net/mlx5: fix shared RSS action release

As shared RSS action will be shared by multiple flows, the action
is created as global standalone action and managed only by the
relevant shared action management functions.

Currently, hrxqs will be created by shared RSS action or general
queue action. For hrxqs created by shared RSS action, they should
also only be released with shared RSS action. It's not correct to
release the shared RSS action hrxqs as general queue actions do
in flow destroy.

This commit adds a new fate action type for shared RSS action to
handle the shared RSS action hrxq release correctly.

Fixes: e1592b6 ("net/mlx5: make Rx queue thread safe")

Signed-off-by: Suanming Mou <suanmingm@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>

drivers/net/mlx5/mlx5.h

@@ -789,11 +789,11 @@ struct mlx5_flow_rss_desc {
 	uint8_t key[MLX5_RSS_HASH_KEY_LEN]; /**< RSS hash key. */
 	uint32_t key_len; /**< RSS hash key len. */
 	uint32_t tunnel; /**< Queue in tunnel. */
+	uint32_t shared_rss; /**< Shared RSS index. */
 	union {
 		uint16_t *queue; /**< Destination queues. */
 		const uint16_t *const_q; /**< Const pointer convert. */
 	};
-	bool standalone; /**< Queue is standalone or not. */
 };
 
 #define MLX5_PROC_PRIV(port_id) \
@@ -836,7 +836,6 @@ struct mlx5_ind_table_obj {
 __extension__
 struct mlx5_hrxq {
 	struct mlx5_cache_entry entry; /* Cache entry. */
-	uint32_t refcnt; /* Reference counter. */
 	uint32_t standalone:1; /* This object used in shared action. */
 	struct mlx5_ind_table_obj *ind_table; /* Indirection table. */
 	RTE_STD_C11

drivers/net/mlx5/mlx5_flow.c

@@ -5597,7 +5597,7 @@ flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
 		buf->entries = 1;
 		buf->entry[0].pattern = (void *)(uintptr_t)items;
 	}
-	flow->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
+	rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
 						      shared_actions_n);
 	/*
 	 * Record the start index when there is a nested call. All sub-flows
@@ -5708,6 +5708,11 @@ flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
 	ret = rte_errno; /* Save rte_errno before cleanup. */
 	flow_mreg_del_copy_action(dev, flow);
 	flow_drv_destroy(dev, flow);
+	if (rss_desc->shared_rss)
+		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
+			mlx5_ipool_get
+			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
+			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
 	rte_errno = ret; /* Restore rte_errno. */
 	ret = rte_errno;

drivers/net/mlx5/mlx5_flow.h

@@ -388,6 +388,7 @@ enum mlx5_flow_fate_type {
 	MLX5_FLOW_FATE_PORT_ID,
 	MLX5_FLOW_FATE_DROP,
 	MLX5_FLOW_FATE_DEFAULT_MISS,
+	MLX5_FLOW_FATE_SHARED_RSS,
 	MLX5_FLOW_FATE_MAX,
 };
 
@@ -651,6 +652,8 @@ struct mlx5_flow_handle {
 		/**< Generic value indicates the fate action. */
 		uint32_t rix_default_fate;
 		/**< Indicates default miss fate action. */
+		uint32_t rix_srss;
+		/**< Indicates shared RSS fate action. */
 	};
 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
 	struct mlx5_flow_handle_dv dvh;
@@ -1027,7 +1030,6 @@ flow_items_to_tunnel(const struct rte_flow_item items[])
 /* Flow structure. */
 struct rte_flow {
 	ILIST_ENTRY(uint32_t)next; /**< Index to the next flow structure. */
-	uint32_t shared_rss; /** < Shared RSS action ID. */
 	uint32_t dev_handles;
 	/**< Device flow handles that are part of the flow. */
 	uint32_t drv_type:2; /**< Driver type. */

drivers/net/mlx5/mlx5_flow_dv.c

@@ -81,6 +81,8 @@ flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
 static int
 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
 					uint32_t port_id);
+static void
+flow_dv_shared_rss_action_release(struct rte_eth_dev *dev, uint32_t srss);
 
 /**
  * Initialize flow attributes structure according to flow items' types.
@@ -8558,7 +8560,7 @@ flow_dv_hrxq_prepare(struct rte_eth_dev *dev,
 	rss_desc->key_len = MLX5_RSS_HASH_KEY_LEN;
 	rss_desc->hash_fields = dev_flow->hash_fields;
 	rss_desc->tunnel = !!(dh->layers & MLX5_FLOW_LAYER_TUNNEL);
-	rss_desc->standalone = false;
+	rss_desc->shared_rss = 0;
 	*hrxq_idx = mlx5_hrxq_get(dev, rss_desc);
 	if (!*hrxq_idx)
 		return NULL;
@@ -9780,7 +9782,9 @@ flow_dv_translate(struct rte_eth_dev *dev,
 			 * when expanding items for RSS.
 			 */
 			action_flags |= MLX5_FLOW_ACTION_RSS;
-			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
+			dev_flow->handle->fate_action = rss_desc->shared_rss ?
+				MLX5_FLOW_FATE_SHARED_RSS :
+				MLX5_FLOW_FATE_QUEUE;
 			break;
 		case MLX5_RTE_FLOW_ACTION_TYPE_AGE:
 			flow->age = (uint32_t)(uintptr_t)(action->conf);
@@ -10574,40 +10578,34 @@ __flow_dv_action_rss_hrxq_lookup(struct rte_eth_dev *dev, uint32_t idx,
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
- * @param[in] flow
- *   Shred RSS action holding hash RX queue objects.
  * @param[in] dev_flow
  *   Pointer to the sub flow.
+ * @param[in] rss_desc
+ *   Pointer to the RSS descriptor.
  * @param[out] hrxq
  *   Pointer to retrieved hash RX queue object.
  *
  * @return
  *   Valid hash RX queue index, otherwise 0 and rte_errno is set.
  */
 static uint32_t
-__flow_dv_rss_get_hrxq(struct rte_eth_dev *dev, struct rte_flow *flow,
-			   struct mlx5_flow *dev_flow,
-			   struct mlx5_hrxq **hrxq)
+__flow_dv_rss_get_hrxq(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
+		       struct mlx5_flow_rss_desc *rss_desc,
+		       struct mlx5_hrxq **hrxq)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
 	uint32_t hrxq_idx;
 
-	if (flow->shared_rss) {
+	if (rss_desc->shared_rss) {
 		hrxq_idx = __flow_dv_action_rss_hrxq_lookup
-				(dev, flow->shared_rss, dev_flow->hash_fields,
+				(dev, rss_desc->shared_rss,
+				 dev_flow->hash_fields,
 				 !!(dev_flow->handle->layers &
 				    MLX5_FLOW_LAYER_TUNNEL));
-		if (hrxq_idx) {
+		if (hrxq_idx)
 			*hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
 					       hrxq_idx);
-			__atomic_fetch_add(&(*hrxq)->refcnt, 1,
-					   __ATOMIC_RELAXED);
-		}
 	} else {
-		struct mlx5_flow_rss_desc *rss_desc =
-				&wks->rss_desc[!!wks->flow_nested_idx];
-
 		*hrxq = flow_dv_hrxq_prepare(dev, dev_flow, rss_desc,
 					     &hrxq_idx);
 	}
@@ -10642,8 +10640,15 @@ flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
 	int err;
 	int idx;
 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
+	struct mlx5_flow_rss_desc *rss_desc =
+				&wks->rss_desc[!!wks->flow_nested_idx];
 
 	MLX5_ASSERT(wks);
+	if (rss_desc->shared_rss) {
+		dh = wks->flows[wks->flow_idx - 1].handle;
+		MLX5_ASSERT(dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS);
+		dh->rix_srss = rss_desc->shared_rss;
+	}
 	for (idx = wks->flow_idx - 1; idx >= wks->flow_nested_idx; idx--) {
 		dev_flow = &wks->flows[idx];
 		dv = &dev_flow->dv;
@@ -10658,19 +10663,21 @@ flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
 				dv->actions[n++] =
 						priv->drop_queue.hrxq->action;
 			}
-		} else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
-			   !dv_h->rix_sample && !dv_h->rix_dest_array) {
+		} else if ((dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
+			   !dv_h->rix_sample && !dv_h->rix_dest_array) ||
+			    (dh->fate_action == MLX5_FLOW_FATE_SHARED_RSS)) {
 			struct mlx5_hrxq *hrxq = NULL;
 			uint32_t hrxq_idx = __flow_dv_rss_get_hrxq
-						(dev, flow, dev_flow, &hrxq);
+					(dev, dev_flow, rss_desc, &hrxq);
 			if (!hrxq) {
 				rte_flow_error_set
 					(error, rte_errno,
 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
 					 "cannot get hash queue");
 				goto error;
 			}
-			dh->rix_hrxq = hrxq_idx;
+			if (dh->fate_action == MLX5_FLOW_FATE_QUEUE)
+				dh->rix_hrxq = hrxq_idx;
 			dv->actions[n++] = hrxq->action;
 		} else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
 			if (!priv->sh->default_miss_action) {
@@ -10716,6 +10723,8 @@ flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
 		if (dh->vf_vlan.tag && dh->vf_vlan.created)
 			mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
 	}
+	if (rss_desc->shared_rss)
+		wks->flows[wks->flow_idx - 1].handle->rix_srss = 0;
 	rte_errno = err; /* Restore rte_errno. */
 	return -rte_errno;
 }
@@ -10900,6 +10909,25 @@ flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
 				     &cache->entry);
 }
 
+/**
+ * Release shared RSS action resource.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param srss
+ *   Shared RSS action index.
+ */
+static void
+flow_dv_shared_rss_action_release(struct rte_eth_dev *dev, uint32_t srss)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_shared_action_rss *shared_rss;
+
+	shared_rss = mlx5_ipool_get
+			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], srss);
+	__atomic_sub_fetch(&shared_rss->refcnt, 1, __ATOMIC_RELAXED);
+}
+
 void
 flow_dv_push_vlan_remove_cb(struct mlx5_cache_list *list,
 			    struct mlx5_cache_entry *entry)
@@ -10964,6 +10992,9 @@ flow_dv_fate_resource_release(struct rte_eth_dev *dev,
 		flow_dv_port_id_action_resource_release(dev,
 				handle->rix_port_id_action);
 		break;
+	case MLX5_FLOW_FATE_SHARED_RSS:
+		flow_dv_shared_rss_action_release(dev, handle->rix_srss);
+		break;
 	default:
 		DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
 		break;
@@ -11130,13 +11161,6 @@ flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
 	if (!flow)
 		return;
 	flow_dv_remove(dev, flow);
-	if (flow->shared_rss) {
-		struct mlx5_shared_action_rss *shared_rss = mlx5_ipool_get
-				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
-							      flow->shared_rss);
-
-		__atomic_sub_fetch(&shared_rss->refcnt, 1, __ATOMIC_RELAXED);
-	}
 	if (flow->counter) {
 		flow_dv_counter_free(dev, flow->counter);
 		flow->counter = 0;
@@ -11241,6 +11265,8 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
+ * @param[in] action_idx
+ *   Shared RSS action ipool index.
  * @param[in, out] action
  *   Partially initialized shared RSS action.
  * @param[out] error
@@ -11252,6 +11278,7 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
  */
 static int
 __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
+			uint32_t action_idx,
 			struct mlx5_shared_action_rss *action,
 			struct rte_flow_error *error)
 {
@@ -11263,7 +11290,8 @@ __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
 	rss_desc.key_len = MLX5_RSS_HASH_KEY_LEN;
 	rss_desc.const_q = action->origin.queue;
 	rss_desc.queue_num = action->origin.queue_num;
-	rss_desc.standalone = true;
+	/* Set non-zero value to indicate a shared RSS. */
+	rss_desc.shared_rss = action_idx;
 	for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
 		uint32_t hrxq_idx;
 		uint64_t hash_fields = mlx5_rss_hash_fields[i];
@@ -11355,7 +11383,7 @@ __flow_dv_action_rss_create(struct rte_eth_dev *dev,
 	memcpy(shared_action->queue, rss->queue, queue_size);
 	origin->queue = shared_action->queue;
 	origin->queue_num = rss->queue_num;
-	if (__flow_dv_action_rss_setup(dev, shared_action, error))
+	if (__flow_dv_action_rss_setup(dev, idx, shared_action, error))
 		goto error_rss_init;
 	__atomic_add_fetch(&shared_action->refcnt, 1, __ATOMIC_RELAXED);
 	rte_spinlock_lock(&priv->shared_act_sl);

drivers/net/mlx5/mlx5_flow_verbs.c

@@ -1972,7 +1972,7 @@ flow_verbs_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
 			rss_desc->hash_fields = dev_flow->hash_fields;
 			rss_desc->tunnel = !!(handle->layers &
 					      MLX5_FLOW_LAYER_TUNNEL);
-			rss_desc->standalone = false;
+			rss_desc->shared_rss = 0;
 			hrxq_idx = mlx5_hrxq_get(dev, rss_desc);
 			hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
 					      hrxq_idx);

drivers/net/mlx5/mlx5_rxq.c

@@ -2211,25 +2211,27 @@ __mlx5_hrxq_create(struct rte_eth_dev *dev,
 	struct mlx5_priv *priv = dev->data->dev_private;
 	const uint8_t *rss_key = rss_desc->key;
 	uint32_t rss_key_len =  rss_desc->key_len;
+	bool standalone = !!rss_desc->shared_rss;
 	const uint16_t *queues =
-		rss_desc->standalone ? rss_desc->const_q : rss_desc->queue;
+		standalone ? rss_desc->const_q : rss_desc->queue;
 	uint32_t queues_n = rss_desc->queue_num;
 	struct mlx5_hrxq *hrxq = NULL;
 	uint32_t hrxq_idx = 0;
 	struct mlx5_ind_table_obj *ind_tbl;
 	int ret;
 
 	queues_n = rss_desc->hash_fields ? queues_n : 1;
-	ind_tbl = mlx5_ind_table_obj_get(dev, queues, queues_n);
+	ind_tbl = standalone ? NULL :
+		  mlx5_ind_table_obj_get(dev, queues, queues_n);
 	if (!ind_tbl)
 		ind_tbl = mlx5_ind_table_obj_new(dev, queues, queues_n,
-						 rss_desc->standalone);
+						 standalone);
 	if (!ind_tbl)
 		return NULL;
 	hrxq = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_HRXQ], &hrxq_idx);
 	if (!hrxq)
 		goto error;
-	hrxq->standalone = rss_desc->standalone;
+	hrxq->standalone = standalone;
 	hrxq->idx = hrxq_idx;
 	hrxq->ind_table = ind_tbl;
 	hrxq->rss_key_len = rss_key_len;
@@ -2240,7 +2242,7 @@ __mlx5_hrxq_create(struct rte_eth_dev *dev,
 		goto error;
 	return hrxq;
 error:
-	mlx5_ind_table_obj_release(dev, ind_tbl, rss_desc->standalone);
+	mlx5_ind_table_obj_release(dev, ind_tbl, standalone);
 	if (hrxq)
 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_HRXQ], hrxq_idx);
 	return NULL;
@@ -2294,7 +2296,7 @@ uint32_t mlx5_hrxq_get(struct rte_eth_dev *dev,
 		.data = rss_desc,
 	};
 
-	if (rss_desc->standalone) {
+	if (rss_desc->shared_rss) {
 		hrxq = __mlx5_hrxq_create(dev, rss_desc);
 	} else {
 		entry = mlx5_cache_register(&priv->hrxqs, &ctx);
@@ -2346,11 +2348,8 @@ mlx5_drop_action_create(struct rte_eth_dev *dev)
 	struct mlx5_hrxq *hrxq = NULL;
 	int ret;
 
-	if (priv->drop_queue.hrxq) {
-		__atomic_fetch_add(&priv->drop_queue.hrxq->refcnt, 1,
-				   __ATOMIC_RELAXED);
+	if (priv->drop_queue.hrxq)
 		return priv->drop_queue.hrxq;
-	}
 	hrxq = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*hrxq), 0, SOCKET_ID_ANY);
 	if (!hrxq) {
 		DRV_LOG(WARNING,
@@ -2369,7 +2368,6 @@ mlx5_drop_action_create(struct rte_eth_dev *dev)
 	ret = priv->obj_ops.drop_action_create(dev);
 	if (ret < 0)
 		goto error;
-	__atomic_store_n(&hrxq->refcnt, 1, __ATOMIC_RELAXED);
 	return hrxq;
 error:
 	if (hrxq) {
@@ -2393,14 +2391,14 @@ mlx5_drop_action_destroy(struct rte_eth_dev *dev)
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
 
-	if (__atomic_sub_fetch(&hrxq->refcnt, 1, __ATOMIC_RELAXED) == 0) {
-		priv->obj_ops.drop_action_destroy(dev);
-		mlx5_free(priv->drop_queue.rxq);
-		mlx5_free(hrxq->ind_table);
-		mlx5_free(hrxq);
-		priv->drop_queue.rxq = NULL;
-		priv->drop_queue.hrxq = NULL;
-	}
+	if (!priv->drop_queue.hrxq)
+		return;
+	priv->obj_ops.drop_action_destroy(dev);
+	mlx5_free(priv->drop_queue.rxq);
+	mlx5_free(hrxq->ind_table);
+	mlx5_free(hrxq);
+	priv->drop_queue.rxq = NULL;
+	priv->drop_queue.hrxq = NULL;
 }
 
 /**