etype, \ #include "trace_tx.h" #include "trace_mmu.h" #include "trace_iowait.h" +#include "trace_tid.h" diff --git a/drivers/infiniband/hw/hfi1/trace_ibhdrs.h b/drivers/infiniband/hw/hfi1/trace_ibhdrs.h index 1dc2c28fc96e..d1372cc66de6 100644 --- a/drivers/infiniband/hw/hfi1/trace_ibhdrs.h +++ b/drivers/infiniband/hw/hfi1/trace_ibhdrs.h @@ -79,6 +79,14 @@ __print_symbolic(opcode, \ ib_opcode_name(RC_ATOMIC_ACKNOWLEDGE), \ ib_opcode_name(RC_COMPARE_SWAP), \ ib_opcode_name(RC_FETCH_ADD), \ + ib_opcode_name(TID_RDMA_WRITE_REQ), \ + ib_opcode_name(TID_RDMA_WRITE_RESP), \ + ib_opcode_name(TID_RDMA_WRITE_DATA), \ + ib_opcode_name(TID_RDMA_WRITE_DATA_LAST), \ + ib_opcode_name(TID_RDMA_READ_REQ), \ + ib_opcode_name(TID_RDMA_READ_RESP), \ + ib_opcode_name(TID_RDMA_RESYNC), \ + ib_opcode_name(TID_RDMA_ACK), \ ib_opcode_name(UC_SEND_FIRST), \ ib_opcode_name(UC_SEND_MIDDLE), \ ib_opcode_name(UC_SEND_LAST), \ diff --git a/drivers/infiniband/hw/hfi1/trace_rc.h b/drivers/infiniband/hw/hfi1/trace_rc.h index 8ce476570462..1ebca37862e0 100644 --- a/drivers/infiniband/hw/hfi1/trace_rc.h +++ b/drivers/infiniband/hw/hfi1/trace_rc.h @@ -109,6 +109,54 @@ DEFINE_EVENT(hfi1_rc_template, hfi1_rcv_error, TP_ARGS(qp, psn) ); +DEFINE_EVENT(/* event */ + hfi1_rc_template, hfi1_rc_completion, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DECLARE_EVENT_CLASS(/* rc_ack */ + hfi1_rc_ack_template, + TP_PROTO(struct rvt_qp *qp, u32 aeth, u32 psn, + struct rvt_swqe *wqe), + TP_ARGS(qp, aeth, psn, wqe), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u32, aeth) + __field(u32, psn) + __field(u8, opcode) + __field(u32, spsn) + __field(u32, lpsn) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)) + __entry->qpn = qp->ibqp.qp_num; + __entry->aeth = aeth; + __entry->psn = psn; + __entry->opcode = wqe->wr.opcode; + __entry->spsn = wqe->psn; + __entry->lpsn = wqe->lpsn; + ), + TP_printk(/* print */ + "[%s] qpn 0x%x aeth 0x%x psn 0x%x opcode 0x%x spsn 0x%x lpsn 0x%x", + __get_str(dev), + __entry->qpn, + __entry->aeth, + __entry->psn, + __entry->opcode, + __entry->spsn, + __entry->lpsn + ) +); + +DEFINE_EVENT(/* do_rc_ack */ + hfi1_rc_ack_template, hfi1_rc_ack_do, + TP_PROTO(struct rvt_qp *qp, u32 aeth, u32 psn, + struct rvt_swqe *wqe), + TP_ARGS(qp, aeth, psn, wqe) +); + #endif /* __HFI1_TRACE_RC_H */ #undef TRACE_INCLUDE_PATH diff --git a/drivers/infiniband/hw/hfi1/trace_rx.h b/drivers/infiniband/hw/hfi1/trace_rx.h index 7eceb57e0415..3cec960e9674 100644 --- a/drivers/infiniband/hw/hfi1/trace_rx.h +++ b/drivers/infiniband/hw/hfi1/trace_rx.h @@ -1,5 +1,5 @@ /* - * Copyright(c) 2015 - 2017 Intel Corporation. + * Copyright(c) 2015 - 2018 Intel Corporation. * * This file is provided under a dual BSD/GPLv2 license. When using or * redistributing this file, you may do so under either license. @@ -128,111 +128,6 @@ TRACE_EVENT(hfi1_receive_interrupt, ) ); -DECLARE_EVENT_CLASS( - hfi1_exp_tid_reg_unreg, - TP_PROTO(unsigned int ctxt, u16 subctxt, u32 rarr, - u32 npages, unsigned long va, unsigned long pa, - dma_addr_t dma), - TP_ARGS(ctxt, subctxt, rarr, npages, va, pa, dma), - TP_STRUCT__entry( - __field(unsigned int, ctxt) - __field(u16, subctxt) - __field(u32, rarr) - __field(u32, npages) - __field(unsigned long, va) - __field(unsigned long, pa) - __field(dma_addr_t, dma) - ), - TP_fast_assign( - __entry->ctxt = ctxt; - __entry->subctxt = subctxt; - __entry->rarr = rarr; - __entry->npages = npages; - __entry->va = va; - __entry->pa = pa; - __entry->dma = dma; - ), - TP_printk("[%u:%u] entry:%u, %u pages @ 0x%lx, va:0x%lx dma:0x%llx", - __entry->ctxt, - __entry->subctxt, - __entry->rarr, - __entry->npages, - __entry->pa, - __entry->va, - __entry->dma - ) - ); - -DEFINE_EVENT( - hfi1_exp_tid_reg_unreg, hfi1_exp_tid_unreg, - TP_PROTO(unsigned int ctxt, u16 subctxt, u32 rarr, u32 npages, - unsigned long va, unsigned long pa, dma_addr_t dma), - TP_ARGS(ctxt, subctxt, rarr, npages, va, pa, dma)); - -DEFINE_EVENT( - hfi1_exp_tid_reg_unreg, hfi1_exp_tid_reg, - TP_PROTO(unsigned int ctxt, u16 subctxt, u32 rarr, u32 npages, - unsigned long va, unsigned long pa, dma_addr_t dma), - TP_ARGS(ctxt, subctxt, rarr, npages, va, pa, dma)); - -TRACE_EVENT( - hfi1_put_tid, - TP_PROTO(struct hfi1_devdata *dd, - u32 index, u32 type, unsigned long pa, u16 order), - TP_ARGS(dd, index, type, pa, order), - TP_STRUCT__entry( - DD_DEV_ENTRY(dd) - __field(unsigned long, pa); - __field(u32, index); - __field(u32, type); - __field(u16, order); - ), - TP_fast_assign( - DD_DEV_ASSIGN(dd); - __entry->pa = pa; - __entry->index = index; - __entry->type = type; - __entry->order = order; - ), - TP_printk("[%s] type %s pa %lx index %u order %u", - __get_str(dev), - show_tidtype(__entry->type), - __entry->pa, - __entry->index, - __entry->order - ) -); - -TRACE_EVENT(hfi1_exp_tid_inval, - TP_PROTO(unsigned int ctxt, u16 subctxt, unsigned long va, u32 rarr, - u32 npages, dma_addr_t dma), - TP_ARGS(ctxt, subctxt, va, rarr, npages, dma), - TP_STRUCT__entry( - __field(unsigned int, ctxt) - __field(u16, subctxt) - __field(unsigned long, va) - __field(u32, rarr) - __field(u32, npages) - __field(dma_addr_t, dma) - ), - TP_fast_assign( - __entry->ctxt = ctxt; - __entry->subctxt = subctxt; - __entry->va = va; - __entry->rarr = rarr; - __entry->npages = npages; - __entry->dma = dma; - ), - TP_printk("[%u:%u] entry:%u, %u pages @ 0x%lx dma: 0x%llx", - __entry->ctxt, - __entry->subctxt, - __entry->rarr, - __entry->npages, - __entry->va, - __entry->dma - ) - ); - TRACE_EVENT(hfi1_mmu_invalidate, TP_PROTO(unsigned int ctxt, u16 subctxt, const char *type, unsigned long start, unsigned long end), diff --git a/drivers/infiniband/hw/hfi1/trace_tid.h b/drivers/infiniband/hw/hfi1/trace_tid.h new file mode 100644 index 000000000000..548dfc45a407 --- /dev/null +++ b/drivers/infiniband/hw/hfi1/trace_tid.h @@ -0,0 +1,1610 @@ +/* SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) */ +/* + * Copyright(c) 2018 Intel Corporation. + * + */ +#if !defined(__HFI1_TRACE_TID_H) || defined(TRACE_HEADER_MULTI_READ) +#define __HFI1_TRACE_TID_H + +#include <linux/tracepoint.h> +#include <linux/trace_seq.h> + +#include "hfi.h" + +#define tidtype_name(type) { PT_##type, #type } +#define show_tidtype(type) \ +__print_symbolic(type, \ + tidtype_name(EXPECTED), \ + tidtype_name(EAGER), \ + tidtype_name(INVALID)) \ + +#undef TRACE_SYSTEM +#define TRACE_SYSTEM hfi1_tid + +u8 hfi1_trace_get_tid_ctrl(u32 ent); +u16 hfi1_trace_get_tid_len(u32 ent); +u16 hfi1_trace_get_tid_idx(u32 ent); + +#define OPFN_PARAM_PRN "[%s] qpn 0x%x %s OPFN: qp 0x%x, max read %u, " \ + "max write %u, max length %u, jkey 0x%x timeout %u " \ + "urg %u" + +#define TID_FLOW_PRN "[%s] qpn 0x%x flow %d: idx %d resp_ib_psn 0x%x " \ + "generation 0x%x fpsn 0x%x-%x r_next_psn 0x%x " \ + "ib_psn 0x%x-%x npagesets %u tnode_cnt %u " \ + "tidcnt %u tid_idx %u tid_offset %u length %u sent %u" + +#define TID_NODE_PRN "[%s] qpn 0x%x %s idx %u grp base 0x%x map 0x%x " \ + "used %u cnt %u" + +#define RSP_INFO_PRN "[%s] qpn 0x%x state 0x%x s_state 0x%x psn 0x%x " \ + "r_psn 0x%x r_state 0x%x r_flags 0x%x " \ + "r_head_ack_queue %u s_tail_ack_queue %u " \ + "s_acked_ack_queue %u s_ack_state 0x%x " \ + "s_nak_state 0x%x s_flags 0x%x ps_flags 0x%x " \ + "iow_flags 0x%lx" + +#define SENDER_INFO_PRN "[%s] qpn 0x%x state 0x%x s_cur %u s_tail %u " \ + "s_head %u s_acked %u s_last %u s_psn 0x%x " \ + "s_last_psn 0x%x s_flags 0x%x ps_flags 0x%x " \ + "iow_flags 0x%lx s_state 0x%x s_num_rd %u s_retry %u" + +#define TID_READ_SENDER_PRN "[%s] qpn 0x%x newreq %u tid_r_reqs %u " \ + "tid_r_comp %u pending_tid_r_segs %u " \ + "s_flags 0x%x ps_flags 0x%x iow_flags 0x%lx " \ + "s_state 0x%x hw_flow_index %u generation 0x%x " \ + "fpsn 0x%x flow_flags 0x%x" + +#define TID_REQ_PRN "[%s] qpn 0x%x newreq %u opcode 0x%x psn 0x%x lpsn 0x%x " \ + "cur_seg %u comp_seg %u ack_seg %u alloc_seg %u " \ + "total_segs %u setup_head %u clear_tail %u flow_idx %u " \ + "acked_tail %u state %u r_ack_psn 0x%x r_flow_psn 0x%x " \ + "r_last_ackd 0x%x s_next_psn 0x%x" + +#define RCV_ERR_PRN "[%s] qpn 0x%x s_flags 0x%x state 0x%x " \ + "s_acked_ack_queue %u s_tail_ack_queue %u " \ + "r_head_ack_queue %u opcode 0x%x psn 0x%x r_psn 0x%x " \ + " diff %d" + +#define TID_WRITE_RSPDR_PRN "[%s] qpn 0x%x r_tid_head %u r_tid_tail %u " \ + "r_tid_ack %u r_tid_alloc %u alloc_w_segs %u " \ + "pending_tid_w_segs %u sync_pt %s " \ + "ps_nak_psn 0x%x ps_nak_state 0x%x " \ + "prnr_nak_state 0x%x hw_flow_index %u generation "\ + "0x%x fpsn 0x%x flow_flags 0x%x resync %s" \ + "r_next_psn_kdeth 0x%x" + +#define TID_WRITE_SENDER_PRN "[%s] qpn 0x%x newreq %u s_tid_cur %u " \ + "s_tid_tail %u s_tid_head %u " \ + "pending_tid_w_resp %u n_requests %u " \ + "n_tid_requests %u s_flags 0x%x ps_flags 0x%x "\ + "iow_flags 0x%lx s_state 0x%x s_retry %u" + +#define KDETH_EFLAGS_ERR_PRN "[%s] qpn 0x%x TID ERR: RcvType 0x%x " \ + "RcvTypeError 0x%x PSN 0x%x" + +DECLARE_EVENT_CLASS(/* class */ + hfi1_exp_tid_reg_unreg, + TP_PROTO(unsigned int ctxt, u16 subctxt, u32 rarr, u32 npages, + unsigned long va, unsigned long pa, dma_addr_t dma), + TP_ARGS(ctxt, subctxt, rarr, npages, va, pa, dma), + TP_STRUCT__entry(/* entry */ + __field(unsigned int, ctxt) + __field(u16, subctxt) + __field(u32, rarr) + __field(u32, npages) + __field(unsigned long, va) + __field(unsigned long, pa) + __field(dma_addr_t, dma) + ), + TP_fast_assign(/* assign */ + __entry->ctxt = ctxt; + __entry->subctxt = subctxt; + __entry->rarr = rarr; + __entry->npages = npages; + __entry->va = va; + __entry->pa = pa; + __entry->dma = dma; + ), + TP_printk("[%u:%u] entry:%u, %u pages @ 0x%lx, va:0x%lx dma:0x%llx", + __entry->ctxt, + __entry->subctxt, + __entry->rarr, + __entry->npages, + __entry->pa, + __entry->va, + __entry->dma + ) +); + +DEFINE_EVENT(/* exp_tid_unreg */ + hfi1_exp_tid_reg_unreg, hfi1_exp_tid_unreg, + TP_PROTO(unsigned int ctxt, u16 subctxt, u32 rarr, u32 npages, + unsigned long va, unsigned long pa, dma_addr_t dma), + TP_ARGS(ctxt, subctxt, rarr, npages, va, pa, dma) +); + +DEFINE_EVENT(/* exp_tid_reg */ + hfi1_exp_tid_reg_unreg, hfi1_exp_tid_reg, + TP_PROTO(unsigned int ctxt, u16 subctxt, u32 rarr, u32 npages, + unsigned long va, unsigned long pa, dma_addr_t dma), + TP_ARGS(ctxt, subctxt, rarr, npages, va, pa, dma) +); + +TRACE_EVENT(/* put_tid */ + hfi1_put_tid, + TP_PROTO(struct hfi1_devdata *dd, + u32 index, u32 type, unsigned long pa, u16 order), + TP_ARGS(dd, index, type, pa, order), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd) + __field(unsigned long, pa); + __field(u32, index); + __field(u32, type); + __field(u16, order); + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd); + __entry->pa = pa; + __entry->index = index; + __entry->type = type; + __entry->order = order; + ), + TP_printk("[%s] type %s pa %lx index %u order %u", + __get_str(dev), + show_tidtype(__entry->type), + __entry->pa, + __entry->index, + __entry->order + ) +); + +TRACE_EVENT(/* exp_tid_inval */ + hfi1_exp_tid_inval, + TP_PROTO(unsigned int ctxt, u16 subctxt, unsigned long va, u32 rarr, + u32 npages, dma_addr_t dma), + TP_ARGS(ctxt, subctxt, va, rarr, npages, dma), + TP_STRUCT__entry(/* entry */ + __field(unsigned int, ctxt) + __field(u16, subctxt) + __field(unsigned long, va) + __field(u32, rarr) + __field(u32, npages) + __field(dma_addr_t, dma) + ), + TP_fast_assign(/* assign */ + __entry->ctxt = ctxt; + __entry->subctxt = subctxt; + __entry->va = va; + __entry->rarr = rarr; + __entry->npages = npages; + __entry->dma = dma; + ), + TP_printk("[%u:%u] entry:%u, %u pages @ 0x%lx dma: 0x%llx", + __entry->ctxt, + __entry->subctxt, + __entry->rarr, + __entry->npages, + __entry->va, + __entry->dma + ) +); + +DECLARE_EVENT_CLASS(/* opfn_state */ + hfi1_opfn_state_template, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u16, requested) + __field(u16, completed) + __field(u8, curr) + ), + TP_fast_assign(/* assign */ + struct hfi1_qp_priv *priv = qp->priv; + + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->requested = priv->opfn.requested; + __entry->completed = priv->opfn.completed; + __entry->curr = priv->opfn.curr; + ), + TP_printk(/* print */ + "[%s] qpn 0x%x requested 0x%x completed 0x%x curr 0x%x", + __get_str(dev), + __entry->qpn, + __entry->requested, + __entry->completed, + __entry->curr + ) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_state_template, hfi1_opfn_state_conn_request, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_state_template, hfi1_opfn_state_sched_conn_request, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_state_template, hfi1_opfn_state_conn_response, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_state_template, hfi1_opfn_state_conn_reply, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_state_template, hfi1_opfn_state_conn_error, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DECLARE_EVENT_CLASS(/* opfn_data */ + hfi1_opfn_data_template, + TP_PROTO(struct rvt_qp *qp, u8 capcode, u64 data), + TP_ARGS(qp, capcode, data), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u32, state) + __field(u8, capcode) + __field(u64, data) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->state = qp->state; + __entry->capcode = capcode; + __entry->data = data; + ), + TP_printk(/* printk */ + "[%s] qpn 0x%x (state 0x%x) Capcode %u data 0x%llx", + __get_str(dev), + __entry->qpn, + __entry->state, + __entry->capcode, + __entry->data + ) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_data_template, hfi1_opfn_data_conn_request, + TP_PROTO(struct rvt_qp *qp, u8 capcode, u64 data), + TP_ARGS(qp, capcode, data) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_data_template, hfi1_opfn_data_conn_response, + TP_PROTO(struct rvt_qp *qp, u8 capcode, u64 data), + TP_ARGS(qp, capcode, data) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_data_template, hfi1_opfn_data_conn_reply, + TP_PROTO(struct rvt_qp *qp, u8 capcode, u64 data), + TP_ARGS(qp, capcode, data) +); + +DECLARE_EVENT_CLASS(/* opfn_param */ + hfi1_opfn_param_template, + TP_PROTO(struct rvt_qp *qp, char remote, + struct tid_rdma_params *param), + TP_ARGS(qp, remote, param), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(char, remote) + __field(u32, param_qp) + __field(u32, max_len) + __field(u16, jkey) + __field(u8, max_read) + __field(u8, max_write) + __field(u8, timeout) + __field(u8, urg) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->remote = remote; + __entry->param_qp = param->qp; + __entry->max_len = param->max_len; + __entry->jkey = param->jkey; + __entry->max_read = param->max_read; + __entry->max_write = param->max_write; + __entry->timeout = param->timeout; + __entry->urg = param->urg; + ), + TP_printk(/* print */ + OPFN_PARAM_PRN, + __get_str(dev), + __entry->qpn, + __entry->remote ? "remote" : "local", + __entry->param_qp, + __entry->max_read, + __entry->max_write, + __entry->max_len, + __entry->jkey, + __entry->timeout, + __entry->urg + ) +); + +DEFINE_EVENT(/* event */ + hfi1_opfn_param_template, hfi1_opfn_param, + TP_PROTO(struct rvt_qp *qp, char remote, + struct tid_rdma_params *param), + TP_ARGS(qp, remote, param) +); + +DECLARE_EVENT_CLASS(/* msg */ + hfi1_msg_template, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more), + TP_STRUCT__entry(/* entry */ + __field(u32, qpn) + __string(msg, msg) + __field(u64, more) + ), + TP_fast_assign(/* assign */ + __entry->qpn = qp ? qp->ibqp.qp_num : 0; + __assign_str(msg, msg); + __entry->more = more; + ), + TP_printk(/* print */ + "qpn 0x%x %s 0x%llx", + __entry->qpn, + __get_str(msg), + __entry->more + ) +); + +DEFINE_EVENT(/* event */ + hfi1_msg_template, hfi1_msg_opfn_conn_request, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more) +); + +DEFINE_EVENT(/* event */ + hfi1_msg_template, hfi1_msg_opfn_conn_error, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more) +); + +DEFINE_EVENT(/* event */ + hfi1_msg_template, hfi1_msg_alloc_tids, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more) +); + +DEFINE_EVENT(/* event */ + hfi1_msg_template, hfi1_msg_tid_restart_req, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more) +); + +DEFINE_EVENT(/* event */ + hfi1_msg_template, hfi1_msg_handle_kdeth_eflags, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more) +); + +DEFINE_EVENT(/* event */ + hfi1_msg_template, hfi1_msg_tid_timeout, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more) +); + +DEFINE_EVENT(/* event */ + hfi1_msg_template, hfi1_msg_tid_retry_timeout, + TP_PROTO(struct rvt_qp *qp, const char *msg, u64 more), + TP_ARGS(qp, msg, more) +); + +DECLARE_EVENT_CLASS(/* tid_flow_page */ + hfi1_tid_flow_page_template, + TP_PROTO(struct rvt_qp *qp, struct tid_rdma_flow *flow, u32 index, + char mtu8k, char v1, void *vaddr), + TP_ARGS(qp, flow, index, mtu8k, v1, vaddr), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(char, mtu8k) + __field(char, v1) + __field(u32, index) + __field(u64, page) + __field(u64, vaddr) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->mtu8k = mtu8k; + __entry->v1 = v1; + __entry->index = index; + __entry->page = vaddr ? (u64)virt_to_page(vaddr) : 0ULL; + __entry->vaddr = (u64)vaddr; + ), + TP_printk(/* print */ + "[%s] qpn 0x%x page[%u]: page 0x%llx %s 0x%llx", + __get_str(dev), + __entry->qpn, + __entry->index, + __entry->page, + __entry->mtu8k ? (__entry->v1 ? "v1" : "v0") : "vaddr", + __entry->vaddr + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_page_template, hfi1_tid_flow_page, + TP_PROTO(struct rvt_qp *qp, struct tid_rdma_flow *flow, u32 index, + char mtu8k, char v1, void *vaddr), + TP_ARGS(qp, flow, index, mtu8k, v1, vaddr) +); + +DECLARE_EVENT_CLASS(/* tid_pageset */ + hfi1_tid_pageset_template, + TP_PROTO(struct rvt_qp *qp, u32 index, u16 idx, u16 count), + TP_ARGS(qp, index, idx, count), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u32, index) + __field(u16, idx) + __field(u16, count) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->index = index; + __entry->idx = idx; + __entry->count = count; + ), + TP_printk(/* print */ + "[%s] qpn 0x%x list[%u]: idx %u count %u", + __get_str(dev), + __entry->qpn, + __entry->index, + __entry->idx, + __entry->count + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_pageset_template, hfi1_tid_pageset, + TP_PROTO(struct rvt_qp *qp, u32 index, u16 idx, u16 count), + TP_ARGS(qp, index, idx, count) +); + +DECLARE_EVENT_CLASS(/* tid_fow */ + hfi1_tid_flow_template, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(int, index) + __field(int, idx) + __field(u32, resp_ib_psn) + __field(u32, generation) + __field(u32, fspsn) + __field(u32, flpsn) + __field(u32, r_next_psn) + __field(u32, ib_spsn) + __field(u32, ib_lpsn) + __field(u32, npagesets) + __field(u32, tnode_cnt) + __field(u32, tidcnt) + __field(u32, tid_idx) + __field(u32, tid_offset) + __field(u32, length) + __field(u32, sent) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->index = index; + __entry->idx = flow->idx; + __entry->resp_ib_psn = flow->flow_state.resp_ib_psn; + __entry->generation = flow->flow_state.generation; + __entry->fspsn = full_flow_psn(flow, + flow->flow_state.spsn); + __entry->flpsn = full_flow_psn(flow, + flow->flow_state.lpsn); + __entry->r_next_psn = flow->flow_state.r_next_psn; + __entry->ib_spsn = flow->flow_state.ib_spsn; + __entry->ib_lpsn = flow->flow_state.ib_lpsn; + __entry->npagesets = flow->npagesets; + __entry->tnode_cnt = flow->tnode_cnt; + __entry->tidcnt = flow->tidcnt; + __entry->tid_idx = flow->tid_idx; + __entry->tid_offset = flow->tid_offset; + __entry->length = flow->length; + __entry->sent = flow->sent; + ), + TP_printk(/* print */ + TID_FLOW_PRN, + __get_str(dev), + __entry->qpn, + __entry->index, + __entry->idx, + __entry->resp_ib_psn, + __entry->generation, + __entry->fspsn, + __entry->flpsn, + __entry->r_next_psn, + __entry->ib_spsn, + __entry->ib_lpsn, + __entry->npagesets, + __entry->tnode_cnt, + __entry->tidcnt, + __entry->tid_idx, + __entry->tid_offset, + __entry->length, + __entry->sent + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_alloc, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_build_read_pkt, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_build_read_resp, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_rcv_read_req, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_rcv_read_resp, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_restart_req, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_build_write_resp, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_rcv_write_resp, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_build_write_data, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_rcv_tid_ack, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_rcv_resync, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_flow_template, hfi1_tid_flow_handle_kdeth_eflags, + TP_PROTO(struct rvt_qp *qp, int index, struct tid_rdma_flow *flow), + TP_ARGS(qp, index, flow) +); + +DECLARE_EVENT_CLASS(/* tid_node */ + hfi1_tid_node_template, + TP_PROTO(struct rvt_qp *qp, const char *msg, u32 index, u32 base, + u8 map, u8 used, u8 cnt), + TP_ARGS(qp, msg, index, base, map, used, cnt), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __string(msg, msg) + __field(u32, index) + __field(u32, base) + __field(u8, map) + __field(u8, used) + __field(u8, cnt) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __assign_str(msg, msg); + __entry->index = index; + __entry->base = base; + __entry->map = map; + __entry->used = used; + __entry->cnt = cnt; + ), + TP_printk(/* print */ + TID_NODE_PRN, + __get_str(dev), + __entry->qpn, + __get_str(msg), + __entry->index, + __entry->base, + __entry->map, + __entry->used, + __entry->cnt + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_node_template, hfi1_tid_node_add, + TP_PROTO(struct rvt_qp *qp, const char *msg, u32 index, u32 base, + u8 map, u8 used, u8 cnt), + TP_ARGS(qp, msg, index, base, map, used, cnt) +); + +DECLARE_EVENT_CLASS(/* tid_entry */ + hfi1_tid_entry_template, + TP_PROTO(struct rvt_qp *qp, int index, u32 ent), + TP_ARGS(qp, index, ent), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(int, index) + __field(u8, ctrl) + __field(u16, idx) + __field(u16, len) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->index = index; + __entry->ctrl = hfi1_trace_get_tid_ctrl(ent); + __entry->idx = hfi1_trace_get_tid_idx(ent); + __entry->len = hfi1_trace_get_tid_len(ent); + ), + TP_printk(/* print */ + "[%s] qpn 0x%x TID entry %d: idx %u len %u ctrl 0x%x", + __get_str(dev), + __entry->qpn, + __entry->index, + __entry->idx, + __entry->len, + __entry->ctrl + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_entry_template, hfi1_tid_entry_alloc, + TP_PROTO(struct rvt_qp *qp, int index, u32 entry), + TP_ARGS(qp, index, entry) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_entry_template, hfi1_tid_entry_build_read_resp, + TP_PROTO(struct rvt_qp *qp, int index, u32 ent), + TP_ARGS(qp, index, ent) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_entry_template, hfi1_tid_entry_rcv_read_req, + TP_PROTO(struct rvt_qp *qp, int index, u32 ent), + TP_ARGS(qp, index, ent) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_entry_template, hfi1_tid_entry_rcv_write_resp, + TP_PROTO(struct rvt_qp *qp, int index, u32 entry), + TP_ARGS(qp, index, entry) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_entry_template, hfi1_tid_entry_build_write_data, + TP_PROTO(struct rvt_qp *qp, int index, u32 entry), + TP_ARGS(qp, index, entry) +); + +DECLARE_EVENT_CLASS(/* rsp_info */ + hfi1_responder_info_template, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u8, state) + __field(u8, s_state) + __field(u32, psn) + __field(u32, r_psn) + __field(u8, r_state) + __field(u8, r_flags) + __field(u8, r_head_ack_queue) + __field(u8, s_tail_ack_queue) + __field(u8, s_acked_ack_queue) + __field(u8, s_ack_state) + __field(u8, s_nak_state) + __field(u8, r_nak_state) + __field(u32, s_flags) + __field(u32, ps_flags) + __field(unsigned long, iow_flags) + ), + TP_fast_assign(/* assign */ + struct hfi1_qp_priv *priv = qp->priv; + + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->state = qp->state; + __entry->s_state = qp->s_state; + __entry->psn = psn; + __entry->r_psn = qp->r_psn; + __entry->r_state = qp->r_state; + __entry->r_flags = qp->r_flags; + __entry->r_head_ack_queue = qp->r_head_ack_queue; + __entry->s_tail_ack_queue = qp->s_tail_ack_queue; + __entry->s_acked_ack_queue = qp->s_acked_ack_queue; + __entry->s_ack_state = qp->s_ack_state; + __entry->s_nak_state = qp->s_nak_state; + __entry->s_flags = qp->s_flags; + __entry->ps_flags = priv->s_flags; + __entry->iow_flags = priv->s_iowait.flags; + ), + TP_printk(/* print */ + RSP_INFO_PRN, + __get_str(dev), + __entry->qpn, + __entry->state, + __entry->s_state, + __entry->psn, + __entry->r_psn, + __entry->r_state, + __entry->r_flags, + __entry->r_head_ack_queue, + __entry->s_tail_ack_queue, + __entry->s_acked_ack_queue, + __entry->s_ack_state, + __entry->s_nak_state, + __entry->s_flags, + __entry->ps_flags, + __entry->iow_flags + ) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_make_rc_ack, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_rcv_tid_read_req, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_tid_rcv_error, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_tid_write_alloc_res, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_rcv_tid_write_req, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_build_tid_write_resp, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_rcv_tid_write_data, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_make_tid_ack, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DEFINE_EVENT(/* event */ + hfi1_responder_info_template, hfi1_rsp_handle_kdeth_eflags, + TP_PROTO(struct rvt_qp *qp, u32 psn), + TP_ARGS(qp, psn) +); + +DECLARE_EVENT_CLASS(/* sender_info */ + hfi1_sender_info_template, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u8, state) + __field(u32, s_cur) + __field(u32, s_tail) + __field(u32, s_head) + __field(u32, s_acked) + __field(u32, s_last) + __field(u32, s_psn) + __field(u32, s_last_psn) + __field(u32, s_flags) + __field(u32, ps_flags) + __field(unsigned long, iow_flags) + __field(u8, s_state) + __field(u8, s_num_rd) + __field(u8, s_retry) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)) + __entry->qpn = qp->ibqp.qp_num; + __entry->state = qp->state; + __entry->s_cur = qp->s_cur; + __entry->s_tail = qp->s_tail; + __entry->s_head = qp->s_head; + __entry->s_acked = qp->s_acked; + __entry->s_last = qp->s_last; + __entry->s_psn = qp->s_psn; + __entry->s_last_psn = qp->s_last_psn; + __entry->s_flags = qp->s_flags; + __entry->ps_flags = ((struct hfi1_qp_priv *)qp->priv)->s_flags; + __entry->iow_flags = + ((struct hfi1_qp_priv *)qp->priv)->s_iowait.flags; + __entry->s_state = qp->s_state; + __entry->s_num_rd = qp->s_num_rd_atomic; + __entry->s_retry = qp->s_retry; + ), + TP_printk(/* print */ + SENDER_INFO_PRN, + __get_str(dev), + __entry->qpn, + __entry->state, + __entry->s_cur, + __entry->s_tail, + __entry->s_head, + __entry->s_acked, + __entry->s_last, + __entry->s_psn, + __entry->s_last_psn, + __entry->s_flags, + __entry->ps_flags, + __entry->iow_flags, + __entry->s_state, + __entry->s_num_rd, + __entry->s_retry + ) +); + +DEFINE_EVENT(/* event */ + hfi1_sender_info_template, hfi1_sender_make_rc_req, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_sender_info_template, hfi1_sender_reset_psn, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_sender_info_template, hfi1_sender_restart_rc, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_sender_info_template, hfi1_sender_do_rc_ack, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_sender_info_template, hfi1_sender_rcv_tid_read_resp, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_sender_info_template, hfi1_sender_rcv_tid_ack, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_sender_info_template, hfi1_sender_make_tid_pkt, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DECLARE_EVENT_CLASS(/* tid_read_sender */ + hfi1_tid_read_sender_template, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(char, newreq) + __field(u32, tid_r_reqs) + __field(u32, tid_r_comp) + __field(u32, pending_tid_r_segs) + __field(u32, s_flags) + __field(u32, ps_flags) + __field(unsigned long, iow_flags) + __field(u8, s_state) + __field(u32, hw_flow_index) + __field(u32, generation) + __field(u32, fpsn) + __field(u32, flow_flags) + ), + TP_fast_assign(/* assign */ + struct hfi1_qp_priv *priv = qp->priv; + + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->newreq = newreq; + __entry->tid_r_reqs = priv->tid_r_reqs; + __entry->tid_r_comp = priv->tid_r_comp; + __entry->pending_tid_r_segs = priv->pending_tid_r_segs; + __entry->s_flags = qp->s_flags; + __entry->ps_flags = priv->s_flags; + __entry->iow_flags = priv->s_iowait.flags; + __entry->s_state = priv->s_state; + __entry->hw_flow_index = priv->flow_state.index; + __entry->generation = priv->flow_state.generation; + __entry->fpsn = priv->flow_state.psn; + __entry->flow_flags = priv->flow_state.flags; + ), + TP_printk(/* print */ + TID_READ_SENDER_PRN, + __get_str(dev), + __entry->qpn, + __entry->newreq, + __entry->tid_r_reqs, + __entry->tid_r_comp, + __entry->pending_tid_r_segs, + __entry->s_flags, + __entry->ps_flags, + __entry->iow_flags, + __entry->s_state, + __entry->hw_flow_index, + __entry->generation, + __entry->fpsn, + __entry->flow_flags + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_read_sender_template, hfi1_tid_read_sender_make_req, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq) +); + +DECLARE_EVENT_CLASS(/* tid_rdma_request */ + hfi1_tid_rdma_request_template, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(char, newreq) + __field(u8, opcode) + __field(u32, psn) + __field(u32, lpsn) + __field(u32, cur_seg) + __field(u32, comp_seg) + __field(u32, ack_seg) + __field(u32, alloc_seg) + __field(u32, total_segs) + __field(u16, setup_head) + __field(u16, clear_tail) + __field(u16, flow_idx) + __field(u16, acked_tail) + __field(u32, state) + __field(u32, r_ack_psn) + __field(u32, r_flow_psn) + __field(u32, r_last_acked) + __field(u32, s_next_psn) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->newreq = newreq; + __entry->opcode = opcode; + __entry->psn = psn; + __entry->lpsn = lpsn; + __entry->cur_seg = req->cur_seg; + __entry->comp_seg = req->comp_seg; + __entry->ack_seg = req->ack_seg; + __entry->alloc_seg = req->alloc_seg; + __entry->total_segs = req->total_segs; + __entry->setup_head = req->setup_head; + __entry->clear_tail = req->clear_tail; + __entry->flow_idx = req->flow_idx; + __entry->acked_tail = req->acked_tail; + __entry->state = req->state; + __entry->r_ack_psn = req->r_ack_psn; + __entry->r_flow_psn = req->r_flow_psn; + __entry->r_last_acked = req->r_last_acked; + __entry->s_next_psn = req->s_next_psn; + ), + TP_printk(/* print */ + TID_REQ_PRN, + __get_str(dev), + __entry->qpn, + __entry->newreq, + __entry->opcode, + __entry->psn, + __entry->lpsn, + __entry->cur_seg, + __entry->comp_seg, + __entry->ack_seg, + __entry->alloc_seg, + __entry->total_segs, + __entry->setup_head, + __entry->clear_tail, + __entry->flow_idx, + __entry->acked_tail, + __entry->state, + __entry->r_ack_psn, + __entry->r_flow_psn, + __entry->r_last_acked, + __entry->s_next_psn + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_make_req_read, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_build_read_req, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_read_req, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_read_resp, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_err, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_restart_req, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_setup_tid_wqe, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_write_alloc_res, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_write_req, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_build_write_resp, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_write_resp, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_write_data, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_tid_ack, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_tid_retry_timeout, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_rcv_resync, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_make_tid_pkt, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_make_tid_ack, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_handle_kdeth_eflags, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_make_rc_ack_write, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_rdma_request_template, hfi1_tid_req_make_req_write, + TP_PROTO(struct rvt_qp *qp, char newreq, u8 opcode, u32 psn, u32 lpsn, + struct tid_rdma_request *req), + TP_ARGS(qp, newreq, opcode, psn, lpsn, req) +); + +DECLARE_EVENT_CLASS(/* rc_rcv_err */ + hfi1_rc_rcv_err_template, + TP_PROTO(struct rvt_qp *qp, u32 opcode, u32 psn, int diff), + TP_ARGS(qp, opcode, psn, diff), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u32, s_flags) + __field(u8, state) + __field(u8, s_acked_ack_queue) + __field(u8, s_tail_ack_queue) + __field(u8, r_head_ack_queue) + __field(u32, opcode) + __field(u32, psn) + __field(u32, r_psn) + __field(int, diff) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)) + __entry->qpn = qp->ibqp.qp_num; + __entry->s_flags = qp->s_flags; + __entry->state = qp->state; + __entry->s_acked_ack_queue = qp->s_acked_ack_queue; + __entry->s_tail_ack_queue = qp->s_tail_ack_queue; + __entry->r_head_ack_queue = qp->r_head_ack_queue; + __entry->opcode = opcode; + __entry->psn = psn; + __entry->r_psn = qp->r_psn; + __entry->diff = diff; + ), + TP_printk(/* print */ + RCV_ERR_PRN, + __get_str(dev), + __entry->qpn, + __entry->s_flags, + __entry->state, + __entry->s_acked_ack_queue, + __entry->s_tail_ack_queue, + __entry->r_head_ack_queue, + __entry->opcode, + __entry->psn, + __entry->r_psn, + __entry->diff + ) +); + +DEFINE_EVENT(/* event */ + hfi1_rc_rcv_err_template, hfi1_tid_rdma_rcv_err, + TP_PROTO(struct rvt_qp *qp, u32 opcode, u32 psn, int diff), + TP_ARGS(qp, opcode, psn, diff) +); + +DECLARE_EVENT_CLASS(/* sge */ + hfi1_sge_template, + TP_PROTO(struct rvt_qp *qp, int index, struct rvt_sge *sge), + TP_ARGS(qp, index, sge), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(int, index) + __field(u64, vaddr) + __field(u32, sge_length) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->index = index; + __entry->vaddr = (u64)sge->vaddr; + __entry->sge_length = sge->sge_length; + ), + TP_printk(/* print */ + "[%s] qpn 0x%x sge %d: vaddr 0x%llx sge_length %u", + __get_str(dev), + __entry->qpn, + __entry->index, + __entry->vaddr, + __entry->sge_length + ) +); + +DEFINE_EVENT(/* event */ + hfi1_sge_template, hfi1_sge_check_align, + TP_PROTO(struct rvt_qp *qp, int index, struct rvt_sge *sge), + TP_ARGS(qp, index, sge) +); + +DECLARE_EVENT_CLASS(/* tid_write_sp */ + hfi1_tid_write_rsp_template, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u32, r_tid_head) + __field(u32, r_tid_tail) + __field(u32, r_tid_ack) + __field(u32, r_tid_alloc) + __field(u32, alloc_w_segs) + __field(u32, pending_tid_w_segs) + __field(bool, sync_pt) + __field(u32, ps_nak_psn) + __field(u8, ps_nak_state) + __field(u8, prnr_nak_state) + __field(u32, hw_flow_index) + __field(u32, generation) + __field(u32, fpsn) + __field(u32, flow_flags) + __field(bool, resync) + __field(u32, r_next_psn_kdeth) + ), + TP_fast_assign(/* assign */ + struct hfi1_qp_priv *priv = qp->priv; + + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->r_tid_head = priv->r_tid_head; + __entry->r_tid_tail = priv->r_tid_tail; + __entry->r_tid_ack = priv->r_tid_ack; + __entry->r_tid_alloc = priv->r_tid_alloc; + __entry->alloc_w_segs = priv->alloc_w_segs; + __entry->pending_tid_w_segs = priv->pending_tid_w_segs; + __entry->sync_pt = priv->sync_pt; + __entry->ps_nak_psn = priv->s_nak_psn; + __entry->ps_nak_state = priv->s_nak_state; + __entry->prnr_nak_state = priv->rnr_nak_state; + __entry->hw_flow_index = priv->flow_state.index; + __entry->generation = priv->flow_state.generation; + __entry->fpsn = priv->flow_state.psn; + __entry->flow_flags = priv->flow_state.flags; + __entry->resync = priv->resync; + __entry->r_next_psn_kdeth = priv->r_next_psn_kdeth; + ), + TP_printk(/* print */ + TID_WRITE_RSPDR_PRN, + __get_str(dev), + __entry->qpn, + __entry->r_tid_head, + __entry->r_tid_tail, + __entry->r_tid_ack, + __entry->r_tid_alloc, + __entry->alloc_w_segs, + __entry->pending_tid_w_segs, + __entry->sync_pt ? "yes" : "no", + __entry->ps_nak_psn, + __entry->ps_nak_state, + __entry->prnr_nak_state, + __entry->hw_flow_index, + __entry->generation, + __entry->fpsn, + __entry->flow_flags, + __entry->resync ? "yes" : "no", + __entry->r_next_psn_kdeth + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_alloc_res, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_rcv_req, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_build_resp, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_rcv_data, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_rcv_resync, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_make_tid_ack, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_handle_kdeth_eflags, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_rsp_template, hfi1_tid_write_rsp_make_rc_ack, + TP_PROTO(struct rvt_qp *qp), + TP_ARGS(qp) +); + +DECLARE_EVENT_CLASS(/* tid_write_sender */ + hfi1_tid_write_sender_template, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(char, newreq) + __field(u32, s_tid_cur) + __field(u32, s_tid_tail) + __field(u32, s_tid_head) + __field(u32, pending_tid_w_resp) + __field(u32, n_requests) + __field(u32, n_tid_requests) + __field(u32, s_flags) + __field(u32, ps_flags) + __field(unsigned long, iow_flags) + __field(u8, s_state) + __field(u8, s_retry) + ), + TP_fast_assign(/* assign */ + struct hfi1_qp_priv *priv = qp->priv; + + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->newreq = newreq; + __entry->s_tid_cur = priv->s_tid_cur; + __entry->s_tid_tail = priv->s_tid_tail; + __entry->s_tid_head = priv->s_tid_head; + __entry->pending_tid_w_resp = priv->pending_tid_w_resp; + __entry->n_requests = atomic_read(&priv->n_requests); + __entry->n_tid_requests = atomic_read(&priv->n_tid_requests); + __entry->s_flags = qp->s_flags; + __entry->ps_flags = priv->s_flags; + __entry->iow_flags = priv->s_iowait.flags; + __entry->s_state = priv->s_state; + __entry->s_retry = priv->s_retry; + ), + TP_printk(/* print */ + TID_WRITE_SENDER_PRN, + __get_str(dev), + __entry->qpn, + __entry->newreq, + __entry->s_tid_cur, + __entry->s_tid_tail, + __entry->s_tid_head, + __entry->pending_tid_w_resp, + __entry->n_requests, + __entry->n_tid_requests, + __entry->s_flags, + __entry->ps_flags, + __entry->iow_flags, + __entry->s_state, + __entry->s_retry + ) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_sender_template, hfi1_tid_write_sender_rcv_resp, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_sender_template, hfi1_tid_write_sender_rcv_tid_ack, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_sender_template, hfi1_tid_write_sender_retry_timeout, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_sender_template, hfi1_tid_write_sender_make_tid_pkt, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_sender_template, hfi1_tid_write_sender_make_req, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq) +); + +DEFINE_EVENT(/* event */ + hfi1_tid_write_sender_template, hfi1_tid_write_sender_restart_rc, + TP_PROTO(struct rvt_qp *qp, char newreq), + TP_ARGS(qp, newreq) +); + +DECLARE_EVENT_CLASS(/* tid_ack */ + hfi1_tid_ack_template, + TP_PROTO(struct rvt_qp *qp, u32 aeth, u32 psn, + u32 req_psn, u32 resync_psn), + TP_ARGS(qp, aeth, psn, req_psn, resync_psn), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u32, aeth) + __field(u32, psn) + __field(u32, req_psn) + __field(u32, resync_psn) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)) + __entry->qpn = qp->ibqp.qp_num; + __entry->aeth = aeth; + __entry->psn = psn; + __entry->req_psn = req_psn; + __entry->resync_psn = resync_psn; + ), + TP_printk(/* print */ + "[%s] qpn 0x%x aeth 0x%x psn 0x%x req_psn 0x%x resync_psn 0x%x", + __get_str(dev), + __entry->qpn, + __entry->aeth, + __entry->psn, + __entry->req_psn, + __entry->resync_psn + ) +); + +DEFINE_EVENT(/* rcv_tid_ack */ + hfi1_tid_ack_template, hfi1_rcv_tid_ack, + TP_PROTO(struct rvt_qp *qp, u32 aeth, u32 psn, + u32 req_psn, u32 resync_psn), + TP_ARGS(qp, aeth, psn, req_psn, resync_psn) +); + +DECLARE_EVENT_CLASS(/* kdeth_eflags_error */ + hfi1_kdeth_eflags_error_template, + TP_PROTO(struct rvt_qp *qp, u8 rcv_type, u8 rte, u32 psn), + TP_ARGS(qp, rcv_type, rte, psn), + TP_STRUCT__entry(/* entry */ + DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device)) + __field(u32, qpn) + __field(u8, rcv_type) + __field(u8, rte) + __field(u32, psn) + ), + TP_fast_assign(/* assign */ + DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)); + __entry->qpn = qp->ibqp.qp_num; + __entry->rcv_type = rcv_type; + __entry->rte = rte; + __entry->psn = psn; + ), + TP_printk(/* print */ + KDETH_EFLAGS_ERR_PRN, + __get_str(dev), + __entry->qpn, + __entry->rcv_type, + __entry->rte, + __entry->psn + ) +); + +DEFINE_EVENT(/* event */ + hfi1_kdeth_eflags_error_template, hfi1_eflags_err_write, + TP_PROTO(struct rvt_qp *qp, u8 rcv_type, u8 rte, u32 psn), + TP_ARGS(qp, rcv_type, rte, psn) +); + +#endif /* __HFI1_TRACE_TID_H */ + +#undef TRACE_INCLUDE_PATH +#undef TRACE_INCLUDE_FILE +#define TRACE_INCLUDE_PATH . +#define TRACE_INCLUDE_FILE trace_tid +#include <trace/define_trace.h> diff --git a/drivers/infiniband/hw/hfi1/trace_tx.h b/drivers/infiniband/hw/hfi1/trace_tx.h index c57af3b31fe1..09eb0c9ada00 100644 --- a/drivers/infiniband/hw/hfi1/trace_tx.h +++ b/drivers/infiniband/hw/hfi1/trace_tx.h @@ -114,19 +114,27 @@ DECLARE_EVENT_CLASS(hfi1_qpsleepwakeup_template, __field(u32, qpn) __field(u32, flags) __field(u32, s_flags) + __field(u32, ps_flags) + __field(unsigned long, iow_flags) ), TP_fast_assign( DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device)) __entry->flags = flags; __entry->qpn = qp->ibqp.qp_num; __entry->s_flags = qp->s_flags; + __entry->ps_flags = + ((struct hfi1_qp_priv *)qp->priv)->s_flags; + __entry->iow_flags = + ((struct hfi1_qp_priv *)qp->priv)->s_iowait.flags; ), TP_printk( - "[%s] qpn 0x%x flags 0x%x s_flags 0x%x", + "[%s] qpn 0x%x flags 0x%x s_flags 0x%x ps_flags 0x%x iow_flags 0x%lx", __get_str(dev), __entry->qpn, __entry->flags, - __entry->s_flags + __entry->s_flags, + __entry->ps_flags, + __entry->iow_flags ) ); @@ -838,6 +846,12 @@ DEFINE_EVENT( TP_ARGS(qp, flag) ); +DEFINE_EVENT(/* event */ + hfi1_do_send_template, hfi1_rc_do_tid_send, + TP_PROTO(struct rvt_qp *qp, bool flag), + TP_ARGS(qp, flag) +); + DEFINE_EVENT( hfi1_do_send_template, hfi1_rc_expired_time_slice, TP_PROTO(struct rvt_qp *qp, bool flag), diff --git a/drivers/infiniband/hw/hfi1/uc.c b/drivers/infiniband/hw/hfi1/uc.c index 6ba47037c424..4ed4fcfabd6c 100644 --- a/drivers/infiniband/hw/hfi1/uc.c +++ b/drivers/infiniband/hw/hfi1/uc.c @@ -271,7 +271,8 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) ps->s_txreq->ss = &qp->s_sge; ps->s_txreq->s_cur_size = len; hfi1_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24), - mask_psn(qp->s_psn++), middle, ps); + qp->remote_qpn, mask_psn(qp->s_psn++), + middle, ps); return 1; done_free_tx: diff --git a/drivers/infiniband/hw/hfi1/ud.c b/drivers/infiniband/hw/hfi1/ud.c index bf96067876c9..f88ad425664a 100644 --- a/drivers/infiniband/hw/hfi1/ud.c +++ b/drivers/infiniband/hw/hfi1/ud.c @@ -222,31 +222,11 @@ static void ud_loopback(struct rvt_qp *sqp, struct rvt_swqe *swqe) ssge.num_sge = swqe->wr.num_sge; sge = &ssge.sge; while (length) { - u32 len = sge->length; + u32 len = rvt_get_sge_length(sge, length); - if (len > length) - len = length; - if (len > sge->sge_length) - len = sge->sge_length; WARN_ON_ONCE(len == 0); rvt_copy_sge(qp, &qp->r_sge, sge->vaddr, len, true, false); - sge->vaddr += len; - sge->length -= len; - sge->sge_length -= len; - if (sge->sge_length == 0) { - if (--ssge.num_sge) - *sge = *ssge.sg_list++; - } else if (sge->length == 0 && sge->mr->lkey) { - if (++sge->n >= RVT_SEGSZ) { - if (++sge->m >= sge->mr->mapsz) - break; - sge->n = 0; - } - sge->vaddr = - sge->mr->map[sge->m]->segs[sge->n].vaddr; - sge->length = - sge->mr->map[sge->m]->segs[sge->n].length; - } + rvt_update_sge(&ssge, len, false); length -= len; } rvt_put_ss(&qp->r_sge); diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.h b/drivers/infiniband/hw/hfi1/user_exp_rcv.h index e383cc01a2bf..43b105de1d54 100644 --- a/drivers/infiniband/hw/hfi1/user_exp_rcv.h +++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.h @@ -48,7 +48,6 @@ */ #include "hfi.h" - #include "exp_rcv.h" struct tid_pageset { diff --git a/drivers/infiniband/hw/hfi1/user_pages.c b/drivers/infiniband/hw/hfi1/user_pages.c index e341e6dcc388..24b592c6522e 100644 --- a/drivers/infiniband/hw/hfi1/user_pages.c +++ b/drivers/infiniband/hw/hfi1/user_pages.c @@ -91,9 +91,7 @@ bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm, /* Convert to number of pages */ size = DIV_ROUND_UP(size, PAGE_SIZE); - down_read(&mm->mmap_sem); - pinned = mm->pinned_vm; - up_read(&mm->mmap_sem); + pinned = atomic64_read(&mm->pinned_vm); /* First, check the absolute limit against all pinned pages. */ if (pinned + npages >= ulimit && !can_lock) @@ -111,9 +109,7 @@ int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr, size_t np if (ret < 0) return ret; - down_write(&mm->mmap_sem); - mm->pinned_vm += ret; - up_write(&mm->mmap_sem); + atomic64_add(ret, &mm->pinned_vm); return ret; } @@ -130,8 +126,6 @@ void hfi1_release_user_pages(struct mm_struct *mm, struct page **p, } if (mm) { /* during close after signal, mm can be NULL */ - down_write(&mm->mmap_sem); - mm->pinned_vm -= npages; - up_write(&mm->mmap_sem); + atomic64_sub(npages, &mm->pinned_vm); } } diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c index e5e7fad09f32..8bfbc6d7ea34 100644 --- a/drivers/infiniband/hw/hfi1/user_sdma.c +++ b/drivers/infiniband/hw/hfi1/user_sdma.c @@ -144,8 +144,10 @@ static int defer_packet_queue( */ xchg(&pq->state, SDMA_PKT_Q_DEFERRED); write_seqlock(&sde->waitlock); - if (list_empty(&pq->busy.list)) + if (list_empty(&pq->busy.list)) { + iowait_get_priority(&pq->busy); iowait_queue(pkts_sent, &pq->busy, &sde->dmawait); + } write_sequnlock(&sde->waitlock); return -EBUSY; eagain: @@ -191,7 +193,7 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, pq->mm = fd->mm; iowait_init(&pq->busy, 0, NULL, NULL, defer_packet_queue, - activate_packet_queue, NULL); + activate_packet_queue, NULL, NULL); pq->reqidx = 0; pq->reqs = kcalloc(hfi1_sdma_comp_ring_size, @@ -1126,7 +1128,8 @@ static inline u32 set_pkt_bth_psn(__be32 bthpsn, u8 expct, u32 frags) 0xffffffull), psn = val & mask; if (expct) - psn = (psn & ~BTH_SEQ_MASK) | ((psn + frags) & BTH_SEQ_MASK); + psn = (psn & ~HFI1_KDETH_BTH_SEQ_MASK) | + ((psn + frags) & HFI1_KDETH_BTH_SEQ_MASK); else psn = psn + frags; return psn & mask; diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c index ec582d86025f..55a56b3d7f83 100644 --- a/drivers/infiniband/hw/hfi1/verbs.c +++ b/drivers/infiniband/hw/hfi1/verbs.c @@ -161,10 +161,12 @@ MODULE_PARM_DESC(wss_clean_period, "Count of verbs copies before an entry in the */ const enum ib_wc_opcode ib_hfi1_wc_opcode[] = { [IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE, + [IB_WR_TID_RDMA_WRITE] = IB_WC_RDMA_WRITE, [IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE, [IB_WR_SEND] = IB_WC_SEND, [IB_WR_SEND_WITH_IMM] = IB_WC_SEND, [IB_WR_RDMA_READ] = IB_WC_RDMA_READ, + [IB_WR_TID_RDMA_READ] = IB_WC_RDMA_READ, [IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP, [IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD, [IB_WR_SEND_WITH_INV] = IB_WC_SEND, @@ -200,6 +202,14 @@ const u8 hdr_len_by_opcode[256] = { [IB_OPCODE_RC_FETCH_ADD] = 12 + 8 + 28, [IB_OPCODE_RC_SEND_LAST_WITH_INVALIDATE] = 12 + 8 + 4, [IB_OPCODE_RC_SEND_ONLY_WITH_INVALIDATE] = 12 + 8 + 4, + [IB_OPCODE_TID_RDMA_READ_REQ] = 12 + 8 + 36, + [IB_OPCODE_TID_RDMA_READ_RESP] = 12 + 8 + 36, + [IB_OPCODE_TID_RDMA_WRITE_REQ] = 12 + 8 + 36, + [IB_OPCODE_TID_RDMA_WRITE_RESP] = 12 + 8 + 36, + [IB_OPCODE_TID_RDMA_WRITE_DATA] = 12 + 8 + 36, + [IB_OPCODE_TID_RDMA_WRITE_DATA_LAST] = 12 + 8 + 36, + [IB_OPCODE_TID_RDMA_ACK] = 12 + 8 + 36, + [IB_OPCODE_TID_RDMA_RESYNC] = 12 + 8 + 36, /* UC */ [IB_OPCODE_UC_SEND_FIRST] = 12 + 8, [IB_OPCODE_UC_SEND_MIDDLE] = 12 + 8, @@ -243,6 +253,17 @@ static const opcode_handler opcode_handler_tbl[256] = { [IB_OPCODE_RC_FETCH_ADD] = &hfi1_rc_rcv, [IB_OPCODE_RC_SEND_LAST_WITH_INVALIDATE] = &hfi1_rc_rcv, [IB_OPCODE_RC_SEND_ONLY_WITH_INVALIDATE] = &hfi1_rc_rcv, + + /* TID RDMA has separate handlers for different opcodes.*/ + [IB_OPCODE_TID_RDMA_WRITE_REQ] = &hfi1_rc_rcv_tid_rdma_write_req, + [IB_OPCODE_TID_RDMA_WRITE_RESP] = &hfi1_rc_rcv_tid_rdma_write_resp, + [IB_OPCODE_TID_RDMA_WRITE_DATA] = &hfi1_rc_rcv_tid_rdma_write_data, + [IB_OPCODE_TID_RDMA_WRITE_DATA_LAST] = &hfi1_rc_rcv_tid_rdma_write_data, + [IB_OPCODE_TID_RDMA_READ_REQ] = &hfi1_rc_rcv_tid_rdma_read_req, + [IB_OPCODE_TID_RDMA_READ_RESP] = &hfi1_rc_rcv_tid_rdma_read_resp, + [IB_OPCODE_TID_RDMA_RESYNC] = &hfi1_rc_rcv_tid_rdma_resync, + [IB_OPCODE_TID_RDMA_ACK] = &hfi1_rc_rcv_tid_rdma_ack, + /* UC */ [IB_OPCODE_UC_SEND_FIRST] = &hfi1_uc_rcv, [IB_OPCODE_UC_SEND_MIDDLE] = &hfi1_uc_rcv, @@ -308,7 +329,7 @@ static inline opcode_handler qp_ok(struct hfi1_packet *packet) static u64 hfi1_fault_tx(struct rvt_qp *qp, u8 opcode, u64 pbc) { #ifdef CONFIG_FAULT_INJECTION - if ((opcode & IB_OPCODE_MSP) == IB_OPCODE_MSP) + if ((opcode & IB_OPCODE_MSP) == IB_OPCODE_MSP) { /* * In order to drop non-IB traffic we * set PbcInsertHrc to NONE (0x2). @@ -319,8 +340,9 @@ static u64 hfi1_fault_tx(struct rvt_qp *qp, u8 opcode, u64 pbc) * packet will not be delivered to the * correct context. */ + pbc &= ~PBC_INSERT_HCRC_SMASK; pbc |= (u64)PBC_IHCRC_NONE << PBC_INSERT_HCRC_SHIFT; - else + } else { /* * In order to drop regular verbs * traffic we set the PbcTestEbp @@ -330,10 +352,129 @@ static u64 hfi1_fault_tx(struct rvt_qp *qp, u8 opcode, u64 pbc) * triggered and will be dropped. */ pbc |= PBC_TEST_EBP; + } #endif return pbc; } +static opcode_handler tid_qp_ok(int opcode, struct hfi1_packet *packet) +{ + if (packet->qp->ibqp.qp_type != IB_QPT_RC || + !(ib_rvt_state_ops[packet->qp->state] & RVT_PROCESS_RECV_OK)) + return NULL; + if ((opcode & RVT_OPCODE_QP_MASK) == IB_OPCODE_TID_RDMA) + return opcode_handler_tbl[opcode]; + return NULL; +} + +void hfi1_kdeth_eager_rcv(struct hfi1_packet *packet) +{ + struct hfi1_ctxtdata *rcd = packet->rcd; + struct ib_header *hdr = packet->hdr; + u32 tlen = packet->tlen; + struct hfi1_pportdata *ppd = rcd->ppd; + struct hfi1_ibport *ibp = &ppd->ibport_data; + struct rvt_dev_info *rdi = &ppd->dd->verbs_dev.rdi; + opcode_handler opcode_handler; + unsigned long flags; + u32 qp_num; + int lnh; + u8 opcode; + + /* DW == LRH (2) + BTH (3) + KDETH (9) + CRC (1) */ + if (unlikely(tlen < 15 * sizeof(u32))) + goto drop; + + lnh = be16_to_cpu(hdr->lrh[0]) & 3; + if (lnh != HFI1_LRH_BTH) + goto drop; + + packet->ohdr = &hdr->u.oth; + trace_input_ibhdr(rcd->dd, packet, !!(rhf_dc_info(packet->rhf))); + + opcode = (be32_to_cpu(packet->ohdr->bth[0]) >> 24); + inc_opstats(tlen, &rcd->opstats->stats[opcode]); + + /* verbs_qp can be picked up from any tid_rdma header struct */ + qp_num = be32_to_cpu(packet->ohdr->u.tid_rdma.r_req.verbs_qp) & + RVT_QPN_MASK; + + rcu_read_lock(); + packet->qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num); + if (!packet->qp) + goto drop_rcu; + spin_lock_irqsave(&packet->qp->r_lock, flags); + opcode_handler = tid_qp_ok(opcode, packet); + if (likely(opcode_handler)) + opcode_handler(packet); + else + goto drop_unlock; + spin_unlock_irqrestore(&packet->qp->r_lock, flags); + rcu_read_unlock(); + + return; +drop_unlock: + spin_unlock_irqrestore(&packet->qp->r_lock, flags); +drop_rcu: + rcu_read_unlock(); +drop: + ibp->rvp.n_pkt_drops++; +} + +void hfi1_kdeth_expected_rcv(struct hfi1_packet *packet) +{ + struct hfi1_ctxtdata *rcd = packet->rcd; + struct ib_header *hdr = packet->hdr; + u32 tlen = packet->tlen; + struct hfi1_pportdata *ppd = rcd->ppd; + struct hfi1_ibport *ibp = &ppd->ibport_data; + struct rvt_dev_info *rdi = &ppd->dd->verbs_dev.rdi; + opcode_handler opcode_handler; + unsigned long flags; + u32 qp_num; + int lnh; + u8 opcode; + + /* DW == LRH (2) + BTH (3) + KDETH (9) + CRC (1) */ + if (unlikely(tlen < 15 * sizeof(u32))) + goto drop; + + lnh = be16_to_cpu(hdr->lrh[0]) & 3; + if (lnh != HFI1_LRH_BTH) + goto drop; + + packet->ohdr = &hdr->u.oth; + trace_input_ibhdr(rcd->dd, packet, !!(rhf_dc_info(packet->rhf))); + + opcode = (be32_to_cpu(packet->ohdr->bth[0]) >> 24); + inc_opstats(tlen, &rcd->opstats->stats[opcode]); + + /* verbs_qp can be picked up from any tid_rdma header struct */ + qp_num = be32_to_cpu(packet->ohdr->u.tid_rdma.r_rsp.verbs_qp) & + RVT_QPN_MASK; + + rcu_read_lock(); + packet->qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num); + if (!packet->qp) + goto drop_rcu; + spin_lock_irqsave(&packet->qp->r_lock, flags); + opcode_handler = tid_qp_ok(opcode, packet); + if (likely(opcode_handler)) + opcode_handler(packet); + else + goto drop_unlock; + spin_unlock_irqrestore(&packet->qp->r_lock, flags); + rcu_read_unlock(); + + return; +drop_unlock: + spin_unlock_irqrestore(&packet->qp->r_lock, flags); +drop_rcu: + rcu_read_unlock(); +drop: + ibp->rvp.n_pkt_drops++; +} + static int hfi1_do_pkey_check(struct hfi1_packet *packet) { struct hfi1_ctxtdata *rcd = packet->rcd; @@ -504,11 +645,28 @@ static void verbs_sdma_complete( hfi1_put_txreq(tx); } +void hfi1_wait_kmem(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + struct ib_qp *ibqp = &qp->ibqp; + struct ib_device *ibdev = ibqp->device; + struct hfi1_ibdev *dev = to_idev(ibdev); + + if (list_empty(&priv->s_iowait.list)) { + if (list_empty(&dev->memwait)) + mod_timer(&dev->mem_timer, jiffies + 1); + qp->s_flags |= RVT_S_WAIT_KMEM; + list_add_tail(&priv->s_iowait.list, &dev->memwait); + priv->s_iowait.lock = &dev->iowait_lock; + trace_hfi1_qpsleep(qp, RVT_S_WAIT_KMEM); + rvt_get_qp(qp); + } +} + static int wait_kmem(struct hfi1_ibdev *dev, struct rvt_qp *qp, struct hfi1_pkt_state *ps) { - struct hfi1_qp_priv *priv = qp->priv; unsigned long flags; int ret = 0; @@ -517,15 +675,7 @@ static int wait_kmem(struct hfi1_ibdev *dev, write_seqlock(&dev->iowait_lock); list_add_tail(&ps->s_txreq->txreq.list, &ps->wait->tx_head); - if (list_empty(&priv->s_iowait.list)) { - if (list_empty(&dev->memwait)) - mod_timer(&dev->mem_timer, jiffies + 1); - qp->s_flags |= RVT_S_WAIT_KMEM; - list_add_tail(&priv->s_iowait.list, &dev->memwait); - priv->s_iowait.lock = &dev->iowait_lock; - trace_hfi1_qpsleep(qp, RVT_S_WAIT_KMEM); - rvt_get_qp(qp); - } + hfi1_wait_kmem(qp); write_sequnlock(&dev->iowait_lock); hfi1_qp_unbusy(qp, ps->wait); ret = -EBUSY; @@ -553,11 +703,7 @@ static noinline int build_verbs_ulp_payload( int ret = 0; while (length) { - len = ss->sge.length; - if (len > length) - len = length; - if (len > ss->sge.sge_length) - len = ss->sge.sge_length; + len = rvt_get_sge_length(&ss->sge, length); WARN_ON_ONCE(len == 0); ret = sdma_txadd_kvaddr( sde->dd, @@ -678,6 +824,15 @@ bail_txadd: return ret; } +static u64 update_hcrc(u8 opcode, u64 pbc) +{ + if ((opcode & IB_OPCODE_TID_RDMA) == IB_OPCODE_TID_RDMA) { + pbc &= ~PBC_INSERT_HCRC_SMASK; + pbc |= (u64)PBC_IHCRC_LKDETH << PBC_INSERT_HCRC_SHIFT; + } + return pbc; +} + int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps, u64 pbc) { @@ -723,6 +878,9 @@ int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps, qp->srate_mbps, vl, plen); + + /* Update HCRC based on packet opcode */ + pbc = update_hcrc(ps->opcode, pbc); } tx->wqe = qp->s_wqe; ret = build_verbs_tx_desc(tx->sde, len, tx, ahg_info, pbc); @@ -787,6 +945,7 @@ static int pio_wait(struct rvt_qp *qp, dev->n_piodrain += !!(flag & HFI1_S_WAIT_PIO_DRAIN); qp->s_flags |= flag; was_empty = list_empty(&sc->piowait); + iowait_get_priority(&priv->s_iowait); iowait_queue(ps->pkts_sent, &priv->s_iowait, &sc->piowait); priv->s_iowait.lock = &sc->waitlock; @@ -871,6 +1030,9 @@ int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps, if (unlikely(hfi1_dbg_should_fault_tx(qp, ps->opcode))) pbc = hfi1_fault_tx(qp, ps->opcode, pbc); pbc = create_pbc(ppd, pbc, qp->srate_mbps, vl, plen); + + /* Update HCRC based on packet opcode */ + pbc = update_hcrc(ps->opcode, pbc); } if (cb) iowait_pio_inc(&priv->s_iowait); @@ -914,12 +1076,8 @@ int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps, if (ss) { while (len) { void *addr = ss->sge.vaddr; - u32 slen = ss->sge.length; + u32 slen = rvt_get_sge_length(&ss->sge, len); - if (slen > len) - slen = len; - if (slen > ss->sge.sge_length) - slen = ss->sge.sge_length; rvt_update_sge(ss, slen, false); seg_pio_copy_mid(pbuf, addr, slen); len -= slen; @@ -1188,7 +1346,9 @@ static void hfi1_fill_device_attr(struct hfi1_devdata *dd) rdi->dparms.props.max_mr_size = U64_MAX; rdi->dparms.props.max_fast_reg_page_list_len = UINT_MAX; rdi->dparms.props.max_qp = hfi1_max_qps; - rdi->dparms.props.max_qp_wr = hfi1_max_qp_wrs; + rdi->dparms.props.max_qp_wr = + (hfi1_max_qp_wrs >= HFI1_QP_WQE_INVALID ? + HFI1_QP_WQE_INVALID - 1 : hfi1_max_qp_wrs); rdi->dparms.props.max_send_sge = hfi1_max_sges; rdi->dparms.props.max_recv_sge = hfi1_max_sges; rdi->dparms.props.max_sge_rd = hfi1_max_sges; @@ -1622,6 +1782,7 @@ static const struct ib_device_ops hfi1_dev_ops = { .alloc_rdma_netdev = hfi1_vnic_alloc_rn, .get_dev_fw_str = hfi1_get_dev_fw_str, .get_hw_stats = get_hw_stats, + .init_port = hfi1_create_port_files, .modify_device = modify_device, /* keep process mad in the driver */ .process_mad = hfi1_process_mad, @@ -1679,7 +1840,6 @@ int hfi1_register_ib_device(struct hfi1_devdata *dd) /* * Fill in rvt info object. */ - dd->verbs_dev.rdi.driver_f.port_callback = hfi1_create_port_files; dd->verbs_dev.rdi.driver_f.get_pci_dev = get_pci_dev; dd->verbs_dev.rdi.driver_f.check_ah = hfi1_check_ah; dd->verbs_dev.rdi.driver_f.notify_new_ah = hfi1_notify_new_ah; @@ -1743,6 +1903,8 @@ int hfi1_register_ib_device(struct hfi1_devdata *dd) dd->verbs_dev.rdi.dparms.sge_copy_mode = sge_copy_mode; dd->verbs_dev.rdi.dparms.wss_threshold = wss_threshold; dd->verbs_dev.rdi.dparms.wss_clean_period = wss_clean_period; + dd->verbs_dev.rdi.dparms.reserved_operations = 1; + dd->verbs_dev.rdi.dparms.extra_rdma_atomic = HFI1_TID_RDMA_WRITE_CNT; /* post send table */ dd->verbs_dev.rdi.post_parms = hfi1_post_parms; diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h index 1ad0b14bdb3c..62ace0b2d17a 100644 --- a/drivers/infiniband/hw/hfi1/verbs.h +++ b/drivers/infiniband/hw/hfi1/verbs.h @@ -72,6 +72,7 @@ struct hfi1_packet; #include "iowait.h" #include "tid_rdma.h" +#include "opfn.h" #define HFI1_MAX_RDMA_ATOMIC 16 @@ -158,10 +159,68 @@ struct hfi1_qp_priv { struct sdma_engine *s_sde; /* current sde */ struct send_context *s_sendcontext; /* current sendcontext */ struct hfi1_ctxtdata *rcd; /* QP's receive context */ + struct page **pages; /* for TID page scan */ + u32 tid_enqueue; /* saved when tid waited */ u8 s_sc; /* SC[0..4] for next packet */ struct iowait s_iowait; + struct timer_list s_tid_timer; /* for timing tid wait */ + struct timer_list s_tid_retry_timer; /* for timing tid ack */ + struct list_head tid_wait; /* for queueing tid space */ + struct hfi1_opfn_data opfn; + struct tid_flow_state flow_state; + struct tid_rdma_qp_params tid_rdma; struct rvt_qp *owner; u8 hdr_type; /* 9B or 16B */ + struct rvt_sge_state tid_ss; /* SGE state pointer for 2nd leg */ + atomic_t n_requests; /* # of TID RDMA requests in the */ + /* queue */ + atomic_t n_tid_requests; /* # of sent TID RDMA requests */ + unsigned long tid_timer_timeout_jiffies; + unsigned long tid_retry_timeout_jiffies; + + /* variables for the TID RDMA SE state machine */ + u8 s_state; + u8 s_retry; + u8 rnr_nak_state; /* RNR NAK state */ + u8 s_nak_state; + u32 s_nak_psn; + u32 s_flags; + u32 s_tid_cur; + u32 s_tid_head; + u32 s_tid_tail; + u32 r_tid_head; /* Most recently added TID RDMA request */ + u32 r_tid_tail; /* the last completed TID RDMA request */ + u32 r_tid_ack; /* the TID RDMA request to be ACK'ed */ + u32 r_tid_alloc; /* Request for which we are allocating resources */ + u32 pending_tid_w_segs; /* Num of pending tid write segments */ + u32 pending_tid_w_resp; /* Num of pending tid write responses */ + u32 alloc_w_segs; /* Number of segments for which write */ + /* resources have been allocated for this QP */ + + /* For TID RDMA READ */ + u32 tid_r_reqs; /* Num of tid reads requested */ + u32 tid_r_comp; /* Num of tid reads completed */ + u32 pending_tid_r_segs; /* Num of pending tid read segments */ + u16 pkts_ps; /* packets per segment */ + u8 timeout_shift; /* account for number of packets per segment */ + + u32 r_next_psn_kdeth; + u32 r_next_psn_kdeth_save; + u32 s_resync_psn; + u8 sync_pt; /* Set when QP reaches sync point */ + u8 resync; +}; + +#define HFI1_QP_WQE_INVALID ((u32)-1) + +struct hfi1_swqe_priv { + struct tid_rdma_request tid_req; + struct rvt_sge_state ss; /* Used for TID RDMA READ Request */ +}; + +struct hfi1_ack_priv { + struct rvt_sge_state ss; /* used for TID WRITE RESP */ + struct tid_rdma_request tid_req; }; /* @@ -225,6 +284,7 @@ struct hfi1_ibdev { struct kmem_cache *verbs_txreq_cache; u64 n_txwait; u64 n_kmem_wait; + u64 n_tidwait; /* protect iowait lists */ seqlock_t iowait_lock ____cacheline_aligned_in_smp; @@ -312,6 +372,31 @@ static inline u32 delta_psn(u32 a, u32 b) return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT; } +static inline struct tid_rdma_request *wqe_to_tid_req(struct rvt_swqe *wqe) +{ + return &((struct hfi1_swqe_priv *)wqe->priv)->tid_req; +} + +static inline struct tid_rdma_request *ack_to_tid_req(struct rvt_ack_entry *e) +{ + return &((struct hfi1_ack_priv *)e->priv)->tid_req; +} + +/* + * Look through all the active flows for a TID RDMA request and find + * the one (if it exists) that contains the specified PSN. + */ +static inline u32 __full_flow_psn(struct flow_state *state, u32 psn) +{ + return mask_psn((state->generation << HFI1_KDETH_BTH_SEQ_SHIFT) | + (psn & HFI1_KDETH_BTH_SEQ_MASK)); +} + +static inline u32 full_flow_psn(struct tid_rdma_flow *flow, u32 psn) +{ + return __full_flow_psn(&flow->flow_state, psn); +} + struct verbs_txreq; void hfi1_put_txreq(struct verbs_txreq *tx); @@ -356,9 +441,12 @@ u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr, const struct ib_global_route *grh, u32 hwords, u32 nwords); void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr, - u32 bth0, u32 bth2, int middle, + u32 bth0, u32 bth1, u32 bth2, int middle, struct hfi1_pkt_state *ps); +bool hfi1_schedule_send_yield(struct rvt_qp *qp, struct hfi1_pkt_state *ps, + bool tid); + void _hfi1_do_send(struct work_struct *work); void hfi1_do_send_from_rvt(struct rvt_qp *qp); @@ -377,6 +465,10 @@ int hfi1_register_ib_device(struct hfi1_devdata *); void hfi1_unregister_ib_device(struct hfi1_devdata *); +void hfi1_kdeth_eager_rcv(struct hfi1_packet *packet); + +void hfi1_kdeth_expected_rcv(struct hfi1_packet *packet); + void hfi1_ib_rcv(struct hfi1_packet *packet); void hfi1_16B_rcv(struct hfi1_packet *packet); @@ -394,6 +486,16 @@ static inline bool opa_bth_is_migration(struct ib_other_headers *ohdr) return ohdr->bth[1] & cpu_to_be32(OPA_BTH_MIG_REQ); } +void hfi1_wait_kmem(struct rvt_qp *qp); + +static inline void hfi1_trdma_send_complete(struct rvt_qp *qp, + struct rvt_swqe *wqe, + enum ib_wc_status status) +{ + trdma_clean_swqe(qp, wqe); + rvt_send_complete(qp, wqe, status); +} + extern const enum ib_wc_opcode ib_hfi1_wc_opcode[]; extern const u8 hdr_len_by_opcode[]; diff --git a/drivers/infiniband/hw/hfi1/verbs_txreq.h b/drivers/infiniband/hw/hfi1/verbs_txreq.h index 2a77af26a231..b002e96eb335 100644 --- a/drivers/infiniband/hw/hfi1/verbs_txreq.h +++ b/drivers/infiniband/hw/hfi1/verbs_txreq.h @@ -94,6 +94,7 @@ static inline struct verbs_txreq *get_txreq(struct hfi1_ibdev *dev, tx->txreq.num_desc = 0; /* Set the header type */ tx->phdr.hdr.hdr_type = priv->hdr_type; + tx->txreq.flags = 0; return tx; } diff --git a/drivers/infiniband/hw/hfi1/vnic_sdma.c b/drivers/infiniband/hw/hfi1/vnic_sdma.c index 1f81c480e028..af1b1ffcb38e 100644 --- a/drivers/infiniband/hw/hfi1/vnic_sdma.c +++ b/drivers/infiniband/hw/hfi1/vnic_sdma.c @@ -240,8 +240,10 @@ static int hfi1_vnic_sdma_sleep(struct sdma_engine *sde, } vnic_sdma->state = HFI1_VNIC_SDMA_Q_DEFERRED; - if (list_empty(&vnic_sdma->wait.list)) + if (list_empty(&vnic_sdma->wait.list)) { + iowait_get_priority(wait->iow); iowait_queue(pkts_sent, wait->iow, &sde->dmawait); + } write_sequnlock(&sde->waitlock); return -EBUSY; } @@ -281,7 +283,7 @@ void hfi1_vnic_sdma_init(struct hfi1_vnic_vport_info *vinfo) iowait_init(&vnic_sdma->wait, 0, NULL, NULL, hfi1_vnic_sdma_sleep, - hfi1_vnic_sdma_wakeup, NULL); + hfi1_vnic_sdma_wakeup, NULL, NULL); vnic_sdma->sde = &vinfo->dd->per_sdma[i]; vnic_sdma->dd = vinfo->dd; vnic_sdma->vinfo = vinfo; diff --git a/drivers/infiniband/hw/hns/Kconfig b/drivers/infiniband/hw/hns/Kconfig index 21c2100b2ea9..fddb5fdf92de 100644 --- a/drivers/infiniband/hw/hns/Kconfig +++ b/drivers/infiniband/hw/hns/Kconfig @@ -1,7 +1,6 @@ config INFINIBAND_HNS tristate "HNS RoCE Driver" depends on NET_VENDOR_HISILICON - depends on INFINIBAND_USER_ACCESS || !INFINIBAND_USER_ACCESS depends on ARM64 || (COMPILE_TEST && 64BIT) ---help--- This is a RoCE/RDMA driver for the Hisilicon RoCE engine. The engine diff --git a/drivers/infiniband/hw/hns/Makefile b/drivers/infiniband/hw/hns/Makefile index 004c88b32e13..e2a7f1488f76 100644 --- a/drivers/infiniband/hw/hns/Makefile +++ b/drivers/infiniband/hw/hns/Makefile @@ -2,7 +2,7 @@ # Makefile for the Hisilicon RoCE drivers. # -ccflags-y := -Idrivers/net/ethernet/hisilicon/hns3 +ccflags-y := -I $(srctree)/drivers/net/ethernet/hisilicon/hns3 obj-$(CONFIG_INFINIBAND_HNS) += hns-roce.o hns-roce-objs := hns_roce_main.o hns_roce_cmd.o hns_roce_pd.o \ diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.c b/drivers/infiniband/hw/hns/hns_roce_cmd.c index a0ba19d4a10e..2acf946d02e5 100644 --- a/drivers/infiniband/hw/hns/hns_roce_cmd.c +++ b/drivers/infiniband/hw/hns/hns_roce_cmd.c @@ -176,17 +176,33 @@ int hns_roce_cmd_mbox(struct hns_roce_dev *hr_dev, u64 in_param, u64 out_param, unsigned long in_modifier, u8 op_modifier, u16 op, unsigned long timeout) { - if (hr_dev->is_reset) - return 0; + int ret; + + if (hr_dev->hw->rst_prc_mbox) { + ret = hr_dev->hw->rst_prc_mbox(hr_dev); + if (ret == CMD_RST_PRC_SUCCESS) + return 0; + else if (ret == CMD_RST_PRC_EBUSY) + return -EBUSY; + } if (hr_dev->cmd.use_events) - return hns_roce_cmd_mbox_wait(hr_dev, in_param, out_param, - in_modifier, op_modifier, op, - timeout); + ret = hns_roce_cmd_mbox_wait(hr_dev, in_param, out_param, + in_modifier, op_modifier, op, + timeout); else - return hns_roce_cmd_mbox_poll(hr_dev, in_param, out_param, - in_modifier, op_modifier, op, - timeout); + ret = hns_roce_cmd_mbox_poll(hr_dev, in_param, out_param, + in_modifier, op_modifier, op, + timeout); + + if (ret == CMD_RST_PRC_EBUSY) + return -EBUSY; + + if (ret && (hr_dev->hw->rst_prc_mbox && + hr_dev->hw->rst_prc_mbox(hr_dev) == CMD_RST_PRC_SUCCESS)) + return 0; + + return ret; } EXPORT_SYMBOL_GPL(hns_roce_cmd_mbox); diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.h b/drivers/infiniband/hw/hns/hns_roce_cmd.h index 927701df5eff..059fd1da493e 100644 --- a/drivers/infiniband/hw/hns/hns_roce_cmd.h +++ b/drivers/infiniband/hw/hns/hns_roce_cmd.h @@ -75,6 +75,10 @@ enum { HNS_ROCE_CMD_DESTROY_MPT_BT1 = 0x29, HNS_ROCE_CMD_DESTROY_MPT_BT2 = 0x2a, + /* CQC TIMER commands */ + HNS_ROCE_CMD_WRITE_CQC_TIMER_BT0 = 0x23, + HNS_ROCE_CMD_READ_CQC_TIMER_BT0 = 0x27, + /* MPT commands */ HNS_ROCE_CMD_QUERY_MPT = 0x62, @@ -89,6 +93,10 @@ enum { HNS_ROCE_CMD_DESTROY_SRQC_BT1 = 0x39, HNS_ROCE_CMD_DESTROY_SRQC_BT2 = 0x3a, + /* QPC TIMER commands */ + HNS_ROCE_CMD_WRITE_QPC_TIMER_BT0 = 0x33, + HNS_ROCE_CMD_READ_QPC_TIMER_BT0 = 0x37, + /* EQC commands */ HNS_ROCE_CMD_CREATE_AEQC = 0x80, HNS_ROCE_CMD_MODIFY_AEQC = 0x81, @@ -98,6 +106,10 @@ enum { HNS_ROCE_CMD_MODIFY_CEQC = 0x91, HNS_ROCE_CMD_QUERY_CEQC = 0x92, HNS_ROCE_CMD_DESTROY_CEQC = 0x93, + + /* SCC CTX BT commands */ + HNS_ROCE_CMD_READ_SCCC_BT0 = 0xa4, + HNS_ROCE_CMD_WRITE_SCCC_BT0 = 0xa5, }; enum { diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c index 3a485f50fede..1dfe5627006c 100644 --- a/drivers/infiniband/hw/hns/hns_roce_cq.c +++ b/drivers/infiniband/hw/hns/hns_roce_cq.c @@ -215,7 +215,7 @@ void hns_roce_free_cq(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq) EXPORT_SYMBOL_GPL(hns_roce_free_cq); static int hns_roce_ib_get_cq_umem(struct hns_roce_dev *hr_dev, - struct ib_ucontext *context, + struct ib_udata *udata, struct hns_roce_cq_buf *buf, struct ib_umem **umem, u64 buf_addr, int cqe) { @@ -223,7 +223,7 @@ static int hns_roce_ib_get_cq_umem(struct hns_roce_dev *hr_dev, u32 page_shift; u32 npages; - *umem = ib_umem_get(context, buf_addr, cqe * hr_dev->caps.cq_entry_sz, + *umem = ib_umem_get(udata, buf_addr, cqe * hr_dev->caps.cq_entry_sz, IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(*umem)) return PTR_ERR(*umem); @@ -347,7 +347,7 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev, } /* Get user space address, write it into mtt table */ - ret = hns_roce_ib_get_cq_umem(hr_dev, context, &hr_cq->hr_buf, + ret = hns_roce_ib_get_cq_umem(hr_dev, udata, &hr_cq->hr_buf, &hr_cq->umem, ucmd.buf_addr, cq_entries); if (ret) { @@ -358,7 +358,8 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev, if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) && (udata->outlen >= sizeof(resp))) { ret = hns_roce_db_map_user(to_hr_ucontext(context), - ucmd.db_addr, &hr_cq->db); + udata, ucmd.db_addr, + &hr_cq->db); if (ret) { dev_err(dev, "cq record doorbell map failed!\n"); goto err_mtt; diff --git a/drivers/infiniband/hw/hns/hns_roce_db.c b/drivers/infiniband/hw/hns/hns_roce_db.c index e2f93c1ce86a..0c6c1fe87705 100644 --- a/drivers/infiniband/hw/hns/hns_roce_db.c +++ b/drivers/infiniband/hw/hns/hns_roce_db.c @@ -8,7 +8,8 @@ #include <rdma/ib_umem.h> #include "hns_roce_device.h" -int hns_roce_db_map_user(struct hns_roce_ucontext *context, unsigned long virt, +int hns_roce_db_map_user(struct hns_roce_ucontext *context, + struct ib_udata *udata, unsigned long virt, struct hns_roce_db *db) { struct hns_roce_user_db_page *page; @@ -28,8 +29,7 @@ int hns_roce_db_map_user(struct hns_roce_ucontext *context, unsigned long virt, refcount_set(&page->refcount, 1); page->user_virt = (virt & PAGE_MASK); - page->umem = ib_umem_get(&context->ibucontext, virt & PAGE_MASK, - PAGE_SIZE, 0, 0); + page->umem = ib_umem_get(udata, virt & PAGE_MASK, PAGE_SIZE, 0, 0); if (IS_ERR(page->umem)) { ret = PTR_ERR(page->umem); kfree(page); diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h index 509e467843f6..9ee86daf1700 100644 --- a/drivers/infiniband/hw/hns/hns_roce_device.h +++ b/drivers/infiniband/hw/hns/hns_roce_device.h @@ -202,6 +202,7 @@ enum { HNS_ROCE_CAP_FLAG_SRQ = BIT(5), HNS_ROCE_CAP_FLAG_MW = BIT(7), HNS_ROCE_CAP_FLAG_FRMR = BIT(8), + HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL = BIT(9), HNS_ROCE_CAP_FLAG_ATOMIC = BIT(10), }; @@ -216,6 +217,32 @@ enum { HNS_ROCE_DB_PER_PAGE = PAGE_SIZE / 4 }; +enum hns_roce_reset_stage { + HNS_ROCE_STATE_NON_RST, + HNS_ROCE_STATE_RST_BEF_DOWN, + HNS_ROCE_STATE_RST_DOWN, + HNS_ROCE_STATE_RST_UNINIT, + HNS_ROCE_STATE_RST_INIT, + HNS_ROCE_STATE_RST_INITED, +}; + +enum hns_roce_instance_state { + HNS_ROCE_STATE_NON_INIT, + HNS_ROCE_STATE_INIT, + HNS_ROCE_STATE_INITED, + HNS_ROCE_STATE_UNINIT, +}; + +enum { + HNS_ROCE_RST_DIRECT_RETURN = 0, +}; + +enum { + CMD_RST_PRC_OTHERS, + CMD_RST_PRC_SUCCESS, + CMD_RST_PRC_EBUSY, +}; + #define HNS_ROCE_CMD_SUCCESS 1 #define HNS_ROCE_PORT_DOWN 0 @@ -482,6 +509,8 @@ struct hns_roce_qp_table { struct hns_roce_hem_table qp_table; struct hns_roce_hem_table irrl_table; struct hns_roce_hem_table trrl_table; + struct hns_roce_hem_table sccc_table; + struct mutex scc_mutex; }; struct hns_roce_cq_table { @@ -729,6 +758,8 @@ struct hns_roce_caps { u32 max_extend_sg; int num_qps; /* 256k */ int reserved_qps; + int num_qpc_timer; + int num_cqc_timer; u32 max_srq_sg; int num_srqs; u32 max_wqes; /* 16k */ @@ -768,6 +799,9 @@ struct hns_roce_caps { int irrl_entry_sz; int trrl_entry_sz; int cqc_entry_sz; + int sccc_entry_sz; + int qpc_timer_entry_sz; + int cqc_timer_entry_sz; int srqc_entry_sz; int idx_entry_sz; u32 pbl_ba_pg_sz; @@ -777,9 +811,12 @@ struct hns_roce_caps { int ceqe_depth; enum ib_mtu max_mtu; u32 qpc_bt_num; + u32 qpc_timer_bt_num; u32 srqc_bt_num; u32 cqc_bt_num; + u32 cqc_timer_bt_num; u32 mpt_bt_num; + u32 sccc_bt_num; u32 qpc_ba_pg_sz; u32 qpc_buf_pg_sz; u32 qpc_hop_num; @@ -795,6 +832,15 @@ struct hns_roce_caps { u32 mtt_ba_pg_sz; u32 mtt_buf_pg_sz; u32 mtt_hop_num; + u32 sccc_ba_pg_sz; + u32 sccc_buf_pg_sz; + u32 sccc_hop_num; + u32 qpc_timer_ba_pg_sz; + u32 qpc_timer_buf_pg_sz; + u32 qpc_timer_hop_num; + u32 cqc_timer_ba_pg_sz; + u32 cqc_timer_buf_pg_sz; + u32 cqc_timer_hop_num; u32 cqe_ba_pg_sz; u32 cqe_buf_pg_sz; u32 cqe_hop_num; @@ -834,6 +880,7 @@ struct hns_roce_hw { u64 out_param, u32 in_modifier, u8 op_modifier, u16 op, u16 token, int event); int (*chk_mbox)(struct hns_roce_dev *hr_dev, unsigned long timeout); + int (*rst_prc_mbox)(struct hns_roce_dev *hr_dev); int (*set_gid)(struct hns_roce_dev *hr_dev, u8 port, int gid_index, const union ib_gid *gid, const struct ib_gid_attr *attr); int (*set_mac)(struct hns_roce_dev *hr_dev, u8 phy_port, u8 *addr); @@ -861,6 +908,8 @@ struct hns_roce_hw { int attr_mask, enum ib_qp_state cur_state, enum ib_qp_state new_state); int (*destroy_qp)(struct ib_qp *ibqp); + int (*qp_flow_control_init)(struct hns_roce_dev *hr_dev, + struct hns_roce_qp *hr_qp); int (*post_send)(struct ib_qp *ibqp, const struct ib_send_wr *wr, const struct ib_send_wr **bad_wr); int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr, @@ -898,6 +947,8 @@ struct hns_roce_dev { spinlock_t bt_cmd_lock; bool active; bool is_reset; + bool dis_db; + unsigned long reset_cnt; struct hns_roce_ib_iboe iboe; struct list_head pgdir_list; @@ -922,6 +973,8 @@ struct hns_roce_dev { struct hns_roce_srq_table srq_table; struct hns_roce_qp_table qp_table; struct hns_roce_eq_table eq_table; + struct hns_roce_hem_table qpc_timer_table; + struct hns_roce_hem_table cqc_timer_table; int cmd_mod; int loop_idc; @@ -1061,10 +1114,9 @@ struct ib_ah *hns_roce_create_ah(struct ib_pd *pd, int hns_roce_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr); int hns_roce_destroy_ah(struct ib_ah *ah, u32 flags); -struct ib_pd *hns_roce_alloc_pd(struct ib_device *ib_dev, - struct ib_ucontext *context, - struct ib_udata *udata); -int hns_roce_dealloc_pd(struct ib_pd *pd); +int hns_roce_alloc_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata); +void hns_roce_dealloc_pd(struct ib_pd *pd); struct ib_mr *hns_roce_get_dma_mr(struct ib_pd *pd, int acc); struct ib_mr *hns_roce_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, @@ -1133,7 +1185,8 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev, int hns_roce_ib_destroy_cq(struct ib_cq *ib_cq); void hns_roce_free_cq(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq); -int hns_roce_db_map_user(struct hns_roce_ucontext *context, unsigned long virt, +int hns_roce_db_map_user(struct hns_roce_ucontext *context, + struct ib_udata *udata, unsigned long virt, struct hns_roce_db *db); void hns_roce_db_unmap_user(struct hns_roce_ucontext *context, struct hns_roce_db *db); diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c index 4cdbcafa5915..f1fec56f3ff4 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hem.c +++ b/drivers/infiniband/hw/hns/hns_roce_hem.c @@ -45,6 +45,9 @@ bool hns_roce_check_whether_mhop(struct hns_roce_dev *hr_dev, u32 type) (hr_dev->caps.mpt_hop_num && type == HEM_TYPE_MTPT) || (hr_dev->caps.cqc_hop_num && type == HEM_TYPE_CQC) || (hr_dev->caps.srqc_hop_num && type == HEM_TYPE_SRQC) || + (hr_dev->caps.sccc_hop_num && type == HEM_TYPE_SCCC) || + (hr_dev->caps.qpc_timer_hop_num && type == HEM_TYPE_QPC_TIMER) || + (hr_dev->caps.cqc_timer_hop_num && type == HEM_TYPE_CQC_TIMER) || (hr_dev->caps.cqe_hop_num && type == HEM_TYPE_CQE) || (hr_dev->caps.mtt_hop_num && type == HEM_TYPE_MTT) || (hr_dev->caps.srqwqe_hop_num && type == HEM_TYPE_SRQWQE) || @@ -125,6 +128,30 @@ int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev, mhop->ba_l0_num = hr_dev->caps.cqc_bt_num; mhop->hop_num = hr_dev->caps.cqc_hop_num; break; + case HEM_TYPE_SCCC: + mhop->buf_chunk_size = 1 << (hr_dev->caps.sccc_buf_pg_sz + + PAGE_SHIFT); + mhop->bt_chunk_size = 1 << (hr_dev->caps.sccc_ba_pg_sz + + PAGE_SHIFT); + mhop->ba_l0_num = hr_dev->caps.sccc_bt_num; + mhop->hop_num = hr_dev->caps.sccc_hop_num; + break; + case HEM_TYPE_QPC_TIMER: + mhop->buf_chunk_size = 1 << (hr_dev->caps.qpc_timer_buf_pg_sz + + PAGE_SHIFT); + mhop->bt_chunk_size = 1 << (hr_dev->caps.qpc_timer_ba_pg_sz + + PAGE_SHIFT); + mhop->ba_l0_num = hr_dev->caps.qpc_timer_bt_num; + mhop->hop_num = hr_dev->caps.qpc_timer_hop_num; + break; + case HEM_TYPE_CQC_TIMER: + mhop->buf_chunk_size = 1 << (hr_dev->caps.cqc_timer_buf_pg_sz + + PAGE_SHIFT); + mhop->bt_chunk_size = 1 << (hr_dev->caps.cqc_timer_ba_pg_sz + + PAGE_SHIFT); + mhop->ba_l0_num = hr_dev->caps.cqc_timer_bt_num; + mhop->hop_num = hr_dev->caps.cqc_timer_hop_num; + break; case HEM_TYPE_SRQC: mhop->buf_chunk_size = 1 << (hr_dev->caps.srqc_buf_pg_sz + PAGE_SHIFT); @@ -175,7 +202,7 @@ int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev, return 0; /* - * QPC/MTPT/CQC/SRQC alloc hem for buffer pages. + * QPC/MTPT/CQC/SRQC/SCCC alloc hem for buffer pages. * MTT/CQE alloc hem for bt pages. */ bt_num = hns_roce_get_bt_num(table->type, mhop->hop_num); @@ -486,7 +513,7 @@ static int hns_roce_table_mhop_get(struct hns_roce_dev *hr_dev, } /* - * alloc buffer space chunk for QPC/MTPT/CQC/SRQC. + * alloc buffer space chunk for QPC/MTPT/CQC/SRQC/SCCC. * alloc bt space chunk for MTT/CQE. */ size = table->type < HEM_TYPE_MTT ? buf_chunk_size : bt_chunk_size; @@ -593,6 +620,7 @@ out: mutex_unlock(&table->mutex); return ret; } +EXPORT_SYMBOL_GPL(hns_roce_table_get); static void hns_roce_table_mhop_put(struct hns_roce_dev *hr_dev, struct hns_roce_hem_table *table, @@ -658,7 +686,7 @@ static void hns_roce_table_mhop_put(struct hns_roce_dev *hr_dev, } /* - * free buffer space chunk for QPC/MTPT/CQC/SRQC. + * free buffer space chunk for QPC/MTPT/CQC/SRQC/SCCC. * free bt space chunk for MTT/CQE. */ hns_roce_free_hem(hr_dev, table->hem[hem_idx]); @@ -735,6 +763,7 @@ void hns_roce_table_put(struct hns_roce_dev *hr_dev, mutex_unlock(&table->mutex); } +EXPORT_SYMBOL_GPL(hns_roce_table_put); void *hns_roce_table_find(struct hns_roce_dev *hr_dev, struct hns_roce_hem_table *table, @@ -904,6 +933,30 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev, num_bt_l0 = hr_dev->caps.cqc_bt_num; hop_num = hr_dev->caps.cqc_hop_num; break; + case HEM_TYPE_SCCC: + buf_chunk_size = 1 << (hr_dev->caps.sccc_buf_pg_sz + + PAGE_SHIFT); + bt_chunk_size = 1 << (hr_dev->caps.sccc_ba_pg_sz + + PAGE_SHIFT); + num_bt_l0 = hr_dev->caps.sccc_bt_num; + hop_num = hr_dev->caps.sccc_hop_num; + break; + case HEM_TYPE_QPC_TIMER: + buf_chunk_size = 1 << (hr_dev->caps.qpc_timer_buf_pg_sz + + PAGE_SHIFT); + bt_chunk_size = 1 << (hr_dev->caps.qpc_timer_ba_pg_sz + + PAGE_SHIFT); + num_bt_l0 = hr_dev->caps.qpc_timer_bt_num; + hop_num = hr_dev->caps.qpc_timer_hop_num; + break; + case HEM_TYPE_CQC_TIMER: + buf_chunk_size = 1 << (hr_dev->caps.cqc_timer_buf_pg_sz + + PAGE_SHIFT); + bt_chunk_size = 1 << (hr_dev->caps.cqc_timer_ba_pg_sz + + PAGE_SHIFT); + num_bt_l0 = hr_dev->caps.cqc_timer_bt_num; + hop_num = hr_dev->caps.cqc_timer_hop_num; + break; case HEM_TYPE_SRQC: buf_chunk_size = 1 << (hr_dev->caps.srqc_buf_pg_sz + PAGE_SHIFT); @@ -1081,6 +1134,15 @@ void hns_roce_cleanup_hem(struct hns_roce_dev *hr_dev) hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table); hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table); + if (hr_dev->caps.qpc_timer_entry_sz) + hns_roce_cleanup_hem_table(hr_dev, + &hr_dev->qpc_timer_table); + if (hr_dev->caps.cqc_timer_entry_sz) + hns_roce_cleanup_hem_table(hr_dev, + &hr_dev->cqc_timer_table); + if (hr_dev->caps.sccc_entry_sz) + hns_roce_cleanup_hem_table(hr_dev, + &hr_dev->qp_table.sccc_table); if (hr_dev->caps.trrl_entry_sz) hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.trrl_table); diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.h b/drivers/infiniband/hw/hns/hns_roce_hem.h index a650278c6fbd..d9d668992e49 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hem.h +++ b/drivers/infiniband/hw/hns/hns_roce_hem.h @@ -44,6 +44,9 @@ enum { HEM_TYPE_MTPT, HEM_TYPE_CQC, HEM_TYPE_SRQC, + HEM_TYPE_SCCC, + HEM_TYPE_QPC_TIMER, + HEM_TYPE_CQC_TIMER, /* UNMAP HEM */ HEM_TYPE_MTT, diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c index b74c742b000c..97515c340134 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c @@ -711,13 +711,14 @@ static int hns_roce_v1_rsv_lp_qp(struct hns_roce_dev *hr_dev) struct ib_qp_attr attr = { 0 }; struct hns_roce_v1_priv *priv; struct hns_roce_qp *hr_qp; + struct ib_device *ibdev; struct ib_cq *cq; struct ib_pd *pd; union ib_gid dgid; u64 subnet_prefix; int attr_mask = 0; + int ret = -ENOMEM; int i, j; - int ret; u8 queue_en[HNS_ROCE_V1_RESV_QP] = { 0 }; u8 phy_port; u8 port = 0; @@ -742,12 +743,16 @@ static int hns_roce_v1_rsv_lp_qp(struct hns_roce_dev *hr_dev) free_mr->mr_free_cq->ib_cq.cq_context = NULL; atomic_set(&free_mr->mr_free_cq->ib_cq.usecnt, 0); - pd = hns_roce_alloc_pd(&hr_dev->ib_dev, NULL, NULL); - if (IS_ERR(pd)) { - dev_err(dev, "Create pd for reserved loop qp failed!"); - ret = -ENOMEM; + ibdev = &hr_dev->ib_dev; + pd = rdma_zalloc_drv_obj(ibdev, ib_pd); + if (!pd) + goto alloc_mem_failed; + + pd->device = ibdev; + ret = hns_roce_alloc_pd(pd, NULL, NULL); + if (ret) goto alloc_pd_failed; - } + free_mr->mr_free_pd = to_hr_pd(pd); free_mr->mr_free_pd->ibpd.device = &hr_dev->ib_dev; free_mr->mr_free_pd->ibpd.uobject = NULL; @@ -854,10 +859,12 @@ create_lp_qp_failed: dev_err(dev, "Destroy qp %d for mr free failed!\n", i); } - if (hns_roce_dealloc_pd(pd)) - dev_err(dev, "Destroy pd for create_lp_qp failed!\n"); + hns_roce_dealloc_pd(pd); alloc_pd_failed: + kfree(pd); + +alloc_mem_failed: if (hns_roce_ib_destroy_cq(cq)) dev_err(dev, "Destroy cq for create_lp_qp failed!\n"); @@ -891,9 +898,7 @@ static void hns_roce_v1_release_lp_qp(struct hns_roce_dev *hr_dev) if (ret) dev_err(dev, "Destroy cq for mr_free failed(%d)!\n", ret); - ret = hns_roce_dealloc_pd(&free_mr->mr_free_pd->ibpd); - if (ret) - dev_err(dev, "Destroy pd for mr_free failed(%d)!\n", ret); + hns_roce_dealloc_pd(&free_mr->mr_free_pd->ibpd); } static int hns_roce_db_init(struct hns_roce_dev *hr_dev) @@ -1866,9 +1871,8 @@ static int hns_roce_v1_write_mtpt(void *mb_buf, struct hns_roce_mr *mr, unsigned long mtpt_idx) { struct hns_roce_v1_mpt_entry *mpt_entry; - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; u64 *pages; - int entry; int i; /* MPT filled into mailbox buf */ @@ -1923,8 +1927,8 @@ static int hns_roce_v1_write_mtpt(void *mb_buf, struct hns_roce_mr *mr, return -ENOMEM; i = 0; - for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) { - pages[i] = ((u64)sg_dma_address(sg)) >> 12; + for_each_sg_dma_page(mr->umem->sg_head.sgl, &sg_iter, mr->umem->nmap, 0) { + pages[i] = ((u64)sg_page_iter_dma_address(&sg_iter)) >> 12; /* Directly record to MTPT table firstly 7 entry */ if (i >= HNS_ROCE_MAX_INNER_MTPT_NUM) @@ -5002,7 +5006,7 @@ static int hns_roce_probe(struct platform_device *pdev) struct hns_roce_dev *hr_dev; struct device *dev = &pdev->dev; - hr_dev = (struct hns_roce_dev *)ib_alloc_device(sizeof(*hr_dev)); + hr_dev = ib_alloc_device(hns_roce_dev, ib_dev); if (!hr_dev) return -ENOMEM; diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index 543fa1504cd3..1c54390e1c85 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -587,7 +587,7 @@ out: roce_set_field(sq_db.parameter, V2_DB_PARAMETER_SL_M, V2_DB_PARAMETER_SL_S, qp->sl); - hns_roce_write64_k((__le32 *)&sq_db, qp->sq.db_reg_l); + hns_roce_write64(hr_dev, (__le32 *)&sq_db, qp->sq.db_reg_l); qp->sq_next_wqe = ind; qp->next_sge = sge_ind; @@ -712,6 +712,113 @@ out: return ret; } +static int hns_roce_v2_cmd_hw_reseted(struct hns_roce_dev *hr_dev, + unsigned long instance_stage, + unsigned long reset_stage) +{ + /* When hardware reset has been completed once or more, we should stop + * sending mailbox&cmq&doorbell to hardware. If now in .init_instance() + * function, we should exit with error. If now at HNAE3_INIT_CLIENT + * stage of soft reset process, we should exit with error, and then + * HNAE3_INIT_CLIENT related process can rollback the operation like + * notifing hardware to free resources, HNAE3_INIT_CLIENT related + * process will exit with error to notify NIC driver to reschedule soft + * reset process once again. + */ + hr_dev->is_reset = true; + hr_dev->dis_db = true; + + if (reset_stage == HNS_ROCE_STATE_RST_INIT || + instance_stage == HNS_ROCE_STATE_INIT) + return CMD_RST_PRC_EBUSY; + + return CMD_RST_PRC_SUCCESS; +} + +static int hns_roce_v2_cmd_hw_resetting(struct hns_roce_dev *hr_dev, + unsigned long instance_stage, + unsigned long reset_stage) +{ + struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv; + struct hnae3_handle *handle = priv->handle; + const struct hnae3_ae_ops *ops = handle->ae_algo->ops; + + /* When hardware reset is detected, we should stop sending mailbox&cmq& + * doorbell to hardware. If now in .init_instance() function, we should + * exit with error. If now at HNAE3_INIT_CLIENT stage of soft reset + * process, we should exit with error, and then HNAE3_INIT_CLIENT + * related process can rollback the operation like notifing hardware to + * free resources, HNAE3_INIT_CLIENT related process will exit with + * error to notify NIC driver to reschedule soft reset process once + * again. + */ + hr_dev->dis_db = true; + if (!ops->get_hw_reset_stat(handle)) + hr_dev->is_reset = true; + + if (!hr_dev->is_reset || reset_stage == HNS_ROCE_STATE_RST_INIT || + instance_stage == HNS_ROCE_STATE_INIT) + return CMD_RST_PRC_EBUSY; + + return CMD_RST_PRC_SUCCESS; +} + +static int hns_roce_v2_cmd_sw_resetting(struct hns_roce_dev *hr_dev) +{ + struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv; + struct hnae3_handle *handle = priv->handle; + const struct hnae3_ae_ops *ops = handle->ae_algo->ops; + + /* When software reset is detected at .init_instance() function, we + * should stop sending mailbox&cmq&doorbell to hardware, and exit + * with error. + */ + hr_dev->dis_db = true; + if (ops->ae_dev_reset_cnt(handle) != hr_dev->reset_cnt) + hr_dev->is_reset = true; + + return CMD_RST_PRC_EBUSY; +} + +static int hns_roce_v2_rst_process_cmd(struct hns_roce_dev *hr_dev) +{ + struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv; + struct hnae3_handle *handle = priv->handle; + const struct hnae3_ae_ops *ops = handle->ae_algo->ops; + unsigned long instance_stage; /* the current instance stage */ + unsigned long reset_stage; /* the current reset stage */ + unsigned long reset_cnt; + bool sw_resetting; + bool hw_resetting; + + if (hr_dev->is_reset) + return CMD_RST_PRC_SUCCESS; + + /* Get information about reset from NIC driver or RoCE driver itself, + * the meaning of the following variables from NIC driver are described + * as below: + * reset_cnt -- The count value of completed hardware reset. + * hw_resetting -- Whether hardware device is resetting now. + * sw_resetting -- Whether NIC's software reset process is running now. + */ + instance_stage = handle->rinfo.instance_state; + reset_stage = handle->rinfo.reset_state; + reset_cnt = ops->ae_dev_reset_cnt(handle); + hw_resetting = ops->get_hw_reset_stat(handle); + sw_resetting = ops->ae_dev_resetting(handle); + + if (reset_cnt != hr_dev->reset_cnt) + return hns_roce_v2_cmd_hw_reseted(hr_dev, instance_stage, + reset_stage); + else if (hw_resetting) + return hns_roce_v2_cmd_hw_resetting(hr_dev, instance_stage, + reset_stage); + else if (sw_resetting && instance_stage == HNS_ROCE_STATE_INIT) + return hns_roce_v2_cmd_sw_resetting(hr_dev); + + return 0; +} + static int hns_roce_cmq_space(struct hns_roce_v2_cmq_ring *ring) { int ntu = ring->next_to_use; @@ -892,8 +999,8 @@ static int hns_roce_cmq_csq_clean(struct hns_roce_dev *hr_dev) return clean; } -static int hns_roce_cmq_send(struct hns_roce_dev *hr_dev, - struct hns_roce_cmq_desc *desc, int num) +static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev, + struct hns_roce_cmq_desc *desc, int num) { struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv; struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq; @@ -905,9 +1012,6 @@ static int hns_roce_cmq_send(struct hns_roce_dev *hr_dev, int ret = 0; int ntc; - if (hr_dev->is_reset) - return 0; - spin_lock_bh(&csq->lock); if (num > hns_roce_cmq_space(csq)) { @@ -982,6 +1086,30 @@ static int hns_roce_cmq_send(struct hns_roce_dev *hr_dev, return ret; } +int hns_roce_cmq_send(struct hns_roce_dev *hr_dev, + struct hns_roce_cmq_desc *desc, int num) +{ + int retval; + int ret; + + ret = hns_roce_v2_rst_process_cmd(hr_dev); + if (ret == CMD_RST_PRC_SUCCESS) + return 0; + if (ret == CMD_RST_PRC_EBUSY) + return ret; + + ret = __hns_roce_cmq_send(hr_dev, desc, num); + if (ret) { + retval = hns_roce_v2_rst_process_cmd(hr_dev); + if (retval == CMD_RST_PRC_SUCCESS) + return 0; + else if (retval == CMD_RST_PRC_EBUSY) + return retval; + } + + return ret; +} + static int hns_roce_cmq_query_hw_info(struct hns_roce_dev *hr_dev) { struct hns_roce_query_version *resp; @@ -1078,6 +1206,44 @@ static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev) hr_dev->caps.sl_num = roce_get_field(req_b->qid_idx_sl_num, PF_RES_DATA_3_PF_SL_NUM_M, PF_RES_DATA_3_PF_SL_NUM_S); + hr_dev->caps.sccc_bt_num = roce_get_field(req_b->sccc_bt_idx_num, + PF_RES_DATA_4_PF_SCCC_BT_NUM_M, + PF_RES_DATA_4_PF_SCCC_BT_NUM_S); + + return 0; +} + +static int hns_roce_query_pf_timer_resource(struct hns_roce_dev *hr_dev) +{ + struct hns_roce_pf_timer_res_a *req_a; + struct hns_roce_cmq_desc desc[2]; + int ret, i; + + for (i = 0; i < 2; i++) { + hns_roce_cmq_setup_basic_desc(&desc[i], + HNS_ROCE_OPC_QUERY_PF_TIMER_RES, + true); + + if (i == 0) + desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT); + else + desc[i].flag &= ~cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT); + } + + ret = hns_roce_cmq_send(hr_dev, desc, 2); + if (ret) + return ret; + + req_a = (struct hns_roce_pf_timer_res_a *)desc[0].data; + + hr_dev->caps.qpc_timer_bt_num = + roce_get_field(req_a->qpc_timer_bt_idx_num, + PF_RES_DATA_1_PF_QPC_TIMER_BT_NUM_M, + PF_RES_DATA_1_PF_QPC_TIMER_BT_NUM_S); + hr_dev->caps.cqc_timer_bt_num = + roce_get_field(req_a->cqc_timer_bt_idx_num, + PF_RES_DATA_2_PF_CQC_TIMER_BT_NUM_M, + PF_RES_DATA_2_PF_CQC_TIMER_BT_NUM_S); return 0; } @@ -1193,6 +1359,14 @@ static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev) VF_RES_B_DATA_3_VF_SL_NUM_M, VF_RES_B_DATA_3_VF_SL_NUM_S, HNS_ROCE_VF_SL_NUM); + + roce_set_field(req_b->vf_sccc_idx_num, + VF_RES_B_DATA_4_VF_SCCC_BT_IDX_M, + VF_RES_B_DATA_4_VF_SCCC_BT_IDX_S, 0); + roce_set_field(req_b->vf_sccc_idx_num, + VF_RES_B_DATA_4_VF_SCCC_BT_NUM_M, + VF_RES_B_DATA_4_VF_SCCC_BT_NUM_S, + HNS_ROCE_VF_SCCC_BT_NUM); } } @@ -1205,6 +1379,7 @@ static int hns_roce_v2_set_bt(struct hns_roce_dev *hr_dev) u8 qpc_hop_num = hr_dev->caps.qpc_hop_num; u8 cqc_hop_num = hr_dev->caps.cqc_hop_num; u8 mpt_hop_num = hr_dev->caps.mpt_hop_num; + u8 sccc_hop_num = hr_dev->caps.sccc_hop_num; struct hns_roce_cfg_bt_attr *req; struct hns_roce_cmq_desc desc; @@ -1252,6 +1427,20 @@ static int hns_roce_v2_set_bt(struct hns_roce_dev *hr_dev) CFG_BT_ATTR_DATA_3_VF_MPT_HOPNUM_S, mpt_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 : mpt_hop_num); + roce_set_field(req->vf_sccc_cfg, + CFG_BT_ATTR_DATA_4_VF_SCCC_BA_PGSZ_M, + CFG_BT_ATTR_DATA_4_VF_SCCC_BA_PGSZ_S, + hr_dev->caps.sccc_ba_pg_sz + PG_SHIFT_OFFSET); + roce_set_field(req->vf_sccc_cfg, + CFG_BT_ATTR_DATA_4_VF_SCCC_BUF_PGSZ_M, + CFG_BT_ATTR_DATA_4_VF_SCCC_BUF_PGSZ_S, + hr_dev->caps.sccc_buf_pg_sz + PG_SHIFT_OFFSET); + roce_set_field(req->vf_sccc_cfg, + CFG_BT_ATTR_DATA_4_VF_SCCC_HOPNUM_M, + CFG_BT_ATTR_DATA_4_VF_SCCC_HOPNUM_S, + sccc_hop_num == + HNS_ROCE_HOP_NUM_0 ? 0 : sccc_hop_num); + return hns_roce_cmq_send(hr_dev, &desc, 1); } @@ -1289,6 +1478,16 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev) return ret; } + if (hr_dev->pci_dev->revision == 0x21) { + ret = hns_roce_query_pf_timer_resource(hr_dev); + if (ret) { + dev_err(hr_dev->dev, + "Query pf timer resource fail, ret = %d.\n", + ret); + return ret; + } + } + ret = hns_roce_alloc_vf_resource(hr_dev); if (ret) { dev_err(hr_dev->dev, "Allocate vf resource fail, ret = %d.\n", @@ -1313,6 +1512,7 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev) caps->max_wqes = HNS_ROCE_V2_MAX_WQE_NUM; caps->num_cqs = HNS_ROCE_V2_MAX_CQ_NUM; caps->num_srqs = HNS_ROCE_V2_MAX_SRQ_NUM; + caps->min_cqes = HNS_ROCE_MIN_CQE_NUM; caps->max_cqes = HNS_ROCE_V2_MAX_CQE_NUM; caps->max_srqwqes = HNS_ROCE_V2_MAX_SRQWQE_NUM; caps->max_sq_sg = HNS_ROCE_V2_MAX_SQ_SGE_NUM; @@ -1366,7 +1566,7 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev) caps->mpt_ba_pg_sz = 0; caps->mpt_buf_pg_sz = 0; caps->mpt_hop_num = HNS_ROCE_CONTEXT_HOP_NUM; - caps->pbl_ba_pg_sz = 0; + caps->pbl_ba_pg_sz = 2; caps->pbl_buf_pg_sz = 0; caps->pbl_hop_num = HNS_ROCE_PBL_HOP_NUM; caps->mtt_ba_pg_sz = 0; @@ -1408,9 +1608,27 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev) caps->max_srq_wrs = HNS_ROCE_V2_MAX_SRQ_WR; caps->max_srq_sges = HNS_ROCE_V2_MAX_SRQ_SGE; - if (hr_dev->pci_dev->revision == 0x21) + if (hr_dev->pci_dev->revision == 0x21) { caps->flags |= HNS_ROCE_CAP_FLAG_ATOMIC | - HNS_ROCE_CAP_FLAG_SRQ; + HNS_ROCE_CAP_FLAG_SRQ | + HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL; + + caps->num_qpc_timer = HNS_ROCE_V2_MAX_QPC_TIMER_NUM; + caps->qpc_timer_entry_sz = HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ; + caps->qpc_timer_ba_pg_sz = 0; + caps->qpc_timer_buf_pg_sz = 0; + caps->qpc_timer_hop_num = HNS_ROCE_HOP_NUM_0; + caps->num_cqc_timer = HNS_ROCE_V2_MAX_CQC_TIMER_NUM; + caps->cqc_timer_entry_sz = HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ; + caps->cqc_timer_ba_pg_sz = 0; + caps->cqc_timer_buf_pg_sz = 0; + caps->cqc_timer_hop_num = HNS_ROCE_HOP_NUM_0; + + caps->sccc_entry_sz = HNS_ROCE_V2_SCCC_ENTRY_SZ; + caps->sccc_ba_pg_sz = 0; + caps->sccc_buf_pg_sz = 0; + caps->sccc_hop_num = HNS_ROCE_SCCC_HOP_NUM; + } ret = hns_roce_v2_set_bt(hr_dev); if (ret) @@ -1611,7 +1829,8 @@ static void hns_roce_free_link_table(struct hns_roce_dev *hr_dev, static int hns_roce_v2_init(struct hns_roce_dev *hr_dev) { struct hns_roce_v2_priv *priv = hr_dev->priv; - int ret; + int qpc_count, cqc_count; + int ret, i; /* TSQ includes SQ doorbell and ack doorbell */ ret = hns_roce_init_link_table(hr_dev, TSQ_LINK_TABLE); @@ -1626,8 +1845,40 @@ static int hns_roce_v2_init(struct hns_roce_dev *hr_dev) goto err_tpq_init_failed; } + /* Alloc memory for QPC Timer buffer space chunk*/ + for (qpc_count = 0; qpc_count < hr_dev->caps.qpc_timer_bt_num; + qpc_count++) { + ret = hns_roce_table_get(hr_dev, &hr_dev->qpc_timer_table, + qpc_count); + if (ret) { + dev_err(hr_dev->dev, "QPC Timer get failed\n"); + goto err_qpc_timer_failed; + } + } + + /* Alloc memory for CQC Timer buffer space chunk*/ + for (cqc_count = 0; cqc_count < hr_dev->caps.cqc_timer_bt_num; + cqc_count++) { + ret = hns_roce_table_get(hr_dev, &hr_dev->cqc_timer_table, + cqc_count); + if (ret) { + dev_err(hr_dev->dev, "CQC Timer get failed\n"); + goto err_cqc_timer_failed; + } + } + return 0; +err_cqc_timer_failed: + for (i = 0; i < cqc_count; i++) + hns_roce_table_put(hr_dev, &hr_dev->cqc_timer_table, i); + +err_qpc_timer_failed: + for (i = 0; i < qpc_count; i++) + hns_roce_table_put(hr_dev, &hr_dev->qpc_timer_table, i); + + hns_roce_free_link_table(hr_dev, &priv->tpq); + err_tpq_init_failed: hns_roce_free_link_table(hr_dev, &priv->tsq); @@ -1735,6 +1986,9 @@ static int hns_roce_v2_chk_mbox(struct hns_roce_dev *hr_dev, status = hns_roce_v2_cmd_complete(hr_dev); if (status != 0x1) { + if (status == CMD_RST_PRC_EBUSY) + return status; + dev_err(dev, "mailbox status 0x%x!\n", status); return -EBUSY; } @@ -1831,12 +2085,10 @@ static int hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port, static int set_mtpt_pbl(struct hns_roce_v2_mpt_entry *mpt_entry, struct hns_roce_mr *mr) { - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; u64 page_addr; u64 *pages; - int i, j; - int len; - int entry; + int i; mpt_entry->pbl_size = cpu_to_le32(mr->pbl_size); mpt_entry->pbl_ba_l = cpu_to_le32(lower_32_bits(mr->pbl_ba >> 3)); @@ -1849,17 +2101,14 @@ static int set_mtpt_pbl(struct hns_roce_v2_mpt_entry *mpt_entry, return -ENOMEM; i = 0; - for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) { - len = sg_dma_len(sg) >> PAGE_SHIFT; - for (j = 0; j < len; ++j) { - page_addr = sg_dma_address(sg) + - (j << mr->umem->page_shift); - pages[i] = page_addr >> 6; - /* Record the first 2 entry directly to MTPT table */ - if (i >= HNS_ROCE_V2_MAX_INNER_MTPT_NUM - 1) - goto found; - i++; - } + for_each_sg_dma_page(mr->umem->sg_head.sgl, &sg_iter, mr->umem->nmap, 0) { + page_addr = sg_page_iter_dma_address(&sg_iter); + pages[i] = page_addr >> 6; + + /* Record the first 2 entry directly to MTPT table */ + if (i >= HNS_ROCE_V2_MAX_INNER_MTPT_NUM - 1) + goto found; + i++; } found: mpt_entry->pa0_l = cpu_to_le32(lower_32_bits(pages[0])); @@ -1941,6 +2190,9 @@ static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev, struct hns_roce_v2_mpt_entry *mpt_entry = mb_buf; int ret = 0; + roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_MPT_ST_M, + V2_MPT_BYTE_4_MPT_ST_S, V2_MPT_ST_VALID); + if (flags & IB_MR_REREG_PD) { roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PD_M, V2_MPT_BYTE_4_PD_S, pdn); @@ -2245,6 +2497,7 @@ static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev, static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags) { + struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device); struct hns_roce_cq *hr_cq = to_hr_cq(ibcq); u32 notification_flag; u32 doorbell[2]; @@ -2270,7 +2523,7 @@ static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq, roce_set_bit(doorbell[1], V2_CQ_DB_PARAMETER_NOTIFY_S, notification_flag); - hns_roce_write64_k(doorbell, hr_cq->cq_db_l); + hns_roce_write64(hr_dev, doorbell, hr_cq->cq_db_l); return 0; } @@ -2663,17 +2916,33 @@ static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev, case HEM_TYPE_SRQC: op = HNS_ROCE_CMD_WRITE_SRQC_BT0; break; + case HEM_TYPE_SCCC: + op = HNS_ROCE_CMD_WRITE_SCCC_BT0; + break; + case HEM_TYPE_QPC_TIMER: + op = HNS_ROCE_CMD_WRITE_QPC_TIMER_BT0; + break; + case HEM_TYPE_CQC_TIMER: + op = HNS_ROCE_CMD_WRITE_CQC_TIMER_BT0; + break; default: dev_warn(dev, "Table %d not to be written by mailbox!\n", table->type); return 0; } + + if (table->type == HEM_TYPE_SCCC && step_idx) + return 0; + op += step_idx; mailbox = hns_roce_alloc_cmd_mailbox(hr_dev); if (IS_ERR(mailbox)) return PTR_ERR(mailbox); + if (table->type == HEM_TYPE_SCCC) + obj = mhop.l0_idx; + if (check_whether_last_step(hop_num, step_idx)) { hem = table->hem[hem_idx]; for (hns_roce_hem_first(hem, &iter); @@ -2722,6 +2991,10 @@ static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev, case HEM_TYPE_CQC: op = HNS_ROCE_CMD_DESTROY_CQC_BT0; break; + case HEM_TYPE_SCCC: + case HEM_TYPE_QPC_TIMER: + case HEM_TYPE_CQC_TIMER: + break; case HEM_TYPE_SRQC: op = HNS_ROCE_CMD_DESTROY_SRQC_BT0; break; @@ -2730,6 +3003,12 @@ static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev, table->type); return 0; } + + if (table->type == HEM_TYPE_SCCC || + table->type == HEM_TYPE_QPC_TIMER || + table->type == HEM_TYPE_CQC_TIMER) + return 0; + op += step_idx; mailbox = hns_roce_alloc_cmd_mailbox(hr_dev); @@ -3686,10 +3965,16 @@ static int modify_qp_rtr_to_rts(struct ib_qp *ibqp, V2_QPC_BYTE_212_LSN_S, 0); if (attr_mask & IB_QP_TIMEOUT) { - roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_AT_M, - V2_QPC_BYTE_28_AT_S, attr->timeout); - roce_set_field(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_AT_M, - V2_QPC_BYTE_28_AT_S, 0); + if (attr->timeout < 31) { + roce_set_field(context->byte_28_at_fl, + V2_QPC_BYTE_28_AT_M, V2_QPC_BYTE_28_AT_S, + attr->timeout); + roce_set_field(qpc_mask->byte_28_at_fl, + V2_QPC_BYTE_28_AT_M, V2_QPC_BYTE_28_AT_S, + 0); + } else { + dev_warn(dev, "Local ACK timeout shall be 0 to 30.\n"); + } } roce_set_field(context->byte_172_sq_psn, V2_QPC_BYTE_172_SQ_CUR_PSN_M, @@ -3742,7 +4027,7 @@ static int hns_roce_v2_modify_qp(struct ib_qp *ibqp, struct device *dev = hr_dev->dev; int ret = -EINVAL; - context = kcalloc(2, sizeof(*context), GFP_KERNEL); + context = kcalloc(2, sizeof(*context), GFP_ATOMIC); if (!context) return -ENOMEM; @@ -3789,13 +4074,16 @@ static int hns_roce_v2_modify_qp(struct ib_qp *ibqp, roce_set_field(qpc_mask->byte_160_sq_ci_pi, V2_QPC_BYTE_160_SQ_PRODUCER_IDX_M, V2_QPC_BYTE_160_SQ_PRODUCER_IDX_S, 0); - roce_set_field(context->byte_84_rq_ci_pi, + + if (!ibqp->srq) { + roce_set_field(context->byte_84_rq_ci_pi, V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M, V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S, hr_qp->rq.head); - roce_set_field(qpc_mask->byte_84_rq_ci_pi, + roce_set_field(qpc_mask->byte_84_rq_ci_pi, V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M, V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S, 0); + } } if (attr_mask & IB_QP_AV) { @@ -4224,6 +4512,59 @@ static int hns_roce_v2_destroy_qp(struct ib_qp *ibqp) return 0; } +static int hns_roce_v2_qp_flow_control_init(struct hns_roce_dev *hr_dev, + struct hns_roce_qp *hr_qp) +{ + struct hns_roce_sccc_clr_done *resp; + struct hns_roce_sccc_clr *clr; + struct hns_roce_cmq_desc desc; + int ret, i; + + mutex_lock(&hr_dev->qp_table.scc_mutex); + + /* set scc ctx clear done flag */ + hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_RESET_SCCC, false); + ret = hns_roce_cmq_send(hr_dev, &desc, 1); + if (ret) { + dev_err(hr_dev->dev, "Reset SCC ctx failed(%d)\n", ret); + goto out; + } + + /* clear scc context */ + hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CLR_SCCC, false); + clr = (struct hns_roce_sccc_clr *)desc.data; + clr->qpn = cpu_to_le32(hr_qp->qpn); + ret = hns_roce_cmq_send(hr_dev, &desc, 1); + if (ret) { + dev_err(hr_dev->dev, "Clear SCC ctx failed(%d)\n", ret); + goto out; + } + + /* query scc context clear is done or not */ + resp = (struct hns_roce_sccc_clr_done *)desc.data; + for (i = 0; i <= HNS_ROCE_CMQ_SCC_CLR_DONE_CNT; i++) { + hns_roce_cmq_setup_basic_desc(&desc, + HNS_ROCE_OPC_QUERY_SCCC, true); + ret = hns_roce_cmq_send(hr_dev, &desc, 1); + if (ret) { + dev_err(hr_dev->dev, "Query clr cmq failed(%d)\n", ret); + goto out; + } + + if (resp->clr_done) + goto out; + + msleep(20); + } + + dev_err(hr_dev->dev, "Query SCC clr done flag overtime.\n"); + ret = -ETIMEDOUT; + +out: + mutex_unlock(&hr_dev->qp_table.scc_mutex); + return ret; +} + static int hns_roce_v2_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period) { struct hns_roce_dev *hr_dev = to_hr_dev(cq->device); @@ -4281,7 +4622,8 @@ static void hns_roce_set_qps_to_err(struct hns_roce_dev *hr_dev, u32 qpn) if (hr_qp->ibqp.uobject) { if (hr_qp->sdb_en == 1) { hr_qp->sq.head = *(int *)(hr_qp->sdb.virt_addr); - hr_qp->rq.head = *(int *)(hr_qp->rdb.virt_addr); + if (hr_qp->rdb_en == 1) + hr_qp->rq.head = *(int *)(hr_qp->rdb.virt_addr); } else { dev_warn(hr_dev->dev, "flush cqe is unsupported in userspace!\n"); return; @@ -4319,64 +4661,19 @@ static void hns_roce_irq_work_handle(struct work_struct *work) dev_warn(dev, "Send queue drained.\n"); break; case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR: - dev_err(dev, "Local work queue catastrophic error.\n"); + dev_err(dev, "Local work queue 0x%x catas error, sub_type:%d\n", + qpn, irq_work->sub_type); hns_roce_set_qps_to_err(irq_work->hr_dev, qpn); - switch (irq_work->sub_type) { - case HNS_ROCE_LWQCE_QPC_ERROR: - dev_err(dev, "QP %d, QPC error.\n", qpn); - break; - case HNS_ROCE_LWQCE_MTU_ERROR: - dev_err(dev, "QP %d, MTU error.\n", qpn); - break; - case HNS_ROCE_LWQCE_WQE_BA_ADDR_ERROR: - dev_err(dev, "QP %d, WQE BA addr error.\n", qpn); - break; - case HNS_ROCE_LWQCE_WQE_ADDR_ERROR: - dev_err(dev, "QP %d, WQE addr error.\n", qpn); - break; - case HNS_ROCE_LWQCE_SQ_WQE_SHIFT_ERROR: - dev_err(dev, "QP %d, WQE shift error.\n", qpn); - break; - default: - dev_err(dev, "Unhandled sub_event type %d.\n", - irq_work->sub_type); - break; - } break; case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR: - dev_err(dev, "Invalid request local work queue error.\n"); + dev_err(dev, "Invalid request local work queue 0x%x error.\n", + qpn); hns_roce_set_qps_to_err(irq_work->hr_dev, qpn); break; case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR: - dev_err(dev, "Local access violation work queue error.\n"); + dev_err(dev, "Local access violation work queue 0x%x error, sub_type:%d\n", + qpn, irq_work->sub_type); hns_roce_set_qps_to_err(irq_work->hr_dev, qpn); - switch (irq_work->sub_type) { - case HNS_ROCE_LAVWQE_R_KEY_VIOLATION: - dev_err(dev, "QP %d, R_key violation.\n", qpn); - break; - case HNS_ROCE_LAVWQE_LENGTH_ERROR: - dev_err(dev, "QP %d, length error.\n", qpn); - break; - case HNS_ROCE_LAVWQE_VA_ERROR: - dev_err(dev, "QP %d, VA error.\n", qpn); - break; - case HNS_ROCE_LAVWQE_PD_ERROR: - dev_err(dev, "QP %d, PD error.\n", qpn); - break; - case HNS_ROCE_LAVWQE_RW_ACC_ERROR: - dev_err(dev, "QP %d, rw acc error.\n", qpn); - break; - case HNS_ROCE_LAVWQE_KEY_STATE_ERROR: - dev_err(dev, "QP %d, key state error.\n", qpn); - break; - case HNS_ROCE_LAVWQE_MR_OPERATION_ERROR: - dev_err(dev, "QP %d, MR operation error.\n", qpn); - break; - default: - dev_err(dev, "Unhandled sub_event type %d.\n", - irq_work->sub_type); - break; - } break; case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH: dev_warn(dev, "SRQ limit reach.\n"); @@ -4427,6 +4724,7 @@ static void hns_roce_v2_init_irq_work(struct hns_roce_dev *hr_dev, static void set_eq_cons_index_v2(struct hns_roce_eq *eq) { + struct hns_roce_dev *hr_dev = eq->hr_dev; u32 doorbell[2]; doorbell[0] = 0; @@ -4453,7 +4751,7 @@ static void set_eq_cons_index_v2(struct hns_roce_eq *eq) HNS_ROCE_V2_EQ_DB_PARA_S, (eq->cons_index & HNS_ROCE_V2_CONS_IDX_M)); - hns_roce_write64_k(doorbell, eq->doorbell); + hns_roce_write64(hr_dev, doorbell, eq->doorbell); } static struct hns_roce_aeqe *get_aeqe_v2(struct hns_roce_eq *eq, u32 entry) @@ -4568,7 +4866,7 @@ static int hns_roce_v2_aeq_int(struct hns_roce_dev *hr_dev, dev_err(dev, "Unhandled event %d on EQ %d at idx %u.\n", event_type, eq->eqn, eq->cons_index); break; - }; + } eq->event_type = event_type; eq->sub_type = sub_type; @@ -4692,11 +4990,22 @@ static irqreturn_t hns_roce_v2_msix_interrupt_abn(int irq, void *dev_id) int_en = roce_read(hr_dev, ROCEE_VF_ABN_INT_EN_REG); if (roce_get_bit(int_st, HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S)) { + struct pci_dev *pdev = hr_dev->pci_dev; + struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev); + const struct hnae3_ae_ops *ops = ae_dev->ops; + dev_err(dev, "AEQ overflow!\n"); roce_set_bit(int_st, HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S, 1); roce_write(hr_dev, ROCEE_VF_ABN_INT_ST_REG, int_st); + /* Set reset level for reset_event() */ + if (ops->set_default_reset_request) + ops->set_default_reset_request(ae_dev, + HNAE3_FUNC_RESET); + if (ops->reset_event) + ops->reset_event(pdev, NULL); + roce_set_bit(int_en, HNS_ROCE_V2_VF_ABN_INT_EN_S, 1); roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, int_en); @@ -5599,7 +5908,7 @@ static int hns_roce_v2_modify_srq(struct ib_srq *ibsrq, return 0; } -int hns_roce_v2_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr) +static int hns_roce_v2_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr) { struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device); struct hns_roce_srq *srq = to_hr_srq(ibsrq); @@ -5664,6 +5973,7 @@ static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr, const struct ib_recv_wr **bad_wr) { + struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device); struct hns_roce_srq *srq = to_hr_srq(ibsrq); struct hns_roce_v2_wqe_data_seg *dseg; struct hns_roce_v2_db srq_db; @@ -5725,7 +6035,7 @@ static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq, srq_db.byte_4 = HNS_ROCE_V2_SRQ_DB << 24 | srq->srqn; srq_db.parameter = srq->head; - hns_roce_write64_k((__le32 *)&srq_db, srq->db_reg_l); + hns_roce_write64(hr_dev, (__le32 *)&srq_db, srq->db_reg_l); } @@ -5758,6 +6068,7 @@ static const struct hns_roce_hw hns_roce_hw_v2 = { .hw_exit = hns_roce_v2_exit, .post_mbox = hns_roce_v2_post_mbox, .chk_mbox = hns_roce_v2_chk_mbox, + .rst_prc_mbox = hns_roce_v2_rst_process_cmd, .set_gid = hns_roce_v2_set_gid, .set_mac = hns_roce_v2_set_mac, .write_mtpt = hns_roce_v2_write_mtpt, @@ -5770,6 +6081,7 @@ static const struct hns_roce_hw hns_roce_hw_v2 = { .modify_qp = hns_roce_v2_modify_qp, .query_qp = hns_roce_v2_query_qp, .destroy_qp = hns_roce_v2_destroy_qp, + .qp_flow_control_init = hns_roce_v2_qp_flow_control_init, .modify_cq = hns_roce_v2_modify_cq, .post_send = hns_roce_v2_post_send, .post_recv = hns_roce_v2_post_recv, @@ -5800,6 +6112,7 @@ MODULE_DEVICE_TABLE(pci, hns_roce_hw_v2_pci_tbl); static int hns_roce_hw_v2_get_cfg(struct hns_roce_dev *hr_dev, struct hnae3_handle *handle) { + struct hns_roce_v2_priv *priv = hr_dev->priv; const struct pci_device_id *id; int i; @@ -5830,15 +6143,18 @@ static int hns_roce_hw_v2_get_cfg(struct hns_roce_dev *hr_dev, hr_dev->cmd_mod = 1; hr_dev->loop_idc = 0; + hr_dev->reset_cnt = handle->ae_algo->ops->ae_dev_reset_cnt(handle); + priv->handle = handle; + return 0; } -static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle) +static int __hns_roce_hw_v2_init_instance(struct hnae3_handle *handle) { struct hns_roce_dev *hr_dev; int ret; - hr_dev = (struct hns_roce_dev *)ib_alloc_device(sizeof(*hr_dev)); + hr_dev = ib_alloc_device(hns_roce_dev, ib_dev); if (!hr_dev) return -ENOMEM; @@ -5850,7 +6166,6 @@ static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle) hr_dev->pci_dev = handle->pdev; hr_dev->dev = &handle->pdev->dev; - handle->priv = hr_dev; ret = hns_roce_hw_v2_get_cfg(hr_dev, handle); if (ret) { @@ -5864,6 +6179,8 @@ static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle) goto error_failed_get_cfg; } + handle->priv = hr_dev; + return 0; error_failed_get_cfg: @@ -5875,7 +6192,7 @@ error_failed_kzalloc: return ret; } -static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle, +static void __hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle, bool reset) { struct hns_roce_dev *hr_dev = (struct hns_roce_dev *)handle->priv; @@ -5883,24 +6200,79 @@ static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle, if (!hr_dev) return; + handle->priv = NULL; hns_roce_exit(hr_dev); kfree(hr_dev->priv); ib_dealloc_device(&hr_dev->ib_dev); } +static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle) +{ + const struct hnae3_ae_ops *ops = handle->ae_algo->ops; + struct device *dev = &handle->pdev->dev; + int ret; + + handle->rinfo.instance_state = HNS_ROCE_STATE_INIT; + + if (ops->ae_dev_resetting(handle) || ops->get_hw_reset_stat(handle)) { + handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT; + goto reset_chk_err; + } + + ret = __hns_roce_hw_v2_init_instance(handle); + if (ret) { + handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT; + dev_err(dev, "RoCE instance init failed! ret = %d\n", ret); + if (ops->ae_dev_resetting(handle) || + ops->get_hw_reset_stat(handle)) + goto reset_chk_err; + else + return ret; + } + + handle->rinfo.instance_state = HNS_ROCE_STATE_INITED; + + + return 0; + +reset_chk_err: + dev_err(dev, "Device is busy in resetting state.\n" + "please retry later.\n"); + + return -EBUSY; +} + +static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle, + bool reset) +{ + if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED) + return; + + handle->rinfo.instance_state = HNS_ROCE_STATE_UNINIT; + + __hns_roce_hw_v2_uninit_instance(handle, reset); + + handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT; +} static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle) { - struct hns_roce_dev *hr_dev = (struct hns_roce_dev *)handle->priv; + struct hns_roce_dev *hr_dev; struct ib_event event; - if (!hr_dev) { - dev_err(&handle->pdev->dev, - "Input parameter handle->priv is NULL!\n"); - return -EINVAL; + if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED) { + set_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state); + return 0; } + handle->rinfo.reset_state = HNS_ROCE_STATE_RST_DOWN; + clear_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state); + + hr_dev = (struct hns_roce_dev *)handle->priv; + if (!hr_dev) + return 0; + hr_dev->active = false; - hr_dev->is_reset = true; + hr_dev->dis_db = true; event.event = IB_EVENT_DEVICE_FATAL; event.device = &hr_dev->ib_dev; @@ -5912,17 +6284,29 @@ static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle) static int hns_roce_hw_v2_reset_notify_init(struct hnae3_handle *handle) { + struct device *dev = &handle->pdev->dev; int ret; - ret = hns_roce_hw_v2_init_instance(handle); + if (test_and_clear_bit(HNS_ROCE_RST_DIRECT_RETURN, + &handle->rinfo.state)) { + handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED; + return 0; + } + + handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INIT; + + dev_info(&handle->pdev->dev, "In reset process RoCE client reinit.\n"); + ret = __hns_roce_hw_v2_init_instance(handle); if (ret) { /* when reset notify type is HNAE3_INIT_CLIENT In reset notify * callback function, RoCE Engine reinitialize. If RoCE reinit * failed, we should inform NIC driver. */ handle->priv = NULL; - dev_err(&handle->pdev->dev, - "In reset process RoCE reinit failed %d.\n", ret); + dev_err(dev, "In reset process RoCE reinit failed %d.\n", ret); + } else { + handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED; + dev_info(dev, "Reset done, RoCE client reinit finished.\n"); } return ret; @@ -5930,8 +6314,14 @@ static int hns_roce_hw_v2_reset_notify_init(struct hnae3_handle *handle) static int hns_roce_hw_v2_reset_notify_uninit(struct hnae3_handle *handle) { + if (test_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state)) + return 0; + + handle->rinfo.reset_state = HNS_ROCE_STATE_RST_UNINIT; + dev_info(&handle->pdev->dev, "In reset process RoCE client uninit.\n"); msleep(100); - hns_roce_hw_v2_uninit_instance(handle, false); + __hns_roce_hw_v2_uninit_instance(handle, false); + return 0; } diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h index b72d0443c835..f1f1b75812f9 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h @@ -36,6 +36,7 @@ #include <linux/bitops.h> #define HNS_ROCE_VF_QPC_BT_NUM 256 +#define HNS_ROCE_VF_SCCC_BT_NUM 64 #define HNS_ROCE_VF_SRQC_BT_NUM 64 #define HNS_ROCE_VF_CQC_BT_NUM 64 #define HNS_ROCE_VF_MPT_BT_NUM 64 @@ -44,12 +45,14 @@ #define HNS_ROCE_VF_SGID_NUM 32 #define HNS_ROCE_VF_SL_NUM 8 -#define HNS_ROCE_V2_MAX_QP_NUM 0x2000 +#define HNS_ROCE_V2_MAX_QP_NUM 0x100000 +#define HNS_ROCE_V2_MAX_QPC_TIMER_NUM 0x200 #define HNS_ROCE_V2_MAX_WQE_NUM 0x8000 #define HNS_ROCE_V2_MAX_SRQ 0x100000 #define HNS_ROCE_V2_MAX_SRQ_WR 0x8000 #define HNS_ROCE_V2_MAX_SRQ_SGE 0x100 -#define HNS_ROCE_V2_MAX_CQ_NUM 0x8000 +#define HNS_ROCE_V2_MAX_CQ_NUM 0x100000 +#define HNS_ROCE_V2_MAX_CQC_TIMER_NUM 0x100 #define HNS_ROCE_V2_MAX_SRQ_NUM 0x100000 #define HNS_ROCE_V2_MAX_CQE_NUM 0x10000 #define HNS_ROCE_V2_MAX_SRQWQE_NUM 0x8000 @@ -64,7 +67,7 @@ #define HNS_ROCE_V2_COMP_VEC_NUM 63 #define HNS_ROCE_V2_AEQE_VEC_NUM 1 #define HNS_ROCE_V2_ABNORMAL_VEC_NUM 1 -#define HNS_ROCE_V2_MAX_MTPT_NUM 0x8000 +#define HNS_ROCE_V2_MAX_MTPT_NUM 0x100000 #define HNS_ROCE_V2_MAX_MTT_SEGS 0x1000000 #define HNS_ROCE_V2_MAX_CQE_SEGS 0x1000000 #define HNS_ROCE_V2_MAX_SRQWQE_SEGS 0x1000000 @@ -83,6 +86,9 @@ #define HNS_ROCE_V2_MTPT_ENTRY_SZ 64 #define HNS_ROCE_V2_MTT_ENTRY_SZ 64 #define HNS_ROCE_V2_CQE_ENTRY_SIZE 32 +#define HNS_ROCE_V2_SCCC_ENTRY_SZ 32 +#define HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ 4096 +#define HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ 4096 #define HNS_ROCE_V2_PAGE_SIZE_SUPPORTED 0xFFFFF000 #define HNS_ROCE_V2_MAX_INNER_MTPT_NUM 2 #define HNS_ROCE_INVALID_LKEY 0x100 @@ -90,7 +96,10 @@ #define HNS_ROCE_V2_UC_RC_SGE_NUM_IN_WQE 2 #define HNS_ROCE_V2_RSV_QPS 8 +#define HNS_ROCE_V2_HW_RST_TIMEOUT 1000 + #define HNS_ROCE_CONTEXT_HOP_NUM 1 +#define HNS_ROCE_SCCC_HOP_NUM 1 #define HNS_ROCE_MTT_HOP_NUM 1 #define HNS_ROCE_CQE_HOP_NUM 1 #define HNS_ROCE_SRQWQE_HOP_NUM 1 @@ -120,6 +129,8 @@ #define HNS_ROCE_CMQ_EN_B 16 #define HNS_ROCE_CMQ_ENABLE BIT(HNS_ROCE_CMQ_EN_B) +#define HNS_ROCE_CMQ_SCC_CLR_DONE_CNT 5 + #define check_whether_last_step(hop_num, step_idx) \ ((step_idx == 0 && hop_num == HNS_ROCE_HOP_NUM_0) || \ (step_idx == 1 && hop_num == 1) || \ @@ -224,11 +235,15 @@ enum hns_roce_opcode_type { HNS_ROCE_OPC_ALLOC_VF_RES = 0x8401, HNS_ROCE_OPC_CFG_EXT_LLM = 0x8403, HNS_ROCE_OPC_CFG_TMOUT_LLM = 0x8404, + HNS_ROCE_OPC_QUERY_PF_TIMER_RES = 0x8406, HNS_ROCE_OPC_CFG_SGID_TB = 0x8500, HNS_ROCE_OPC_CFG_SMAC_TB = 0x8501, HNS_ROCE_OPC_POST_MB = 0x8504, HNS_ROCE_OPC_QUERY_MB_ST = 0x8505, HNS_ROCE_OPC_CFG_BT_ATTR = 0x8506, + HNS_ROCE_OPC_CLR_SCCC = 0x8509, + HNS_ROCE_OPC_QUERY_SCCC = 0x850a, + HNS_ROCE_OPC_RESET_SCCC = 0x850b, HNS_SWITCH_PARAMETER_CFG = 0x1033, }; @@ -1300,7 +1315,8 @@ struct hns_roce_pf_res_b { __le32 smac_idx_num; __le32 sgid_idx_num; __le32 qid_idx_sl_num; - __le32 rsv[2]; + __le32 sccc_bt_idx_num; + __le32 rsv; }; #define PF_RES_DATA_1_PF_SMAC_IDX_S 0 @@ -1321,6 +1337,31 @@ struct hns_roce_pf_res_b { #define PF_RES_DATA_3_PF_SL_NUM_S 16 #define PF_RES_DATA_3_PF_SL_NUM_M GENMASK(26, 16) +#define PF_RES_DATA_4_PF_SCCC_BT_IDX_S 0 +#define PF_RES_DATA_4_PF_SCCC_BT_IDX_M GENMASK(8, 0) + +#define PF_RES_DATA_4_PF_SCCC_BT_NUM_S 9 +#define PF_RES_DATA_4_PF_SCCC_BT_NUM_M GENMASK(17, 9) + +struct hns_roce_pf_timer_res_a { + __le32 rsv0; + __le32 qpc_timer_bt_idx_num; + __le32 cqc_timer_bt_idx_num; + __le32 rsv[3]; +}; + +#define PF_RES_DATA_1_PF_QPC_TIMER_BT_IDX_S 0 +#define PF_RES_DATA_1_PF_QPC_TIMER_BT_IDX_M GENMASK(11, 0) + +#define PF_RES_DATA_1_PF_QPC_TIMER_BT_NUM_S 16 +#define PF_RES_DATA_1_PF_QPC_TIMER_BT_NUM_M GENMASK(28, 16) + +#define PF_RES_DATA_2_PF_CQC_TIMER_BT_IDX_S 0 +#define PF_RES_DATA_2_PF_CQC_TIMER_BT_IDX_M GENMASK(10, 0) + +#define PF_RES_DATA_2_PF_CQC_TIMER_BT_NUM_S 16 +#define PF_RES_DATA_2_PF_CQC_TIMER_BT_NUM_M GENMASK(27, 16) + struct hns_roce_vf_res_a { __le32 vf_id; __le32 vf_qpc_bt_idx_num; @@ -1365,7 +1406,8 @@ struct hns_roce_vf_res_b { __le32 vf_smac_idx_num; __le32 vf_sgid_idx_num; __le32 vf_qid_idx_sl_num; - __le32 rsv[2]; + __le32 vf_sccc_idx_num; + __le32 rsv1; }; #define VF_RES_B_DATA_0_VF_ID_S 0 @@ -1389,6 +1431,12 @@ struct hns_roce_vf_res_b { #define VF_RES_B_DATA_3_VF_SL_NUM_S 16 #define VF_RES_B_DATA_3_VF_SL_NUM_M GENMASK(19, 16) +#define VF_RES_B_DATA_4_VF_SCCC_BT_IDX_S 0 +#define VF_RES_B_DATA_4_VF_SCCC_BT_IDX_M GENMASK(8, 0) + +#define VF_RES_B_DATA_4_VF_SCCC_BT_NUM_S 9 +#define VF_RES_B_DATA_4_VF_SCCC_BT_NUM_M GENMASK(17, 9) + struct hns_roce_vf_switch { __le32 rocee_sel; __le32 fun_id; @@ -1424,7 +1472,8 @@ struct hns_roce_cfg_bt_attr { __le32 vf_srqc_cfg; __le32 vf_cqc_cfg; __le32 vf_mpt_cfg; - __le32 rsv[2]; + __le32 vf_sccc_cfg; + __le32 rsv; }; #define CFG_BT_ATTR_DATA_0_VF_QPC_BA_PGSZ_S 0 @@ -1463,6 +1512,15 @@ struct hns_roce_cfg_bt_attr { #define CFG_BT_ATTR_DATA_3_VF_MPT_HOPNUM_S 8 #define CFG_BT_ATTR_DATA_3_VF_MPT_HOPNUM_M GENMASK(9, 8) +#define CFG_BT_ATTR_DATA_4_VF_SCCC_BA_PGSZ_S 0 +#define CFG_BT_ATTR_DATA_4_VF_SCCC_BA_PGSZ_M GENMASK(3, 0) + +#define CFG_BT_ATTR_DATA_4_VF_SCCC_BUF_PGSZ_S 4 +#define CFG_BT_ATTR_DATA_4_VF_SCCC_BUF_PGSZ_M GENMASK(7, 4) + +#define CFG_BT_ATTR_DATA_4_VF_SCCC_HOPNUM_S 8 +#define CFG_BT_ATTR_DATA_4_VF_SCCC_HOPNUM_M GENMASK(9, 8) + struct hns_roce_cfg_sgid_tb { __le32 table_idx_rsv; __le32 vf_sgid_l; @@ -1546,6 +1604,7 @@ struct hns_roce_link_table_entry { #define HNS_ROCE_LINK_TABLE_NXT_PTR_M GENMASK(31, 20) struct hns_roce_v2_priv { + struct hnae3_handle *handle; struct hns_roce_v2_cmq cmq; struct hns_roce_link_table tsq; struct hns_roce_link_table tpq; @@ -1730,4 +1789,25 @@ struct hns_roce_wqe_atomic_seg { __le64 cmp_data; }; +struct hns_roce_sccc_clr { + __le32 qpn; + __le32 rsv[5]; +}; + +struct hns_roce_sccc_clr_done { + __le32 clr_done; + __le32 rsv[5]; +}; + +static inline void hns_roce_write64(struct hns_roce_dev *hr_dev, __le32 val[2], + void __iomem *dest) +{ + struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv; + struct hnae3_handle *handle = priv->handle; + const struct hnae3_ae_ops *ops = handle->ae_algo->ops; + + if (!hr_dev->dis_db && !ops->get_hw_reset_stat(handle)) + hns_roce_write64_k(val, dest); +} + #endif diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c index c79054ba9495..c929125da84b 100644 --- a/drivers/infiniband/hw/hns/hns_roce_main.c +++ b/drivers/infiniband/hw/hns/hns_roce_main.c @@ -226,6 +226,11 @@ static int hns_roce_query_device(struct ib_device *ib_dev, props->max_srq_sge = hr_dev->caps.max_srq_sges; } + if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR) { + props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS; + props->max_fast_reg_page_list_len = HNS_ROCE_FRMR_MAX_PA; + } + return 0; } @@ -330,23 +335,19 @@ static int hns_roce_modify_port(struct ib_device *ib_dev, u8 port_num, int mask, return 0; } -static struct ib_ucontext *hns_roce_alloc_ucontext(struct ib_device *ib_dev, - struct ib_udata *udata) +static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx, + struct ib_udata *udata) { int ret = 0; - struct hns_roce_ucontext *context; + struct hns_roce_ucontext *context = to_hr_ucontext(uctx); struct hns_roce_ib_alloc_ucontext_resp resp = {}; - struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev); + struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device); if (!hr_dev->active) - return ERR_PTR(-EAGAIN); + return -EAGAIN; resp.qp_tab_size = hr_dev->caps.num_qps; - context = kmalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); - ret = hns_roce_uar_alloc(hr_dev, &context->uar); if (ret) goto error_fail_uar_alloc; @@ -360,25 +361,20 @@ static struct ib_ucontext *hns_roce_alloc_ucontext(struct ib_device *ib_dev, if (ret) goto error_fail_copy_to_udata; - return &context->ibucontext; + return 0; error_fail_copy_to_udata: hns_roce_uar_free(hr_dev, &context->uar); error_fail_uar_alloc: - kfree(context); - - return ERR_PTR(ret); + return ret; } -static int hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext) +static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext) { struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext); hns_roce_uar_free(to_hr_dev(ibcontext->device), &context->uar); - kfree(context); - - return 0; } static int hns_roce_mmap(struct ib_ucontext *context, @@ -472,6 +468,8 @@ static const struct ib_device_ops hns_roce_dev_ops = { .query_pkey = hns_roce_query_pkey, .query_port = hns_roce_query_port, .reg_user_mr = hns_roce_reg_user_mr, + INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext), }; static const struct ib_device_ops hns_roce_dev_mr_ops = { @@ -564,7 +562,7 @@ static int hns_roce_register_device(struct hns_roce_dev *hr_dev) ib_dev->driver_id = RDMA_DRIVER_HNS; ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops); ib_set_device_ops(ib_dev, &hns_roce_dev_ops); - ret = ib_register_device(ib_dev, "hns_%d", NULL); + ret = ib_register_device(ib_dev, "hns_%d"); if (ret) { dev_err(dev, "ib_register_device failed!\n"); return ret; @@ -702,8 +700,62 @@ static int hns_roce_init_hem(struct hns_roce_dev *hr_dev) } } + if (hr_dev->caps.sccc_entry_sz) { + ret = hns_roce_init_hem_table(hr_dev, + &hr_dev->qp_table.sccc_table, + HEM_TYPE_SCCC, + hr_dev->caps.sccc_entry_sz, + hr_dev->caps.num_qps, 1); + if (ret) { + dev_err(dev, + "Failed to init SCC context memory, aborting.\n"); + goto err_unmap_idx; + } + } + + if (hr_dev->caps.qpc_timer_entry_sz) { + ret = hns_roce_init_hem_table(hr_dev, + &hr_dev->qpc_timer_table, + HEM_TYPE_QPC_TIMER, + hr_dev->caps.qpc_timer_entry_sz, + hr_dev->caps.num_qpc_timer, 1); + if (ret) { + dev_err(dev, + "Failed to init QPC timer memory, aborting.\n"); + goto err_unmap_ctx; + } + } + + if (hr_dev->caps.cqc_timer_entry_sz) { + ret = hns_roce_init_hem_table(hr_dev, + &hr_dev->cqc_timer_table, + HEM_TYPE_CQC_TIMER, + hr_dev->caps.cqc_timer_entry_sz, + hr_dev->caps.num_cqc_timer, 1); + if (ret) { + dev_err(dev, + "Failed to init CQC timer memory, aborting.\n"); + goto err_unmap_qpc_timer; + } + } + return 0; +err_unmap_qpc_timer: + if (hr_dev->caps.qpc_timer_entry_sz) + hns_roce_cleanup_hem_table(hr_dev, + &hr_dev->qpc_timer_table); + +err_unmap_ctx: + if (hr_dev->caps.sccc_entry_sz) + hns_roce_cleanup_hem_table(hr_dev, + &hr_dev->qp_table.sccc_table); + +err_unmap_idx: + if (hr_dev->caps.num_idx_segs) + hns_roce_cleanup_hem_table(hr_dev, + &hr_dev->mr_table.mtt_idx_table); + err_unmap_srqwqe: if (hr_dev->caps.num_srqwqe_segs) hns_roce_cleanup_hem_table(hr_dev, diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c index ee5991bd4171..b09f1cde2ff5 100644 --- a/drivers/infiniband/hw/hns/hns_roce_mr.c +++ b/drivers/infiniband/hw/hns/hns_roce_mr.c @@ -976,12 +976,11 @@ int hns_roce_ib_umem_write_mtt(struct hns_roce_dev *hr_dev, struct hns_roce_mtt *mtt, struct ib_umem *umem) { struct device *dev = hr_dev->dev; - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; unsigned int order; - int i, k, entry; int npage = 0; int ret = 0; - int len; + int i; u64 page_addr; u64 *pages; u32 bt_page_size; @@ -1014,29 +1013,25 @@ int hns_roce_ib_umem_write_mtt(struct hns_roce_dev *hr_dev, i = n = 0; - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { - len = sg_dma_len(sg) >> PAGE_SHIFT; - for (k = 0; k < len; ++k) { - page_addr = - sg_dma_address(sg) + (k << umem->page_shift); - if (!(npage % (1 << (mtt->page_shift - PAGE_SHIFT)))) { - if (page_addr & ((1 << mtt->page_shift) - 1)) { - dev_err(dev, "page_addr 0x%llx is not page_shift %d alignment!\n", - page_addr, mtt->page_shift); - ret = -EINVAL; - goto out; - } - pages[i++] = page_addr; - } - npage++; - if (i == bt_page_size / sizeof(u64)) { - ret = hns_roce_write_mtt(hr_dev, mtt, n, i, - pages); - if (ret) - goto out; - n += i; - i = 0; + for_each_sg_dma_page(umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + page_addr = sg_page_iter_dma_address(&sg_iter); + if (!(npage % (1 << (mtt->page_shift - PAGE_SHIFT)))) { + if (page_addr & ((1 << mtt->page_shift) - 1)) { + dev_err(dev, + "page_addr 0x%llx is not page_shift %d alignment!\n", + page_addr, mtt->page_shift); + ret = -EINVAL; + goto out; } + pages[i++] = page_addr; + } + npage++; + if (i == bt_page_size / sizeof(u64)) { + ret = hns_roce_write_mtt(hr_dev, mtt, n, i, pages); + if (ret) + goto out; + n += i; + i = 0; } } @@ -1052,10 +1047,8 @@ static int hns_roce_ib_umem_write_mr(struct hns_roce_dev *hr_dev, struct hns_roce_mr *mr, struct ib_umem *umem) { - struct scatterlist *sg; - int i = 0, j = 0, k; - int entry; - int len; + struct sg_dma_page_iter sg_iter; + int i = 0, j = 0; u64 page_addr; u32 pbl_bt_sz; @@ -1063,27 +1056,22 @@ static int hns_roce_ib_umem_write_mr(struct hns_roce_dev *hr_dev, return 0; pbl_bt_sz = 1 << (hr_dev->caps.pbl_ba_pg_sz + PAGE_SHIFT); - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { - len = sg_dma_len(sg) >> PAGE_SHIFT; - for (k = 0; k < len; ++k) { - page_addr = sg_dma_address(sg) + - (k << umem->page_shift); - - if (!hr_dev->caps.pbl_hop_num) { - mr->pbl_buf[i++] = page_addr >> 12; - } else if (hr_dev->caps.pbl_hop_num == 1) { - mr->pbl_buf[i++] = page_addr; - } else { - if (hr_dev->caps.pbl_hop_num == 2) - mr->pbl_bt_l1[i][j] = page_addr; - else if (hr_dev->caps.pbl_hop_num == 3) - mr->pbl_bt_l2[i][j] = page_addr; - - j++; - if (j >= (pbl_bt_sz / 8)) { - i++; - j = 0; - } + for_each_sg_dma_page(umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + page_addr = sg_page_iter_dma_address(&sg_iter); + if (!hr_dev->caps.pbl_hop_num) { + mr->pbl_buf[i++] = page_addr >> 12; + } else if (hr_dev->caps.pbl_hop_num == 1) { + mr->pbl_buf[i++] = page_addr; + } else { + if (hr_dev->caps.pbl_hop_num == 2) + mr->pbl_bt_l1[i][j] = page_addr; + else if (hr_dev->caps.pbl_hop_num == 3) + mr->pbl_bt_l2[i][j] = page_addr; + + j++; + if (j >= (pbl_bt_sz / 8)) { + i++; + j = 0; } } } @@ -1110,8 +1098,7 @@ struct ib_mr *hns_roce_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, if (!mr) return ERR_PTR(-ENOMEM); - mr->umem = ib_umem_get(pd->uobject->context, start, length, - access_flags, 0); + mr->umem = ib_umem_get(udata, start, length, access_flags, 0); if (IS_ERR(mr->umem)) { ret = PTR_ERR(mr->umem); goto err_free; @@ -1220,8 +1207,8 @@ int hns_roce_rereg_user_mr(struct ib_mr *ibmr, int flags, u64 start, u64 length, } ib_umem_release(mr->umem); - mr->umem = ib_umem_get(ibmr->uobject->context, start, length, - mr_access_flags, 0); + mr->umem = + ib_umem_get(udata, start, length, mr_access_flags, 0); if (IS_ERR(mr->umem)) { ret = PTR_ERR(mr->umem); mr->umem = NULL; diff --git a/drivers/infiniband/hw/hns/hns_roce_pd.c b/drivers/infiniband/hw/hns/hns_roce_pd.c index e11c149da04d..b9b97c5e97e6 100644 --- a/drivers/infiniband/hw/hns/hns_roce_pd.c +++ b/drivers/infiniband/hw/hns/hns_roce_pd.c @@ -57,24 +57,19 @@ void hns_roce_cleanup_pd_table(struct hns_roce_dev *hr_dev) hns_roce_bitmap_cleanup(&hr_dev->pd_bitmap); } -struct ib_pd *hns_roce_alloc_pd(struct ib_device *ib_dev, - struct ib_ucontext *context, - struct ib_udata *udata) +int hns_roce_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { + struct ib_device *ib_dev = ibpd->device; struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev); struct device *dev = hr_dev->dev; - struct hns_roce_pd *pd; + struct hns_roce_pd *pd = to_hr_pd(ibpd); int ret; - pd = kmalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - ret = hns_roce_pd_alloc(to_hr_dev(ib_dev), &pd->pdn); if (ret) { - kfree(pd); dev_err(dev, "[alloc_pd]hns_roce_pd_alloc failed!\n"); - return ERR_PTR(ret); + return ret; } if (context) { @@ -83,21 +78,17 @@ struct ib_pd *hns_roce_alloc_pd(struct ib_device *ib_dev, if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) { hns_roce_pd_free(to_hr_dev(ib_dev), pd->pdn); dev_err(dev, "[alloc_pd]ib_copy_to_udata failed!\n"); - kfree(pd); - return ERR_PTR(-EFAULT); + return -EFAULT; } } - return &pd->ibpd; + return 0; } EXPORT_SYMBOL_GPL(hns_roce_alloc_pd); -int hns_roce_dealloc_pd(struct ib_pd *pd) +void hns_roce_dealloc_pd(struct ib_pd *pd) { hns_roce_pd_free(to_hr_dev(pd->device), to_hr_pd(pd)->pdn); - kfree(to_hr_pd(pd)); - - return 0; } EXPORT_SYMBOL_GPL(hns_roce_dealloc_pd); diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c index 54031c5b53fa..57c76eafef2f 100644 --- a/drivers/infiniband/hw/hns/hns_roce_qp.c +++ b/drivers/infiniband/hw/hns/hns_roce_qp.c @@ -35,6 +35,7 @@ #include <linux/platform_device.h> #include <rdma/ib_addr.h> #include <rdma/ib_umem.h> +#include <rdma/uverbs_ioctl.h> #include "hns_roce_common.h" #include "hns_roce_device.h" #include "hns_roce_hem.h" @@ -209,13 +210,23 @@ static int hns_roce_qp_alloc(struct hns_roce_dev *hr_dev, unsigned long qpn, } } + if (hr_dev->caps.sccc_entry_sz) { + /* Alloc memory for SCC CTX */ + ret = hns_roce_table_get(hr_dev, &qp_table->sccc_table, + hr_qp->qpn); + if (ret) { + dev_err(dev, "SCC CTX table get failed\n"); + goto err_put_trrl; + } + } + spin_lock_irq(&qp_table->lock); ret = radix_tree_insert(&hr_dev->qp_table_tree, hr_qp->qpn & (hr_dev->caps.num_qps - 1), hr_qp); spin_unlock_irq(&qp_table->lock); if (ret) { dev_err(dev, "QPC radix_tree_insert failed\n"); - goto err_put_trrl; + goto err_put_sccc; } atomic_set(&hr_qp->refcount, 1); @@ -223,6 +234,11 @@ static int hns_roce_qp_alloc(struct hns_roce_dev *hr_dev, unsigned long qpn, return 0; +err_put_sccc: + if (hr_dev->caps.sccc_entry_sz) + hns_roce_table_put(hr_dev, &qp_table->sccc_table, + hr_qp->qpn); + err_put_trrl: if (hr_dev->caps.trrl_entry_sz) hns_roce_table_put(hr_dev, &qp_table->trrl_table, hr_qp->qpn); @@ -258,6 +274,9 @@ void hns_roce_qp_free(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp) wait_for_completion(&hr_qp->free); if ((hr_qp->ibqp.qp_type) != IB_QPT_GSI) { + if (hr_dev->caps.sccc_entry_sz) + hns_roce_table_put(hr_dev, &qp_table->sccc_table, + hr_qp->qpn); if (hr_dev->caps.trrl_entry_sz) hns_roce_table_put(hr_dev, &qp_table->trrl_table, hr_qp->qpn); @@ -526,7 +545,8 @@ static int hns_roce_qp_has_sq(struct ib_qp_init_attr *attr) static int hns_roce_qp_has_rq(struct ib_qp_init_attr *attr) { if (attr->qp_type == IB_QPT_XRC_INI || - attr->qp_type == IB_QPT_XRC_TGT || attr->srq) + attr->qp_type == IB_QPT_XRC_TGT || attr->srq || + !attr->cap.max_recv_wr) return 0; return 1; @@ -541,6 +561,8 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, struct device *dev = hr_dev->dev; struct hns_roce_ib_create_qp ucmd; struct hns_roce_ib_create_qp_resp resp = {}; + struct hns_roce_ucontext *uctx = rdma_udata_to_drv_context( + udata, struct hns_roce_ucontext, ibucontext); unsigned long qpn = 0; int ret = 0; u32 page_shift; @@ -612,9 +634,8 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, goto err_rq_sge_list; } - hr_qp->umem = ib_umem_get(ib_pd->uobject->context, - ucmd.buf_addr, hr_qp->buff_size, 0, - 0); + hr_qp->umem = ib_umem_get(udata, ucmd.buf_addr, + hr_qp->buff_size, 0, 0); if (IS_ERR(hr_qp->umem)) { dev_err(dev, "ib_umem_get error for create qp\n"); ret = PTR_ERR(hr_qp->umem); @@ -622,19 +643,19 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, } hr_qp->mtt.mtt_type = MTT_TYPE_WQE; + page_shift = PAGE_SHIFT; if (hr_dev->caps.mtt_buf_pg_sz) { npages = (ib_umem_page_count(hr_qp->umem) + (1 << hr_dev->caps.mtt_buf_pg_sz) - 1) / - (1 << hr_dev->caps.mtt_buf_pg_sz); - page_shift = PAGE_SHIFT + hr_dev->caps.mtt_buf_pg_sz; + (1 << hr_dev->caps.mtt_buf_pg_sz); + page_shift += hr_dev->caps.mtt_buf_pg_sz; ret = hns_roce_mtt_init(hr_dev, npages, page_shift, &hr_qp->mtt); } else { ret = hns_roce_mtt_init(hr_dev, - ib_umem_page_count(hr_qp->umem), - hr_qp->umem->page_shift, - &hr_qp->mtt); + ib_umem_page_count(hr_qp->umem), + page_shift, &hr_qp->mtt); } if (ret) { dev_err(dev, "hns_roce_mtt_init error for create qp\n"); @@ -652,9 +673,8 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, (udata->inlen >= sizeof(ucmd)) && (udata->outlen >= sizeof(resp)) && hns_roce_qp_has_sq(init_attr)) { - ret = hns_roce_db_map_user( - to_hr_ucontext(ib_pd->uobject->context), - ucmd.sdb_addr, &hr_qp->sdb); + ret = hns_roce_db_map_user(uctx, udata, ucmd.sdb_addr, + &hr_qp->sdb); if (ret) { dev_err(dev, "sq record doorbell map failed!\n"); goto err_mtt; @@ -668,13 +688,16 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) && (udata->outlen >= sizeof(resp)) && hns_roce_qp_has_rq(init_attr)) { - ret = hns_roce_db_map_user( - to_hr_ucontext(ib_pd->uobject->context), - ucmd.db_addr, &hr_qp->rdb); + ret = hns_roce_db_map_user(uctx, udata, ucmd.db_addr, + &hr_qp->rdb); if (ret) { dev_err(dev, "rq record doorbell map failed!\n"); goto err_sq_dbmap; } + + /* indicate kernel supports rq record db */ + resp.cap_flags |= HNS_ROCE_SUPPORT_RQ_RECORD_DB; + hr_qp->rdb_en = 1; } } else { if (init_attr->create_flags & @@ -741,10 +764,10 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, goto err_mtt; } - hr_qp->sq.wrid = kmalloc_array(hr_qp->sq.wqe_cnt, sizeof(u64), - GFP_KERNEL); - hr_qp->rq.wrid = kmalloc_array(hr_qp->rq.wqe_cnt, sizeof(u64), - GFP_KERNEL); + hr_qp->sq.wrid = kcalloc(hr_qp->sq.wqe_cnt, sizeof(u64), + GFP_KERNEL); + hr_qp->rq.wrid = kcalloc(hr_qp->rq.wqe_cnt, sizeof(u64), + GFP_KERNEL); if (!hr_qp->sq.wrid || !hr_qp->rq.wrid) { ret = -ENOMEM; goto err_wrid; @@ -783,17 +806,19 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev, else hr_qp->doorbell_qpn = cpu_to_le64(hr_qp->qpn); - if (udata && (udata->outlen >= sizeof(resp)) && - (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB)) { - - /* indicate kernel supports rq record db */ - resp.cap_flags |= HNS_ROCE_SUPPORT_RQ_RECORD_DB; - ret = ib_copy_to_udata(udata, &resp, sizeof(resp)); + if (udata) { + ret = ib_copy_to_udata(udata, &resp, + min(udata->outlen, sizeof(resp))); if (ret) goto err_qp; + } - hr_qp->rdb_en = 1; + if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) { + ret = hr_dev->hw->qp_flow_control_init(hr_dev, hr_qp); + if (ret) + goto err_qp; } + hr_qp->event = hns_roce_ib_qp_event; return 0; @@ -814,9 +839,7 @@ err_wrid: if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) && (udata->outlen >= sizeof(resp)) && hns_roce_qp_has_rq(init_attr)) - hns_roce_db_unmap_user( - to_hr_ucontext(ib_pd->uobject->context), - &hr_qp->rdb); + hns_roce_db_unmap_user(uctx, &hr_qp->rdb); } else { kfree(hr_qp->sq.wrid); kfree(hr_qp->rq.wrid); @@ -828,9 +851,7 @@ err_sq_dbmap: (udata->inlen >= sizeof(ucmd)) && (udata->outlen >= sizeof(resp)) && hns_roce_qp_has_sq(init_attr)) - hns_roce_db_unmap_user( - to_hr_ucontext(ib_pd->uobject->context), - &hr_qp->sdb); + hns_roce_db_unmap_user(uctx, &hr_qp->sdb); err_mtt: hns_roce_mtt_cleanup(hr_dev, &hr_qp->mtt); @@ -969,7 +990,9 @@ int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, (attr_mask & IB_QP_STATE) && new_state == IB_QPS_ERR) { if (hr_qp->sdb_en == 1) { hr_qp->sq.head = *(int *)(hr_qp->sdb.virt_addr); - hr_qp->rq.head = *(int *)(hr_qp->rdb.virt_addr); + + if (hr_qp->rdb_en == 1) + hr_qp->rq.head = *(int *)(hr_qp->rdb.virt_addr); } else { dev_warn(dev, "flush cqe is not supported in userspace!\n"); goto out; @@ -1133,6 +1156,7 @@ int hns_roce_init_qp_table(struct hns_roce_dev *hr_dev) int reserved_from_bot; int ret; + mutex_init(&qp_table->scc_mutex); spin_lock_init(&qp_table->lock); INIT_RADIX_TREE(&hr_dev->qp_table_tree, GFP_ATOMIC); diff --git a/drivers/infiniband/hw/hns/hns_roce_srq.c b/drivers/infiniband/hw/hns/hns_roce_srq.c index 12deacf442cf..a8ee2f6da967 100644 --- a/drivers/infiniband/hw/hns/hns_roce_srq.c +++ b/drivers/infiniband/hw/hns/hns_roce_srq.c @@ -78,9 +78,9 @@ static int hns_roce_hw2sw_srq(struct hns_roce_dev *dev, HNS_ROCE_CMD_TIMEOUT_MSECS); } -int hns_roce_srq_alloc(struct hns_roce_dev *hr_dev, u32 pdn, u32 cqn, u16 xrcd, - struct hns_roce_mtt *hr_mtt, u64 db_rec_addr, - struct hns_roce_srq *srq) +static int hns_roce_srq_alloc(struct hns_roce_dev *hr_dev, u32 pdn, u32 cqn, + u16 xrcd, struct hns_roce_mtt *hr_mtt, + u64 db_rec_addr, struct hns_roce_srq *srq) { struct hns_roce_srq_table *srq_table = &hr_dev->srq_table; struct hns_roce_cmd_mailbox *mailbox; @@ -155,7 +155,8 @@ err_out: return ret; } -void hns_roce_srq_free(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq) +static void hns_roce_srq_free(struct hns_roce_dev *hr_dev, + struct hns_roce_srq *srq) { struct hns_roce_srq_table *srq_table = &hr_dev->srq_table; int ret; @@ -253,8 +254,8 @@ struct ib_srq *hns_roce_create_srq(struct ib_pd *pd, goto err_srq; } - srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, - srq_buf_size, 0, 0); + srq->umem = + ib_umem_get(udata, ucmd.buf_addr, srq_buf_size, 0, 0); if (IS_ERR(srq->umem)) { ret = PTR_ERR(srq->umem); goto err_srq; @@ -281,8 +282,7 @@ struct ib_srq *hns_roce_create_srq(struct ib_pd *pd, goto err_srq_mtt; /* config index queue BA */ - srq->idx_que.umem = ib_umem_get(pd->uobject->context, - ucmd.que_addr, + srq->idx_que.umem = ib_umem_get(udata, ucmd.que_addr, srq->idx_que.buf_size, 0, 0); if (IS_ERR(srq->idx_que.umem)) { dev_err(hr_dev->dev, diff --git a/drivers/infiniband/hw/i40iw/Makefile b/drivers/infiniband/hw/i40iw/Makefile index 5a8a7a3f28ae..8942f8229945 100644 --- a/drivers/infiniband/hw/i40iw/Makefile +++ b/drivers/infiniband/hw/i40iw/Makefile @@ -1,5 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 -ccflags-y := -Idrivers/net/ethernet/intel/i40e +ccflags-y := -I $(srctree)/drivers/net/ethernet/intel/i40e obj-$(CONFIG_INFINIBAND_I40IW) += i40iw.o diff --git a/drivers/infiniband/hw/i40iw/i40iw_utils.c b/drivers/infiniband/hw/i40iw/i40iw_utils.c index 59e978141ad4..c5a881172524 100644 --- a/drivers/infiniband/hw/i40iw/i40iw_utils.c +++ b/drivers/infiniband/hw/i40iw/i40iw_utils.c @@ -601,7 +601,6 @@ void i40iw_rem_pdusecount(struct i40iw_pd *iwpd, struct i40iw_device *iwdev) if (!atomic_dec_and_test(&iwpd->usecount)) return; i40iw_free_resource(iwdev, iwdev->allocated_pds, iwpd->sc_pd.pd_id); - kfree(iwpd); } /** diff --git a/drivers/infiniband/hw/i40iw/i40iw_verbs.c b/drivers/infiniband/hw/i40iw/i40iw_verbs.c index 0b675b0742c2..a8352e3ca23d 100644 --- a/drivers/infiniband/hw/i40iw/i40iw_verbs.c +++ b/drivers/infiniband/hw/i40iw/i40iw_verbs.c @@ -45,6 +45,7 @@ #include <rdma/iw_cm.h> #include <rdma/ib_user_verbs.h> #include <rdma/ib_umem.h> +#include <rdma/uverbs_ioctl.h> #include "i40iw.h" /** @@ -120,78 +121,55 @@ static int i40iw_query_port(struct ib_device *ibdev, /** * i40iw_alloc_ucontext - Allocate the user context data structure - * @ibdev: device pointer from stack + * @uctx: Uverbs context pointer from stack * @udata: user data * * This keeps track of all objects associated with a particular * user-mode client. */ -static struct ib_ucontext *i40iw_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +static int i40iw_alloc_ucontext(struct ib_ucontext *uctx, + struct ib_udata *udata) { + struct ib_device *ibdev = uctx->device; struct i40iw_device *iwdev = to_iwdev(ibdev); struct i40iw_alloc_ucontext_req req; - struct i40iw_alloc_ucontext_resp uresp; - struct i40iw_ucontext *ucontext; + struct i40iw_alloc_ucontext_resp uresp = {}; + struct i40iw_ucontext *ucontext = to_ucontext(uctx); if (ib_copy_from_udata(&req, udata, sizeof(req))) - return ERR_PTR(-EINVAL); + return -EINVAL; if (req.userspace_ver < 4 || req.userspace_ver > I40IW_ABI_VER) { i40iw_pr_err("Unsupported provider library version %u.\n", req.userspace_ver); - return ERR_PTR(-EINVAL); + return -EINVAL; } - memset(&uresp, 0, sizeof(uresp)); uresp.max_qps = iwdev->max_qp; uresp.max_pds = iwdev->max_pd; uresp.wq_size = iwdev->max_qp_wr * 2; uresp.kernel_ver = req.userspace_ver; - ucontext = kzalloc(sizeof(*ucontext), GFP_KERNEL); - if (!ucontext) - return ERR_PTR(-ENOMEM); - ucontext->iwdev = iwdev; ucontext->abi_ver = req.userspace_ver; - if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) { - kfree(ucontext); - return ERR_PTR(-EFAULT); - } + if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) + return -EFAULT; INIT_LIST_HEAD(&ucontext->cq_reg_mem_list); spin_lock_init(&ucontext->cq_reg_mem_list_lock); INIT_LIST_HEAD(&ucontext->qp_reg_mem_list); spin_lock_init(&ucontext->qp_reg_mem_list_lock); - return &ucontext->ibucontext; + return 0; } /** * i40iw_dealloc_ucontext - deallocate the user context data structure * @context: user context created during alloc */ -static int i40iw_dealloc_ucontext(struct ib_ucontext *context) +static void i40iw_dealloc_ucontext(struct ib_ucontext *context) { - struct i40iw_ucontext *ucontext = to_ucontext(context); - unsigned long flags; - - spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags); - if (!list_empty(&ucontext->cq_reg_mem_list)) { - spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags); - return -EBUSY; - } - spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags); - spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags); - if (!list_empty(&ucontext->qp_reg_mem_list)) { - spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags); - return -EBUSY; - } - spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags); - - kfree(ucontext); - return 0; + return; } /** @@ -312,16 +290,15 @@ static void i40iw_dealloc_push_page(struct i40iw_device *iwdev, struct i40iw_sc_ /** * i40iw_alloc_pd - allocate protection domain - * @ibdev: device pointer from stack + * @pd: PD pointer * @context: user context created during alloc * @udata: user data */ -static struct ib_pd *i40iw_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +static int i40iw_alloc_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct i40iw_pd *iwpd; - struct i40iw_device *iwdev = to_iwdev(ibdev); + struct i40iw_pd *iwpd = to_iwpd(pd); + struct i40iw_device *iwdev = to_iwdev(pd->device); struct i40iw_sc_dev *dev = &iwdev->sc_dev; struct i40iw_alloc_pd_resp uresp; struct i40iw_sc_pd *sc_pd; @@ -330,19 +307,13 @@ static struct ib_pd *i40iw_alloc_pd(struct ib_device *ibdev, int err; if (iwdev->closing) - return ERR_PTR(-ENODEV); + return -ENODEV; err = i40iw_alloc_resource(iwdev, iwdev->allocated_pds, iwdev->max_pd, &pd_id, &iwdev->next_pd); if (err) { i40iw_pr_err("alloc resource failed\n"); - return ERR_PTR(err); - } - - iwpd = kzalloc(sizeof(*iwpd), GFP_KERNEL); - if (!iwpd) { - err = -ENOMEM; - goto free_res; + return err; } sc_pd = &iwpd->sc_pd; @@ -361,25 +332,23 @@ static struct ib_pd *i40iw_alloc_pd(struct ib_device *ibdev, } i40iw_add_pdusecount(iwpd); - return &iwpd->ibpd; + return 0; + error: - kfree(iwpd); -free_res: i40iw_free_resource(iwdev, iwdev->allocated_pds, pd_id); - return ERR_PTR(err); + return err; } /** * i40iw_dealloc_pd - deallocate pd * @ibpd: ptr of pd to be deallocated */ -static int i40iw_dealloc_pd(struct ib_pd *ibpd) +static void i40iw_dealloc_pd(struct ib_pd *ibpd) { struct i40iw_pd *iwpd = to_iwpd(ibpd); struct i40iw_device *iwdev = to_iwdev(ibpd->device); i40iw_rem_pdusecount(iwpd, iwdev); - return 0; } /** @@ -565,7 +534,8 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd, struct i40iw_device *iwdev = to_iwdev(ibpd->device); struct i40iw_cqp *iwcqp = &iwdev->cqp; struct i40iw_qp *iwqp; - struct i40iw_ucontext *ucontext; + struct i40iw_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct i40iw_ucontext, ibucontext); struct i40iw_create_qp_req req; struct i40iw_create_qp_resp uresp; u32 qp_num = 0; @@ -674,7 +644,6 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd, } iwqp->ctx_info.qp_compl_ctx = req.user_compl_ctx; iwqp->user_mode = 1; - ucontext = to_ucontext(ibpd->uobject->context); if (req.user_wqe_buffers) { struct i40iw_pbl *iwpbl; @@ -1369,32 +1338,29 @@ static void i40iw_copy_user_pgaddrs(struct i40iw_mr *iwmr, { struct ib_umem *region = iwmr->region; struct i40iw_pbl *iwpbl = &iwmr->iwpbl; - int chunk_pages, entry, i; struct i40iw_pble_alloc *palloc = &iwpbl->pble_alloc; struct i40iw_pble_info *pinfo; - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; u64 pg_addr = 0; u32 idx = 0; + bool first_pg = true; pinfo = (level == I40IW_LEVEL_1) ? NULL : palloc->level2.leaf; - for_each_sg(region->sg_head.sgl, sg, region->nmap, entry) { - chunk_pages = sg_dma_len(sg) >> region->page_shift; - if ((iwmr->type == IW_MEMREG_TYPE_QP) && - !iwpbl->qp_mr.sq_page) - iwpbl->qp_mr.sq_page = sg_page(sg); - for (i = 0; i < chunk_pages; i++) { - pg_addr = sg_dma_address(sg) + - (i << region->page_shift); - - if ((entry + i) == 0) - *pbl = cpu_to_le64(pg_addr & iwmr->page_msk); - else if (!(pg_addr & ~iwmr->page_msk)) - *pbl = cpu_to_le64(pg_addr); - else - continue; - pbl = i40iw_next_pbl_addr(pbl, &pinfo, &idx); - } + if (iwmr->type == IW_MEMREG_TYPE_QP) + iwpbl->qp_mr.sq_page = sg_page(region->sg_head.sgl); + + for_each_sg_dma_page (region->sg_head.sgl, &sg_iter, region->nmap, 0) { + pg_addr = sg_page_iter_dma_address(&sg_iter); + if (first_pg) + *pbl = cpu_to_le64(pg_addr & iwmr->page_msk); + else if (!(pg_addr & ~iwmr->page_msk)) + *pbl = cpu_to_le64(pg_addr); + else + continue; + + first_pg = false; + pbl = i40iw_next_pbl_addr(pbl, &pinfo, &idx); } } @@ -1831,7 +1797,8 @@ static struct ib_mr *i40iw_reg_user_mr(struct ib_pd *pd, { struct i40iw_pd *iwpd = to_iwpd(pd); struct i40iw_device *iwdev = to_iwdev(pd->device); - struct i40iw_ucontext *ucontext; + struct i40iw_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct i40iw_ucontext, ibucontext); struct i40iw_pble_alloc *palloc; struct i40iw_pbl *iwpbl; struct i40iw_mr *iwmr; @@ -1852,7 +1819,7 @@ static struct ib_mr *i40iw_reg_user_mr(struct ib_pd *pd, if (length > I40IW_MAX_MR_SIZE) return ERR_PTR(-EINVAL); - region = ib_umem_get(pd->uobject->context, start, length, acc, 0); + region = ib_umem_get(udata, start, length, acc, 0); if (IS_ERR(region)) return (struct ib_mr *)region; @@ -1872,7 +1839,6 @@ static struct ib_mr *i40iw_reg_user_mr(struct ib_pd *pd, iwmr->region = region; iwmr->ibmr.pd = pd; iwmr->ibmr.device = pd->device; - ucontext = to_ucontext(pd->uobject->context); iwmr->page_size = PAGE_SIZE; iwmr->page_msk = PAGE_MASK; @@ -2139,9 +2105,8 @@ static int i40iw_dereg_mr(struct ib_mr *ib_mr) static ssize_t hw_rev_show(struct device *dev, struct device_attribute *attr, char *buf) { - struct i40iw_ib_device *iwibdev = container_of(dev, - struct i40iw_ib_device, - ibdev.dev); + struct i40iw_ib_device *iwibdev = + rdma_device_to_drv_device(dev, struct i40iw_ib_device, ibdev); u32 hw_rev = iwibdev->iwdev->sc_dev.hw_rev; return sprintf(buf, "%x\n", hw_rev); @@ -2751,6 +2716,8 @@ static const struct ib_device_ops i40iw_dev_ops = { .query_qp = i40iw_query_qp, .reg_user_mr = i40iw_reg_user_mr, .req_notify_cq = i40iw_req_notify_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, i40iw_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, i40iw_ucontext, ibucontext), }; /** @@ -2763,7 +2730,7 @@ static struct i40iw_ib_device *i40iw_init_rdma_device(struct i40iw_device *iwdev struct net_device *netdev = iwdev->netdev; struct pci_dev *pcidev = (struct pci_dev *)iwdev->hw.dev_context; - iwibdev = (struct i40iw_ib_device *)ib_alloc_device(sizeof(*iwibdev)); + iwibdev = ib_alloc_device(i40iw_ib_device, ibdev); if (!iwibdev) { i40iw_pr_err("iwdev == NULL\n"); return NULL; @@ -2868,7 +2835,7 @@ int i40iw_register_rdma_device(struct i40iw_device *iwdev) iwibdev = iwdev->iwibdev; rdma_set_device_sysfs_group(&iwibdev->ibdev, &i40iw_attr_group); iwibdev->ibdev.driver_id = RDMA_DRIVER_I40IW; - ret = ib_register_device(&iwibdev->ibdev, "i40iw%d", NULL); + ret = ib_register_device(&iwibdev->ibdev, "i40iw%d"); if (ret) goto error; diff --git a/drivers/infiniband/hw/mlx4/Kconfig b/drivers/infiniband/hw/mlx4/Kconfig index 4e9936731867..fc01deac1d3c 100644 --- a/drivers/infiniband/hw/mlx4/Kconfig +++ b/drivers/infiniband/hw/mlx4/Kconfig @@ -1,7 +1,6 @@ config MLX4_INFINIBAND tristate "Mellanox ConnectX HCA support" depends on NETDEVICES && ETHERNET && PCI && INET - depends on INFINIBAND_USER_ACCESS || !INFINIBAND_USER_ACCESS select NET_VENDOR_MELLANOX select MLX4_CORE ---help--- diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c index fedaf8260105..8c79a480f2b7 100644 --- a/drivers/infiniband/hw/mlx4/cm.c +++ b/drivers/infiniband/hw/mlx4/cm.c @@ -39,7 +39,7 @@ #include "mlx4_ib.h" -#define CM_CLEANUP_CACHE_TIMEOUT (5 * HZ) +#define CM_CLEANUP_CACHE_TIMEOUT (30 * HZ) struct id_map_entry { struct rb_node node; diff --git a/drivers/infiniband/hw/mlx4/cq.c b/drivers/infiniband/hw/mlx4/cq.c index 43512347b4f0..03ac72339dd2 100644 --- a/drivers/infiniband/hw/mlx4/cq.c +++ b/drivers/infiniband/hw/mlx4/cq.c @@ -134,16 +134,16 @@ static void mlx4_ib_free_cq_buf(struct mlx4_ib_dev *dev, struct mlx4_ib_cq_buf * mlx4_buf_free(dev->dev, (cqe + 1) * buf->entry_size, &buf->buf); } -static int mlx4_ib_get_cq_umem(struct mlx4_ib_dev *dev, struct ib_ucontext *context, - struct mlx4_ib_cq_buf *buf, struct ib_umem **umem, - u64 buf_addr, int cqe) +static int mlx4_ib_get_cq_umem(struct mlx4_ib_dev *dev, struct ib_udata *udata, + struct mlx4_ib_cq_buf *buf, + struct ib_umem **umem, u64 buf_addr, int cqe) { int err; int cqe_size = dev->dev->caps.cqe_size; int shift; int n; - *umem = ib_umem_get(context, buf_addr, cqe * cqe_size, + *umem = ib_umem_get(udata, buf_addr, cqe * cqe_size, IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(*umem)) return PTR_ERR(*umem); @@ -190,7 +190,7 @@ struct ib_cq *mlx4_ib_create_cq(struct ib_device *ibdev, if (attr->flags & ~CQ_CREATE_FLAGS_SUPPORTED) return ERR_PTR(-EINVAL); - cq = kmalloc(sizeof *cq, GFP_KERNEL); + cq = kzalloc(sizeof(*cq), GFP_KERNEL); if (!cq) return ERR_PTR(-ENOMEM); @@ -213,14 +213,13 @@ struct ib_cq *mlx4_ib_create_cq(struct ib_device *ibdev, } buf_addr = (void *)(unsigned long)ucmd.buf_addr; - - err = mlx4_ib_get_cq_umem(dev, context, &cq->buf, &cq->umem, + err = mlx4_ib_get_cq_umem(dev, udata, &cq->buf, &cq->umem, ucmd.buf_addr, entries); if (err) goto err_cq; - err = mlx4_ib_db_map_user(to_mucontext(context), ucmd.db_addr, - &cq->db); + err = mlx4_ib_db_map_user(to_mucontext(context), udata, + ucmd.db_addr, &cq->db); if (err) goto err_mtt; @@ -336,7 +335,7 @@ static int mlx4_alloc_resize_umem(struct mlx4_ib_dev *dev, struct mlx4_ib_cq *cq if (!cq->resize_buf) return -ENOMEM; - err = mlx4_ib_get_cq_umem(dev, cq->umem->context, &cq->resize_buf->buf, + err = mlx4_ib_get_cq_umem(dev, udata, &cq->resize_buf->buf, &cq->resize_umem, ucmd.buf_addr, entries); if (err) { kfree(cq->resize_buf); diff --git a/drivers/infiniband/hw/mlx4/doorbell.c b/drivers/infiniband/hw/mlx4/doorbell.c index c51740986367..3aab71b29ce8 100644 --- a/drivers/infiniband/hw/mlx4/doorbell.c +++ b/drivers/infiniband/hw/mlx4/doorbell.c @@ -41,7 +41,8 @@ struct mlx4_ib_user_db_page { int refcnt; }; -int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt, +int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, + struct ib_udata *udata, unsigned long virt, struct mlx4_db *db) { struct mlx4_ib_user_db_page *page; @@ -61,8 +62,7 @@ int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt, page->user_virt = (virt & PAGE_MASK); page->refcnt = 0; - page->umem = ib_umem_get(&context->ibucontext, virt & PAGE_MASK, - PAGE_SIZE, 0, 0); + page->umem = ib_umem_get(udata, virt & PAGE_MASK, PAGE_SIZE, 0, 0); if (IS_ERR(page->umem)) { err = PTR_ERR(page->umem); kfree(page); diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index 1f15ec3e2b83..733f7bbd5901 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -1076,17 +1076,18 @@ out: return err; } -static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx, + struct ib_udata *udata) { + struct ib_device *ibdev = uctx->device; struct mlx4_ib_dev *dev = to_mdev(ibdev); - struct mlx4_ib_ucontext *context; + struct mlx4_ib_ucontext *context = to_mucontext(uctx); struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3; struct mlx4_ib_alloc_ucontext_resp resp; int err; if (!dev->ib_active) - return ERR_PTR(-EAGAIN); + return -EAGAIN; if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) { resp_v3.qp_tab_size = dev->dev->caps.num_qps; @@ -1100,15 +1101,9 @@ static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev, resp.cqe_size = dev->dev->caps.cqe_size; } - context = kzalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); - err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar); - if (err) { - kfree(context); - return ERR_PTR(err); - } + if (err) + return err; INIT_LIST_HEAD(&context->db_page_list); mutex_init(&context->db_page_mutex); @@ -1123,21 +1118,17 @@ static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev, if (err) { mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar); - kfree(context); - return ERR_PTR(-EFAULT); + return -EFAULT; } - return &context->ibucontext; + return err; } -static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) +static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) { struct mlx4_ib_ucontext *context = to_mucontext(ibcontext); mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar); - kfree(context); - - return 0; } static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext) @@ -1186,38 +1177,27 @@ static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) } } -static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct mlx4_ib_pd *pd; + struct mlx4_ib_pd *pd = to_mpd(ibpd); + struct ib_device *ibdev = ibpd->device; int err; - pd = kzalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn); - if (err) { - kfree(pd); - return ERR_PTR(err); - } + if (err) + return err; - if (context) - if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) { - mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn); - kfree(pd); - return ERR_PTR(-EFAULT); - } - return &pd->ibpd; + if (context && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) { + mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn); + return -EFAULT; + } + return 0; } -static int mlx4_ib_dealloc_pd(struct ib_pd *pd) +static void mlx4_ib_dealloc_pd(struct ib_pd *pd) { mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn); - kfree(pd); - - return 0; } static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev, @@ -2043,7 +2023,7 @@ static ssize_t hca_type_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx4_ib_dev *dev = - container_of(device, struct mlx4_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev); return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device); } static DEVICE_ATTR_RO(hca_type); @@ -2052,7 +2032,7 @@ static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx4_ib_dev *dev = - container_of(device, struct mlx4_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev); return sprintf(buf, "%x\n", dev->dev->rev_id); } static DEVICE_ATTR_RO(hw_rev); @@ -2061,7 +2041,8 @@ static ssize_t board_id_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx4_ib_dev *dev = - container_of(device, struct mlx4_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev); + return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN, dev->dev->board_id); } @@ -2579,6 +2560,8 @@ static const struct ib_device_ops mlx4_ib_dev_ops = { .req_notify_cq = mlx4_ib_arm_cq, .rereg_user_mr = mlx4_ib_rereg_user_mr, .resize_cq = mlx4_ib_resize_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext), }; static const struct ib_device_ops mlx4_ib_dev_wq_ops = { @@ -2634,7 +2617,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev) if (num_ports == 0) return NULL; - ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev); + ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev); if (!ibdev) { dev_err(&dev->persist->pdev->dev, "Device struct alloc failed\n"); @@ -2856,7 +2839,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev) rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group); ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4; - if (ib_register_device(&ibdev->ib_dev, "mlx4_%d", NULL)) + if (ib_register_device(&ibdev->ib_dev, "mlx4_%d")) goto err_diag_counters; if (mlx4_ib_mad_init(ibdev)) diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index e491f3eda6e7..60dc1347c5ab 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -722,7 +722,8 @@ static inline u8 mlx4_ib_bond_next_port(struct mlx4_ib_dev *dev) int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev); void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev); -int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt, +int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, + struct ib_udata *udata, unsigned long virt, struct mlx4_db *db); void mlx4_ib_db_unmap_user(struct mlx4_ib_ucontext *context, struct mlx4_db *db); diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c index c7c85c22e4e3..395379a480cb 100644 --- a/drivers/infiniband/hw/mlx4/mr.c +++ b/drivers/infiniband/hw/mlx4/mr.c @@ -367,7 +367,7 @@ end: return block_shift; } -static struct ib_umem *mlx4_get_umem_mr(struct ib_ucontext *context, u64 start, +static struct ib_umem *mlx4_get_umem_mr(struct ib_udata *udata, u64 start, u64 length, u64 virt_addr, int access_flags) { @@ -398,7 +398,7 @@ static struct ib_umem *mlx4_get_umem_mr(struct ib_ucontext *context, u64 start, up_read(¤t->mm->mmap_sem); } - return ib_umem_get(context, start, length, access_flags, 0); + return ib_umem_get(udata, start, length, access_flags, 0); } struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, @@ -415,8 +415,8 @@ struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, if (!mr) return ERR_PTR(-ENOMEM); - mr->umem = mlx4_get_umem_mr(pd->uobject->context, start, length, - virt_addr, access_flags); + mr->umem = + mlx4_get_umem_mr(udata, start, length, virt_addr, access_flags); if (IS_ERR(mr->umem)) { err = PTR_ERR(mr->umem); goto err_free; @@ -505,9 +505,8 @@ int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags, mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr); ib_umem_release(mmr->umem); - mmr->umem = - mlx4_get_umem_mr(mr->uobject->context, start, length, - virt_addr, mr_access_flags); + mmr->umem = mlx4_get_umem_mr(udata, start, length, virt_addr, + mr_access_flags); if (IS_ERR(mmr->umem)) { err = PTR_ERR(mmr->umem); /* Prevent mlx4_ib_dereg_mr from free'ing invalid pointer */ diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index 971e9a9ebdaf..429a59c5801c 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -41,6 +41,7 @@ #include <rdma/ib_pack.h> #include <rdma/ib_addr.h> #include <rdma/ib_mad.h> +#include <rdma/uverbs_ioctl.h> #include <linux/mlx4/driver.h> #include <linux/mlx4/qp.h> @@ -52,7 +53,8 @@ static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq); static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq); -static int _mlx4_ib_modify_wq(struct ib_wq *ibwq, enum ib_wq_state new_state); +static int _mlx4_ib_modify_wq(struct ib_wq *ibwq, enum ib_wq_state new_state, + struct ib_udata *udata); enum { MLX4_IB_ACK_REQ_FREQ = 8, @@ -863,6 +865,8 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, int err; struct mlx4_ib_sqp *sqp = NULL; struct mlx4_ib_qp *qp; + struct mlx4_ib_ucontext *context = rdma_udata_to_drv_context( + udata, struct mlx4_ib_ucontext, ibucontext); enum mlx4_ib_qp_type qp_type = (enum mlx4_ib_qp_type) init_attr->qp_type; struct mlx4_ib_cq *mcq; unsigned long flags; @@ -1015,9 +1019,11 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, (qp->sq.wqe_cnt << qp->sq.wqe_shift); } - qp->umem = ib_umem_get(pd->uobject->context, - (src == MLX4_IB_QP_SRC) ? ucmd.qp.buf_addr : - ucmd.wq.buf_addr, qp->buf_size, 0, 0); + qp->umem = + ib_umem_get(udata, + (src == MLX4_IB_QP_SRC) ? ucmd.qp.buf_addr : + ucmd.wq.buf_addr, + qp->buf_size, 0, 0); if (IS_ERR(qp->umem)) { err = PTR_ERR(qp->umem); goto err; @@ -1035,9 +1041,11 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, goto err_mtt; if (qp_has_rq(init_attr)) { - err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context), + err = mlx4_ib_db_map_user( + context, udata, (src == MLX4_IB_QP_SRC) ? ucmd.qp.db_addr : - ucmd.wq.db_addr, &qp->db); + ucmd.wq.db_addr, + &qp->db); if (err) goto err_mtt; } @@ -1108,8 +1116,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, } } } else if (src == MLX4_IB_RWQ_SRC) { - err = mlx4_ib_alloc_wqn(to_mucontext(pd->uobject->context), qp, - range_size, &qpn); + err = mlx4_ib_alloc_wqn(context, qp, range_size, &qpn); if (err) goto err_wrid; } else { @@ -1180,8 +1187,7 @@ err_qpn: if (qp->flags & MLX4_IB_QP_NETIF) mlx4_ib_steer_qp_free(dev, qpn, 1); else if (src == MLX4_IB_RWQ_SRC) - mlx4_ib_release_wqn(to_mucontext(pd->uobject->context), - qp, 0); + mlx4_ib_release_wqn(context, qp, 0); else mlx4_qp_release_range(dev->dev, qpn, 1); } @@ -1191,7 +1197,7 @@ err_proxy: err_wrid: if (udata) { if (qp_has_rq(init_attr)) - mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &qp->db); + mlx4_ib_db_unmap_user(context, &qp->db); } else { kvfree(qp->sq.wrid); kvfree(qp->rq.wrid); @@ -1938,7 +1944,8 @@ static u8 gid_type_to_qpc(enum ib_gid_type gid_type) * Go over all RSS QP's childes (WQs) and apply their HW state according to * their logic state if the RSS QP is the first RSS QP associated for the WQ. */ -static int bringup_rss_rwqs(struct ib_rwq_ind_table *ind_tbl, u8 port_num) +static int bringup_rss_rwqs(struct ib_rwq_ind_table *ind_tbl, u8 port_num, + struct ib_udata *udata) { int err = 0; int i; @@ -1962,7 +1969,7 @@ static int bringup_rss_rwqs(struct ib_rwq_ind_table *ind_tbl, u8 port_num) } wq->port = port_num; if ((wq->rss_usecnt == 0) && (ibwq->state == IB_WQS_RDY)) { - err = _mlx4_ib_modify_wq(ibwq, IB_WQS_RDY); + err = _mlx4_ib_modify_wq(ibwq, IB_WQS_RDY, udata); if (err) { mutex_unlock(&wq->mutex); break; @@ -1984,7 +1991,8 @@ static int bringup_rss_rwqs(struct ib_rwq_ind_table *ind_tbl, u8 port_num) if ((wq->rss_usecnt == 1) && (ibwq->state == IB_WQS_RDY)) - if (_mlx4_ib_modify_wq(ibwq, IB_WQS_RESET)) + if (_mlx4_ib_modify_wq(ibwq, IB_WQS_RESET, + udata)) pr_warn("failed to reverse WQN=0x%06x\n", ibwq->wq_num); wq->rss_usecnt--; @@ -1996,7 +2004,8 @@ static int bringup_rss_rwqs(struct ib_rwq_ind_table *ind_tbl, u8 port_num) return err; } -static void bring_down_rss_rwqs(struct ib_rwq_ind_table *ind_tbl) +static void bring_down_rss_rwqs(struct ib_rwq_ind_table *ind_tbl, + struct ib_udata *udata) { int i; @@ -2007,7 +2016,7 @@ static void bring_down_rss_rwqs(struct ib_rwq_ind_table *ind_tbl) mutex_lock(&wq->mutex); if ((wq->rss_usecnt == 1) && (ibwq->state == IB_WQS_RDY)) - if (_mlx4_ib_modify_wq(ibwq, IB_WQS_RESET)) + if (_mlx4_ib_modify_wq(ibwq, IB_WQS_RESET, udata)) pr_warn("failed to reverse WQN=%x\n", ibwq->wq_num); wq->rss_usecnt--; @@ -2039,9 +2048,10 @@ static void fill_qp_rss_context(struct mlx4_qp_context *context, static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, const struct ib_qp_attr *attr, int attr_mask, - enum ib_qp_state cur_state, enum ib_qp_state new_state) + enum ib_qp_state cur_state, + enum ib_qp_state new_state, + struct ib_udata *udata) { - struct ib_uobject *ibuobject; struct ib_srq *ibsrq; const struct ib_gid_attr *gid_attr = NULL; struct ib_rwq_ind_table *rwq_ind_tbl; @@ -2050,6 +2060,8 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, struct mlx4_ib_qp *qp; struct mlx4_ib_pd *pd; struct mlx4_ib_cq *send_cq, *recv_cq; + struct mlx4_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx4_ib_ucontext, ibucontext); struct mlx4_qp_context *context; enum mlx4_qp_optpar optpar = 0; int sqd_event; @@ -2061,7 +2073,6 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, struct ib_wq *ibwq; ibwq = (struct ib_wq *)src; - ibuobject = ibwq->uobject; ibsrq = NULL; rwq_ind_tbl = NULL; qp_type = IB_QPT_RAW_PACKET; @@ -2072,7 +2083,6 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, struct ib_qp *ibqp; ibqp = (struct ib_qp *)src; - ibuobject = ibqp->uobject; ibsrq = ibqp->srq; rwq_ind_tbl = ibqp->rwq_ind_tbl; qp_type = ibqp->qp_type; @@ -2157,11 +2167,9 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, context->param3 |= cpu_to_be32(1 << 30); } - if (ibuobject) + if (ucontext) context->usr_page = cpu_to_be32( - mlx4_to_hw_uar_index(dev->dev, - to_mucontext(ibuobject->context) - ->uar.index)); + mlx4_to_hw_uar_index(dev->dev, ucontext->uar.index)); else context->usr_page = cpu_to_be32( mlx4_to_hw_uar_index(dev->dev, dev->priv_uar.index)); @@ -2293,7 +2301,7 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, context->cqn_recv = cpu_to_be32(recv_cq->mcq.cqn); /* Set "fast registration enabled" for all kernel QPs */ - if (!ibuobject) + if (!ucontext) context->params1 |= cpu_to_be32(1 << 11); if (attr_mask & IB_QP_RNR_RETRY) { @@ -2430,7 +2438,7 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, else sqd_event = 0; - if (!ibuobject && + if (!ucontext && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) context->rlkey_roce_mode |= (1 << 4); @@ -2441,7 +2449,7 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, * headroom is stamped so that the hardware doesn't start * processing stale work requests. */ - if (!ibuobject && + if (!ucontext && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) { struct mlx4_wqe_ctrl_seg *ctrl; @@ -2505,7 +2513,7 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type, * entries and reinitialize the QP. */ if (new_state == IB_QPS_RESET) { - if (!ibuobject) { + if (!ucontext) { mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn, ibsrq ? to_msrq(ibsrq) : NULL); if (send_cq != recv_cq) @@ -2731,16 +2739,17 @@ static int _mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, } if (ibqp->rwq_ind_tbl && (new_state == IB_QPS_INIT)) { - err = bringup_rss_rwqs(ibqp->rwq_ind_tbl, attr->port_num); + err = bringup_rss_rwqs(ibqp->rwq_ind_tbl, attr->port_num, + udata); if (err) goto out; } err = __mlx4_ib_modify_qp(ibqp, MLX4_IB_QP_SRC, attr, attr_mask, - cur_state, new_state); + cur_state, new_state, udata); if (ibqp->rwq_ind_tbl && err) - bring_down_rss_rwqs(ibqp->rwq_ind_tbl); + bring_down_rss_rwqs(ibqp->rwq_ind_tbl, udata); if (mlx4_is_bonded(dev->dev) && (attr_mask & IB_QP_PORT)) attr->port_num = 1; @@ -4118,7 +4127,8 @@ static int ib_wq2qp_state(enum ib_wq_state state) } } -static int _mlx4_ib_modify_wq(struct ib_wq *ibwq, enum ib_wq_state new_state) +static int _mlx4_ib_modify_wq(struct ib_wq *ibwq, enum ib_wq_state new_state, + struct ib_udata *udata) { struct mlx4_ib_qp *qp = to_mqp((struct ib_qp *)ibwq); enum ib_qp_state qp_cur_state; @@ -4142,7 +4152,8 @@ static int _mlx4_ib_modify_wq(struct ib_wq *ibwq, enum ib_wq_state new_state) attr_mask = IB_QP_PORT; err = __mlx4_ib_modify_qp(ibwq, MLX4_IB_RWQ_SRC, &attr, - attr_mask, IB_QPS_RESET, IB_QPS_INIT); + attr_mask, IB_QPS_RESET, IB_QPS_INIT, + udata); if (err) { pr_debug("WQN=0x%06x failed to apply RST->INIT on the HW QP\n", ibwq->wq_num); @@ -4154,12 +4165,13 @@ static int _mlx4_ib_modify_wq(struct ib_wq *ibwq, enum ib_wq_state new_state) attr_mask = 0; err = __mlx4_ib_modify_qp(ibwq, MLX4_IB_RWQ_SRC, NULL, attr_mask, - qp_cur_state, qp_new_state); + qp_cur_state, qp_new_state, udata); if (err && (qp_cur_state == IB_QPS_INIT)) { qp_new_state = IB_QPS_RESET; if (__mlx4_ib_modify_qp(ibwq, MLX4_IB_RWQ_SRC, NULL, - attr_mask, IB_QPS_INIT, IB_QPS_RESET)) { + attr_mask, IB_QPS_INIT, IB_QPS_RESET, + udata)) { pr_warn("WQN=0x%06x failed with reverting HW's resources failure\n", ibwq->wq_num); qp_new_state = IB_QPS_INIT; @@ -4222,7 +4234,7 @@ int mlx4_ib_modify_wq(struct ib_wq *ibwq, struct ib_wq_attr *wq_attr, * WQ, so we can apply its port on the WQ. */ if (qp->rss_usecnt) - err = _mlx4_ib_modify_wq(ibwq, new_state); + err = _mlx4_ib_modify_wq(ibwq, new_state, udata); if (!err) ibwq->state = new_state; diff --git a/drivers/infiniband/hw/mlx4/srq.c b/drivers/infiniband/hw/mlx4/srq.c index 4456f1b8921d..381cf899bcef 100644 --- a/drivers/infiniband/hw/mlx4/srq.c +++ b/drivers/infiniband/hw/mlx4/srq.c @@ -37,6 +37,7 @@ #include "mlx4_ib.h" #include <rdma/mlx4-abi.h> +#include <rdma/uverbs_ioctl.h> static void *get_wqe(struct mlx4_ib_srq *srq, int n) { @@ -73,6 +74,8 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd, struct ib_udata *udata) { struct mlx4_ib_dev *dev = to_mdev(pd->device); + struct mlx4_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx4_ib_ucontext, ibucontext); struct mlx4_ib_srq *srq; struct mlx4_wqe_srq_next_seg *next; struct mlx4_wqe_data_seg *scatter; @@ -113,8 +116,7 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd, goto err_srq; } - srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, - buf_size, 0, 0); + srq->umem = ib_umem_get(udata, ucmd.buf_addr, buf_size, 0, 0); if (IS_ERR(srq->umem)) { err = PTR_ERR(srq->umem); goto err_srq; @@ -129,8 +131,8 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd, if (err) goto err_mtt; - err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context), - ucmd.db_addr, &srq->db); + err = mlx4_ib_db_map_user(ucontext, udata, ucmd.db_addr, + &srq->db); if (err) goto err_mtt; } else { @@ -203,7 +205,7 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd, err_wrid: if (udata) - mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db); + mlx4_ib_db_unmap_user(ucontext, &srq->db); else kvfree(srq->wrid); diff --git a/drivers/infiniband/hw/mlx5/Kconfig b/drivers/infiniband/hw/mlx5/Kconfig index 0440966bc6ec..8d651c05de62 100644 --- a/drivers/infiniband/hw/mlx5/Kconfig +++ b/drivers/infiniband/hw/mlx5/Kconfig @@ -1,7 +1,6 @@ config MLX5_INFINIBAND tristate "Mellanox 5th generation network adapters (ConnectX series) support" depends on NETDEVICES && ETHERNET && PCI && MLX5_CORE - depends on INFINIBAND_USER_ACCESS || INFINIBAND_USER_ACCESS=n ---help--- This driver provides low-level InfiniBand support for Mellanox Connect-IB PCI Express host channel adapters (HCAs). diff --git a/drivers/infiniband/hw/mlx5/cong.c b/drivers/infiniband/hw/mlx5/cong.c index 7e4e358a4fd8..8ba439fabf7f 100644 --- a/drivers/infiniband/hw/mlx5/cong.c +++ b/drivers/infiniband/hw/mlx5/cong.c @@ -389,19 +389,19 @@ void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num) dev->port[port_num].dbg_cc_params = NULL; } -int mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num) +void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num) { struct mlx5_ib_dbg_cc_params *dbg_cc_params; struct mlx5_core_dev *mdev; int i; if (!mlx5_debugfs_root) - goto out; + return; /* Takes a 1-based port number */ mdev = mlx5_ib_get_native_port_mdev(dev, port_num + 1, NULL); if (!mdev) - goto out; + return; if (!MLX5_CAP_GEN(mdev, cc_query_allowed) || !MLX5_CAP_GEN(mdev, cc_modify_allowed)) @@ -415,8 +415,6 @@ int mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num) dbg_cc_params->root = debugfs_create_dir("cc_params", mdev->priv.dbg_root); - if (!dbg_cc_params->root) - goto err; for (i = 0; i < MLX5_IB_DBG_CC_MAX; i++) { dbg_cc_params->params[i].offset = i; @@ -427,14 +425,11 @@ int mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num) 0600, dbg_cc_params->root, &dbg_cc_params->params[i], &dbg_cc_fops); - if (!dbg_cc_params->params[i].dentry) - goto err; } put_mdev: mlx5_ib_put_native_port_mdev(dev, port_num + 1); -out: - return 0; + return; err: mlx5_ib_warn(dev, "cong debugfs failure\n"); @@ -445,5 +440,5 @@ err: * We don't want to fail driver if debugfs failed to initialize, * so we are not forwarding error to the user. */ - return 0; + return; } diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c index 90f1b0bae5b5..18704e503508 100644 --- a/drivers/infiniband/hw/mlx5/cq.c +++ b/drivers/infiniband/hw/mlx5/cq.c @@ -187,8 +187,8 @@ static void handle_responder(struct ib_wc *wc, struct mlx5_cqe64 *cqe, wqe_ctr = be16_to_cpu(cqe->wqe_counter); wc->wr_id = srq->wrid[wqe_ctr]; mlx5_ib_free_srq_wqe(srq, wqe_ctr); - if (msrq && atomic_dec_and_test(&msrq->refcount)) - complete(&msrq->free); + if (msrq) + mlx5_core_res_put(&msrq->common); } } else { wq = &qp->rq; @@ -707,15 +707,15 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata, *cqe_size = ucmd.cqe_size; - cq->buf.umem = ib_umem_get(context, ucmd.buf_addr, - entries * ucmd.cqe_size, - IB_ACCESS_LOCAL_WRITE, 1); + cq->buf.umem = + ib_umem_get(udata, ucmd.buf_addr, entries * ucmd.cqe_size, + IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(cq->buf.umem)) { err = PTR_ERR(cq->buf.umem); return err; } - err = mlx5_ib_db_map_user(to_mucontext(context), ucmd.db_addr, + err = mlx5_ib_db_map_user(to_mucontext(context), udata, ucmd.db_addr, &cq->db); if (err) goto err_umem; @@ -1111,7 +1111,6 @@ static int resize_user(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq, struct ib_umem *umem; int err; int npages; - struct ib_ucontext *context = cq->buf.umem->context; err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd)); if (err) @@ -1124,7 +1123,7 @@ static int resize_user(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq, if (ucmd.cqe_size && SIZE_MAX / ucmd.cqe_size <= entries - 1) return -EINVAL; - umem = ib_umem_get(context, ucmd.buf_addr, + umem = ib_umem_get(udata, ucmd.buf_addr, (size_t)ucmd.cqe_size * entries, IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(umem)) { diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c index 5a588f3cfb1b..eaa055007f28 100644 --- a/drivers/infiniband/hw/mlx5/devx.c +++ b/drivers/infiniband/hw/mlx5/devx.c @@ -8,6 +8,7 @@ #include <rdma/uverbs_types.h> #include <rdma/uverbs_ioctl.h> #include <rdma/mlx5_user_ioctl_cmds.h> +#include <rdma/mlx5_user_ioctl_verbs.h> #include <rdma/ib_umem.h> #include <rdma/uverbs_std_types.h> #include <linux/mlx5/driver.h> @@ -17,12 +18,28 @@ #define UVERBS_MODULE_NAME mlx5_ib #include <rdma/uverbs_named_ioctl.h> +enum devx_obj_flags { + DEVX_OBJ_FLAGS_INDIRECT_MKEY = 1 << 0, +}; + +struct devx_async_data { + struct mlx5_ib_dev *mdev; + struct list_head list; + struct ib_uobject *fd_uobj; + struct mlx5_async_work cb_work; + u16 cmd_out_len; + /* must be last field in this structure */ + struct mlx5_ib_uapi_devx_async_cmd_hdr hdr; +}; + #define MLX5_MAX_DESTROY_INBOX_SIZE_DW MLX5_ST_SZ_DW(delete_fte_in) struct devx_obj { struct mlx5_core_dev *mdev; u64 obj_id; u32 dinlen; /* destroy inbox length */ u32 dinbox[MLX5_MAX_DESTROY_INBOX_SIZE_DW]; + u32 flags; + struct mlx5_ib_devx_mr devx_mr; }; struct devx_umem { @@ -1011,6 +1028,92 @@ static void devx_obj_build_destroy_cmd(void *in, void *out, void *din, } } +static int devx_handle_mkey_indirect(struct devx_obj *obj, + struct mlx5_ib_dev *dev, + void *in, void *out) +{ + struct mlx5_mkey_table *table = &dev->mdev->priv.mkey_table; + struct mlx5_ib_devx_mr *devx_mr = &obj->devx_mr; + unsigned long flags; + struct mlx5_core_mkey *mkey; + void *mkc; + u8 key; + int err; + + mkey = &devx_mr->mmkey; + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + key = MLX5_GET(mkc, mkc, mkey_7_0); + mkey->key = mlx5_idx_to_mkey( + MLX5_GET(create_mkey_out, out, mkey_index)) | key; + mkey->type = MLX5_MKEY_INDIRECT_DEVX; + mkey->iova = MLX5_GET64(mkc, mkc, start_addr); + mkey->size = MLX5_GET64(mkc, mkc, len); + mkey->pd = MLX5_GET(mkc, mkc, pd); + devx_mr->ndescs = MLX5_GET(mkc, mkc, translations_octword_size); + + write_lock_irqsave(&table->lock, flags); + err = radix_tree_insert(&table->tree, mlx5_base_mkey(mkey->key), + mkey); + write_unlock_irqrestore(&table->lock, flags); + return err; +} + +static int devx_handle_mkey_create(struct mlx5_ib_dev *dev, + struct devx_obj *obj, + void *in, int in_len) +{ + int min_len = MLX5_BYTE_OFF(create_mkey_in, memory_key_mkey_entry) + + MLX5_FLD_SZ_BYTES(create_mkey_in, + memory_key_mkey_entry); + void *mkc; + u8 access_mode; + + if (in_len < min_len) + return -EINVAL; + + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + + access_mode = MLX5_GET(mkc, mkc, access_mode_1_0); + access_mode |= MLX5_GET(mkc, mkc, access_mode_4_2) << 2; + + if (access_mode == MLX5_MKC_ACCESS_MODE_KLMS || + access_mode == MLX5_MKC_ACCESS_MODE_KSM) { + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) + obj->flags |= DEVX_OBJ_FLAGS_INDIRECT_MKEY; + return 0; + } + + MLX5_SET(create_mkey_in, in, mkey_umem_valid, 1); + return 0; +} + +static void devx_free_indirect_mkey(struct rcu_head *rcu) +{ + kfree(container_of(rcu, struct devx_obj, devx_mr.rcu)); +} + +/* This function to delete from the radix tree needs to be called before + * destroying the underlying mkey. Otherwise a race might occur in case that + * other thread will get the same mkey before this one will be deleted, + * in that case it will fail via inserting to the tree its own data. + * + * Note: + * An error in the destroy is not expected unless there is some other indirect + * mkey which points to this one. In a kernel cleanup flow it will be just + * destroyed in the iterative destruction call. In a user flow, in case + * the application didn't close in the expected order it's its own problem, + * the mkey won't be part of the tree, in both cases the kernel is safe. + */ +static void devx_cleanup_mkey(struct devx_obj *obj) +{ + struct mlx5_mkey_table *table = &obj->mdev->priv.mkey_table; + unsigned long flags; + + write_lock_irqsave(&table->lock, flags); + radix_tree_delete(&table->tree, mlx5_base_mkey(obj->devx_mr.mmkey.key)); + write_unlock_irqrestore(&table->lock, flags); +} + static int devx_obj_cleanup(struct ib_uobject *uobject, enum rdma_remove_reason why) { @@ -1018,10 +1121,21 @@ static int devx_obj_cleanup(struct ib_uobject *uobject, struct devx_obj *obj = uobject->object; int ret; + if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) + devx_cleanup_mkey(obj); + ret = mlx5_cmd_exec(obj->mdev, obj->dinbox, obj->dinlen, out, sizeof(out)); if (ib_is_destroy_retryable(ret, why, uobject)) return ret; + if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) { + struct mlx5_ib_dev *dev = to_mdev(uobject->context->device); + + call_srcu(&dev->mr_srcu, &obj->devx_mr.rcu, + devx_free_indirect_mkey); + return ret; + } + kfree(obj); return ret; } @@ -1032,10 +1146,13 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)( void *cmd_in = uverbs_attr_get_alloced_ptr(attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_IN); int cmd_out_len = uverbs_attr_get_len(attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_OUT); + int cmd_in_len = uverbs_attr_get_len(attrs, + MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_IN); void *cmd_out; struct ib_uobject *uobj = uverbs_attr_get_uobject( attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_HANDLE); - struct mlx5_ib_ucontext *c = to_mucontext(uobj->context); + struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context( + &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext); struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device); u32 out[MLX5_ST_SZ_DW(general_obj_out_cmd_hdr)]; struct devx_obj *obj; @@ -1060,10 +1177,16 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)( return -ENOMEM; MLX5_SET(general_obj_in_cmd_hdr, cmd_in, uid, uid); - devx_set_umem_valid(cmd_in); + if (opcode == MLX5_CMD_OP_CREATE_MKEY) { + err = devx_handle_mkey_create(dev, obj, cmd_in, cmd_in_len); + if (err) + goto obj_free; + } else { + devx_set_umem_valid(cmd_in); + } err = mlx5_cmd_exec(dev->mdev, cmd_in, - uverbs_attr_get_len(attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_IN), + cmd_in_len, cmd_out, cmd_out_len); if (err) goto obj_free; @@ -1074,13 +1197,22 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)( &obj_id); WARN_ON(obj->dinlen > MLX5_MAX_DESTROY_INBOX_SIZE_DW * sizeof(u32)); + if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) { + err = devx_handle_mkey_indirect(obj, dev, cmd_in, cmd_out); + if (err) + goto obj_destroy; + } + err = uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_OUT, cmd_out, cmd_out_len); if (err) - goto obj_destroy; + goto err_copy; obj->obj_id = get_enc_obj_id(opcode, obj_id); return 0; +err_copy: + if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) + devx_cleanup_mkey(obj); obj_destroy: mlx5_cmd_exec(obj->mdev, obj->dinbox, obj->dinlen, out, sizeof(out)); obj_free: @@ -1096,8 +1228,9 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_MODIFY)( MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_OUT); struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs, MLX5_IB_ATTR_DEVX_OBJ_MODIFY_HANDLE); - struct mlx5_ib_ucontext *c = to_mucontext(uobj->context); - struct mlx5_ib_dev *mdev = to_mdev(uobj->context->device); + struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context( + &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext); + struct mlx5_ib_dev *mdev = to_mdev(c->ibucontext.device); void *cmd_out; int err; int uid; @@ -1137,11 +1270,12 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_QUERY)( MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_OUT); struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs, MLX5_IB_ATTR_DEVX_OBJ_QUERY_HANDLE); - struct mlx5_ib_ucontext *c = to_mucontext(uobj->context); + struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context( + &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext); void *cmd_out; int err; int uid; - struct mlx5_ib_dev *mdev = to_mdev(uobj->context->device); + struct mlx5_ib_dev *mdev = to_mdev(c->ibucontext.device); uid = devx_get_uid(c, cmd_in); if (uid < 0) @@ -1168,6 +1302,154 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_QUERY)( cmd_out, cmd_out_len); } +struct devx_async_event_queue { + spinlock_t lock; + wait_queue_head_t poll_wait; + struct list_head event_list; + atomic_t bytes_in_use; + u8 is_destroyed:1; +}; + +struct devx_async_cmd_event_file { + struct ib_uobject uobj; + struct devx_async_event_queue ev_queue; + struct mlx5_async_ctx async_ctx; +}; + +static void devx_init_event_queue(struct devx_async_event_queue *ev_queue) +{ + spin_lock_init(&ev_queue->lock); + INIT_LIST_HEAD(&ev_queue->event_list); + init_waitqueue_head(&ev_queue->poll_wait); + atomic_set(&ev_queue->bytes_in_use, 0); + ev_queue->is_destroyed = 0; +} + +static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC)( + struct uverbs_attr_bundle *attrs) +{ + struct devx_async_cmd_event_file *ev_file; + + struct ib_uobject *uobj = uverbs_attr_get_uobject( + attrs, MLX5_IB_ATTR_DEVX_ASYNC_CMD_FD_ALLOC_HANDLE); + struct mlx5_ib_dev *mdev = to_mdev(uobj->context->device); + + ev_file = container_of(uobj, struct devx_async_cmd_event_file, + uobj); + devx_init_event_queue(&ev_file->ev_queue); + mlx5_cmd_init_async_ctx(mdev->mdev, &ev_file->async_ctx); + return 0; +} + +static void devx_query_callback(int status, struct mlx5_async_work *context) +{ + struct devx_async_data *async_data = + container_of(context, struct devx_async_data, cb_work); + struct ib_uobject *fd_uobj = async_data->fd_uobj; + struct devx_async_cmd_event_file *ev_file; + struct devx_async_event_queue *ev_queue; + unsigned long flags; + + ev_file = container_of(fd_uobj, struct devx_async_cmd_event_file, + uobj); + ev_queue = &ev_file->ev_queue; + + spin_lock_irqsave(&ev_queue->lock, flags); + list_add_tail(&async_data->list, &ev_queue->event_list); + spin_unlock_irqrestore(&ev_queue->lock, flags); + + wake_up_interruptible(&ev_queue->poll_wait); + fput(fd_uobj->object); +} + +#define MAX_ASYNC_BYTES_IN_USE (1024 * 1024) /* 1MB */ + +static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY)( + struct uverbs_attr_bundle *attrs) +{ + void *cmd_in = uverbs_attr_get_alloced_ptr(attrs, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_CMD_IN); + struct ib_uobject *uobj = uverbs_attr_get_uobject( + attrs, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_HANDLE); + u16 cmd_out_len; + struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context( + &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext); + struct ib_uobject *fd_uobj; + int err; + int uid; + struct mlx5_ib_dev *mdev = to_mdev(c->ibucontext.device); + struct devx_async_cmd_event_file *ev_file; + struct devx_async_data *async_data; + + uid = devx_get_uid(c, cmd_in); + if (uid < 0) + return uid; + + if (!devx_is_obj_query_cmd(cmd_in)) + return -EINVAL; + + err = uverbs_get_const(&cmd_out_len, attrs, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_OUT_LEN); + if (err) + return err; + + if (!devx_is_valid_obj_id(uobj, cmd_in)) + return -EINVAL; + + fd_uobj = uverbs_attr_get_uobject(attrs, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_FD); + if (IS_ERR(fd_uobj)) + return PTR_ERR(fd_uobj); + + ev_file = container_of(fd_uobj, struct devx_async_cmd_event_file, + uobj); + + if (atomic_add_return(cmd_out_len, &ev_file->ev_queue.bytes_in_use) > + MAX_ASYNC_BYTES_IN_USE) { + atomic_sub(cmd_out_len, &ev_file->ev_queue.bytes_in_use); + return -EAGAIN; + } + + async_data = kvzalloc(struct_size(async_data, hdr.out_data, + cmd_out_len), GFP_KERNEL); + if (!async_data) { + err = -ENOMEM; + goto sub_bytes; + } + + err = uverbs_copy_from(&async_data->hdr.wr_id, attrs, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_WR_ID); + if (err) + goto free_async; + + async_data->cmd_out_len = cmd_out_len; + async_data->mdev = mdev; + async_data->fd_uobj = fd_uobj; + + get_file(fd_uobj->object); + MLX5_SET(general_obj_in_cmd_hdr, cmd_in, uid, uid); + err = mlx5_cmd_exec_cb(&ev_file->async_ctx, cmd_in, + uverbs_attr_get_len(attrs, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_CMD_IN), + async_data->hdr.out_data, + async_data->cmd_out_len, + devx_query_callback, &async_data->cb_work); + + if (err) + goto cb_err; + + return 0; + +cb_err: + fput(fd_uobj->object); +free_async: + kvfree(async_data); +sub_bytes: + atomic_sub(cmd_out_len, &ev_file->ev_queue.bytes_in_use); + return err; +} + static int devx_umem_get(struct mlx5_ib_dev *dev, struct ib_ucontext *ucontext, struct uverbs_attr_bundle *attrs, struct devx_umem *obj) @@ -1195,7 +1477,7 @@ static int devx_umem_get(struct mlx5_ib_dev *dev, struct ib_ucontext *ucontext, if (err) return err; - obj->umem = ib_umem_get(ucontext, addr, size, access, 0); + obj->umem = ib_umem_get(&attrs->driver_udata, addr, size, access, 0); if (IS_ERR(obj->umem)) return PTR_ERR(obj->umem); @@ -1252,7 +1534,8 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_UMEM_REG)( struct ib_uobject *uobj = uverbs_attr_get_uobject( attrs, MLX5_IB_ATTR_DEVX_UMEM_REG_HANDLE); u32 obj_id; - struct mlx5_ib_ucontext *c = to_mucontext(uobj->context); + struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context( + &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext); struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device); int err; @@ -1313,6 +1596,123 @@ static int devx_umem_cleanup(struct ib_uobject *uobject, return 0; } +static ssize_t devx_async_cmd_event_read(struct file *filp, char __user *buf, + size_t count, loff_t *pos) +{ + struct devx_async_cmd_event_file *comp_ev_file = filp->private_data; + struct devx_async_event_queue *ev_queue = &comp_ev_file->ev_queue; + struct devx_async_data *event; + int ret = 0; + size_t eventsz; + + spin_lock_irq(&ev_queue->lock); + + while (list_empty(&ev_queue->event_list)) { + spin_unlock_irq(&ev_queue->lock); + + if (filp->f_flags & O_NONBLOCK) + return -EAGAIN; + + if (wait_event_interruptible( + ev_queue->poll_wait, + (!list_empty(&ev_queue->event_list) || + ev_queue->is_destroyed))) { + return -ERESTARTSYS; + } + + if (list_empty(&ev_queue->event_list) && + ev_queue->is_destroyed) + return -EIO; + + spin_lock_irq(&ev_queue->lock); + } + + event = list_entry(ev_queue->event_list.next, + struct devx_async_data, list); + eventsz = event->cmd_out_len + + sizeof(struct mlx5_ib_uapi_devx_async_cmd_hdr); + + if (eventsz > count) { + spin_unlock_irq(&ev_queue->lock); + return -ENOSPC; + } + + list_del(ev_queue->event_list.next); + spin_unlock_irq(&ev_queue->lock); + + if (copy_to_user(buf, &event->hdr, eventsz)) + ret = -EFAULT; + else + ret = eventsz; + + atomic_sub(event->cmd_out_len, &ev_queue->bytes_in_use); + kvfree(event); + return ret; +} + +static int devx_async_cmd_event_close(struct inode *inode, struct file *filp) +{ + struct ib_uobject *uobj = filp->private_data; + struct devx_async_cmd_event_file *comp_ev_file = container_of( + uobj, struct devx_async_cmd_event_file, uobj); + struct devx_async_data *entry, *tmp; + + spin_lock_irq(&comp_ev_file->ev_queue.lock); + list_for_each_entry_safe(entry, tmp, + &comp_ev_file->ev_queue.event_list, list) + kvfree(entry); + spin_unlock_irq(&comp_ev_file->ev_queue.lock); + + uverbs_close_fd(filp); + return 0; +} + +static __poll_t devx_async_cmd_event_poll(struct file *filp, + struct poll_table_struct *wait) +{ + struct devx_async_cmd_event_file *comp_ev_file = filp->private_data; + struct devx_async_event_queue *ev_queue = &comp_ev_file->ev_queue; + __poll_t pollflags = 0; + + poll_wait(filp, &ev_queue->poll_wait, wait); + + spin_lock_irq(&ev_queue->lock); + if (ev_queue->is_destroyed) + pollflags = EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; + else if (!list_empty(&ev_queue->event_list)) + pollflags = EPOLLIN | EPOLLRDNORM; + spin_unlock_irq(&ev_queue->lock); + + return pollflags; +} + +const struct file_operations devx_async_cmd_event_fops = { + .owner = THIS_MODULE, + .read = devx_async_cmd_event_read, + .poll = devx_async_cmd_event_poll, + .release = devx_async_cmd_event_close, + .llseek = no_llseek, +}; + +static int devx_hot_unplug_async_cmd_event_file(struct ib_uobject *uobj, + enum rdma_remove_reason why) +{ + struct devx_async_cmd_event_file *comp_ev_file = + container_of(uobj, struct devx_async_cmd_event_file, + uobj); + struct devx_async_event_queue *ev_queue = &comp_ev_file->ev_queue; + + spin_lock_irq(&ev_queue->lock); + ev_queue->is_destroyed = 1; + spin_unlock_irq(&ev_queue->lock); + + if (why == RDMA_REMOVE_DRIVER_REMOVE) + wake_up_interruptible(&ev_queue->poll_wait); + + mlx5_cmd_cleanup_async_ctx(&comp_ev_file->async_ctx); + return 0; +}; + DECLARE_UVERBS_NAMED_METHOD( MLX5_IB_METHOD_DEVX_UMEM_REG, UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_UMEM_REG_HANDLE, @@ -1423,6 +1823,27 @@ DECLARE_UVERBS_NAMED_METHOD( UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_out_cmd_hdr)), UA_MANDATORY)); +DECLARE_UVERBS_NAMED_METHOD( + MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY, + UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_OBJ_QUERY_HANDLE, + UVERBS_IDR_ANY_OBJECT, + UVERBS_ACCESS_READ, + UA_MANDATORY), + UVERBS_ATTR_PTR_IN( + MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_IN, + UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_in_cmd_hdr)), + UA_MANDATORY, + UA_ALLOC_AND_COPY), + UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_OUT_LEN, + u16, UA_MANDATORY), + UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_FD, + MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD, + UVERBS_ACCESS_READ, + UA_MANDATORY), + UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_WR_ID, + UVERBS_ATTR_TYPE(u64), + UA_MANDATORY)); + DECLARE_UVERBS_GLOBAL_METHODS(MLX5_IB_OBJECT_DEVX, &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OTHER), &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_QUERY_UAR), @@ -1433,13 +1854,30 @@ DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_DEVX_OBJ, &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_CREATE), &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_DESTROY), &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_MODIFY), - &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_QUERY)); + &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_QUERY), + &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY)); DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_DEVX_UMEM, UVERBS_TYPE_ALLOC_IDR(devx_umem_cleanup), &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_UMEM_REG), &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_UMEM_DEREG)); + +DECLARE_UVERBS_NAMED_METHOD( + MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC, + UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_ASYNC_CMD_FD_ALLOC_HANDLE, + MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD, + UVERBS_ACCESS_NEW, + UA_MANDATORY)); + +DECLARE_UVERBS_NAMED_OBJECT( + MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD, + UVERBS_TYPE_ALLOC_FD(sizeof(struct devx_async_cmd_event_file), + devx_hot_unplug_async_cmd_event_file, + &devx_async_cmd_event_fops, "[devx_async_cmd]", + O_RDONLY), + &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC)); + static bool devx_is_supported(struct ib_device *device) { struct mlx5_ib_dev *dev = to_mdev(device); @@ -1457,5 +1895,8 @@ const struct uapi_definition mlx5_ib_devx_defs[] = { UAPI_DEF_CHAIN_OBJ_TREE_NAMED( MLX5_IB_OBJECT_DEVX_UMEM, UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)), + UAPI_DEF_CHAIN_OBJ_TREE_NAMED( + MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD, + UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)), {}, }; diff --git a/drivers/infiniband/hw/mlx5/doorbell.c b/drivers/infiniband/hw/mlx5/doorbell.c index a0e4e6ddb71a..8f4e5f22b84c 100644 --- a/drivers/infiniband/hw/mlx5/doorbell.c +++ b/drivers/infiniband/hw/mlx5/doorbell.c @@ -43,7 +43,8 @@ struct mlx5_ib_user_db_page { int refcnt; }; -int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt, +int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, + struct ib_udata *udata, unsigned long virt, struct mlx5_db *db) { struct mlx5_ib_user_db_page *page; @@ -63,8 +64,7 @@ int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt, page->user_virt = (virt & PAGE_MASK); page->refcnt = 0; - page->umem = ib_umem_get(&context->ibucontext, virt & PAGE_MASK, - PAGE_SIZE, 0, 0); + page->umem = ib_umem_get(udata, virt & PAGE_MASK, PAGE_SIZE, 0, 0); if (IS_ERR(page->umem)) { err = PTR_ERR(page->umem); kfree(page); diff --git a/drivers/infiniband/hw/mlx5/ib_rep.c b/drivers/infiniband/hw/mlx5/ib_rep.c index 4700cffb5a00..b8639ac71336 100644 --- a/drivers/infiniband/hw/mlx5/ib_rep.c +++ b/drivers/infiniband/hw/mlx5/ib_rep.c @@ -57,7 +57,7 @@ mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep) else profile = &vf_rep_profile; - ibdev = (struct mlx5_ib_dev *)ib_alloc_device(sizeof(*ibdev)); + ibdev = ib_alloc_device(mlx5_ib_dev, ib_dev); if (!ibdev) return -ENOMEM; @@ -65,8 +65,10 @@ mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep) ibdev->mdev = dev; ibdev->num_ports = max(MLX5_CAP_GEN(dev, num_ports), MLX5_CAP_GEN(dev, num_vhca_ports)); - if (!__mlx5_ib_add(ibdev, profile)) + if (!__mlx5_ib_add(ibdev, profile)) { + ib_dealloc_device(&ibdev->ib_dev); return -EINVAL; + } rep->rep_if[REP_IB].priv = ibdev; diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 581ae11e2fc9..994c19d01211 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -535,24 +535,51 @@ out: return err; } +struct mlx5_ib_vlan_info { + u16 vlan_id; + bool vlan; +}; + +static int get_lower_dev_vlan(struct net_device *lower_dev, void *data) +{ + struct mlx5_ib_vlan_info *vlan_info = data; + + if (is_vlan_dev(lower_dev)) { + vlan_info->vlan = true; + vlan_info->vlan_id = vlan_dev_vlan_id(lower_dev); + } + /* We are interested only in first level vlan device, so + * always return 1 to stop iterating over next level devices. + */ + return 1; +} + static int set_roce_addr(struct mlx5_ib_dev *dev, u8 port_num, unsigned int index, const union ib_gid *gid, const struct ib_gid_attr *attr) { enum ib_gid_type gid_type = IB_GID_TYPE_IB; + struct mlx5_ib_vlan_info vlan_info = { }; u8 roce_version = 0; u8 roce_l3_type = 0; - bool vlan = false; u8 mac[ETH_ALEN]; - u16 vlan_id = 0; if (gid) { gid_type = attr->gid_type; ether_addr_copy(mac, attr->ndev->dev_addr); if (is_vlan_dev(attr->ndev)) { - vlan = true; - vlan_id = vlan_dev_vlan_id(attr->ndev); + vlan_info.vlan = true; + vlan_info.vlan_id = vlan_dev_vlan_id(attr->ndev); + } else { + /* If the netdev is upper device and if it's lower + * lower device is vlan device, consider vlan id of + * the lower vlan device for this gid entry. + */ + rcu_read_lock(); + netdev_walk_all_lower_dev_rcu(attr->ndev, + get_lower_dev_vlan, &vlan_info); + rcu_read_unlock(); } } @@ -573,8 +600,9 @@ static int set_roce_addr(struct mlx5_ib_dev *dev, u8 port_num, } return mlx5_core_roce_gid_set(dev->mdev, index, roce_version, - roce_l3_type, gid->raw, mac, vlan, - vlan_id, port_num); + roce_l3_type, gid->raw, mac, + vlan_info.vlan, vlan_info.vlan_id, + port_num); } static int mlx5_ib_add_gid(const struct ib_gid_attr *attr, @@ -982,11 +1010,11 @@ static int mlx5_ib_query_device(struct ib_device *ibdev, props->hca_core_clock = MLX5_CAP_GEN(mdev, device_frequency_khz); props->timestamp_mask = 0x7FFFFFFFFFFFFFFFULL; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - if (MLX5_CAP_GEN(mdev, pg)) - props->device_cap_flags |= IB_DEVICE_ON_DEMAND_PAGING; - props->odp_caps = dev->odp_caps; -#endif + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) { + if (MLX5_CAP_GEN(mdev, pg)) + props->device_cap_flags |= IB_DEVICE_ON_DEMAND_PAGING; + props->odp_caps = dev->odp_caps; + } if (MLX5_CAP_GEN(mdev, cd)) props->device_cap_flags |= IB_DEVICE_CROSS_CHANNEL; @@ -1717,14 +1745,15 @@ static void mlx5_ib_dealloc_transport_domain(struct mlx5_ib_dev *dev, u32 tdn, mlx5_ib_disable_lb(dev, true, false); } -static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx, + struct ib_udata *udata) { + struct ib_device *ibdev = uctx->device; struct mlx5_ib_dev *dev = to_mdev(ibdev); struct mlx5_ib_alloc_ucontext_req_v2 req = {}; struct mlx5_ib_alloc_ucontext_resp resp = {}; struct mlx5_core_dev *mdev = dev->mdev; - struct mlx5_ib_ucontext *context; + struct mlx5_ib_ucontext *context = to_mucontext(uctx); struct mlx5_bfreg_info *bfregi; int ver; int err; @@ -1734,29 +1763,29 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev, bool lib_uar_4k; if (!dev->ib_active) - return ERR_PTR(-EAGAIN); + return -EAGAIN; if (udata->inlen == sizeof(struct mlx5_ib_alloc_ucontext_req)) ver = 0; else if (udata->inlen >= min_req_v2) ver = 2; else - return ERR_PTR(-EINVAL); + return -EINVAL; err = ib_copy_from_udata(&req, udata, min(udata->inlen, sizeof(req))); if (err) - return ERR_PTR(err); + return err; if (req.flags & ~MLX5_IB_ALLOC_UCTX_DEVX) - return ERR_PTR(-EOPNOTSUPP); + return -EOPNOTSUPP; if (req.comp_mask || req.reserved0 || req.reserved1 || req.reserved2) - return ERR_PTR(-EOPNOTSUPP); + return -EOPNOTSUPP; req.total_num_bfregs = ALIGN(req.total_num_bfregs, MLX5_NON_FP_BFREGS_PER_UAR); if (req.num_low_latency_bfregs > req.total_num_bfregs - 1) - return ERR_PTR(-EINVAL); + return -EINVAL; resp.qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp); if (mlx5_core_is_pf(dev->mdev) && MLX5_CAP_GEN(dev->mdev, bf)) @@ -1789,10 +1818,6 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev, /* MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_FULL_OFFLOAD is currently always 0 */ } - context = kzalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); - lib_uar_4k = req.lib_caps & MLX5_LIB_CAP_4K_UAR; bfregi = &context->bfregi; @@ -1822,9 +1847,9 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev, if (err) goto out_sys_pages; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - context->ibucontext.invalidate_range = &mlx5_ib_invalidate_range; -#endif + if (ibdev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING) + context->ibucontext.invalidate_range = + &mlx5_ib_invalidate_range; if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX) { err = mlx5_ib_devx_create(dev, true); @@ -1927,7 +1952,7 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev, 1, &dev->roce[port].tx_port_affinity)); } - return &context->ibucontext; + return 0; out_mdev: mlx5_ib_dealloc_transport_domain(dev, context->tdn, context->devx_uid); @@ -1945,23 +1970,19 @@ out_count: kfree(bfregi->count); out_ctx: - kfree(context); - - return ERR_PTR(err); + return err; } -static int mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) +static void mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) { struct mlx5_ib_ucontext *context = to_mucontext(ibcontext); struct mlx5_ib_dev *dev = to_mdev(ibcontext->device); struct mlx5_bfreg_info *bfregi; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING /* All umem's must be destroyed before destroying the ucontext. */ mutex_lock(&ibcontext->per_mm_list_lock); WARN_ON(!list_empty(&ibcontext->per_mm_list)); mutex_unlock(&ibcontext->per_mm_list_lock); -#endif bfregi = &context->bfregi; mlx5_ib_dealloc_transport_domain(dev, context->tdn, context->devx_uid); @@ -1972,9 +1993,6 @@ static int mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) deallocate_uars(dev, context); kfree(bfregi->sys_pages); kfree(bfregi->count); - kfree(context); - - return 0; } static phys_addr_t uar_index2pfn(struct mlx5_ib_dev *dev, @@ -2313,30 +2331,24 @@ int mlx5_ib_dealloc_dm(struct ib_dm *ibdm) return 0; } -static struct ib_pd *mlx5_ib_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +static int mlx5_ib_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { + struct mlx5_ib_pd *pd = to_mpd(ibpd); + struct ib_device *ibdev = ibpd->device; struct mlx5_ib_alloc_pd_resp resp; - struct mlx5_ib_pd *pd; int err; u32 out[MLX5_ST_SZ_DW(alloc_pd_out)] = {}; u32 in[MLX5_ST_SZ_DW(alloc_pd_in)] = {}; u16 uid = 0; - pd = kmalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - uid = context ? to_mucontext(context)->devx_uid : 0; MLX5_SET(alloc_pd_in, in, opcode, MLX5_CMD_OP_ALLOC_PD); MLX5_SET(alloc_pd_in, in, uid, uid); err = mlx5_cmd_exec(to_mdev(ibdev)->mdev, in, sizeof(in), out, sizeof(out)); - if (err) { - kfree(pd); - return ERR_PTR(err); - } + if (err) + return err; pd->pdn = MLX5_GET(alloc_pd_out, out, pd); pd->uid = uid; @@ -2344,23 +2356,19 @@ static struct ib_pd *mlx5_ib_alloc_pd(struct ib_device *ibdev, resp.pdn = pd->pdn; if (ib_copy_to_udata(udata, &resp, sizeof(resp))) { mlx5_cmd_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn, uid); - kfree(pd); - return ERR_PTR(-EFAULT); + return -EFAULT; } } - return &pd->ibpd; + return 0; } -static int mlx5_ib_dealloc_pd(struct ib_pd *pd) +static void mlx5_ib_dealloc_pd(struct ib_pd *pd) { struct mlx5_ib_dev *mdev = to_mdev(pd->device); struct mlx5_ib_pd *mpd = to_mpd(pd); mlx5_cmd_dealloc_pd(mdev->mdev, mpd->pdn, mpd->uid); - kfree(mpd); - - return 0; } enum { @@ -2394,10 +2402,29 @@ static u8 get_match_criteria_enable(u32 *match_criteria) return match_criteria_enable; } -static void set_proto(void *outer_c, void *outer_v, u8 mask, u8 val) +static int set_proto(void *outer_c, void *outer_v, u8 mask, u8 val) { - MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_protocol, mask); - MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_protocol, val); + u8 entry_mask; + u8 entry_val; + int err = 0; + + if (!mask) + goto out; + + entry_mask = MLX5_GET(fte_match_set_lyr_2_4, outer_c, + ip_protocol); + entry_val = MLX5_GET(fte_match_set_lyr_2_4, outer_v, + ip_protocol); + if (!entry_mask) { + MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_protocol, mask); + MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_protocol, val); + goto out; + } + /* Don't override existing ip protocol */ + if (mask != entry_mask || val != entry_val) + err = -EINVAL; +out: + return err; } static void set_flow_label(void *misc_c, void *misc_v, u32 mask, u32 val, @@ -2631,8 +2658,10 @@ static int parse_flow_attr(struct mlx5_core_dev *mdev, u32 *match_c, set_tos(headers_c, headers_v, ib_spec->ipv4.mask.tos, ib_spec->ipv4.val.tos); - set_proto(headers_c, headers_v, - ib_spec->ipv4.mask.proto, ib_spec->ipv4.val.proto); + if (set_proto(headers_c, headers_v, + ib_spec->ipv4.mask.proto, + ib_spec->ipv4.val.proto)) + return -EINVAL; break; case IB_FLOW_SPEC_IPV6: if (FIELDS_NOT_SUPPORTED(ib_spec->ipv6.mask, LAST_IPV6_FIELD)) @@ -2671,9 +2700,10 @@ static int parse_flow_attr(struct mlx5_core_dev *mdev, u32 *match_c, ib_spec->ipv6.mask.traffic_class, ib_spec->ipv6.val.traffic_class); - set_proto(headers_c, headers_v, - ib_spec->ipv6.mask.next_hdr, - ib_spec->ipv6.val.next_hdr); + if (set_proto(headers_c, headers_v, + ib_spec->ipv6.mask.next_hdr, + ib_spec->ipv6.val.next_hdr)) + return -EINVAL; set_flow_label(misc_params_c, misc_params_v, ntohl(ib_spec->ipv6.mask.flow_label), @@ -2694,10 +2724,8 @@ static int parse_flow_attr(struct mlx5_core_dev *mdev, u32 *match_c, LAST_TCP_UDP_FIELD)) return -EOPNOTSUPP; - MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol, - 0xff); - MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, - IPPROTO_TCP); + if (set_proto(headers_c, headers_v, 0xff, IPPROTO_TCP)) + return -EINVAL; MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_sport, ntohs(ib_spec->tcp_udp.mask.src_port)); @@ -2714,10 +2742,8 @@ static int parse_flow_attr(struct mlx5_core_dev *mdev, u32 *match_c, LAST_TCP_UDP_FIELD)) return -EOPNOTSUPP; - MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol, - 0xff); - MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, - IPPROTO_UDP); + if (set_proto(headers_c, headers_v, 0xff, IPPROTO_UDP)) + return -EINVAL; MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_sport, ntohs(ib_spec->tcp_udp.mask.src_port)); @@ -2733,6 +2759,9 @@ static int parse_flow_attr(struct mlx5_core_dev *mdev, u32 *match_c, if (ib_spec->gre.mask.c_ks_res0_ver) return -EOPNOTSUPP; + if (set_proto(headers_c, headers_v, 0xff, IPPROTO_GRE)) + return -EINVAL; + MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol, 0xff); MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, @@ -3884,7 +3913,7 @@ mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev, if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO) return ERR_PTR(-ENOMEM); - dst = kzalloc(sizeof(*dst) * 2, GFP_KERNEL); + dst = kcalloc(2, sizeof(*dst), GFP_KERNEL); if (!dst) return ERR_PTR(-ENOMEM); @@ -4165,7 +4194,7 @@ static ssize_t fw_pages_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx5_ib_dev *dev = - container_of(device, struct mlx5_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev); return sprintf(buf, "%d\n", dev->mdev->priv.fw_pages); } @@ -4175,7 +4204,7 @@ static ssize_t reg_pages_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx5_ib_dev *dev = - container_of(device, struct mlx5_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev); return sprintf(buf, "%d\n", atomic_read(&dev->mdev->priv.reg_pages)); } @@ -4185,7 +4214,8 @@ static ssize_t hca_type_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx5_ib_dev *dev = - container_of(device, struct mlx5_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev); + return sprintf(buf, "MT%d\n", dev->mdev->pdev->device); } static DEVICE_ATTR_RO(hca_type); @@ -4194,7 +4224,8 @@ static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx5_ib_dev *dev = - container_of(device, struct mlx5_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev); + return sprintf(buf, "%x\n", dev->mdev->rev_id); } static DEVICE_ATTR_RO(hw_rev); @@ -4203,7 +4234,8 @@ static ssize_t board_id_show(struct device *device, struct device_attribute *attr, char *buf) { struct mlx5_ib_dev *dev = - container_of(device, struct mlx5_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev); + return sprintf(buf, "%.*s\n", MLX5_BOARD_ID_LEN, dev->mdev->board_id); } @@ -4689,23 +4721,28 @@ static int create_dev_resources(struct mlx5_ib_resources *devr) { struct ib_srq_init_attr attr; struct mlx5_ib_dev *dev; + struct ib_device *ibdev; struct ib_cq_init_attr cq_attr = {.cqe = 1}; int port; int ret = 0; dev = container_of(devr, struct mlx5_ib_dev, devr); + ibdev = &dev->ib_dev; mutex_init(&devr->mutex); - devr->p0 = mlx5_ib_alloc_pd(&dev->ib_dev, NULL, NULL); - if (IS_ERR(devr->p0)) { - ret = PTR_ERR(devr->p0); - goto error0; - } - devr->p0->device = &dev->ib_dev; + devr->p0 = rdma_zalloc_drv_obj(ibdev, ib_pd); + if (!devr->p0) + return -ENOMEM; + + devr->p0->device = ibdev; devr->p0->uobject = NULL; atomic_set(&devr->p0->usecnt, 0); + ret = mlx5_ib_alloc_pd(devr->p0, NULL, NULL); + if (ret) + goto error0; + devr->c0 = mlx5_ib_create_cq(&dev->ib_dev, &cq_attr, NULL, NULL); if (IS_ERR(devr->c0)) { ret = PTR_ERR(devr->c0); @@ -4803,6 +4840,7 @@ error2: error1: mlx5_ib_dealloc_pd(devr->p0); error0: + kfree(devr->p0); return ret; } @@ -4818,6 +4856,7 @@ static void destroy_dev_resources(struct mlx5_ib_resources *devr) mlx5_ib_dealloc_xrcd(devr->x1); mlx5_ib_destroy_cq(devr->c0); mlx5_ib_dealloc_pd(devr->p0); + kfree(devr->p0); /* Make sure no change P_Key work items are still executing */ for (port = 0; port < dev->num_ports; ++port) @@ -5567,9 +5606,7 @@ static bool mlx5_ib_bind_slave_port(struct mlx5_ib_dev *ibdev, mpi->mdev_events.notifier_call = mlx5_ib_event_slave_port; mlx5_notifier_register(mpi->mdev, &mpi->mdev_events); - err = mlx5_ib_init_cong_debugfs(ibdev, port_num); - if (err) - goto unbind; + mlx5_ib_init_cong_debugfs(ibdev, port_num); return true; @@ -5781,11 +5818,10 @@ static struct ib_counters *mlx5_ib_create_counters(struct ib_device *device, void mlx5_ib_stage_init_cleanup(struct mlx5_ib_dev *dev) { mlx5_ib_cleanup_multiport_master(dev); -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - cleanup_srcu_struct(&dev->mr_srcu); - drain_workqueue(dev->advise_mr_wq); - destroy_workqueue(dev->advise_mr_wq); -#endif + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) { + srcu_barrier(&dev->mr_srcu); + cleanup_srcu_struct(&dev->mr_srcu); + } kfree(dev->port); } @@ -5838,19 +5874,11 @@ int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev) spin_lock_init(&dev->memic.memic_lock); dev->memic.dev = mdev; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - dev->advise_mr_wq = alloc_ordered_workqueue("mlx5_ib_advise_mr_wq", 0); - if (!dev->advise_mr_wq) { - err = -ENOMEM; - goto err_mp; - } - - err = init_srcu_struct(&dev->mr_srcu); - if (err) { - destroy_workqueue(dev->advise_mr_wq); - goto err_mp; + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) { + err = init_srcu_struct(&dev->mr_srcu); + if (err) + goto err_mp; } -#endif return 0; err_mp: @@ -5947,6 +5975,8 @@ static const struct ib_device_ops mlx5_ib_dev_ops = { .req_notify_cq = mlx5_ib_arm_cq, .rereg_user_mr = mlx5_ib_rereg_user_mr, .resize_cq = mlx5_ib_resize_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, mlx5_ib_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx5_ib_ucontext, ibucontext), }; static const struct ib_device_ops mlx5_ib_dev_flow_ipsec_ops = { @@ -6213,7 +6243,7 @@ static int mlx5_ib_stage_odp_init(struct mlx5_ib_dev *dev) return mlx5_ib_odp_init_one(dev); } -void mlx5_ib_stage_odp_cleanup(struct mlx5_ib_dev *dev) +static void mlx5_ib_stage_odp_cleanup(struct mlx5_ib_dev *dev) { mlx5_ib_odp_cleanup_one(dev); } @@ -6242,8 +6272,9 @@ void mlx5_ib_stage_counters_cleanup(struct mlx5_ib_dev *dev) static int mlx5_ib_stage_cong_debugfs_init(struct mlx5_ib_dev *dev) { - return mlx5_ib_init_cong_debugfs(dev, - mlx5_core_native_port_num(dev->mdev) - 1); + mlx5_ib_init_cong_debugfs(dev, + mlx5_core_native_port_num(dev->mdev) - 1); + return 0; } static void mlx5_ib_stage_cong_debugfs_cleanup(struct mlx5_ib_dev *dev) @@ -6293,7 +6324,7 @@ int mlx5_ib_stage_ib_reg_init(struct mlx5_ib_dev *dev) name = "mlx5_%d"; else name = "mlx5_bond_%d"; - return ib_register_device(&dev->ib_dev, name, NULL); + return ib_register_device(&dev->ib_dev, name); } void mlx5_ib_stage_pre_ib_reg_umr_cleanup(struct mlx5_ib_dev *dev) @@ -6550,7 +6581,7 @@ static void *mlx5_ib_add(struct mlx5_core_dev *mdev) if (mlx5_core_is_mp_slave(mdev) && ll == IB_LINK_LAYER_ETHERNET) return mlx5_ib_add_slave_port(mdev); - dev = (struct mlx5_ib_dev *)ib_alloc_device(sizeof(*dev)); + dev = ib_alloc_device(mlx5_ib_dev, ib_dev); if (!dev) return NULL; diff --git a/drivers/infiniband/hw/mlx5/mem.c b/drivers/infiniband/hw/mlx5/mem.c index 549234988bb4..9f90be296ee0 100644 --- a/drivers/infiniband/hw/mlx5/mem.c +++ b/drivers/infiniband/hw/mlx5/mem.c @@ -111,7 +111,6 @@ void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, *count = i; } -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING static u64 umem_dma_to_mtt(dma_addr_t umem_dma) { u64 mtt_entry = umem_dma & ODP_DMA_ADDR_MASK; @@ -123,7 +122,6 @@ static u64 umem_dma_to_mtt(dma_addr_t umem_dma) return mtt_entry; } -#endif /* * Populate the given array with bus addresses from the umem. @@ -151,7 +149,7 @@ void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, int len; struct scatterlist *sg; int entry; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + if (umem->is_odp) { WARN_ON(shift != 0); WARN_ON(access_flags != (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)); @@ -164,7 +162,6 @@ void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, } return; } -#endif i = 0; for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index eedba0d2ec4b..4a617d78eae1 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -36,6 +36,7 @@ #include <linux/kernel.h> #include <linux/sched.h> #include <rdma/ib_verbs.h> +#include <rdma/ib_umem.h> #include <rdma/ib_smi.h> #include <linux/mlx5/driver.h> #include <linux/mlx5/cq.h> @@ -588,14 +589,27 @@ struct mlx5_ib_mr { atomic_t num_leaf_free; wait_queue_head_t q_leaf_free; struct mlx5_async_work cb_work; + atomic_t num_pending_prefetch; }; +static inline bool is_odp_mr(struct mlx5_ib_mr *mr) +{ + return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem && + mr->umem->is_odp; +} + struct mlx5_ib_mw { struct ib_mw ibmw; struct mlx5_core_mkey mmkey; int ndescs; }; +struct mlx5_ib_devx_mr { + struct mlx5_core_mkey mmkey; + int ndescs; + struct rcu_head rcu; +}; + struct mlx5_ib_umr_context { struct ib_cqe cqe; enum ib_wc_status status; @@ -624,7 +638,6 @@ struct mlx5_cache_ent { spinlock_t lock; - struct dentry *dir; char name[4]; u32 order; u32 xlt; @@ -636,11 +649,6 @@ struct mlx5_cache_ent { u32 miss; u32 limit; - struct dentry *fsize; - struct dentry *fcur; - struct dentry *fmiss; - struct dentry *flimit; - struct mlx5_ib_dev *dev; struct work_struct work; struct delayed_work dwork; @@ -912,7 +920,6 @@ struct mlx5_ib_dev { /* Prevents soft lock on massive reg MRs */ struct mutex slow_path_mutex; int fill_delay; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING struct ib_odp_caps odp_caps; u64 odp_max_size; struct mlx5_ib_pf_eq odp_pf_eq; @@ -923,8 +930,6 @@ struct mlx5_ib_dev { */ struct srcu_struct mr_srcu; u32 null_mkey; - struct workqueue_struct *advise_mr_wq; -#endif struct mlx5_ib_flow_db *flow_db; /* protect resources needed as part of reset flow */ spinlock_t reset_flow_resource_lock; @@ -1034,7 +1039,8 @@ to_mflow_act(struct ib_flow_action *ibact) return container_of(ibact, struct mlx5_ib_flow_action, ib_action); } -int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt, +int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, + struct ib_udata *udata, unsigned long virt, struct mlx5_db *db); void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db); void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq); @@ -1069,9 +1075,12 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, const struct ib_send_wr **bad_wr); int mlx5_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr, const struct ib_recv_wr **bad_wr); -int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index, - void *buffer, u32 length, - struct mlx5_ib_qp_base *base); +int mlx5_ib_read_user_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer, + int buflen, size_t *bc); +int mlx5_ib_read_user_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer, + int buflen, size_t *bc); +int mlx5_ib_read_user_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, + void *buffer, int buflen, size_t *bc); struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev, const struct ib_cq_init_attr *attr, struct ib_ucontext *context, @@ -1097,6 +1106,7 @@ int mlx5_ib_dealloc_mw(struct ib_mw *mw); int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages, int page_shift, int flags); struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, + struct ib_udata *udata, int access_flags); void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr); int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, @@ -1214,6 +1224,9 @@ mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, { return -EOPNOTSUPP; } +static inline void mlx5_ib_invalidate_range(struct ib_umem_odp *umem_odp, + unsigned long start, + unsigned long end){}; #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ /* Needed for rep profile */ @@ -1253,7 +1266,7 @@ __be16 mlx5_get_roce_udp_sport(struct mlx5_ib_dev *dev, const struct ib_gid_attr *attr); void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num); -int mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num); +void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num); /* GSI QP helper functions */ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd, diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index bf2b6ea23851..c85f00255884 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -71,10 +71,9 @@ static int destroy_mkey(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr) { int err = mlx5_core_destroy_mkey(dev->mdev, &mr->mmkey); -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - /* Wait until all page fault handlers using the mr complete. */ - synchronize_srcu(&dev->mr_srcu); -#endif + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) + /* Wait until all page fault handlers using the mr complete. */ + synchronize_srcu(&dev->mr_srcu); return err; } @@ -95,10 +94,9 @@ static bool use_umr_mtt_update(struct mlx5_ib_mr *mr, u64 start, u64 length) length + (start & (MLX5_ADAPTER_PAGE_SIZE - 1)); } -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING static void update_odp_mr(struct mlx5_ib_mr *mr) { - if (mr->umem->is_odp) { + if (is_odp_mr(mr)) { /* * This barrier prevents the compiler from moving the * setting of umem->odp_data->private to point to our @@ -121,7 +119,6 @@ static void update_odp_mr(struct mlx5_ib_mr *mr) smp_wmb(); } } -#endif static void reg_mr_callback(int status, struct mlx5_async_work *context) { @@ -257,9 +254,8 @@ static void remove_keys(struct mlx5_ib_dev *dev, int c, int num) mlx5_core_destroy_mkey(dev->mdev, &mr->mmkey); } -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - synchronize_srcu(&dev->mr_srcu); -#endif + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) + synchronize_srcu(&dev->mr_srcu); list_for_each_entry_safe(mr, tmp_mr, &del_list, list) { list_del(&mr->list); @@ -611,52 +607,27 @@ static void mlx5_mr_cache_debugfs_cleanup(struct mlx5_ib_dev *dev) dev->cache.root = NULL; } -static int mlx5_mr_cache_debugfs_init(struct mlx5_ib_dev *dev) +static void mlx5_mr_cache_debugfs_init(struct mlx5_ib_dev *dev) { struct mlx5_mr_cache *cache = &dev->cache; struct mlx5_cache_ent *ent; + struct dentry *dir; int i; if (!mlx5_debugfs_root || dev->rep) - return 0; + return; cache->root = debugfs_create_dir("mr_cache", dev->mdev->priv.dbg_root); - if (!cache->root) - return -ENOMEM; for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++) { ent = &cache->ent[i]; sprintf(ent->name, "%d", ent->order); - ent->dir = debugfs_create_dir(ent->name, cache->root); - if (!ent->dir) - goto err; - - ent->fsize = debugfs_create_file("size", 0600, ent->dir, ent, - &size_fops); - if (!ent->fsize) - goto err; - - ent->flimit = debugfs_create_file("limit", 0600, ent->dir, ent, - &limit_fops); - if (!ent->flimit) - goto err; - - ent->fcur = debugfs_create_u32("cur", 0400, ent->dir, - &ent->cur); - if (!ent->fcur) - goto err; - - ent->fmiss = debugfs_create_u32("miss", 0600, ent->dir, - &ent->miss); - if (!ent->fmiss) - goto err; + dir = debugfs_create_dir(ent->name, cache->root); + debugfs_create_file("size", 0600, dir, ent, &size_fops); + debugfs_create_file("limit", 0600, dir, ent, &limit_fops); + debugfs_create_u32("cur", 0400, dir, &ent->cur); + debugfs_create_u32("miss", 0600, dir, &ent->miss); } - - return 0; -err: - mlx5_mr_cache_debugfs_cleanup(dev); - - return -ENOMEM; } static void delay_time_func(struct timer_list *t) @@ -670,7 +641,6 @@ int mlx5_mr_cache_init(struct mlx5_ib_dev *dev) { struct mlx5_mr_cache *cache = &dev->cache; struct mlx5_cache_ent *ent; - int err; int i; mutex_init(&dev->slow_path_mutex); @@ -715,14 +685,7 @@ int mlx5_mr_cache_init(struct mlx5_ib_dev *dev) queue_work(cache->wq, &ent->work); } - err = mlx5_mr_cache_debugfs_init(dev); - if (err) - mlx5_ib_warn(dev, "cache debugfs failure\n"); - - /* - * We don't want to fail driver if debugfs failed to initialize, - * so we are not forwarding error to the user. - */ + mlx5_mr_cache_debugfs_init(dev); return 0; } @@ -822,18 +785,17 @@ static int mr_cache_max_order(struct mlx5_ib_dev *dev) return MLX5_MAX_UMR_SHIFT; } -static int mr_umem_get(struct ib_pd *pd, u64 start, u64 length, - int access_flags, struct ib_umem **umem, - int *npages, int *page_shift, int *ncont, - int *order) +static int mr_umem_get(struct mlx5_ib_dev *dev, struct ib_udata *udata, + u64 start, u64 length, int access_flags, + struct ib_umem **umem, int *npages, int *page_shift, + int *ncont, int *order) { - struct mlx5_ib_dev *dev = to_mdev(pd->device); struct ib_umem *u; int err; *umem = NULL; - u = ib_umem_get(pd->uobject->context, start, length, access_flags, 0); + u = ib_umem_get(udata, start, length, access_flags, 0); err = PTR_ERR_OR_ZERO(u); if (err) { mlx5_ib_dbg(dev, "umem get failed (%d)\n", err); @@ -1306,21 +1268,20 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mlx5_ib_dbg(dev, "start 0x%llx, virt_addr 0x%llx, length 0x%llx, access_flags 0x%x\n", start, virt_addr, length, access_flags); -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - if (!start && length == U64_MAX) { + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && !start && + length == U64_MAX) { if (!(access_flags & IB_ACCESS_ON_DEMAND) || !(dev->odp_caps.general_caps & IB_ODP_SUPPORT_IMPLICIT)) return ERR_PTR(-EINVAL); - mr = mlx5_ib_alloc_implicit_mr(to_mpd(pd), access_flags); + mr = mlx5_ib_alloc_implicit_mr(to_mpd(pd), udata, access_flags); if (IS_ERR(mr)) return ERR_CAST(mr); return &mr->ibmr; } -#endif - err = mr_umem_get(pd, start, length, access_flags, &umem, &npages, - &page_shift, &ncont, &order); + err = mr_umem_get(dev, udata, start, length, access_flags, &umem, + &npages, &page_shift, &ncont, &order); if (err < 0) return ERR_PTR(err); @@ -1361,9 +1322,7 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mr->umem = umem; set_mr_fields(dev, mr, npages, length, access_flags); -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING update_odp_mr(mr); -#endif if (!populate_mtts) { int update_xlt_flags = MLX5_IB_UPD_XLT_ENABLE; @@ -1380,9 +1339,11 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, } } -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - mr->live = 1; -#endif + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) { + mr->live = 1; + atomic_set(&mr->num_pending_prefetch, 0); + } + return &mr->ibmr; error: ib_umem_release(umem); @@ -1470,8 +1431,9 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, flags |= IB_MR_REREG_TRANS; ib_umem_release(mr->umem); mr->umem = NULL; - err = mr_umem_get(pd, addr, len, access_flags, &mr->umem, - &npages, &page_shift, &ncont, &order); + err = mr_umem_get(dev, udata, addr, len, access_flags, + &mr->umem, &npages, &page_shift, &ncont, + &order); if (err) goto err; } @@ -1497,9 +1459,8 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, } mr->allocated_from_cache = 0; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - mr->live = 1; -#endif + if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) + mr->live = 1; } else { /* * Send a UMR WQE @@ -1528,9 +1489,7 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, set_mr_fields(dev, mr, npages, len, access_flags); -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING update_odp_mr(mr); -#endif return 0; err: @@ -1616,12 +1575,19 @@ static void dereg_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr) int npages = mr->npages; struct ib_umem *umem = mr->umem; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - if (umem && umem->is_odp) { + if (is_odp_mr(mr)) { struct ib_umem_odp *umem_odp = to_ib_umem_odp(umem); - /* Prevent new page faults from succeeding */ + /* Prevent new page faults and + * prefetch requests from succeeding + */ mr->live = 0; + + /* dequeue pending prefetch requests for the mr */ + if (atomic_read(&mr->num_pending_prefetch)) + flush_workqueue(system_unbound_wq); + WARN_ON(atomic_read(&mr->num_pending_prefetch)); + /* Wait for all running page-fault handlers to finish. */ synchronize_srcu(&dev->mr_srcu); /* Destroy all page mappings */ @@ -1641,7 +1607,7 @@ static void dereg_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr) /* Avoid double-freeing the umem. */ umem = NULL; } -#endif + clean_mr(dev, mr); /* diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index 4ee32964e1dd..c20bfc41ecf1 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -101,9 +101,9 @@ static int check_parent(struct ib_umem_odp *odp, return mr && mr->parent == parent && !odp->dying; } -struct ib_ucontext_per_mm *mr_to_per_mm(struct mlx5_ib_mr *mr) +static struct ib_ucontext_per_mm *mr_to_per_mm(struct mlx5_ib_mr *mr) { - if (WARN_ON(!mr || !mr->umem || !mr->umem->is_odp)) + if (WARN_ON(!mr || !is_odp_mr(mr))) return NULL; return to_ib_umem_odp(mr->umem)->per_mm; @@ -315,6 +315,9 @@ void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev) if (MLX5_CAP_ODP(dev->mdev, ud_odp_caps.send)) caps->per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SEND; + if (MLX5_CAP_ODP(dev->mdev, ud_odp_caps.srq_receive)) + caps->per_transport_caps.ud_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV; + if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.send)) caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SEND; @@ -330,6 +333,27 @@ void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev) if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.atomic)) caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_ATOMIC; + if (MLX5_CAP_ODP(dev->mdev, rc_odp_caps.srq_receive)) + caps->per_transport_caps.rc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV; + + if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.send)) + caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_SEND; + + if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.receive)) + caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_RECV; + + if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.write)) + caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_WRITE; + + if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.read)) + caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_READ; + + if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.atomic)) + caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_ATOMIC; + + if (MLX5_CAP_ODP(dev->mdev, xrc_odp_caps.srq_receive)) + caps->per_transport_caps.xrc_odp_caps |= IB_ODP_SUPPORT_SRQ_RECV; + if (MLX5_CAP_GEN(dev->mdev, fixed_buffer_size) && MLX5_CAP_GEN(dev->mdev, null_mkey) && MLX5_CAP_GEN(dev->mdev, umr_extended_translation_offset)) @@ -439,7 +463,7 @@ next_mr: if (nentries) nentries++; } else { - odp = ib_alloc_odp_umem(odp_mr->per_mm, addr, + odp = ib_alloc_odp_umem(odp_mr, addr, MLX5_IMR_MTT_SIZE); if (IS_ERR(odp)) { mutex_unlock(&odp_mr->umem_mutex); @@ -492,13 +516,13 @@ next_mr: } struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, + struct ib_udata *udata, int access_flags) { - struct ib_ucontext *ctx = pd->ibpd.uobject->context; struct mlx5_ib_mr *imr; struct ib_umem *umem; - umem = ib_umem_get(ctx, 0, 0, IB_ACCESS_ON_DEMAND, 0); + umem = ib_umem_get(udata, 0, 0, access_flags, 0); if (IS_ERR(umem)) return ERR_CAST(umem); @@ -511,6 +535,7 @@ struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, imr->umem = umem; init_waitqueue_head(&imr->q_leaf_free); atomic_set(&imr->num_leaf_free, 0); + atomic_set(&imr->num_pending_prefetch, 0); return imr; } @@ -685,6 +710,21 @@ struct pf_frame { int depth; }; +static int get_indirect_num_descs(struct mlx5_core_mkey *mmkey) +{ + struct mlx5_ib_mw *mw; + struct mlx5_ib_devx_mr *devx_mr; + + if (mmkey->type == MLX5_MKEY_MW) { + mw = container_of(mmkey, struct mlx5_ib_mw, mmkey); + return mw->ndescs; + } + + devx_mr = container_of(mmkey, struct mlx5_ib_devx_mr, + mmkey); + return devx_mr->ndescs; +} + /* * Handle a single data segment in a page-fault WQE or RDMA region. * @@ -696,7 +736,8 @@ struct pf_frame { * -EFAULT when there's an error mapping the requested pages. The caller will * abort the page fault handling. */ -static int pagefault_single_data_segment(struct mlx5_ib_dev *dev, u32 key, +static int pagefault_single_data_segment(struct mlx5_ib_dev *dev, + struct ib_pd *pd, u32 key, u64 io_virt, size_t bcnt, u32 *bytes_committed, u32 *bytes_mapped, u32 flags) @@ -705,11 +746,11 @@ static int pagefault_single_data_segment(struct mlx5_ib_dev *dev, u32 key, bool prefetch = flags & MLX5_PF_FLAGS_PREFETCH; struct pf_frame *head = NULL, *frame; struct mlx5_core_mkey *mmkey; - struct mlx5_ib_mw *mw; struct mlx5_ib_mr *mr; struct mlx5_klm *pklm; u32 *out = NULL; size_t offset; + int ndescs; srcu_key = srcu_read_lock(&dev->mr_srcu); @@ -739,12 +780,18 @@ next_mr: goto srcu_unlock; } - if (prefetch && !mr->umem->is_odp) { - ret = -EINVAL; - goto srcu_unlock; + if (prefetch) { + if (!is_odp_mr(mr) || + mr->ibmr.pd != pd) { + mlx5_ib_dbg(dev, "Invalid prefetch request: %s\n", + is_odp_mr(mr) ? "MR is not ODP" : + "PD is not of the MR"); + ret = -EINVAL; + goto srcu_unlock; + } } - if (!mr->umem->is_odp) { + if (!is_odp_mr(mr)) { mlx5_ib_dbg(dev, "skipping non ODP MR (lkey=0x%06x) in page fault handler.\n", key); if (bytes_mapped) @@ -762,7 +809,8 @@ next_mr: break; case MLX5_MKEY_MW: - mw = container_of(mmkey, struct mlx5_ib_mw, mmkey); + case MLX5_MKEY_INDIRECT_DEVX: + ndescs = get_indirect_num_descs(mmkey); if (depth >= MLX5_CAP_GEN(dev->mdev, max_indirection)) { mlx5_ib_dbg(dev, "indirection level exceeded\n"); @@ -771,7 +819,7 @@ next_mr: } outlen = MLX5_ST_SZ_BYTES(query_mkey_out) + - sizeof(*pklm) * (mw->ndescs - 2); + sizeof(*pklm) * (ndescs - 2); if (outlen > cur_outlen) { kfree(out); @@ -786,14 +834,14 @@ next_mr: pklm = (struct mlx5_klm *)MLX5_ADDR_OF(query_mkey_out, out, bsf0_klm0_pas_mtt0_1); - ret = mlx5_core_query_mkey(dev->mdev, &mw->mmkey, out, outlen); + ret = mlx5_core_query_mkey(dev->mdev, mmkey, out, outlen); if (ret) goto srcu_unlock; offset = io_virt - MLX5_GET64(query_mkey_out, out, memory_key_mkey_entry.start_addr); - for (i = 0; bcnt && i < mw->ndescs; i++, pklm++) { + for (i = 0; bcnt && i < ndescs; i++, pklm++) { if (offset >= be32_to_cpu(pklm->bcount)) { offset -= be32_to_cpu(pklm->bcount); continue; @@ -853,7 +901,6 @@ srcu_unlock: /** * Parse a series of data segments for page fault handling. * - * @qp the QP on which the fault occurred. * @pfault contains page fault information. * @wqe points at the first data segment in the WQE. * @wqe_end points after the end of the WQE. @@ -870,7 +917,7 @@ srcu_unlock: */ static int pagefault_data_segments(struct mlx5_ib_dev *dev, struct mlx5_pagefault *pfault, - struct mlx5_ib_qp *qp, void *wqe, + void *wqe, void *wqe_end, u32 *bytes_mapped, u32 *total_wqe_bytes, int receive_queue) { @@ -881,10 +928,6 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, size_t bcnt; int inline_segment; - /* Skip SRQ next-WQE segment. */ - if (receive_queue && qp->ibqp.srq) - wqe += sizeof(struct mlx5_wqe_srq_next_seg); - if (bytes_mapped) *bytes_mapped = 0; if (total_wqe_bytes) @@ -928,7 +971,8 @@ static int pagefault_data_segments(struct mlx5_ib_dev *dev, continue; } - ret = pagefault_single_data_segment(dev, key, io_virt, bcnt, + ret = pagefault_single_data_segment(dev, NULL, key, + io_virt, bcnt, &pfault->bytes_committed, bytes_mapped, 0); if (ret < 0) @@ -1009,6 +1053,10 @@ static int mlx5_ib_mr_initiator_pfault_handler( MLX5_WQE_CTRL_OPCODE_MASK; switch (qp->ibqp.qp_type) { + case IB_QPT_XRC_INI: + *wqe += sizeof(struct mlx5_wqe_xrc_seg); + transport_caps = dev->odp_caps.per_transport_caps.xrc_odp_caps; + break; case IB_QPT_RC: transport_caps = dev->odp_caps.per_transport_caps.rc_odp_caps; break; @@ -1028,7 +1076,7 @@ static int mlx5_ib_mr_initiator_pfault_handler( return -EFAULT; } - if (qp->ibqp.qp_type != IB_QPT_RC) { + if (qp->ibqp.qp_type == IB_QPT_UD) { av = *wqe; if (av->dqp_dct & cpu_to_be32(MLX5_EXTENDED_UD_AV)) *wqe += sizeof(struct mlx5_av); @@ -1053,21 +1101,34 @@ static int mlx5_ib_mr_initiator_pfault_handler( } /* - * Parse responder WQE. Advances the wqe pointer to point at the - * scatter-gather list, and set wqe_end to the end of the WQE. + * Parse responder WQE and set wqe_end to the end of the WQE. */ -static int mlx5_ib_mr_responder_pfault_handler( - struct mlx5_ib_dev *dev, struct mlx5_pagefault *pfault, - struct mlx5_ib_qp *qp, void **wqe, void **wqe_end, int wqe_length) +static int mlx5_ib_mr_responder_pfault_handler_srq(struct mlx5_ib_dev *dev, + struct mlx5_ib_srq *srq, + void **wqe, void **wqe_end, + int wqe_length) { - struct mlx5_ib_wq *wq = &qp->rq; - int wqe_size = 1 << wq->wqe_shift; + int wqe_size = 1 << srq->msrq.wqe_shift; - if (qp->ibqp.srq) { - mlx5_ib_err(dev, "ODP fault on SRQ is not supported\n"); + if (wqe_size > wqe_length) { + mlx5_ib_err(dev, "Couldn't read all of the receive WQE's content\n"); return -EFAULT; } + *wqe_end = *wqe + wqe_size; + *wqe += sizeof(struct mlx5_wqe_srq_next_seg); + + return 0; +} + +static int mlx5_ib_mr_responder_pfault_handler_rq(struct mlx5_ib_dev *dev, + struct mlx5_ib_qp *qp, + void *wqe, void **wqe_end, + int wqe_length) +{ + struct mlx5_ib_wq *wq = &qp->rq; + int wqe_size = 1 << wq->wqe_shift; + if (qp->wq_sig) { mlx5_ib_err(dev, "ODP fault with WQE signatures is not supported\n"); return -EFAULT; @@ -1091,7 +1152,7 @@ invalid_transport_or_opcode: return -EFAULT; } - *wqe_end = *wqe + wqe_size; + *wqe_end = wqe + wqe_size; return 0; } @@ -1099,22 +1160,25 @@ invalid_transport_or_opcode: static inline struct mlx5_core_rsc_common *odp_get_rsc(struct mlx5_ib_dev *dev, u32 wq_num, int pf_type) { - enum mlx5_res_type res_type; + struct mlx5_core_rsc_common *common = NULL; + struct mlx5_core_srq *srq; switch (pf_type) { case MLX5_WQE_PF_TYPE_RMP: - res_type = MLX5_RES_SRQ; + srq = mlx5_cmd_get_srq(dev, wq_num); + if (srq) + common = &srq->common; break; case MLX5_WQE_PF_TYPE_REQ_SEND_OR_WRITE: case MLX5_WQE_PF_TYPE_RESP: case MLX5_WQE_PF_TYPE_REQ_READ_OR_ATOMIC: - res_type = MLX5_RES_QP; + common = mlx5_core_res_hold(dev->mdev, wq_num, MLX5_RES_QP); break; default: - return NULL; + break; } - return mlx5_core_res_hold(dev->mdev, wq_num, res_type); + return common; } static inline struct mlx5_ib_qp *res_to_qp(struct mlx5_core_rsc_common *res) @@ -1124,6 +1188,14 @@ static inline struct mlx5_ib_qp *res_to_qp(struct mlx5_core_rsc_common *res) return to_mibqp(mqp); } +static inline struct mlx5_ib_srq *res_to_srq(struct mlx5_core_rsc_common *res) +{ + struct mlx5_core_srq *msrq = + container_of(res, struct mlx5_core_srq, common); + + return to_mibsrq(msrq); +} + static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_dev *dev, struct mlx5_pagefault *pfault) { @@ -1134,8 +1206,10 @@ static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_dev *dev, int resume_with_error = 1; u16 wqe_index = pfault->wqe.wqe_index; int requestor = pfault->type & MLX5_PFAULT_REQUESTOR; - struct mlx5_core_rsc_common *res; - struct mlx5_ib_qp *qp; + struct mlx5_core_rsc_common *res = NULL; + struct mlx5_ib_qp *qp = NULL; + struct mlx5_ib_srq *srq = NULL; + size_t bytes_copied; res = odp_get_rsc(dev, pfault->wqe.wq_num, pfault->type); if (!res) { @@ -1147,6 +1221,10 @@ static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_dev *dev, case MLX5_RES_QP: qp = res_to_qp(res); break; + case MLX5_RES_SRQ: + case MLX5_RES_XSRQ: + srq = res_to_srq(res); + break; default: mlx5_ib_err(dev, "wqe page fault for unsupported type %d\n", pfault->type); goto resolve_page_fault; @@ -1158,9 +1236,23 @@ static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_dev *dev, goto resolve_page_fault; } - ret = mlx5_ib_read_user_wqe(qp, requestor, wqe_index, buffer, - PAGE_SIZE, &qp->trans_qp.base); - if (ret < 0) { + if (qp) { + if (requestor) { + ret = mlx5_ib_read_user_wqe_sq(qp, wqe_index, + buffer, PAGE_SIZE, + &bytes_copied); + } else { + ret = mlx5_ib_read_user_wqe_rq(qp, wqe_index, + buffer, PAGE_SIZE, + &bytes_copied); + } + } else { + ret = mlx5_ib_read_user_wqe_srq(srq, wqe_index, + buffer, PAGE_SIZE, + &bytes_copied); + } + + if (ret) { mlx5_ib_err(dev, "Failed reading a WQE following page fault, error=%d, wqe_index=%x, qpn=%x\n", ret, wqe_index, pfault->token); goto resolve_page_fault; @@ -1168,11 +1260,18 @@ static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_dev *dev, wqe = buffer; if (requestor) - ret = mlx5_ib_mr_initiator_pfault_handler(dev, pfault, qp, &wqe, - &wqe_end, ret); + ret = mlx5_ib_mr_initiator_pfault_handler(dev, pfault, qp, + &wqe, &wqe_end, + bytes_copied); + else if (qp) + ret = mlx5_ib_mr_responder_pfault_handler_rq(dev, qp, + wqe, &wqe_end, + bytes_copied); else - ret = mlx5_ib_mr_responder_pfault_handler(dev, pfault, qp, &wqe, - &wqe_end, ret); + ret = mlx5_ib_mr_responder_pfault_handler_srq(dev, srq, + &wqe, &wqe_end, + bytes_copied); + if (ret < 0) goto resolve_page_fault; @@ -1181,7 +1280,7 @@ static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_dev *dev, goto resolve_page_fault; } - ret = pagefault_data_segments(dev, pfault, qp, wqe, wqe_end, + ret = pagefault_data_segments(dev, pfault, wqe, wqe_end, &bytes_mapped, &total_wqe_bytes, !requestor); if (ret == -EAGAIN) { @@ -1240,7 +1339,7 @@ static void mlx5_ib_mr_rdma_pfault_handler(struct mlx5_ib_dev *dev, prefetch_len = min(MAX_PREFETCH_LEN, prefetch_len); } - ret = pagefault_single_data_segment(dev, rkey, address, length, + ret = pagefault_single_data_segment(dev, NULL, rkey, address, length, &pfault->bytes_committed, NULL, 0); if (ret == -EAGAIN) { @@ -1267,7 +1366,7 @@ static void mlx5_ib_mr_rdma_pfault_handler(struct mlx5_ib_dev *dev, if (prefetch_activated) { u32 bytes_committed = 0; - ret = pagefault_single_data_segment(dev, rkey, address, + ret = pagefault_single_data_segment(dev, NULL, rkey, address, prefetch_len, &bytes_committed, NULL, 0); @@ -1564,30 +1663,98 @@ int mlx5_ib_odp_init(void) struct prefetch_mr_work { struct work_struct work; - struct mlx5_ib_dev *dev; + struct ib_pd *pd; u32 pf_flags; u32 num_sge; struct ib_sge sg_list[0]; }; -static int mlx5_ib_prefetch_sg_list(struct mlx5_ib_dev *dev, u32 pf_flags, +static void num_pending_prefetch_dec(struct mlx5_ib_dev *dev, + struct ib_sge *sg_list, u32 num_sge, + u32 from) +{ + u32 i; + int srcu_key; + + srcu_key = srcu_read_lock(&dev->mr_srcu); + + for (i = from; i < num_sge; ++i) { + struct mlx5_core_mkey *mmkey; + struct mlx5_ib_mr *mr; + + mmkey = __mlx5_mr_lookup(dev->mdev, + mlx5_base_mkey(sg_list[i].lkey)); + mr = container_of(mmkey, struct mlx5_ib_mr, mmkey); + atomic_dec(&mr->num_pending_prefetch); + } + + srcu_read_unlock(&dev->mr_srcu, srcu_key); +} + +static bool num_pending_prefetch_inc(struct ib_pd *pd, + struct ib_sge *sg_list, u32 num_sge) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + bool ret = true; + u32 i; + + for (i = 0; i < num_sge; ++i) { + struct mlx5_core_mkey *mmkey; + struct mlx5_ib_mr *mr; + + mmkey = __mlx5_mr_lookup(dev->mdev, + mlx5_base_mkey(sg_list[i].lkey)); + if (!mmkey || mmkey->key != sg_list[i].lkey) { + ret = false; + break; + } + + if (mmkey->type != MLX5_MKEY_MR) { + ret = false; + break; + } + + mr = container_of(mmkey, struct mlx5_ib_mr, mmkey); + + if (mr->ibmr.pd != pd) { + ret = false; + break; + } + + if (!mr->live) { + ret = false; + break; + } + + atomic_inc(&mr->num_pending_prefetch); + } + + if (!ret) + num_pending_prefetch_dec(dev, sg_list, i, 0); + + return ret; +} + +static int mlx5_ib_prefetch_sg_list(struct ib_pd *pd, u32 pf_flags, struct ib_sge *sg_list, u32 num_sge) { - int i; + u32 i; + int ret = 0; + struct mlx5_ib_dev *dev = to_mdev(pd->device); for (i = 0; i < num_sge; ++i) { struct ib_sge *sg = &sg_list[i]; int bytes_committed = 0; - int ret; - ret = pagefault_single_data_segment(dev, sg->lkey, sg->addr, + ret = pagefault_single_data_segment(dev, pd, sg->lkey, sg->addr, sg->length, &bytes_committed, NULL, pf_flags); if (ret < 0) - return ret; + break; } - return 0; + + return ret < 0 ? ret : 0; } static void mlx5_ib_prefetch_mr_work(struct work_struct *work) @@ -1595,12 +1762,14 @@ static void mlx5_ib_prefetch_mr_work(struct work_struct *work) struct prefetch_mr_work *w = container_of(work, struct prefetch_mr_work, work); - if (ib_device_try_get(&w->dev->ib_dev)) { - mlx5_ib_prefetch_sg_list(w->dev, w->pf_flags, w->sg_list, + if (ib_device_try_get(w->pd->device)) { + mlx5_ib_prefetch_sg_list(w->pd, w->pf_flags, w->sg_list, w->num_sge); - ib_device_put(&w->dev->ib_dev); + ib_device_put(w->pd->device); } - put_device(&w->dev->ib_dev.dev); + + num_pending_prefetch_dec(to_mdev(w->pd->device), w->sg_list, + w->num_sge, 0); kfree(w); } @@ -1611,12 +1780,14 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, struct mlx5_ib_dev *dev = to_mdev(pd->device); u32 pf_flags = MLX5_PF_FLAGS_PREFETCH; struct prefetch_mr_work *work; + bool valid_req; + int srcu_key; if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH) pf_flags |= MLX5_PF_FLAGS_DOWNGRADE; if (flags & IB_UVERBS_ADVISE_MR_FLAG_FLUSH) - return mlx5_ib_prefetch_sg_list(dev, pf_flags, sg_list, + return mlx5_ib_prefetch_sg_list(pd, pf_flags, sg_list, num_sge); work = kvzalloc(struct_size(work, sg_list, num_sge), GFP_KERNEL); @@ -1625,12 +1796,25 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, memcpy(work->sg_list, sg_list, num_sge * sizeof(struct ib_sge)); - get_device(&dev->ib_dev.dev); - work->dev = dev; + /* It is guaranteed that the pd when work is executed is the pd when + * work was queued since pd can't be destroyed while it holds MRs and + * destroying a MR leads to flushing the workquque + */ + work->pd = pd; work->pf_flags = pf_flags; work->num_sge = num_sge; INIT_WORK(&work->work, mlx5_ib_prefetch_mr_work); - schedule_work(&work->work); - return 0; + + srcu_key = srcu_read_lock(&dev->mr_srcu); + + valid_req = num_pending_prefetch_inc(pd, sg_list, num_sge); + if (valid_req) + queue_work(system_unbound_wq, &work->work); + else + kfree(work); + + srcu_read_unlock(&dev->mr_srcu, srcu_key); + + return valid_req ? 0 : -EINVAL; } diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index 7db778d96ef5..6b1f0e76900b 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -109,75 +109,173 @@ static int is_sqp(enum ib_qp_type qp_type) } /** - * mlx5_ib_read_user_wqe() - Copy a user-space WQE to kernel space. + * mlx5_ib_read_user_wqe_common() - Copy a WQE (or part of) from user WQ + * to kernel buffer * - * @qp: QP to copy from. - * @send: copy from the send queue when non-zero, use the receive queue - * otherwise. - * @wqe_index: index to start copying from. For send work queues, the - * wqe_index is in units of MLX5_SEND_WQE_BB. - * For receive work queue, it is the number of work queue - * element in the queue. - * @buffer: destination buffer. - * @length: maximum number of bytes to copy. + * @umem: User space memory where the WQ is + * @buffer: buffer to copy to + * @buflen: buffer length + * @wqe_index: index of WQE to copy from + * @wq_offset: offset to start of WQ + * @wq_wqe_cnt: number of WQEs in WQ + * @wq_wqe_shift: log2 of WQE size + * @bcnt: number of bytes to copy + * @bytes_copied: number of bytes to copy (return value) * - * Copies at least a single WQE, but may copy more data. + * Copies from start of WQE bcnt or less bytes. + * Does not gurantee to copy the entire WQE. * - * Return: the number of bytes copied, or an error code. + * Return: zero on success, or an error code. */ -int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index, - void *buffer, u32 length, - struct mlx5_ib_qp_base *base) +static int mlx5_ib_read_user_wqe_common(struct ib_umem *umem, + void *buffer, + u32 buflen, + int wqe_index, + int wq_offset, + int wq_wqe_cnt, + int wq_wqe_shift, + int bcnt, + size_t *bytes_copied) +{ + size_t offset = wq_offset + ((wqe_index % wq_wqe_cnt) << wq_wqe_shift); + size_t wq_end = wq_offset + (wq_wqe_cnt << wq_wqe_shift); + size_t copy_length; + int ret; + + /* don't copy more than requested, more than buffer length or + * beyond WQ end + */ + copy_length = min_t(u32, buflen, wq_end - offset); + copy_length = min_t(u32, copy_length, bcnt); + + ret = ib_umem_copy_from(buffer, umem, offset, copy_length); + if (ret) + return ret; + + if (!ret && bytes_copied) + *bytes_copied = copy_length; + + return 0; +} + +int mlx5_ib_read_user_wqe_sq(struct mlx5_ib_qp *qp, + int wqe_index, + void *buffer, + int buflen, + size_t *bc) { - struct ib_device *ibdev = qp->ibqp.device; - struct mlx5_ib_dev *dev = to_mdev(ibdev); - struct mlx5_ib_wq *wq = send ? &qp->sq : &qp->rq; - size_t offset; - size_t wq_end; + struct mlx5_ib_qp_base *base = &qp->trans_qp.base; struct ib_umem *umem = base->ubuffer.umem; - u32 first_copy_length; - int wqe_length; + struct mlx5_ib_wq *wq = &qp->sq; + struct mlx5_wqe_ctrl_seg *ctrl; + size_t bytes_copied; + size_t bytes_copied2; + size_t wqe_length; int ret; + int ds; - if (wq->wqe_cnt == 0) { - mlx5_ib_dbg(dev, "mlx5_ib_read_user_wqe for a QP with wqe_cnt == 0. qp_type: 0x%x\n", - qp->ibqp.qp_type); + if (buflen < sizeof(*ctrl)) return -EINVAL; - } - offset = wq->offset + ((wqe_index % wq->wqe_cnt) << wq->wqe_shift); - wq_end = wq->offset + (wq->wqe_cnt << wq->wqe_shift); + /* at first read as much as possible */ + ret = mlx5_ib_read_user_wqe_common(umem, + buffer, + buflen, + wqe_index, + wq->offset, + wq->wqe_cnt, + wq->wqe_shift, + buflen, + &bytes_copied); + if (ret) + return ret; - if (send && length < sizeof(struct mlx5_wqe_ctrl_seg)) + /* we need at least control segment size to proceed */ + if (bytes_copied < sizeof(*ctrl)) return -EINVAL; - if (offset > umem->length || - (send && offset + sizeof(struct mlx5_wqe_ctrl_seg) > umem->length)) - return -EINVAL; + ctrl = buffer; + ds = be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_DS_MASK; + wqe_length = ds * MLX5_WQE_DS_UNITS; + + /* if we copied enough then we are done */ + if (bytes_copied >= wqe_length) { + *bc = bytes_copied; + return 0; + } + + /* otherwise this a wrapped around wqe + * so read the remaining bytes starting + * from wqe_index 0 + */ + ret = mlx5_ib_read_user_wqe_common(umem, + buffer + bytes_copied, + buflen - bytes_copied, + 0, + wq->offset, + wq->wqe_cnt, + wq->wqe_shift, + wqe_length - bytes_copied, + &bytes_copied2); - first_copy_length = min_t(u32, offset + length, wq_end) - offset; - ret = ib_umem_copy_from(buffer, umem, offset, first_copy_length); if (ret) return ret; + *bc = bytes_copied + bytes_copied2; + return 0; +} - if (send) { - struct mlx5_wqe_ctrl_seg *ctrl = buffer; - int ds = be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_DS_MASK; - - wqe_length = ds * MLX5_WQE_DS_UNITS; - } else { - wqe_length = 1 << wq->wqe_shift; - } +int mlx5_ib_read_user_wqe_rq(struct mlx5_ib_qp *qp, + int wqe_index, + void *buffer, + int buflen, + size_t *bc) +{ + struct mlx5_ib_qp_base *base = &qp->trans_qp.base; + struct ib_umem *umem = base->ubuffer.umem; + struct mlx5_ib_wq *wq = &qp->rq; + size_t bytes_copied; + int ret; - if (wqe_length <= first_copy_length) - return first_copy_length; + ret = mlx5_ib_read_user_wqe_common(umem, + buffer, + buflen, + wqe_index, + wq->offset, + wq->wqe_cnt, + wq->wqe_shift, + buflen, + &bytes_copied); - ret = ib_umem_copy_from(buffer + first_copy_length, umem, wq->offset, - wqe_length - first_copy_length); if (ret) return ret; + *bc = bytes_copied; + return 0; +} - return wqe_length; +int mlx5_ib_read_user_wqe_srq(struct mlx5_ib_srq *srq, + int wqe_index, + void *buffer, + int buflen, + size_t *bc) +{ + struct ib_umem *umem = srq->umem; + size_t bytes_copied; + int ret; + + ret = mlx5_ib_read_user_wqe_common(umem, + buffer, + buflen, + wqe_index, + 0, + srq->msrq.max, + srq->msrq.wqe_shift, + buflen, + &bytes_copied); + + if (ret) + return ret; + *bc = bytes_copied; + return 0; } static void mlx5_ib_qp_event(struct mlx5_core_qp *qp, int type) @@ -435,9 +533,9 @@ static int set_user_buf_size(struct mlx5_ib_dev *dev, return -EINVAL; } - if (ucmd->sq_wqe_count && ((1 << ilog2(ucmd->sq_wqe_count)) != ucmd->sq_wqe_count)) { - mlx5_ib_warn(dev, "sq_wqe_count %d, sq_wqe_count %d\n", - ucmd->sq_wqe_count, ucmd->sq_wqe_count); + if (ucmd->sq_wqe_count && !is_power_of_2(ucmd->sq_wqe_count)) { + mlx5_ib_warn(dev, "sq_wqe_count %d is not a power of two\n", + ucmd->sq_wqe_count); return -EINVAL; } @@ -645,16 +743,14 @@ int bfregn_to_uar_index(struct mlx5_ib_dev *dev, return bfregi->sys_pages[index_of_sys_page] + offset; } -static int mlx5_ib_umem_get(struct mlx5_ib_dev *dev, - struct ib_pd *pd, +static int mlx5_ib_umem_get(struct mlx5_ib_dev *dev, struct ib_udata *udata, unsigned long addr, size_t size, - struct ib_umem **umem, - int *npages, int *page_shift, int *ncont, - u32 *offset) + struct ib_umem **umem, int *npages, int *page_shift, + int *ncont, u32 *offset) { int err; - *umem = ib_umem_get(pd->uobject->context, addr, size, 0, 0); + *umem = ib_umem_get(udata, addr, size, 0, 0); if (IS_ERR(*umem)) { mlx5_ib_dbg(dev, "umem_get failed\n"); return PTR_ERR(*umem); @@ -695,10 +791,11 @@ static void destroy_user_rq(struct mlx5_ib_dev *dev, struct ib_pd *pd, } static int create_user_rq(struct mlx5_ib_dev *dev, struct ib_pd *pd, - struct mlx5_ib_rwq *rwq, + struct ib_udata *udata, struct mlx5_ib_rwq *rwq, struct mlx5_ib_create_wq *ucmd) { - struct mlx5_ib_ucontext *context; + struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); int page_shift = 0; int npages; u32 offset = 0; @@ -708,9 +805,7 @@ static int create_user_rq(struct mlx5_ib_dev *dev, struct ib_pd *pd, if (!ucmd->buf_addr) return -EINVAL; - context = to_mucontext(pd->uobject->context); - rwq->umem = ib_umem_get(pd->uobject->context, ucmd->buf_addr, - rwq->buf_size, 0, 0); + rwq->umem = ib_umem_get(udata, ucmd->buf_addr, rwq->buf_size, 0, 0); if (IS_ERR(rwq->umem)) { mlx5_ib_dbg(dev, "umem_get failed\n"); err = PTR_ERR(rwq->umem); @@ -735,7 +830,7 @@ static int create_user_rq(struct mlx5_ib_dev *dev, struct ib_pd *pd, (unsigned long long)ucmd->buf_addr, rwq->buf_size, npages, page_shift, ncont, offset); - err = mlx5_ib_db_map_user(context, ucmd->db_addr, &rwq->db); + err = mlx5_ib_db_map_user(ucontext, udata, ucmd->db_addr, &rwq->db); if (err) { mlx5_ib_dbg(dev, "map failed\n"); goto err_umem; @@ -783,7 +878,8 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd, return err; } - context = to_mucontext(pd->uobject->context); + context = rdma_udata_to_drv_context(udata, struct mlx5_ib_ucontext, + ibucontext); if (ucmd.flags & MLX5_QP_FLAG_BFREG_INDEX) { uar_index = bfregn_to_uar_index(dev, &context->bfregi, ucmd.bfreg_index, true); @@ -819,10 +915,9 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd, if (ucmd.buf_addr && ubuffer->buf_size) { ubuffer->buf_addr = ucmd.buf_addr; - err = mlx5_ib_umem_get(dev, pd, ubuffer->buf_addr, - ubuffer->buf_size, - &ubuffer->umem, &npages, &page_shift, - &ncont, &offset); + err = mlx5_ib_umem_get(dev, udata, ubuffer->buf_addr, + ubuffer->buf_size, &ubuffer->umem, + &npages, &page_shift, &ncont, &offset); if (err) goto err_bfreg; } else { @@ -856,7 +951,7 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd, resp->bfreg_index = MLX5_IB_INVALID_BFREG; qp->bfregn = bfregn; - err = mlx5_ib_db_map_user(context, ucmd.db_addr, &qp->db); + err = mlx5_ib_db_map_user(context, udata, ucmd.db_addr, &qp->db); if (err) { mlx5_ib_dbg(dev, "map failed\n"); goto err_free; @@ -1119,6 +1214,7 @@ static void destroy_flow_rule_vport_sq(struct mlx5_ib_dev *dev, } static int create_raw_packet_qp_sq(struct mlx5_ib_dev *dev, + struct ib_udata *udata, struct mlx5_ib_sq *sq, void *qpin, struct ib_pd *pd) { @@ -1135,9 +1231,9 @@ static int create_raw_packet_qp_sq(struct mlx5_ib_dev *dev, int ncont = 0; u32 offset = 0; - err = mlx5_ib_umem_get(dev, pd, ubuffer->buf_addr, ubuffer->buf_size, - &sq->ubuffer.umem, &npages, &page_shift, - &ncont, &offset); + err = mlx5_ib_umem_get(dev, udata, ubuffer->buf_addr, ubuffer->buf_size, + &sq->ubuffer.umem, &npages, &page_shift, &ncont, + &offset); if (err) return err; @@ -1362,9 +1458,8 @@ static int create_raw_packet_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp, struct mlx5_ib_raw_packet_qp *raw_packet_qp = &qp->raw_packet_qp; struct mlx5_ib_sq *sq = &raw_packet_qp->sq; struct mlx5_ib_rq *rq = &raw_packet_qp->rq; - struct ib_uobject *uobj = pd->uobject; - struct ib_ucontext *ucontext = uobj->context; - struct mlx5_ib_ucontext *mucontext = to_mucontext(ucontext); + struct mlx5_ib_ucontext *mucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); int err; u32 tdn = mucontext->tdn; u16 uid = to_mpd(pd)->uid; @@ -1374,7 +1469,7 @@ static int create_raw_packet_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp, if (err) return err; - err = create_raw_packet_qp_sq(dev, sq, in, pd); + err = create_raw_packet_qp_sq(dev, udata, sq, in, pd); if (err) goto err_destroy_tis; @@ -1478,9 +1573,8 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp, struct ib_qp_init_attr *init_attr, struct ib_udata *udata) { - struct ib_uobject *uobj = pd->uobject; - struct ib_ucontext *ucontext = uobj->context; - struct mlx5_ib_ucontext *mucontext = to_mucontext(ucontext); + struct mlx5_ib_ucontext *mucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); struct mlx5_ib_create_qp_resp resp = {}; int inlen; int err; @@ -1822,6 +1916,8 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd, int inlen = MLX5_ST_SZ_BYTES(create_qp_in); struct mlx5_core_dev *mdev = dev->mdev; struct mlx5_ib_create_qp_resp resp = {}; + struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); struct mlx5_ib_cq *send_cq; struct mlx5_ib_cq *recv_cq; unsigned long flags; @@ -1924,8 +2020,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd, MLX5_QP_FLAG_TYPE_DCT)) return -EINVAL; - err = get_qp_user_index(to_mucontext(pd->uobject->context), - &ucmd, udata->inlen, &uidx); + err = get_qp_user_index(ucontext, &ucmd, udata->inlen, &uidx); if (err) return err; @@ -2409,8 +2504,11 @@ static const char *ib_qp_type_str(enum ib_qp_type type) static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd, struct ib_qp_init_attr *attr, - struct mlx5_ib_create_qp *ucmd) + struct mlx5_ib_create_qp *ucmd, + struct ib_udata *udata) { + struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); struct mlx5_ib_qp *qp; int err = 0; u32 uidx = MLX5_IB_DEFAULT_UIDX; @@ -2419,8 +2517,7 @@ static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd, if (!attr->srq || !attr->recv_cq) return ERR_PTR(-EINVAL); - err = get_qp_user_index(to_mucontext(pd->uobject->context), - ucmd, sizeof(*ucmd), &uidx); + err = get_qp_user_index(ucontext, ucmd, sizeof(*ucmd), &uidx); if (err) return ERR_PTR(err); @@ -2502,15 +2599,17 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, int err; struct ib_qp_init_attr mlx_init_attr; struct ib_qp_init_attr *init_attr = verbs_init_attr; + struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); if (pd) { dev = to_mdev(pd->device); if (init_attr->qp_type == IB_QPT_RAW_PACKET) { - if (!udata) { + if (!ucontext) { mlx5_ib_dbg(dev, "Raw Packet QP is not supported for kernel consumers\n"); return ERR_PTR(-EINVAL); - } else if (!to_mucontext(pd->uobject->context)->cqe_version) { + } else if (!ucontext->cqe_version) { mlx5_ib_dbg(dev, "Raw Packet QP is only supported for CQE version > 0\n"); return ERR_PTR(-EINVAL); } @@ -2542,7 +2641,7 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, return ERR_PTR(-EINVAL); } } else { - return mlx5_ib_create_dct(pd, init_attr, &ucmd); + return mlx5_ib_create_dct(pd, init_attr, &ucmd, udata); } } @@ -2653,10 +2752,10 @@ int mlx5_ib_destroy_qp(struct ib_qp *qp) static int to_mlx5_access_flags(struct mlx5_ib_qp *qp, const struct ib_qp_attr *attr, - int attr_mask, __be32 *hw_access_flags) + int attr_mask, __be32 *hw_access_flags_be) { u8 dest_rd_atomic; - u32 access_flags; + u32 access_flags, hw_access_flags = 0; struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.device); @@ -2674,7 +2773,7 @@ static int to_mlx5_access_flags(struct mlx5_ib_qp *qp, access_flags &= IB_ACCESS_REMOTE_WRITE; if (access_flags & IB_ACCESS_REMOTE_READ) - *hw_access_flags |= MLX5_QP_BIT_RRE; + hw_access_flags |= MLX5_QP_BIT_RRE; if (access_flags & IB_ACCESS_REMOTE_ATOMIC) { int atomic_mode; @@ -2682,14 +2781,14 @@ static int to_mlx5_access_flags(struct mlx5_ib_qp *qp, if (atomic_mode < 0) return -EOPNOTSUPP; - *hw_access_flags |= MLX5_QP_BIT_RAE; - *hw_access_flags |= atomic_mode << MLX5_ATOMIC_MODE_OFFSET; + hw_access_flags |= MLX5_QP_BIT_RAE; + hw_access_flags |= atomic_mode << MLX5_ATOMIC_MODE_OFFSET; } if (access_flags & IB_ACCESS_REMOTE_WRITE) - *hw_access_flags |= MLX5_QP_BIT_RWE; + hw_access_flags |= MLX5_QP_BIT_RWE; - *hw_access_flags = cpu_to_be32(*hw_access_flags); + *hw_access_flags_be = cpu_to_be32(hw_access_flags); return 0; } @@ -3180,14 +3279,12 @@ static int modify_raw_packet_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp, static unsigned int get_tx_affinity(struct mlx5_ib_dev *dev, struct mlx5_ib_pd *pd, struct mlx5_ib_qp_base *qp_base, - u8 port_num) + u8 port_num, struct ib_udata *udata) { - struct mlx5_ib_ucontext *ucontext = NULL; + struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); unsigned int tx_port_affinity; - if (pd && pd->ibpd.uobject && pd->ibpd.uobject->context) - ucontext = to_mucontext(pd->ibpd.uobject->context); - if (ucontext) { tx_port_affinity = (unsigned int)atomic_add_return( 1, &ucontext->tx_port_affinity) % @@ -3210,8 +3307,10 @@ static unsigned int get_tx_affinity(struct mlx5_ib_dev *dev, static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr, int attr_mask, - enum ib_qp_state cur_state, enum ib_qp_state new_state, - const struct mlx5_ib_modify_qp *ucmd) + enum ib_qp_state cur_state, + enum ib_qp_state new_state, + const struct mlx5_ib_modify_qp *ucmd, + struct ib_udata *udata) { static const u16 optab[MLX5_QP_NUM_STATE][MLX5_QP_NUM_STATE] = { [MLX5_QP_STATE_RST] = { @@ -3302,7 +3401,8 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, (ibqp->qp_type == IB_QPT_XRC_TGT)) { if (dev->lag_active) { u8 p = mlx5_core_native_port_num(dev->mdev); - tx_affinity = get_tx_affinity(dev, pd, base, p); + tx_affinity = get_tx_affinity(dev, pd, base, p, + udata); context->flags |= cpu_to_be32(tx_affinity << 24); } } @@ -3390,7 +3490,7 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, } if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC)) { - __be32 access_flags = 0; + __be32 access_flags; err = to_mlx5_access_flags(qp, attr, attr_mask, &access_flags); if (err) @@ -3785,7 +3885,7 @@ int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, } err = __mlx5_ib_modify_qp(ibqp, attr, attr_mask, cur_state, - new_state, &ucmd); + new_state, &ucmd, udata); out: mutex_unlock(&qp->mutex); @@ -5795,7 +5895,7 @@ static int prepare_user_rq(struct ib_pd *pd, return err; } - err = create_user_rq(dev, pd, rwq, &ucmd); + err = create_user_rq(dev, pd, udata, rwq, &ucmd); if (err) { mlx5_ib_dbg(dev, "err %d\n", err); return err; diff --git a/drivers/infiniband/hw/mlx5/srq.c b/drivers/infiniband/hw/mlx5/srq.c index 4e8d18009f58..1ec1beb1296b 100644 --- a/drivers/infiniband/hw/mlx5/srq.c +++ b/drivers/infiniband/hw/mlx5/srq.c @@ -47,6 +47,8 @@ static int create_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq, { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_ib_create_srq ucmd = {}; + struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mlx5_ib_ucontext, ibucontext); size_t ucmdlen; int err; int npages; @@ -71,16 +73,14 @@ static int create_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq, return -EINVAL; if (in->type != IB_SRQT_BASIC) { - err = get_srq_user_index(to_mucontext(pd->uobject->context), - &ucmd, udata->inlen, &uidx); + err = get_srq_user_index(ucontext, &ucmd, udata->inlen, &uidx); if (err) return err; } srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE); - srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, buf_size, - 0, 0); + srq->umem = ib_umem_get(udata, ucmd.buf_addr, buf_size, 0, 0); if (IS_ERR(srq->umem)) { mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size); err = PTR_ERR(srq->umem); @@ -104,8 +104,7 @@ static int create_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq, mlx5_ib_populate_pas(dev, srq->umem, page_shift, in->pas, 0); - err = mlx5_ib_db_map_user(to_mucontext(pd->uobject->context), - ucmd.db_addr, &srq->db); + err = mlx5_ib_db_map_user(ucontext, udata, ucmd.db_addr, &srq->db); if (err) { mlx5_ib_dbg(dev, "map doorbell failed\n"); goto err_in; diff --git a/drivers/infiniband/hw/mlx5/srq.h b/drivers/infiniband/hw/mlx5/srq.h index 75eb5839ae95..c330af35ff10 100644 --- a/drivers/infiniband/hw/mlx5/srq.h +++ b/drivers/infiniband/hw/mlx5/srq.h @@ -46,8 +46,6 @@ struct mlx5_core_srq { int wqe_shift; void (*event)(struct mlx5_core_srq *srq, enum mlx5_event e); - atomic_t refcount; - struct completion free; u16 uid; }; diff --git a/drivers/infiniband/hw/mlx5/srq_cmd.c b/drivers/infiniband/hw/mlx5/srq_cmd.c index 7aaaffbd4afa..63ac38bb3498 100644 --- a/drivers/infiniband/hw/mlx5/srq_cmd.c +++ b/drivers/infiniband/hw/mlx5/srq_cmd.c @@ -87,7 +87,7 @@ struct mlx5_core_srq *mlx5_cmd_get_srq(struct mlx5_ib_dev *dev, u32 srqn) srq = radix_tree_lookup(&table->tree, srqn); if (srq) - atomic_inc(&srq->refcount); + atomic_inc(&srq->common.refcount); spin_unlock(&table->lock); @@ -594,8 +594,8 @@ int mlx5_cmd_create_srq(struct mlx5_ib_dev *dev, struct mlx5_core_srq *srq, if (err) return err; - atomic_set(&srq->refcount, 1); - init_completion(&srq->free); + atomic_set(&srq->common.refcount, 1); + init_completion(&srq->common.free); spin_lock_irq(&table->lock); err = radix_tree_insert(&table->tree, srq->srqn, srq); @@ -627,9 +627,8 @@ int mlx5_cmd_destroy_srq(struct mlx5_ib_dev *dev, struct mlx5_core_srq *srq) if (err) return err; - if (atomic_dec_and_test(&srq->refcount)) - complete(&srq->free); - wait_for_completion(&srq->free); + mlx5_core_res_put(&srq->common); + wait_for_completion(&srq->common.free); return 0; } @@ -685,7 +684,7 @@ static int srq_event_notifier(struct notifier_block *nb, srq = radix_tree_lookup(&table->tree, srqn); if (srq) - atomic_inc(&srq->refcount); + atomic_inc(&srq->common.refcount); spin_unlock(&table->lock); @@ -694,8 +693,7 @@ static int srq_event_notifier(struct notifier_block *nb, srq->event(srq, eqe->type); - if (atomic_dec_and_test(&srq->refcount)) - complete(&srq->free); + mlx5_core_res_put(&srq->common); return NOTIFY_OK; } diff --git a/drivers/infiniband/hw/mthca/mthca_main.c b/drivers/infiniband/hw/mthca/mthca_main.c index 92c49bff22bc..fe9654a7af71 100644 --- a/drivers/infiniband/hw/mthca/mthca_main.c +++ b/drivers/infiniband/hw/mthca/mthca_main.c @@ -961,7 +961,7 @@ static int __mthca_init_one(struct pci_dev *pdev, int hca_type) /* We can handle large RDMA requests, so allow larger segments. */ dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024); - mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev); + mdev = ib_alloc_device(mthca_dev, ib_dev); if (!mdev) { dev_err(&pdev->dev, "Device struct alloc failed, " "aborting.\n"); diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c index e3e9dd54caa2..d063d7a37762 100644 --- a/drivers/infiniband/hw/mthca/mthca_provider.c +++ b/drivers/infiniband/hw/mthca/mthca_provider.c @@ -37,6 +37,7 @@ #include <rdma/ib_smi.h> #include <rdma/ib_umem.h> #include <rdma/ib_user_verbs.h> +#include <rdma/uverbs_ioctl.h> #include <linux/sched.h> #include <linux/slab.h> @@ -300,17 +301,16 @@ static int mthca_query_gid(struct ib_device *ibdev, u8 port, return err; } -static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +static int mthca_alloc_ucontext(struct ib_ucontext *uctx, + struct ib_udata *udata) { - struct mthca_alloc_ucontext_resp uresp; - struct mthca_ucontext *context; + struct ib_device *ibdev = uctx->device; + struct mthca_alloc_ucontext_resp uresp = {}; + struct mthca_ucontext *context = to_mucontext(uctx); int err; if (!(to_mdev(ibdev)->active)) - return ERR_PTR(-EAGAIN); - - memset(&uresp, 0, sizeof uresp); + return -EAGAIN; uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps; if (mthca_is_memfree(to_mdev(ibdev))) @@ -318,44 +318,33 @@ static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev, else uresp.uarc_size = 0; - context = kmalloc(sizeof *context, GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); - err = mthca_uar_alloc(to_mdev(ibdev), &context->uar); - if (err) { - kfree(context); - return ERR_PTR(err); - } + if (err) + return err; context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev)); if (IS_ERR(context->db_tab)) { err = PTR_ERR(context->db_tab); mthca_uar_free(to_mdev(ibdev), &context->uar); - kfree(context); - return ERR_PTR(err); + return err; } - if (ib_copy_to_udata(udata, &uresp, sizeof uresp)) { + if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) { mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab); mthca_uar_free(to_mdev(ibdev), &context->uar); - kfree(context); - return ERR_PTR(-EFAULT); + return -EFAULT; } context->reg_mr_warned = 0; - return &context->ibucontext; + return 0; } -static int mthca_dealloc_ucontext(struct ib_ucontext *context) +static void mthca_dealloc_ucontext(struct ib_ucontext *context) { mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar, to_mucontext(context)->db_tab); mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar); - kfree(to_mucontext(context)); - - return 0; } static int mthca_mmap_uar(struct ib_ucontext *context, @@ -374,40 +363,30 @@ static int mthca_mmap_uar(struct ib_ucontext *context, return 0; } -static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +static int mthca_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct mthca_pd *pd; + struct ib_device *ibdev = ibpd->device; + struct mthca_pd *pd = to_mpd(ibpd); int err; - pd = kmalloc(sizeof *pd, GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - err = mthca_pd_alloc(to_mdev(ibdev), !context, pd); - if (err) { - kfree(pd); - return ERR_PTR(err); - } + if (err) + return err; if (context) { if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) { mthca_pd_free(to_mdev(ibdev), pd); - kfree(pd); - return ERR_PTR(-EFAULT); + return -EFAULT; } } - return &pd->ibpd; + return 0; } -static int mthca_dealloc_pd(struct ib_pd *pd) +static void mthca_dealloc_pd(struct ib_pd *pd) { mthca_pd_free(to_mdev(pd->device), to_mpd(pd)); - kfree(pd); - - return 0; } static struct ib_ah *mthca_ah_create(struct ib_pd *pd, @@ -445,7 +424,8 @@ static struct ib_srq *mthca_create_srq(struct ib_pd *pd, struct ib_udata *udata) { struct mthca_create_srq ucmd; - struct mthca_ucontext *context = NULL; + struct mthca_ucontext *context = rdma_udata_to_drv_context( + udata, struct mthca_ucontext, ibucontext); struct mthca_srq *srq; int err; @@ -457,8 +437,6 @@ static struct ib_srq *mthca_create_srq(struct ib_pd *pd, return ERR_PTR(-ENOMEM); if (udata) { - context = to_mucontext(pd->uobject->context); - if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) { err = -EFAULT; goto err_free; @@ -520,6 +498,8 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *init_attr, struct ib_udata *udata) { + struct mthca_ucontext *context = rdma_udata_to_drv_context( + udata, struct mthca_ucontext, ibucontext); struct mthca_create_qp ucmd; struct mthca_qp *qp; int err; @@ -532,15 +512,11 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd, case IB_QPT_UC: case IB_QPT_UD: { - struct mthca_ucontext *context; - qp = kzalloc(sizeof(*qp), GFP_KERNEL); if (!qp) return ERR_PTR(-ENOMEM); if (udata) { - context = to_mucontext(pd->uobject->context); - if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) { kfree(qp); return ERR_PTR(-EFAULT); @@ -578,8 +554,6 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd, &init_attr->cap, qp, udata); if (err && udata) { - context = to_mucontext(pd->uobject->context); - mthca_unmap_user_db(to_mdev(pd->device), &context->uar, context->db_tab, @@ -684,7 +658,7 @@ static struct ib_cq *mthca_create_cq(struct ib_device *ibdev, goto err_unmap_set; } - cq = kmalloc(sizeof *cq, GFP_KERNEL); + cq = kzalloc(sizeof(*cq), GFP_KERNEL); if (!cq) { err = -ENOMEM; goto err_unmap_arm; @@ -907,22 +881,23 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt, int acc, struct ib_udata *udata) { struct mthca_dev *dev = to_mdev(pd->device); - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; + struct mthca_ucontext *context = rdma_udata_to_drv_context( + udata, struct mthca_ucontext, ibucontext); struct mthca_mr *mr; struct mthca_reg_mr ucmd; u64 *pages; - int shift, n, len; - int i, k, entry; + int n, i; int err = 0; int write_mtt_size; if (udata->inlen < sizeof ucmd) { - if (!to_mucontext(pd->uobject->context)->reg_mr_warned) { + if (!context->reg_mr_warned) { mthca_warn(dev, "Process '%s' did not pass in MR attrs.\n", current->comm); mthca_warn(dev, " Update libmthca to fix this.\n"); } - ++to_mucontext(pd->uobject->context)->reg_mr_warned; + ++context->reg_mr_warned; ucmd.mr_attrs = 0; } else if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) return ERR_PTR(-EFAULT); @@ -931,7 +906,7 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, if (!mr) return ERR_PTR(-ENOMEM); - mr->umem = ib_umem_get(pd->uobject->context, start, length, acc, + mr->umem = ib_umem_get(udata, start, length, acc, ucmd.mr_attrs & MTHCA_MR_DMASYNC); if (IS_ERR(mr->umem)) { @@ -939,7 +914,6 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, goto err; } - shift = mr->umem->page_shift; n = mr->umem->nmap; mr->mtt = mthca_alloc_mtt(dev, n); @@ -958,21 +932,19 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages)); - for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) { - len = sg_dma_len(sg) >> shift; - for (k = 0; k < len; ++k) { - pages[i++] = sg_dma_address(sg) + (k << shift); - /* - * Be friendly to write_mtt and pass it chunks - * of appropriate size. - */ - if (i == write_mtt_size) { - err = mthca_write_mtt(dev, mr->mtt, n, pages, i); - if (err) - goto mtt_done; - n += i; - i = 0; - } + for_each_sg_dma_page(mr->umem->sg_head.sgl, &sg_iter, mr->umem->nmap, 0) { + pages[i++] = sg_page_iter_dma_address(&sg_iter); + + /* + * Be friendly to write_mtt and pass it chunks + * of appropriate size. + */ + if (i == write_mtt_size) { + err = mthca_write_mtt(dev, mr->mtt, n, pages, i); + if (err) + goto mtt_done; + n += i; + i = 0; } } @@ -983,7 +955,7 @@ mtt_done: if (err) goto err_mtt; - err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, shift, virt, length, + err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, PAGE_SHIFT, virt, length, convert_access(acc), mr); if (err) @@ -1081,7 +1053,8 @@ static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { struct mthca_dev *dev = - container_of(device, struct mthca_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mthca_dev, ib_dev); + return sprintf(buf, "%x\n", dev->rev_id); } static DEVICE_ATTR_RO(hw_rev); @@ -1090,7 +1063,8 @@ static ssize_t hca_type_show(struct device *device, struct device_attribute *attr, char *buf) { struct mthca_dev *dev = - container_of(device, struct mthca_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mthca_dev, ib_dev); + switch (dev->pdev->device) { case PCI_DEVICE_ID_MELLANOX_TAVOR: return sprintf(buf, "MT23108\n"); @@ -1111,7 +1085,8 @@ static ssize_t board_id_show(struct device *device, struct device_attribute *attr, char *buf) { struct mthca_dev *dev = - container_of(device, struct mthca_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct mthca_dev, ib_dev); + return sprintf(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id); } static DEVICE_ATTR_RO(board_id); @@ -1225,6 +1200,8 @@ static const struct ib_device_ops mthca_dev_ops = { .query_qp = mthca_query_qp, .reg_user_mr = mthca_reg_user_mr, .resize_cq = mthca_resize_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, mthca_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, mthca_ucontext, ibucontext), }; static const struct ib_device_ops mthca_dev_arbel_srq_ops = { @@ -1338,7 +1315,7 @@ int mthca_register_device(struct mthca_dev *dev) rdma_set_device_sysfs_group(&dev->ib_dev, &mthca_attr_group); dev->ib_dev.driver_id = RDMA_DRIVER_MTHCA; - ret = ib_register_device(&dev->ib_dev, "mthca%d", NULL); + ret = ib_register_device(&dev->ib_dev, "mthca%d"); if (ret) return ret; diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c index 4e5b5cc17f1d..7a5b25d13faa 100644 --- a/drivers/infiniband/hw/mthca/mthca_qp.c +++ b/drivers/infiniband/hw/mthca/mthca_qp.c @@ -42,6 +42,7 @@ #include <rdma/ib_verbs.h> #include <rdma/ib_cache.h> #include <rdma/ib_pack.h> +#include <rdma/uverbs_ioctl.h> #include "mthca_dev.h" #include "mthca_cmd.h" @@ -554,10 +555,14 @@ static int mthca_path_set(struct mthca_dev *dev, const struct rdma_ah_attr *ah, static int __mthca_modify_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr, int attr_mask, - enum ib_qp_state cur_state, enum ib_qp_state new_state) + enum ib_qp_state cur_state, + enum ib_qp_state new_state, + struct ib_udata *udata) { struct mthca_dev *dev = to_mdev(ibqp->device); struct mthca_qp *qp = to_mqp(ibqp); + struct mthca_ucontext *context = rdma_udata_to_drv_context( + udata, struct mthca_ucontext, ibucontext); struct mthca_mailbox *mailbox; struct mthca_qp_param *qp_param; struct mthca_qp_context *qp_context; @@ -619,8 +624,7 @@ static int __mthca_modify_qp(struct ib_qp *ibqp, /* leave arbel_sched_queue as 0 */ if (qp->ibqp.uobject) - qp_context->usr_page = - cpu_to_be32(to_mucontext(qp->ibqp.uobject->context)->uar.index); + qp_context->usr_page = cpu_to_be32(context->uar.index); else qp_context->usr_page = cpu_to_be32(dev->driver_uar.index); qp_context->local_qpn = cpu_to_be32(qp->qpn); @@ -913,7 +917,8 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask, goto out; } - err = __mthca_modify_qp(ibqp, attr, attr_mask, cur_state, new_state); + err = __mthca_modify_qp(ibqp, attr, attr_mask, cur_state, new_state, + udata); out: mutex_unlock(&qp->mutex); diff --git a/drivers/infiniband/hw/mthca/mthca_srq.c b/drivers/infiniband/hw/mthca/mthca_srq.c index b8333c79e3fa..06b920385512 100644 --- a/drivers/infiniband/hw/mthca/mthca_srq.c +++ b/drivers/infiniband/hw/mthca/mthca_srq.c @@ -36,6 +36,8 @@ #include <asm/io.h> +#include <rdma/uverbs_ioctl.h> + #include "mthca_dev.h" #include "mthca_cmd.h" #include "mthca_memfree.h" @@ -96,17 +98,19 @@ static void mthca_tavor_init_srq_context(struct mthca_dev *dev, struct mthca_pd *pd, struct mthca_srq *srq, struct mthca_tavor_srq_context *context, - bool is_user) + struct ib_udata *udata) { + struct mthca_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mthca_ucontext, ibucontext); + memset(context, 0, sizeof *context); context->wqe_base_ds = cpu_to_be64(1 << (srq->wqe_shift - 4)); context->state_pd = cpu_to_be32(pd->pd_num); context->lkey = cpu_to_be32(srq->mr.ibmr.lkey); - if (is_user) - context->uar = - cpu_to_be32(to_mucontext(pd->ibpd.uobject->context)->uar.index); + if (udata) + context->uar = cpu_to_be32(ucontext->uar.index); else context->uar = cpu_to_be32(dev->driver_uar.index); } @@ -115,8 +119,10 @@ static void mthca_arbel_init_srq_context(struct mthca_dev *dev, struct mthca_pd *pd, struct mthca_srq *srq, struct mthca_arbel_srq_context *context, - bool is_user) + struct ib_udata *udata) { + struct mthca_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct mthca_ucontext, ibucontext); int logsize, max; memset(context, 0, sizeof *context); @@ -131,9 +137,8 @@ static void mthca_arbel_init_srq_context(struct mthca_dev *dev, context->lkey = cpu_to_be32(srq->mr.ibmr.lkey); context->db_index = cpu_to_be32(srq->db_index); context->logstride_usrpage = cpu_to_be32((srq->wqe_shift - 4) << 29); - if (is_user) - context->logstride_usrpage |= - cpu_to_be32(to_mucontext(pd->ibpd.uobject->context)->uar.index); + if (udata) + context->logstride_usrpage |= cpu_to_be32(ucontext->uar.index); else context->logstride_usrpage |= cpu_to_be32(dev->driver_uar.index); context->eq_pd = cpu_to_be32(MTHCA_EQ_ASYNC << 24 | pd->pd_num); diff --git a/drivers/infiniband/hw/nes/Kconfig b/drivers/infiniband/hw/nes/Kconfig index 7964eba8e7ed..52caae954e4a 100644 --- a/drivers/infiniband/hw/nes/Kconfig +++ b/drivers/infiniband/hw/nes/Kconfig @@ -1,6 +1,6 @@ config INFINIBAND_NES tristate "NetEffect RNIC Driver" - depends on PCI && INET && INFINIBAND + depends on PCI && INET select LIBCRC32C ---help--- This is the RDMA Network Interface Card (RNIC) driver for diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c index 4e7f08ee1907..828e4af3f951 100644 --- a/drivers/infiniband/hw/nes/nes_verbs.c +++ b/drivers/infiniband/hw/nes/nes_verbs.c @@ -41,6 +41,7 @@ #include <rdma/ib_verbs.h> #include <rdma/iw_cm.h> #include <rdma/ib_user_verbs.h> +#include <rdma/uverbs_ioctl.h> #include "nes.h" @@ -528,42 +529,36 @@ static int nes_query_gid(struct ib_device *ibdev, u8 port, * nes_alloc_ucontext - Allocate the user context data structure. This keeps track * of all objects associated with a particular user-mode client. */ -static struct ib_ucontext *nes_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +static int nes_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { + struct ib_device *ibdev = uctx->device; struct nes_vnic *nesvnic = to_nesvnic(ibdev); struct nes_device *nesdev = nesvnic->nesdev; struct nes_adapter *nesadapter = nesdev->nesadapter; struct nes_alloc_ucontext_req req; - struct nes_alloc_ucontext_resp uresp; - struct nes_ucontext *nes_ucontext; + struct nes_alloc_ucontext_resp uresp = {}; + struct nes_ucontext *nes_ucontext = to_nesucontext(uctx); struct nes_ib_device *nesibdev = nesvnic->nesibdev; if (ib_copy_from_udata(&req, udata, sizeof(struct nes_alloc_ucontext_req))) { printk(KERN_ERR PFX "Invalid structure size on allocate user context.\n"); - return ERR_PTR(-EINVAL); + return -EINVAL; } if (req.userspace_ver != NES_ABI_USERSPACE_VER) { printk(KERN_ERR PFX "Invalid userspace driver version detected. Detected version %d, should be %d\n", req.userspace_ver, NES_ABI_USERSPACE_VER); - return ERR_PTR(-EINVAL); + return -EINVAL; } - memset(&uresp, 0, sizeof uresp); - uresp.max_qps = nesibdev->max_qp; uresp.max_pds = nesibdev->max_pd; uresp.wq_size = nesdev->nesadapter->max_qp_wr * 2; uresp.virtwq = nesadapter->virtwq; uresp.kernel_ver = NES_ABI_KERNEL_VER; - nes_ucontext = kzalloc(sizeof *nes_ucontext, GFP_KERNEL); - if (!nes_ucontext) - return ERR_PTR(-ENOMEM); - nes_ucontext->nesdev = nesdev; nes_ucontext->mmap_wq_offset = uresp.max_pds; nes_ucontext->mmap_cq_offset = nes_ucontext->mmap_wq_offset + @@ -571,34 +566,22 @@ static struct ib_ucontext *nes_alloc_ucontext(struct ib_device *ibdev, PAGE_SIZE; - if (ib_copy_to_udata(udata, &uresp, sizeof uresp)) { - kfree(nes_ucontext); - return ERR_PTR(-EFAULT); - } + if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) + return -EFAULT; INIT_LIST_HEAD(&nes_ucontext->cq_reg_mem_list); INIT_LIST_HEAD(&nes_ucontext->qp_reg_mem_list); - atomic_set(&nes_ucontext->usecnt, 1); - return &nes_ucontext->ibucontext; + return 0; } - /** * nes_dealloc_ucontext */ -static int nes_dealloc_ucontext(struct ib_ucontext *context) +static void nes_dealloc_ucontext(struct ib_ucontext *context) { - /* struct nes_vnic *nesvnic = to_nesvnic(context->device); */ - /* struct nes_device *nesdev = nesvnic->nesdev; */ - struct nes_ucontext *nes_ucontext = to_nesucontext(context); - - if (!atomic_dec_and_test(&nes_ucontext->usecnt)) - return 0; - kfree(nes_ucontext); - return 0; + return; } - /** * nes_mmap */ @@ -658,10 +641,11 @@ static int nes_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) /** * nes_alloc_pd */ -static struct ib_pd *nes_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, struct ib_udata *udata) +static int nes_alloc_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct nes_pd *nespd; + struct ib_device *ibdev = pd->device; + struct nes_pd *nespd = to_nespd(pd); struct nes_vnic *nesvnic = to_nesvnic(ibdev); struct nes_device *nesdev = nesvnic->nesdev; struct nes_adapter *nesadapter = nesdev->nesadapter; @@ -676,15 +660,8 @@ static struct ib_pd *nes_alloc_pd(struct ib_device *ibdev, err = nes_alloc_resource(nesadapter, nesadapter->allocated_pds, nesadapter->max_pd, &pd_num, &nesadapter->next_pd, NES_RESOURCE_PD); - if (err) { - return ERR_PTR(err); - } - - nespd = kzalloc(sizeof (struct nes_pd), GFP_KERNEL); - if (!nespd) { - nes_free_resource(nesadapter, nesadapter->allocated_pds, pd_num); - return ERR_PTR(-ENOMEM); - } + if (err) + return err; nes_debug(NES_DBG_PD, "Allocating PD (%p) for ib device %s\n", nespd, dev_name(&nesvnic->nesibdev->ibdev.dev)); @@ -700,16 +677,14 @@ static struct ib_pd *nes_alloc_pd(struct ib_device *ibdev, if (nespd->mmap_db_index >= NES_MAX_USER_DB_REGIONS) { nes_debug(NES_DBG_PD, "mmap_db_index > MAX\n"); nes_free_resource(nesadapter, nesadapter->allocated_pds, pd_num); - kfree(nespd); - return ERR_PTR(-ENOMEM); + return -ENOMEM; } uresp.pd_id = nespd->pd_id; uresp.mmap_db_index = nespd->mmap_db_index; if (ib_copy_to_udata(udata, &uresp, sizeof (struct nes_alloc_pd_resp))) { nes_free_resource(nesadapter, nesadapter->allocated_pds, pd_num); - kfree(nespd); - return ERR_PTR(-EFAULT); + return -EFAULT; } set_bit(nespd->mmap_db_index, nesucontext->allocated_doorbells); @@ -718,14 +693,14 @@ static struct ib_pd *nes_alloc_pd(struct ib_device *ibdev, } nes_debug(NES_DBG_PD, "PD%u structure located @%p.\n", nespd->pd_id, nespd); - return &nespd->ibpd; + return 0; } /** * nes_dealloc_pd */ -static int nes_dealloc_pd(struct ib_pd *ibpd) +static void nes_dealloc_pd(struct ib_pd *ibpd) { struct nes_ucontext *nesucontext; struct nes_pd *nespd = to_nespd(ibpd); @@ -748,9 +723,6 @@ static int nes_dealloc_pd(struct ib_pd *ibpd) nespd->pd_id, nespd); nes_free_resource(nesadapter, nesadapter->allocated_pds, (nespd->pd_id-nesadapter->base_pd)>>(PAGE_SHIFT-12)); - kfree(nespd); - - return 0; } @@ -985,7 +957,8 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd, struct nes_adapter *nesadapter = nesdev->nesadapter; struct nes_qp *nesqp; struct nes_cq *nescq; - struct nes_ucontext *nes_ucontext; + struct nes_ucontext *nes_ucontext = rdma_udata_to_drv_context( + udata, struct nes_ucontext, ibucontext); struct nes_hw_cqp_wqe *cqp_wqe; struct nes_cqp_request *cqp_request; struct nes_create_qp_req req; @@ -1066,9 +1039,8 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd, } if (req.user_qp_buffer) nesqp->nesuqp_addr = req.user_qp_buffer; - if (udata && (ibpd->uobject->context)) { + if (udata) { nesqp->user_mode = 1; - nes_ucontext = to_nesucontext(ibpd->uobject->context); if (virt_wqs) { err = 1; list_for_each_entry(nespbl, &nes_ucontext->qp_reg_mem_list, list) { @@ -1089,7 +1061,6 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd, } } - nes_ucontext = to_nesucontext(ibpd->uobject->context); nesqp->mmap_sq_db_index = find_next_zero_bit(nes_ucontext->allocated_wqs, NES_MAX_USER_WQ_REGIONS, nes_ucontext->first_free_wq); @@ -2109,18 +2080,18 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, struct nes_device *nesdev = nesvnic->nesdev; struct nes_adapter *nesadapter = nesdev->nesadapter; struct ib_mr *ibmr = ERR_PTR(-EINVAL); - struct scatterlist *sg; - struct nes_ucontext *nes_ucontext; + struct sg_dma_page_iter dma_iter; + struct nes_ucontext *nes_ucontext = rdma_udata_to_drv_context( + udata, struct nes_ucontext, ibucontext); struct nes_pbl *nespbl; struct nes_mr *nesmr; struct ib_umem *region; struct nes_mem_reg_req req; struct nes_vpbl vpbl; struct nes_root_vpbl root_vpbl; - int entry, page_index; + int page_index; int page_count = 0; int err, pbl_depth = 0; - int chunk_pages; int ret; u32 stag; u32 stag_index = 0; @@ -2132,9 +2103,8 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u16 pbl_count; u8 single_page = 1; u8 stag_key; - int first_page = 1; - region = ib_umem_get(pd->uobject->context, start, length, acc, 0); + region = ib_umem_get(udata, start, length, acc, 0); if (IS_ERR(region)) { return (struct ib_mr *)region; } @@ -2183,127 +2153,99 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, } nesmr->region = region; - for_each_sg(region->sg_head.sgl, sg, region->nmap, entry) { - if (sg_dma_address(sg) & ~PAGE_MASK) { - ib_umem_release(region); - nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index); - nes_debug(NES_DBG_MR, "Unaligned Memory Buffer: 0x%x\n", - (unsigned int) sg_dma_address(sg)); - ibmr = ERR_PTR(-EINVAL); - kfree(nesmr); - goto reg_user_mr_err; - } + for_each_sg_dma_page (region->sg_head.sgl, &dma_iter, region->nmap, 0) { - if (!sg_dma_len(sg)) { - ib_umem_release(region); - nes_free_resource(nesadapter, nesadapter->allocated_mrs, - stag_index); - nes_debug(NES_DBG_MR, "Invalid Buffer Size\n"); - ibmr = ERR_PTR(-EINVAL); - kfree(nesmr); - goto reg_user_mr_err; - } - - region_length += sg_dma_len(sg); - chunk_pages = sg_dma_len(sg) >> 12; + region_length += PAGE_SIZE; region_length -= skip_pages << 12; - for (page_index = skip_pages; page_index < chunk_pages; page_index++) { - skip_pages = 0; - if ((page_count != 0) && (page_count << 12) - (ib_umem_offset(region) & (4096 - 1)) >= region->length) - goto enough_pages; - if ((page_count&0x01FF) == 0) { - if (page_count >= 1024 * 512) { + skip_pages = 0; + if ((page_count != 0) && (page_count << 12) - (ib_umem_offset(region) & (4096 - 1)) >= region->length) + goto enough_pages; + if ((page_count & 0x01FF) == 0) { + if (page_count >= 1024 * 512) { + ib_umem_release(region); + nes_free_resource(nesadapter, + nesadapter->allocated_mrs, stag_index); + kfree(nesmr); + ibmr = ERR_PTR(-E2BIG); + goto reg_user_mr_err; + } + if (root_pbl_index == 1) { + root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, + 8192, &root_vpbl.pbl_pbase); + nes_debug(NES_DBG_MR, "Allocating root PBL, va = %p, pa = 0x%08X\n", + root_vpbl.pbl_vbase, (unsigned int)root_vpbl.pbl_pbase); + if (!root_vpbl.pbl_vbase) { ib_umem_release(region); - nes_free_resource(nesadapter, - nesadapter->allocated_mrs, stag_index); + pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase, + vpbl.pbl_pbase); + nes_free_resource(nesadapter, nesadapter->allocated_mrs, + stag_index); kfree(nesmr); - ibmr = ERR_PTR(-E2BIG); + ibmr = ERR_PTR(-ENOMEM); goto reg_user_mr_err; } - if (root_pbl_index == 1) { - root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, - 8192, &root_vpbl.pbl_pbase); - nes_debug(NES_DBG_MR, "Allocating root PBL, va = %p, pa = 0x%08X\n", - root_vpbl.pbl_vbase, (unsigned int)root_vpbl.pbl_pbase); - if (!root_vpbl.pbl_vbase) { - ib_umem_release(region); - pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase, - vpbl.pbl_pbase); - nes_free_resource(nesadapter, nesadapter->allocated_mrs, - stag_index); - kfree(nesmr); - ibmr = ERR_PTR(-ENOMEM); - goto reg_user_mr_err; - } - root_vpbl.leaf_vpbl = kcalloc(1024, - sizeof(*root_vpbl.leaf_vpbl), - GFP_KERNEL); - if (!root_vpbl.leaf_vpbl) { - ib_umem_release(region); - pci_free_consistent(nesdev->pcidev, 8192, root_vpbl.pbl_vbase, - root_vpbl.pbl_pbase); - pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase, - vpbl.pbl_pbase); - nes_free_resource(nesadapter, nesadapter->allocated_mrs, - stag_index); - kfree(nesmr); - ibmr = ERR_PTR(-ENOMEM); - goto reg_user_mr_err; - } - root_vpbl.pbl_vbase[0].pa_low = - cpu_to_le32((u32)vpbl.pbl_pbase); - root_vpbl.pbl_vbase[0].pa_high = - cpu_to_le32((u32)((((u64)vpbl.pbl_pbase) >> 32))); - root_vpbl.leaf_vpbl[0] = vpbl; - } - vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096, - &vpbl.pbl_pbase); - nes_debug(NES_DBG_MR, "Allocating leaf PBL, va = %p, pa = 0x%08X\n", - vpbl.pbl_vbase, (unsigned int)vpbl.pbl_pbase); - if (!vpbl.pbl_vbase) { + root_vpbl.leaf_vpbl = kcalloc(1024, + sizeof(*root_vpbl.leaf_vpbl), + GFP_KERNEL); + if (!root_vpbl.leaf_vpbl) { ib_umem_release(region); - nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index); - ibmr = ERR_PTR(-ENOMEM); + pci_free_consistent(nesdev->pcidev, 8192, root_vpbl.pbl_vbase, + root_vpbl.pbl_pbase); + pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase, + vpbl.pbl_pbase); + nes_free_resource(nesadapter, nesadapter->allocated_mrs, + stag_index); kfree(nesmr); + ibmr = ERR_PTR(-ENOMEM); goto reg_user_mr_err; } - if (1 <= root_pbl_index) { - root_vpbl.pbl_vbase[root_pbl_index].pa_low = - cpu_to_le32((u32)vpbl.pbl_pbase); - root_vpbl.pbl_vbase[root_pbl_index].pa_high = - cpu_to_le32((u32)((((u64)vpbl.pbl_pbase)>>32))); - root_vpbl.leaf_vpbl[root_pbl_index] = vpbl; - } - root_pbl_index++; - cur_pbl_index = 0; + root_vpbl.pbl_vbase[0].pa_low = + cpu_to_le32((u32)vpbl.pbl_pbase); + root_vpbl.pbl_vbase[0].pa_high = + cpu_to_le32((u32)((((u64)vpbl.pbl_pbase) >> 32))); + root_vpbl.leaf_vpbl[0] = vpbl; } - if (single_page) { - if (page_count != 0) { - if ((last_dma_addr+4096) != - (sg_dma_address(sg)+ - (page_index*4096))) - single_page = 0; - last_dma_addr = sg_dma_address(sg)+ - (page_index*4096); - } else { - first_dma_addr = sg_dma_address(sg)+ - (page_index*4096); - last_dma_addr = first_dma_addr; - } + vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096, + &vpbl.pbl_pbase); + nes_debug(NES_DBG_MR, "Allocating leaf PBL, va = %p, pa = 0x%08X\n", + vpbl.pbl_vbase, (unsigned int)vpbl.pbl_pbase); + if (!vpbl.pbl_vbase) { + ib_umem_release(region); + nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index); + ibmr = ERR_PTR(-ENOMEM); + kfree(nesmr); + goto reg_user_mr_err; } - - vpbl.pbl_vbase[cur_pbl_index].pa_low = - cpu_to_le32((u32)(sg_dma_address(sg)+ - (page_index*4096))); - vpbl.pbl_vbase[cur_pbl_index].pa_high = - cpu_to_le32((u32)((((u64)(sg_dma_address(sg)+ - (page_index*4096))) >> 32))); - cur_pbl_index++; - page_count++; + if (1 <= root_pbl_index) { + root_vpbl.pbl_vbase[root_pbl_index].pa_low = + cpu_to_le32((u32)vpbl.pbl_pbase); + root_vpbl.pbl_vbase[root_pbl_index].pa_high = + cpu_to_le32((u32)((((u64)vpbl.pbl_pbase) >> 32))); + root_vpbl.leaf_vpbl[root_pbl_index] = vpbl; + } + root_pbl_index++; + cur_pbl_index = 0; } + if (single_page) { + if (page_count != 0) { + if ((last_dma_addr + 4096) != sg_page_iter_dma_address(&dma_iter)) + single_page = 0; + last_dma_addr = sg_page_iter_dma_address(&dma_iter); + } else { + first_dma_addr = sg_page_iter_dma_address(&dma_iter); + last_dma_addr = first_dma_addr; + } + } + + vpbl.pbl_vbase[cur_pbl_index].pa_low = + cpu_to_le32((u32)(sg_page_iter_dma_address(&dma_iter))); + vpbl.pbl_vbase[cur_pbl_index].pa_high = + cpu_to_le32((u32)((u64)(sg_page_iter_dma_address(&dma_iter)))); + cur_pbl_index++; + page_count++; } - enough_pages: +enough_pages: nes_debug(NES_DBG_MR, "calculating stag, stag_index=0x%08x, driver_key=0x%08x," " stag_key=0x%08x\n", stag_index, driver_key, stag_key); @@ -2345,7 +2287,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, ibmr = ERR_PTR(-ENOMEM); } - reg_user_mr_err: +reg_user_mr_err: /* free the resources */ if (root_pbl_index == 1) { pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase, @@ -2383,7 +2325,6 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, return ERR_PTR(-ENOMEM); } nesmr->region = region; - nes_ucontext = to_nesucontext(pd->uobject->context); pbl_depth = region->length >> 12; pbl_depth += (region->length & (4096-1)) ? 1 : 0; nespbl->pbl_size = pbl_depth*sizeof(u64); @@ -2412,26 +2353,14 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, nespbl->pbl_size, (unsigned long) nespbl->pbl_pbase, (void *) nespbl->pbl_vbase, nespbl->user_base); - for_each_sg(region->sg_head.sgl, sg, region->nmap, entry) { - chunk_pages = sg_dma_len(sg) >> 12; - chunk_pages += (sg_dma_len(sg) & (4096-1)) ? 1 : 0; - if (first_page) { - nespbl->page = sg_page(sg); - first_page = 0; - } - - for (page_index = 0; page_index < chunk_pages; page_index++) { - ((__le32 *)pbl)[0] = cpu_to_le32((u32) - (sg_dma_address(sg)+ - (page_index*4096))); - ((__le32 *)pbl)[1] = cpu_to_le32(((u64) - (sg_dma_address(sg)+ - (page_index*4096)))>>32); - nes_debug(NES_DBG_MR, "pbl=%p, *pbl=0x%016llx, 0x%08x%08x\n", pbl, - (unsigned long long)*pbl, - le32_to_cpu(((__le32 *)pbl)[1]), le32_to_cpu(((__le32 *)pbl)[0])); - pbl++; - } + nespbl->page = sg_page(region->sg_head.sgl); + for_each_sg_dma_page(region->sg_head.sgl, &dma_iter, region->nmap, 0) { + ((__le32 *)pbl)[0] = cpu_to_le32((u32)(sg_page_iter_dma_address(&dma_iter))); + ((__le32 *)pbl)[1] = cpu_to_le32(((u64)(sg_page_iter_dma_address(&dma_iter)))>>32); + nes_debug(NES_DBG_MR, "pbl=%p, *pbl=0x%016llx, 0x%08x%08x\n", pbl, + (unsigned long long)*pbl, + le32_to_cpu(((__le32 *)pbl)[1]), le32_to_cpu(((__le32 *)pbl)[0])); + pbl++; } if (req.reg_type == IWNES_MEMREG_TYPE_QP) { @@ -2560,7 +2489,7 @@ static ssize_t hw_rev_show(struct device *dev, struct device_attribute *attr, char *buf) { struct nes_ib_device *nesibdev = - container_of(dev, struct nes_ib_device, ibdev.dev); + rdma_device_to_drv_device(dev, struct nes_ib_device, ibdev); struct nes_vnic *nesvnic = nesibdev->nesvnic; nes_debug(NES_DBG_INIT, "\n"); @@ -3658,6 +3587,8 @@ static const struct ib_device_ops nes_dev_ops = { .query_qp = nes_query_qp, .reg_user_mr = nes_reg_user_mr, .req_notify_cq = nes_req_notify_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, nes_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, nes_ucontext, ibucontext), }; /** @@ -3669,7 +3600,7 @@ struct nes_ib_device *nes_init_ofa_device(struct net_device *netdev) struct nes_vnic *nesvnic = netdev_priv(netdev); struct nes_device *nesdev = nesvnic->nesdev; - nesibdev = (struct nes_ib_device *)ib_alloc_device(sizeof(struct nes_ib_device)); + nesibdev = ib_alloc_device(nes_ib_device, ibdev); if (nesibdev == NULL) { return NULL; } @@ -3801,7 +3732,7 @@ int nes_register_ofa_device(struct nes_ib_device *nesibdev) rdma_set_device_sysfs_group(&nesvnic->nesibdev->ibdev, &nes_attr_group); nesvnic->nesibdev->ibdev.driver_id = RDMA_DRIVER_NES; - ret = ib_register_device(&nesvnic->nesibdev->ibdev, "nes%d", NULL); + ret = ib_register_device(&nesvnic->nesibdev->ibdev, "nes%d"); if (ret) { return ret; } diff --git a/drivers/infiniband/hw/nes/nes_verbs.h b/drivers/infiniband/hw/nes/nes_verbs.h index e02a5662dc20..114a9b59fefd 100644 --- a/drivers/infiniband/hw/nes/nes_verbs.h +++ b/drivers/infiniband/hw/nes/nes_verbs.h @@ -59,7 +59,6 @@ struct nes_ucontext { struct list_head cq_reg_mem_list; struct list_head qp_reg_mem_list; u32 mcrqf; - atomic_t usecnt; }; struct nes_pd { diff --git a/drivers/infiniband/hw/ocrdma/Makefile b/drivers/infiniband/hw/ocrdma/Makefile index d1bfd4f4cdde..e3f20ca15462 100644 --- a/drivers/infiniband/hw/ocrdma/Makefile +++ b/drivers/infiniband/hw/ocrdma/Makefile @@ -1,4 +1,4 @@ -ccflags-y := -Idrivers/net/ethernet/emulex/benet +ccflags-y := -I $(srctree)/drivers/net/ethernet/emulex/benet obj-$(CONFIG_INFINIBAND_OCRDMA) += ocrdma.o diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_main.c b/drivers/infiniband/hw/ocrdma/ocrdma_main.c index 1f393842453a..b9e10d55a58e 100644 --- a/drivers/infiniband/hw/ocrdma/ocrdma_main.c +++ b/drivers/infiniband/hw/ocrdma/ocrdma_main.c @@ -118,7 +118,8 @@ static void get_dev_fw_str(struct ib_device *device, char *str) static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { - struct ocrdma_dev *dev = dev_get_drvdata(device); + struct ocrdma_dev *dev = + rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev); return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor); } @@ -127,7 +128,8 @@ static DEVICE_ATTR_RO(hw_rev); static ssize_t hca_type_show(struct device *device, struct device_attribute *attr, char *buf) { - struct ocrdma_dev *dev = dev_get_drvdata(device); + struct ocrdma_dev *dev = + rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev); return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->model_number[0]); } @@ -177,6 +179,8 @@ static const struct ib_device_ops ocrdma_dev_ops = { .reg_user_mr = ocrdma_reg_user_mr, .req_notify_cq = ocrdma_arm_cq, .resize_cq = ocrdma_resize_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, ocrdma_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, ocrdma_ucontext, ibucontext), }; static const struct ib_device_ops ocrdma_dev_srq_ops = { @@ -243,7 +247,7 @@ static int ocrdma_register_device(struct ocrdma_dev *dev) } rdma_set_device_sysfs_group(&dev->ibdev, &ocrdma_attr_group); dev->ibdev.driver_id = RDMA_DRIVER_OCRDMA; - return ib_register_device(&dev->ibdev, "ocrdma%d", NULL); + return ib_register_device(&dev->ibdev, "ocrdma%d"); } static int ocrdma_alloc_resources(struct ocrdma_dev *dev) @@ -295,7 +299,7 @@ static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info) u8 lstate = 0; struct ocrdma_dev *dev; - dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev)); + dev = ib_alloc_device(ocrdma_dev, ibdev); if (!dev) { pr_err("Unable to allocate ib device\n"); return NULL; diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_stats.c b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c index 6be0ea109138..a902942adb5d 100644 --- a/drivers/infiniband/hw/ocrdma/ocrdma_stats.c +++ b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c @@ -767,88 +767,65 @@ void ocrdma_add_port_stats(struct ocrdma_dev *dev) /* Create post stats base dir */ dev->dir = debugfs_create_dir(pci_name(pdev), ocrdma_dbgfs_dir); - if (!dev->dir) - goto err; dev->rsrc_stats.type = OCRDMA_RSRC_STATS; dev->rsrc_stats.dev = dev; - if (!debugfs_create_file("resource_stats", S_IRUSR, dev->dir, - &dev->rsrc_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("resource_stats", S_IRUSR, dev->dir, + &dev->rsrc_stats, &ocrdma_dbg_ops); dev->rx_stats.type = OCRDMA_RXSTATS; dev->rx_stats.dev = dev; - if (!debugfs_create_file("rx_stats", S_IRUSR, dev->dir, - &dev->rx_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("rx_stats", S_IRUSR, dev->dir, &dev->rx_stats, + &ocrdma_dbg_ops); dev->wqe_stats.type = OCRDMA_WQESTATS; dev->wqe_stats.dev = dev; - if (!debugfs_create_file("wqe_stats", S_IRUSR, dev->dir, - &dev->wqe_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("wqe_stats", S_IRUSR, dev->dir, &dev->wqe_stats, + &ocrdma_dbg_ops); dev->tx_stats.type = OCRDMA_TXSTATS; dev->tx_stats.dev = dev; - if (!debugfs_create_file("tx_stats", S_IRUSR, dev->dir, - &dev->tx_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("tx_stats", S_IRUSR, dev->dir, &dev->tx_stats, + &ocrdma_dbg_ops); dev->db_err_stats.type = OCRDMA_DB_ERRSTATS; dev->db_err_stats.dev = dev; - if (!debugfs_create_file("db_err_stats", S_IRUSR, dev->dir, - &dev->db_err_stats, &ocrdma_dbg_ops)) - goto err; - + debugfs_create_file("db_err_stats", S_IRUSR, dev->dir, + &dev->db_err_stats, &ocrdma_dbg_ops); dev->tx_qp_err_stats.type = OCRDMA_TXQP_ERRSTATS; dev->tx_qp_err_stats.dev = dev; - if (!debugfs_create_file("tx_qp_err_stats", S_IRUSR, dev->dir, - &dev->tx_qp_err_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("tx_qp_err_stats", S_IRUSR, dev->dir, + &dev->tx_qp_err_stats, &ocrdma_dbg_ops); dev->rx_qp_err_stats.type = OCRDMA_RXQP_ERRSTATS; dev->rx_qp_err_stats.dev = dev; - if (!debugfs_create_file("rx_qp_err_stats", S_IRUSR, dev->dir, - &dev->rx_qp_err_stats, &ocrdma_dbg_ops)) - goto err; - + debugfs_create_file("rx_qp_err_stats", S_IRUSR, dev->dir, + &dev->rx_qp_err_stats, &ocrdma_dbg_ops); dev->tx_dbg_stats.type = OCRDMA_TX_DBG_STATS; dev->tx_dbg_stats.dev = dev; - if (!debugfs_create_file("tx_dbg_stats", S_IRUSR, dev->dir, - &dev->tx_dbg_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("tx_dbg_stats", S_IRUSR, dev->dir, + &dev->tx_dbg_stats, &ocrdma_dbg_ops); dev->rx_dbg_stats.type = OCRDMA_RX_DBG_STATS; dev->rx_dbg_stats.dev = dev; - if (!debugfs_create_file("rx_dbg_stats", S_IRUSR, dev->dir, - &dev->rx_dbg_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("rx_dbg_stats", S_IRUSR, dev->dir, + &dev->rx_dbg_stats, &ocrdma_dbg_ops); dev->driver_stats.type = OCRDMA_DRV_STATS; dev->driver_stats.dev = dev; - if (!debugfs_create_file("driver_dbg_stats", S_IRUSR, dev->dir, - &dev->driver_stats, &ocrdma_dbg_ops)) - goto err; + debugfs_create_file("driver_dbg_stats", S_IRUSR, dev->dir, + &dev->driver_stats, &ocrdma_dbg_ops); dev->reset_stats.type = OCRDMA_RESET_STATS; dev->reset_stats.dev = dev; - if (!debugfs_create_file("reset_stats", 0200, dev->dir, - &dev->reset_stats, &ocrdma_dbg_ops)) - goto err; - - - return; -err: - debugfs_remove_recursive(dev->dir); - dev->dir = NULL; + debugfs_create_file("reset_stats", 0200, dev->dir, &dev->reset_stats, + &ocrdma_dbg_ops); } void ocrdma_rem_port_stats(struct ocrdma_dev *dev) { - if (!dev->dir) - return; debugfs_remove_recursive(dev->dir); } diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c index 287c332ff0e6..b4e1777c2c97 100644 --- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c +++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c @@ -55,7 +55,7 @@ int ocrdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) { - if (index > 1) + if (index > 0) return -EINVAL; *pkey = 0xffff; @@ -367,17 +367,12 @@ static int ocrdma_get_pd_num(struct ocrdma_dev *dev, struct ocrdma_pd *pd) return status; } -static struct ocrdma_pd *_ocrdma_alloc_pd(struct ocrdma_dev *dev, - struct ocrdma_ucontext *uctx, - struct ib_udata *udata) +static int _ocrdma_alloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd, + struct ocrdma_ucontext *uctx, + struct ib_udata *udata) { - struct ocrdma_pd *pd = NULL; int status; - pd = kzalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - if (udata && uctx && dev->attr.max_dpp_pds) { pd->dpp_enabled = ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R; @@ -386,15 +381,8 @@ static struct ocrdma_pd *_ocrdma_alloc_pd(struct ocrdma_dev *dev, dev->attr.wqe_size) : 0; } - if (dev->pd_mgr->pd_prealloc_valid) { - status = ocrdma_get_pd_num(dev, pd); - if (status == 0) { - return pd; - } else { - kfree(pd); - return ERR_PTR(status); - } - } + if (dev->pd_mgr->pd_prealloc_valid) + return ocrdma_get_pd_num(dev, pd); retry: status = ocrdma_mbx_alloc_pd(dev, pd); @@ -403,13 +391,11 @@ retry: pd->dpp_enabled = false; pd->num_dpp_qp = 0; goto retry; - } else { - kfree(pd); - return ERR_PTR(status); } + return status; } - return pd; + return 0; } static inline int is_ucontext_pd(struct ocrdma_ucontext *uctx, @@ -418,30 +404,33 @@ static inline int is_ucontext_pd(struct ocrdma_ucontext *uctx, return (uctx->cntxt_pd == pd); } -static int _ocrdma_dealloc_pd(struct ocrdma_dev *dev, +static void _ocrdma_dealloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd) { - int status; - if (dev->pd_mgr->pd_prealloc_valid) - status = ocrdma_put_pd_num(dev, pd->id, pd->dpp_enabled); + ocrdma_put_pd_num(dev, pd->id, pd->dpp_enabled); else - status = ocrdma_mbx_dealloc_pd(dev, pd); - - kfree(pd); - return status; + ocrdma_mbx_dealloc_pd(dev, pd); } static int ocrdma_alloc_ucontext_pd(struct ocrdma_dev *dev, struct ocrdma_ucontext *uctx, struct ib_udata *udata) { - int status = 0; + struct ib_device *ibdev = &dev->ibdev; + struct ib_pd *pd; + int status; + + pd = rdma_zalloc_drv_obj(ibdev, ib_pd); + if (!pd) + return -ENOMEM; - uctx->cntxt_pd = _ocrdma_alloc_pd(dev, uctx, udata); - if (IS_ERR(uctx->cntxt_pd)) { - status = PTR_ERR(uctx->cntxt_pd); - uctx->cntxt_pd = NULL; + pd->device = ibdev; + uctx->cntxt_pd = get_ocrdma_pd(pd); + + status = _ocrdma_alloc_pd(dev, uctx->cntxt_pd, uctx, udata); + if (status) { + kfree(uctx->cntxt_pd); goto err; } @@ -451,7 +440,7 @@ err: return status; } -static int ocrdma_dealloc_ucontext_pd(struct ocrdma_ucontext *uctx) +static void ocrdma_dealloc_ucontext_pd(struct ocrdma_ucontext *uctx) { struct ocrdma_pd *pd = uctx->cntxt_pd; struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device); @@ -460,9 +449,9 @@ static int ocrdma_dealloc_ucontext_pd(struct ocrdma_ucontext *uctx) pr_err("%s(%d) Freeing in use pdid=0x%x.\n", __func__, dev->id, pd->id); } + kfree(uctx->cntxt_pd); uctx->cntxt_pd = NULL; - (void)_ocrdma_dealloc_pd(dev, pd); - return 0; + _ocrdma_dealloc_pd(dev, pd); } static struct ocrdma_pd *ocrdma_get_ucontext_pd(struct ocrdma_ucontext *uctx) @@ -486,33 +475,28 @@ static void ocrdma_release_ucontext_pd(struct ocrdma_ucontext *uctx) mutex_unlock(&uctx->mm_list_lock); } -struct ib_ucontext *ocrdma_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +int ocrdma_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { + struct ib_device *ibdev = uctx->device; int status; - struct ocrdma_ucontext *ctx; - struct ocrdma_alloc_ucontext_resp resp; + struct ocrdma_ucontext *ctx = get_ocrdma_ucontext(uctx); + struct ocrdma_alloc_ucontext_resp resp = {}; struct ocrdma_dev *dev = get_ocrdma_dev(ibdev); struct pci_dev *pdev = dev->nic_info.pdev; u32 map_len = roundup(sizeof(u32) * 2048, PAGE_SIZE); if (!udata) - return ERR_PTR(-EFAULT); - ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); - if (!ctx) - return ERR_PTR(-ENOMEM); + return -EFAULT; INIT_LIST_HEAD(&ctx->mm_head); mutex_init(&ctx->mm_list_lock); ctx->ah_tbl.va = dma_alloc_coherent(&pdev->dev, map_len, &ctx->ah_tbl.pa, GFP_KERNEL); - if (!ctx->ah_tbl.va) { - kfree(ctx); - return ERR_PTR(-ENOMEM); - } + if (!ctx->ah_tbl.va) + return -ENOMEM; + ctx->ah_tbl.len = map_len; - memset(&resp, 0, sizeof(resp)); resp.ah_tbl_len = ctx->ah_tbl.len; resp.ah_tbl_page = virt_to_phys(ctx->ah_tbl.va); @@ -534,27 +518,26 @@ struct ib_ucontext *ocrdma_alloc_ucontext(struct ib_device *ibdev, status = ib_copy_to_udata(udata, &resp, sizeof(resp)); if (status) goto cpy_err; - return &ctx->ibucontext; + return 0; cpy_err: + ocrdma_dealloc_ucontext_pd(ctx); pd_err: ocrdma_del_mmap(ctx, ctx->ah_tbl.pa, ctx->ah_tbl.len); map_err: dma_free_coherent(&pdev->dev, ctx->ah_tbl.len, ctx->ah_tbl.va, ctx->ah_tbl.pa); - kfree(ctx); - return ERR_PTR(status); + return status; } -int ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx) +void ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx) { - int status; struct ocrdma_mm *mm, *tmp; struct ocrdma_ucontext *uctx = get_ocrdma_ucontext(ibctx); struct ocrdma_dev *dev = get_ocrdma_dev(ibctx->device); struct pci_dev *pdev = dev->nic_info.pdev; - status = ocrdma_dealloc_ucontext_pd(uctx); + ocrdma_dealloc_ucontext_pd(uctx); ocrdma_del_mmap(uctx, uctx->ah_tbl.pa, uctx->ah_tbl.len); dma_free_coherent(&pdev->dev, uctx->ah_tbl.len, uctx->ah_tbl.va, @@ -564,8 +547,6 @@ int ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx) list_del(&mm->entry); kfree(mm); } - kfree(uctx); - return status; } int ocrdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) @@ -658,10 +639,10 @@ dpp_map_err: return status; } -struct ib_pd *ocrdma_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +int ocrdma_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { + struct ib_device *ibdev = ibpd->device; struct ocrdma_dev *dev = get_ocrdma_dev(ibdev); struct ocrdma_pd *pd; struct ocrdma_ucontext *uctx = NULL; @@ -677,11 +658,10 @@ struct ib_pd *ocrdma_alloc_pd(struct ib_device *ibdev, } } - pd = _ocrdma_alloc_pd(dev, uctx, udata); - if (IS_ERR(pd)) { - status = PTR_ERR(pd); + pd = get_ocrdma_pd(ibpd); + status = _ocrdma_alloc_pd(dev, pd, uctx, udata); + if (status) goto exit; - } pd_mapping: if (udata && context) { @@ -689,25 +669,22 @@ pd_mapping: if (status) goto err; } - return &pd->ibpd; + return 0; err: - if (is_uctx_pd) { + if (is_uctx_pd) ocrdma_release_ucontext_pd(uctx); - } else { - if (_ocrdma_dealloc_pd(dev, pd)) - pr_err("%s: _ocrdma_dealloc_pd() failed\n", __func__); - } + else + _ocrdma_dealloc_pd(dev, pd); exit: - return ERR_PTR(status); + return status; } -int ocrdma_dealloc_pd(struct ib_pd *ibpd) +void ocrdma_dealloc_pd(struct ib_pd *ibpd) { struct ocrdma_pd *pd = get_ocrdma_pd(ibpd); struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device); struct ocrdma_ucontext *uctx = NULL; - int status = 0; u64 usr_db; uctx = pd->uctx; @@ -721,11 +698,10 @@ int ocrdma_dealloc_pd(struct ib_pd *ibpd) if (is_ucontext_pd(uctx, pd)) { ocrdma_release_ucontext_pd(uctx); - return status; + return; } } - status = _ocrdma_dealloc_pd(dev, pd); - return status; + _ocrdma_dealloc_pd(dev, pd); } static int ocrdma_alloc_lkey(struct ocrdma_dev *dev, struct ocrdma_mr *mr, @@ -854,10 +830,11 @@ static void build_user_pbes(struct ocrdma_dev *dev, struct ocrdma_mr *mr, u32 num_pbes) { struct ocrdma_pbe *pbe; - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; struct ocrdma_pbl *pbl_tbl = mr->hwmr.pbl_table; struct ib_umem *umem = mr->umem; - int shift, pg_cnt, pages, pbe_cnt, entry, total_num_pbes = 0; + int pbe_cnt, total_num_pbes = 0; + u64 pg_addr; if (!mr->hwmr.num_pbes) return; @@ -865,36 +842,26 @@ static void build_user_pbes(struct ocrdma_dev *dev, struct ocrdma_mr *mr, pbe = (struct ocrdma_pbe *)pbl_tbl->va; pbe_cnt = 0; - shift = umem->page_shift; - - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { - pages = sg_dma_len(sg) >> shift; - for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) { - /* store the page address in pbe */ - pbe->pa_lo = - cpu_to_le32(sg_dma_address(sg) + - (pg_cnt << shift)); - pbe->pa_hi = - cpu_to_le32(upper_32_bits(sg_dma_address(sg) + - (pg_cnt << shift))); - pbe_cnt += 1; - total_num_pbes += 1; - pbe++; - - /* if done building pbes, issue the mbx cmd. */ - if (total_num_pbes == num_pbes) - return; - - /* if the given pbl is full storing the pbes, - * move to next pbl. - */ - if (pbe_cnt == - (mr->hwmr.pbl_size / sizeof(u64))) { - pbl_tbl++; - pbe = (struct ocrdma_pbe *)pbl_tbl->va; - pbe_cnt = 0; - } + for_each_sg_dma_page (umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + /* store the page address in pbe */ + pg_addr = sg_page_iter_dma_address(&sg_iter); + pbe->pa_lo = cpu_to_le32(pg_addr); + pbe->pa_hi = cpu_to_le32(upper_32_bits(pg_addr)); + pbe_cnt += 1; + total_num_pbes += 1; + pbe++; + + /* if done building pbes, issue the mbx cmd. */ + if (total_num_pbes == num_pbes) + return; + /* if the given pbl is full storing the pbes, + * move to next pbl. + */ + if (pbe_cnt == (mr->hwmr.pbl_size / sizeof(u64))) { + pbl_tbl++; + pbe = (struct ocrdma_pbe *)pbl_tbl->va; + pbe_cnt = 0; } } } @@ -916,7 +883,7 @@ struct ib_mr *ocrdma_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len, mr = kzalloc(sizeof(*mr), GFP_KERNEL); if (!mr) return ERR_PTR(status); - mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0); + mr->umem = ib_umem_get(udata, start, len, acc, 0); if (IS_ERR(mr->umem)) { status = -EFAULT; goto umem_err; @@ -926,7 +893,7 @@ struct ib_mr *ocrdma_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len, if (status) goto umem_err; - mr->hwmr.pbe_size = BIT(mr->umem->page_shift); + mr->hwmr.pbe_size = PAGE_SIZE; mr->hwmr.fbo = ib_umem_offset(mr->umem); mr->hwmr.va = usr_addr; mr->hwmr.len = len; diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.h b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.h index b69cfdce7970..4c04ab40798e 100644 --- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.h +++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.h @@ -64,15 +64,14 @@ void ocrdma_get_guid(struct ocrdma_dev *, u8 *guid); struct net_device *ocrdma_get_netdev(struct ib_device *device, u8 port_num); int ocrdma_query_pkey(struct ib_device *, u8 port, u16 index, u16 *pkey); -struct ib_ucontext *ocrdma_alloc_ucontext(struct ib_device *, - struct ib_udata *); -int ocrdma_dealloc_ucontext(struct ib_ucontext *); +int ocrdma_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata); +void ocrdma_dealloc_ucontext(struct ib_ucontext *uctx); int ocrdma_mmap(struct ib_ucontext *, struct vm_area_struct *vma); -struct ib_pd *ocrdma_alloc_pd(struct ib_device *, - struct ib_ucontext *, struct ib_udata *); -int ocrdma_dealloc_pd(struct ib_pd *pd); +int ocrdma_alloc_pd(struct ib_pd *pd, struct ib_ucontext *uctx, + struct ib_udata *udata); +void ocrdma_dealloc_pd(struct ib_pd *pd); struct ib_cq *ocrdma_create_cq(struct ib_device *ibdev, const struct ib_cq_init_attr *attr, diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c index 75940e2a8791..996d9ecd93e0 100644 --- a/drivers/infiniband/hw/qedr/main.c +++ b/drivers/infiniband/hw/qedr/main.c @@ -137,7 +137,8 @@ static int qedr_iw_port_immutable(struct ib_device *ibdev, u8 port_num, static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { - struct qedr_dev *dev = dev_get_drvdata(device); + struct qedr_dev *dev = + rdma_device_to_drv_device(device, struct qedr_dev, ibdev); return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor); } @@ -238,6 +239,8 @@ static const struct ib_device_ops qedr_dev_ops = { .reg_user_mr = qedr_reg_user_mr, .req_notify_cq = qedr_arm_cq, .resize_cq = qedr_resize_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, qedr_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, qedr_ucontext, ibucontext), }; static int qedr_register_device(struct qedr_dev *dev) @@ -290,7 +293,7 @@ static int qedr_register_device(struct qedr_dev *dev) ib_set_device_ops(&dev->ibdev, &qedr_dev_ops); dev->ibdev.driver_id = RDMA_DRIVER_QEDR; - return ib_register_device(&dev->ibdev, "qedr%d", NULL); + return ib_register_device(&dev->ibdev, "qedr%d"); } /* This function allocates fast-path status block memory */ @@ -852,7 +855,7 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev, struct qedr_dev *dev; int rc = 0; - dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev)); + dev = ib_alloc_device(qedr_dev, ibdev); if (!dev) { pr_err("Unable to allocate ib device\n"); return NULL; diff --git a/drivers/infiniband/hw/qedr/qedr_iw_cm.c b/drivers/infiniband/hw/qedr/qedr_iw_cm.c index 93b16237b767..0555e5a8c9ed 100644 --- a/drivers/infiniband/hw/qedr/qedr_iw_cm.c +++ b/drivers/infiniband/hw/qedr/qedr_iw_cm.c @@ -349,7 +349,7 @@ qedr_iw_event_handler(void *context, struct qed_iwarp_cm_event_params *params) default: DP_NOTICE(dev, "Unknown event received %d\n", params->event); break; - }; + } return 0; } diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c index e1ccf32b1c3d..59ad4202422c 100644 --- a/drivers/infiniband/hw/qedr/verbs.c +++ b/drivers/infiniband/hw/qedr/verbs.c @@ -67,7 +67,7 @@ static inline int qedr_ib_copy_to_udata(struct ib_udata *udata, void *src, int qedr_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) { - if (index > QEDR_ROCE_PKEY_TABLE_LEN) + if (index >= QEDR_ROCE_PKEY_TABLE_LEN) return -EINVAL; *pkey = QEDR_ROCE_PKEY_DEFAULT; @@ -316,28 +316,24 @@ static bool qedr_search_mmap(struct qedr_ucontext *uctx, u64 phy_addr, return found; } -struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +int qedr_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { + struct ib_device *ibdev = uctx->device; int rc; - struct qedr_ucontext *ctx; - struct qedr_alloc_ucontext_resp uresp; + struct qedr_ucontext *ctx = get_qedr_ucontext(uctx); + struct qedr_alloc_ucontext_resp uresp = {}; struct qedr_dev *dev = get_qedr_dev(ibdev); struct qed_rdma_add_user_out_params oparams; if (!udata) - return ERR_PTR(-EFAULT); - - ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); - if (!ctx) - return ERR_PTR(-ENOMEM); + return -EFAULT; rc = dev->ops->rdma_add_user(dev->rdma_ctx, &oparams); if (rc) { DP_ERR(dev, "failed to allocate a DPI for a new RoCE application, rc=%d. To overcome this consider to increase the number of DPIs, increase the doorbell BAR size or just close unnecessary RoCE applications. In order to increase the number of DPIs consult the qedr readme\n", rc); - goto err; + return rc; } ctx->dpi = oparams.dpi; @@ -347,8 +343,6 @@ struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *ibdev, INIT_LIST_HEAD(&ctx->mm_head); mutex_init(&ctx->mm_list_lock); - memset(&uresp, 0, sizeof(uresp)); - uresp.dpm_enabled = dev->user_dpm_enabled; uresp.wids_enabled = 1; uresp.wid_count = oparams.wid_count; @@ -364,28 +358,23 @@ struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *ibdev, rc = qedr_ib_copy_to_udata(udata, &uresp, sizeof(uresp)); if (rc) - goto err; + return rc; ctx->dev = dev; rc = qedr_add_mmap(ctx, ctx->dpi_phys_addr, ctx->dpi_size); if (rc) - goto err; + return rc; DP_DEBUG(dev, QEDR_MSG_INIT, "Allocating user context %p\n", &ctx->ibucontext); - return &ctx->ibucontext; - -err: - kfree(ctx); - return ERR_PTR(rc); + return 0; } -int qedr_dealloc_ucontext(struct ib_ucontext *ibctx) +void qedr_dealloc_ucontext(struct ib_ucontext *ibctx) { struct qedr_ucontext *uctx = get_qedr_ucontext(ibctx); struct qedr_mm *mm, *tmp; - int status = 0; DP_DEBUG(uctx->dev, QEDR_MSG_INIT, "Deallocating user context %p\n", uctx); @@ -398,9 +387,6 @@ int qedr_dealloc_ucontext(struct ib_ucontext *ibctx) list_del(&mm->entry); kfree(mm); } - - kfree(uctx); - return status; } int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) @@ -450,11 +436,12 @@ int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) vma->vm_page_prot); } -struct ib_pd *qedr_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, struct ib_udata *udata) +int qedr_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { + struct ib_device *ibdev = ibpd->device; struct qedr_dev *dev = get_qedr_dev(ibdev); - struct qedr_pd *pd; + struct qedr_pd *pd = get_qedr_pd(ibpd); u16 pd_id; int rc; @@ -463,16 +450,12 @@ struct ib_pd *qedr_alloc_pd(struct ib_device *ibdev, if (!dev->rdma_ctx) { DP_ERR(dev, "invalid RDMA context\n"); - return ERR_PTR(-EINVAL); + return -EINVAL; } - pd = kzalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - rc = dev->ops->rdma_alloc_pd(dev->rdma_ctx, &pd_id); if (rc) - goto err; + return rc; pd->pd_id = pd_id; @@ -485,36 +468,23 @@ struct ib_pd *qedr_alloc_pd(struct ib_device *ibdev, if (rc) { DP_ERR(dev, "copy error pd_id=0x%x.\n", pd_id); dev->ops->rdma_dealloc_pd(dev->rdma_ctx, pd_id); - goto err; + return rc; } pd->uctx = get_qedr_ucontext(context); pd->uctx->pd = pd; } - return &pd->ibpd; - -err: - kfree(pd); - return ERR_PTR(rc); + return 0; } -int qedr_dealloc_pd(struct ib_pd *ibpd) +void qedr_dealloc_pd(struct ib_pd *ibpd) { struct qedr_dev *dev = get_qedr_dev(ibpd->device); struct qedr_pd *pd = get_qedr_pd(ibpd); - if (!pd) { - pr_err("Invalid PD received in dealloc_pd\n"); - return -EINVAL; - } - DP_DEBUG(dev, QEDR_MSG_INIT, "Deallocating PD %d\n", pd->pd_id); dev->ops->rdma_dealloc_pd(dev->rdma_ctx, pd->pd_id); - - kfree(pd); - - return 0; } static void qedr_free_pbl(struct qedr_dev *dev, @@ -636,13 +606,12 @@ static void qedr_populate_pbls(struct qedr_dev *dev, struct ib_umem *umem, struct qedr_pbl *pbl, struct qedr_pbl_info *pbl_info, u32 pg_shift) { - int shift, pg_cnt, pages, pbe_cnt, total_num_pbes = 0; + int pbe_cnt, total_num_pbes = 0; u32 fw_pg_cnt, fw_pg_per_umem_pg; struct qedr_pbl *pbl_tbl; - struct scatterlist *sg; + struct sg_dma_page_iter sg_iter; struct regpair *pbe; u64 pg_addr; - int entry; if (!pbl_info->num_pbes) return; @@ -663,38 +632,32 @@ static void qedr_populate_pbls(struct qedr_dev *dev, struct ib_umem *umem, pbe_cnt = 0; - shift = umem->page_shift; - - fw_pg_per_umem_pg = BIT(umem->page_shift - pg_shift); - - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { - pages = sg_dma_len(sg) >> shift; - pg_addr = sg_dma_address(sg); - for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) { - for (fw_pg_cnt = 0; fw_pg_cnt < fw_pg_per_umem_pg;) { - pbe->lo = cpu_to_le32(pg_addr); - pbe->hi = cpu_to_le32(upper_32_bits(pg_addr)); - - pg_addr += BIT(pg_shift); - pbe_cnt++; - total_num_pbes++; - pbe++; - - if (total_num_pbes == pbl_info->num_pbes) - return; - - /* If the given pbl is full storing the pbes, - * move to next pbl. - */ - if (pbe_cnt == - (pbl_info->pbl_size / sizeof(u64))) { - pbl_tbl++; - pbe = (struct regpair *)pbl_tbl->va; - pbe_cnt = 0; - } + fw_pg_per_umem_pg = BIT(PAGE_SHIFT - pg_shift); + + for_each_sg_dma_page (umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + pg_addr = sg_page_iter_dma_address(&sg_iter); + for (fw_pg_cnt = 0; fw_pg_cnt < fw_pg_per_umem_pg;) { + pbe->lo = cpu_to_le32(pg_addr); + pbe->hi = cpu_to_le32(upper_32_bits(pg_addr)); + + pg_addr += BIT(pg_shift); + pbe_cnt++; + total_num_pbes++; + pbe++; + + if (total_num_pbes == pbl_info->num_pbes) + return; - fw_pg_cnt++; + /* If the given pbl is full storing the pbes, + * move to next pbl. + */ + if (pbe_cnt == (pbl_info->pbl_size / sizeof(u64))) { + pbl_tbl++; + pbe = (struct regpair *)pbl_tbl->va; + pbe_cnt = 0; } + + fw_pg_cnt++; } } } @@ -736,11 +699,10 @@ static inline int qedr_align_cq_entries(int entries) return aligned_size / QEDR_CQE_SIZE; } -static inline int qedr_init_user_queue(struct ib_ucontext *ib_ctx, +static inline int qedr_init_user_queue(struct ib_udata *udata, struct qedr_dev *dev, - struct qedr_userq *q, - u64 buf_addr, size_t buf_len, - int access, int dmasync, + struct qedr_userq *q, u64 buf_addr, + size_t buf_len, int access, int dmasync, int alloc_and_init) { u32 fw_pages; @@ -748,7 +710,7 @@ static inline int qedr_init_user_queue(struct ib_ucontext *ib_ctx, q->buf_addr = buf_addr; q->buf_len = buf_len; - q->umem = ib_umem_get(ib_ctx, q->buf_addr, q->buf_len, access, dmasync); + q->umem = ib_umem_get(udata, q->buf_addr, q->buf_len, access, dmasync); if (IS_ERR(q->umem)) { DP_ERR(dev, "create user queue: failed ib_umem_get, got %ld\n", PTR_ERR(q->umem)); @@ -756,7 +718,7 @@ static inline int qedr_init_user_queue(struct ib_ucontext *ib_ctx, } fw_pages = ib_umem_page_count(q->umem) << - (q->umem->page_shift - FW_PAGE_SHIFT); + (PAGE_SHIFT - FW_PAGE_SHIFT); rc = qedr_prepare_pbl_tbl(dev, &q->pbl_info, fw_pages, 0); if (rc) @@ -905,9 +867,9 @@ struct ib_cq *qedr_create_cq(struct ib_device *ibdev, cq->cq_type = QEDR_CQ_TYPE_USER; - rc = qedr_init_user_queue(ib_ctx, dev, &cq->q, ureq.addr, - ureq.len, IB_ACCESS_LOCAL_WRITE, - 1, 1); + rc = qedr_init_user_queue(udata, dev, &cq->q, ureq.addr, + ureq.len, IB_ACCESS_LOCAL_WRITE, 1, + 1); if (rc) goto err0; @@ -1344,7 +1306,7 @@ static void qedr_free_srq_kernel_params(struct qedr_srq *srq) hw_srq->phy_prod_pair_addr); } -static int qedr_init_srq_user_params(struct ib_ucontext *ib_ctx, +static int qedr_init_srq_user_params(struct ib_udata *udata, struct qedr_srq *srq, struct qedr_create_srq_ureq *ureq, int access, int dmasync) @@ -1352,14 +1314,14 @@ static int qedr_init_srq_user_params(struct ib_ucontext *ib_ctx, struct scatterlist *sg; int rc; - rc = qedr_init_user_queue(ib_ctx, srq->dev, &srq->usrq, ureq->srq_addr, + rc = qedr_init_user_queue(udata, srq->dev, &srq->usrq, ureq->srq_addr, ureq->srq_len, access, dmasync, 1); if (rc) return rc; - srq->prod_umem = ib_umem_get(ib_ctx, ureq->prod_pair_addr, - sizeof(struct rdma_srq_producers), - access, dmasync); + srq->prod_umem = + ib_umem_get(udata, ureq->prod_pair_addr, + sizeof(struct rdma_srq_producers), access, dmasync); if (IS_ERR(srq->prod_umem)) { qedr_free_pbl(srq->dev, &srq->usrq.pbl_info, srq->usrq.pbl_tbl); ib_umem_release(srq->usrq.umem); @@ -1434,7 +1396,6 @@ struct ib_srq *qedr_create_srq(struct ib_pd *ibpd, struct qedr_pd *pd = get_qedr_pd(ibpd); struct qedr_create_srq_ureq ureq = {}; u64 pbl_base_addr, phy_prod_pair_addr; - struct ib_ucontext *ib_ctx = NULL; struct qedr_srq_hwq_info *hw_srq; u32 page_cnt, page_size; struct qedr_srq *srq; @@ -1459,23 +1420,21 @@ struct ib_srq *qedr_create_srq(struct ib_pd *ibpd, hw_srq->max_wr = init_attr->attr.max_wr; hw_srq->max_sges = init_attr->attr.max_sge; - if (udata && ibpd->uobject && ibpd->uobject->context) { - ib_ctx = ibpd->uobject->context; - + if (udata) { if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) { DP_ERR(dev, "create srq: problem copying data from user space\n"); goto err0; } - rc = qedr_init_srq_user_params(ib_ctx, srq, &ureq, 0, 0); + rc = qedr_init_srq_user_params(udata, srq, &ureq, 0, 0); if (rc) goto err0; page_cnt = srq->usrq.pbl_info.num_pbes; pbl_base_addr = srq->usrq.pbl_tbl->pa; phy_prod_pair_addr = hw_srq->phy_prod_pair_addr; - page_size = BIT(srq->usrq.umem->page_shift); + page_size = PAGE_SIZE; } else { struct qed_chain *pbl; @@ -1699,13 +1658,10 @@ static int qedr_create_user_qp(struct qedr_dev *dev, struct qed_rdma_create_qp_in_params in_params; struct qed_rdma_create_qp_out_params out_params; struct qedr_pd *pd = get_qedr_pd(ibpd); - struct ib_ucontext *ib_ctx = NULL; struct qedr_create_qp_ureq ureq; int alloc_and_init = rdma_protocol_roce(&dev->ibdev, 1); int rc = -EINVAL; - ib_ctx = ibpd->uobject->context; - memset(&ureq, 0, sizeof(ureq)); rc = ib_copy_from_udata(&ureq, udata, sizeof(ureq)); if (rc) { @@ -1714,14 +1670,14 @@ static int qedr_create_user_qp(struct qedr_dev *dev, } /* SQ - read access only (0), dma sync not required (0) */ - rc = qedr_init_user_queue(ib_ctx, dev, &qp->usq, ureq.sq_addr, + rc = qedr_init_user_queue(udata, dev, &qp->usq, ureq.sq_addr, ureq.sq_len, 0, 0, alloc_and_init); if (rc) return rc; if (!qp->srq) { /* RQ - read access only (0), dma sync not required (0) */ - rc = qedr_init_user_queue(ib_ctx, dev, &qp->urq, ureq.rq_addr, + rc = qedr_init_user_queue(udata, dev, &qp->urq, ureq.rq_addr, ureq.rq_len, 0, 0, alloc_and_init); if (rc) return rc; @@ -2117,7 +2073,7 @@ static int qedr_update_qp_state(struct qedr_dev *dev, default: status = -EINVAL; break; - }; + } break; case QED_ROCE_QP_STATE_INIT: switch (new_state) { @@ -2138,7 +2094,7 @@ static int qedr_update_qp_state(struct qedr_dev *dev, /* Invalid state change. */ status = -EINVAL; break; - }; + } break; case QED_ROCE_QP_STATE_RTR: /* RTR->XXX */ @@ -2151,7 +2107,7 @@ static int qedr_update_qp_state(struct qedr_dev *dev, /* Invalid state change. */ status = -EINVAL; break; - }; + } break; case QED_ROCE_QP_STATE_RTS: /* RTS->XXX */ @@ -2164,7 +2120,7 @@ static int qedr_update_qp_state(struct qedr_dev *dev, /* Invalid state change. */ status = -EINVAL; break; - }; + } break; case QED_ROCE_QP_STATE_SQD: /* SQD->XXX */ @@ -2176,7 +2132,7 @@ static int qedr_update_qp_state(struct qedr_dev *dev, /* Invalid state change. */ status = -EINVAL; break; - }; + } break; case QED_ROCE_QP_STATE_ERR: /* ERR->XXX */ @@ -2194,12 +2150,12 @@ static int qedr_update_qp_state(struct qedr_dev *dev, default: status = -EINVAL; break; - }; + } break; default: status = -EINVAL; break; - }; + } return status; } @@ -2719,7 +2675,7 @@ struct ib_mr *qedr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len, mr->type = QEDR_MR_USER; - mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0); + mr->umem = ib_umem_get(udata, start, len, acc, 0); if (IS_ERR(mr->umem)) { rc = -EFAULT; goto err0; @@ -2730,7 +2686,7 @@ struct ib_mr *qedr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len, goto err1; qedr_populate_pbls(dev, mr->umem, mr->info.pbl_table, - &mr->info.pbl_info, mr->umem->page_shift); + &mr->info.pbl_info, PAGE_SHIFT); rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid); if (rc) { @@ -2751,7 +2707,7 @@ struct ib_mr *qedr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len, mr->hw_mr.pbl_ptr = mr->info.pbl_table[0].pa; mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered; mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size); - mr->hw_mr.page_size_log = mr->umem->page_shift; + mr->hw_mr.page_size_log = PAGE_SHIFT; mr->hw_mr.fbo = ib_umem_offset(mr->umem); mr->hw_mr.length = len; mr->hw_mr.vaddr = usr_addr; diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h index 1852b7012bf4..f0c05f4771ac 100644 --- a/drivers/infiniband/hw/qedr/verbs.h +++ b/drivers/infiniband/hw/qedr/verbs.h @@ -43,13 +43,13 @@ int qedr_iw_query_gid(struct ib_device *ibdev, u8 port, int qedr_query_pkey(struct ib_device *, u8 port, u16 index, u16 *pkey); -struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *, struct ib_udata *); -int qedr_dealloc_ucontext(struct ib_ucontext *); +int qedr_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata); +void qedr_dealloc_ucontext(struct ib_ucontext *uctx); int qedr_mmap(struct ib_ucontext *, struct vm_area_struct *vma); -struct ib_pd *qedr_alloc_pd(struct ib_device *, - struct ib_ucontext *, struct ib_udata *); -int qedr_dealloc_pd(struct ib_pd *pd); +int qedr_alloc_pd(struct ib_pd *pd, struct ib_ucontext *uctx, + struct ib_udata *udata); +void qedr_dealloc_pd(struct ib_pd *pd); struct ib_cq *qedr_create_cq(struct ib_device *ibdev, const struct ib_cq_init_attr *attr, diff --git a/drivers/infiniband/hw/qib/qib_debugfs.c b/drivers/infiniband/hw/qib/qib_debugfs.c index 5ed1ed93380f..caeb77d07a58 100644 --- a/drivers/infiniband/hw/qib/qib_debugfs.c +++ b/drivers/infiniband/hw/qib/qib_debugfs.c @@ -66,15 +66,6 @@ static const struct file_operations _##name##_file_ops = { \ .release = seq_release \ }; -#define DEBUGFS_FILE_CREATE(name) \ -do { \ - struct dentry *ent; \ - ent = debugfs_create_file(#name , 0400, ibd->qib_ibdev_dbg, \ - ibd, &_##name##_file_ops); \ - if (!ent) \ - pr_warn("create of " #name " failed\n"); \ -} while (0) - static void *_opcode_stats_seq_start(struct seq_file *s, loff_t *pos) { struct qib_opcode_stats_perctx *opstats; @@ -249,17 +240,17 @@ DEBUGFS_FILE(qp_stats) void qib_dbg_ibdev_init(struct qib_ibdev *ibd) { + struct dentry *root; char name[10]; snprintf(name, sizeof(name), "qib%d", dd_from_dev(ibd)->unit); - ibd->qib_ibdev_dbg = debugfs_create_dir(name, qib_dbg_root); - if (!ibd->qib_ibdev_dbg) { - pr_warn("create of %s failed\n", name); - return; - } - DEBUGFS_FILE_CREATE(opcode_stats); - DEBUGFS_FILE_CREATE(ctx_stats); - DEBUGFS_FILE_CREATE(qp_stats); + root = debugfs_create_dir(name, qib_dbg_root); + ibd->qib_ibdev_dbg = root; + + debugfs_create_file("opcode_stats", 0400, root, ibd, + &_opcode_stats_file_ops); + debugfs_create_file("ctx_stats", 0400, root, ibd, &_ctx_stats_file_ops); + debugfs_create_file("qp_stats", 0400, root, ibd, &_qp_stats_file_ops); } void qib_dbg_ibdev_exit(struct qib_ibdev *ibd) @@ -274,8 +265,6 @@ out: void qib_dbg_init(void) { qib_dbg_root = debugfs_create_dir(QIB_DRV_NAME, NULL); - if (!qib_dbg_root) - pr_warn("init of debugfs failed\n"); } void qib_dbg_exit(void) diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c index 6fa002940451..50dd9811b088 100644 --- a/drivers/infiniband/hw/qib/qib_rc.c +++ b/drivers/infiniband/hw/qib/qib_rc.c @@ -45,12 +45,7 @@ static u32 restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe, u32 len; len = ((psn - wqe->psn) & QIB_PSN_MASK) * pmtu; - ss->sge = wqe->sg_list[0]; - ss->sg_list = wqe->sg_list + 1; - ss->num_sge = wqe->wr.num_sge; - ss->total_len = wqe->length; - rvt_skip_sge(ss, len, false); - return wqe->length - len; + return rvt_restart_sge(ss, wqe, len); } /** diff --git a/drivers/infiniband/hw/qib/qib_sdma.c b/drivers/infiniband/hw/qib/qib_sdma.c index 3d64081c4819..99e11c347130 100644 --- a/drivers/infiniband/hw/qib/qib_sdma.c +++ b/drivers/infiniband/hw/qib/qib_sdma.c @@ -565,13 +565,8 @@ retry: sge = &ss->sge; while (dwords) { u32 dw; - u32 len; + u32 len = rvt_get_sge_length(sge, dwords << 2); - len = dwords << 2; - if (len > sge->length) - len = sge->length; - if (len > sge->sge_length) - len = sge->sge_length; dw = (len + 3) >> 2; addr = dma_map_single(&ppd->dd->pcidev->dev, sge->vaddr, dw << 2, DMA_TO_DEVICE); @@ -594,24 +589,7 @@ retry: descqp = &ppd->sdma_descq[0].qw[0]; ++ppd->sdma_generation; } - sge->vaddr += len; - sge->length -= len; - sge->sge_length -= len; - if (sge->sge_length == 0) { - if (--ss->num_sge) - *sge = *ss->sg_list++; - } else if (sge->length == 0 && sge->mr->lkey) { - if (++sge->n >= RVT_SEGSZ) { - if (++sge->m >= sge->mr->mapsz) - break; - sge->n = 0; - } - sge->vaddr = - sge->mr->map[sge->m]->segs[sge->n].vaddr; - sge->length = - sge->mr->map[sge->m]->segs[sge->n].length; - } - + rvt_update_sge(ss, len, false); dwoffset += dw; dwords -= dw; } diff --git a/drivers/infiniband/hw/qib/qib_sysfs.c b/drivers/infiniband/hw/qib/qib_sysfs.c index 1cf4ca3f23e3..905206a0c2d5 100644 --- a/drivers/infiniband/hw/qib/qib_sysfs.c +++ b/drivers/infiniband/hw/qib/qib_sysfs.c @@ -555,7 +555,7 @@ static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); return sprintf(buf, "%x\n", dd_from_dev(dev)->minrev); } @@ -565,7 +565,7 @@ static ssize_t hca_type_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); int ret; @@ -590,7 +590,7 @@ static ssize_t boardversion_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); /* The string printed here is already newline-terminated. */ @@ -602,7 +602,7 @@ static ssize_t localbus_info_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); /* The string printed here is already newline-terminated. */ @@ -614,7 +614,7 @@ static ssize_t nctxts_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); /* Return the number of user ports (contexts) available. */ @@ -630,7 +630,7 @@ static ssize_t nfreectxts_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); /* Return the number of free user ports (contexts) available. */ @@ -642,7 +642,7 @@ static ssize_t serial_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); buf[sizeof(dd->serial)] = '\0'; @@ -657,7 +657,7 @@ static ssize_t chip_reset_store(struct device *device, size_t count) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); int ret; @@ -679,7 +679,7 @@ static ssize_t tempsense_show(struct device *device, struct device_attribute *attr, char *buf) { struct qib_ibdev *dev = - container_of(device, struct qib_ibdev, rdi.ibdev.dev); + rdma_device_to_drv_device(device, struct qib_ibdev, rdi.ibdev); struct qib_devdata *dd = dd_from_dev(dev); int ret; int idx; diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c index 445ea19a2ec8..5cdedba2d164 100644 --- a/drivers/infiniband/hw/qib/qib_ud.c +++ b/drivers/infiniband/hw/qib/qib_ud.c @@ -172,12 +172,8 @@ static void qib_ud_loopback(struct rvt_qp *sqp, struct rvt_swqe *swqe) ssge.num_sge = swqe->wr.num_sge; sge = &ssge.sge; while (length) { - u32 len = sge->length; + u32 len = rvt_get_sge_length(sge, length); - if (len > length) - len = length; - if (len > sge->sge_length) - len = sge->sge_length; rvt_copy_sge(qp, &qp->r_sge, sge->vaddr, len, true, false); sge->vaddr += len; sge->length -= len; diff --git a/drivers/infiniband/hw/qib/qib_user_pages.c b/drivers/infiniband/hw/qib/qib_user_pages.c index 16543d5e80c3..123ca8f64f75 100644 --- a/drivers/infiniband/hw/qib/qib_user_pages.c +++ b/drivers/infiniband/hw/qib/qib_user_pages.c @@ -49,43 +49,6 @@ static void __qib_release_user_pages(struct page **p, size_t num_pages, } } -/* - * Call with current->mm->mmap_sem held. - */ -static int __qib_get_user_pages(unsigned long start_page, size_t num_pages, - struct page **p) -{ - unsigned long lock_limit; - size_t got; - int ret; - - lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; - - if (num_pages > lock_limit && !capable(CAP_IPC_LOCK)) { - ret = -ENOMEM; - goto bail; - } - - for (got = 0; got < num_pages; got += ret) { - ret = get_user_pages(start_page + got * PAGE_SIZE, - num_pages - got, - FOLL_WRITE | FOLL_FORCE, - p + got, NULL); - if (ret < 0) - goto bail_release; - } - - current->mm->pinned_vm += num_pages; - - ret = 0; - goto bail; - -bail_release: - __qib_release_user_pages(p, got, 0); -bail: - return ret; -} - /** * qib_map_page - a safety wrapper around pci_map_page() * @@ -137,26 +100,44 @@ int qib_map_page(struct pci_dev *hwdev, struct page *page, dma_addr_t *daddr) int qib_get_user_pages(unsigned long start_page, size_t num_pages, struct page **p) { + unsigned long locked, lock_limit; + size_t got; int ret; - down_write(¤t->mm->mmap_sem); + lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; + locked = atomic64_add_return(num_pages, ¤t->mm->pinned_vm); - ret = __qib_get_user_pages(start_page, num_pages, p); + if (locked > lock_limit && !capable(CAP_IPC_LOCK)) { + ret = -ENOMEM; + goto bail; + } - up_write(¤t->mm->mmap_sem); + down_read(¤t->mm->mmap_sem); + for (got = 0; got < num_pages; got += ret) { + ret = get_user_pages_longterm(start_page + got * PAGE_SIZE, + num_pages - got, + FOLL_WRITE | FOLL_FORCE, + p + got, NULL); + if (ret < 0) { + up_read(¤t->mm->mmap_sem); + goto bail_release; + } + } + up_read(¤t->mm->mmap_sem); + return 0; +bail_release: + __qib_release_user_pages(p, got, 0); +bail: + atomic64_sub(num_pages, ¤t->mm->pinned_vm); return ret; } void qib_release_user_pages(struct page **p, size_t num_pages) { - if (current->mm) /* during close after signal, mm can be NULL */ - down_write(¤t->mm->mmap_sem); - __qib_release_user_pages(p, num_pages, 1); - if (current->mm) { - current->mm->pinned_vm -= num_pages; - up_write(¤t->mm->mmap_sem); - } + /* during close after signal, mm can be NULL */ + if (current->mm) + atomic64_sub(num_pages, ¤t->mm->pinned_vm); } diff --git a/drivers/infiniband/hw/qib/qib_verbs.c b/drivers/infiniband/hw/qib/qib_verbs.c index 276304f611ab..5ff32d32c61c 100644 --- a/drivers/infiniband/hw/qib/qib_verbs.c +++ b/drivers/infiniband/hw/qib/qib_verbs.c @@ -144,12 +144,8 @@ static u32 qib_count_sge(struct rvt_sge_state *ss, u32 length) u32 ndesc = 1; /* count the header */ while (length) { - u32 len = sge.length; + u32 len = rvt_get_sge_length(&sge, length); - if (len > length) - len = length; - if (len > sge.sge_length) - len = sge.sge_length; if (((long) sge.vaddr & (sizeof(u32) - 1)) || (len != length && (len & (sizeof(u32) - 1)))) { ndesc = 0; @@ -186,12 +182,8 @@ static void qib_copy_from_sge(void *data, struct rvt_sge_state *ss, u32 length) struct rvt_sge *sge = &ss->sge; while (length) { - u32 len = sge->length; + u32 len = rvt_get_sge_length(sge, length); - if (len > length) - len = length; - if (len > sge->sge_length) - len = sge->sge_length; memcpy(data, sge->vaddr, len); sge->vaddr += len; sge->length -= len; @@ -440,13 +432,9 @@ static void copy_io(u32 __iomem *piobuf, struct rvt_sge_state *ss, u32 last; while (1) { - u32 len = ss->sge.length; + u32 len = rvt_get_sge_length(&ss->sge, length); u32 off; - if (len > length) - len = length; - if (len > ss->sge.sge_length) - len = ss->sge.sge_length; /* If the source address is not aligned, try to align it. */ off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1); if (off) { @@ -1494,6 +1482,7 @@ static void qib_fill_device_attr(struct qib_devdata *dd) } static const struct ib_device_ops qib_dev_ops = { + .init_port = qib_create_port_files, .modify_device = qib_modify_device, .process_mad = qib_process_mad, }; @@ -1567,7 +1556,6 @@ int qib_register_ib_device(struct qib_devdata *dd) /* * Fill in rvt info object. */ - dd->verbs_dev.rdi.driver_f.port_callback = qib_create_port_files; dd->verbs_dev.rdi.driver_f.get_pci_dev = qib_get_pci_dev; dd->verbs_dev.rdi.driver_f.check_ah = qib_check_ah; dd->verbs_dev.rdi.driver_f.setup_wqe = qib_check_send_wqe; diff --git a/drivers/infiniband/hw/usnic/Makefile b/drivers/infiniband/hw/usnic/Makefile index 94ae7a1a6950..f12a4938ffd2 100644 --- a/drivers/infiniband/hw/usnic/Makefile +++ b/drivers/infiniband/hw/usnic/Makefile @@ -1,5 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 -ccflags-y := -Idrivers/net/ethernet/cisco/enic +ccflags-y := -I $(srctree)/drivers/net/ethernet/cisco/enic obj-$(CONFIG_INFINIBAND_USNIC)+= usnic_verbs.o diff --git a/drivers/infiniband/hw/usnic/usnic_debugfs.c b/drivers/infiniband/hw/usnic/usnic_debugfs.c index a3115709fb03..e5a3f02fb078 100644 --- a/drivers/infiniband/hw/usnic/usnic_debugfs.c +++ b/drivers/infiniband/hw/usnic/usnic_debugfs.c @@ -113,42 +113,21 @@ static const struct file_operations flowinfo_ops = { void usnic_debugfs_init(void) { debugfs_root = debugfs_create_dir(DRV_NAME, NULL); - if (IS_ERR(debugfs_root)) { - usnic_err("Failed to create debugfs root dir, check if debugfs is enabled in kernel configuration\n"); - goto out_clear_root; - } flows_dentry = debugfs_create_dir("flows", debugfs_root); - if (IS_ERR_OR_NULL(flows_dentry)) { - usnic_err("Failed to create debugfs flow dir with err %ld\n", - PTR_ERR(flows_dentry)); - goto out_free_root; - } debugfs_create_file("build-info", S_IRUGO, debugfs_root, NULL, &usnic_debugfs_buildinfo_ops); - return; - -out_free_root: - debugfs_remove_recursive(debugfs_root); -out_clear_root: - debugfs_root = NULL; } void usnic_debugfs_exit(void) { - if (!debugfs_root) - return; - debugfs_remove_recursive(debugfs_root); debugfs_root = NULL; } void usnic_debugfs_flow_add(struct usnic_ib_qp_grp_flow *qp_flow) { - if (IS_ERR_OR_NULL(flows_dentry)) - return; - scnprintf(qp_flow->dentry_name, sizeof(qp_flow->dentry_name), "%u", qp_flow->flow->flow_id); qp_flow->dbgfs_dentry = debugfs_create_file(qp_flow->dentry_name, @@ -156,11 +135,6 @@ void usnic_debugfs_flow_add(struct usnic_ib_qp_grp_flow *qp_flow) flows_dentry, qp_flow, &flowinfo_ops); - if (IS_ERR_OR_NULL(qp_flow->dbgfs_dentry)) { - usnic_err("Failed to create dbg fs entry for flow %u with error %ld\n", - qp_flow->flow->flow_id, - PTR_ERR(qp_flow->dbgfs_dentry)); - } } void usnic_debugfs_flow_remove(struct usnic_ib_qp_grp_flow *qp_flow) diff --git a/drivers/infiniband/hw/usnic/usnic_ib_main.c b/drivers/infiniband/hw/usnic/usnic_ib_main.c index b2323a52a0dd..d88d9f8a7f9a 100644 --- a/drivers/infiniband/hw/usnic/usnic_ib_main.c +++ b/drivers/infiniband/hw/usnic/usnic_ib_main.c @@ -216,18 +216,17 @@ static int usnic_ib_netdevice_event(struct notifier_block *notifier, unsigned long event, void *ptr) { struct usnic_ib_dev *us_ibdev; + struct ib_device *ibdev; struct net_device *netdev = netdev_notifier_info_to_dev(ptr); - mutex_lock(&usnic_ib_ibdev_list_lock); - list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) { - if (us_ibdev->netdev == netdev) { - usnic_ib_handle_usdev_event(us_ibdev, event); - break; - } - } - mutex_unlock(&usnic_ib_ibdev_list_lock); + ibdev = ib_device_get_by_netdev(netdev, RDMA_DRIVER_USNIC); + if (!ibdev) + return NOTIFY_DONE; + us_ibdev = container_of(ibdev, struct usnic_ib_dev, ib_dev); + usnic_ib_handle_usdev_event(us_ibdev, event); + ib_device_put(ibdev); return NOTIFY_DONE; } @@ -282,16 +281,15 @@ static int usnic_ib_inetaddr_event(struct notifier_block *notifier, struct usnic_ib_dev *us_ibdev; struct in_ifaddr *ifa = ptr; struct net_device *netdev = ifa->ifa_dev->dev; + struct ib_device *ibdev; - mutex_lock(&usnic_ib_ibdev_list_lock); - list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) { - if (us_ibdev->netdev == netdev) { - usnic_ib_handle_inet_event(us_ibdev, event, ptr); - break; - } - } - mutex_unlock(&usnic_ib_ibdev_list_lock); + ibdev = ib_device_get_by_netdev(netdev, RDMA_DRIVER_USNIC); + if (!ibdev) + return NOTIFY_DONE; + us_ibdev = container_of(ibdev, struct usnic_ib_dev, ib_dev); + usnic_ib_handle_inet_event(us_ibdev, event, ptr); + ib_device_put(ibdev); return NOTIFY_DONE; } static struct notifier_block usnic_ib_inetaddr_notifier = { @@ -333,32 +331,26 @@ static void usnic_get_dev_fw_str(struct ib_device *device, char *str) static const struct ib_device_ops usnic_dev_ops = { .alloc_pd = usnic_ib_alloc_pd, .alloc_ucontext = usnic_ib_alloc_ucontext, - .create_ah = usnic_ib_create_ah, .create_cq = usnic_ib_create_cq, .create_qp = usnic_ib_create_qp, .dealloc_pd = usnic_ib_dealloc_pd, .dealloc_ucontext = usnic_ib_dealloc_ucontext, .dereg_mr = usnic_ib_dereg_mr, - .destroy_ah = usnic_ib_destroy_ah, .destroy_cq = usnic_ib_destroy_cq, .destroy_qp = usnic_ib_destroy_qp, .get_dev_fw_str = usnic_get_dev_fw_str, - .get_dma_mr = usnic_ib_get_dma_mr, .get_link_layer = usnic_ib_port_link_layer, - .get_netdev = usnic_get_netdev, .get_port_immutable = usnic_port_immutable, .mmap = usnic_ib_mmap, .modify_qp = usnic_ib_modify_qp, - .poll_cq = usnic_ib_poll_cq, - .post_recv = usnic_ib_post_recv, - .post_send = usnic_ib_post_send, .query_device = usnic_ib_query_device, .query_gid = usnic_ib_query_gid, .query_pkey = usnic_ib_query_pkey, .query_port = usnic_ib_query_port, .query_qp = usnic_ib_query_qp, .reg_user_mr = usnic_ib_reg_mr, - .req_notify_cq = usnic_ib_req_notify_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, usnic_ib_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, usnic_ib_ucontext, ibucontext), }; /* Start of PF discovery section */ @@ -368,11 +360,12 @@ static void *usnic_ib_device_add(struct pci_dev *dev) union ib_gid gid; struct in_device *ind; struct net_device *netdev; + int ret; usnic_dbg("\n"); netdev = pci_get_drvdata(dev); - us_ibdev = (struct usnic_ib_dev *)ib_alloc_device(sizeof(*us_ibdev)); + us_ibdev = ib_alloc_device(usnic_ib_dev, ib_dev); if (!us_ibdev) { usnic_err("Device %s context alloc failed\n", netdev_name(pci_get_drvdata(dev))); @@ -422,7 +415,11 @@ static void *usnic_ib_device_add(struct pci_dev *dev) us_ibdev->ib_dev.driver_id = RDMA_DRIVER_USNIC; rdma_set_device_sysfs_group(&us_ibdev->ib_dev, &usnic_attr_group); - if (ib_register_device(&us_ibdev->ib_dev, "usnic_%d", NULL)) + ret = ib_device_set_netdev(&us_ibdev->ib_dev, us_ibdev->netdev, 1); + if (ret) + goto err_fwd_dealloc; + + if (ib_register_device(&us_ibdev->ib_dev, "usnic_%d")) goto err_fwd_dealloc; usnic_fwd_set_mtu(us_ibdev->ufdev, us_ibdev->netdev->mtu); @@ -477,15 +474,17 @@ static void usnic_ib_undiscover_pf(struct kref *kref) &usnic_ib_ibdev_list, ib_dev_link) { if (us_ibdev->pdev == dev) { list_del(&us_ibdev->ib_dev_link); - usnic_ib_device_remove(us_ibdev); found = true; break; } } - WARN(!found, "Failed to remove PF %s\n", pci_name(dev)); mutex_unlock(&usnic_ib_ibdev_list_lock); + if (found) + usnic_ib_device_remove(us_ibdev); + else + WARN(1, "Failed to remove PF %s\n", pci_name(dev)); } static struct usnic_ib_dev *usnic_ib_discover_pf(struct usnic_vnic *vnic) @@ -691,7 +690,6 @@ out_unreg_netdev_notifier: out_pci_unreg: pci_unregister_driver(&usnic_ib_pci_driver); out_umem_fini: - usnic_uiom_fini(); return err; } @@ -704,7 +702,6 @@ static void __exit usnic_ib_destroy(void) unregister_inetaddr_notifier(&usnic_ib_inetaddr_notifier); unregister_netdevice_notifier(&usnic_ib_netdevice_notifier); pci_unregister_driver(&usnic_ib_pci_driver); - usnic_uiom_fini(); } MODULE_DESCRIPTION("Cisco VIC (usNIC) Verbs Driver"); diff --git a/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c index a7e4b2ccfaf8..c85d48ae7442 100644 --- a/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c +++ b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c @@ -50,7 +50,7 @@ static ssize_t board_id_show(struct device *device, struct device_attribute *attr, char *buf) { struct usnic_ib_dev *us_ibdev = - container_of(device, struct usnic_ib_dev, ib_dev.dev); + rdma_device_to_drv_device(device, struct usnic_ib_dev, ib_dev); unsigned short subsystem_device_id; mutex_lock(&us_ibdev->usdev_lock); @@ -67,14 +67,13 @@ static DEVICE_ATTR_RO(board_id); static ssize_t config_show(struct device *device, struct device_attribute *attr, char *buf) { - struct usnic_ib_dev *us_ibdev; + struct usnic_ib_dev *us_ibdev = + rdma_device_to_drv_device(device, struct usnic_ib_dev, ib_dev); char *ptr; unsigned left; unsigned n; enum usnic_vnic_res_type res_type; - us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev); - /* Buffer space limit is 1 page */ ptr = buf; left = PAGE_SIZE; @@ -130,9 +129,8 @@ static DEVICE_ATTR_RO(config); static ssize_t iface_show(struct device *device, struct device_attribute *attr, char *buf) { - struct usnic_ib_dev *us_ibdev; - - us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev); + struct usnic_ib_dev *us_ibdev = + rdma_device_to_drv_device(device, struct usnic_ib_dev, ib_dev); return scnprintf(buf, PAGE_SIZE, "%s\n", netdev_name(us_ibdev->netdev)); @@ -142,9 +140,8 @@ static DEVICE_ATTR_RO(iface); static ssize_t max_vf_show(struct device *device, struct device_attribute *attr, char *buf) { - struct usnic_ib_dev *us_ibdev; - - us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev); + struct usnic_ib_dev *us_ibdev = + rdma_device_to_drv_device(device, struct usnic_ib_dev, ib_dev); return scnprintf(buf, PAGE_SIZE, "%u\n", kref_read(&us_ibdev->vf_cnt)); @@ -154,10 +151,10 @@ static DEVICE_ATTR_RO(max_vf); static ssize_t qp_per_vf_show(struct device *device, struct device_attribute *attr, char *buf) { - struct usnic_ib_dev *us_ibdev; + struct usnic_ib_dev *us_ibdev = + rdma_device_to_drv_device(device, struct usnic_ib_dev, ib_dev); int qp_per_vf; - us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev); qp_per_vf = max(us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_WQ], us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_RQ]); @@ -169,9 +166,8 @@ static DEVICE_ATTR_RO(qp_per_vf); static ssize_t cq_per_vf_show(struct device *device, struct device_attribute *attr, char *buf) { - struct usnic_ib_dev *us_ibdev; - - us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev); + struct usnic_ib_dev *us_ibdev = + rdma_device_to_drv_device(device, struct usnic_ib_dev, ib_dev); return scnprintf(buf, PAGE_SIZE, "%d\n", us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_CQ]); diff --git a/drivers/infiniband/hw/usnic/usnic_ib_verbs.c b/drivers/infiniband/hw/usnic/usnic_ib_verbs.c index 1d4abef17e38..bd4521b2cc5f 100644 --- a/drivers/infiniband/hw/usnic/usnic_ib_verbs.c +++ b/drivers/infiniband/hw/usnic/usnic_ib_verbs.c @@ -37,6 +37,7 @@ #include <rdma/ib_user_verbs.h> #include <rdma/ib_addr.h> +#include <rdma/uverbs_ioctl.h> #include "usnic_abi.h" #include "usnic_ib.h" @@ -436,57 +437,33 @@ int usnic_ib_query_gid(struct ib_device *ibdev, u8 port, int index, return 0; } -struct net_device *usnic_get_netdev(struct ib_device *device, u8 port_num) -{ - struct usnic_ib_dev *us_ibdev = to_usdev(device); - - if (us_ibdev->netdev) - dev_hold(us_ibdev->netdev); - - return us_ibdev->netdev; -} - int usnic_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) { - if (index > 1) + if (index > 0) return -EINVAL; *pkey = 0xffff; return 0; } -struct ib_pd *usnic_ib_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +int usnic_ib_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct usnic_ib_pd *pd; + struct usnic_ib_pd *pd = to_upd(ibpd); void *umem_pd; - usnic_dbg("\n"); - - pd = kzalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) - return ERR_PTR(-ENOMEM); - umem_pd = pd->umem_pd = usnic_uiom_alloc_pd(); if (IS_ERR_OR_NULL(umem_pd)) { - kfree(pd); - return ERR_PTR(umem_pd ? PTR_ERR(umem_pd) : -ENOMEM); + return umem_pd ? PTR_ERR(umem_pd) : -ENOMEM; } - usnic_info("domain 0x%p allocated for context 0x%p and device %s\n", - pd, context, dev_name(&ibdev->dev)); - return &pd->ibpd; + return 0; } -int usnic_ib_dealloc_pd(struct ib_pd *pd) +void usnic_ib_dealloc_pd(struct ib_pd *pd) { - usnic_info("freeing domain 0x%p\n", pd); - usnic_uiom_dealloc_pd((to_upd(pd))->umem_pd); - kfree(pd); - return 0; } struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd, @@ -496,7 +473,8 @@ struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd, int err; struct usnic_ib_dev *us_ibdev; struct usnic_ib_qp_grp *qp_grp; - struct usnic_ib_ucontext *ucontext; + struct usnic_ib_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct usnic_ib_ucontext, ibucontext); int cq_cnt; struct usnic_vnic_res_spec res_spec; struct usnic_ib_create_qp_cmd cmd; @@ -504,7 +482,6 @@ struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd, usnic_dbg("\n"); - ucontext = to_uucontext(pd->uobject->context); us_ibdev = to_usdev(pd->device); if (init_attr->create_flags) @@ -676,37 +653,31 @@ int usnic_ib_dereg_mr(struct ib_mr *ibmr) return 0; } -struct ib_ucontext *usnic_ib_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +int usnic_ib_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { - struct usnic_ib_ucontext *context; + struct ib_device *ibdev = uctx->device; + struct usnic_ib_ucontext *context = to_ucontext(uctx); struct usnic_ib_dev *us_ibdev = to_usdev(ibdev); usnic_dbg("\n"); - context = kmalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); - INIT_LIST_HEAD(&context->qp_grp_list); mutex_lock(&us_ibdev->usdev_lock); list_add_tail(&context->link, &us_ibdev->ctx_list); mutex_unlock(&us_ibdev->usdev_lock); - return &context->ibucontext; + return 0; } -int usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) +void usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) { struct usnic_ib_ucontext *context = to_uucontext(ibcontext); struct usnic_ib_dev *us_ibdev = to_usdev(ibcontext->device); usnic_dbg("\n"); mutex_lock(&us_ibdev->usdev_lock); - BUG_ON(!list_empty(&context->qp_grp_list)); + WARN_ON_ONCE(!list_empty(&context->qp_grp_list)); list_del(&context->link); mutex_unlock(&us_ibdev->usdev_lock); - kfree(context); - return 0; } int usnic_ib_mmap(struct ib_ucontext *context, @@ -760,57 +731,4 @@ int usnic_ib_mmap(struct ib_ucontext *context, return -EINVAL; } -/* In ib callbacks section - Start of stub funcs */ -struct ib_ah *usnic_ib_create_ah(struct ib_pd *pd, - struct rdma_ah_attr *ah_attr, - u32 flags, - struct ib_udata *udata) - -{ - usnic_dbg("\n"); - return ERR_PTR(-EPERM); -} - -int usnic_ib_destroy_ah(struct ib_ah *ah, u32 flags) -{ - usnic_dbg("\n"); - return -EINVAL; -} - -int usnic_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, - const struct ib_send_wr **bad_wr) -{ - usnic_dbg("\n"); - return -EINVAL; -} - -int usnic_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr, - const struct ib_recv_wr **bad_wr) -{ - usnic_dbg("\n"); - return -EINVAL; -} - -int usnic_ib_poll_cq(struct ib_cq *ibcq, int num_entries, - struct ib_wc *wc) -{ - usnic_dbg("\n"); - return -EINVAL; -} - -int usnic_ib_req_notify_cq(struct ib_cq *cq, - enum ib_cq_notify_flags flags) -{ - usnic_dbg("\n"); - return -EINVAL; -} - -struct ib_mr *usnic_ib_get_dma_mr(struct ib_pd *pd, int acc) -{ - usnic_dbg("\n"); - return ERR_PTR(-ENOMEM); -} - - -/* In ib callbacks section - End of stub funcs */ /* End of ib callbacks section */ diff --git a/drivers/infiniband/hw/usnic/usnic_ib_verbs.h b/drivers/infiniband/hw/usnic/usnic_ib_verbs.h index e33144261b9a..c40e89b6246f 100644 --- a/drivers/infiniband/hw/usnic/usnic_ib_verbs.h +++ b/drivers/infiniband/hw/usnic/usnic_ib_verbs.h @@ -48,13 +48,11 @@ int usnic_ib_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr, struct ib_qp_init_attr *qp_init_attr); int usnic_ib_query_gid(struct ib_device *ibdev, u8 port, int index, union ib_gid *gid); -struct net_device *usnic_get_netdev(struct ib_device *device, u8 port_num); int usnic_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey); -struct ib_pd *usnic_ib_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata); -int usnic_ib_dealloc_pd(struct ib_pd *pd); +int usnic_ib_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata); +void usnic_ib_dealloc_pd(struct ib_pd *pd); struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *init_attr, struct ib_udata *udata); @@ -70,24 +68,8 @@ struct ib_mr *usnic_ib_reg_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_udata *udata); int usnic_ib_dereg_mr(struct ib_mr *ibmr); -struct ib_ucontext *usnic_ib_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata); -int usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext); +int usnic_ib_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata); +void usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext); int usnic_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); -struct ib_ah *usnic_ib_create_ah(struct ib_pd *pd, - struct rdma_ah_attr *ah_attr, - u32 flags, - struct ib_udata *udata); - -int usnic_ib_destroy_ah(struct ib_ah *ah, u32 flags); -int usnic_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, - const struct ib_send_wr **bad_wr); -int usnic_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr, - const struct ib_recv_wr **bad_wr); -int usnic_ib_poll_cq(struct ib_cq *ibcq, int num_entries, - struct ib_wc *wc); -int usnic_ib_req_notify_cq(struct ib_cq *cq, - enum ib_cq_notify_flags flags); -struct ib_mr *usnic_ib_get_dma_mr(struct ib_pd *pd, int acc); #endif /* !USNIC_IB_VERBS_H */ diff --git a/drivers/infiniband/hw/usnic/usnic_uiom.c b/drivers/infiniband/hw/usnic/usnic_uiom.c index 49275a548751..06862a6af185 100644 --- a/drivers/infiniband/hw/usnic/usnic_uiom.c +++ b/drivers/infiniband/hw/usnic/usnic_uiom.c @@ -47,8 +47,6 @@ #include "usnic_uiom.h" #include "usnic_uiom_interval_tree.h" -static struct workqueue_struct *usnic_uiom_wq; - #define USNIC_UIOM_PAGE_CHUNK \ ((PAGE_SIZE - offsetof(struct usnic_uiom_chunk, page_list)) /\ ((void *) &((struct usnic_uiom_chunk *) 0)->page_list[1] - \ @@ -127,9 +125,9 @@ static int usnic_uiom_get_pages(unsigned long addr, size_t size, int writable, npages = PAGE_ALIGN(size + (addr & ~PAGE_MASK)) >> PAGE_SHIFT; uiomr->owning_mm = mm = current->mm; - down_write(&mm->mmap_sem); + down_read(&mm->mmap_sem); - locked = npages + current->mm->pinned_vm; + locked = atomic64_add_return(npages, ¤t->mm->pinned_vm); lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { @@ -157,9 +155,8 @@ static int usnic_uiom_get_pages(unsigned long addr, size_t size, int writable, off = 0; while (ret) { - chunk = kmalloc(sizeof(*chunk) + - sizeof(struct scatterlist) * - min_t(int, ret, USNIC_UIOM_PAGE_CHUNK), + chunk = kmalloc(struct_size(chunk, page_list, + min_t(int, ret, USNIC_UIOM_PAGE_CHUNK)), GFP_KERNEL); if (!chunk) { ret = -ENOMEM; @@ -185,14 +182,13 @@ static int usnic_uiom_get_pages(unsigned long addr, size_t size, int writable, } out: - if (ret < 0) + if (ret < 0) { usnic_uiom_put_pages(chunk_list, 0); - else { - mm->pinned_vm = locked; + atomic64_sub(npages, ¤t->mm->pinned_vm); + } else mmgrab(uiomr->owning_mm); - } - up_write(&mm->mmap_sem); + up_read(&mm->mmap_sem); free_page((unsigned long) page_list); return ret; } @@ -436,43 +432,12 @@ static inline size_t usnic_uiom_num_pages(struct usnic_uiom_reg *uiomr) return PAGE_ALIGN(uiomr->length + uiomr->offset) >> PAGE_SHIFT; } -static void usnic_uiom_release_defer(struct work_struct *work) -{ - struct usnic_uiom_reg *uiomr = - container_of(work, struct usnic_uiom_reg, work); - - down_write(&uiomr->owning_mm->mmap_sem); - uiomr->owning_mm->pinned_vm -= usnic_uiom_num_pages(uiomr); - up_write(&uiomr->owning_mm->mmap_sem); - - __usnic_uiom_release_tail(uiomr); -} - void usnic_uiom_reg_release(struct usnic_uiom_reg *uiomr, struct ib_ucontext *context) { __usnic_uiom_reg_release(uiomr->pd, uiomr, 1); - /* - * We may be called with the mm's mmap_sem already held. This - * can happen when a userspace munmap() is the call that drops - * the last reference to our file and calls our release - * method. If there are memory regions to destroy, we'll end - * up here and not be able to take the mmap_sem. In that case - * we defer the vm_locked accounting to a workqueue. - */ - if (context->closing) { - if (!down_write_trylock(&uiomr->owning_mm->mmap_sem)) { - INIT_WORK(&uiomr->work, usnic_uiom_release_defer); - queue_work(usnic_uiom_wq, &uiomr->work); - return; - } - } else { - down_write(&uiomr->owning_mm->mmap_sem); - } - uiomr->owning_mm->pinned_vm -= usnic_uiom_num_pages(uiomr); - up_write(&uiomr->owning_mm->mmap_sem); - + atomic64_sub(usnic_uiom_num_pages(uiomr), &uiomr->owning_mm->pinned_vm); __usnic_uiom_release_tail(uiomr); } @@ -601,17 +566,5 @@ int usnic_uiom_init(char *drv_name) return -EPERM; } - usnic_uiom_wq = create_workqueue(drv_name); - if (!usnic_uiom_wq) { - usnic_err("Unable to alloc wq for drv %s\n", drv_name); - return -ENOMEM; - } - return 0; } - -void usnic_uiom_fini(void) -{ - flush_workqueue(usnic_uiom_wq); - destroy_workqueue(usnic_uiom_wq); -} diff --git a/drivers/infiniband/hw/usnic/usnic_uiom.h b/drivers/infiniband/hw/usnic/usnic_uiom.h index b86a9731071b..c88cfa087e3a 100644 --- a/drivers/infiniband/hw/usnic/usnic_uiom.h +++ b/drivers/infiniband/hw/usnic/usnic_uiom.h @@ -93,5 +93,4 @@ struct usnic_uiom_reg *usnic_uiom_reg_get(struct usnic_uiom_pd *pd, void usnic_uiom_reg_release(struct usnic_uiom_reg *uiomr, struct ib_ucontext *ucontext); int usnic_uiom_init(char *drv_name); -void usnic_uiom_fini(void); #endif /* USNIC_UIOM_H_ */ diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_cq.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_cq.c index 0f004c737620..104c7db4704f 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_cq.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_cq.c @@ -141,7 +141,7 @@ struct ib_cq *pvrdma_create_cq(struct ib_device *ibdev, goto err_cq; } - cq->umem = ib_umem_get(context, ucmd.buf_addr, ucmd.buf_size, + cq->umem = ib_umem_get(udata, ucmd.buf_addr, ucmd.buf_size, IB_ACCESS_LOCAL_WRITE, 1); if (IS_ERR(cq->umem)) { ret = PTR_ERR(cq->umem); diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h index 6fd5a8f4e2f6..8f9749d54688 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h @@ -57,7 +57,8 @@ #define PVRDMA_ROCEV1_VERSION 17 #define PVRDMA_ROCEV2_VERSION 18 -#define PVRDMA_VERSION PVRDMA_ROCEV2_VERSION +#define PVRDMA_PPN64_VERSION 19 +#define PVRDMA_VERSION PVRDMA_PPN64_VERSION #define PVRDMA_BOARD_ID 1 #define PVRDMA_REV_ID 1 @@ -279,8 +280,10 @@ struct pvrdma_device_shared_region { /* W: Async ring page info. */ struct pvrdma_ring_page_info cq_ring_pages; /* W: CQ ring page info. */ - u32 uar_pfn; /* W: UAR pageframe. */ - u32 pad2; /* Pad to 8-byte align. */ + union { + u32 uar_pfn; /* W: UAR pageframe. */ + u64 uar_pfn64; /* W: 64-bit UAR page frame. */ + }; struct pvrdma_device_caps caps; /* R: Device capabilities. */ }; @@ -411,8 +414,10 @@ struct pvrdma_cmd_query_pkey_resp { struct pvrdma_cmd_create_uc { struct pvrdma_cmd_hdr hdr; - u32 pfn; /* UAR page frame number */ - u8 reserved[4]; + union { + u32 pfn; /* UAR page frame number */ + u64 pfn64; /* 64-bit UAR page frame number */ + }; }; struct pvrdma_cmd_create_uc_resp { diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c index 39c37b6fd715..6d8b3e0de57a 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c @@ -195,6 +195,8 @@ static const struct ib_device_ops pvrdma_dev_ops = { .query_qp = pvrdma_query_qp, .reg_user_mr = pvrdma_reg_user_mr, .req_notify_cq = pvrdma_req_notify_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, pvrdma_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, pvrdma_ucontext, ibucontext), }; static const struct ib_device_ops pvrdma_dev_srq_ops = { @@ -278,7 +280,7 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) spin_lock_init(&dev->srq_tbl_lock); rdma_set_device_sysfs_group(&dev->ib_dev, &pvrdma_attr_group); - ret = ib_register_device(&dev->ib_dev, "vmw_pvrdma%d", NULL); + ret = ib_register_device(&dev->ib_dev, "vmw_pvrdma%d"); if (ret) goto err_srq_free; @@ -795,7 +797,7 @@ static int pvrdma_pci_probe(struct pci_dev *pdev, dev_dbg(&pdev->dev, "initializing driver %s\n", pci_name(pdev)); /* Allocate zero-out device */ - dev = (struct pvrdma_dev *)ib_alloc_device(sizeof(*dev)); + dev = ib_alloc_device(pvrdma_dev, ib_dev); if (!dev) { dev_err(&pdev->dev, "failed to allocate IB device\n"); return -ENOMEM; @@ -905,7 +907,11 @@ static int pvrdma_pci_probe(struct pci_dev *pdev, PVRDMA_GOS_BITS_64; dev->dsr->gos_info.gos_type = PVRDMA_GOS_TYPE_LINUX; dev->dsr->gos_info.gos_ver = 1; - dev->dsr->uar_pfn = dev->driver_uar.pfn; + + if (dev->dsr_version < PVRDMA_PPN64_VERSION) + dev->dsr->uar_pfn = dev->driver_uar.pfn; + else + dev->dsr->uar_pfn64 = dev->driver_uar.pfn; /* Command slot. */ dev->cmd_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE, diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_misc.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_misc.c index fb0c5c0976b3..7944c58ded0e 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_misc.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_misc.c @@ -183,25 +183,20 @@ int pvrdma_page_dir_insert_umem(struct pvrdma_page_dir *pdir, struct ib_umem *umem, u64 offset) { u64 i = offset; - int j, entry; - int ret = 0, len = 0; - struct scatterlist *sg; + int ret = 0; + struct sg_dma_page_iter sg_iter; if (offset >= pdir->npages) return -EINVAL; - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { - len = sg_dma_len(sg) >> PAGE_SHIFT; - for (j = 0; j < len; j++) { - dma_addr_t addr = sg_dma_address(sg) + - (j << umem->page_shift); + for_each_sg_dma_page(umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + dma_addr_t addr = sg_page_iter_dma_address(&sg_iter); - ret = pvrdma_page_dir_insert_dma(pdir, i, addr); - if (ret) - goto exit; + ret = pvrdma_page_dir_insert_dma(pdir, i, addr); + if (ret) + goto exit; - i++; - } + i++; } exit: diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c index fa96fa4fb829..a85884e90e84 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_mr.c @@ -126,8 +126,7 @@ struct ib_mr *pvrdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, return ERR_PTR(-EINVAL); } - umem = ib_umem_get(pd->uobject->context, start, - length, access_flags, 0); + umem = ib_umem_get(udata, start, length, access_flags, 0); if (IS_ERR(umem)) { dev_warn(&dev->pdev->dev, "could not get umem for mem region\n"); diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c index 1ec3646087ba..08f4257169bd 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c @@ -262,8 +262,7 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, if (!is_srq) { /* set qp->sq.wqe_cnt, shift, buf_size.. */ - qp->rumem = ib_umem_get(pd->uobject->context, - ucmd.rbuf_addr, + qp->rumem = ib_umem_get(udata, ucmd.rbuf_addr, ucmd.rbuf_size, 0, 0); if (IS_ERR(qp->rumem)) { ret = PTR_ERR(qp->rumem); @@ -275,8 +274,7 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, qp->srq = to_vsrq(init_attr->srq); } - qp->sumem = ib_umem_get(pd->uobject->context, - ucmd.sbuf_addr, + qp->sumem = ib_umem_get(udata, ucmd.sbuf_addr, ucmd.sbuf_size, 0, 0); if (IS_ERR(qp->sumem)) { if (!is_srq) diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c index 06ba7c7a2235..951d9d68107a 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c @@ -153,9 +153,7 @@ struct ib_srq *pvrdma_create_srq(struct ib_pd *pd, goto err_srq; } - srq->umem = ib_umem_get(pd->uobject->context, - ucmd.buf_addr, - ucmd.buf_size, 0, 0); + srq->umem = ib_umem_get(udata, ucmd.buf_addr, ucmd.buf_size, 0, 0); if (IS_ERR(srq->umem)) { ret = PTR_ERR(srq->umem); goto err_srq; diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c index 4d238d0e484b..42fe821f8d58 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c @@ -306,47 +306,42 @@ out: /** * pvrdma_alloc_ucontext - allocate ucontext - * @ibdev: the IB device + * @uctx: the uverbs countext * @udata: user data * - * @return: the ib_ucontext pointer on success, otherwise errno. + * @return: zero on success, otherwise errno. */ -struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +int pvrdma_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { + struct ib_device *ibdev = uctx->device; struct pvrdma_dev *vdev = to_vdev(ibdev); - struct pvrdma_ucontext *context; - union pvrdma_cmd_req req; - union pvrdma_cmd_resp rsp; + struct pvrdma_ucontext *context = to_vucontext(uctx); + union pvrdma_cmd_req req = {}; + union pvrdma_cmd_resp rsp = {}; struct pvrdma_cmd_create_uc *cmd = &req.create_uc; struct pvrdma_cmd_create_uc_resp *resp = &rsp.create_uc_resp; - struct pvrdma_alloc_ucontext_resp uresp = {0}; + struct pvrdma_alloc_ucontext_resp uresp = {}; int ret; - void *ptr; if (!vdev->ib_active) - return ERR_PTR(-EAGAIN); - - context = kmalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); + return -EAGAIN; context->dev = vdev; ret = pvrdma_uar_alloc(vdev, &context->uar); - if (ret) { - kfree(context); - return ERR_PTR(-ENOMEM); - } + if (ret) + return -ENOMEM; /* get ctx_handle from host */ - memset(cmd, 0, sizeof(*cmd)); - cmd->pfn = context->uar.pfn; + if (vdev->dsr_version < PVRDMA_PPN64_VERSION) + cmd->pfn = context->uar.pfn; + else + cmd->pfn64 = context->uar.pfn; + cmd->hdr.cmd = PVRDMA_CMD_CREATE_UC; ret = pvrdma_cmd_post(vdev, &req, &rsp, PVRDMA_CMD_CREATE_UC_RESP); if (ret < 0) { dev_warn(&vdev->pdev->dev, "could not create ucontext, error: %d\n", ret); - ptr = ERR_PTR(ret); goto err; } @@ -357,33 +352,28 @@ struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev, ret = ib_copy_to_udata(udata, &uresp, sizeof(uresp)); if (ret) { pvrdma_uar_free(vdev, &context->uar); - context->ibucontext.device = ibdev; pvrdma_dealloc_ucontext(&context->ibucontext); - return ERR_PTR(-EFAULT); + return -EFAULT; } - return &context->ibucontext; + return 0; err: pvrdma_uar_free(vdev, &context->uar); - kfree(context); - return ptr; + return ret; } /** * pvrdma_dealloc_ucontext - deallocate ucontext * @ibcontext: the ucontext - * - * @return: 0 on success, otherwise errno. */ -int pvrdma_dealloc_ucontext(struct ib_ucontext *ibcontext) +void pvrdma_dealloc_ucontext(struct ib_ucontext *ibcontext) { struct pvrdma_ucontext *context = to_vucontext(ibcontext); - union pvrdma_cmd_req req; + union pvrdma_cmd_req req = {}; struct pvrdma_cmd_destroy_uc *cmd = &req.destroy_uc; int ret; - memset(cmd, 0, sizeof(*cmd)); cmd->hdr.cmd = PVRDMA_CMD_DESTROY_UC; cmd->ctx_handle = context->ctx_handle; @@ -394,9 +384,6 @@ int pvrdma_dealloc_ucontext(struct ib_ucontext *ibcontext) /* Free the UAR even if the device command failed */ pvrdma_uar_free(to_vdev(ibcontext->device), &context->uar); - kfree(context); - - return ret; } /** @@ -433,37 +420,29 @@ int pvrdma_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma) /** * pvrdma_alloc_pd - allocate protection domain - * @ibdev: the IB device + * @ibpd: PD pointer * @context: user context * @udata: user data * * @return: the ib_pd protection domain pointer on success, otherwise errno. */ -struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +int pvrdma_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct pvrdma_pd *pd; + struct ib_device *ibdev = ibpd->device; + struct pvrdma_pd *pd = to_vpd(ibpd); struct pvrdma_dev *dev = to_vdev(ibdev); - union pvrdma_cmd_req req; - union pvrdma_cmd_resp rsp; + union pvrdma_cmd_req req = {}; + union pvrdma_cmd_resp rsp = {}; struct pvrdma_cmd_create_pd *cmd = &req.create_pd; struct pvrdma_cmd_create_pd_resp *resp = &rsp.create_pd_resp; struct pvrdma_alloc_pd_resp pd_resp = {0}; int ret; - void *ptr; /* Check allowed max pds */ if (!atomic_add_unless(&dev->num_pds, 1, dev->dsr->caps.max_pd)) - return ERR_PTR(-ENOMEM); - - pd = kmalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) { - ptr = ERR_PTR(-ENOMEM); - goto err; - } + return -ENOMEM; - memset(cmd, 0, sizeof(*cmd)); cmd->hdr.cmd = PVRDMA_CMD_CREATE_PD; cmd->ctx_handle = (context) ? to_vucontext(context)->ctx_handle : 0; ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_PD_RESP); @@ -471,8 +450,7 @@ struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev, dev_warn(&dev->pdev->dev, "failed to allocate protection domain, error: %d\n", ret); - ptr = ERR_PTR(ret); - goto freepd; + goto err; } pd->privileged = !context; @@ -485,18 +463,16 @@ struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev, dev_warn(&dev->pdev->dev, "failed to copy back protection domain\n"); pvrdma_dealloc_pd(&pd->ibpd); - return ERR_PTR(-EFAULT); + return -EFAULT; } } /* u32 pd handle */ - return &pd->ibpd; + return 0; -freepd: - kfree(pd); err: atomic_dec(&dev->num_pds); - return ptr; + return ret; } /** @@ -505,14 +481,13 @@ err: * * @return: 0 on success, otherwise errno. */ -int pvrdma_dealloc_pd(struct ib_pd *pd) +void pvrdma_dealloc_pd(struct ib_pd *pd) { struct pvrdma_dev *dev = to_vdev(pd->device); - union pvrdma_cmd_req req; + union pvrdma_cmd_req req = {}; struct pvrdma_cmd_destroy_pd *cmd = &req.destroy_pd; int ret; - memset(cmd, 0, sizeof(*cmd)); cmd->hdr.cmd = PVRDMA_CMD_DESTROY_PD; cmd->pd_handle = to_vpd(pd)->pd_handle; @@ -522,10 +497,7 @@ int pvrdma_dealloc_pd(struct ib_pd *pd) "could not dealloc protection domain, error: %d\n", ret); - kfree(to_vpd(pd)); atomic_dec(&dev->num_pds); - - return 0; } /** diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h index f7f758d60110..607aa131d67c 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h @@ -396,13 +396,11 @@ int pvrdma_modify_device(struct ib_device *ibdev, int mask, int pvrdma_modify_port(struct ib_device *ibdev, u8 port, int mask, struct ib_port_modify *props); int pvrdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); -struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata); -int pvrdma_dealloc_ucontext(struct ib_ucontext *context); -struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata); -int pvrdma_dealloc_pd(struct ib_pd *ibpd); +int pvrdma_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata); +void pvrdma_dealloc_ucontext(struct ib_ucontext *context); +int pvrdma_alloc_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata); +void pvrdma_dealloc_pd(struct ib_pd *ibpd); struct ib_mr *pvrdma_get_dma_mr(struct ib_pd *pd, int acc); struct ib_mr *pvrdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int access_flags, diff --git a/drivers/infiniband/sw/rdmavt/mr.c b/drivers/infiniband/sw/rdmavt/mr.c index 49c9541050d4..728795043496 100644 --- a/drivers/infiniband/sw/rdmavt/mr.c +++ b/drivers/infiniband/sw/rdmavt/mr.c @@ -381,15 +381,14 @@ struct ib_mr *rvt_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, { struct rvt_mr *mr; struct ib_umem *umem; - struct scatterlist *sg; - int n, m, entry; + struct sg_page_iter sg_iter; + int n, m; struct ib_mr *ret; if (length == 0) return ERR_PTR(-EINVAL); - umem = ib_umem_get(pd->uobject->context, start, length, - mr_access_flags, 0); + umem = ib_umem_get(udata, start, length, mr_access_flags, 0); if (IS_ERR(umem)) return (void *)umem; @@ -408,23 +407,21 @@ struct ib_mr *rvt_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mr->mr.access_flags = mr_access_flags; mr->umem = umem; - mr->mr.page_shift = umem->page_shift; + mr->mr.page_shift = PAGE_SHIFT; m = 0; n = 0; - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { + for_each_sg_page (umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { void *vaddr; - vaddr = page_address(sg_page(sg)); + vaddr = page_address(sg_page_iter_page(&sg_iter)); if (!vaddr) { ret = ERR_PTR(-EINVAL); goto bail_inval; } mr->mr.map[m]->segs[n].vaddr = vaddr; - mr->mr.map[m]->segs[n].length = BIT(umem->page_shift); - trace_rvt_mr_user_seg(&mr->mr, m, n, vaddr, - BIT(umem->page_shift)); - n++; - if (n == RVT_SEGSZ) { + mr->mr.map[m]->segs[n].length = PAGE_SIZE; + trace_rvt_mr_user_seg(&mr->mr, m, n, vaddr, PAGE_SIZE); + if (++n == RVT_SEGSZ) { m++; n = 0; } diff --git a/drivers/infiniband/sw/rdmavt/pd.c b/drivers/infiniband/sw/rdmavt/pd.c index 8a89afff3363..6033054b22fa 100644 --- a/drivers/infiniband/sw/rdmavt/pd.c +++ b/drivers/infiniband/sw/rdmavt/pd.c @@ -50,7 +50,7 @@ /** * rvt_alloc_pd - allocate a protection domain - * @ibdev: ib device + * @ibpd: PD * @context: optional user context * @udata: optional user data * @@ -58,19 +58,14 @@ * * Return: 0 on success */ -struct ib_pd *rvt_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata) +int rvt_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { + struct ib_device *ibdev = ibpd->device; struct rvt_dev_info *dev = ib_to_rvt(ibdev); - struct rvt_pd *pd; - struct ib_pd *ret; + struct rvt_pd *pd = ibpd_to_rvtpd(ibpd); + int ret = 0; - pd = kmalloc(sizeof(*pd), GFP_KERNEL); - if (!pd) { - ret = ERR_PTR(-ENOMEM); - goto bail; - } /* * While we could continue allocating protecetion domains, being * constrained only by system resources. The IBTA spec defines that @@ -81,8 +76,7 @@ struct ib_pd *rvt_alloc_pd(struct ib_device *ibdev, spin_lock(&dev->n_pds_lock); if (dev->n_pds_allocated == dev->dparms.props.max_pd) { spin_unlock(&dev->n_pds_lock); - kfree(pd); - ret = ERR_PTR(-ENOMEM); + ret = -ENOMEM; goto bail; } @@ -92,8 +86,6 @@ struct ib_pd *rvt_alloc_pd(struct ib_device *ibdev, /* ib_alloc_pd() will initialize pd->ibpd. */ pd->user = !!udata; - ret = &pd->ibpd; - bail: return ret; } @@ -104,16 +96,11 @@ bail: * * Return: always 0 */ -int rvt_dealloc_pd(struct ib_pd *ibpd) +void rvt_dealloc_pd(struct ib_pd *ibpd) { - struct rvt_pd *pd = ibpd_to_rvtpd(ibpd); struct rvt_dev_info *dev = ib_to_rvt(ibpd->device); spin_lock(&dev->n_pds_lock); dev->n_pds_allocated--; spin_unlock(&dev->n_pds_lock); - - kfree(pd); - - return 0; } diff --git a/drivers/infiniband/sw/rdmavt/pd.h b/drivers/infiniband/sw/rdmavt/pd.h index 1892ca4a9746..7a887e4a45e7 100644 --- a/drivers/infiniband/sw/rdmavt/pd.h +++ b/drivers/infiniband/sw/rdmavt/pd.h @@ -50,9 +50,8 @@ #include <rdma/rdma_vt.h> -struct ib_pd *rvt_alloc_pd(struct ib_device *ibdev, - struct ib_ucontext *context, - struct ib_udata *udata); -int rvt_dealloc_pd(struct ib_pd *ibpd); +int rvt_alloc_pd(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata); +void rvt_dealloc_pd(struct ib_pd *ibpd); #endif /* DEF_RDMAVTPD_H */ diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c index c6cc3e4ab71d..a34b9a2a32b6 100644 --- a/drivers/infiniband/sw/rdmavt/qp.c +++ b/drivers/infiniband/sw/rdmavt/qp.c @@ -53,6 +53,7 @@ #include <rdma/ib_verbs.h> #include <rdma/ib_hdrs.h> #include <rdma/opa_addr.h> +#include <rdma/uverbs_ioctl.h> #include "qp.h" #include "vt.h" #include "trace.h" @@ -854,6 +855,7 @@ static void rvt_init_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp, qp->s_mig_state = IB_MIG_MIGRATED; qp->r_head_ack_queue = 0; qp->s_tail_ack_queue = 0; + qp->s_acked_ack_queue = 0; qp->s_num_rd_atomic = 0; if (qp->r_rq.wq) { qp->r_rq.wq->head = 0; @@ -955,6 +957,8 @@ struct ib_qp *rvt_create_qp(struct ib_pd *ibpd, size_t sg_list_sz; struct ib_qp *ret = ERR_PTR(-ENOMEM); struct rvt_dev_info *rdi = ib_to_rvt(ibpd->device); + struct rvt_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct rvt_ucontext, ibucontext); void *priv = NULL; size_t sqsize; @@ -1128,7 +1132,7 @@ struct ib_qp *rvt_create_qp(struct ib_pd *ibpd, u32 s = sizeof(struct rvt_rwq) + qp->r_rq.size * sz; qp->ip = rvt_create_mmap_info(rdi, s, - ibpd->uobject->context, + &ucontext->ibucontext, qp->r_rq.wq); if (!qp->ip) { ret = ERR_PTR(-ENOMEM); @@ -1642,11 +1646,11 @@ int rvt_destroy_qp(struct ib_qp *ibqp) kref_put(&qp->ip->ref, rvt_release_mmap_info); else vfree(qp->r_rq.wq); - vfree(qp->s_wq); rdi->driver_f.qp_priv_free(rdi, qp); kfree(qp->s_ack_queue); rdma_destroy_ah_attr(&qp->remote_ah_attr); rdma_destroy_ah_attr(&qp->alt_ah_attr); + vfree(qp->s_wq); kfree(qp); return 0; } @@ -2393,11 +2397,12 @@ static inline unsigned long rvt_aeth_to_usec(u32 aeth) } /* - * rvt_add_retry_timer - add/start a retry timer + * rvt_add_retry_timer_ext - add/start a retry timer * @qp - the QP + * @shift - timeout shift to wait for multiple packets * add a retry timer on the QP */ -void rvt_add_retry_timer(struct rvt_qp *qp) +void rvt_add_retry_timer_ext(struct rvt_qp *qp, u8 shift) { struct ib_qp *ibqp = &qp->ibqp; struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device); @@ -2405,11 +2410,11 @@ void rvt_add_retry_timer(struct rvt_qp *qp) lockdep_assert_held(&qp->s_lock); qp->s_flags |= RVT_S_TIMER; /* 4.096 usec. * (1 << qp->timeout) */ - qp->s_timer.expires = jiffies + qp->timeout_jiffies + - rdi->busy_jiffies; + qp->s_timer.expires = jiffies + rdi->busy_jiffies + + (qp->timeout_jiffies << shift); add_timer(&qp->s_timer); } -EXPORT_SYMBOL(rvt_add_retry_timer); +EXPORT_SYMBOL(rvt_add_retry_timer_ext); /** * rvt_add_rnr_timer - add/start an rnr timer @@ -2785,6 +2790,18 @@ again: } EXPORT_SYMBOL(rvt_copy_sge); +static enum ib_wc_status loopback_qp_drop(struct rvt_ibport *rvp, + struct rvt_qp *sqp) +{ + rvp->n_pkt_drops++; + /* + * For RC, the requester would timeout and retry so + * shortcut the timeouts and just signal too many retries. + */ + return sqp->ibqp.qp_type == IB_QPT_RC ? + IB_WC_RETRY_EXC_ERR : IB_WC_SUCCESS; +} + /** * ruc_loopback - handle UC and RC loopback requests * @sqp: the sending QP @@ -2857,17 +2874,14 @@ again: } spin_unlock_irqrestore(&sqp->s_lock, flags); - if (!qp || !(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK) || + if (!qp) { + send_status = loopback_qp_drop(rvp, sqp); + goto serr_no_r_lock; + } + spin_lock_irqsave(&qp->r_lock, flags); + if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK) || qp->ibqp.qp_type != sqp->ibqp.qp_type) { - rvp->n_pkt_drops++; - /* - * For RC, the requester would timeout and retry so - * shortcut the timeouts and just signal too many retries. - */ - if (sqp->ibqp.qp_type == IB_QPT_RC) - send_status = IB_WC_RETRY_EXC_ERR; - else - send_status = IB_WC_SUCCESS; + send_status = loopback_qp_drop(rvp, sqp); goto serr; } @@ -2893,18 +2907,8 @@ again: goto send_comp; case IB_WR_SEND_WITH_INV: - if (!rvt_invalidate_rkey(qp, wqe->wr.ex.invalidate_rkey)) { - wc.wc_flags = IB_WC_WITH_INVALIDATE; - wc.ex.invalidate_rkey = wqe->wr.ex.invalidate_rkey; - } - goto send; - case IB_WR_SEND_WITH_IMM: - wc.wc_flags = IB_WC_WITH_IMM; - wc.ex.imm_data = wqe->wr.ex.imm_data; - /* FALLTHROUGH */ case IB_WR_SEND: -send: ret = rvt_get_rwqe(qp, false); if (ret < 0) goto op_err; @@ -2912,6 +2916,22 @@ send: goto rnr_nak; if (wqe->length > qp->r_len) goto inv_err; + switch (wqe->wr.opcode) { + case IB_WR_SEND_WITH_INV: + if (!rvt_invalidate_rkey(qp, + wqe->wr.ex.invalidate_rkey)) { + wc.wc_flags = IB_WC_WITH_INVALIDATE; + wc.ex.invalidate_rkey = + wqe->wr.ex.invalidate_rkey; + } + break; + case IB_WR_SEND_WITH_IMM: + wc.wc_flags = IB_WC_WITH_IMM; + wc.ex.imm_data = wqe->wr.ex.imm_data; + break; + default: + break; + } break; case IB_WR_RDMA_WRITE_WITH_IMM: @@ -2988,34 +3008,12 @@ do_write: sge = &sqp->s_sge.sge; while (sqp->s_len) { - u32 len = sqp->s_len; + u32 len = rvt_get_sge_length(sge, sqp->s_len); - if (len > sge->length) - len = sge->length; - if (len > sge->sge_length) - len = sge->sge_length; WARN_ON_ONCE(len == 0); rvt_copy_sge(qp, &qp->r_sge, sge->vaddr, len, release, copy_last); - sge->vaddr += len; - sge->length -= len; - sge->sge_length -= len; - if (sge->sge_length == 0) { - if (!release) - rvt_put_mr(sge->mr); - if (--sqp->s_sge.num_sge) - *sge = *sqp->s_sge.sg_list++; - } else if (sge->length == 0 && sge->mr->lkey) { - if (++sge->n >= RVT_SEGSZ) { - if (++sge->m >= sge->mr->mapsz) - break; - sge->n = 0; - } - sge->vaddr = - sge->mr->map[sge->m]->segs[sge->n].vaddr; - sge->length = - sge->mr->map[sge->m]->segs[sge->n].length; - } + rvt_update_sge(&sqp->s_sge, len, !release); sqp->s_len -= len; } if (release) @@ -3041,6 +3039,7 @@ do_write: wqe->wr.send_flags & IB_SEND_SOLICITED); send_comp: + spin_unlock_irqrestore(&qp->r_lock, flags); spin_lock_irqsave(&sqp->s_lock, flags); rvp->n_loop_pkts++; flush_send: @@ -3067,6 +3066,7 @@ rnr_nak: } if (sqp->s_rnr_retry_cnt < 7) sqp->s_rnr_retry--; + spin_unlock_irqrestore(&qp->r_lock, flags); spin_lock_irqsave(&sqp->s_lock, flags); if (!(ib_rvt_state_ops[sqp->state] & RVT_PROCESS_RECV_OK)) goto clr_busy; @@ -3095,6 +3095,8 @@ err: rvt_rc_error(qp, wc.status); serr: + spin_unlock_irqrestore(&qp->r_lock, flags); +serr_no_r_lock: spin_lock_irqsave(&sqp->s_lock, flags); rvt_send_complete(sqp, wqe, send_status); if (sqp->ibqp.qp_type == IB_QPT_RC) { diff --git a/drivers/infiniband/sw/rdmavt/rc.c b/drivers/infiniband/sw/rdmavt/rc.c index 6131cc558bdb..8d71647820a8 100644 --- a/drivers/infiniband/sw/rdmavt/rc.c +++ b/drivers/infiniband/sw/rdmavt/rc.c @@ -187,3 +187,16 @@ void rvt_get_credit(struct rvt_qp *qp, u32 aeth) } } EXPORT_SYMBOL(rvt_get_credit); + +/* rvt_restart_sge - rewind the sge state for a wqe */ +u32 rvt_restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe, u32 len) +{ + ss->sge = wqe->sg_list[0]; + ss->sg_list = wqe->sg_list + 1; + ss->num_sge = wqe->wr.num_sge; + ss->total_len = wqe->length; + rvt_skip_sge(ss, len, false); + return wqe->length - len; +} +EXPORT_SYMBOL(rvt_restart_sge); + diff --git a/drivers/infiniband/sw/rdmavt/srq.c b/drivers/infiniband/sw/rdmavt/srq.c index 78e06fc456c5..895b3fabd0bf 100644 --- a/drivers/infiniband/sw/rdmavt/srq.c +++ b/drivers/infiniband/sw/rdmavt/srq.c @@ -48,6 +48,7 @@ #include <linux/err.h> #include <linux/slab.h> #include <linux/vmalloc.h> +#include <rdma/uverbs_ioctl.h> #include "srq.h" #include "vt.h" @@ -77,6 +78,8 @@ struct ib_srq *rvt_create_srq(struct ib_pd *ibpd, struct ib_udata *udata) { struct rvt_dev_info *dev = ib_to_rvt(ibpd->device); + struct rvt_ucontext *ucontext = rdma_udata_to_drv_context( + udata, struct rvt_ucontext, ibucontext); struct rvt_srq *srq; u32 sz; struct ib_srq *ret; @@ -119,7 +122,7 @@ struct ib_srq *rvt_create_srq(struct ib_pd *ibpd, u32 s = sizeof(struct rvt_rwq) + srq->rq.size * sz; srq->ip = - rvt_create_mmap_info(dev, s, ibpd->uobject->context, + rvt_create_mmap_info(dev, s, &ucontext->ibucontext, srq->rq.wq); if (!srq->ip) { ret = ERR_PTR(-ENOMEM); diff --git a/drivers/infiniband/sw/rdmavt/trace_cq.h b/drivers/infiniband/sw/rdmavt/trace_cq.h index df8e1adbef9d..e3c416c6f900 100644 --- a/drivers/infiniband/sw/rdmavt/trace_cq.h +++ b/drivers/infiniband/sw/rdmavt/trace_cq.h @@ -105,7 +105,7 @@ DEFINE_EVENT(rvt_cq_template, rvt_create_cq, TP_ARGS(cq, attr)); #define CQ_PRN \ -"[%s] idx %u wr_id %llx status %u opcode %u,%s length %u qpn %x" +"[%s] idx %u wr_id %llx status %u opcode %u,%s length %u qpn %x flags %x imm %x" DECLARE_EVENT_CLASS( rvt_cq_entry_template, @@ -119,6 +119,8 @@ DECLARE_EVENT_CLASS( __field(u32, qpn) __field(u32, length) __field(u32, idx) + __field(u32, flags) + __field(u32, imm) ), TP_fast_assign( RDI_DEV_ASSIGN(cq->rdi) @@ -128,6 +130,8 @@ DECLARE_EVENT_CLASS( __entry->length = wc->byte_len; __entry->qpn = wc->qp->qp_num; __entry->idx = idx; + __entry->flags = wc->wc_flags; + __entry->imm = be32_to_cpu(wc->ex.imm_data); ), TP_printk( CQ_PRN, @@ -137,7 +141,9 @@ DECLARE_EVENT_CLASS( __entry->status, __entry->opcode, show_wc_opcode(__entry->opcode), __entry->length, - __entry->qpn + __entry->qpn, + __entry->flags, + __entry->imm ) ); diff --git a/drivers/infiniband/sw/rdmavt/vt.c b/drivers/infiniband/sw/rdmavt/vt.c index aef3aa3fe667..42c9d35f832d 100644 --- a/drivers/infiniband/sw/rdmavt/vt.c +++ b/drivers/infiniband/sw/rdmavt/vt.c @@ -91,7 +91,7 @@ struct rvt_dev_info *rvt_alloc_device(size_t size, int nports) { struct rvt_dev_info *rdi; - rdi = (struct rvt_dev_info *)ib_alloc_device(size); + rdi = container_of(_ib_alloc_device(size), struct rvt_dev_info, ibdev); if (!rdi) return rdi; @@ -284,10 +284,6 @@ static int rvt_query_gid(struct ib_device *ibdev, u8 port_num, &gid->global.interface_id); } -struct rvt_ucontext { - struct ib_ucontext ibucontext; -}; - static inline struct rvt_ucontext *to_iucontext(struct ib_ucontext *ibucontext) { @@ -296,28 +292,21 @@ static inline struct rvt_ucontext *to_iucontext(struct ib_ucontext /** * rvt_alloc_ucontext - Allocate a user context - * @ibdev: Verbs IB dev + * @uctx: Verbs context * @udata: User data allocated */ -static struct ib_ucontext *rvt_alloc_ucontext(struct ib_device *ibdev, - struct ib_udata *udata) +static int rvt_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { - struct rvt_ucontext *context; - - context = kmalloc(sizeof(*context), GFP_KERNEL); - if (!context) - return ERR_PTR(-ENOMEM); - return &context->ibucontext; + return 0; } /** - *rvt_dealloc_ucontext - Free a user context - *@context - Free this + * rvt_dealloc_ucontext - Free a user context + * @context - Free this */ -static int rvt_dealloc_ucontext(struct ib_ucontext *context) +static void rvt_dealloc_ucontext(struct ib_ucontext *context) { - kfree(to_iucontext(context)); - return 0; + return; } static int rvt_get_port_immutable(struct ib_device *ibdev, u8 port_num, @@ -436,6 +425,8 @@ static const struct ib_device_ops rvt_dev_ops = { .req_notify_cq = rvt_req_notify_cq, .resize_cq = rvt_resize_cq, .unmap_fmr = rvt_unmap_fmr, + INIT_RDMA_OBJ_SIZE(ib_pd, rvt_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, rvt_ucontext, ibucontext), }; static noinline int check_support(struct rvt_dev_info *rdi, int verb) @@ -446,7 +437,7 @@ static noinline int check_support(struct rvt_dev_info *rdi, int verb) * These functions are not part of verbs specifically but are * required for rdmavt to function. */ - if ((!rdi->driver_f.port_callback) || + if ((!rdi->ibdev.ops.init_port) || (!rdi->driver_f.get_pci_dev)) return -EINVAL; break; @@ -644,8 +635,7 @@ int rvt_register_device(struct rvt_dev_info *rdi, u32 driver_id) rdi->ibdev.driver_id = driver_id; /* We are now good to announce we exist */ - ret = ib_register_device(&rdi->ibdev, dev_name(&rdi->ibdev.dev), - rdi->driver_f.port_callback); + ret = ib_register_device(&rdi->ibdev, dev_name(&rdi->ibdev.dev)); if (ret) { rvt_pr_err(rdi, "Failed to register driver with ib core.\n"); goto bail_wss; diff --git a/drivers/infiniband/sw/rxe/rxe.c b/drivers/infiniband/sw/rxe/rxe.c index 383e65c7bbc0..a8c11b5e1e94 100644 --- a/drivers/infiniband/sw/rxe/rxe.c +++ b/drivers/infiniband/sw/rxe/rxe.c @@ -31,6 +31,7 @@ * SOFTWARE. */ +#include <rdma/rdma_netlink.h> #include <net/addrconf.h> #include "rxe.h" #include "rxe_loc.h" @@ -50,8 +51,10 @@ static void rxe_cleanup_ports(struct rxe_dev *rxe) /* free resources for a rxe device all objects created for this device must * have been destroyed */ -static void rxe_cleanup(struct rxe_dev *rxe) +void rxe_dealloc(struct ib_device *ib_dev) { + struct rxe_dev *rxe = container_of(ib_dev, struct rxe_dev, ib_dev); + rxe_pool_cleanup(&rxe->uc_pool); rxe_pool_cleanup(&rxe->pd_pool); rxe_pool_cleanup(&rxe->ah_pool); @@ -65,16 +68,8 @@ static void rxe_cleanup(struct rxe_dev *rxe) rxe_cleanup_ports(rxe); - crypto_free_shash(rxe->tfm); -} - -/* called when all references have been dropped */ -void rxe_release(struct kref *kref) -{ - struct rxe_dev *rxe = container_of(kref, struct rxe_dev, ref_cnt); - - rxe_cleanup(rxe); - ib_dealloc_device(&rxe->ib_dev); + if (rxe->tfm) + crypto_free_shash(rxe->tfm); } /* initialize rxe device parameters */ @@ -279,7 +274,6 @@ static int rxe_init(struct rxe_dev *rxe) spin_lock_init(&rxe->mmap_offset_lock); spin_lock_init(&rxe->pending_lock); INIT_LIST_HEAD(&rxe->pending_mmaps); - INIT_LIST_HEAD(&rxe->list); mutex_init(&rxe->usdev_lock); @@ -308,37 +302,46 @@ void rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu) /* called by ifc layer to create new rxe device. * The caller should allocate memory for rxe by calling ib_alloc_device. */ -int rxe_add(struct rxe_dev *rxe, unsigned int mtu) +int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name) { int err; - kref_init(&rxe->ref_cnt); - err = rxe_init(rxe); if (err) - goto err1; + return err; rxe_set_mtu(rxe, mtu); - err = rxe_register_device(rxe); - if (err) - goto err1; - - return 0; - -err1: - rxe_dev_put(rxe); - return err; + return rxe_register_device(rxe, ibdev_name); } -/* called by the ifc layer to remove a device */ -void rxe_remove(struct rxe_dev *rxe) +static int rxe_newlink(const char *ibdev_name, struct net_device *ndev) { - rxe_unregister_device(rxe); + struct rxe_dev *exists; + int err = 0; + + exists = rxe_get_dev_from_net(ndev); + if (exists) { + ib_device_put(&exists->ib_dev); + pr_err("already configured on %s\n", ndev->name); + err = -EEXIST; + goto err; + } - rxe_dev_put(rxe); + err = rxe_net_add(ibdev_name, ndev); + if (err) { + pr_err("failed to add %s\n", ndev->name); + goto err; + } +err: + return err; } +static struct rdma_link_ops rxe_link_ops = { + .type = "rxe", + .newlink = rxe_newlink, +}; + static int __init rxe_module_init(void) { int err; @@ -354,13 +357,15 @@ static int __init rxe_module_init(void) if (err) return err; + rdma_link_register(&rxe_link_ops); pr_info("loaded\n"); return 0; } static void __exit rxe_module_exit(void) { - rxe_remove_all(); + rdma_link_unregister(&rxe_link_ops); + ib_unregister_driver(RDMA_DRIVER_RXE); rxe_net_exit(); rxe_cache_exit(); @@ -369,3 +374,5 @@ static void __exit rxe_module_exit(void) late_initcall(rxe_module_init); module_exit(rxe_module_exit); + +MODULE_ALIAS_RDMA_LINK("rxe"); diff --git a/drivers/infiniband/sw/rxe/rxe.h b/drivers/infiniband/sw/rxe/rxe.h index 5bde2ad964d2..2e2dff478833 100644 --- a/drivers/infiniband/sw/rxe/rxe.h +++ b/drivers/infiniband/sw/rxe/rxe.h @@ -95,18 +95,20 @@ static inline u32 rxe_crc32(struct rxe_dev *rxe, void rxe_set_mtu(struct rxe_dev *rxe, unsigned int dev_mtu); -int rxe_add(struct rxe_dev *rxe, unsigned int mtu); -void rxe_remove(struct rxe_dev *rxe); -void rxe_remove_all(void); +int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name); void rxe_rcv(struct sk_buff *skb); -static inline void rxe_dev_put(struct rxe_dev *rxe) +/* The caller must do a matching ib_device_put(&dev->ib_dev) */ +static inline struct rxe_dev *rxe_get_dev_from_net(struct net_device *ndev) { - kref_put(&rxe->ref_cnt, rxe_release); + struct ib_device *ibdev = + ib_device_get_by_netdev(ndev, RDMA_DRIVER_RXE); + + if (!ibdev) + return NULL; + return container_of(ibdev, struct rxe_dev, ib_dev); } -struct rxe_dev *net_to_rxe(struct net_device *ndev); -struct rxe_dev *get_rxe_by_name(const char *name); void rxe_port_up(struct rxe_dev *rxe); void rxe_port_down(struct rxe_dev *rxe); diff --git a/drivers/infiniband/sw/rxe/rxe_av.c b/drivers/infiniband/sw/rxe/rxe_av.c index 26fe8d7dbc55..81ee756c19b8 100644 --- a/drivers/infiniband/sw/rxe/rxe_av.c +++ b/drivers/infiniband/sw/rxe/rxe_av.c @@ -34,6 +34,13 @@ #include "rxe.h" #include "rxe_loc.h" +void rxe_init_av(struct rdma_ah_attr *attr, struct rxe_av *av) +{ + rxe_av_from_attr(rdma_ah_get_port_num(attr), av, attr); + rxe_av_fill_ip_info(av, attr); + memcpy(av->dmac, attr->roce.dmac, ETH_ALEN); +} + int rxe_av_chk_attr(struct rxe_dev *rxe, struct rdma_ah_attr *attr) { struct rxe_port *port; diff --git a/drivers/infiniband/sw/rxe/rxe_comp.c b/drivers/infiniband/sw/rxe/rxe_comp.c index e996da67a851..00eb99d3df86 100644 --- a/drivers/infiniband/sw/rxe/rxe_comp.c +++ b/drivers/infiniband/sw/rxe/rxe_comp.c @@ -146,8 +146,7 @@ void retransmit_timer(struct timer_list *t) } } -void rxe_comp_queue_pkt(struct rxe_dev *rxe, struct rxe_qp *qp, - struct sk_buff *skb) +void rxe_comp_queue_pkt(struct rxe_qp *qp, struct sk_buff *skb) { int must_sched; @@ -155,7 +154,8 @@ void rxe_comp_queue_pkt(struct rxe_dev *rxe, struct rxe_qp *qp, must_sched = skb_queue_len(&qp->resp_pkts) > 1; if (must_sched != 0) - rxe_counter_inc(rxe, RXE_CNT_COMPLETER_SCHED); + rxe_counter_inc(SKB_TO_PKT(skb)->rxe, RXE_CNT_COMPLETER_SCHED); + rxe_run_task(&qp->comp.task, must_sched); } diff --git a/drivers/infiniband/sw/rxe/rxe_loc.h b/drivers/infiniband/sw/rxe/rxe_loc.h index 01b74597b36a..3d8cef836f0d 100644 --- a/drivers/infiniband/sw/rxe/rxe_loc.h +++ b/drivers/infiniband/sw/rxe/rxe_loc.h @@ -35,6 +35,7 @@ #define RXE_LOC_H /* rxe_av.c */ +void rxe_init_av(struct rdma_ah_attr *attr, struct rxe_av *av); int rxe_av_chk_attr(struct rxe_dev *rxe, struct rdma_ah_attr *attr); @@ -231,7 +232,7 @@ int rxe_srq_from_attr(struct rxe_dev *rxe, struct rxe_srq *srq, struct ib_srq_attr *attr, enum ib_srq_attr_mask mask, struct rxe_modify_srq_cmd *ucmd); -void rxe_release(struct kref *kref); +void rxe_dealloc(struct ib_device *ib_dev); int rxe_completer(void *arg); int rxe_requester(void *arg); @@ -239,11 +240,9 @@ int rxe_responder(void *arg); u32 rxe_icrc_hdr(struct rxe_pkt_info *pkt, struct sk_buff *skb); -void rxe_resp_queue_pkt(struct rxe_dev *rxe, - struct rxe_qp *qp, struct sk_buff *skb); +void rxe_resp_queue_pkt(struct rxe_qp *qp, struct sk_buff *skb); -void rxe_comp_queue_pkt(struct rxe_dev *rxe, - struct rxe_qp *qp, struct sk_buff *skb); +void rxe_comp_queue_pkt(struct rxe_qp *qp, struct sk_buff *skb); static inline unsigned int wr_opcode_mask(int opcode, struct rxe_qp *qp) { diff --git a/drivers/infiniband/sw/rxe/rxe_mr.c b/drivers/infiniband/sw/rxe/rxe_mr.c index 9d3916b93f23..42f0f25e396c 100644 --- a/drivers/infiniband/sw/rxe/rxe_mr.c +++ b/drivers/infiniband/sw/rxe/rxe_mr.c @@ -162,16 +162,15 @@ int rxe_mem_init_user(struct rxe_pd *pd, u64 start, u64 length, u64 iova, int access, struct ib_udata *udata, struct rxe_mem *mem) { - int entry; struct rxe_map **map; struct rxe_phys_buf *buf = NULL; struct ib_umem *umem; - struct scatterlist *sg; + struct sg_page_iter sg_iter; int num_buf; void *vaddr; int err; - umem = ib_umem_get(pd->ibpd.uobject->context, start, length, access, 0); + umem = ib_umem_get(udata, start, length, access, 0); if (IS_ERR(umem)) { pr_warn("err %d from rxe_umem_get\n", (int)PTR_ERR(umem)); @@ -191,16 +190,16 @@ int rxe_mem_init_user(struct rxe_pd *pd, u64 start, goto err1; } - mem->page_shift = umem->page_shift; - mem->page_mask = BIT(umem->page_shift) - 1; + mem->page_shift = PAGE_SHIFT; + mem->page_mask = PAGE_SIZE - 1; num_buf = 0; map = mem->map; if (length > 0) { buf = map[0]->buf; - for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { - vaddr = page_address(sg_page(sg)); + for_each_sg_page(umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + vaddr = page_address(sg_page_iter_page(&sg_iter)); if (!vaddr) { pr_warn("null vaddr\n"); err = -ENOMEM; @@ -208,7 +207,7 @@ int rxe_mem_init_user(struct rxe_pd *pd, u64 start, } buf->addr = (uintptr_t)vaddr; - buf->size = BIT(umem->page_shift); + buf->size = PAGE_SIZE; num_buf++; buf++; diff --git a/drivers/infiniband/sw/rxe/rxe_net.c b/drivers/infiniband/sw/rxe/rxe_net.c index 8fd03ae20efc..753cabcd441c 100644 --- a/drivers/infiniband/sw/rxe/rxe_net.c +++ b/drivers/infiniband/sw/rxe/rxe_net.c @@ -45,43 +45,6 @@ #include "rxe_net.h" #include "rxe_loc.h" -static LIST_HEAD(rxe_dev_list); -static DEFINE_SPINLOCK(dev_list_lock); /* spinlock for device list */ - -struct rxe_dev *net_to_rxe(struct net_device *ndev) -{ - struct rxe_dev *rxe; - struct rxe_dev *found = NULL; - - spin_lock_bh(&dev_list_lock); - list_for_each_entry(rxe, &rxe_dev_list, list) { - if (rxe->ndev == ndev) { - found = rxe; - break; - } - } - spin_unlock_bh(&dev_list_lock); - - return found; -} - -struct rxe_dev *get_rxe_by_name(const char *name) -{ - struct rxe_dev *rxe; - struct rxe_dev *found = NULL; - - spin_lock_bh(&dev_list_lock); - list_for_each_entry(rxe, &rxe_dev_list, list) { - if (!strcmp(name, dev_name(&rxe->ib_dev.dev))) { - found = rxe; - break; - } - } - spin_unlock_bh(&dev_list_lock); - return found; -} - - static struct rxe_recv_sockets recv_sockets; struct device *rxe_dma_device(struct rxe_dev *rxe) @@ -229,18 +192,19 @@ static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb) struct udphdr *udph; struct net_device *ndev = skb->dev; struct net_device *rdev = ndev; - struct rxe_dev *rxe = net_to_rxe(ndev); + struct rxe_dev *rxe = rxe_get_dev_from_net(ndev); struct rxe_pkt_info *pkt = SKB_TO_PKT(skb); if (!rxe && is_vlan_dev(rdev)) { rdev = vlan_dev_real_dev(ndev); - rxe = net_to_rxe(rdev); + rxe = rxe_get_dev_from_net(rdev); } if (!rxe) goto drop; if (skb_linearize(skb)) { pr_err("skb_linearize failed\n"); + ib_device_put(&rxe->ib_dev); goto drop; } @@ -253,6 +217,12 @@ static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb) rxe_rcv(skb); + /* + * FIXME: this is in the wrong place, it needs to be done when pkt is + * destroyed + */ + ib_device_put(&rxe->ib_dev); + return 0; drop: kfree_skb(skb); @@ -384,9 +354,6 @@ static int prepare4(struct rxe_pkt_info *pkt, struct sk_buff *skb, return -EHOSTUNREACH; } - if (!memcmp(saddr, daddr, sizeof(*daddr))) - pkt->mask |= RXE_LOOPBACK_MASK; - prepare_udp_hdr(skb, cpu_to_be16(qp->src_port), cpu_to_be16(ROCE_V2_UDP_DPORT)); @@ -411,9 +378,6 @@ static int prepare6(struct rxe_pkt_info *pkt, struct sk_buff *skb, return -EHOSTUNREACH; } - if (!memcmp(saddr, daddr, sizeof(*daddr))) - pkt->mask |= RXE_LOOPBACK_MASK; - prepare_udp_hdr(skb, cpu_to_be16(qp->src_port), cpu_to_be16(ROCE_V2_UDP_DPORT)); @@ -437,6 +401,9 @@ int rxe_prepare(struct rxe_pkt_info *pkt, struct sk_buff *skb, u32 *crc) *crc = rxe_icrc_hdr(pkt, skb); + if (ether_addr_equal(skb->dev->dev_addr, av->dmac)) + pkt->mask |= RXE_LOOPBACK_MASK; + return err; } @@ -550,42 +517,24 @@ enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num) return IB_LINK_LAYER_ETHERNET; } -struct rxe_dev *rxe_net_add(struct net_device *ndev) +int rxe_net_add(const char *ibdev_name, struct net_device *ndev) { int err; struct rxe_dev *rxe = NULL; - rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe)); + rxe = ib_alloc_device(rxe_dev, ib_dev); if (!rxe) - return NULL; + return -ENOMEM; rxe->ndev = ndev; - err = rxe_add(rxe, ndev->mtu); + err = rxe_add(rxe, ndev->mtu, ibdev_name); if (err) { ib_dealloc_device(&rxe->ib_dev); - return NULL; + return err; } - spin_lock_bh(&dev_list_lock); - list_add_tail(&rxe->list, &rxe_dev_list); - spin_unlock_bh(&dev_list_lock); - return rxe; -} - -void rxe_remove_all(void) -{ - spin_lock_bh(&dev_list_lock); - while (!list_empty(&rxe_dev_list)) { - struct rxe_dev *rxe = - list_first_entry(&rxe_dev_list, struct rxe_dev, list); - - list_del(&rxe->list); - spin_unlock_bh(&dev_list_lock); - rxe_remove(rxe); - spin_lock_bh(&dev_list_lock); - } - spin_unlock_bh(&dev_list_lock); + return 0; } static void rxe_port_event(struct rxe_dev *rxe, @@ -638,15 +587,14 @@ static int rxe_notify(struct notifier_block *not_blk, void *arg) { struct net_device *ndev = netdev_notifier_info_to_dev(arg); - struct rxe_dev *rxe = net_to_rxe(ndev); + struct rxe_dev *rxe = rxe_get_dev_from_net(ndev); if (!rxe) - goto out; + return NOTIFY_OK; switch (event) { case NETDEV_UNREGISTER: - list_del(&rxe->list); - rxe_remove(rxe); + ib_unregister_device_queued(&rxe->ib_dev); break; case NETDEV_UP: rxe_port_up(rxe); @@ -671,7 +619,8 @@ static int rxe_notify(struct notifier_block *not_blk, event, ndev->name); break; } -out: + + ib_device_put(&rxe->ib_dev); return NOTIFY_OK; } diff --git a/drivers/infiniband/sw/rxe/rxe_net.h b/drivers/infiniband/sw/rxe/rxe_net.h index 106c586dbb26..2ca71d3d245c 100644 --- a/drivers/infiniband/sw/rxe/rxe_net.h +++ b/drivers/infiniband/sw/rxe/rxe_net.h @@ -43,7 +43,7 @@ struct rxe_recv_sockets { struct socket *sk6; }; -struct rxe_dev *rxe_net_add(struct net_device *ndev); +int rxe_net_add(const char *ibdev_name, struct net_device *ndev); int rxe_net_init(void); void rxe_net_exit(void); diff --git a/drivers/infiniband/sw/rxe/rxe_param.h b/drivers/infiniband/sw/rxe/rxe_param.h index bdea899a58ac..1abed47ca221 100644 --- a/drivers/infiniband/sw/rxe/rxe_param.h +++ b/drivers/infiniband/sw/rxe/rxe_param.h @@ -78,7 +78,8 @@ enum rxe_device_param { | IB_DEVICE_SYS_IMAGE_GUID | IB_DEVICE_RC_RNR_NAK_GEN | IB_DEVICE_SRQ_RESIZE - | IB_DEVICE_MEM_MGT_EXTENSIONS, + | IB_DEVICE_MEM_MGT_EXTENSIONS + | IB_DEVICE_ALLOW_USER_UNREG, RXE_MAX_SGE = 32, RXE_MAX_SGE_RD = 32, RXE_MAX_CQ = 16384, diff --git a/drivers/infiniband/sw/rxe/rxe_pool.c b/drivers/infiniband/sw/rxe/rxe_pool.c index b5c91df22047..120fa9005954 100644 --- a/drivers/infiniband/sw/rxe/rxe_pool.c +++ b/drivers/infiniband/sw/rxe/rxe_pool.c @@ -42,10 +42,12 @@ struct rxe_type_info rxe_type_info[RXE_NUM_TYPES] = { [RXE_TYPE_UC] = { .name = "rxe-uc", .size = sizeof(struct rxe_ucontext), + .flags = RXE_POOL_NO_ALLOC, }, [RXE_TYPE_PD] = { .name = "rxe-pd", .size = sizeof(struct rxe_pd), + .flags = RXE_POOL_NO_ALLOC, }, [RXE_TYPE_AH] = { .name = "rxe-ah", @@ -119,8 +121,10 @@ static void rxe_cache_clean(size_t cnt) for (i = 0; i < cnt; i++) { type = &rxe_type_info[i]; - kmem_cache_destroy(type->cache); - type->cache = NULL; + if (!(type->flags & RXE_POOL_NO_ALLOC)) { + kmem_cache_destroy(type->cache); + type->cache = NULL; + } } } @@ -134,14 +138,17 @@ int rxe_cache_init(void) for (i = 0; i < RXE_NUM_TYPES; i++) { type = &rxe_type_info[i]; size = ALIGN(type->size, RXE_POOL_ALIGN); - type->cache = kmem_cache_create(type->name, size, - RXE_POOL_ALIGN, - RXE_POOL_CACHE_FLAGS, NULL); - if (!type->cache) { - pr_err("Unable to init kmem cache for %s\n", - type->name); - err = -ENOMEM; - goto err1; + if (!(type->flags & RXE_POOL_NO_ALLOC)) { + type->cache = + kmem_cache_create(type->name, size, + RXE_POOL_ALIGN, + RXE_POOL_CACHE_FLAGS, NULL); + if (!type->cache) { + pr_err("Unable to init kmem cache for %s\n", + type->name); + err = -ENOMEM; + goto err1; + } } } @@ -392,29 +399,64 @@ void *rxe_alloc(struct rxe_pool *pool) kref_get(&pool->ref_cnt); read_unlock_irqrestore(&pool->pool_lock, flags); - kref_get(&pool->rxe->ref_cnt); + if (!ib_device_try_get(&pool->rxe->ib_dev)) + goto out_put_pool; if (atomic_inc_return(&pool->num_elem) > pool->max_elem) - goto out_put_pool; + goto out_cnt; elem = kmem_cache_zalloc(pool_cache(pool), (pool->flags & RXE_POOL_ATOMIC) ? GFP_ATOMIC : GFP_KERNEL); if (!elem) - goto out_put_pool; + goto out_cnt; elem->pool = pool; kref_init(&elem->ref_cnt); return elem; -out_put_pool: +out_cnt: atomic_dec(&pool->num_elem); - rxe_dev_put(pool->rxe); + ib_device_put(&pool->rxe->ib_dev); +out_put_pool: rxe_pool_put(pool); return NULL; } +int rxe_add_to_pool(struct rxe_pool *pool, struct rxe_pool_entry *elem) +{ + unsigned long flags; + + might_sleep_if(!(pool->flags & RXE_POOL_ATOMIC)); + + read_lock_irqsave(&pool->pool_lock, flags); + if (pool->state != RXE_POOL_STATE_VALID) { + read_unlock_irqrestore(&pool->pool_lock, flags); + return -EINVAL; + } + kref_get(&pool->ref_cnt); + read_unlock_irqrestore(&pool->pool_lock, flags); + + if (!ib_device_try_get(&pool->rxe->ib_dev)) + goto out_put_pool; + + if (atomic_inc_return(&pool->num_elem) > pool->max_elem) + goto out_cnt; + + elem->pool = pool; + kref_init(&elem->ref_cnt); + + return 0; + +out_cnt: + atomic_dec(&pool->num_elem); + ib_device_put(&pool->rxe->ib_dev); +out_put_pool: + rxe_pool_put(pool); + return -EINVAL; +} + void rxe_elem_release(struct kref *kref) { struct rxe_pool_entry *elem = @@ -424,9 +466,10 @@ void rxe_elem_release(struct kref *kref) if (pool->cleanup) pool->cleanup(elem); - kmem_cache_free(pool_cache(pool), elem); + if (!(pool->flags & RXE_POOL_NO_ALLOC)) + kmem_cache_free(pool_cache(pool), elem); atomic_dec(&pool->num_elem); - rxe_dev_put(pool->rxe); + ib_device_put(&pool->rxe->ib_dev); rxe_pool_put(pool); } diff --git a/drivers/infiniband/sw/rxe/rxe_pool.h b/drivers/infiniband/sw/rxe/rxe_pool.h index 72968c29e01f..2f2cff1cbe43 100644 --- a/drivers/infiniband/sw/rxe/rxe_pool.h +++ b/drivers/infiniband/sw/rxe/rxe_pool.h @@ -41,6 +41,7 @@ enum rxe_pool_flags { RXE_POOL_ATOMIC = BIT(0), RXE_POOL_INDEX = BIT(1), RXE_POOL_KEY = BIT(2), + RXE_POOL_NO_ALLOC = BIT(4), }; enum rxe_elem_type { @@ -131,6 +132,9 @@ void rxe_pool_cleanup(struct rxe_pool *pool); /* allocate an object from pool */ void *rxe_alloc(struct rxe_pool *pool); +/* connect already allocated object to pool */ +int rxe_add_to_pool(struct rxe_pool *pool, struct rxe_pool_entry *elem); + /* assign an index to an indexed object and insert object into * pool's rb tree */ diff --git a/drivers/infiniband/sw/rxe/rxe_qp.c b/drivers/infiniband/sw/rxe/rxe_qp.c index fd86fd2fbb26..09ede70dc1e8 100644 --- a/drivers/infiniband/sw/rxe/rxe_qp.c +++ b/drivers/infiniband/sw/rxe/rxe_qp.c @@ -35,6 +35,7 @@ #include <linux/delay.h> #include <linux/sched.h> #include <linux/vmalloc.h> +#include <rdma/uverbs_ioctl.h> #include "rxe.h" #include "rxe_loc.h" @@ -343,7 +344,8 @@ int rxe_qp_from_init(struct rxe_dev *rxe, struct rxe_qp *qp, struct rxe_pd *pd, struct rxe_cq *rcq = to_rcq(init->recv_cq); struct rxe_cq *scq = to_rcq(init->send_cq); struct rxe_srq *srq = init->srq ? to_rsrq(init->srq) : NULL; - struct ib_ucontext *context = udata ? ibpd->uobject->context : NULL; + struct rxe_ucontext *ucontext = + rdma_udata_to_drv_context(udata, struct rxe_ucontext, ibuc); rxe_add_ref(pd); rxe_add_ref(rcq); @@ -358,11 +360,11 @@ int rxe_qp_from_init(struct rxe_dev *rxe, struct rxe_qp *qp, struct rxe_pd *pd, rxe_qp_init_misc(rxe, qp, init); - err = rxe_qp_init_req(rxe, qp, init, context, uresp); + err = rxe_qp_init_req(rxe, qp, init, &ucontext->ibuc, uresp); if (err) goto err1; - err = rxe_qp_init_resp(rxe, qp, init, context, uresp); + err = rxe_qp_init_resp(rxe, qp, init, &ucontext->ibuc, uresp); if (err) goto err2; @@ -631,14 +633,11 @@ int rxe_qp_from_attr(struct rxe_qp *qp, struct ib_qp_attr *attr, int mask, qp->attr.qkey = attr->qkey; if (mask & IB_QP_AV) { - rxe_av_from_attr(attr->port_num, &qp->pri_av, &attr->ah_attr); - rxe_av_fill_ip_info(&qp->pri_av, &attr->ah_attr); + rxe_init_av(&attr->ah_attr, &qp->pri_av); } if (mask & IB_QP_ALT_PATH) { - rxe_av_from_attr(attr->alt_port_num, &qp->alt_av, - &attr->alt_ah_attr); - rxe_av_fill_ip_info(&qp->alt_av, &attr->alt_ah_attr); + rxe_init_av(&attr->alt_ah_attr, &qp->alt_av); qp->attr.alt_port_num = attr->alt_port_num; qp->attr.alt_pkey_index = attr->alt_pkey_index; qp->attr.alt_timeout = attr->alt_timeout; diff --git a/drivers/infiniband/sw/rxe/rxe_recv.c b/drivers/infiniband/sw/rxe/rxe_recv.c index 5c29a1bb575a..f9a492ed900b 100644 --- a/drivers/infiniband/sw/rxe/rxe_recv.c +++ b/drivers/infiniband/sw/rxe/rxe_recv.c @@ -266,14 +266,12 @@ err1: return -EINVAL; } -static inline void rxe_rcv_pkt(struct rxe_dev *rxe, - struct rxe_pkt_info *pkt, - struct sk_buff *skb) +static inline void rxe_rcv_pkt(struct rxe_pkt_info *pkt, struct sk_buff *skb) { if (pkt->mask & RXE_REQ_MASK) - rxe_resp_queue_pkt(rxe, pkt->qp, skb); + rxe_resp_queue_pkt(pkt->qp, skb); else - rxe_comp_queue_pkt(rxe, pkt->qp, skb); + rxe_comp_queue_pkt(pkt->qp, skb); } static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb) @@ -319,7 +317,7 @@ static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb) pkt->qp = qp; rxe_add_ref(qp); - rxe_rcv_pkt(rxe, pkt, skb); + rxe_rcv_pkt(pkt, skb); } spin_unlock_bh(&mcg->mcg_lock); @@ -411,7 +409,7 @@ void rxe_rcv(struct sk_buff *skb) if (unlikely(bth_qpn(pkt) == IB_MULTICAST_QPN)) rxe_rcv_mcast_pkt(rxe, skb); else - rxe_rcv_pkt(rxe, pkt, skb); + rxe_rcv_pkt(pkt, skb); return; diff --git a/drivers/infiniband/sw/rxe/rxe_resp.c b/drivers/infiniband/sw/rxe/rxe_resp.c index 231528188250..aca9f60f9b21 100644 --- a/drivers/infiniband/sw/rxe/rxe_resp.c +++ b/drivers/infiniband/sw/rxe/rxe_resp.c @@ -104,8 +104,7 @@ static char *resp_state_name[] = { }; /* rxe_recv calls here to add a request packet to the input queue */ -void rxe_resp_queue_pkt(struct rxe_dev *rxe, struct rxe_qp *qp, - struct sk_buff *skb) +void rxe_resp_queue_pkt(struct rxe_qp *qp, struct sk_buff *skb) { int must_sched; struct rxe_pkt_info *pkt = SKB_TO_PKT(skb); diff --git a/drivers/infiniband/sw/rxe/rxe_sysfs.c b/drivers/infiniband/sw/rxe/rxe_sysfs.c index 95a15892f7e6..ccda5f5a3bc0 100644 --- a/drivers/infiniband/sw/rxe/rxe_sysfs.c +++ b/drivers/infiniband/sw/rxe/rxe_sysfs.c @@ -58,41 +58,37 @@ static int rxe_param_set_add(const char *val, const struct kernel_param *kp) int len; int err = 0; char intf[32]; - struct net_device *ndev = NULL; - struct rxe_dev *rxe; + struct net_device *ndev; + struct rxe_dev *exists; len = sanitize_arg(val, intf, sizeof(intf)); if (!len) { pr_err("add: invalid interface name\n"); - err = -EINVAL; - goto err; + return -EINVAL; } ndev = dev_get_by_name(&init_net, intf); if (!ndev) { pr_err("interface %s not found\n", intf); - err = -EINVAL; - goto err; + return -EINVAL; } - if (net_to_rxe(ndev)) { + exists = rxe_get_dev_from_net(ndev); + if (exists) { + ib_device_put(&exists->ib_dev); pr_err("already configured on %s\n", intf); err = -EINVAL; goto err; } - rxe = rxe_net_add(ndev); - if (!rxe) { + err = rxe_net_add("rxe%d", ndev); + if (err) { pr_err("failed to add %s\n", intf); - err = -EINVAL; goto err; } - rxe_set_port_state(rxe); - dev_info(&rxe->ib_dev.dev, "added %s\n", intf); err: - if (ndev) - dev_put(ndev); + dev_put(ndev); return err; } @@ -100,7 +96,7 @@ static int rxe_param_set_remove(const char *val, const struct kernel_param *kp) { int len; char intf[32]; - struct rxe_dev *rxe; + struct ib_device *ib_dev; len = sanitize_arg(val, intf, sizeof(intf)); if (!len) { @@ -110,19 +106,17 @@ static int rxe_param_set_remove(const char *val, const struct kernel_param *kp) if (strncmp("all", intf, len) == 0) { pr_info("rxe_sys: remove all"); - rxe_remove_all(); + ib_unregister_driver(RDMA_DRIVER_RXE); return 0; } - rxe = get_rxe_by_name(intf); - - if (!rxe) { + ib_dev = ib_device_get_by_name(intf, RDMA_DRIVER_RXE); + if (!ib_dev) { pr_err("not configured on %s\n", intf); return -EINVAL; } - list_del(&rxe->list); - rxe_remove(rxe); + ib_unregister_device_and_put(ib_dev); return 0; } @@ -136,6 +130,6 @@ static const struct kernel_param_ops rxe_remove_ops = { }; module_param_cb(add, &rxe_add_ops, NULL, 0200); -MODULE_PARM_DESC(add, "Create RXE device over network interface"); +MODULE_PARM_DESC(add, "DEPRECATED. Create RXE device over network interface"); module_param_cb(remove, &rxe_remove_ops, NULL, 0200); -MODULE_PARM_DESC(remove, "Remove RXE device over network interface"); +MODULE_PARM_DESC(remove, "DEPRECATED. Remove RXE device over network interface"); diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.c b/drivers/infiniband/sw/rxe/rxe_verbs.c index b20e6e0415f5..6ecf28570ff0 100644 --- a/drivers/infiniband/sw/rxe/rxe_verbs.c +++ b/drivers/infiniband/sw/rxe/rxe_verbs.c @@ -33,6 +33,7 @@ #include <linux/dma-mapping.h> #include <net/addrconf.h> +#include <rdma/uverbs_ioctl.h> #include "rxe.h" #include "rxe_loc.h" #include "rxe_queue.h" @@ -79,19 +80,6 @@ static int rxe_query_port(struct ib_device *dev, return rc; } -static struct net_device *rxe_get_netdev(struct ib_device *device, - u8 port_num) -{ - struct rxe_dev *rxe = to_rdev(device); - - if (rxe->ndev) { - dev_hold(rxe->ndev); - return rxe->ndev; - } - - return NULL; -} - static int rxe_query_pkey(struct ib_device *device, u8 port_num, u16 index, u16 *pkey) { @@ -154,22 +142,19 @@ static enum rdma_link_layer rxe_get_link_layer(struct ib_device *dev, return rxe_link_layer(rxe, port_num); } -static struct ib_ucontext *rxe_alloc_ucontext(struct ib_device *dev, - struct ib_udata *udata) +static int rxe_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata) { - struct rxe_dev *rxe = to_rdev(dev); - struct rxe_ucontext *uc; + struct rxe_dev *rxe = to_rdev(uctx->device); + struct rxe_ucontext *uc = to_ruc(uctx); - uc = rxe_alloc(&rxe->uc_pool); - return uc ? &uc->ibuc : ERR_PTR(-ENOMEM); + return rxe_add_to_pool(&rxe->uc_pool, &uc->pelem); } -static int rxe_dealloc_ucontext(struct ib_ucontext *ibuc) +static void rxe_dealloc_ucontext(struct ib_ucontext *ibuc) { struct rxe_ucontext *uc = to_ruc(ibuc); rxe_drop_ref(uc); - return 0; } static int rxe_port_immutable(struct ib_device *dev, u8 port_num, @@ -191,30 +176,20 @@ static int rxe_port_immutable(struct ib_device *dev, u8 port_num, return 0; } -static struct ib_pd *rxe_alloc_pd(struct ib_device *dev, - struct ib_ucontext *context, - struct ib_udata *udata) +static int rxe_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context, + struct ib_udata *udata) { - struct rxe_dev *rxe = to_rdev(dev); - struct rxe_pd *pd; + struct rxe_dev *rxe = to_rdev(ibpd->device); + struct rxe_pd *pd = to_rpd(ibpd); - pd = rxe_alloc(&rxe->pd_pool); - return pd ? &pd->ibpd : ERR_PTR(-ENOMEM); + return rxe_add_to_pool(&rxe->pd_pool, &pd->pelem); } -static int rxe_dealloc_pd(struct ib_pd *ibpd) +static void rxe_dealloc_pd(struct ib_pd *ibpd) { struct rxe_pd *pd = to_rpd(ibpd); rxe_drop_ref(pd); - return 0; -} - -static void rxe_init_av(struct rxe_dev *rxe, struct rdma_ah_attr *attr, - struct rxe_av *av) -{ - rxe_av_from_attr(rdma_ah_get_port_num(attr), av, attr); - rxe_av_fill_ip_info(av, attr); } static struct ib_ah *rxe_create_ah(struct ib_pd *ibpd, @@ -239,7 +214,7 @@ static struct ib_ah *rxe_create_ah(struct ib_pd *ibpd, rxe_add_ref(pd); ah->pd = pd; - rxe_init_av(rxe, attr, &ah->av); + rxe_init_av(attr, &ah->av); return &ah->ibah; } @@ -253,7 +228,7 @@ static int rxe_modify_ah(struct ib_ah *ibah, struct rdma_ah_attr *attr) if (err) return err; - rxe_init_av(rxe, attr, &ah->av); + rxe_init_av(attr, &ah->av); return 0; } @@ -330,8 +305,9 @@ static struct ib_srq *rxe_create_srq(struct ib_pd *ibpd, int err; struct rxe_dev *rxe = to_rdev(ibpd->device); struct rxe_pd *pd = to_rpd(ibpd); + struct rxe_ucontext *ucontext = + rdma_udata_to_drv_context(udata, struct rxe_ucontext, ibuc); struct rxe_srq *srq; - struct ib_ucontext *context = udata ? ibpd->uobject->context : NULL; struct rxe_create_srq_resp __user *uresp = NULL; if (udata) { @@ -354,7 +330,7 @@ static struct ib_srq *rxe_create_srq(struct ib_pd *ibpd, rxe_add_ref(pd); srq->pd = pd; - err = rxe_srq_from_init(rxe, srq, init, context, uresp); + err = rxe_srq_from_init(rxe, srq, init, &ucontext->ibuc, uresp); if (err) goto err2; @@ -1129,8 +1105,8 @@ static int rxe_detach_mcast(struct ib_qp *ibqp, union ib_gid *mgid, u16 mlid) static ssize_t parent_show(struct device *device, struct device_attribute *attr, char *buf) { - struct rxe_dev *rxe = container_of(device, struct rxe_dev, - ib_dev.dev); + struct rxe_dev *rxe = + rdma_device_to_drv_device(device, struct rxe_dev, ib_dev); return snprintf(buf, 16, "%s\n", rxe_parent_name(rxe, 1)); } @@ -1146,6 +1122,15 @@ static const struct attribute_group rxe_attr_group = { .attrs = rxe_dev_attributes, }; +static int rxe_enable_driver(struct ib_device *ib_dev) +{ + struct rxe_dev *rxe = container_of(ib_dev, struct rxe_dev, ib_dev); + + rxe_set_port_state(rxe); + dev_info(&rxe->ib_dev.dev, "added %s\n", netdev_name(rxe->ndev)); + return 0; +} + static const struct ib_device_ops rxe_dev_ops = { .alloc_hw_stats = rxe_ib_alloc_hw_stats, .alloc_mr = rxe_alloc_mr, @@ -1156,6 +1141,7 @@ static const struct ib_device_ops rxe_dev_ops = { .create_cq = rxe_create_cq, .create_qp = rxe_create_qp, .create_srq = rxe_create_srq, + .dealloc_driver = rxe_dealloc, .dealloc_pd = rxe_dealloc_pd, .dealloc_ucontext = rxe_dealloc_ucontext, .dereg_mr = rxe_dereg_mr, @@ -1164,10 +1150,10 @@ static const struct ib_device_ops rxe_dev_ops = { .destroy_qp = rxe_destroy_qp, .destroy_srq = rxe_destroy_srq, .detach_mcast = rxe_detach_mcast, + .enable_driver = rxe_enable_driver, .get_dma_mr = rxe_get_dma_mr, .get_hw_stats = rxe_ib_get_hw_stats, .get_link_layer = rxe_get_link_layer, - .get_netdev = rxe_get_netdev, .get_port_immutable = rxe_port_immutable, .map_mr_sg = rxe_map_mr_sg, .mmap = rxe_mmap, @@ -1190,9 +1176,11 @@ static const struct ib_device_ops rxe_dev_ops = { .reg_user_mr = rxe_reg_user_mr, .req_notify_cq = rxe_req_notify_cq, .resize_cq = rxe_resize_cq, + INIT_RDMA_OBJ_SIZE(ib_pd, rxe_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, rxe_ucontext, ibuc), }; -int rxe_register_device(struct rxe_dev *rxe) +int rxe_register_device(struct rxe_dev *rxe, const char *ibdev_name) { int err; struct ib_device *dev = &rxe->ib_dev; @@ -1247,6 +1235,9 @@ int rxe_register_device(struct rxe_dev *rxe) ; ib_set_device_ops(dev, &rxe_dev_ops); + err = ib_device_set_netdev(&rxe->ib_dev, rxe->ndev, 1); + if (err) + return err; tfm = crypto_alloc_shash("crc32", 0, 0); if (IS_ERR(tfm)) { @@ -1258,23 +1249,13 @@ int rxe_register_device(struct rxe_dev *rxe) rdma_set_device_sysfs_group(dev, &rxe_attr_group); dev->driver_id = RDMA_DRIVER_RXE; - err = ib_register_device(dev, "rxe%d", NULL); - if (err) { + err = ib_register_device(dev, ibdev_name); + if (err) pr_warn("%s failed with error %d\n", __func__, err); - goto err1; - } - - return 0; - -err1: - crypto_free_shash(rxe->tfm); + /* + * Note that rxe may be invalid at this point if another thread + * unregistered it. + */ return err; } - -void rxe_unregister_device(struct rxe_dev *rxe) -{ - struct ib_device *dev = &rxe->ib_dev; - - ib_unregister_device(dev); -} diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.h b/drivers/infiniband/sw/rxe/rxe_verbs.h index 74e04801d34d..157e51aeb1e1 100644 --- a/drivers/infiniband/sw/rxe/rxe_verbs.h +++ b/drivers/infiniband/sw/rxe/rxe_verbs.h @@ -61,13 +61,13 @@ static inline int psn_compare(u32 psn_a, u32 psn_b) } struct rxe_ucontext { + struct ib_ucontext ibuc; struct rxe_pool_entry pelem; - struct ib_ucontext ibuc; }; struct rxe_pd { + struct ib_pd ibpd; struct rxe_pool_entry pelem; - struct ib_pd ibpd; }; struct rxe_ah { @@ -385,7 +385,6 @@ struct rxe_dev { struct ib_device_attr attr; int max_ucontext; int max_inline_data; - struct kref ref_cnt; struct mutex usdev_lock; struct net_device *ndev; @@ -412,7 +411,6 @@ struct rxe_dev { atomic64_t stats_counters[RXE_NUM_OF_COUNTERS]; struct rxe_port port; - struct list_head list; struct crypto_shash *tfm; }; @@ -466,8 +464,7 @@ static inline struct rxe_mem *to_rmw(struct ib_mw *mw) return mw ? container_of(mw, struct rxe_mem, ibmw) : NULL; } -int rxe_register_device(struct rxe_dev *rxe); -void rxe_unregister_device(struct rxe_dev *rxe); +int rxe_register_device(struct rxe_dev *rxe, const char *ibdev_name); void rxe_mc_cleanup(struct rxe_pool_entry *arg); diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h index 73e808c1e6ad..2aa3457a30ce 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib.h +++ b/drivers/infiniband/ulp/ipoib/ipoib.h @@ -780,12 +780,12 @@ static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *w #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG void ipoib_create_debug_files(struct net_device *dev); void ipoib_delete_debug_files(struct net_device *dev); -int ipoib_register_debugfs(void); +void ipoib_register_debugfs(void); void ipoib_unregister_debugfs(void); #else static inline void ipoib_create_debug_files(struct net_device *dev) { } static inline void ipoib_delete_debug_files(struct net_device *dev) { } -static inline int ipoib_register_debugfs(void) { return 0; } +static inline void ipoib_register_debugfs(void) { } static inline void ipoib_unregister_debugfs(void) { } #endif diff --git a/drivers/infiniband/ulp/ipoib/ipoib_fs.c b/drivers/infiniband/ulp/ipoib/ipoib_fs.c index 178488028734..64c19f6fa931 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_fs.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_fs.c @@ -267,14 +267,10 @@ void ipoib_create_debug_files(struct net_device *dev) snprintf(name, sizeof(name), "%s_mcg", dev->name); priv->mcg_dentry = debugfs_create_file(name, S_IFREG | S_IRUGO, ipoib_root, dev, &ipoib_mcg_fops); - if (!priv->mcg_dentry) - ipoib_warn(priv, "failed to create mcg debug file\n"); snprintf(name, sizeof(name), "%s_path", dev->name); priv->path_dentry = debugfs_create_file(name, S_IFREG | S_IRUGO, ipoib_root, dev, &ipoib_path_fops); - if (!priv->path_dentry) - ipoib_warn(priv, "failed to create path debug file\n"); } void ipoib_delete_debug_files(struct net_device *dev) @@ -286,10 +282,9 @@ void ipoib_delete_debug_files(struct net_device *dev) priv->mcg_dentry = priv->path_dentry = NULL; } -int ipoib_register_debugfs(void) +void ipoib_register_debugfs(void) { ipoib_root = debugfs_create_dir("ipoib", NULL); - return ipoib_root ? 0 : -ENOMEM; } void ipoib_unregister_debugfs(void) diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index d932f99201d1..48eda16db1a7 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -613,7 +613,7 @@ static void path_free(struct net_device *dev, struct ipoib_path *path) while ((skb = __skb_dequeue(&path->queue))) dev_kfree_skb_irq(skb); - ipoib_dbg(ipoib_priv(dev), "path_free\n"); + ipoib_dbg(ipoib_priv(dev), "%s\n", __func__); /* remove all neigh connected to this path */ ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw); @@ -1641,7 +1641,7 @@ static void ipoib_neigh_hash_uninit(struct net_device *dev) { struct ipoib_dev_priv *priv = ipoib_priv(dev); - ipoib_dbg(priv, "ipoib_neigh_hash_uninit\n"); + ipoib_dbg(priv, "%s\n", __func__); init_completion(&priv->ntbl.deleted); cancel_delayed_work_sync(&priv->neigh_reap_task); @@ -2411,7 +2411,7 @@ static ssize_t dev_id_show(struct device *dev, } static DEVICE_ATTR_RO(dev_id); -int ipoib_intercept_dev_id_attr(struct net_device *dev) +static int ipoib_intercept_dev_id_attr(struct net_device *dev) { device_remove_file(&dev->dev, &dev_attr_dev_id); return device_create_file(&dev->dev, &dev_attr_dev_id); @@ -2495,7 +2495,7 @@ static void ipoib_add_one(struct ib_device *device) struct list_head *dev_list; struct net_device *dev; struct ipoib_dev_priv *priv; - int p; + unsigned int p; int count = 0; dev_list = kmalloc(sizeof(*dev_list), GFP_KERNEL); @@ -2504,7 +2504,7 @@ static void ipoib_add_one(struct ib_device *device) INIT_LIST_HEAD(dev_list); - for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) { + rdma_for_each_port (device, p) { if (!rdma_protocol_ib(device, p)) continue; dev = ipoib_add_port("ib%d", device, p); @@ -2577,9 +2577,7 @@ static int __init ipoib_init_module(void) */ BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE); - ret = ipoib_register_debugfs(); - if (ret) - return ret; + ipoib_register_debugfs(); /* * We create a global workqueue here that is used for all flush diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h index 120b40829560..a7aeaa0c6fbc 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.h +++ b/drivers/infiniband/ulp/iser/iscsi_iser.h @@ -197,7 +197,7 @@ struct iser_data_buf { struct scatterlist *sg; int size; unsigned long data_len; - unsigned int dma_nents; + int dma_nents; }; /* fwd declarations */ diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c index e9b7efc302d0..2ba70729d7b0 100644 --- a/drivers/infiniband/ulp/iser/iser_memory.c +++ b/drivers/infiniband/ulp/iser/iser_memory.c @@ -145,9 +145,8 @@ static void iser_data_buf_dump(struct iser_data_buf *data, for_each_sg(data->sg, sg, data->dma_nents, i) iser_dbg("sg[%d] dma_addr:0x%lX page:0x%p " "off:0x%x sz:0x%x dma_len:0x%x\n", - i, (unsigned long)ib_sg_dma_address(ibdev, sg), - sg_page(sg), sg->offset, - sg->length, ib_sg_dma_len(ibdev, sg)); + i, (unsigned long)sg_dma_address(sg), + sg_page(sg), sg->offset, sg->length, sg_dma_len(sg)); } static void iser_dump_page_vec(struct iser_page_vec *page_vec) @@ -204,8 +203,8 @@ iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem, reg->rkey = device->pd->unsafe_global_rkey; else reg->rkey = 0; - reg->sge.addr = ib_sg_dma_address(device->ib_device, &sg[0]); - reg->sge.length = ib_sg_dma_len(device->ib_device, &sg[0]); + reg->sge.addr = sg_dma_address(&sg[0]); + reg->sge.length = sg_dma_len(&sg[0]); iser_dbg("Single DMA entry: lkey=0x%x, rkey=0x%x, addr=0x%llx," " length=0x%x\n", reg->sge.lkey, reg->rkey, @@ -240,8 +239,8 @@ int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task, page_vec->npages = 0; page_vec->fake_mr.page_size = SIZE_4K; plen = ib_sg_to_pages(&page_vec->fake_mr, mem->sg, - mem->size, NULL, iser_set_page); - if (unlikely(plen < mem->size)) { + mem->dma_nents, NULL, iser_set_page); + if (unlikely(plen < mem->dma_nents)) { iser_err("page vec too short to hold this SG\n"); iser_data_buf_dump(mem, device->ib_device); iser_dump_page_vec(page_vec); @@ -448,10 +447,10 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task, ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey)); - n = ib_map_mr_sg(mr, mem->sg, mem->size, NULL, SIZE_4K); - if (unlikely(n != mem->size)) { + n = ib_map_mr_sg(mr, mem->sg, mem->dma_nents, NULL, SIZE_4K); + if (unlikely(n != mem->dma_nents)) { iser_err("failed to map sg (%d/%d)\n", - n, mem->size); + n, mem->dma_nents); return n < 0 ? n : -EINVAL; } diff --git a/drivers/infiniband/ulp/isert/Makefile b/drivers/infiniband/ulp/isert/Makefile index c8bf2421f5bc..a4a4766e3e18 100644 --- a/drivers/infiniband/ulp/isert/Makefile +++ b/drivers/infiniband/ulp/isert/Makefile @@ -1,2 +1 @@ -ccflags-y := -Idrivers/target -Idrivers/target/iscsi obj-$(CONFIG_INFINIBAND_ISERT) += ib_isert.o diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index 694324b37480..be9ddcad8f28 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -443,8 +443,7 @@ static struct srp_fr_pool *srp_create_fr_pool(struct ib_device *device, if (pool_size <= 0) goto err; ret = -ENOMEM; - pool = kzalloc(sizeof(struct srp_fr_pool) + - pool_size * sizeof(struct srp_fr_desc), GFP_KERNEL); + pool = kzalloc(struct_size(pool, desc, pool_size), GFP_KERNEL); if (!pool) goto err; pool->size = pool_size; @@ -1601,9 +1600,8 @@ static int srp_map_sg_entry(struct srp_map_state *state, { struct srp_target_port *target = ch->target; struct srp_device *dev = target->srp_host->srp_dev; - struct ib_device *ibdev = dev->dev; - dma_addr_t dma_addr = ib_sg_dma_address(ibdev, sg); - unsigned int dma_len = ib_sg_dma_len(ibdev, sg); + dma_addr_t dma_addr = sg_dma_address(sg); + unsigned int dma_len = sg_dma_len(sg); unsigned int len = 0; int ret; @@ -1697,13 +1695,11 @@ static int srp_map_sg_dma(struct srp_map_state *state, struct srp_rdma_ch *ch, int count) { struct srp_target_port *target = ch->target; - struct srp_device *dev = target->srp_host->srp_dev; struct scatterlist *sg; int i; for_each_sg(scat, sg, count, i) { - srp_map_desc(state, ib_sg_dma_address(dev->dev, sg), - ib_sg_dma_len(dev->dev, sg), + srp_map_desc(state, sg_dma_address(sg), sg_dma_len(sg), target->global_rkey); } @@ -1853,8 +1849,8 @@ static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_rdma_ch *ch, buf->len = cpu_to_be32(data_len); WARN_ON_ONCE((void *)(buf + 1) > (void *)cmd + len); for_each_sg(scat, sg, count, i) { - sge[i].addr = ib_sg_dma_address(ibdev, sg); - sge[i].length = ib_sg_dma_len(ibdev, sg); + sge[i].addr = sg_dma_address(sg); + sge[i].length = sg_dma_len(sg); sge[i].lkey = target->lkey; } req->cmd->num_sge += count; @@ -1875,9 +1871,9 @@ static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_rdma_ch *ch, struct srp_direct_buf *buf; buf = (void *)cmd->add_data + cmd->add_cdb_len; - buf->va = cpu_to_be64(ib_sg_dma_address(ibdev, scat)); + buf->va = cpu_to_be64(sg_dma_address(scat)); buf->key = cpu_to_be32(target->global_rkey); - buf->len = cpu_to_be32(ib_sg_dma_len(ibdev, scat)); + buf->len = cpu_to_be32(sg_dma_len(scat)); req->nmdesc = 0; goto map_complete; @@ -3814,6 +3810,7 @@ static ssize_t srp_create_target(struct device *dev, target_host->max_id = 1; target_host->max_lun = -1LL; target_host->max_cmd_len = sizeof ((struct srp_cmd *) (void *) 0L)->cdb; + target_host->max_segment_size = ib_dma_max_seg_size(ibdev); target = host_to_target(target_host); @@ -4120,7 +4117,8 @@ static void srp_add_one(struct ib_device *device) struct srp_device *srp_dev; struct ib_device_attr *attr = &device->attrs; struct srp_host *host; - int mr_page_shift, p; + int mr_page_shift; + unsigned int p; u64 max_pages_per_mr; unsigned int flags = 0; @@ -4187,7 +4185,7 @@ static void srp_add_one(struct ib_device *device) WARN_ON_ONCE(srp_dev->global_rkey == 0); } - for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) { + rdma_for_each_port (device, p) { host = srp_add_port(srp_dev, p); if (host) list_add_tail(&host->list, &srp_dev->dev_list); diff --git a/drivers/infiniband/ulp/srpt/Makefile b/drivers/infiniband/ulp/srpt/Makefile index e3ee4bdfffa5..43fbde42c58b 100644 --- a/drivers/infiniband/ulp/srpt/Makefile +++ b/drivers/infiniband/ulp/srpt/Makefile @@ -1,2 +1 @@ -ccflags-y := -Idrivers/target obj-$(CONFIG_INFINIBAND_SRPT) += ib_srpt.o diff --git a/drivers/media/pci/intel/ipu3/ipu3-cio2.c b/drivers/media/pci/intel/ipu3/ipu3-cio2.c index 617fb2e944dc..f8020ebe9f05 100644 --- a/drivers/media/pci/intel/ipu3/ipu3-cio2.c +++ b/drivers/media/pci/intel/ipu3/ipu3-cio2.c @@ -846,7 +846,7 @@ static int cio2_vb2_buf_init(struct vb2_buffer *vb) unsigned int pages = DIV_ROUND_UP(vb->planes[0].length, CIO2_PAGE_SIZE); unsigned int lops = DIV_ROUND_UP(pages + 1, entries_per_page); struct sg_table *sg; - struct sg_page_iter sg_iter; + struct sg_dma_page_iter sg_iter; int i, j; if (lops <= 0 || lops > CIO2_MAX_LOPS) { @@ -873,7 +873,7 @@ static int cio2_vb2_buf_init(struct vb2_buffer *vb) b->offset = sg->sgl->offset; i = j = 0; - for_each_sg_page(sg->sgl, &sg_iter, sg->nents, 0) { + for_each_sg_dma_page (sg->sgl, &sg_iter, sg->nents, 0) { if (!pages--) break; b->lop[i][j] = sg_page_iter_dma_address(&sg_iter) >> PAGE_SHIFT; diff --git a/drivers/misc/mic/scif/scif_rma.c b/drivers/misc/mic/scif/scif_rma.c index f62216628fa6..0fb9b440dc70 100644 --- a/drivers/misc/mic/scif/scif_rma.c +++ b/drivers/misc/mic/scif/scif_rma.c @@ -272,21 +272,12 @@ static inline void __scif_release_mm(struct mm_struct *mm) static inline int __scif_dec_pinned_vm_lock(struct mm_struct *mm, - int nr_pages, bool try_lock) + int nr_pages) { if (!mm || !nr_pages || !scif_ulimit_check) return 0; - if (try_lock) { - if (!down_write_trylock(&mm->mmap_sem)) { - dev_err(scif_info.mdev.this_device, - "%s %d err\n", __func__, __LINE__); - return -1; - } - } else { - down_write(&mm->mmap_sem); - } - mm->pinned_vm -= nr_pages; - up_write(&mm->mmap_sem); + + atomic64_sub(nr_pages, &mm->pinned_vm); return 0; } @@ -298,16 +289,16 @@ static inline int __scif_check_inc_pinned_vm(struct mm_struct *mm, if (!mm || !nr_pages || !scif_ulimit_check) return 0; - locked = nr_pages; - locked += mm->pinned_vm; lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; + locked = atomic64_add_return(nr_pages, &mm->pinned_vm); + if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { + atomic64_sub(nr_pages, &mm->pinned_vm); dev_err(scif_info.mdev.this_device, "locked(%lu) > lock_limit(%lu)\n", locked, lock_limit); return -ENOMEM; } - mm->pinned_vm = locked; return 0; } @@ -326,7 +317,7 @@ int scif_destroy_window(struct scif_endpt *ep, struct scif_window *window) might_sleep(); if (!window->temp && window->mm) { - __scif_dec_pinned_vm_lock(window->mm, window->nr_pages, 0); + __scif_dec_pinned_vm_lock(window->mm, window->nr_pages); __scif_release_mm(window->mm); window->mm = NULL; } @@ -737,7 +728,7 @@ done: ep->rma_info.dma_chan); } else { if (!__scif_dec_pinned_vm_lock(window->mm, - window->nr_pages, 1)) { + window->nr_pages)) { __scif_release_mm(window->mm); window->mm = NULL; } @@ -1385,28 +1376,23 @@ int __scif_pin_pages(void *addr, size_t len, int *out_prot, prot |= SCIF_PROT_WRITE; retry: mm = current->mm; - down_write(&mm->mmap_sem); if (ulimit) { err = __scif_check_inc_pinned_vm(mm, nr_pages); if (err) { - up_write(&mm->mmap_sem); pinned_pages->nr_pages = 0; goto error_unmap; } } - pinned_pages->nr_pages = get_user_pages( + pinned_pages->nr_pages = get_user_pages_fast( (u64)addr, nr_pages, (prot & SCIF_PROT_WRITE) ? FOLL_WRITE : 0, - pinned_pages->pages, - NULL); - up_write(&mm->mmap_sem); + pinned_pages->pages); if (nr_pages != pinned_pages->nr_pages) { if (try_upgrade) { if (ulimit) - __scif_dec_pinned_vm_lock(mm, - nr_pages, 0); + __scif_dec_pinned_vm_lock(mm, nr_pages); /* Roll back any pinned pages */ for (i = 0; i < pinned_pages->nr_pages; i++) { if (pinned_pages->pages[i]) @@ -1433,7 +1419,7 @@ retry: return err; dec_pinned: if (ulimit) - __scif_dec_pinned_vm_lock(mm, nr_pages, 0); + __scif_dec_pinned_vm_lock(mm, nr_pages); /* Something went wrong! Rollback */ error_unmap: pinned_pages->nr_pages = nr_pages; diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c index ea45a9b8179e..cf475873ce81 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c @@ -43,9 +43,6 @@ static int bnxt_register_dev(struct bnxt_en_dev *edev, int ulp_id, if (ulp_id == BNXT_ROCE_ULP) { unsigned int max_stat_ctxs; - if (bp->flags & BNXT_FLAG_CHIP_P5) - return -EOPNOTSUPP; - max_stat_ctxs = bnxt_get_max_func_stat_ctxs(bp); if (max_stat_ctxs <= BNXT_MIN_ROCE_STAT_CTXS || bp->cp_nr_rings == max_stat_ctxs) diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h index c62a0c830705..38dd41eb959e 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h +++ b/drivers/net/ethernet/chelsio/cxgb4/t4_msg.h @@ -56,6 +56,7 @@ enum { CPL_TX_DATA_ISO = 0x1F, CPL_CLOSE_LISTSRV_RPL = 0x20, + CPL_GET_TCB_RPL = 0x22, CPL_L2T_WRITE_RPL = 0x23, CPL_PASS_OPEN_RPL = 0x24, CPL_ACT_OPEN_RPL = 0x25, @@ -688,6 +689,13 @@ struct cpl_get_tcb { #define NO_REPLY_V(x) ((x) << NO_REPLY_S) #define NO_REPLY_F NO_REPLY_V(1U) +struct cpl_get_tcb_rpl { + union opcode_tid ot; + __u8 cookie; + __u8 status; + __be16 len; +}; + struct cpl_set_tcb_field { WR_HDR; union opcode_tid ot; diff --git a/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h b/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h index 3297ce025e8b..1b9afb192f7f 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h +++ b/drivers/net/ethernet/chelsio/cxgb4/t4_tcb.h @@ -41,6 +41,14 @@ #define TCB_SMAC_SEL_V(x) ((x) << TCB_SMAC_SEL_S) #define TCB_T_FLAGS_W 1 +#define TCB_T_FLAGS_S 0 +#define TCB_T_FLAGS_M 0xffffffffffffffffULL +#define TCB_T_FLAGS_V(x) ((__u64)(x) << TCB_T_FLAGS_S) + +#define TCB_RQ_START_W 30 +#define TCB_RQ_START_S 0 +#define TCB_RQ_START_M 0x3ffffffULL +#define TCB_RQ_START_V(x) ((x) << TCB_RQ_START_S) #define TF_CCTRL_ECE_S 60 #define TF_CCTRL_CWR_S 61 @@ -66,4 +74,8 @@ #define TCB_RX_FRAG3_LEN_RAW_W 29 #define TCB_RX_FRAG3_START_IDX_OFFSET_RAW_W 30 #define TCB_PDU_HDR_LEN_W 31 + +#define TF_RX_PDU_OUT_S 49 +#define TF_RX_PDU_OUT_V(x) ((__u64)(x) << TF_RX_PDU_OUT_S) + #endif /* __T4_TCB_H */ diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c index 8391dde869a7..70cc906a102b 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/main.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c @@ -465,20 +465,17 @@ static int handle_hca_cap_odp(struct mlx5_core_dev *dev) void *set_hca_cap; void *set_ctx; int set_sz; + bool do_set = false; int err; - if (!MLX5_CAP_GEN(dev, pg)) + if (!IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) || + !MLX5_CAP_GEN(dev, pg)) return 0; err = mlx5_core_get_caps(dev, MLX5_CAP_ODP); if (err) return err; - if (!(MLX5_CAP_ODP_MAX(dev, ud_odp_caps.srq_receive) || - MLX5_CAP_ODP_MAX(dev, rc_odp_caps.srq_receive) || - MLX5_CAP_ODP_MAX(dev, xrc_odp_caps.srq_receive))) - return 0; - set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in); set_ctx = kzalloc(set_sz, GFP_KERNEL); if (!set_ctx) @@ -488,19 +485,30 @@ static int handle_hca_cap_odp(struct mlx5_core_dev *dev) memcpy(set_hca_cap, dev->caps.hca_cur[MLX5_CAP_ODP], MLX5_ST_SZ_BYTES(odp_cap)); - /* set ODP SRQ support for RC/UD and XRC transports */ - MLX5_SET(odp_cap, set_hca_cap, ud_odp_caps.srq_receive, - MLX5_CAP_ODP_MAX(dev, ud_odp_caps.srq_receive)); - - MLX5_SET(odp_cap, set_hca_cap, rc_odp_caps.srq_receive, - MLX5_CAP_ODP_MAX(dev, rc_odp_caps.srq_receive)); - - MLX5_SET(odp_cap, set_hca_cap, xrc_odp_caps.srq_receive, - MLX5_CAP_ODP_MAX(dev, xrc_odp_caps.srq_receive)); - - err = set_caps(dev, set_ctx, set_sz, MLX5_SET_HCA_CAP_OP_MOD_ODP); +#define ODP_CAP_SET_MAX(dev, field) \ + do { \ + u32 _res = MLX5_CAP_ODP_MAX(dev, field); \ + if (_res) { \ + do_set = true; \ + MLX5_SET(odp_cap, set_hca_cap, field, _res); \ + } \ + } while (0) + + ODP_CAP_SET_MAX(dev, ud_odp_caps.srq_receive); + ODP_CAP_SET_MAX(dev, rc_odp_caps.srq_receive); + ODP_CAP_SET_MAX(dev, xrc_odp_caps.srq_receive); + ODP_CAP_SET_MAX(dev, xrc_odp_caps.send); + ODP_CAP_SET_MAX(dev, xrc_odp_caps.receive); + ODP_CAP_SET_MAX(dev, xrc_odp_caps.write); + ODP_CAP_SET_MAX(dev, xrc_odp_caps.read); + ODP_CAP_SET_MAX(dev, xrc_odp_caps.atomic); + + if (do_set) + err = set_caps(dev, set_ctx, set_sz, + MLX5_SET_HCA_CAP_OP_MOD_ODP); kfree(set_ctx); + return err; } @@ -577,6 +585,33 @@ query_ex: return err; } +static int set_hca_cap(struct mlx5_core_dev *dev) +{ + struct pci_dev *pdev = dev->pdev; + int err; + + err = handle_hca_cap(dev); + if (err) { + dev_err(&pdev->dev, "handle_hca_cap failed\n"); + goto out; + } + + err = handle_hca_cap_atomic(dev); + if (err) { + dev_err(&pdev->dev, "handle_hca_cap_atomic failed\n"); + goto out; + } + + err = handle_hca_cap_odp(dev); + if (err) { + dev_err(&pdev->dev, "handle_hca_cap_odp failed\n"); + goto out; + } + +out: + return err; +} + static int set_hca_ctrl(struct mlx5_core_dev *dev) { struct mlx5_reg_host_endianness he_in; @@ -969,21 +1004,9 @@ static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, goto reclaim_boot_pages; } - err = handle_hca_cap(dev); + err = set_hca_cap(dev); if (err) { - dev_err(&pdev->dev, "handle_hca_cap failed\n"); - goto reclaim_boot_pages; - } - - err = handle_hca_cap_atomic(dev); - if (err) { - dev_err(&pdev->dev, "handle_hca_cap_atomic failed\n"); - goto reclaim_boot_pages; - } - - err = handle_hca_cap_odp(dev); - if (err) { - dev_err(&pdev->dev, "handle_hca_cap_odp failed\n"); + dev_err(&pdev->dev, "set_hca_cap failed\n"); goto reclaim_boot_pages; } diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c index beccb0b1d57c..92a91e7816d8 100644 --- a/fs/proc/task_mmu.c +++ b/fs/proc/task_mmu.c @@ -59,7 +59,7 @@ void task_mem(struct seq_file *m, struct mm_struct *mm) SEQ_PUT_DEC("VmPeak:\t", hiwater_vm); SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm); SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm); - SEQ_PUT_DEC(" kB\nVmPin:\t", mm->pinned_vm); + SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm)); SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss); SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss); SEQ_PUT_DEC(" kB\nRssAnon:\t", anon); diff --git a/include/linux/cgroup_rdma.h b/include/linux/cgroup_rdma.h index e94290b29e99..ef1bae2983f3 100644 --- a/include/linux/cgroup_rdma.h +++ b/include/linux/cgroup_rdma.h @@ -39,7 +39,7 @@ struct rdmacg_device { * APIs for RDMA/IB stack to publish when a device wants to * participate in resource accounting */ -int rdmacg_register_device(struct rdmacg_device *device); +void rdmacg_register_device(struct rdmacg_device *device); void rdmacg_unregister_device(struct rdmacg_device *device); /* APIs for RDMA/IB stack to charge/uncharge pool specific resources */ diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h index 5ffb5df1a2c2..022541dc5dbf 100644 --- a/include/linux/mlx5/driver.h +++ b/include/linux/mlx5/driver.h @@ -365,6 +365,7 @@ struct mlx5_core_sig_ctx { enum { MLX5_MKEY_MR = 1, MLX5_MKEY_MW, + MLX5_MKEY_INDIRECT_DEVX, }; struct mlx5_core_mkey { @@ -960,10 +961,6 @@ int mlx5_query_odp_caps(struct mlx5_core_dev *dev, struct mlx5_odp_caps *odp_caps); int mlx5_core_query_ib_ppcnt(struct mlx5_core_dev *dev, u8 port_num, void *out, size_t sz); -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING -int mlx5_core_page_fault_resume(struct mlx5_core_dev *dev, u32 token, - u32 wq_num, u8 type, int error); -#endif int mlx5_init_rl_table(struct mlx5_core_dev *dev); void mlx5_cleanup_rl_table(struct mlx5_core_dev *dev); diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index 86e7a7a46353..7eade9132f02 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -411,7 +411,7 @@ struct mm_struct { unsigned long total_vm; /* Total pages mapped */ unsigned long locked_vm; /* Pages that have PG_mlocked set */ - unsigned long pinned_vm; /* Refcount permanently increased */ + atomic64_t pinned_vm; /* Refcount permanently increased */ unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ unsigned long stack_vm; /* VM_STACK */ diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h index b96f0d0b5b8f..b4be960c7e5d 100644 --- a/include/linux/scatterlist.h +++ b/include/linux/scatterlist.h @@ -339,12 +339,12 @@ int sg_alloc_table_chained(struct sg_table *table, int nents, /* * sg page iterator * - * Iterates over sg entries page-by-page. On each successful iteration, - * you can call sg_page_iter_page(@piter) and sg_page_iter_dma_address(@piter) - * to get the current page and its dma address. @piter->sg will point to the - * sg holding this page and @piter->sg_pgoffset to the page's page offset - * within the sg. The iteration will stop either when a maximum number of sg - * entries was reached or a terminating sg (sg_last(sg) == true) was reached. + * Iterates over sg entries page-by-page. On each successful iteration, you + * can call sg_page_iter_page(@piter) to get the current page and its dma + * address. @piter->sg will point to the sg holding this page and + * @piter->sg_pgoffset to the page's page offset within the sg. The iteration + * will stop either when a maximum number of sg entries was reached or a + * terminating sg (sg_last(sg) == true) was reached. */ struct sg_page_iter { struct scatterlist *sg; /* sg holding the page */ @@ -356,7 +356,19 @@ struct sg_page_iter { * next step */ }; +/* + * sg page iterator for DMA addresses + * + * This is the same as sg_page_iter however you can call + * sg_page_iter_dma_address(@dma_iter) to get the page's DMA + * address. sg_page_iter_page() cannot be called on this iterator. + */ +struct sg_dma_page_iter { + struct sg_page_iter base; +}; + bool __sg_page_iter_next(struct sg_page_iter *piter); +bool __sg_page_iter_dma_next(struct sg_dma_page_iter *dma_iter); void __sg_page_iter_start(struct sg_page_iter *piter, struct scatterlist *sglist, unsigned int nents, unsigned long pgoffset); @@ -372,11 +384,13 @@ static inline struct page *sg_page_iter_page(struct sg_page_iter *piter) /** * sg_page_iter_dma_address - get the dma address of the current page held by * the page iterator. - * @piter: page iterator holding the page + * @dma_iter: page iterator holding the page */ -static inline dma_addr_t sg_page_iter_dma_address(struct sg_page_iter *piter) +static inline dma_addr_t +sg_page_iter_dma_address(struct sg_dma_page_iter *dma_iter) { - return sg_dma_address(piter->sg) + (piter->sg_pgoffset << PAGE_SHIFT); + return sg_dma_address(dma_iter->base.sg) + + (dma_iter->base.sg_pgoffset << PAGE_SHIFT); } /** @@ -385,11 +399,28 @@ static inline dma_addr_t sg_page_iter_dma_address(struct sg_page_iter *piter) * @piter: page iterator to hold current page, sg, sg_pgoffset * @nents: maximum number of sg entries to iterate over * @pgoffset: starting page offset + * + * Callers may use sg_page_iter_page() to get each page pointer. */ #define for_each_sg_page(sglist, piter, nents, pgoffset) \ for (__sg_page_iter_start((piter), (sglist), (nents), (pgoffset)); \ __sg_page_iter_next(piter);) +/** + * for_each_sg_dma_page - iterate over the pages of the given sg list + * @sglist: sglist to iterate over + * @dma_iter: page iterator to hold current page + * @dma_nents: maximum number of sg entries to iterate over, this is the value + * returned from dma_map_sg + * @pgoffset: starting page offset + * + * Callers may use sg_page_iter_dma_address() to get each page's DMA address. + */ +#define for_each_sg_dma_page(sglist, dma_iter, dma_nents, pgoffset) \ + for (__sg_page_iter_start(&(dma_iter)->base, sglist, dma_nents, \ + pgoffset); \ + __sg_page_iter_dma_next(dma_iter);) + /* * Mapping sg iterator * diff --git a/include/rdma/ib_hdrs.h b/include/rdma/ib_hdrs.h index 6e35416170a3..9a90bd031e8c 100644 --- a/include/rdma/ib_hdrs.h +++ b/include/rdma/ib_hdrs.h @@ -1,5 +1,5 @@ /* - * Copyright(c) 2016 Intel Corporation. + * Copyright(c) 2016 - 2018 Intel Corporation. * * This file is provided under a dual BSD/GPLv2 license. When using or * redistributing this file, you may do so under either license. @@ -100,6 +100,8 @@ struct ib_atomic_eth { __be64 compare_data; /* potentially unaligned */ } __packed; +#include <rdma/tid_rdma_defs.h> + union ib_ehdrs { struct { __be32 deth[2]; @@ -117,6 +119,16 @@ union ib_ehdrs { __be32 aeth; __be32 ieth; struct ib_atomic_eth atomic_eth; + /* TID RDMA headers */ + union { + struct tid_rdma_read_req r_req; + struct tid_rdma_read_resp r_rsp; + struct tid_rdma_write_req w_req; + struct tid_rdma_write_resp w_rsp; + struct tid_rdma_write_data w_data; + struct tid_rdma_resync resync; + struct tid_rdma_ack ack; + } tid_rdma; } __packed; struct ib_other_headers { diff --git a/include/rdma/ib_mad.h b/include/rdma/ib_mad.h index fdef558e3a2d..79ba8219e7dc 100644 --- a/include/rdma/ib_mad.h +++ b/include/rdma/ib_mad.h @@ -616,12 +616,11 @@ struct ib_mad_agent { void *context; u32 hi_tid; u32 flags; + void *security; + struct list_head mad_agent_sec_list; u8 port_num; u8 rmpp_version; - void *security; bool smp_allowed; - bool lsm_nb_reg; - struct notifier_block lsm_nb; }; /** diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h index 5d3755ec5afa..73af05db04c7 100644 --- a/include/rdma/ib_umem.h +++ b/include/rdma/ib_umem.h @@ -36,6 +36,7 @@ #include <linux/list.h> #include <linux/scatterlist.h> #include <linux/workqueue.h> +#include <rdma/ib_verbs.h> struct ib_ucontext; struct ib_umem_odp; @@ -80,7 +81,7 @@ static inline size_t ib_umem_num_pages(struct ib_umem *umem) #ifdef CONFIG_INFINIBAND_USER_MEM -struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, +struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr, size_t size, int access, int dmasync); void ib_umem_release(struct ib_umem *umem); int ib_umem_page_count(struct ib_umem *umem); @@ -91,9 +92,10 @@ int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset, #include <linux/err.h> -static inline struct ib_umem *ib_umem_get(struct ib_ucontext *context, +static inline struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr, size_t size, - int access, int dmasync) { + int access, int dmasync) +{ return ERR_PTR(-EINVAL); } static inline void ib_umem_release(struct ib_umem *umem) { } diff --git a/include/rdma/ib_umem_odp.h b/include/rdma/ib_umem_odp.h index 0b1446fe2fab..dadc96dea39c 100644 --- a/include/rdma/ib_umem_odp.h +++ b/include/rdma/ib_umem_odp.h @@ -83,6 +83,19 @@ static inline struct ib_umem_odp *to_ib_umem_odp(struct ib_umem *umem) return container_of(umem, struct ib_umem_odp, umem); } +/* + * The lower 2 bits of the DMA address signal the R/W permissions for + * the entry. To upgrade the permissions, provide the appropriate + * bitmask to the map_dma_pages function. + * + * Be aware that upgrading a mapped address might result in change of + * the DMA address for the page. + */ +#define ODP_READ_ALLOWED_BIT (1<<0ULL) +#define ODP_WRITE_ALLOWED_BIT (1<<1ULL) + +#define ODP_DMA_ADDR_MASK (~(ODP_READ_ALLOWED_BIT | ODP_WRITE_ALLOWED_BIT)) + #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING struct ib_ucontext_per_mm { @@ -103,23 +116,10 @@ struct ib_ucontext_per_mm { }; int ib_umem_odp_get(struct ib_umem_odp *umem_odp, int access); -struct ib_umem_odp *ib_alloc_odp_umem(struct ib_ucontext_per_mm *per_mm, +struct ib_umem_odp *ib_alloc_odp_umem(struct ib_umem_odp *root_umem, unsigned long addr, size_t size); void ib_umem_odp_release(struct ib_umem_odp *umem_odp); -/* - * The lower 2 bits of the DMA address signal the R/W permissions for - * the entry. To upgrade the permissions, provide the appropriate - * bitmask to the map_dma_pages function. - * - * Be aware that upgrading a mapped address might result in change of - * the DMA address for the page. - */ -#define ODP_READ_ALLOWED_BIT (1<<0ULL) -#define ODP_WRITE_ALLOWED_BIT (1<<1ULL) - -#define ODP_DMA_ADDR_MASK (~(ODP_READ_ALLOWED_BIT | ODP_WRITE_ALLOWED_BIT)) - int ib_umem_odp_map_dma_pages(struct ib_umem_odp *umem_odp, u64 start_offset, u64 bcnt, u64 access_mask, unsigned long current_seq); @@ -169,12 +169,6 @@ static inline int ib_umem_odp_get(struct ib_umem_odp *umem_odp, int access) return -EINVAL; } -static inline struct ib_umem_odp * -ib_alloc_odp_umem(struct ib_ucontext *context, unsigned long addr, size_t size) -{ - return ERR_PTR(-EINVAL); -} - static inline void ib_umem_odp_release(struct ib_umem_odp *umem_odp) {} #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 80debf5982ac..9b9e17bcc201 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -238,6 +238,7 @@ enum ib_device_cap_flags { IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), /* The device supports padding incoming writes to cacheline. */ IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), + IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37), }; enum ib_signature_prot_cap { @@ -268,6 +269,7 @@ enum ib_odp_transport_cap_bits { IB_ODP_SUPPORT_WRITE = 1 << 2, IB_ODP_SUPPORT_READ = 1 << 3, IB_ODP_SUPPORT_ATOMIC = 1 << 4, + IB_ODP_SUPPORT_SRQ_RECV = 1 << 5, }; struct ib_odp_caps { @@ -276,6 +278,7 @@ struct ib_odp_caps { uint32_t rc_odp_caps; uint32_t uc_odp_caps; uint32_t ud_odp_caps; + uint32_t xrc_odp_caps; } per_transport_caps; }; @@ -1504,12 +1507,10 @@ struct ib_ucontext { bool cleanup_retryable; -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING void (*invalidate_range)(struct ib_umem_odp *umem_odp, unsigned long start, unsigned long end); struct mutex per_mm_list_lock; struct list_head per_mm_list; -#endif struct ib_rdmacg_object cg_obj; /* @@ -2186,7 +2187,6 @@ struct ib_port_cache { struct ib_cache { rwlock_t lock; struct ib_event_handler event_handler; - struct ib_port_cache *ports; }; struct iw_cm_verbs; @@ -2198,6 +2198,21 @@ struct ib_port_immutable { u32 max_mad_size; }; +struct ib_port_data { + struct ib_device *ib_dev; + + struct ib_port_immutable immutable; + + spinlock_t pkey_list_lock; + struct list_head pkey_list; + + struct ib_port_cache cache; + + spinlock_t netdev_lock; + struct net_device __rcu *netdev; + struct hlist_node ndev_hash_link; +}; + /* rdma netdev type - specifies protocol type */ enum rdma_netdev_t { RDMA_NETDEV_OPA_VNIC, @@ -2243,12 +2258,6 @@ struct rdma_netdev_alloc_params { struct net_device *netdev, void *param); }; -struct ib_port_pkey_list { - /* Lock to hold while modifying the list. */ - spinlock_t list_lock; - struct list_head pkey_list; -}; - struct ib_counters { struct ib_device *device; struct ib_uobject *uobject; @@ -2264,6 +2273,19 @@ struct ib_counters_read_attr { struct uverbs_attr_bundle; +#define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \ + .size_##ib_struct = \ + (sizeof(struct drv_struct) + \ + BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \ + BUILD_BUG_ON_ZERO( \ + !__same_type(((struct drv_struct *)NULL)->member, \ + struct ib_struct))) + +#define rdma_zalloc_drv_obj(ib_dev, ib_type) \ + ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, GFP_KERNEL)) + +#define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct + /** * struct ib_device_ops - InfiniBand device operations * This structure defines all the InfiniBand device operations, providers will @@ -2367,15 +2389,14 @@ struct ib_device_ops { int (*del_gid)(const struct ib_gid_attr *attr, void **context); int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index, u16 *pkey); - struct ib_ucontext *(*alloc_ucontext)(struct ib_device *device, - struct ib_udata *udata); - int (*dealloc_ucontext)(struct ib_ucontext *context); + int (*alloc_ucontext)(struct ib_ucontext *context, + struct ib_udata *udata); + void (*dealloc_ucontext)(struct ib_ucontext *context); int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma); void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); - struct ib_pd *(*alloc_pd)(struct ib_device *device, - struct ib_ucontext *context, - struct ib_udata *udata); - int (*dealloc_pd)(struct ib_pd *pd); + int (*alloc_pd)(struct ib_pd *pd, struct ib_ucontext *context, + struct ib_udata *udata); + void (*dealloc_pd)(struct ib_pd *pd); struct ib_ah *(*create_ah)(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, u32 flags, struct ib_udata *udata); @@ -2506,34 +2527,57 @@ struct ib_device_ops { */ int (*get_hw_stats)(struct ib_device *device, struct rdma_hw_stats *stats, u8 port, int index); + /* + * This function is called once for each port when a ib device is + * registered. + */ + int (*init_port)(struct ib_device *device, u8 port_num, + struct kobject *port_sysfs); + /** + * Allows rdma drivers to add their own restrack attributes. + */ + int (*fill_res_entry)(struct sk_buff *msg, + struct rdma_restrack_entry *entry); + + /* Device lifecycle callbacks */ + /* + * Called after the device becomes registered, before clients are + * attached + */ + int (*enable_driver)(struct ib_device *dev); + /* + * This is called as part of ib_dealloc_device(). + */ + void (*dealloc_driver)(struct ib_device *dev); + + DECLARE_RDMA_OBJ_SIZE(ib_pd); + DECLARE_RDMA_OBJ_SIZE(ib_ucontext); }; +struct rdma_restrack_root; + struct ib_device { /* Do not access @dma_device directly from ULP nor from HW drivers. */ struct device *dma_device; struct ib_device_ops ops; char name[IB_DEVICE_NAME_MAX]; + struct rcu_head rcu_head; struct list_head event_handler_list; spinlock_t event_handler_lock; - rwlock_t client_data_lock; - struct list_head core_list; - /* Access to the client_data_list is protected by the client_data_lock - * rwlock and the lists_rwsem read-write semaphore - */ - struct list_head client_data_list; + struct rw_semaphore client_data_rwsem; + struct xarray client_data; + struct mutex unregistration_lock; struct ib_cache cache; /** - * port_immutable is indexed by port number + * port_data is indexed by port number */ - struct ib_port_immutable *port_immutable; + struct ib_port_data *port_data; int num_comp_vectors; - struct ib_port_pkey_list *port_pkey_list; - struct iw_cm_verbs *iwcm; struct module *owner; @@ -2547,12 +2591,6 @@ struct ib_device { struct kobject *ports_kobj; struct list_head port_list; - enum { - IB_DEV_UNINITIALIZED, - IB_DEV_REGISTERED, - IB_DEV_UNREGISTERED - } reg_state; - int uverbs_abi_ver; u64 uverbs_cmd_mask; u64 uverbs_ex_cmd_mask; @@ -2561,6 +2599,8 @@ struct ib_device { __be64 node_guid; u32 local_dma_lkey; u16 is_switch:1; + /* Indicates kernel verbs support, should not be used in drivers */ + u16 kverbs_provider:1; u8 node_type; u8 phys_port_cnt; struct ib_device_attr attrs; @@ -2572,10 +2612,7 @@ struct ib_device { #endif u32 index; - /* - * Implementation details of the RDMA core, don't use in drivers - */ - struct rdma_restrack_root res; + struct rdma_restrack_root *res; const struct uapi_definition *driver_def; enum rdma_driver_id driver_id; @@ -2586,10 +2623,13 @@ struct ib_device { */ refcount_t refcount; struct completion unreg_completion; + struct work_struct unregistration_work; + + const struct rdma_link_ops *link_ops; }; struct ib_client { - char *name; + const char *name; void (*add) (struct ib_device *); void (*remove)(struct ib_device *, void *client_data); @@ -2616,22 +2656,47 @@ struct ib_client { const struct sockaddr *addr, void *client_data); struct list_head list; + u32 client_id; + + /* kverbs are not required by the client */ + u8 no_kverbs_req:1; }; -struct ib_device *ib_alloc_device(size_t size); +struct ib_device *_ib_alloc_device(size_t size); +#define ib_alloc_device(drv_struct, member) \ + container_of(_ib_alloc_device(sizeof(struct drv_struct) + \ + BUILD_BUG_ON_ZERO(offsetof( \ + struct drv_struct, member))), \ + struct drv_struct, member) + void ib_dealloc_device(struct ib_device *device); void ib_get_device_fw_str(struct ib_device *device, char *str); -int ib_register_device(struct ib_device *device, const char *name, - int (*port_callback)(struct ib_device *, u8, - struct kobject *)); +int ib_register_device(struct ib_device *device, const char *name); void ib_unregister_device(struct ib_device *device); +void ib_unregister_driver(enum rdma_driver_id driver_id); +void ib_unregister_device_and_put(struct ib_device *device); +void ib_unregister_device_queued(struct ib_device *ib_dev); int ib_register_client (struct ib_client *client); void ib_unregister_client(struct ib_client *client); -void *ib_get_client_data(struct ib_device *device, struct ib_client *client); +/** + * ib_get_client_data - Get IB client context + * @device:Device to get context for + * @client:Client to get context for + * + * ib_get_client_data() returns the client context data set with + * ib_set_client_data(). This can only be called while the client is + * registered to the device, once the ib_client remove() callback returns this + * cannot be called. + */ +static inline void *ib_get_client_data(struct ib_device *device, + struct ib_client *client) +{ + return xa_load(&device->client_data, client->client_id); +} void ib_set_client_data(struct ib_device *device, struct ib_client *client, void *data); void ib_set_device_ops(struct ib_device *device, @@ -2790,6 +2855,16 @@ static inline u8 rdma_start_port(const struct ib_device *device) } /** + * rdma_for_each_port - Iterate over all valid port numbers of the IB device + * @device - The struct ib_device * to iterate over + * @iter - The unsigned int to store the port number + */ +#define rdma_for_each_port(device, iter) \ + for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \ + unsigned int, iter))); \ + iter <= rdma_end_port(device); (iter)++) + +/** * rdma_end_port - Return the last valid port number for the device * specified * @@ -2812,34 +2887,38 @@ static inline int rdma_is_port_valid(const struct ib_device *device, static inline bool rdma_is_grh_required(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & - RDMA_CORE_PORT_IB_GRH_REQUIRED; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_PORT_IB_GRH_REQUIRED; } static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_PROT_IB; } static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & - (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); + return device->port_data[port_num].immutable.core_cap_flags & + (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); } static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; } static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_PROT_ROCE; } static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_PROT_IWARP; } static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) @@ -2850,12 +2929,14 @@ static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_PROT_RAW_PACKET; } static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_PROT_USNIC; } /** @@ -2872,7 +2953,8 @@ static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_n */ static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_IB_MAD; } /** @@ -2896,8 +2978,8 @@ static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) */ static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) { - return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) - == RDMA_CORE_CAP_OPA_MAD; + return (device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_OPA_MAD) == RDMA_CORE_CAP_OPA_MAD; } /** @@ -2922,7 +3004,8 @@ static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) */ static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_IB_SMI; } /** @@ -2942,7 +3025,8 @@ static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) */ static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_IB_CM; } /** @@ -2959,7 +3043,8 @@ static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) */ static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_IW_CM; } /** @@ -2979,7 +3064,8 @@ static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) */ static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_IB_SA; } /** @@ -3019,7 +3105,8 @@ static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num */ static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_AF_IB; } /** @@ -3040,7 +3127,8 @@ static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) */ static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; + return device->port_data[port_num].immutable.core_cap_flags & + RDMA_CORE_CAP_ETH_AH; } /** @@ -3054,7 +3142,7 @@ static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) */ static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) { - return (device->port_immutable[port_num].core_cap_flags & + return (device->port_data[port_num].immutable.core_cap_flags & RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; } @@ -3072,7 +3160,7 @@ static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) */ static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) { - return device->port_immutable[port_num].max_mad_size; + return device->port_data[port_num].immutable.max_mad_size; } /** @@ -3686,32 +3774,18 @@ static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, { dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); } -/** - * ib_sg_dma_address - Return the DMA address from a scatter/gather entry - * @dev: The device for which the DMA addresses were created - * @sg: The scatter/gather entry - * - * Note: this function is obsolete. To do: change all occurrences of - * ib_sg_dma_address() into sg_dma_address(). - */ -static inline u64 ib_sg_dma_address(struct ib_device *dev, - struct scatterlist *sg) -{ - return sg_dma_address(sg); -} /** - * ib_sg_dma_len - Return the DMA length from a scatter/gather entry - * @dev: The device for which the DMA addresses were created - * @sg: The scatter/gather entry + * ib_dma_max_seg_size - Return the size limit of a single DMA transfer + * @dev: The device to query * - * Note: this function is obsolete. To do: change all occurrences of - * ib_sg_dma_len() into sg_dma_len(). + * The returned value represents a size in bytes. */ -static inline unsigned int ib_sg_dma_len(struct ib_device *dev, - struct scatterlist *sg) +static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev) { - return sg_dma_len(sg); + struct device_dma_parameters *p = dev->dma_device->dma_parms; + + return p ? p->max_segment_size : UINT_MAX; } /** @@ -3946,9 +4020,17 @@ static inline bool ib_device_try_get(struct ib_device *dev) } void ib_device_put(struct ib_device *device); +struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, + enum rdma_driver_id driver_id); +struct ib_device *ib_device_get_by_name(const char *name, + enum rdma_driver_id driver_id); struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, u16 pkey, const union ib_gid *gid, const struct sockaddr *addr); +int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, + unsigned int port); +struct net_device *ib_device_netdev(struct ib_device *dev, u8 port); + struct ib_wq *ib_create_wq(struct ib_pd *pd, struct ib_wq_init_attr *init_attr); int ib_destroy_wq(struct ib_wq *wq); @@ -4222,7 +4304,6 @@ void rdma_roce_rescan_device(struct ib_device *ibdev); struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile); - int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs); struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num, @@ -4258,4 +4339,27 @@ rdma_set_device_sysfs_group(struct ib_device *dev, dev->groups[1] = group; } +/** + * rdma_device_to_ibdev - Get ib_device pointer from device pointer + * + * @device: device pointer for which ib_device pointer to retrieve + * + * rdma_device_to_ibdev() retrieves ib_device pointer from device. + * + */ +static inline struct ib_device *rdma_device_to_ibdev(struct device *device) +{ + return container_of(device, struct ib_device, dev); +} + +/** + * rdma_device_to_drv_device - Helper macro to reach back to driver's + * ib_device holder structure from device pointer. + * + * NOTE: New drivers should not make use of this API; This API is only for + * existing drivers who have exposed sysfs entries using + * rdma_set_device_sysfs_group(). + */ +#define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \ + container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member) #endif /* IB_VERBS_H */ diff --git a/include/rdma/iw_cm.h b/include/rdma/iw_cm.h index 5cd7701db148..0e1f02815643 100644 --- a/include/rdma/iw_cm.h +++ b/include/rdma/iw_cm.h @@ -94,7 +94,8 @@ struct iw_cm_id { void (*add_ref)(struct iw_cm_id *); void (*rem_ref)(struct iw_cm_id *); u8 tos; - bool mapped; + bool tos_set:1; + bool mapped:1; }; struct iw_cm_conn_param { @@ -105,6 +106,18 @@ struct iw_cm_conn_param { u32 qpn; }; +enum iw_flags { + + /* + * This flag allows the iwcm and iwpmd to still advertise + * mappings but the real and mapped port numbers are the + * same. Further, iwpmd will not bind any user socket to + * reserve the port. This is required for soft iwarp + * to play in the port mapped iwarp space. + */ + IW_F_NO_PORT_MAP = (1 << 0), +}; + struct iw_cm_verbs { void (*add_ref)(struct ib_qp *qp); @@ -127,6 +140,7 @@ struct iw_cm_verbs { int (*destroy_listen)(struct iw_cm_id *cm_id); char ifname[IFNAMSIZ]; + enum iw_flags driver_flags; }; /** diff --git a/include/rdma/iw_portmap.h b/include/rdma/iw_portmap.h index fda31673a562..b9fee7feeeb5 100644 --- a/include/rdma/iw_portmap.h +++ b/include/rdma/iw_portmap.h @@ -58,167 +58,31 @@ struct iwpm_sa_data { struct sockaddr_storage mapped_loc_addr; struct sockaddr_storage rem_addr; struct sockaddr_storage mapped_rem_addr; + u32 flags; }; -/** - * iwpm_init - Allocate resources for the iwarp port mapper - * - * Should be called when network interface goes up. - */ int iwpm_init(u8); - -/** - * iwpm_exit - Deallocate resources for the iwarp port mapper - * - * Should be called when network interface goes down. - */ int iwpm_exit(u8); - -/** - * iwpm_valid_pid - Check if the userspace iwarp port mapper pid is valid - * - * Returns true if the pid is greater than zero, otherwise returns false - */ int iwpm_valid_pid(void); - -/** - * iwpm_register_pid - Send a netlink query to userspace - * to get the iwarp port mapper pid - * @pm_msg: Contains driver info to send to the userspace port mapper - * @nl_client: The index of the netlink client - */ int iwpm_register_pid(struct iwpm_dev_data *pm_msg, u8 nl_client); - -/** - * iwpm_add_mapping - Send a netlink add mapping request to - * the userspace port mapper - * @pm_msg: Contains the local ip/tcp address info to send - * @nl_client: The index of the netlink client - * - * If the request is successful, the pm_msg stores - * the port mapper response (mapped address info) - */ int iwpm_add_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client); - -/** - * iwpm_add_and_query_mapping - Send a netlink add and query mapping request - * to the userspace port mapper - * @pm_msg: Contains the local and remote ip/tcp address info to send - * @nl_client: The index of the netlink client - * - * If the request is successful, the pm_msg stores the - * port mapper response (mapped local and remote address info) - */ int iwpm_add_and_query_mapping(struct iwpm_sa_data *pm_msg, u8 nl_client); - -/** - * iwpm_remove_mapping - Send a netlink remove mapping request - * to the userspace port mapper - * - * @local_addr: Local ip/tcp address to remove - * @nl_client: The index of the netlink client - */ int iwpm_remove_mapping(struct sockaddr_storage *local_addr, u8 nl_client); - -/** - * iwpm_register_pid_cb - Process the port mapper response to - * iwpm_register_pid query - * @skb: - * @cb: Contains the received message (payload and netlink header) - * - * If successful, the function receives the userspace port mapper pid - * which is used in future communication with the port mapper - */ int iwpm_register_pid_cb(struct sk_buff *, struct netlink_callback *); - -/** - * iwpm_add_mapping_cb - Process the port mapper response to - * iwpm_add_mapping request - * @skb: - * @cb: Contains the received message (payload and netlink header) - */ int iwpm_add_mapping_cb(struct sk_buff *, struct netlink_callback *); - -/** - * iwpm_add_and_query_mapping_cb - Process the port mapper response to - * iwpm_add_and_query_mapping request - * @skb: - * @cb: Contains the received message (payload and netlink header) - */ int iwpm_add_and_query_mapping_cb(struct sk_buff *, struct netlink_callback *); - -/** - * iwpm_remote_info_cb - Process remote connecting peer address info, which - * the port mapper has received from the connecting peer - * - * @cb: Contains the received message (payload and netlink header) - * - * Stores the IPv4/IPv6 address info in a hash table - */ int iwpm_remote_info_cb(struct sk_buff *, struct netlink_callback *); - -/** - * iwpm_mapping_error_cb - Process port mapper notification for error - * - * @skb: - * @cb: Contains the received message (payload and netlink header) - */ int iwpm_mapping_error_cb(struct sk_buff *, struct netlink_callback *); - -/** - * iwpm_mapping_info_cb - Process a notification that the userspace - * port mapper daemon is started - * @skb: - * @cb: Contains the received message (payload and netlink header) - * - * Using the received port mapper pid, send all the local mapping - * info records to the userspace port mapper - */ int iwpm_mapping_info_cb(struct sk_buff *, struct netlink_callback *); - -/** - * iwpm_ack_mapping_info_cb - Process the port mapper ack for - * the provided local mapping info records - * @skb: - * @cb: Contains the received message (payload and netlink header) - */ int iwpm_ack_mapping_info_cb(struct sk_buff *, struct netlink_callback *); - -/** - * iwpm_get_remote_info - Get the remote connecting peer address info - * - * @mapped_loc_addr: Mapped local address of the listening peer - * @mapped_rem_addr: Mapped remote address of the connecting peer - * @remote_addr: To store the remote address of the connecting peer - * @nl_client: The index of the netlink client - * - * The remote address info is retrieved and provided to the client in - * the remote_addr. After that it is removed from the hash table - */ int iwpm_get_remote_info(struct sockaddr_storage *mapped_loc_addr, struct sockaddr_storage *mapped_rem_addr, struct sockaddr_storage *remote_addr, u8 nl_client); - -/** - * iwpm_create_mapinfo - Store local and mapped IPv4/IPv6 address - * info in a hash table - * @local_addr: Local ip/tcp address - * @mapped_addr: Mapped local ip/tcp address - * @nl_client: The index of the netlink client - */ int iwpm_create_mapinfo(struct sockaddr_storage *local_addr, - struct sockaddr_storage *mapped_addr, u8 nl_client); - -/** - * iwpm_remove_mapinfo - Remove local and mapped IPv4/IPv6 address - * info from the hash table - * @local_addr: Local ip/tcp address - * @mapped_addr: Mapped local ip/tcp address - * - * Returns err code if mapping info is not found in the hash table, - * otherwise returns 0 - */ + struct sockaddr_storage *mapped_addr, u8 nl_client, + u32 map_flags); int iwpm_remove_mapinfo(struct sockaddr_storage *local_addr, struct sockaddr_storage *mapped_addr); +int iwpm_hello_cb(struct sk_buff *skb, struct netlink_callback *cb); #endif /* _IW_PORTMAP_H */ diff --git a/include/rdma/rdma_cm.h b/include/rdma/rdma_cm.h index 60987a5903b7..71f48cfdc24c 100644 --- a/include/rdma/rdma_cm.h +++ b/include/rdma/rdma_cm.h @@ -374,6 +374,7 @@ int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse); */ int rdma_set_afonly(struct rdma_cm_id *id, int afonly); +int rdma_set_ack_timeout(struct rdma_cm_id *id, u8 timeout); /** * rdma_get_service_id - Return the IB service ID for a specified address. * @id: Communication identifier associated with the address. diff --git a/include/rdma/rdma_netlink.h b/include/rdma/rdma_netlink.h index 70218e6b5187..10732ab31ba2 100644 --- a/include/rdma/rdma_netlink.h +++ b/include/rdma/rdma_netlink.h @@ -99,4 +99,15 @@ int rdma_nl_multicast(struct sk_buff *skb, unsigned int group, gfp_t flags); * Returns true on success or false if no listeners. */ bool rdma_nl_chk_listeners(unsigned int group); + +struct rdma_link_ops { + struct list_head list; + const char *type; + int (*newlink)(const char *ibdev_name, struct net_device *ndev); +}; + +void rdma_link_register(struct rdma_link_ops *ops); +void rdma_link_unregister(struct rdma_link_ops *ops); + +#define MODULE_ALIAS_RDMA_LINK(type) MODULE_ALIAS("rdma-link-" type) #endif /* _RDMA_NETLINK_H */ diff --git a/include/rdma/rdma_vt.h b/include/rdma/rdma_vt.h index dd0ed8048bb4..4c257aff7d32 100644 --- a/include/rdma/rdma_vt.h +++ b/include/rdma/rdma_vt.h @@ -182,9 +182,15 @@ struct rvt_driver_params { u32 max_mad_size; u8 qos_shift; u8 max_rdma_atomic; + u8 extra_rdma_atomic; u8 reserved_operations; }; +/* User context */ +struct rvt_ucontext { + struct ib_ucontext ibucontext; +}; + /* Protection domain */ struct rvt_pd { struct ib_pd ibpd; @@ -250,9 +256,6 @@ struct rvt_driver_provided { */ void (*do_send)(struct rvt_qp *qp); - /* Passed to ib core registration. Callback to create syfs files */ - int (*port_callback)(struct ib_device *, u8, struct kobject *); - /* * Returns a pointer to the undelying hardware's PCI device. This is * used to display information as to what hardware is being referenced @@ -522,7 +525,14 @@ static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi) */ static inline unsigned int rvt_max_atomic(struct rvt_dev_info *rdi) { - return rdi->dparms.max_rdma_atomic + 1; + return rdi->dparms.max_rdma_atomic + + rdi->dparms.extra_rdma_atomic + 1; +} + +static inline unsigned int rvt_size_atomic(struct rvt_dev_info *rdi) +{ + return rdi->dparms.max_rdma_atomic + + rdi->dparms.extra_rdma_atomic; } /* @@ -569,9 +579,10 @@ static inline struct rvt_qp *rvt_lookup_qpn(struct rvt_dev_info *rdi, /** * rvt_mod_retry_timer - mod a retry timer * @qp - the QP + * @shift - timeout shift to wait for multiple packets * Modify a potentially already running retry timer */ -static inline void rvt_mod_retry_timer(struct rvt_qp *qp) +static inline void rvt_mod_retry_timer_ext(struct rvt_qp *qp, u8 shift) { struct ib_qp *ibqp = &qp->ibqp; struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device); @@ -579,8 +590,13 @@ static inline void rvt_mod_retry_timer(struct rvt_qp *qp) lockdep_assert_held(&qp->s_lock); qp->s_flags |= RVT_S_TIMER; /* 4.096 usec. * (1 << qp->timeout) */ - mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies + - rdi->busy_jiffies); + mod_timer(&qp->s_timer, jiffies + rdi->busy_jiffies + + (qp->timeout_jiffies << shift)); +} + +static inline void rvt_mod_retry_timer(struct rvt_qp *qp) +{ + return rvt_mod_retry_timer_ext(qp, 0); } struct rvt_dev_info *rvt_alloc_device(size_t size, int nports); diff --git a/include/rdma/rdmavt_qp.h b/include/rdma/rdmavt_qp.h index cbafb1878669..f0fbd4063fef 100644 --- a/include/rdma/rdmavt_qp.h +++ b/include/rdma/rdmavt_qp.h @@ -174,6 +174,7 @@ struct rvt_swqe { u32 lpsn; /* last packet sequence number */ u32 ssn; /* send sequence number */ u32 length; /* total length of data in sg_list */ + void *priv; /* driver dependent field */ struct rvt_sge sg_list[0]; }; @@ -235,6 +236,7 @@ struct rvt_ack_entry { u32 lpsn; u8 opcode; u8 sent; + void *priv; }; #define RC_QP_SCALING_INTERVAL 5 @@ -244,6 +246,7 @@ struct rvt_ack_entry { #define RVT_OPERATION_ATOMIC_SGE 0x00000004 #define RVT_OPERATION_LOCAL 0x00000008 #define RVT_OPERATION_USE_RESERVE 0x00000010 +#define RVT_OPERATION_IGN_RNR_CNT 0x00000020 #define RVT_OPERATION_MAX (IB_WR_RESERVED10 + 1) @@ -373,6 +376,7 @@ struct rvt_qp { u8 s_rnr_retry; /* requester RNR retry counter */ u8 s_num_rd_atomic; /* number of RDMA read/atomic pending */ u8 s_tail_ack_queue; /* index into s_ack_queue[] */ + u8 s_acked_ack_queue; /* index into s_ack_queue[] */ struct rvt_sge_state s_ack_rdma_sge; struct timer_list s_timer; @@ -629,6 +633,16 @@ __be32 rvt_compute_aeth(struct rvt_qp *qp); void rvt_get_credit(struct rvt_qp *qp, u32 aeth); /** + * rvt_restart_sge - rewind the sge state for a wqe + * @ss: the sge state pointer + * @wqe: the wqe to rewind + * @len: the data length from the start of the wqe in bytes + * + * Returns the remaining data length. + */ +u32 rvt_restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe, u32 len); + +/** * @qp - the qp pair * @len - the length * @@ -676,7 +690,11 @@ enum hrtimer_restart rvt_rc_rnr_retry(struct hrtimer *t); void rvt_add_rnr_timer(struct rvt_qp *qp, u32 aeth); void rvt_del_timers_sync(struct rvt_qp *qp); void rvt_stop_rc_timers(struct rvt_qp *qp); -void rvt_add_retry_timer(struct rvt_qp *qp); +void rvt_add_retry_timer_ext(struct rvt_qp *qp, u8 shift); +static inline void rvt_add_retry_timer(struct rvt_qp *qp) +{ + rvt_add_retry_timer_ext(qp, 0); +} void rvt_copy_sge(struct rvt_qp *qp, struct rvt_sge_state *ss, void *data, u32 length, diff --git a/include/rdma/restrack.h b/include/rdma/restrack.h index 8f179be9d9a9..ecf3c7702a4f 100644 --- a/include/rdma/restrack.h +++ b/include/rdma/restrack.h @@ -7,12 +7,12 @@ #define _RDMA_RESTRACK_H_ #include <linux/typecheck.h> -#include <linux/rwsem.h> #include <linux/sched.h> #include <linux/kref.h> #include <linux/completion.h> #include <linux/sched/task.h> #include <uapi/rdma/rdma_netlink.h> +#include <linux/xarray.h> /** * enum rdma_restrack_type - HW objects to track @@ -48,30 +48,7 @@ enum rdma_restrack_type { RDMA_RESTRACK_MAX }; -#define RDMA_RESTRACK_HASH_BITS 8 -struct rdma_restrack_entry; - -/** - * struct rdma_restrack_root - main resource tracking management - * entity, per-device - */ -struct rdma_restrack_root { - /* - * @rwsem: Read/write lock to protect lists - */ - struct rw_semaphore rwsem; - /** - * @hash: global database for all resources per-device - */ - DECLARE_HASHTABLE(hash, RDMA_RESTRACK_HASH_BITS); - /** - * @fill_res_entry: driver-specific fill function - * - * Allows rdma drivers to add their own restrack attributes. - */ - int (*fill_res_entry)(struct sk_buff *msg, - struct rdma_restrack_entry *entry); -}; +struct ib_device; /** * struct rdma_restrack_entry - metadata per-entry @@ -109,10 +86,6 @@ struct rdma_restrack_entry { */ const char *kern_name; /** - * @node: hash table entry - */ - struct hlist_node node; - /** * @type: various objects in restrack database */ enum rdma_restrack_type type; @@ -120,27 +93,13 @@ struct rdma_restrack_entry { * @user: user resource */ bool user; + /** + * @id: ID to expose to users + */ + u32 id; }; -/** - * rdma_restrack_init() - initialize resource tracking - * @res: resource tracking root - */ -void rdma_restrack_init(struct rdma_restrack_root *res); - -/** - * rdma_restrack_clean() - clean resource tracking - * @res: resource tracking root - */ -void rdma_restrack_clean(struct rdma_restrack_root *res); - -/** - * rdma_restrack_count() - the current usage of specific object - * @res: resource entry - * @type: actual type of object to operate - * @ns: PID namespace - */ -int rdma_restrack_count(struct rdma_restrack_root *res, +int rdma_restrack_count(struct ib_device *dev, enum rdma_restrack_type type, struct pid_namespace *ns); @@ -193,4 +152,7 @@ int rdma_nl_put_driver_u32_hex(struct sk_buff *msg, const char *name, int rdma_nl_put_driver_u64(struct sk_buff *msg, const char *name, u64 value); int rdma_nl_put_driver_u64_hex(struct sk_buff *msg, const char *name, u64 value); +struct rdma_restrack_entry *rdma_restrack_get_byid(struct ib_device *dev, + enum rdma_restrack_type type, + u32 id); #endif /* _RDMA_RESTRACK_H_ */ diff --git a/include/rdma/tid_rdma_defs.h b/include/rdma/tid_rdma_defs.h new file mode 100644 index 000000000000..08fe47c7ad2c --- /dev/null +++ b/include/rdma/tid_rdma_defs.h @@ -0,0 +1,108 @@ +/* SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) */ +/* + * Copyright(c) 2018 Intel Corporation. + * + */ + +#ifndef TID_RDMA_DEFS_H +#define TID_RDMA_DEFS_H + +#include <rdma/ib_pack.h> + +struct tid_rdma_read_req { + __le32 kdeth0; + __le32 kdeth1; + struct ib_reth reth; + __be32 tid_flow_psn; + __be32 tid_flow_qp; + __be32 verbs_qp; +}; + +struct tid_rdma_read_resp { + __le32 kdeth0; + __le32 kdeth1; + __be32 aeth; + __be32 reserved[4]; + __be32 verbs_psn; + __be32 verbs_qp; +}; + +struct tid_rdma_write_req { + __le32 kdeth0; + __le32 kdeth1; + struct ib_reth reth; + __be32 reserved[2]; + __be32 verbs_qp; +}; + +struct tid_rdma_write_resp { + __le32 kdeth0; + __le32 kdeth1; + __be32 aeth; + __be32 reserved[3]; + __be32 tid_flow_psn; + __be32 tid_flow_qp; + __be32 verbs_qp; +}; + +struct tid_rdma_write_data { + __le32 kdeth0; + __le32 kdeth1; + __be32 reserved[6]; + __be32 verbs_qp; +}; + +struct tid_rdma_resync { + __le32 kdeth0; + __le32 kdeth1; + __be32 reserved[6]; + __be32 verbs_qp; +}; + +struct tid_rdma_ack { + __le32 kdeth0; + __le32 kdeth1; + __be32 aeth; + __be32 reserved[2]; + __be32 tid_flow_psn; + __be32 verbs_psn; + __be32 tid_flow_qp; + __be32 verbs_qp; +}; + +/* + * TID RDMA Opcodes + */ +#define IB_OPCODE_TID_RDMA 0xe0 +enum { + IB_OPCODE_WRITE_REQ = 0x0, + IB_OPCODE_WRITE_RESP = 0x1, + IB_OPCODE_WRITE_DATA = 0x2, + IB_OPCODE_WRITE_DATA_LAST = 0x3, + IB_OPCODE_READ_REQ = 0x4, + IB_OPCODE_READ_RESP = 0x5, + IB_OPCODE_RESYNC = 0x6, + IB_OPCODE_ACK = 0x7, + + IB_OPCODE(TID_RDMA, WRITE_REQ), + IB_OPCODE(TID_RDMA, WRITE_RESP), + IB_OPCODE(TID_RDMA, WRITE_DATA), + IB_OPCODE(TID_RDMA, WRITE_DATA_LAST), + IB_OPCODE(TID_RDMA, READ_REQ), + IB_OPCODE(TID_RDMA, READ_RESP), + IB_OPCODE(TID_RDMA, RESYNC), + IB_OPCODE(TID_RDMA, ACK), +}; + +#define TID_OP(x) IB_OPCODE_TID_RDMA_##x + +/* + * Define TID RDMA specific WR opcodes. The ib_wr_opcode + * enum already provides some reserved values for use by + * low level drivers. Two of those are used but renamed + * to be more descriptive. + */ +#define IB_WR_TID_RDMA_WRITE IB_WR_RESERVED1 +#define IB_WR_TID_RDMA_READ IB_WR_RESERVED2 + +#endif /* TID_RDMA_DEFS_H */ diff --git a/include/rdma/uverbs_ioctl.h b/include/rdma/uverbs_ioctl.h index 27da906beea7..28570ac2b6a0 100644 --- a/include/rdma/uverbs_ioctl.h +++ b/include/rdma/uverbs_ioctl.h @@ -652,6 +652,7 @@ struct uverbs_attr_bundle { struct ib_udata driver_udata; struct ib_udata ucore; struct ib_uverbs_file *ufile; + struct ib_ucontext *context; DECLARE_BITMAP(attr_present, UVERBS_API_ATTR_BKEY_LEN); struct uverbs_attr attrs[]; }; @@ -663,6 +664,23 @@ static inline bool uverbs_attr_is_valid(const struct uverbs_attr_bundle *attrs_b attrs_bundle->attr_present); } +/** + * rdma_udata_to_drv_context - Helper macro to get the driver's context out of + * ib_udata which is embedded in uverbs_attr_bundle. + * + * If udata is not NULL this cannot fail. Otherwise a NULL udata will result + * in a NULL ucontext pointer, as a safety precaution. Callers should be using + * 'udata' to determine if the driver call is in user or kernel mode, not + * 'ucontext'. + * + */ +#define rdma_udata_to_drv_context(udata, drv_dev_struct, member) \ + (udata ? container_of(container_of(udata, struct uverbs_attr_bundle, \ + driver_udata) \ + ->context, \ + drv_dev_struct, member) : \ + (drv_dev_struct *)NULL) + #define IS_UVERBS_COPY_ERR(_ret) ((_ret) && (_ret) != -ENOENT) static inline const struct uverbs_attr *uverbs_attr_get(const struct uverbs_attr_bundle *attrs_bundle, diff --git a/include/rdma/uverbs_std_types.h b/include/rdma/uverbs_std_types.h index 883abcf6d36e..794c47565971 100644 --- a/include/rdma/uverbs_std_types.h +++ b/include/rdma/uverbs_std_types.h @@ -48,9 +48,12 @@ #define uobj_get_type(_attrs, _object) \ uapi_get_object((_attrs)->ufile->device->uapi, _object) +struct ib_uobject *_uobj_get_read(enum uverbs_default_objects type, + u32 object_id, + struct uverbs_attr_bundle *attrs); + #define uobj_get_read(_type, _id, _attrs) \ - rdma_lookup_get_uobject(uobj_get_type(_attrs, _type), (_attrs)->ufile, \ - _uobj_check_id(_id), UVERBS_LOOKUP_READ) + _uobj_get_read(_type, _uobj_check_id(_id), _attrs) #define ufd_get_read(_type, _fdnum, _attrs) \ rdma_lookup_get_uobject(uobj_get_type(_attrs, _type), (_attrs)->ufile, \ @@ -67,9 +70,12 @@ static inline void *_uobj_get_obj_read(struct ib_uobject *uobj) ((struct ib_##_object *)_uobj_get_obj_read( \ uobj_get_read(_type, _id, _attrs))) +struct ib_uobject *_uobj_get_write(enum uverbs_default_objects type, + u32 object_id, + struct uverbs_attr_bundle *attrs); + #define uobj_get_write(_type, _id, _attrs) \ - rdma_lookup_get_uobject(uobj_get_type(_attrs, _type), (_attrs)->ufile, \ - _uobj_check_id(_id), UVERBS_LOOKUP_WRITE) + _uobj_get_write(_type, _uobj_check_id(_id), _attrs) int __uobj_perform_destroy(const struct uverbs_api_object *obj, u32 id, const struct uverbs_attr_bundle *attrs); @@ -123,8 +129,10 @@ __uobj_alloc(const struct uverbs_api_object *obj, { struct ib_uobject *uobj = rdma_alloc_begin_uobject(obj, attrs->ufile); - if (!IS_ERR(uobj)) + if (!IS_ERR(uobj)) { *ib_dev = uobj->context->device; + attrs->context = uobj->context; + } return uobj; } diff --git a/include/rdma/uverbs_types.h b/include/rdma/uverbs_types.h index acb1bfa3cc99..175d761695e1 100644 --- a/include/rdma/uverbs_types.h +++ b/include/rdma/uverbs_types.h @@ -157,6 +157,7 @@ struct uverbs_obj_fd_type { extern const struct uverbs_obj_type_class uverbs_idr_class; extern const struct uverbs_obj_type_class uverbs_fd_class; +void uverbs_close_fd(struct file *f); #define UVERBS_BUILD_BUG_ON(cond) (sizeof(char[1 - 2 * !!(cond)]) - \ sizeof(char)) diff --git a/include/uapi/rdma/bnxt_re-abi.h b/include/uapi/rdma/bnxt_re-abi.h index a7a6111e50c7..dc52e3cf574c 100644 --- a/include/uapi/rdma/bnxt_re-abi.h +++ b/include/uapi/rdma/bnxt_re-abi.h @@ -44,6 +44,14 @@ #define BNXT_RE_ABI_VERSION 1 +#define BNXT_RE_CHIP_ID0_CHIP_NUM_SFT 0x00 +#define BNXT_RE_CHIP_ID0_CHIP_REV_SFT 0x10 +#define BNXT_RE_CHIP_ID0_CHIP_MET_SFT 0x18 + +enum { + BNXT_RE_UCNTX_CMASK_HAVE_CCTX = 0x1ULL +}; + struct bnxt_re_uctx_resp { __u32 dev_id; __u32 max_qp; @@ -51,6 +59,9 @@ struct bnxt_re_uctx_resp { __u32 cqe_sz; __u32 max_cqd; __u32 rsvd; + __aligned_u64 comp_mask; + __u32 chip_id0; + __u32 chip_id1; }; /* diff --git a/include/uapi/rdma/ib_user_verbs.h b/include/uapi/rdma/ib_user_verbs.h index 480d9a60b68e..0474c7400268 100644 --- a/include/uapi/rdma/ib_user_verbs.h +++ b/include/uapi/rdma/ib_user_verbs.h @@ -270,6 +270,8 @@ struct ib_uverbs_ex_query_device_resp { struct ib_uverbs_tm_caps tm_caps; struct ib_uverbs_cq_moderation_caps cq_moderation_caps; __aligned_u64 max_dm_size; + __u32 xrc_odp_caps; + __u32 reserved; }; struct ib_uverbs_query_port { diff --git a/include/uapi/rdma/mlx5_user_ioctl_cmds.h b/include/uapi/rdma/mlx5_user_ioctl_cmds.h index b8d121d457f1..8149d224030b 100644 --- a/include/uapi/rdma/mlx5_user_ioctl_cmds.h +++ b/include/uapi/rdma/mlx5_user_ioctl_cmds.h @@ -84,6 +84,14 @@ enum mlx5_ib_devx_obj_query_attrs { MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_OUT, }; +enum mlx5_ib_devx_obj_query_async_attrs { + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_HANDLE = (1U << UVERBS_ID_NS_SHIFT), + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_CMD_IN, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_FD, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_WR_ID, + MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_OUT_LEN, +}; + enum mlx5_ib_devx_query_eqn_attrs { MLX5_IB_ATTR_DEVX_QUERY_EQN_USER_VEC = (1U << UVERBS_ID_NS_SHIFT), MLX5_IB_ATTR_DEVX_QUERY_EQN_DEV_EQN, @@ -94,6 +102,7 @@ enum mlx5_ib_devx_obj_methods { MLX5_IB_METHOD_DEVX_OBJ_DESTROY, MLX5_IB_METHOD_DEVX_OBJ_MODIFY, MLX5_IB_METHOD_DEVX_OBJ_QUERY, + MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY, }; enum mlx5_ib_devx_umem_reg_attrs { @@ -113,11 +122,20 @@ enum mlx5_ib_devx_umem_methods { MLX5_IB_METHOD_DEVX_UMEM_DEREG, }; +enum mlx5_ib_devx_async_cmd_fd_alloc_attrs { + MLX5_IB_ATTR_DEVX_ASYNC_CMD_FD_ALLOC_HANDLE = (1U << UVERBS_ID_NS_SHIFT), +}; + +enum mlx5_ib_devx_async_cmd_fd_methods { + MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC = (1U << UVERBS_ID_NS_SHIFT), +}; + enum mlx5_ib_objects { MLX5_IB_OBJECT_DEVX = (1U << UVERBS_ID_NS_SHIFT), MLX5_IB_OBJECT_DEVX_OBJ, MLX5_IB_OBJECT_DEVX_UMEM, MLX5_IB_OBJECT_FLOW_MATCHER, + MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD, }; enum mlx5_ib_flow_matcher_create_attrs { diff --git a/include/uapi/rdma/mlx5_user_ioctl_verbs.h b/include/uapi/rdma/mlx5_user_ioctl_verbs.h index 4ef62c0e8452..4a701033b93f 100644 --- a/include/uapi/rdma/mlx5_user_ioctl_verbs.h +++ b/include/uapi/rdma/mlx5_user_ioctl_verbs.h @@ -51,5 +51,10 @@ enum mlx5_ib_uapi_flow_action_packet_reformat_type { MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL = 0x3, }; +struct mlx5_ib_uapi_devx_async_cmd_hdr { + __aligned_u64 wr_id; + __u8 out_data[]; +}; + #endif diff --git a/include/uapi/rdma/rdma_netlink.h b/include/uapi/rdma/rdma_netlink.h index 2e18b77a817f..5cc592728071 100644 --- a/include/uapi/rdma/rdma_netlink.h +++ b/include/uapi/rdma/rdma_netlink.h @@ -5,8 +5,7 @@ #include <linux/types.h> enum { - RDMA_NL_RDMA_CM = 1, - RDMA_NL_IWCM, + RDMA_NL_IWCM = 2, RDMA_NL_RSVD, RDMA_NL_LS, /* RDMA Local Services */ RDMA_NL_NLDEV, /* RDMA device interface */ @@ -14,8 +13,7 @@ enum { }; enum { - RDMA_NL_GROUP_CM = 1, - RDMA_NL_GROUP_IWPM, + RDMA_NL_GROUP_IWPM = 2, RDMA_NL_GROUP_LS, RDMA_NL_NUM_GROUPS }; @@ -24,15 +22,17 @@ enum { #define RDMA_NL_GET_OP(type) (type & ((1 << 10) - 1)) #define RDMA_NL_GET_TYPE(client, op) ((client << 10) + op) -enum { - RDMA_NL_RDMA_CM_ID_STATS = 0, - RDMA_NL_RDMA_CM_NUM_OPS -}; +/* The minimum version that the iwpm kernel supports */ +#define IWPM_UABI_VERSION_MIN 3 + +/* The latest version that the iwpm kernel supports */ +#define IWPM_UABI_VERSION 4 +/* iwarp port mapper message flags */ enum { - RDMA_NL_RDMA_CM_ATTR_SRC_ADDR = 1, - RDMA_NL_RDMA_CM_ATTR_DST_ADDR, - RDMA_NL_RDMA_CM_NUM_ATTR, + + /* Do not map the port for this IWPM request */ + IWPM_FLAGS_NO_PORT_MAP = (1 << 0), }; /* iwarp port mapper op-codes */ @@ -45,6 +45,7 @@ enum { RDMA_NL_IWPM_HANDLE_ERR, RDMA_NL_IWPM_MAPINFO, RDMA_NL_IWPM_MAPINFO_NUM, + RDMA_NL_IWPM_HELLO, RDMA_NL_IWPM_NUM_OPS }; @@ -83,20 +84,38 @@ enum { IWPM_NLA_MANAGE_MAPPING_UNSPEC = 0, IWPM_NLA_MANAGE_MAPPING_SEQ, IWPM_NLA_MANAGE_ADDR, - IWPM_NLA_MANAGE_MAPPED_LOC_ADDR, + IWPM_NLA_MANAGE_FLAGS, + IWPM_NLA_MANAGE_MAPPING_MAX +}; + +enum { + IWPM_NLA_RMANAGE_MAPPING_UNSPEC = 0, + IWPM_NLA_RMANAGE_MAPPING_SEQ, + IWPM_NLA_RMANAGE_ADDR, + IWPM_NLA_RMANAGE_MAPPED_LOC_ADDR, + /* The following maintains bisectability of rdma-core */ + IWPM_NLA_MANAGE_MAPPED_LOC_ADDR = IWPM_NLA_RMANAGE_MAPPED_LOC_ADDR, IWPM_NLA_RMANAGE_MAPPING_ERR, IWPM_NLA_RMANAGE_MAPPING_MAX }; -#define IWPM_NLA_MANAGE_MAPPING_MAX 3 -#define IWPM_NLA_QUERY_MAPPING_MAX 4 #define IWPM_NLA_MAPINFO_SEND_MAX 3 +#define IWPM_NLA_REMOVE_MAPPING_MAX 3 enum { IWPM_NLA_QUERY_MAPPING_UNSPEC = 0, IWPM_NLA_QUERY_MAPPING_SEQ, IWPM_NLA_QUERY_LOCAL_ADDR, IWPM_NLA_QUERY_REMOTE_ADDR, + IWPM_NLA_QUERY_FLAGS, + IWPM_NLA_QUERY_MAPPING_MAX, +}; + +enum { + IWPM_NLA_RQUERY_MAPPING_UNSPEC = 0, + IWPM_NLA_RQUERY_MAPPING_SEQ, + IWPM_NLA_RQUERY_LOCAL_ADDR, + IWPM_NLA_RQUERY_REMOTE_ADDR, IWPM_NLA_RQUERY_MAPPED_LOC_ADDR, IWPM_NLA_RQUERY_MAPPED_REM_ADDR, IWPM_NLA_RQUERY_MAPPING_ERR, @@ -114,6 +133,7 @@ enum { IWPM_NLA_MAPINFO_UNSPEC = 0, IWPM_NLA_MAPINFO_LOCAL_ADDR, IWPM_NLA_MAPINFO_MAPPED_ADDR, + IWPM_NLA_MAPINFO_FLAGS, IWPM_NLA_MAPINFO_MAX }; @@ -132,6 +152,12 @@ enum { IWPM_NLA_ERR_MAX }; +enum { + IWPM_NLA_HELLO_UNSPEC = 0, + IWPM_NLA_HELLO_ABI_VERSION, + IWPM_NLA_HELLO_MAX +}; + /* * Local service operations: * RESOLVE - The client requests the local service to resolve a path. @@ -229,9 +255,11 @@ enum rdma_nldev_command { RDMA_NLDEV_CMD_GET, /* can dump */ RDMA_NLDEV_CMD_SET, - /* 3 - 4 are free to use */ + RDMA_NLDEV_CMD_NEWLINK, - RDMA_NLDEV_CMD_PORT_GET = 5, /* can dump */ + RDMA_NLDEV_CMD_DELLINK, + + RDMA_NLDEV_CMD_PORT_GET, /* can dump */ /* 6 - 8 are free to use */ @@ -431,6 +459,20 @@ enum rdma_nldev_attr { RDMA_NLDEV_ATTR_DRIVER_U64, /* u64 */ /* + * Indexes to get/set secific entry, + * for QP use RDMA_NLDEV_ATTR_RES_LQPN + */ + RDMA_NLDEV_ATTR_RES_PDN, /* u32 */ + RDMA_NLDEV_ATTR_RES_CQN, /* u32 */ + RDMA_NLDEV_ATTR_RES_MRN, /* u32 */ + RDMA_NLDEV_ATTR_RES_CM_IDN, /* u32 */ + RDMA_NLDEV_ATTR_RES_CTXN, /* u32 */ + /* + * Identifies the rdma driver. eg: "rxe" or "siw" + */ + RDMA_NLDEV_ATTR_LINK_TYPE, /* string */ + + /* * Always the end */ RDMA_NLDEV_ATTR_MAX diff --git a/include/uapi/rdma/rdma_user_cm.h b/include/uapi/rdma/rdma_user_cm.h index 0d1e78ebad05..e42940a215a3 100644 --- a/include/uapi/rdma/rdma_user_cm.h +++ b/include/uapi/rdma/rdma_user_cm.h @@ -300,6 +300,10 @@ enum { RDMA_OPTION_ID_TOS = 0, RDMA_OPTION_ID_REUSEADDR = 1, RDMA_OPTION_ID_AFONLY = 2, + RDMA_OPTION_ID_ACK_TIMEOUT = 3 +}; + +enum { RDMA_OPTION_IB_PATH = 1 }; diff --git a/include/uapi/rdma/rdma_user_rxe.h b/include/uapi/rdma/rdma_user_rxe.h index 44ef6a3b7afc..aae2e696bb38 100644 --- a/include/uapi/rdma/rdma_user_rxe.h +++ b/include/uapi/rdma/rdma_user_rxe.h @@ -58,8 +58,7 @@ struct rxe_global_route { struct rxe_av { __u8 port_num; __u8 network_type; - __u16 reserved1; - __u32 reserved2; + __u8 dmac[6]; struct rxe_global_route grh; union { struct sockaddr_in _sockaddr_in; diff --git a/kernel/cgroup/rdma.c b/kernel/cgroup/rdma.c index d3bbb757ee49..1d75ae7f1cb7 100644 --- a/kernel/cgroup/rdma.c +++ b/kernel/cgroup/rdma.c @@ -313,10 +313,8 @@ EXPORT_SYMBOL(rdmacg_try_charge); * If IB stack wish a device to participate in rdma cgroup resource * tracking, it must invoke this API to register with rdma cgroup before * any user space application can start using the RDMA resources. - * Returns 0 on success or EINVAL when table length given is beyond - * supported size. */ -int rdmacg_register_device(struct rdmacg_device *device) +void rdmacg_register_device(struct rdmacg_device *device) { INIT_LIST_HEAD(&device->dev_node); INIT_LIST_HEAD(&device->rpools); @@ -324,7 +322,6 @@ int rdmacg_register_device(struct rdmacg_device *device) mutex_lock(&rdmacg_mutex); list_add_tail(&device->dev_node, &rdmacg_devices); mutex_unlock(&rdmacg_mutex); - return 0; } EXPORT_SYMBOL(rdmacg_register_device); diff --git a/kernel/events/core.c b/kernel/events/core.c index 5f59d848171e..f7086c388781 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -5473,7 +5473,7 @@ static void perf_mmap_close(struct vm_area_struct *vma) /* now it's safe to free the pages */ atomic_long_sub(rb->aux_nr_pages, &mmap_user->locked_vm); - vma->vm_mm->pinned_vm -= rb->aux_mmap_locked; + atomic64_sub(rb->aux_mmap_locked, &vma->vm_mm->pinned_vm); /* this has to be the last one */ rb_free_aux(rb); @@ -5546,7 +5546,7 @@ again: */ atomic_long_sub((size >> PAGE_SHIFT) + 1, &mmap_user->locked_vm); - vma->vm_mm->pinned_vm -= mmap_locked; + atomic64_sub(mmap_locked, &vma->vm_mm->pinned_vm); free_uid(mmap_user); out_put: @@ -5694,7 +5694,7 @@ accounting: lock_limit = rlimit(RLIMIT_MEMLOCK); lock_limit >>= PAGE_SHIFT; - locked = vma->vm_mm->pinned_vm + extra; + locked = atomic64_read(&vma->vm_mm->pinned_vm) + extra; if ((locked > lock_limit) && perf_paranoid_tracepoint_raw() && !capable(CAP_IPC_LOCK)) { @@ -5735,7 +5735,7 @@ accounting: unlock: if (!ret) { atomic_long_add(user_extra, &user->locked_vm); - vma->vm_mm->pinned_vm += extra; + atomic64_add(extra, &vma->vm_mm->pinned_vm); atomic_inc(&event->mmap_count); } else if (rb) { diff --git a/kernel/fork.c b/kernel/fork.c index 874e48c410f8..9dcd18aa210b 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -980,7 +980,7 @@ static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p, mm_pgtables_bytes_init(mm); mm->map_count = 0; mm->locked_vm = 0; - mm->pinned_vm = 0; + atomic64_set(&mm->pinned_vm, 0); memset(&mm->rss_stat, 0, sizeof(mm->rss_stat)); spin_lock_init(&mm->page_table_lock); spin_lock_init(&mm->arg_lock); diff --git a/lib/irq_poll.c b/lib/irq_poll.c index 86a709954f5a..2f17b488d58e 100644 --- a/lib/irq_poll.c +++ b/lib/irq_poll.c @@ -35,7 +35,7 @@ void irq_poll_sched(struct irq_poll *iop) local_irq_save(flags); list_add_tail(&iop->list, this_cpu_ptr(&blk_cpu_iopoll)); - __raise_softirq_irqoff(IRQ_POLL_SOFTIRQ); + raise_softirq_irqoff(IRQ_POLL_SOFTIRQ); local_irq_restore(flags); } EXPORT_SYMBOL(irq_poll_sched); diff --git a/lib/scatterlist.c b/lib/scatterlist.c index 9ba349e775ef..739dc9fe2c55 100644 --- a/lib/scatterlist.c +++ b/lib/scatterlist.c @@ -625,6 +625,32 @@ bool __sg_page_iter_next(struct sg_page_iter *piter) } EXPORT_SYMBOL(__sg_page_iter_next); +static int sg_dma_page_count(struct scatterlist *sg) +{ + return PAGE_ALIGN(sg->offset + sg_dma_len(sg)) >> PAGE_SHIFT; +} + +bool __sg_page_iter_dma_next(struct sg_dma_page_iter *dma_iter) +{ + struct sg_page_iter *piter = &dma_iter->base; + + if (!piter->__nents || !piter->sg) + return false; + + piter->sg_pgoffset += piter->__pg_advance; + piter->__pg_advance = 1; + + while (piter->sg_pgoffset >= sg_dma_page_count(piter->sg)) { + piter->sg_pgoffset -= sg_dma_page_count(piter->sg); + piter->sg = sg_next(piter->sg); + if (!--piter->__nents || !piter->sg) + return false; + } + + return true; +} +EXPORT_SYMBOL(__sg_page_iter_dma_next); + /** * sg_miter_start - start mapping iteration over a sg list * @miter: sg mapping iter to be started diff --git a/mm/debug.c b/mm/debug.c index 1611cf00a137..c0b31b6c3877 100644 --- a/mm/debug.c +++ b/mm/debug.c @@ -137,7 +137,7 @@ void dump_mm(const struct mm_struct *mm) "mmap_base %lu mmap_legacy_base %lu highest_vm_end %lu\n" "pgd %px mm_users %d mm_count %d pgtables_bytes %lu map_count %d\n" "hiwater_rss %lx hiwater_vm %lx total_vm %lx locked_vm %lx\n" - "pinned_vm %lx data_vm %lx exec_vm %lx stack_vm %lx\n" + "pinned_vm %llx data_vm %lx exec_vm %lx stack_vm %lx\n" "start_code %lx end_code %lx start_data %lx end_data %lx\n" "start_brk %lx brk %lx start_stack %lx\n" "arg_start %lx arg_end %lx env_start %lx env_end %lx\n" @@ -168,7 +168,8 @@ void dump_mm(const struct mm_struct *mm) mm_pgtables_bytes(mm), mm->map_count, mm->hiwater_rss, mm->hiwater_vm, mm->total_vm, mm->locked_vm, - mm->pinned_vm, mm->data_vm, mm->exec_vm, mm->stack_vm, + atomic64_read(&mm->pinned_vm), + mm->data_vm, mm->exec_vm, mm->stack_vm, mm->start_code, mm->end_code, mm->start_data, mm->end_data, mm->start_brk, mm->brk, mm->start_stack, mm->arg_start, mm->arg_end, mm->env_start, mm->env_end, diff --git a/net/rds/ib.h b/net/rds/ib.h index 752f92235a38..67a715b076ca 100644 --- a/net/rds/ib.h +++ b/net/rds/ib.h @@ -333,10 +333,8 @@ static inline void rds_ib_dma_sync_sg_for_cpu(struct ib_device *dev, unsigned int i; for_each_sg(sglist, sg, sg_dma_len, i) { - ib_dma_sync_single_for_cpu(dev, - ib_sg_dma_address(dev, sg), - ib_sg_dma_len(dev, sg), - direction); + ib_dma_sync_single_for_cpu(dev, sg_dma_address(sg), + sg_dma_len(sg), direction); } } #define ib_dma_sync_sg_for_cpu rds_ib_dma_sync_sg_for_cpu @@ -350,10 +348,8 @@ static inline void rds_ib_dma_sync_sg_for_device(struct ib_device *dev, unsigned int i; for_each_sg(sglist, sg, sg_dma_len, i) { - ib_dma_sync_single_for_device(dev, - ib_sg_dma_address(dev, sg), - ib_sg_dma_len(dev, sg), - direction); + ib_dma_sync_single_for_device(dev, sg_dma_address(sg), + sg_dma_len(sg), direction); } } #define ib_dma_sync_sg_for_device rds_ib_dma_sync_sg_for_device diff --git a/net/rds/ib_fmr.c b/net/rds/ib_fmr.c index e0f70c4051b6..31cf37da4510 100644 --- a/net/rds/ib_fmr.c +++ b/net/rds/ib_fmr.c @@ -108,8 +108,8 @@ static int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, page_cnt = 0; for (i = 0; i < sg_dma_len; ++i) { - unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]); - u64 dma_addr = ib_sg_dma_address(dev, &scat[i]); + unsigned int dma_len = sg_dma_len(&scat[i]); + u64 dma_addr = sg_dma_address(&scat[i]); if (dma_addr & ~PAGE_MASK) { if (i > 0) { @@ -148,8 +148,8 @@ static int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, page_cnt = 0; for (i = 0; i < sg_dma_len; ++i) { - unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]); - u64 dma_addr = ib_sg_dma_address(dev, &scat[i]); + unsigned int dma_len = sg_dma_len(&scat[i]); + u64 dma_addr = sg_dma_address(&scat[i]); for (j = 0; j < dma_len; j += PAGE_SIZE) dma_pages[page_cnt++] = diff --git a/net/rds/ib_frmr.c b/net/rds/ib_frmr.c index 6431a023ac89..688dcd68d4ea 100644 --- a/net/rds/ib_frmr.c +++ b/net/rds/ib_frmr.c @@ -181,8 +181,8 @@ static int rds_ib_map_frmr(struct rds_ib_device *rds_ibdev, ret = -EINVAL; for (i = 0; i < ibmr->sg_dma_len; ++i) { - unsigned int dma_len = ib_sg_dma_len(dev, &ibmr->sg[i]); - u64 dma_addr = ib_sg_dma_address(dev, &ibmr->sg[i]); + unsigned int dma_len = sg_dma_len(&ibmr->sg[i]); + u64 dma_addr = sg_dma_address(&ibmr->sg[i]); frmr->sg_byte_len += dma_len; if (dma_addr & ~PAGE_MASK) { diff --git a/net/rds/ib_recv.c b/net/rds/ib_recv.c index d395eec98959..70559854837e 100644 --- a/net/rds/ib_recv.c +++ b/net/rds/ib_recv.c @@ -346,8 +346,8 @@ static int rds_ib_recv_refill_one(struct rds_connection *conn, sge->length = sizeof(struct rds_header); sge = &recv->r_sge[1]; - sge->addr = ib_sg_dma_address(ic->i_cm_id->device, &recv->r_frag->f_sg); - sge->length = ib_sg_dma_len(ic->i_cm_id->device, &recv->r_frag->f_sg); + sge->addr = sg_dma_address(&recv->r_frag->f_sg); + sge->length = sg_dma_len(&recv->r_frag->f_sg); ret = 0; out: @@ -409,9 +409,7 @@ void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp) rdsdebug("recv %p ibinc %p page %p addr %lu\n", recv, recv->r_ibinc, sg_page(&recv->r_frag->f_sg), - (long) ib_sg_dma_address( - ic->i_cm_id->device, - &recv->r_frag->f_sg)); + (long)sg_dma_address(&recv->r_frag->f_sg)); /* XXX when can this fail? */ ret = ib_post_recv(ic->i_cm_id->qp, &recv->r_wr, NULL); diff --git a/net/rds/ib_send.c b/net/rds/ib_send.c index 09c46f2e97fa..18f2341202f8 100644 --- a/net/rds/ib_send.c +++ b/net/rds/ib_send.c @@ -646,16 +646,16 @@ int rds_ib_xmit(struct rds_connection *conn, struct rds_message *rm, if (i < work_alloc && scat != &rm->data.op_sg[rm->data.op_count]) { len = min(RDS_FRAG_SIZE, - ib_sg_dma_len(dev, scat) - rm->data.op_dmaoff); + sg_dma_len(scat) - rm->data.op_dmaoff); send->s_wr.num_sge = 2; - send->s_sge[1].addr = ib_sg_dma_address(dev, scat); + send->s_sge[1].addr = sg_dma_address(scat); send->s_sge[1].addr += rm->data.op_dmaoff; send->s_sge[1].length = len; bytes_sent += len; rm->data.op_dmaoff += len; - if (rm->data.op_dmaoff == ib_sg_dma_len(dev, scat)) { + if (rm->data.op_dmaoff == sg_dma_len(scat)) { scat++; rm->data.op_dmasg++; rm->data.op_dmaoff = 0; @@ -809,8 +809,8 @@ int rds_ib_xmit_atomic(struct rds_connection *conn, struct rm_atomic_op *op) } /* Convert our struct scatterlist to struct ib_sge */ - send->s_sge[0].addr = ib_sg_dma_address(ic->i_cm_id->device, op->op_sg); - send->s_sge[0].length = ib_sg_dma_len(ic->i_cm_id->device, op->op_sg); + send->s_sge[0].addr = sg_dma_address(op->op_sg); + send->s_sge[0].length = sg_dma_len(op->op_sg); send->s_sge[0].lkey = ic->i_pd->local_dma_lkey; rdsdebug("rva %Lx rpa %Lx len %u\n", op->op_remote_addr, @@ -922,9 +922,8 @@ int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op) for (j = 0; j < send->s_rdma_wr.wr.num_sge && scat != &op->op_sg[op->op_count]; j++) { - len = ib_sg_dma_len(ic->i_cm_id->device, scat); - send->s_sge[j].addr = - ib_sg_dma_address(ic->i_cm_id->device, scat); + len = sg_dma_len(scat); + send->s_sge[j].addr = sg_dma_address(scat); send->s_sge[j].length = len; send->s_sge[j].lkey = ic->i_pd->local_dma_lkey; |