Merge branch 'cxgb4-Reference-count-MPS-TCAM-entries-within-a-PF'

Raju Rangoju says:

====================
cxgb4: Reference count MPS TCAM entries within a PF

Firmware reference counts the MPS TCAM entries by PF and VF,
but it does not do it for usage within a PF or VF. This patch
adds the support to track MPS TCAM entries within a PF.

v2->v3:
 Fixed the compiler errors due to incorrect patch
 Also, removed the new blank line at EOF
v1->v2:
 Use refcount_t type instead of atomic_t for mps reference count
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

5 files changed, 312 insertions, 32 deletions

diff --git a/drivers/net/ethernet/chelsio/cxgb4/Makefile b/drivers/net/ethernet/chelsio/cxgb4/Makefile
index 91d8a885deba..20390f6afbb4 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/Makefile
+++ b/drivers/net/ethernet/chelsio/cxgb4/Makefile
@@ -7,7 +7,7 @@ obj-$(CONFIG_CHELSIO_T4) += cxgb4.o
 
 cxgb4-objs := cxgb4_main.o l2t.o smt.o t4_hw.o sge.o clip_tbl.o cxgb4_ethtool.o \
 	      cxgb4_uld.o srq.o sched.o cxgb4_filter.o cxgb4_tc_u32.o \
-	      cxgb4_ptp.o cxgb4_tc_flower.o cxgb4_cudbg.o \
+	      cxgb4_ptp.o cxgb4_tc_flower.o cxgb4_cudbg.o cxgb4_mps.o \
 	      cudbg_common.o cudbg_lib.o cudbg_zlib.o
 cxgb4-$(CONFIG_CHELSIO_T4_DCB) +=  cxgb4_dcb.o
 cxgb4-$(CONFIG_CHELSIO_T4_FCOE) +=  cxgb4_fcoe.o
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
index db2ec46ba6b6..1fbb640e896a 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
@@ -905,10 +905,6 @@ struct mbox_list {
 	struct list_head list;
 };
 
-struct mps_encap_entry {
-	atomic_t refcnt;
-};
-
 #if IS_ENABLED(CONFIG_THERMAL)
 struct ch_thermal {
 	struct thermal_zone_device *tzdev;
@@ -917,6 +913,14 @@ struct ch_thermal {
 };
 #endif
 
+struct mps_entries_ref {
+	struct list_head list;
+	u8 addr[ETH_ALEN];
+	u8 mask[ETH_ALEN];
+	u16 idx;
+	refcount_t refcnt;
+};
+
 struct adapter {
 	void __iomem *regs;
 	void __iomem *bar2;
@@ -969,7 +973,6 @@ struct adapter {
 	unsigned int rawf_start;
 	unsigned int rawf_cnt;
 	struct smt_data *smt;
-	struct mps_encap_entry *mps_encap;
 	struct cxgb4_uld_info *uld;
 	void *uld_handle[CXGB4_ULD_MAX];
 	unsigned int num_uld;
@@ -977,6 +980,8 @@ struct adapter {
 	struct list_head list_node;
 	struct list_head rcu_node;
 	struct list_head mac_hlist; /* list of MAC addresses in MPS Hash */
+	struct list_head mps_ref;
+	spinlock_t mps_ref_lock; /* lock for syncing mps ref/def activities */
 
 	void *iscsi_ppm;
 
@@ -1906,4 +1911,42 @@ int cxgb4_set_msix_aff(struct adapter *adap, unsigned short vec,
 		       cpumask_var_t *aff_mask, int idx);
 void cxgb4_clear_msix_aff(unsigned short vec, cpumask_var_t aff_mask);
 
+int cxgb4_change_mac(struct port_info *pi, unsigned int viid,
+		     int *tcam_idx, const u8 *addr,
+		     bool persistent, u8 *smt_idx);
+
+int cxgb4_alloc_mac_filt(struct adapter *adap, unsigned int viid,
+			 bool free, unsigned int naddr,
+			 const u8 **addr, u16 *idx,
+			 u64 *hash, bool sleep_ok);
+int cxgb4_free_mac_filt(struct adapter *adap, unsigned int viid,
+			unsigned int naddr, const u8 **addr, bool sleep_ok);
+int cxgb4_init_mps_ref_entries(struct adapter *adap);
+void cxgb4_free_mps_ref_entries(struct adapter *adap);
+int cxgb4_alloc_encap_mac_filt(struct adapter *adap, unsigned int viid,
+			       const u8 *addr, const u8 *mask,
+			       unsigned int vni, unsigned int vni_mask,
+			       u8 dip_hit, u8 lookup_type, bool sleep_ok);
+int cxgb4_free_encap_mac_filt(struct adapter *adap, unsigned int viid,
+			      int idx, bool sleep_ok);
+int cxgb4_free_raw_mac_filt(struct adapter *adap,
+			    unsigned int viid,
+			    const u8 *addr,
+			    const u8 *mask,
+			    unsigned int idx,
+			    u8 lookup_type,
+			    u8 port_id,
+			    bool sleep_ok);
+int cxgb4_alloc_raw_mac_filt(struct adapter *adap,
+			     unsigned int viid,
+			     const u8 *addr,
+			     const u8 *mask,
+			     unsigned int idx,
+			     u8 lookup_type,
+			     u8 port_id,
+			     bool sleep_ok);
+int cxgb4_update_mac_filt(struct port_info *pi, unsigned int viid,
+			  int *tcam_idx, const u8 *addr,
+			  bool persistent, u8 *smt_idx);
+
 #endif /* __CXGB4_H__ */
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
index 6232236d7abc..43b0f8c57da7 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_filter.c
@@ -727,10 +727,8 @@ void clear_filter(struct adapter *adap, struct filter_entry *f)
 		cxgb4_smt_release(f->smt);
 
 	if (f->fs.val.encap_vld && f->fs.val.ovlan_vld)
-		if (atomic_dec_and_test(&adap->mps_encap[f->fs.val.ovlan &
-							 0x1ff].refcnt))
-			t4_free_encap_mac_filt(adap, pi->viid,
-					       f->fs.val.ovlan & 0x1ff, 0);
+		t4_free_encap_mac_filt(adap, pi->viid,
+				       f->fs.val.ovlan & 0x1ff, 0);
 
 	if ((f->fs.hash || is_t6(adap->params.chip)) && f->fs.type)
 		cxgb4_clip_release(f->dev, (const u32 *)&f->fs.val.lip, 1);
@@ -1177,7 +1175,6 @@ static int cxgb4_set_hash_filter(struct net_device *dev,
 			if (ret < 0)
 				goto free_atid;
 
-			atomic_inc(&adapter->mps_encap[ret].refcnt);
 			f->fs.val.ovlan = ret;
 			f->fs.mask.ovlan = 0xffff;
 			f->fs.val.ovlan_vld = 1;
@@ -1420,7 +1417,6 @@ int __cxgb4_set_filter(struct net_device *dev, int filter_id,
 			if (ret < 0)
 				goto free_clip;
 
-			atomic_inc(&adapter->mps_encap[ret].refcnt);
 			f->fs.val.ovlan = ret;
 			f->fs.mask.ovlan = 0x1ff;
 			f->fs.val.ovlan_vld = 1;
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
index 54908002c786..b08efc48d42f 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
@@ -366,13 +366,19 @@ static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
 	int ret;
 	u64 mhash = 0;
 	u64 uhash = 0;
+	/* idx stores the index of allocated filters,
+	 * its size should be modified based on the number of
+	 * MAC addresses that we allocate filters for
+	 */
+
+	u16 idx[1] = {};
 	bool free = false;
 	bool ucast = is_unicast_ether_addr(mac_addr);
 	const u8 *maclist[1] = {mac_addr};
 	struct hash_mac_addr *new_entry;
 
-	ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
-				NULL, ucast ? &uhash : &mhash, false);
+	ret = cxgb4_alloc_mac_filt(adap, pi->viid, free, 1, maclist,
+				   idx, ucast ? &uhash : &mhash, false);
 	if (ret < 0)
 		goto out;
 	/* if hash != 0, then add the addr to hash addr list
@@ -410,7 +416,7 @@ static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
 		}
 	}
 
-	ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
+	ret = cxgb4_free_mac_filt(adap, pi->viid, 1, maclist, false);
 	return ret < 0 ? -EINVAL : 0;
 }
 
@@ -449,9 +455,9 @@ static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
  *	Addresses are programmed to hash region, if tcam runs out of entries.
  *
  */
-static int cxgb4_change_mac(struct port_info *pi, unsigned int viid,
-			    int *tcam_idx, const u8 *addr, bool persist,
-			    u8 *smt_idx)
+int cxgb4_change_mac(struct port_info *pi, unsigned int viid,
+		     int *tcam_idx, const u8 *addr, bool persist,
+		     u8 *smt_idx)
 {
 	struct adapter *adapter = pi->adapter;
 	struct hash_mac_addr *entry, *new_entry;
@@ -505,8 +511,8 @@ static int link_start(struct net_device *dev)
 	ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
 			    !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
 	if (ret == 0)
-		ret = cxgb4_change_mac(pi, pi->viid, &pi->xact_addr_filt,
-				       dev->dev_addr, true, &pi->smt_idx);
+		ret = cxgb4_update_mac_filt(pi, pi->viid, &pi->xact_addr_filt,
+					    dev->dev_addr, true, &pi->smt_idx);
 	if (ret == 0)
 		ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
 				    &pi->link_cfg);
@@ -3020,8 +3026,8 @@ static int cxgb_set_mac_addr(struct net_device *dev, void *p)
 	if (!is_valid_ether_addr(addr->sa_data))
 		return -EADDRNOTAVAIL;
 
-	ret = cxgb4_change_mac(pi, pi->viid, &pi->xact_addr_filt,
-			       addr->sa_data, true, &pi->smt_idx);
+	ret = cxgb4_update_mac_filt(pi, pi->viid, &pi->xact_addr_filt,
+				    addr->sa_data, true, &pi->smt_idx);
 	if (ret < 0)
 		return ret;
 
@@ -3273,8 +3279,6 @@ static void cxgb_del_udp_tunnel(struct net_device *netdev,
 				    i);
 			return;
 		}
-		atomic_dec(&adapter->mps_encap[adapter->rawf_start +
-			   pi->port_id].refcnt);
 	}
 }
 
@@ -3363,7 +3367,6 @@ static void cxgb_add_udp_tunnel(struct net_device *netdev,
 			cxgb_del_udp_tunnel(netdev, ti);
 			return;
 		}
-		atomic_inc(&adapter->mps_encap[ret].refcnt);
 	}
 }
 
@@ -5446,7 +5449,6 @@ static void free_some_resources(struct adapter *adapter)
 {
 	unsigned int i;
 
-	kvfree(adapter->mps_encap);
 	kvfree(adapter->smt);
 	kvfree(adapter->l2t);
 	kvfree(adapter->srq);
@@ -5972,12 +5974,6 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 		adapter->params.offload = 0;
 	}
 
-	adapter->mps_encap = kvcalloc(adapter->params.arch.mps_tcam_size,
-				      sizeof(struct mps_encap_entry),
-				      GFP_KERNEL);
-	if (!adapter->mps_encap)
-		dev_warn(&pdev->dev, "could not allocate MPS Encap entries, continuing\n");
-
 #if IS_ENABLED(CONFIG_IPV6)
 	if (chip_ver <= CHELSIO_T5 &&
 	    (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
@@ -6053,6 +6049,8 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	/* check for PCI Express bandwidth capabiltites */
 	pcie_print_link_status(pdev);
 
+	cxgb4_init_mps_ref_entries(adapter);
+
 	err = init_rss(adapter);
 	if (err)
 		goto out_free_dev;
@@ -6179,6 +6177,8 @@ static void remove_one(struct pci_dev *pdev)
 
 		disable_interrupts(adapter);
 
+		cxgb4_free_mps_ref_entries(adapter);
+
 		for_each_port(adapter, i)
 			if (adapter->port[i]->reg_state == NETREG_REGISTERED)
 				unregister_netdev(adapter->port[i]);
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_mps.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_mps.c
new file mode 100644
index 000000000000..b1a073eea60b
--- /dev/null
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_mps.c
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2019 Chelsio Communications, Inc. All rights reserved. */
+
+#include "cxgb4.h"
+
+static int cxgb4_mps_ref_dec_by_mac(struct adapter *adap,
+				    const u8 *addr, const u8 *mask)
+{
+	u8 bitmask[] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+	struct mps_entries_ref *mps_entry, *tmp;
+	int ret = -EINVAL;
+
+	spin_lock_bh(&adap->mps_ref_lock);
+	list_for_each_entry_safe(mps_entry, tmp, &adap->mps_ref, list) {
+		if (ether_addr_equal(mps_entry->addr, addr) &&
+		    ether_addr_equal(mps_entry->mask, mask ? mask : bitmask)) {
+			if (!refcount_dec_and_test(&mps_entry->refcnt)) {
+				spin_unlock_bh(&adap->mps_ref_lock);
+				return -EBUSY;
+			}
+			list_del(&mps_entry->list);
+			kfree(mps_entry);
+			ret = 0;
+			break;
+		}
+	}
+	spin_unlock_bh(&adap->mps_ref_lock);
+	return ret;
+}
+
+static int cxgb4_mps_ref_dec(struct adapter *adap, u16 idx)
+{
+	struct mps_entries_ref *mps_entry, *tmp;
+	int ret = -EINVAL;
+
+	spin_lock(&adap->mps_ref_lock);
+	list_for_each_entry_safe(mps_entry, tmp, &adap->mps_ref, list) {
+		if (mps_entry->idx == idx) {
+			if (!refcount_dec_and_test(&mps_entry->refcnt)) {
+				spin_unlock(&adap->mps_ref_lock);
+				return -EBUSY;
+			}
+			list_del(&mps_entry->list);
+			kfree(mps_entry);
+			ret = 0;
+			break;
+		}
+	}
+	spin_unlock(&adap->mps_ref_lock);
+	return ret;
+}
+
+static int cxgb4_mps_ref_inc(struct adapter *adap, const u8 *mac_addr,
+			     u16 idx, const u8 *mask)
+{
+	u8 bitmask[] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+	struct mps_entries_ref *mps_entry;
+	int ret = 0;
+
+	spin_lock_bh(&adap->mps_ref_lock);
+	list_for_each_entry(mps_entry, &adap->mps_ref, list) {
+		if (mps_entry->idx == idx) {
+			refcount_inc(&mps_entry->refcnt);
+			goto unlock;
+		}
+	}
+	mps_entry = kzalloc(sizeof(*mps_entry), GFP_ATOMIC);
+	if (!mps_entry) {
+		ret = -ENOMEM;
+		goto unlock;
+	}
+	ether_addr_copy(mps_entry->mask, mask ? mask : bitmask);
+	ether_addr_copy(mps_entry->addr, mac_addr);
+	mps_entry->idx = idx;
+	refcount_set(&mps_entry->refcnt, 1);
+	list_add_tail(&mps_entry->list, &adap->mps_ref);
+unlock:
+	spin_unlock_bh(&adap->mps_ref_lock);
+	return ret;
+}
+
+int cxgb4_free_mac_filt(struct adapter *adap, unsigned int viid,
+			unsigned int naddr, const u8 **addr, bool sleep_ok)
+{
+	int ret, i;
+
+	for (i = 0; i < naddr; i++) {
+		if (!cxgb4_mps_ref_dec_by_mac(adap, addr[i], NULL)) {
+			ret = t4_free_mac_filt(adap, adap->mbox, viid,
+					       1, &addr[i], sleep_ok);
+			if (ret < 0)
+				return ret;
+		}
+	}
+
+	/* return number of filters freed */
+	return naddr;
+}
+
+int cxgb4_alloc_mac_filt(struct adapter *adap, unsigned int viid,
+			 bool free, unsigned int naddr, const u8 **addr,
+			 u16 *idx, u64 *hash, bool sleep_ok)
+{
+	int ret, i;
+
+	ret = t4_alloc_mac_filt(adap, adap->mbox, viid, free,
+				naddr, addr, idx, hash, sleep_ok);
+	if (ret < 0)
+		return ret;
+
+	for (i = 0; i < naddr; i++) {
+		if (idx[i] != 0xffff) {
+			if (cxgb4_mps_ref_inc(adap, addr[i], idx[i], NULL)) {
+				ret = -ENOMEM;
+				goto error;
+			}
+		}
+	}
+
+	goto out;
+error:
+	cxgb4_free_mac_filt(adap, viid, naddr, addr, sleep_ok);
+
+out:
+	/* Returns a negative error number or the number of filters allocated */
+	return ret;
+}
+
+int cxgb4_update_mac_filt(struct port_info *pi, unsigned int viid,
+			  int *tcam_idx, const u8 *addr,
+			  bool persistent, u8 *smt_idx)
+{
+	int ret;
+
+	ret = cxgb4_change_mac(pi, viid, tcam_idx,
+			       addr, persistent, smt_idx);
+	if (ret < 0)
+		return ret;
+
+	cxgb4_mps_ref_inc(pi->adapter, addr, *tcam_idx, NULL);
+	return ret;
+}
+
+int cxgb4_free_raw_mac_filt(struct adapter *adap,
+			    unsigned int viid,
+			    const u8 *addr,
+			    const u8 *mask,
+			    unsigned int idx,
+			    u8 lookup_type,
+			    u8 port_id,
+			    bool sleep_ok)
+{
+	int ret = 0;
+
+	if (!cxgb4_mps_ref_dec(adap, idx))
+		ret = t4_free_raw_mac_filt(adap, viid, addr,
+					   mask, idx, lookup_type,
+					   port_id, sleep_ok);
+
+	return ret;
+}
+
+int cxgb4_alloc_raw_mac_filt(struct adapter *adap,
+			     unsigned int viid,
+			     const u8 *addr,
+			     const u8 *mask,
+			     unsigned int idx,
+			     u8 lookup_type,
+			     u8 port_id,
+			     bool sleep_ok)
+{
+	int ret;
+
+	ret = t4_alloc_raw_mac_filt(adap, viid, addr,
+				    mask, idx, lookup_type,
+				    port_id, sleep_ok);
+	if (ret < 0)
+		return ret;
+
+	if (cxgb4_mps_ref_inc(adap, addr, ret, mask)) {
+		ret = -ENOMEM;
+		t4_free_raw_mac_filt(adap, viid, addr,
+				     mask, idx, lookup_type,
+				     port_id, sleep_ok);
+	}
+
+	return ret;
+}
+
+int cxgb4_free_encap_mac_filt(struct adapter *adap, unsigned int viid,
+			      int idx, bool sleep_ok)
+{
+	int ret = 0;
+
+	if (!cxgb4_mps_ref_dec(adap, idx))
+		ret = t4_free_encap_mac_filt(adap, viid, idx, sleep_ok);
+
+	return ret;
+}
+
+int cxgb4_alloc_encap_mac_filt(struct adapter *adap, unsigned int viid,
+			       const u8 *addr, const u8 *mask,
+			       unsigned int vni, unsigned int vni_mask,
+			       u8 dip_hit, u8 lookup_type, bool sleep_ok)
+{
+	int ret;
+
+	ret = t4_alloc_encap_mac_filt(adap, viid, addr, mask, vni, vni_mask,
+				      dip_hit, lookup_type, sleep_ok);
+	if (ret < 0)
+		return ret;
+
+	if (cxgb4_mps_ref_inc(adap, addr, ret, mask)) {
+		ret = -ENOMEM;
+		t4_free_encap_mac_filt(adap, viid, ret, sleep_ok);
+	}
+	return ret;
+}
+
+int cxgb4_init_mps_ref_entries(struct adapter *adap)
+{
+	spin_lock_init(&adap->mps_ref_lock);
+	INIT_LIST_HEAD(&adap->mps_ref);
+
+	return 0;
+}
+
+void cxgb4_free_mps_ref_entries(struct adapter *adap)
+{
+	struct mps_entries_ref *mps_entry, *tmp;
+
+	if (!list_empty(&adap->mps_ref))
+		return;
+
+	spin_lock(&adap->mps_ref_lock);
+	list_for_each_entry_safe(mps_entry, tmp, &adap->mps_ref, list) {
+		list_del(&mps_entry->list);
+		kfree(mps_entry);
+	}
+	spin_unlock(&adap->mps_ref_lock);
+}