net/mlx5e: Allow concurrent creation of hairpin entries

Hairpin entries creation is fully synchronized by hairpin_tbl_lock. In
order to allow concurrent initialization of mlx5e_hairpin structure
instances and provisioning of hairpin entries to hardware, extend
mlx5e_hairpin_entry with 'res_ready' completion. Move call to
mlx5e_hairpin_create() out of hairpin_tbl_lock critical section. Modify
code that attaches new flows to existing hpe to wait for 'res_ready'
completion before using the hpe. Insert hpe to hairpin table before
provisioning it to hardware and modify all users of hairpin table to verify
that hpe was fully initialized by checking hpe->hp pointer (and to wait for
'res_ready' completion, if necessary).

Modify dead peer update event handling function to save hpe's to temporary
list with their reference counter incremented. Wait for completion of hpe's
in temporary list and update their 'peer_gone' flag outside of
hairpin_tbl_lock critical section.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

drivers/net/ethernet/mellanox/mlx5/core/en_tc.c

@@ -39,6 +39,7 @@
 #include <linux/mlx5/device.h>
 #include <linux/rhashtable.h>
 #include <linux/refcount.h>
+#include <linux/completion.h>
 #include <net/tc_act/tc_mirred.h>
 #include <net/tc_act/tc_vlan.h>
 #include <net/tc_act/tc_tunnel_key.h>
@@ -166,11 +167,16 @@ struct mlx5e_hairpin_entry {
 	spinlock_t flows_lock;
 	/* flows sharing the same hairpin */
 	struct list_head flows;
+	/* hpe's that were not fully initialized when dead peer update event
+	 * function traversed them.
+	 */
+	struct list_head dead_peer_wait_list;
 
 	u16 peer_vhca_id;
 	u8 prio;
 	struct mlx5e_hairpin *hp;
 	refcount_t refcnt;
+	struct completion res_ready;
 };
 
 struct mod_hdr_key {
@@ -657,11 +663,14 @@ static void mlx5e_hairpin_put(struct mlx5e_priv *priv,
 	hash_del(&hpe->hairpin_hlist);
 	mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 
+	if (!IS_ERR_OR_NULL(hpe->hp)) {
 		netdev_dbg(priv->netdev, "del hairpin: peer %s\n",
 			   dev_name(hpe->hp->pair->peer_mdev->device));
 
-	WARN_ON(!list_empty(&hpe->flows));
 		mlx5e_hairpin_destroy(hpe->hp);
+	}
+
+	WARN_ON(!list_empty(&hpe->flows));
 	kfree(hpe);
 }
 
@@ -733,20 +742,34 @@ static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 
 	mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
 	hpe = mlx5e_hairpin_get(priv, peer_id, match_prio);
-	if (hpe)
+	if (hpe) {
+		mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
+		wait_for_completion(&hpe->res_ready);
+
+		if (IS_ERR(hpe->hp)) {
+			err = -EREMOTEIO;
+			goto out_err;
+		}
 		goto attach_flow;
+	}
 
 	hpe = kzalloc(sizeof(*hpe), GFP_KERNEL);
 	if (!hpe) {
-		err = -ENOMEM;
-		goto create_hairpin_err;
+		mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
+		return -ENOMEM;
 	}
 
 	spin_lock_init(&hpe->flows_lock);
 	INIT_LIST_HEAD(&hpe->flows);
+	INIT_LIST_HEAD(&hpe->dead_peer_wait_list);
 	hpe->peer_vhca_id = peer_id;
 	hpe->prio = match_prio;
 	refcount_set(&hpe->refcnt, 1);
+	init_completion(&hpe->res_ready);
+
+	hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
+		 hash_hairpin_info(peer_id, match_prio));
+	mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 
 	params.log_data_size = 15;
 	params.log_data_size = min_t(u8, params.log_data_size,
@@ -768,9 +791,11 @@ static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 	params.num_channels = link_speed64;
 
 	hp = mlx5e_hairpin_create(priv, &params, peer_ifindex);
+	hpe->hp = hp;
+	complete_all(&hpe->res_ready);
 	if (IS_ERR(hp)) {
 		err = PTR_ERR(hp);
-		goto create_hairpin_err;
+		goto out_err;
 	}
 
 	netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d (log) data %d packets %d\n",
@@ -778,10 +803,6 @@ static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 		   dev_name(hp->pair->peer_mdev->device),
 		   hp->pair->sqn[0], match_prio, params.log_data_size, params.log_num_packets);
 
-	hpe->hp = hp;
-	hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,
-		 hash_hairpin_info(peer_id, match_prio));
-
 attach_flow:
 	if (hpe->hp->num_channels > 1) {
 		flow_flag_set(flow, HAIRPIN_RSS);
@@ -789,7 +810,6 @@ static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 	} else {
 		flow->nic_attr->hairpin_tirn = hpe->hp->tirn;
 	}
-	mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 
 	flow->hpe = hpe;
 	spin_lock(&hpe->flows_lock);
@@ -798,9 +818,8 @@ static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 
 	return 0;
 
-create_hairpin_err:
-	mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
-	kfree(hpe);
+out_err:
+	mlx5e_hairpin_put(priv, hpe);
 	return err;
 }
 
@@ -3767,7 +3786,8 @@ static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
 					      struct mlx5e_priv *peer_priv)
 {
 	struct mlx5_core_dev *peer_mdev = peer_priv->mdev;
-	struct mlx5e_hairpin_entry *hpe;
+	struct mlx5e_hairpin_entry *hpe, *tmp;
+	LIST_HEAD(init_wait_list);
 	u16 peer_vhca_id;
 	int bkt;
 
@@ -3777,11 +3797,18 @@ static void mlx5e_tc_hairpin_update_dead_peer(struct mlx5e_priv *priv,
 	peer_vhca_id = MLX5_CAP_GEN(peer_mdev, vhca_id);
 
 	mutex_lock(&priv->fs.tc.hairpin_tbl_lock);
-	hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist) {
-		if (hpe->peer_vhca_id == peer_vhca_id)
+	hash_for_each(priv->fs.tc.hairpin_tbl, bkt, hpe, hairpin_hlist)
+		if (refcount_inc_not_zero(&hpe->refcnt))
+			list_add(&hpe->dead_peer_wait_list, &init_wait_list);
+	mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
+
+	list_for_each_entry_safe(hpe, tmp, &init_wait_list, dead_peer_wait_list) {
+		wait_for_completion(&hpe->res_ready);
+		if (!IS_ERR_OR_NULL(hpe->hp) && hpe->peer_vhca_id == peer_vhca_id)
 			hpe->hp->pair->peer_gone = true;
+
+		mlx5e_hairpin_put(priv, hpe);
 	}
-	mutex_unlock(&priv->fs.tc.hairpin_tbl_lock);
 }
 
 static int mlx5e_tc_netdev_event(struct notifier_block *this,