diff options
author | Mark Brown | 2017-04-25 16:35:35 +0100 |
---|---|---|
committer | Mark Brown | 2017-04-25 16:35:35 +0100 |
commit | 9095bf25ea08135a5b74875dd0e3eeaddc4218a0 (patch) | |
tree | a4e6976eab0e7a5b6258281077cd5853c7e69514 /drivers/ntb | |
parent | cdf4275e957c6bad3756e98942341667f1d7de7d (diff) | |
parent | c1ae3cfa0e89fa1a7ecc4c99031f5e9ae99d9201 (diff) |
Merge tag 'v4.11-rc1' into regulator-arizona
Linux 4.11-rc1
Diffstat (limited to 'drivers/ntb')
-rw-r--r-- | drivers/ntb/hw/amd/ntb_hw_amd.c | 25 | ||||
-rw-r--r-- | drivers/ntb/hw/amd/ntb_hw_amd.h | 5 | ||||
-rw-r--r-- | drivers/ntb/hw/intel/ntb_hw_intel.c | 684 | ||||
-rw-r--r-- | drivers/ntb/hw/intel/ntb_hw_intel.h | 48 | ||||
-rw-r--r-- | drivers/ntb/ntb_transport.c | 34 | ||||
-rw-r--r-- | drivers/ntb/test/ntb_perf.c | 2 |
6 files changed, 763 insertions, 35 deletions
diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.c b/drivers/ntb/hw/amd/ntb_hw_amd.c index 6ccba0d862df..019a158e1128 100644 --- a/drivers/ntb/hw/amd/ntb_hw_amd.c +++ b/drivers/ntb/hw/amd/ntb_hw_amd.c @@ -138,11 +138,11 @@ static int amd_ntb_mw_set_trans(struct ntb_dev *ntb, int idx, base_addr = pci_resource_start(ndev->ntb.pdev, bar); if (bar != 1) { - xlat_reg = AMD_BAR23XLAT_OFFSET + ((bar - 2) << 3); - limit_reg = AMD_BAR23LMT_OFFSET + ((bar - 2) << 3); + xlat_reg = AMD_BAR23XLAT_OFFSET + ((bar - 2) << 2); + limit_reg = AMD_BAR23LMT_OFFSET + ((bar - 2) << 2); /* Set the limit if supported */ - limit = base_addr + size; + limit = size; /* set and verify setting the translation address */ write64(addr, peer_mmio + xlat_reg); @@ -164,14 +164,8 @@ static int amd_ntb_mw_set_trans(struct ntb_dev *ntb, int idx, xlat_reg = AMD_BAR1XLAT_OFFSET; limit_reg = AMD_BAR1LMT_OFFSET; - /* split bar addr range must all be 32 bit */ - if (addr & (~0ull << 32)) - return -EINVAL; - if ((addr + size) & (~0ull << 32)) - return -EINVAL; - /* Set the limit if supported */ - limit = base_addr + size; + limit = size; /* set and verify setting the translation address */ write64(addr, peer_mmio + xlat_reg); @@ -199,6 +193,11 @@ static int amd_link_is_up(struct amd_ntb_dev *ndev) if (!ndev->peer_sta) return NTB_LNK_STA_ACTIVE(ndev->cntl_sta); + if (ndev->peer_sta & AMD_LINK_UP_EVENT) { + ndev->peer_sta = 0; + return 1; + } + /* If peer_sta is reset or D0 event, the ISR has * started a timer to check link status of hardware. * So here just clear status bit. And if peer_sta is @@ -207,7 +206,7 @@ static int amd_link_is_up(struct amd_ntb_dev *ndev) */ if (ndev->peer_sta & AMD_PEER_RESET_EVENT) ndev->peer_sta &= ~AMD_PEER_RESET_EVENT; - else if (ndev->peer_sta & AMD_PEER_D0_EVENT) + else if (ndev->peer_sta & (AMD_PEER_D0_EVENT | AMD_LINK_DOWN_EVENT)) ndev->peer_sta = 0; return 0; @@ -491,6 +490,8 @@ static void amd_handle_event(struct amd_ntb_dev *ndev, int vec) break; case AMD_PEER_D3_EVENT: case AMD_PEER_PMETO_EVENT: + case AMD_LINK_UP_EVENT: + case AMD_LINK_DOWN_EVENT: amd_ack_smu(ndev, status); /* link down */ @@ -598,7 +599,7 @@ static int ndev_init_isr(struct amd_ntb_dev *ndev, err_msix_request: while (i-- > 0) - free_irq(ndev->msix[i].vector, ndev); + free_irq(ndev->msix[i].vector, &ndev->vec[i]); pci_disable_msix(pdev); err_msix_enable: kfree(ndev->msix); diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.h b/drivers/ntb/hw/amd/ntb_hw_amd.h index 2eac3cd3e646..13d73ed94a52 100644 --- a/drivers/ntb/hw/amd/ntb_hw_amd.h +++ b/drivers/ntb/hw/amd/ntb_hw_amd.h @@ -148,9 +148,12 @@ enum { AMD_PEER_D3_EVENT = BIT(2), AMD_PEER_PMETO_EVENT = BIT(3), AMD_PEER_D0_EVENT = BIT(4), + AMD_LINK_UP_EVENT = BIT(5), + AMD_LINK_DOWN_EVENT = BIT(6), AMD_EVENT_INTMASK = (AMD_PEER_FLUSH_EVENT | AMD_PEER_RESET_EVENT | AMD_PEER_D3_EVENT | - AMD_PEER_PMETO_EVENT | AMD_PEER_D0_EVENT), + AMD_PEER_PMETO_EVENT | AMD_PEER_D0_EVENT | + AMD_LINK_UP_EVENT | AMD_LINK_DOWN_EVENT), AMD_PMESTAT_OFFSET = 0x480, AMD_PMSGTRIG_OFFSET = 0x490, diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.c b/drivers/ntb/hw/intel/ntb_hw_intel.c index 7310a261c858..c00238491673 100644 --- a/drivers/ntb/hw/intel/ntb_hw_intel.c +++ b/drivers/ntb/hw/intel/ntb_hw_intel.c @@ -86,7 +86,12 @@ static const struct intel_ntb_xlat_reg xeon_pri_xlat; static const struct intel_ntb_xlat_reg xeon_sec_xlat; static struct intel_b2b_addr xeon_b2b_usd_addr; static struct intel_b2b_addr xeon_b2b_dsd_addr; +static const struct intel_ntb_reg skx_reg; +static const struct intel_ntb_alt_reg skx_pri_reg; +static const struct intel_ntb_alt_reg skx_b2b_reg; +static const struct intel_ntb_xlat_reg skx_sec_xlat; static const struct ntb_dev_ops intel_ntb_ops; +static const struct ntb_dev_ops intel_ntb3_ops; static const struct file_operations intel_ntb_debugfs_info; static struct dentry *debugfs_dir; @@ -145,6 +150,9 @@ module_param_named(xeon_b2b_dsd_bar5_addr32, MODULE_PARM_DESC(xeon_b2b_dsd_bar5_addr32, "XEON B2B DSD split-BAR 5 32-bit address"); +static inline enum ntb_topo xeon_ppd_topo(struct intel_ntb_dev *ndev, u8 ppd); +static int xeon_init_isr(struct intel_ntb_dev *ndev); + #ifndef ioread64 #ifdef readq #define ioread64 readq @@ -206,6 +214,14 @@ static inline int pdev_is_xeon(struct pci_dev *pdev) return 0; } +static inline int pdev_is_skx_xeon(struct pci_dev *pdev) +{ + if (pdev->device == PCI_DEVICE_ID_INTEL_NTB_B2B_SKX) + return 1; + + return 0; +} + static inline void ndev_reset_unsafe_flags(struct intel_ntb_dev *ndev) { ndev->unsafe_flags = 0; @@ -390,6 +406,9 @@ static irqreturn_t ndev_interrupt(struct intel_ntb_dev *ndev, int vec) vec_mask = ndev_vec_mask(ndev, vec); + if ((ndev->hwerr_flags & NTB_HWERR_MSIX_VECTOR32_BAD) && (vec == 31)) + vec_mask |= ndev->db_link_mask; + dev_dbg(ndev_dev(ndev), "vec %d vec_mask %llx\n", vec, vec_mask); ndev->last_ts = jiffies; @@ -409,6 +428,9 @@ static irqreturn_t ndev_vec_isr(int irq, void *dev) { struct intel_ntb_vec *nvec = dev; + dev_dbg(ndev_dev(nvec->ndev), "irq: %d nvec->num: %d\n", + irq, nvec->num); + return ndev_interrupt(nvec->ndev, nvec->num); } @@ -465,14 +487,14 @@ static int ndev_init_isr(struct intel_ntb_dev *ndev, goto err_msix_request; } - dev_dbg(ndev_dev(ndev), "Using msix interrupts\n"); + dev_dbg(ndev_dev(ndev), "Using %d msix interrupts\n", msix_count); ndev->db_vec_count = msix_count; ndev->db_vec_shift = msix_shift; return 0; err_msix_request: while (i-- > 0) - free_irq(ndev->msix[i].vector, ndev); + free_irq(ndev->msix[i].vector, &ndev->vec[i]); pci_disable_msix(pdev); err_msix_enable: kfree(ndev->msix); @@ -547,8 +569,171 @@ static void ndev_deinit_isr(struct intel_ntb_dev *ndev) } } -static ssize_t ndev_debugfs_read(struct file *filp, char __user *ubuf, - size_t count, loff_t *offp) +static ssize_t ndev_ntb3_debugfs_read(struct file *filp, char __user *ubuf, + size_t count, loff_t *offp) +{ + struct intel_ntb_dev *ndev; + void __iomem *mmio; + char *buf; + size_t buf_size; + ssize_t ret, off; + union { u64 v64; u32 v32; u16 v16; } u; + + ndev = filp->private_data; + mmio = ndev->self_mmio; + + buf_size = min(count, 0x800ul); + + buf = kmalloc(buf_size, GFP_KERNEL); + if (!buf) + return -ENOMEM; + + off = 0; + + off += scnprintf(buf + off, buf_size - off, + "NTB Device Information:\n"); + + off += scnprintf(buf + off, buf_size - off, + "Connection Topology -\t%s\n", + ntb_topo_string(ndev->ntb.topo)); + + off += scnprintf(buf + off, buf_size - off, + "NTB CTL -\t\t%#06x\n", ndev->ntb_ctl); + off += scnprintf(buf + off, buf_size - off, + "LNK STA -\t\t%#06x\n", ndev->lnk_sta); + + if (!ndev->reg->link_is_up(ndev)) + off += scnprintf(buf + off, buf_size - off, + "Link Status -\t\tDown\n"); + else { + off += scnprintf(buf + off, buf_size - off, + "Link Status -\t\tUp\n"); + off += scnprintf(buf + off, buf_size - off, + "Link Speed -\t\tPCI-E Gen %u\n", + NTB_LNK_STA_SPEED(ndev->lnk_sta)); + off += scnprintf(buf + off, buf_size - off, + "Link Width -\t\tx%u\n", + NTB_LNK_STA_WIDTH(ndev->lnk_sta)); + } + + off += scnprintf(buf + off, buf_size - off, + "Memory Window Count -\t%u\n", ndev->mw_count); + off += scnprintf(buf + off, buf_size - off, + "Scratchpad Count -\t%u\n", ndev->spad_count); + off += scnprintf(buf + off, buf_size - off, + "Doorbell Count -\t%u\n", ndev->db_count); + off += scnprintf(buf + off, buf_size - off, + "Doorbell Vector Count -\t%u\n", ndev->db_vec_count); + off += scnprintf(buf + off, buf_size - off, + "Doorbell Vector Shift -\t%u\n", ndev->db_vec_shift); + + off += scnprintf(buf + off, buf_size - off, + "Doorbell Valid Mask -\t%#llx\n", ndev->db_valid_mask); + off += scnprintf(buf + off, buf_size - off, + "Doorbell Link Mask -\t%#llx\n", ndev->db_link_mask); + off += scnprintf(buf + off, buf_size - off, + "Doorbell Mask Cached -\t%#llx\n", ndev->db_mask); + + u.v64 = ndev_db_read(ndev, mmio + ndev->self_reg->db_mask); + off += scnprintf(buf + off, buf_size - off, + "Doorbell Mask -\t\t%#llx\n", u.v64); + + u.v64 = ndev_db_read(ndev, mmio + ndev->self_reg->db_bell); + off += scnprintf(buf + off, buf_size - off, + "Doorbell Bell -\t\t%#llx\n", u.v64); + + off += scnprintf(buf + off, buf_size - off, + "\nNTB Incoming XLAT:\n"); + + u.v64 = ioread64(mmio + SKX_IMBAR1XBASE_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "IMBAR1XBASE -\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_IMBAR2XBASE_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "IMBAR2XBASE -\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_IMBAR1XLMT_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "IMBAR1XLMT -\t\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_IMBAR2XLMT_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "IMBAR2XLMT -\t\t\t%#018llx\n", u.v64); + + if (ntb_topo_is_b2b(ndev->ntb.topo)) { + off += scnprintf(buf + off, buf_size - off, + "\nNTB Outgoing B2B XLAT:\n"); + + u.v64 = ioread64(mmio + SKX_EMBAR1XBASE_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "EMBAR1XBASE -\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_EMBAR2XBASE_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "EMBAR2XBASE -\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_EMBAR1XLMT_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "EMBAR1XLMT -\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_EMBAR2XLMT_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "EMBAR2XLMT -\t\t%#018llx\n", u.v64); + + off += scnprintf(buf + off, buf_size - off, + "\nNTB Secondary BAR:\n"); + + u.v64 = ioread64(mmio + SKX_EMBAR0_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "EMBAR0 -\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_EMBAR1_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "EMBAR1 -\t\t%#018llx\n", u.v64); + + u.v64 = ioread64(mmio + SKX_EMBAR2_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "EMBAR2 -\t\t%#018llx\n", u.v64); + } + + off += scnprintf(buf + off, buf_size - off, + "\nNTB Statistics:\n"); + + u.v16 = ioread16(mmio + SKX_USMEMMISS_OFFSET); + off += scnprintf(buf + off, buf_size - off, + "Upstream Memory Miss -\t%u\n", u.v16); + + off += scnprintf(buf + off, buf_size - off, + "\nNTB Hardware Errors:\n"); + + if (!pci_read_config_word(ndev->ntb.pdev, + SKX_DEVSTS_OFFSET, &u.v16)) + off += scnprintf(buf + off, buf_size - off, + "DEVSTS -\t\t%#06x\n", u.v16); + + if (!pci_read_config_word(ndev->ntb.pdev, + SKX_LINK_STATUS_OFFSET, &u.v16)) + off += scnprintf(buf + off, buf_size - off, + "LNKSTS -\t\t%#06x\n", u.v16); + + if (!pci_read_config_dword(ndev->ntb.pdev, + SKX_UNCERRSTS_OFFSET, &u.v32)) + off += scnprintf(buf + off, buf_size - off, + "UNCERRSTS -\t\t%#06x\n", u.v32); + + if (!pci_read_config_dword(ndev->ntb.pdev, + SKX_CORERRSTS_OFFSET, &u.v32)) + off += scnprintf(buf + off, buf_size - off, + "CORERRSTS -\t\t%#06x\n", u.v32); + + ret = simple_read_from_buffer(ubuf, count, offp, buf, off); + kfree(buf); + return ret; +} + +static ssize_t ndev_ntb_debugfs_read(struct file *filp, char __user *ubuf, + size_t count, loff_t *offp) { struct intel_ntb_dev *ndev; struct pci_dev *pdev; @@ -813,6 +998,20 @@ static ssize_t ndev_debugfs_read(struct file *filp, char __user *ubuf, return ret; } +static ssize_t ndev_debugfs_read(struct file *filp, char __user *ubuf, + size_t count, loff_t *offp) +{ + struct intel_ntb_dev *ndev = filp->private_data; + + if (pdev_is_xeon(ndev->ntb.pdev) || + pdev_is_atom(ndev->ntb.pdev)) + return ndev_ntb_debugfs_read(filp, ubuf, count, offp); + else if (pdev_is_skx_xeon(ndev->ntb.pdev)) + return ndev_ntb3_debugfs_read(filp, ubuf, count, offp); + + return -ENXIO; +} + static void ndev_init_debugfs(struct intel_ntb_dev *ndev) { if (!debugfs_dir) { @@ -1428,6 +1627,405 @@ static void atom_deinit_dev(struct intel_ntb_dev *ndev) atom_deinit_isr(ndev); } +/* Skylake Xeon NTB */ + +static int skx_poll_link(struct intel_ntb_dev *ndev) +{ + u16 reg_val; + int rc; + + ndev->reg->db_iowrite(ndev->db_link_mask, + ndev->self_mmio + + ndev->self_reg->db_clear); + + rc = pci_read_config_word(ndev->ntb.pdev, + SKX_LINK_STATUS_OFFSET, ®_val); + if (rc) + return 0; + + if (reg_val == ndev->lnk_sta) + return 0; + + ndev->lnk_sta = reg_val; + + return 1; +} + +static u64 skx_db_ioread(void __iomem *mmio) +{ + return ioread64(mmio); +} + +static void skx_db_iowrite(u64 bits, void __iomem *mmio) +{ + iowrite64(bits, mmio); +} + +static int skx_init_isr(struct intel_ntb_dev *ndev) +{ + int i; + + /* + * The MSIX vectors and the interrupt status bits are not lined up + * on Skylake. By default the link status bit is bit 32, however it + * is by default MSIX vector0. We need to fixup to line them up. + * The vectors at reset is 1-32,0. We need to reprogram to 0-32. + */ + + for (i = 0; i < SKX_DB_MSIX_VECTOR_COUNT; i++) + iowrite8(i, ndev->self_mmio + SKX_INTVEC_OFFSET + i); + + /* move link status down one as workaround */ + if (ndev->hwerr_flags & NTB_HWERR_MSIX_VECTOR32_BAD) { + iowrite8(SKX_DB_MSIX_VECTOR_COUNT - 2, + ndev->self_mmio + SKX_INTVEC_OFFSET + + (SKX_DB_MSIX_VECTOR_COUNT - 1)); + } + + return ndev_init_isr(ndev, SKX_DB_MSIX_VECTOR_COUNT, + SKX_DB_MSIX_VECTOR_COUNT, + SKX_DB_MSIX_VECTOR_SHIFT, + SKX_DB_TOTAL_SHIFT); +} + +static int skx_setup_b2b_mw(struct intel_ntb_dev *ndev, + const struct intel_b2b_addr *addr, + const struct intel_b2b_addr *peer_addr) +{ + struct pci_dev *pdev; + void __iomem *mmio; + resource_size_t bar_size; + phys_addr_t bar_addr; + int b2b_bar; + u8 bar_sz; + + pdev = ndev_pdev(ndev); + mmio = ndev->self_mmio; + + if (ndev->b2b_idx == UINT_MAX) { + dev_dbg(ndev_dev(ndev), "not using b2b mw\n"); + b2b_bar = 0; + ndev->b2b_off = 0; + } else { + b2b_bar = ndev_mw_to_bar(ndev, ndev->b2b_idx); + if (b2b_bar < 0) + return -EIO; + + dev_dbg(ndev_dev(ndev), "using b2b mw bar %d\n", b2b_bar); + + bar_size = pci_resource_len(ndev->ntb.pdev, b2b_bar); + + dev_dbg(ndev_dev(ndev), "b2b bar size %#llx\n", bar_size); + + if (b2b_mw_share && ((bar_size >> 1) >= XEON_B2B_MIN_SIZE)) { + dev_dbg(ndev_dev(ndev), + "b2b using first half of bar\n"); + ndev->b2b_off = bar_size >> 1; + } else if (bar_size >= XEON_B2B_MIN_SIZE) { + dev_dbg(ndev_dev(ndev), + "b2b using whole bar\n"); + ndev->b2b_off = 0; + --ndev->mw_count; + } else { + dev_dbg(ndev_dev(ndev), + "b2b bar size is too small\n"); + return -EIO; + } + } + + /* + * Reset the secondary bar sizes to match the primary bar sizes, + * except disable or halve the size of the b2b secondary bar. + */ + pci_read_config_byte(pdev, SKX_IMBAR1SZ_OFFSET, &bar_sz); + dev_dbg(ndev_dev(ndev), "IMBAR1SZ %#x\n", bar_sz); + if (b2b_bar == 1) { + if (ndev->b2b_off) + bar_sz -= 1; + else + bar_sz = 0; + } + + pci_write_config_byte(pdev, SKX_EMBAR1SZ_OFFSET, bar_sz); + pci_read_config_byte(pdev, SKX_EMBAR1SZ_OFFSET, &bar_sz); + dev_dbg(ndev_dev(ndev), "EMBAR1SZ %#x\n", bar_sz); + + pci_read_config_byte(pdev, SKX_IMBAR2SZ_OFFSET, &bar_sz); + dev_dbg(ndev_dev(ndev), "IMBAR2SZ %#x\n", bar_sz); + if (b2b_bar == 2) { + if (ndev->b2b_off) + bar_sz -= 1; + else + bar_sz = 0; + } + + pci_write_config_byte(pdev, SKX_EMBAR2SZ_OFFSET, bar_sz); + pci_read_config_byte(pdev, SKX_EMBAR2SZ_OFFSET, &bar_sz); + dev_dbg(ndev_dev(ndev), "EMBAR2SZ %#x\n", bar_sz); + + /* SBAR01 hit by first part of the b2b bar */ + if (b2b_bar == 0) + bar_addr = addr->bar0_addr; + else if (b2b_bar == 1) + bar_addr = addr->bar2_addr64; + else if (b2b_bar == 2) + bar_addr = addr->bar4_addr64; + else + return -EIO; + + /* setup incoming bar limits == base addrs (zero length windows) */ + bar_addr = addr->bar2_addr64 + (b2b_bar == 1 ? ndev->b2b_off : 0); + iowrite64(bar_addr, mmio + SKX_IMBAR1XLMT_OFFSET); + bar_addr = ioread64(mmio + SKX_IMBAR1XLMT_OFFSET); + dev_dbg(ndev_dev(ndev), "IMBAR1XLMT %#018llx\n", bar_addr); + + bar_addr = addr->bar4_addr64 + (b2b_bar == 2 ? ndev->b2b_off : 0); + iowrite64(bar_addr, mmio + SKX_IMBAR2XLMT_OFFSET); + bar_addr = ioread64(mmio + SKX_IMBAR2XLMT_OFFSET); + dev_dbg(ndev_dev(ndev), "IMBAR2XLMT %#018llx\n", bar_addr); + + /* zero incoming translation addrs */ + iowrite64(0, mmio + SKX_IMBAR1XBASE_OFFSET); + iowrite64(0, mmio + SKX_IMBAR2XBASE_OFFSET); + + ndev->peer_mmio = ndev->self_mmio; + + return 0; +} + +static int skx_init_ntb(struct intel_ntb_dev *ndev) +{ + int rc; + + + ndev->mw_count = XEON_MW_COUNT; + ndev->spad_count = SKX_SPAD_COUNT; + ndev->db_count = SKX_DB_COUNT; + ndev->db_link_mask = SKX_DB_LINK_BIT; + + /* DB fixup for using 31 right now */ + if (ndev->hwerr_flags & NTB_HWERR_MSIX_VECTOR32_BAD) + ndev->db_link_mask |= BIT_ULL(31); + + switch (ndev->ntb.topo) { + case NTB_TOPO_B2B_USD: + case NTB_TOPO_B2B_DSD: + ndev->self_reg = &skx_pri_reg; + ndev->peer_reg = &skx_b2b_reg; + ndev->xlat_reg = &skx_sec_xlat; + + if (ndev->ntb.topo == NTB_TOPO_B2B_USD) { + rc = skx_setup_b2b_mw(ndev, + &xeon_b2b_dsd_addr, + &xeon_b2b_usd_addr); + } else { + rc = skx_setup_b2b_mw(ndev, + &xeon_b2b_usd_addr, + &xeon_b2b_dsd_addr); + } + + if (rc) + return rc; + + /* Enable Bus Master and Memory Space on the secondary side */ + iowrite16(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER, + ndev->self_mmio + SKX_SPCICMD_OFFSET); + + break; + + default: + return -EINVAL; + } + + ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1; + + ndev->reg->db_iowrite(ndev->db_valid_mask, + ndev->self_mmio + + ndev->self_reg->db_mask); + + return 0; +} + +static int skx_init_dev(struct intel_ntb_dev *ndev) +{ + struct pci_dev *pdev; + u8 ppd; + int rc; + + pdev = ndev_pdev(ndev); + + ndev->reg = &skx_reg; + + rc = pci_read_config_byte(pdev, XEON_PPD_OFFSET, &ppd); + if (rc) + return -EIO; + + ndev->ntb.topo = xeon_ppd_topo(ndev, ppd); + dev_dbg(ndev_dev(ndev), "ppd %#x topo %s\n", ppd, + ntb_topo_string(ndev->ntb.topo)); + if (ndev->ntb.topo == NTB_TOPO_NONE) + return -EINVAL; + + if (pdev_is_skx_xeon(pdev)) + ndev->hwerr_flags |= NTB_HWERR_MSIX_VECTOR32_BAD; + + rc = skx_init_ntb(ndev); + if (rc) + return rc; + + return skx_init_isr(ndev); +} + +static int intel_ntb3_link_enable(struct ntb_dev *ntb, + enum ntb_speed max_speed, + enum ntb_width max_width) +{ + struct intel_ntb_dev *ndev; + u32 ntb_ctl; + + ndev = container_of(ntb, struct intel_ntb_dev, ntb); + + dev_dbg(ndev_dev(ndev), + "Enabling link with max_speed %d max_width %d\n", + max_speed, max_width); + + if (max_speed != NTB_SPEED_AUTO) + dev_dbg(ndev_dev(ndev), "ignoring max_speed %d\n", max_speed); + if (max_width != NTB_WIDTH_AUTO) + dev_dbg(ndev_dev(ndev), "ignoring max_width %d\n", max_width); + + ntb_ctl = ioread32(ndev->self_mmio + ndev->reg->ntb_ctl); + ntb_ctl &= ~(NTB_CTL_DISABLE | NTB_CTL_CFG_LOCK); + ntb_ctl |= NTB_CTL_P2S_BAR2_SNOOP | NTB_CTL_S2P_BAR2_SNOOP; + ntb_ctl |= NTB_CTL_P2S_BAR4_SNOOP | NTB_CTL_S2P_BAR4_SNOOP; + iowrite32(ntb_ctl, ndev->self_mmio + ndev->reg->ntb_ctl); + + return 0; +} +static int intel_ntb3_mw_set_trans(struct ntb_dev *ntb, int idx, + dma_addr_t addr, resource_size_t size) +{ + struct intel_ntb_dev *ndev = ntb_ndev(ntb); + unsigned long xlat_reg, limit_reg; + resource_size_t bar_size, mw_size; + void __iomem *mmio; + u64 base, limit, reg_val; + int bar; + + if (idx >= ndev->b2b_idx && !ndev->b2b_off) + idx += 1; + + bar = ndev_mw_to_bar(ndev, idx); + if (bar < 0) + return bar; + + bar_size = pci_resource_len(ndev->ntb.pdev, bar); + + if (idx == ndev->b2b_idx) + mw_size = bar_size - ndev->b2b_off; + else + mw_size = bar_size; + + /* hardware requires that addr is aligned to bar size */ + if (addr & (bar_size - 1)) + return -EINVAL; + + /* make sure the range fits in the usable mw size */ + if (size > mw_size) + return -EINVAL; + + mmio = ndev->self_mmio; + xlat_reg = ndev->xlat_reg->bar2_xlat + (idx * 0x10); + limit_reg = ndev->xlat_reg->bar2_limit + (idx * 0x10); + base = pci_resource_start(ndev->ntb.pdev, bar); + + /* Set the limit if supported, if size is not mw_size */ + if (limit_reg && size != mw_size) + limit = base + size; + else + limit = base + mw_size; + + /* set and verify setting the translation address */ + iowrite64(addr, mmio + xlat_reg); + reg_val = ioread64(mmio + xlat_reg); + if (reg_val != addr) { + iowrite64(0, mmio + xlat_reg); + return -EIO; + } + + dev_dbg(ndev_dev(ndev), "BAR %d IMBARXBASE: %#Lx\n", bar, reg_val); + + /* set and verify setting the limit */ + iowrite64(limit, mmio + limit_reg); + reg_val = ioread64(mmio + limit_reg); + if (reg_val != limit) { + iowrite64(base, mmio + limit_reg); + iowrite64(0, mmio + xlat_reg); + return -EIO; + } + + dev_dbg(ndev_dev(ndev), "BAR %d IMBARXLMT: %#Lx\n", bar, reg_val); + + /* setup the EP */ + limit_reg = ndev->xlat_reg->bar2_limit + (idx * 0x10) + 0x4000; + base = ioread64(mmio + SKX_EMBAR1_OFFSET + (8 * idx)); + base &= ~0xf; + + if (limit_reg && size != mw_size) + limit = base + size; + else + limit = base + mw_size; + + /* set and verify setting the limit */ + iowrite64(limit, mmio + limit_reg); + reg_val = ioread64(mmio + limit_reg); + if (reg_val != limit) { + iowrite64(base, mmio + limit_reg); + iowrite64(0, mmio + xlat_reg); + return -EIO; + } + + dev_dbg(ndev_dev(ndev), "BAR %d EMBARXLMT: %#Lx\n", bar, reg_val); + + return 0; +} + +static int intel_ntb3_peer_db_set(struct ntb_dev *ntb, u64 db_bits) +{ + struct intel_ntb_dev *ndev = ntb_ndev(ntb); + int bit; + + if (db_bits & ~ndev->db_valid_mask) + return -EINVAL; + + while (db_bits) { + bit = __ffs(db_bits); + iowrite32(1, ndev->peer_mmio + + ndev->peer_reg->db_bell + (bit * 4)); + db_bits &= db_bits - 1; + } + + return 0; +} + +static u64 intel_ntb3_db_read(struct ntb_dev *ntb) +{ + struct intel_ntb_dev *ndev = ntb_ndev(ntb); + + return ndev_db_read(ndev, + ndev->self_mmio + + ndev->self_reg->db_clear); +} + +static int intel_ntb3_db_clear(struct ntb_dev *ntb, u64 db_bits) +{ + struct intel_ntb_dev *ndev = ntb_ndev(ntb); + + return ndev_db_write(ndev, db_bits, + ndev->self_mmio + + ndev->self_reg->db_clear); +} + /* XEON */ static u64 xeon_db_ioread(void __iomem *mmio) @@ -2120,6 +2718,24 @@ static int intel_ntb_pci_probe(struct pci_dev *pdev, if (rc) goto err_init_dev; + } else if (pdev_is_skx_xeon(pdev)) { + ndev = kzalloc_node(sizeof(*ndev), GFP_KERNEL, node); + if (!ndev) { + rc = -ENOMEM; + goto err_ndev; + } + + ndev_init_struct(ndev, pdev); + ndev->ntb.ops = &intel_ntb3_ops; + + rc = intel_ntb_init_pci(ndev, pdev); + if (rc) + goto err_init_pci; + + rc = skx_init_dev(ndev); + if (rc) + goto err_init_dev; + } else { rc = -EINVAL; goto err_ndev; @@ -2143,7 +2759,7 @@ err_register: ndev_deinit_debugfs(ndev); if (pdev_is_atom(pdev)) atom_deinit_dev(ndev); - else if (pdev_is_xeon(pdev)) + else if (pdev_is_xeon(pdev) || pdev_is_skx_xeon(pdev)) xeon_deinit_dev(ndev); err_init_dev: intel_ntb_deinit_pci(ndev); @@ -2161,7 +2777,7 @@ static void intel_ntb_pci_remove(struct pci_dev *pdev) ndev_deinit_debugfs(ndev); if (pdev_is_atom(pdev)) atom_deinit_dev(ndev); - else if (pdev_is_xeon(pdev)) + else if (pdev_is_xeon(pdev) || pdev_is_skx_xeon(pdev)) xeon_deinit_dev(ndev); intel_ntb_deinit_pci(ndev); kfree(ndev); @@ -2257,6 +2873,36 @@ static struct intel_b2b_addr xeon_b2b_dsd_addr = { .bar5_addr32 = XEON_B2B_BAR5_ADDR32, }; +static const struct intel_ntb_reg skx_reg = { + .poll_link = skx_poll_link, + .link_is_up = xeon_link_is_up, + .db_ioread = skx_db_ioread, + .db_iowrite = skx_db_iowrite, + .db_size = sizeof(u64), + .ntb_ctl = SKX_NTBCNTL_OFFSET, + .mw_bar = {2, 4}, +}; + +static const struct intel_ntb_alt_reg skx_pri_reg = { + .db_bell = SKX_EM_DOORBELL_OFFSET, + .db_clear = SKX_IM_INT_STATUS_OFFSET, + .db_mask = SKX_IM_INT_DISABLE_OFFSET, + .spad = SKX_IM_SPAD_OFFSET, +}; + +static const struct intel_ntb_alt_reg skx_b2b_reg = { + .db_bell = SKX_IM_DOORBELL_OFFSET, + .db_clear = SKX_EM_INT_STATUS_OFFSET, + .db_mask = SKX_EM_INT_DISABLE_OFFSET, + .spad = SKX_B2B_SPAD_OFFSET, +}; + +static const struct intel_ntb_xlat_reg skx_sec_xlat = { +/* .bar0_base = SKX_EMBAR0_OFFSET, */ + .bar2_limit = SKX_IMBAR1XLMT_OFFSET, + .bar2_xlat = SKX_IMBAR1XBASE_OFFSET, +}; + /* operations for primary side of local ntb */ static const struct ntb_dev_ops intel_ntb_ops = { .mw_count = intel_ntb_mw_count, @@ -2284,6 +2930,31 @@ static const struct ntb_dev_ops intel_ntb_ops = { .peer_spad_write = intel_ntb_peer_spad_write, }; +static const struct ntb_dev_ops intel_ntb3_ops = { + .mw_count = intel_ntb_mw_count, + .mw_get_range = intel_ntb_mw_get_range, + .mw_set_trans = intel_ntb3_mw_set_trans, + .link_is_up = intel_ntb_link_is_up, + .link_enable = intel_ntb3_link_enable, + .link_disable = intel_ntb_link_disable, + .db_valid_mask = intel_ntb_db_valid_mask, + .db_vector_count = intel_ntb_db_vector_count, + .db_vector_mask = intel_ntb_db_vector_mask, + .db_read = intel_ntb3_db_read, + .db_clear = intel_ntb3_db_clear, + .db_set_mask = intel_ntb_db_set_mask, + .db_clear_mask = intel_ntb_db_clear_mask, + .peer_db_addr = intel_ntb_peer_db_addr, + .peer_db_set = intel_ntb3_peer_db_set, + .spad_is_unsafe = intel_ntb_spad_is_unsafe, + .spad_count = intel_ntb_spad_count, + .spad_read = intel_ntb_spad_read, + .spad_write = intel_ntb_spad_write, + .peer_spad_addr = intel_ntb_peer_spad_addr, + .peer_spad_read = intel_ntb_peer_spad_read, + .peer_spad_write = intel_ntb_peer_spad_write, +}; + static const struct file_operations intel_ntb_debugfs_info = { .owner = THIS_MODULE, .open = simple_open, @@ -2307,6 +2978,7 @@ static const struct pci_device_id intel_ntb_pci_tbl[] = { {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_IVT)}, {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_HSX)}, {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_BDX)}, + {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_SKX)}, {0} }; MODULE_DEVICE_TABLE(pci, intel_ntb_pci_tbl); diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.h b/drivers/ntb/hw/intel/ntb_hw_intel.h index 3ec149cf6562..f2cf8a783f1e 100644 --- a/drivers/ntb/hw/intel/ntb_hw_intel.h +++ b/drivers/ntb/hw/intel/ntb_hw_intel.h @@ -70,6 +70,7 @@ #define PCI_DEVICE_ID_INTEL_NTB_B2B_BDX 0x6F0D #define PCI_DEVICE_ID_INTEL_NTB_PS_BDX 0x6F0E #define PCI_DEVICE_ID_INTEL_NTB_SS_BDX 0x6F0F +#define PCI_DEVICE_ID_INTEL_NTB_B2B_SKX 0x201C /* Intel Xeon hardware */ @@ -150,6 +151,51 @@ #define XEON_DB_TOTAL_SHIFT 16 #define XEON_SPAD_COUNT 16 +/* Intel Skylake Xeon hardware */ +#define SKX_IMBAR1SZ_OFFSET 0x00d0 +#define SKX_IMBAR2SZ_OFFSET 0x00d1 +#define SKX_EMBAR1SZ_OFFSET 0x00d2 +#define SKX_EMBAR2SZ_OFFSET 0x00d3 +#define SKX_DEVCTRL_OFFSET 0x0098 +#define SKX_DEVSTS_OFFSET 0x009a +#define SKX_UNCERRSTS_OFFSET 0x014c +#define SKX_CORERRSTS_OFFSET 0x0158 +#define SKX_LINK_STATUS_OFFSET 0x01a2 + +#define SKX_NTBCNTL_OFFSET 0x0000 +#define SKX_IMBAR1XBASE_OFFSET 0x0010 /* SBAR2XLAT */ +#define SKX_IMBAR1XLMT_OFFSET 0x0018 /* SBAR2LMT */ +#define SKX_IMBAR2XBASE_OFFSET 0x0020 /* SBAR4XLAT */ +#define SKX_IMBAR2XLMT_OFFSET 0x0028 /* SBAR4LMT */ +#define SKX_IM_INT_STATUS_OFFSET 0x0040 +#define SKX_IM_INT_DISABLE_OFFSET 0x0048 +#define SKX_IM_SPAD_OFFSET 0x0080 /* SPAD */ +#define SKX_USMEMMISS_OFFSET 0x0070 +#define SKX_INTVEC_OFFSET 0x00d0 +#define SKX_IM_DOORBELL_OFFSET 0x0100 /* SDOORBELL0 */ +#define SKX_B2B_SPAD_OFFSET 0x0180 /* B2B SPAD */ +#define SKX_EMBAR0XBASE_OFFSET 0x4008 /* B2B_XLAT */ +#define SKX_EMBAR1XBASE_OFFSET 0x4010 /* PBAR2XLAT */ +#define SKX_EMBAR1XLMT_OFFSET 0x4018 /* PBAR2LMT */ +#define SKX_EMBAR2XBASE_OFFSET 0x4020 /* PBAR4XLAT */ +#define SKX_EMBAR2XLMT_OFFSET 0x4028 /* PBAR4LMT */ +#define SKX_EM_INT_STATUS_OFFSET 0x4040 +#define SKX_EM_INT_DISABLE_OFFSET 0x4048 +#define SKX_EM_SPAD_OFFSET 0x4080 /* remote SPAD */ +#define SKX_EM_DOORBELL_OFFSET 0x4100 /* PDOORBELL0 */ +#define SKX_SPCICMD_OFFSET 0x4504 /* SPCICMD */ +#define SKX_EMBAR0_OFFSET 0x4510 /* SBAR0BASE */ +#define SKX_EMBAR1_OFFSET 0x4518 /* SBAR23BASE */ +#define SKX_EMBAR2_OFFSET 0x4520 /* SBAR45BASE */ + +#define SKX_DB_COUNT 32 +#define SKX_DB_LINK 32 +#define SKX_DB_LINK_BIT BIT_ULL(SKX_DB_LINK) +#define SKX_DB_MSIX_VECTOR_COUNT 33 +#define SKX_DB_MSIX_VECTOR_SHIFT 1 +#define SKX_DB_TOTAL_SHIFT 33 +#define SKX_SPAD_COUNT 16 + /* Intel Atom hardware */ #define ATOM_SBAR2XLAT_OFFSET 0x0008 @@ -240,6 +286,7 @@ #define NTB_HWERR_SDOORBELL_LOCKUP BIT_ULL(0) #define NTB_HWERR_SB01BASE_LOCKUP BIT_ULL(1) #define NTB_HWERR_B2BDOORBELL_BIT14 BIT_ULL(2) +#define NTB_HWERR_MSIX_VECTOR32_BAD BIT_ULL(3) /* flags to indicate unsafe api */ #define NTB_UNSAFE_DB BIT_ULL(0) @@ -263,6 +310,7 @@ struct intel_ntb_reg { struct intel_ntb_alt_reg { unsigned long db_bell; unsigned long db_mask; + unsigned long db_clear; unsigned long spad; }; diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c index 4eb8adb34508..02ca45fdd892 100644 --- a/drivers/ntb/ntb_transport.c +++ b/drivers/ntb/ntb_transport.c @@ -66,6 +66,7 @@ #define NTB_TRANSPORT_VER "4" #define NTB_TRANSPORT_NAME "ntb_transport" #define NTB_TRANSPORT_DESC "Software Queue-Pair Transport over NTB" +#define NTB_TRANSPORT_MIN_SPADS (MW0_SZ_HIGH + 2) MODULE_DESCRIPTION(NTB_TRANSPORT_DESC); MODULE_VERSION(NTB_TRANSPORT_VER); @@ -242,9 +243,6 @@ enum { NUM_MWS, MW0_SZ_HIGH, MW0_SZ_LOW, - MW1_SZ_HIGH, - MW1_SZ_LOW, - MAX_SPAD, }; #define dev_client_dev(__dev) \ @@ -811,7 +809,7 @@ static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt) { struct ntb_transport_qp *qp; u64 qp_bitmap_alloc; - int i; + unsigned int i, count; qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; @@ -831,7 +829,8 @@ static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt) * goes down, blast them now to give them a sane value the next * time they are accessed */ - for (i = 0; i < MAX_SPAD; i++) + count = ntb_spad_count(nt->ndev); + for (i = 0; i < count; i++) ntb_spad_write(nt->ndev, i, 0); } @@ -960,7 +959,6 @@ static void ntb_qp_link_work(struct work_struct *work) ntb_peer_spad_write(nt->ndev, QP_LINKS, val | BIT(qp->qp_num)); /* query remote spad for qp ready bits */ - ntb_peer_spad_read(nt->ndev, QP_LINKS); dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val); /* See if the remote side is up */ @@ -1064,17 +1062,12 @@ static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) { struct ntb_transport_ctx *nt; struct ntb_transport_mw *mw; - unsigned int mw_count, qp_count; + unsigned int mw_count, qp_count, spad_count, max_mw_count_for_spads; u64 qp_bitmap; int node; int rc, i; mw_count = ntb_mw_count(ndev); - if (ntb_spad_count(ndev) < (NUM_MWS + 1 + mw_count * 2)) { - dev_err(&ndev->dev, "Not enough scratch pad registers for %s", - NTB_TRANSPORT_NAME); - return -EIO; - } if (ntb_db_is_unsafe(ndev)) dev_dbg(&ndev->dev, @@ -1090,8 +1083,18 @@ static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) return -ENOMEM; nt->ndev = ndev; + spad_count = ntb_spad_count(ndev); + + /* Limit the MW's based on the availability of scratchpads */ + + if (spad_count < NTB_TRANSPORT_MIN_SPADS) { + nt->mw_count = 0; + rc = -EINVAL; + goto err; + } - nt->mw_count = mw_count; + max_mw_count_for_spads = (spad_count - MW0_SZ_HIGH) / 2; + nt->mw_count = min(mw_count, max_mw_count_for_spads); nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec), GFP_KERNEL, node); @@ -1799,7 +1802,7 @@ ntb_transport_create_queue(void *data, struct device *client_dev, node = dev_to_node(&ndev->dev); - free_queue = ffs(nt->qp_bitmap); + free_queue = ffs(nt->qp_bitmap_free); if (!free_queue) goto err; @@ -2270,9 +2273,8 @@ module_init(ntb_transport_init); static void __exit ntb_transport_exit(void) { - debugfs_remove_recursive(nt_debugfs_dir); - ntb_unregister_client(&ntb_transport_client); bus_unregister(&ntb_transport_bus); + debugfs_remove_recursive(nt_debugfs_dir); } module_exit(ntb_transport_exit); diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c index e75d4fdc0866..434e1d474f33 100644 --- a/drivers/ntb/test/ntb_perf.c +++ b/drivers/ntb/test/ntb_perf.c @@ -265,6 +265,8 @@ static ssize_t perf_copy(struct pthr_ctx *pctx, char __iomem *dst, if (dma_submit_error(cookie)) goto err_set_unmap; + dmaengine_unmap_put(unmap); + atomic_inc(&pctx->dma_sync); dma_async_issue_pending(chan); |