patches for DPDK stable branches
 help / color / mirror / Atom feed
* [dpdk-stable] [PATCH 3/5] net/mlx5: discover max flow priority using DevX
       [not found] <20211005005216.2427489-1-dkozlyuk@nvidia.com>
@ 2021-10-05  0:52 ` dkozlyuk
  2021-10-05  0:52 ` [dpdk-stable] [PATCH 4/5] net/mlx5: create drop queue " dkozlyuk
                   ` (2 subsequent siblings)
  3 siblings, 0 replies; 18+ messages in thread
From: dkozlyuk @ 2021-10-05  0:52 UTC (permalink / raw)
  To: dev; +Cc: Dmitry Kozlyuk, stable, Matan Azrad, Viacheslav Ovsiienko

From: Dmitry Kozlyuk <dkozlyuk@nvidia.com>

Maximum available flow priority was discovered using Verbs API
regardless of the selected flow engine. This required some Verbs
objects to be initialized in order to use DevX engine. Make priority
discovery an engine method and implement it for DevX using its API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c   |   1 -
 drivers/net/mlx5/mlx5_flow.c       |  98 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow.h       |   4 ++
 drivers/net/mlx5/mlx5_flow_dv.c    | 103 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  77 +++------------------
 5 files changed, 215 insertions(+), 68 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 3746057673..8ee7ada51b 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1830,7 +1830,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	priv->drop_queue.hrxq = mlx5_drop_action_create(eth_dev);
 	if (!priv->drop_queue.hrxq)
 		goto error;
-	/* Supported Verbs flow priority number detection. */
 	err = mlx5_flow_discover_priorities(eth_dev);
 	if (err < 0) {
 		err = -err;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index c914a7120c..bfc3e20c9a 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -9416,3 +9416,101 @@ mlx5_dbg__print_pattern(const struct rte_flow_item *item)
 	}
 	printf("END\n");
 }
+
+/* Map of Verbs to Flow priority with 8 Verbs priorities. */
+static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
+};
+
+/* Map of Verbs to Flow priority with 16 Verbs priorities. */
+static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
+	{ 9, 10, 11 }, { 12, 13, 14 },
+};
+
+/**
+ * Discover the number of available flow priorities.
+ *
+ * @param dev
+ *   Ethernet device.
+ *
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+int
+mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+{
+	static const uint16_t vprio[] = {8, 16};
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	const struct mlx5_flow_driver_ops *fops;
+	enum mlx5_flow_drv_type type;
+	int ret;
+
+	type = mlx5_flow_os_get_type();
+	if (type == MLX5_FLOW_TYPE_MAX) {
+		type = MLX5_FLOW_TYPE_VERBS;
+		if (priv->config.devx && priv->config.dv_flow_en)
+			type = MLX5_FLOW_TYPE_DV;
+	}
+	fops = flow_get_drv_ops(type);
+	if (fops->discover_priorities == NULL) {
+		DRV_LOG(ERR, "Priority discovery not supported");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
+	if (ret < 0)
+		return ret;
+	switch (ret) {
+	case 8:
+		ret = RTE_DIM(priority_map_3);
+		break;
+	case 16:
+		ret = RTE_DIM(priority_map_5);
+		break;
+	default:
+		rte_errno = ENOTSUP;
+		DRV_LOG(ERR,
+			"port %u maximum priority: %d expected 8/16",
+			dev->data->port_id, ret);
+		return -rte_errno;
+	}
+	DRV_LOG(INFO, "port %u supported flow priorities:"
+		" 0-%d for ingress or egress root table,"
+		" 0-%d for non-root table or transfer root table.",
+		dev->data->port_id, ret - 2,
+		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
+	return ret;
+}
+
+/**
+ * Adjust flow priority based on the highest layer and the request priority.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] priority
+ *   The rule base priority.
+ * @param[in] subpriority
+ *   The priority based on the items.
+ *
+ * @return
+ *   The new priority.
+ */
+uint32_t
+mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
+			  uint32_t subpriority)
+{
+	uint32_t res = 0;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	switch (priv->config.flow_prio) {
+	case RTE_DIM(priority_map_3):
+		res = priority_map_3[priority][subpriority];
+		break;
+	case RTE_DIM(priority_map_5):
+		res = priority_map_5[priority][subpriority];
+		break;
+	}
+	return  res;
+}
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 5c68d4f7d7..8f94125f26 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1226,6 +1226,9 @@ typedef int (*mlx5_flow_create_def_policy_t)
 			(struct rte_eth_dev *dev);
 typedef void (*mlx5_flow_destroy_def_policy_t)
 			(struct rte_eth_dev *dev);
+typedef int (*mlx5_flow_discover_priorities_t)
+			(struct rte_eth_dev *dev,
+			 const uint16_t *vprio, int vprio_n);
 
 struct mlx5_flow_driver_ops {
 	mlx5_flow_validate_t validate;
@@ -1260,6 +1263,7 @@ struct mlx5_flow_driver_ops {
 	mlx5_flow_action_update_t action_update;
 	mlx5_flow_action_query_t action_query;
 	mlx5_flow_sync_domain_t sync_domain;
+	mlx5_flow_discover_priorities_t discover_priorities;
 };
 
 /* mlx5_flow.c */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index c6370cd1d6..155745748f 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -17978,6 +17978,108 @@ flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
 	return 0;
 }
 
+/**
+ * Discover the number of available flow priorities
+ * by trying to create a flow with the highest priority value
+ * for each possible number.
+ *
+ * @param[in] dev
+ *   Ethernet device.
+ * @param[in] vprio
+ *   List of possible number of available priorities.
+ * @param[in] vprio_n
+ *   Size of @p vprio array.
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+static int
+flow_dv_discover_priorities(struct rte_eth_dev *dev,
+			    const uint16_t *vprio, int vprio_n)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+	struct mlx5_flow_dv_matcher matcher = {
+		.mask = {
+			.size = sizeof(matcher.mask.buf),
+		},
+	};
+	union mlx5_flow_tbl_key tbl_key;
+	struct mlx5_flow flow;
+	void *action;
+	struct rte_flow_error error;
+	uint8_t misc_mask;
+	int i, err, ret = -ENOTSUP;
+
+	/*
+	 * Prepare a flow with a catch-all pattern and a drop action.
+	 * Use drop queue, because shared drop action may be unavailable.
+	 */
+	action = priv->drop_queue.hrxq->action;
+	if (action == NULL) {
+		DRV_LOG(ERR, "Priority discovery requires a drop action");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	memset(&flow, 0, sizeof(flow));
+	flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
+	if (flow.handle == NULL) {
+		DRV_LOG(ERR, "Cannot create flow handle");
+		rte_errno = ENOMEM;
+		return -rte_errno;
+	}
+	flow.ingress = true;
+	flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
+	flow.dv.actions[0] = action;
+	flow.dv.actions_n = 1;
+	memset(&eth, 0, sizeof(eth));
+	flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
+				   &item, /* inner */ false, /* group */ 0);
+	matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
+	for (i = 0; i < vprio_n; i++) {
+		/* Configure the next proposed maximum priority. */
+		matcher.priority = vprio[i] - 1;
+		memset(&tbl_key, 0, sizeof(tbl_key));
+		err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
+					       /* tunnel */ NULL,
+					       /* group */ 0,
+					       &error);
+		if (err != 0) {
+			/* This action is pure SW and must always succeed. */
+			DRV_LOG(ERR, "Cannot register matcher");
+			ret = -rte_errno;
+			break;
+		}
+		/* Try to apply the flow to HW. */
+		misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
+		__flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
+		err = mlx5_flow_os_create_flow
+				(flow.handle->dvh.matcher->matcher_object,
+				 (void *)&flow.dv.value, flow.dv.actions_n,
+				 flow.dv.actions, &flow.handle->drv_flow);
+		if (err == 0) {
+			claim_zero(mlx5_flow_os_destroy_flow
+						(flow.handle->drv_flow));
+			flow.handle->drv_flow = NULL;
+		}
+		claim_zero(flow_dv_matcher_release(dev, flow.handle));
+		if (err != 0)
+			break;
+		ret = vprio[i];
+	}
+	mlx5_ipool_free(pool, flow.handle_idx);
+	/* Set rte_errno if no expected priority value matched. */
+	if (ret < 0)
+		rte_errno = -ret;
+	return ret;
+}
+
 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.validate = flow_dv_validate,
 	.prepare = flow_dv_prepare,
@@ -18011,6 +18113,7 @@ const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.action_update = flow_dv_action_update,
 	.action_query = flow_dv_action_query,
 	.sync_domain = flow_dv_sync_domain,
+	.discover_priorities = flow_dv_discover_priorities,
 };
 
 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
diff --git a/drivers/net/mlx5/mlx5_flow_verbs.c b/drivers/net/mlx5/mlx5_flow_verbs.c
index b93fd4d2c9..72b9db6c7f 100644
--- a/drivers/net/mlx5/mlx5_flow_verbs.c
+++ b/drivers/net/mlx5/mlx5_flow_verbs.c
@@ -28,17 +28,6 @@
 #define VERBS_SPEC_INNER(item_flags) \
 	(!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
 
-/* Map of Verbs to Flow priority with 8 Verbs priorities. */
-static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
-};
-
-/* Map of Verbs to Flow priority with 16 Verbs priorities. */
-static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
-	{ 9, 10, 11 }, { 12, 13, 14 },
-};
-
 /* Verbs specification header. */
 struct ibv_spec_header {
 	enum ibv_flow_spec_type type;
@@ -50,13 +39,17 @@ struct ibv_spec_header {
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
- *
+ * @param[in] vprio
+ *   Expected result variants.
+ * @param[in] vprio_n
+ *   Number of entries in @p vprio array.
  * @return
- *   number of supported flow priority on success, a negative errno
+ *   Number of supported flow priority on success, a negative errno
  *   value otherwise and rte_errno is set.
  */
-int
-mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+static int
+flow_verbs_discover_priorities(struct rte_eth_dev *dev,
+			       const uint16_t *vprio, int vprio_n)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct {
@@ -79,7 +72,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	};
 	struct ibv_flow *flow;
 	struct mlx5_hrxq *drop = priv->drop_queue.hrxq;
-	uint16_t vprio[] = { 8, 16 };
 	int i;
 	int priority = 0;
 
@@ -87,7 +79,7 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		rte_errno = ENOTSUP;
 		return -rte_errno;
 	}
-	for (i = 0; i != RTE_DIM(vprio); i++) {
+	for (i = 0; i != vprio_n; i++) {
 		flow_attr.attr.priority = vprio[i] - 1;
 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 		if (!flow)
@@ -95,59 +87,9 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		claim_zero(mlx5_glue->destroy_flow(flow));
 		priority = vprio[i];
 	}
-	switch (priority) {
-	case 8:
-		priority = RTE_DIM(priority_map_3);
-		break;
-	case 16:
-		priority = RTE_DIM(priority_map_5);
-		break;
-	default:
-		rte_errno = ENOTSUP;
-		DRV_LOG(ERR,
-			"port %u verbs maximum priority: %d expected 8/16",
-			dev->data->port_id, priority);
-		return -rte_errno;
-	}
-	DRV_LOG(INFO, "port %u supported flow priorities:"
-		" 0-%d for ingress or egress root table,"
-		" 0-%d for non-root table or transfer root table.",
-		dev->data->port_id, priority - 2,
-		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
 	return priority;
 }
 
-/**
- * Adjust flow priority based on the highest layer and the request priority.
- *
- * @param[in] dev
- *   Pointer to the Ethernet device structure.
- * @param[in] priority
- *   The rule base priority.
- * @param[in] subpriority
- *   The priority based on the items.
- *
- * @return
- *   The new priority.
- */
-uint32_t
-mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
-				   uint32_t subpriority)
-{
-	uint32_t res = 0;
-	struct mlx5_priv *priv = dev->data->dev_private;
-
-	switch (priv->config.flow_prio) {
-	case RTE_DIM(priority_map_3):
-		res = priority_map_3[priority][subpriority];
-		break;
-	case RTE_DIM(priority_map_5):
-		res = priority_map_5[priority][subpriority];
-		break;
-	}
-	return  res;
-}
-
 /**
  * Get Verbs flow counter by index.
  *
@@ -2105,4 +2047,5 @@ const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
 	.destroy = flow_verbs_destroy,
 	.query = flow_verbs_query,
 	.sync_domain = flow_verbs_sync_domain,
+	.discover_priorities = flow_verbs_discover_priorities,
 };
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH 4/5] net/mlx5: create drop queue using DevX
       [not found] <20211005005216.2427489-1-dkozlyuk@nvidia.com>
  2021-10-05  0:52 ` [dpdk-stable] [PATCH 3/5] net/mlx5: discover max flow priority using DevX dkozlyuk
@ 2021-10-05  0:52 ` dkozlyuk
  2021-10-05  0:52 ` [dpdk-stable] [PATCH 5/5] net/mlx5: preserve indirect actions on restart dkozlyuk
       [not found] ` <20211015161822.3099818-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: dkozlyuk @ 2021-10-05  0:52 UTC (permalink / raw)
  To: dev; +Cc: Dmitry Kozlyuk, stable, Matan Azrad, Viacheslav Ovsiienko

From: Dmitry Kozlyuk <dkozlyuk@nvidia.com>

Drop queue creation and destruction were not implemented for DevX
flow engine and Verbs engine methods were used as a workaround.
Implement these methods for DevX so that there is a valid queue ID
that can be used regardless of queue configuration via API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c |   4 -
 drivers/net/mlx5/mlx5_devx.c     | 204 ++++++++++++++++++++++++++-----
 2 files changed, 176 insertions(+), 32 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 8ee7ada51b..985f0bd489 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1790,10 +1790,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	}
 	if (config->devx && config->dv_flow_en && config->dest_tir) {
 		priv->obj_ops = devx_obj_ops;
-		priv->obj_ops.drop_action_create =
-						ibv_obj_ops.drop_action_create;
-		priv->obj_ops.drop_action_destroy =
-						ibv_obj_ops.drop_action_destroy;
 #ifndef HAVE_MLX5DV_DEVX_UAR_OFFSET
 		priv->obj_ops.txq_obj_modify = ibv_obj_ops.txq_obj_modify;
 #else
diff --git a/drivers/net/mlx5/mlx5_devx.c b/drivers/net/mlx5/mlx5_devx.c
index a1db53577a..447d6bafb9 100644
--- a/drivers/net/mlx5/mlx5_devx.c
+++ b/drivers/net/mlx5/mlx5_devx.c
@@ -226,17 +226,17 @@ mlx5_rx_devx_get_event(struct mlx5_rxq_obj *rxq_obj)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	struct mlx5_devx_create_rq_attr rq_attr = { 0 };
@@ -289,20 +289,20 @@ mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_devx_cq *cq_obj = 0;
 	struct mlx5_devx_cq_attr cq_attr = { 0 };
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_dev_ctx_shared *sh = priv->sh;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	unsigned int cqe_n = mlx5_rxq_cqe_num(rxq_data);
@@ -497,13 +497,13 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
 		tmpl->fd = mlx5_os_get_devx_channel_fd(tmpl->devx_channel);
 	}
 	/* Create CQ using DevX API. */
-	ret = mlx5_rxq_create_devx_cq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Failed to create CQ.");
 		goto error;
 	}
 	/* Create RQ using DevX API. */
-	ret = mlx5_rxq_create_devx_rq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Port %u Rx queue %u RQ creation failure.",
 			dev->data->port_id, idx);
@@ -536,6 +536,11 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
  *   Pointer to Ethernet device.
  * @param log_n
  *   Log of number of queues in the array.
+ * @param queues
+ *   List of RX queue indices or NULL, in which case
+ *   the attribute will be filled by drop queue ID.
+ * @param queues_n
+ *   Size of @p queues array or 0 if it is NULL.
  * @param ind_tbl
  *   DevX indirection table object.
  *
@@ -563,6 +568,11 @@ mlx5_devx_ind_table_create_rqt_attr(struct rte_eth_dev *dev,
 	}
 	rqt_attr->rqt_max_size = priv->config.ind_table_max_size;
 	rqt_attr->rqt_actual_size = rqt_n;
+	if (queues == NULL) {
+		for (i = 0; i < rqt_n; i++)
+			rqt_attr->rq_list[i] = priv->drop_queue.rxq->rq->id;
+		return rqt_attr;
+	}
 	for (i = 0; i != queues_n; ++i) {
 		struct mlx5_rxq_data *rxq = (*priv->rxqs)[queues[i]];
 		struct mlx5_rxq_ctrl *rxq_ctrl =
@@ -670,7 +680,8 @@ mlx5_devx_ind_table_destroy(struct mlx5_ind_table_obj *ind_tbl)
  * @param[in] hash_fields
  *   Verbs protocol hash field to make the RSS on.
  * @param[in] ind_tbl
- *   Indirection table for TIR.
+ *   Indirection table for TIR. If table queues array is NULL,
+ *   a TIR for drop queue is assumed.
  * @param[in] tunnel
  *   Tunnel type.
  * @param[out] tir_attr
@@ -686,19 +697,27 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 		       int tunnel, struct mlx5_devx_tir_attr *tir_attr)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[ind_tbl->queues[0]];
-	struct mlx5_rxq_ctrl *rxq_ctrl =
-		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
-	enum mlx5_rxq_type rxq_obj_type = rxq_ctrl->type;
+	enum mlx5_rxq_type rxq_obj_type;
 	bool lro = true;
 	uint32_t i;
 
-	/* Enable TIR LRO only if all the queues were configured for. */
-	for (i = 0; i < ind_tbl->queues_n; ++i) {
-		if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
-			lro = false;
-			break;
+	/* NULL queues designate drop queue. */
+	if (ind_tbl->queues != NULL) {
+		struct mlx5_rxq_data *rxq_data =
+					(*priv->rxqs)[ind_tbl->queues[0]];
+		struct mlx5_rxq_ctrl *rxq_ctrl =
+			container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+		rxq_obj_type = rxq_ctrl->type;
+
+		/* Enable TIR LRO only if all the queues were configured for. */
+		for (i = 0; i < ind_tbl->queues_n; ++i) {
+			if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
+				lro = false;
+				break;
+			}
 		}
+	} else {
+		rxq_obj_type = priv->drop_queue.rxq->rxq_ctrl->type;
 	}
 	memset(tir_attr, 0, sizeof(*tir_attr));
 	tir_attr->disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
@@ -857,7 +876,7 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
 }
 
 /**
- * Create a DevX drop action for Rx Hash queue.
+ * Create a DevX drop Rx queue.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -866,14 +885,99 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+mlx5_rxq_devx_obj_drop_create(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	int socket_id = dev->device->numa_node;
+	struct mlx5_rxq_ctrl *rxq_ctrl;
+	struct mlx5_rxq_data *rxq_data;
+	struct mlx5_rxq_obj *rxq = NULL;
+	int ret;
+
+	/*
+	 * Initialize dummy control structures.
+	 * They are required to hold pointers for cleanup
+	 * and are only accessible via drop queue DevX objects.
+	 */
+	rxq_ctrl = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq_ctrl),
+			       0, socket_id);
+	if (rxq_ctrl == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue control",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq), 0, socket_id);
+	if (rxq == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue object",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq->rxq_ctrl = rxq_ctrl;
+	rxq_ctrl->type = MLX5_RXQ_TYPE_STANDARD;
+	rxq_ctrl->priv = priv;
+	rxq_ctrl->obj = rxq;
+	rxq_data = &rxq_ctrl->rxq;
+	/* Create CQ using DevX API. */
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue CQ creation failed.",
+			dev->data->port_id);
+		goto error;
+	}
+	/* Create RQ using DevX API. */
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue RQ creation failed.",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	/* Change queue state to ready. */
+	ret = mlx5_devx_modify_rq(rxq, MLX5_RXQ_MOD_RST2RDY);
+	if (ret != 0)
+		goto error;
+	/* Initialize drop queue. */
+	priv->drop_queue.rxq = rxq;
+	return 0;
+error:
+	ret = rte_errno; /* Save rte_errno before cleanup. */
+	if (rxq != NULL) {
+		if (rxq->rq_obj.rq != NULL)
+			mlx5_devx_rq_destroy(&rxq->rq_obj);
+		if (rxq->cq_obj.cq != NULL)
+			mlx5_devx_cq_destroy(&rxq->cq_obj);
+		if (rxq->devx_channel)
+			mlx5_os_devx_destroy_event_channel
+							(rxq->devx_channel);
+		mlx5_free(rxq);
+	}
+	if (rxq_ctrl != NULL)
+		mlx5_free(rxq_ctrl);
+	rte_errno = ret; /* Restore rte_errno. */
 	return -rte_errno;
 }
 
+/**
+ * Release drop Rx queue resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ */
+static void
+mlx5_rxq_devx_obj_drop_release(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_obj *rxq = priv->drop_queue.rxq;
+	struct mlx5_rxq_ctrl *rxq_ctrl = rxq->rxq_ctrl;
+
+	mlx5_rxq_devx_obj_release(rxq);
+	mlx5_free(rxq);
+	mlx5_free(rxq_ctrl);
+	priv->drop_queue.rxq = NULL;
+}
+
 /**
  * Release a drop hash Rx queue.
  *
@@ -883,9 +987,53 @@ mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
 static void
 mlx5_devx_drop_action_destroy(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+
+	if (hrxq->tir != NULL)
+		mlx5_devx_tir_destroy(hrxq);
+	if (hrxq->ind_table->ind_table != NULL)
+		mlx5_devx_ind_table_destroy(hrxq->ind_table);
+	if (priv->drop_queue.rxq->rq != NULL)
+		mlx5_rxq_devx_obj_drop_release(dev);
+}
+
+/**
+ * Create a DevX drop action for Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+	int ret;
+
+	ret = mlx5_rxq_devx_obj_drop_create(dev);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop RX queue");
+		return ret;
+	}
+	/* hrxq->ind_table queues are NULL, drop RX queue ID will be used */
+	ret = mlx5_devx_ind_table_new(dev, 0, hrxq->ind_table);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue indirection table");
+		goto error;
+	}
+	ret = mlx5_devx_hrxq_new(dev, hrxq, /* tunnel */ false);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue");
+		goto error;
+	}
+	return 0;
+error:
+	mlx5_devx_drop_action_destroy(dev);
+	return ret;
 }
 
 /**
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH 5/5] net/mlx5: preserve indirect actions on restart
       [not found] <20211005005216.2427489-1-dkozlyuk@nvidia.com>
  2021-10-05  0:52 ` [dpdk-stable] [PATCH 3/5] net/mlx5: discover max flow priority using DevX dkozlyuk
  2021-10-05  0:52 ` [dpdk-stable] [PATCH 4/5] net/mlx5: create drop queue " dkozlyuk
@ 2021-10-05  0:52 ` dkozlyuk
       [not found] ` <20211015161822.3099818-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: dkozlyuk @ 2021-10-05  0:52 UTC (permalink / raw)
  To: dev; +Cc: Dmitry Kozlyuk, bingz, stable, Matan Azrad, Viacheslav Ovsiienko

From: Dmitry Kozlyuk <dkozlyuk@nvidia.com>

MLX5 PMD uses reference counting to manage RX queue resources.
After port stop shared RSS actions kept references to RX queues,
preventing resource release. As a result, internal PMD mempool
for such queues had been exhausted after a number of port restarts.
Diagnostic message from rte_eth_dev_start():

    Rx queue allocation failed: Cannot allocate memory

Dereference RX queues used by indirect actions on port stop (detach)
and restore references on port start (attach) in order to allow RX queue
resource release, but keep indirect RSS across the port restart.
Replace queue IDs in HW by drop queue ID on detach and restore actual
queue IDs on attach.

As a result, MLX5 PMD is able to keep all its indirect actions
across port restart. Advertise this capability to the ethdev layer.

Fixes: 4b61b8774be9 ("ethdev: introduce indirect flow action")
Cc: bingz@nvidia.com
Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/mlx5_ethdev.c  |   1 +
 drivers/net/mlx5/mlx5_flow.c    | 194 ++++++++++++++++++++++++++++----
 drivers/net/mlx5/mlx5_flow.h    |   2 +
 drivers/net/mlx5/mlx5_rx.h      |   4 +
 drivers/net/mlx5/mlx5_rxq.c     |  99 ++++++++++++++--
 drivers/net/mlx5/mlx5_trigger.c |  10 ++
 6 files changed, 276 insertions(+), 34 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 82e2284d98..8ebfd0bccb 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -325,6 +325,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 	info->reta_size = priv->reta_idx_n ?
 		priv->reta_idx_n : config->ind_table_max_size;
 	info->hash_key_size = MLX5_RSS_HASH_KEY_LEN;
+	info->dev_capa = RTE_ETH_DEV_CAPA_FLOW_SHARED_OBJECT_KEEP;
 	info->speed_capa = priv->link_speed_capa;
 	info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK;
 	mlx5_set_default_params(dev, info);
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index bfc3e20c9a..c10b911259 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1560,6 +1560,58 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 	return 0;
 }
 
+/**
+ * Validate queue numbers for device RSS.
+ *
+ * @param[in] dev
+ *   Configured device.
+ * @param[in] queues
+ *   Array of queue numbers.
+ * @param[in] queues_n
+ *   Size of the @p queues array.
+ * @param[out] error
+ *   On error, filled with a textual error description.
+ * @param[out] queue
+ *   On error, filled with an offending queue index in @p queues array.
+ *
+ * @return
+ *   0 on success, a negative errno code on error.
+ */
+static int
+mlx5_validate_rss_queues(const struct rte_eth_dev *dev,
+			 const uint16_t *queues, uint32_t queues_n,
+			 const char **error, uint32_t *queue_idx)
+{
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
+	uint32_t i;
+
+	for (i = 0; i != queues_n; ++i) {
+		struct mlx5_rxq_ctrl *rxq_ctrl;
+
+		if (queues[i] >= priv->rxqs_n) {
+			*error = "queue index out of range";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		if (!(*priv->rxqs)[queues[i]]) {
+			*error =  "queue is not configured";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		rxq_ctrl = container_of((*priv->rxqs)[queues[i]],
+					struct mlx5_rxq_ctrl, rxq);
+		if (i == 0)
+			rxq_type = rxq_ctrl->type;
+		if (rxq_type != rxq_ctrl->type) {
+			*error = "combining hairpin and regular RSS queues is not supported";
+			*queue_idx = i;
+			return -ENOTSUP;
+		}
+	}
+	return 0;
+}
+
 /*
  * Validate the rss action.
  *
@@ -1580,8 +1632,9 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_rss *rss = action->conf;
-	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
-	unsigned int i;
+	int ret;
+	const char *message;
+	uint32_t queue_idx;
 
 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
@@ -1645,27 +1698,12 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  NULL, "No queues configured");
-	for (i = 0; i != rss->queue_num; ++i) {
-		struct mlx5_rxq_ctrl *rxq_ctrl;
-
-		if (rss->queue[i] >= priv->rxqs_n)
-			return rte_flow_error_set
-				(error, EINVAL,
-				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue index out of range");
-		if (!(*priv->rxqs)[rss->queue[i]])
-			return rte_flow_error_set
-				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue is not configured");
-		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
-					struct mlx5_rxq_ctrl, rxq);
-		if (i == 0)
-			rxq_type = rxq_ctrl->type;
-		if (rxq_type != rxq_ctrl->type)
-			return rte_flow_error_set
-				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i],
-				 "combining hairpin and regular RSS queues is not supported");
+	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
+				       &message, &queue_idx);
+	if (ret != 0) {
+		return rte_flow_error_set(error, -ret,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &rss->queue[queue_idx], message);
 	}
 	return 0;
 }
@@ -8547,6 +8585,116 @@ mlx5_action_handle_flush(struct rte_eth_dev *dev)
 	return ret;
 }
 
+/**
+ * Validate existing indirect actions against current device configuration
+ * and attach them to device resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_attach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+		const char *message;
+		uint32_t queue_idx;
+
+		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
+					       ind_tbl->queues_n,
+					       &message, &queue_idx);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
+				dev->data->port_id, ind_tbl->queues[queue_idx],
+				message);
+			break;
+		}
+	}
+	if (ret != 0)
+		return ret;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not attach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not detach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
+/**
+ * Detach indirect actions of the device from its resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_detach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not detach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not attach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
 #ifndef HAVE_MLX5DV_DR
 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
 #else
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 8f94125f26..6bc7946cc3 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1574,6 +1574,8 @@ void mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
 		struct mlx5_flow_meter_policy *mtr_policy);
 int mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev);
 int mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev);
+int mlx5_action_handle_attach(struct rte_eth_dev *dev);
+int mlx5_action_handle_detach(struct rte_eth_dev *dev);
 int mlx5_action_handle_flush(struct rte_eth_dev *dev);
 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id);
 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh);
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 3f2b99fb65..7319ad0264 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -222,6 +222,10 @@ int mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 			      struct mlx5_ind_table_obj *ind_tbl,
 			      uint16_t *queues, const uint32_t queues_n,
 			      bool standalone);
+int mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
+int mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
 struct mlx5_list_entry *mlx5_hrxq_create_cb(void *tool_ctx, void *cb_ctx);
 int mlx5_hrxq_match_cb(void *tool_ctx, struct mlx5_list_entry *entry,
 		       void *cb_ctx);
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index abd8ce7989..cf4a29772c 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -2018,6 +2018,26 @@ mlx5_ind_table_obj_new(struct rte_eth_dev *dev, const uint16_t *queues,
 	return ind_tbl;
 }
 
+static int
+mlx5_ind_table_obj_check_standalone(struct rte_eth_dev *dev __rte_unused,
+				    struct mlx5_ind_table_obj *ind_tbl)
+{
+	uint32_t refcnt;
+
+	refcnt = __atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED);
+	if (refcnt <= 1)
+		return 0;
+	/*
+	 * Modification of indirection tables having more than 1
+	 * reference is unsupported.
+	 */
+	DRV_LOG(DEBUG,
+		"Port %u cannot modify indirection table %p (refcnt %u > 1).",
+		dev->data->port_id, (void *)ind_tbl, refcnt);
+	rte_errno = EINVAL;
+	return -rte_errno;
+}
+
 /**
  * Modify an indirection table.
  *
@@ -2050,18 +2070,8 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 
 	MLX5_ASSERT(standalone);
 	RTE_SET_USED(standalone);
-	if (__atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED) > 1) {
-		/*
-		 * Modification of indirection ntables having more than 1
-		 * reference unsupported. Intended for standalone indirection
-		 * tables only.
-		 */
-		DRV_LOG(DEBUG,
-			"Port %u cannot modify indirection table (refcnt> 1).",
-			dev->data->port_id);
-		rte_errno = EINVAL;
+	if (mlx5_ind_table_obj_check_standalone(dev, ind_tbl) < 0)
 		return -rte_errno;
-	}
 	for (i = 0; i != queues_n; ++i) {
 		if (!mlx5_rxq_get(dev, queues[i])) {
 			ret = -rte_errno;
@@ -2087,6 +2097,73 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 	return ret;
 }
 
+/**
+ * Attach an indirection table to its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to attach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_modify(dev, ind_tbl, ind_tbl->queues,
+					ind_tbl->queues_n, true);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_get(dev, ind_tbl->queues[i]);
+	return 0;
+}
+
+/**
+ * Detach an indirection table from its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to detach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	const unsigned int n = rte_is_power_of_2(ind_tbl->queues_n) ?
+			       log2above(ind_tbl->queues_n) :
+			       log2above(priv->config.ind_table_max_size);
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_check_standalone(dev, ind_tbl);
+	if (ret != 0)
+		return ret;
+	MLX5_ASSERT(priv->obj_ops.ind_table_modify);
+	ret = priv->obj_ops.ind_table_modify(dev, n, NULL, 0, ind_tbl);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_release(dev, ind_tbl->queues[i]);
+	return ret;
+}
+
 int
 mlx5_hrxq_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
 		   void *cb_ctx)
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 54173bfacb..c3adf5082e 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -14,6 +14,7 @@
 #include <mlx5_malloc.h>
 
 #include "mlx5.h"
+#include "mlx5_flow.h"
 #include "mlx5_mr.h"
 #include "mlx5_rx.h"
 #include "mlx5_tx.h"
@@ -1113,6 +1114,14 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 	mlx5_rxq_timestamp_set(dev);
 	/* Set a mask and offset of scheduling on timestamp into Tx queues. */
 	mlx5_txq_dynf_timestamp_set(dev);
+	/* Attach indirection table objects detached on port stop. */
+	ret = mlx5_action_handle_attach(dev);
+	if (ret) {
+		DRV_LOG(ERR,
+			"port %u failed to attach indirect actions: %s",
+			dev->data->port_id, rte_strerror(rte_errno));
+		goto error;
+	}
 	/*
 	 * In non-cached mode, it only needs to start the default mreg copy
 	 * action and no flow created by application exists anymore.
@@ -1185,6 +1194,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	/* All RX queue flags will be cleared in the flush interface. */
 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, true);
 	mlx5_flow_meter_rxq_flush(dev);
+	mlx5_action_handle_detach(dev);
 	mlx5_rx_intr_vec_disable(dev);
 	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
 	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v2 3/5] net/mlx5: discover max flow priority using DevX
       [not found] ` <20211015161822.3099818-1-dkozlyuk@nvidia.com>
@ 2021-10-15 16:18   ` Dmitry Kozlyuk
  2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 4/5] net/mlx5: create drop queue " Dmitry Kozlyuk
                     ` (2 subsequent siblings)
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-15 16:18 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Maximum available flow priority was discovered using Verbs API
regardless of the selected flow engine. This required some Verbs
objects to be initialized in order to use DevX engine. Make priority
discovery an engine method and implement it for DevX using its API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c   |   1 -
 drivers/net/mlx5/mlx5_flow.c       |  98 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow.h       |   4 ++
 drivers/net/mlx5/mlx5_flow_dv.c    | 103 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  77 +++------------------
 5 files changed, 215 insertions(+), 68 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 3746057673..8ee7ada51b 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1830,7 +1830,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	priv->drop_queue.hrxq = mlx5_drop_action_create(eth_dev);
 	if (!priv->drop_queue.hrxq)
 		goto error;
-	/* Supported Verbs flow priority number detection. */
 	err = mlx5_flow_discover_priorities(eth_dev);
 	if (err < 0) {
 		err = -err;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index c914a7120c..bfc3e20c9a 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -9416,3 +9416,101 @@ mlx5_dbg__print_pattern(const struct rte_flow_item *item)
 	}
 	printf("END\n");
 }
+
+/* Map of Verbs to Flow priority with 8 Verbs priorities. */
+static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
+};
+
+/* Map of Verbs to Flow priority with 16 Verbs priorities. */
+static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
+	{ 9, 10, 11 }, { 12, 13, 14 },
+};
+
+/**
+ * Discover the number of available flow priorities.
+ *
+ * @param dev
+ *   Ethernet device.
+ *
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+int
+mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+{
+	static const uint16_t vprio[] = {8, 16};
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	const struct mlx5_flow_driver_ops *fops;
+	enum mlx5_flow_drv_type type;
+	int ret;
+
+	type = mlx5_flow_os_get_type();
+	if (type == MLX5_FLOW_TYPE_MAX) {
+		type = MLX5_FLOW_TYPE_VERBS;
+		if (priv->config.devx && priv->config.dv_flow_en)
+			type = MLX5_FLOW_TYPE_DV;
+	}
+	fops = flow_get_drv_ops(type);
+	if (fops->discover_priorities == NULL) {
+		DRV_LOG(ERR, "Priority discovery not supported");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
+	if (ret < 0)
+		return ret;
+	switch (ret) {
+	case 8:
+		ret = RTE_DIM(priority_map_3);
+		break;
+	case 16:
+		ret = RTE_DIM(priority_map_5);
+		break;
+	default:
+		rte_errno = ENOTSUP;
+		DRV_LOG(ERR,
+			"port %u maximum priority: %d expected 8/16",
+			dev->data->port_id, ret);
+		return -rte_errno;
+	}
+	DRV_LOG(INFO, "port %u supported flow priorities:"
+		" 0-%d for ingress or egress root table,"
+		" 0-%d for non-root table or transfer root table.",
+		dev->data->port_id, ret - 2,
+		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
+	return ret;
+}
+
+/**
+ * Adjust flow priority based on the highest layer and the request priority.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] priority
+ *   The rule base priority.
+ * @param[in] subpriority
+ *   The priority based on the items.
+ *
+ * @return
+ *   The new priority.
+ */
+uint32_t
+mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
+			  uint32_t subpriority)
+{
+	uint32_t res = 0;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	switch (priv->config.flow_prio) {
+	case RTE_DIM(priority_map_3):
+		res = priority_map_3[priority][subpriority];
+		break;
+	case RTE_DIM(priority_map_5):
+		res = priority_map_5[priority][subpriority];
+		break;
+	}
+	return  res;
+}
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 5c68d4f7d7..8f94125f26 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1226,6 +1226,9 @@ typedef int (*mlx5_flow_create_def_policy_t)
 			(struct rte_eth_dev *dev);
 typedef void (*mlx5_flow_destroy_def_policy_t)
 			(struct rte_eth_dev *dev);
+typedef int (*mlx5_flow_discover_priorities_t)
+			(struct rte_eth_dev *dev,
+			 const uint16_t *vprio, int vprio_n);
 
 struct mlx5_flow_driver_ops {
 	mlx5_flow_validate_t validate;
@@ -1260,6 +1263,7 @@ struct mlx5_flow_driver_ops {
 	mlx5_flow_action_update_t action_update;
 	mlx5_flow_action_query_t action_query;
 	mlx5_flow_sync_domain_t sync_domain;
+	mlx5_flow_discover_priorities_t discover_priorities;
 };
 
 /* mlx5_flow.c */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index c6370cd1d6..155745748f 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -17978,6 +17978,108 @@ flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
 	return 0;
 }
 
+/**
+ * Discover the number of available flow priorities
+ * by trying to create a flow with the highest priority value
+ * for each possible number.
+ *
+ * @param[in] dev
+ *   Ethernet device.
+ * @param[in] vprio
+ *   List of possible number of available priorities.
+ * @param[in] vprio_n
+ *   Size of @p vprio array.
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+static int
+flow_dv_discover_priorities(struct rte_eth_dev *dev,
+			    const uint16_t *vprio, int vprio_n)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+	struct mlx5_flow_dv_matcher matcher = {
+		.mask = {
+			.size = sizeof(matcher.mask.buf),
+		},
+	};
+	union mlx5_flow_tbl_key tbl_key;
+	struct mlx5_flow flow;
+	void *action;
+	struct rte_flow_error error;
+	uint8_t misc_mask;
+	int i, err, ret = -ENOTSUP;
+
+	/*
+	 * Prepare a flow with a catch-all pattern and a drop action.
+	 * Use drop queue, because shared drop action may be unavailable.
+	 */
+	action = priv->drop_queue.hrxq->action;
+	if (action == NULL) {
+		DRV_LOG(ERR, "Priority discovery requires a drop action");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	memset(&flow, 0, sizeof(flow));
+	flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
+	if (flow.handle == NULL) {
+		DRV_LOG(ERR, "Cannot create flow handle");
+		rte_errno = ENOMEM;
+		return -rte_errno;
+	}
+	flow.ingress = true;
+	flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
+	flow.dv.actions[0] = action;
+	flow.dv.actions_n = 1;
+	memset(&eth, 0, sizeof(eth));
+	flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
+				   &item, /* inner */ false, /* group */ 0);
+	matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
+	for (i = 0; i < vprio_n; i++) {
+		/* Configure the next proposed maximum priority. */
+		matcher.priority = vprio[i] - 1;
+		memset(&tbl_key, 0, sizeof(tbl_key));
+		err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
+					       /* tunnel */ NULL,
+					       /* group */ 0,
+					       &error);
+		if (err != 0) {
+			/* This action is pure SW and must always succeed. */
+			DRV_LOG(ERR, "Cannot register matcher");
+			ret = -rte_errno;
+			break;
+		}
+		/* Try to apply the flow to HW. */
+		misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
+		__flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
+		err = mlx5_flow_os_create_flow
+				(flow.handle->dvh.matcher->matcher_object,
+				 (void *)&flow.dv.value, flow.dv.actions_n,
+				 flow.dv.actions, &flow.handle->drv_flow);
+		if (err == 0) {
+			claim_zero(mlx5_flow_os_destroy_flow
+						(flow.handle->drv_flow));
+			flow.handle->drv_flow = NULL;
+		}
+		claim_zero(flow_dv_matcher_release(dev, flow.handle));
+		if (err != 0)
+			break;
+		ret = vprio[i];
+	}
+	mlx5_ipool_free(pool, flow.handle_idx);
+	/* Set rte_errno if no expected priority value matched. */
+	if (ret < 0)
+		rte_errno = -ret;
+	return ret;
+}
+
 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.validate = flow_dv_validate,
 	.prepare = flow_dv_prepare,
@@ -18011,6 +18113,7 @@ const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.action_update = flow_dv_action_update,
 	.action_query = flow_dv_action_query,
 	.sync_domain = flow_dv_sync_domain,
+	.discover_priorities = flow_dv_discover_priorities,
 };
 
 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
diff --git a/drivers/net/mlx5/mlx5_flow_verbs.c b/drivers/net/mlx5/mlx5_flow_verbs.c
index b93fd4d2c9..72b9db6c7f 100644
--- a/drivers/net/mlx5/mlx5_flow_verbs.c
+++ b/drivers/net/mlx5/mlx5_flow_verbs.c
@@ -28,17 +28,6 @@
 #define VERBS_SPEC_INNER(item_flags) \
 	(!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
 
-/* Map of Verbs to Flow priority with 8 Verbs priorities. */
-static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
-};
-
-/* Map of Verbs to Flow priority with 16 Verbs priorities. */
-static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
-	{ 9, 10, 11 }, { 12, 13, 14 },
-};
-
 /* Verbs specification header. */
 struct ibv_spec_header {
 	enum ibv_flow_spec_type type;
@@ -50,13 +39,17 @@ struct ibv_spec_header {
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
- *
+ * @param[in] vprio
+ *   Expected result variants.
+ * @param[in] vprio_n
+ *   Number of entries in @p vprio array.
  * @return
- *   number of supported flow priority on success, a negative errno
+ *   Number of supported flow priority on success, a negative errno
  *   value otherwise and rte_errno is set.
  */
-int
-mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+static int
+flow_verbs_discover_priorities(struct rte_eth_dev *dev,
+			       const uint16_t *vprio, int vprio_n)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct {
@@ -79,7 +72,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	};
 	struct ibv_flow *flow;
 	struct mlx5_hrxq *drop = priv->drop_queue.hrxq;
-	uint16_t vprio[] = { 8, 16 };
 	int i;
 	int priority = 0;
 
@@ -87,7 +79,7 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		rte_errno = ENOTSUP;
 		return -rte_errno;
 	}
-	for (i = 0; i != RTE_DIM(vprio); i++) {
+	for (i = 0; i != vprio_n; i++) {
 		flow_attr.attr.priority = vprio[i] - 1;
 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 		if (!flow)
@@ -95,59 +87,9 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		claim_zero(mlx5_glue->destroy_flow(flow));
 		priority = vprio[i];
 	}
-	switch (priority) {
-	case 8:
-		priority = RTE_DIM(priority_map_3);
-		break;
-	case 16:
-		priority = RTE_DIM(priority_map_5);
-		break;
-	default:
-		rte_errno = ENOTSUP;
-		DRV_LOG(ERR,
-			"port %u verbs maximum priority: %d expected 8/16",
-			dev->data->port_id, priority);
-		return -rte_errno;
-	}
-	DRV_LOG(INFO, "port %u supported flow priorities:"
-		" 0-%d for ingress or egress root table,"
-		" 0-%d for non-root table or transfer root table.",
-		dev->data->port_id, priority - 2,
-		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
 	return priority;
 }
 
-/**
- * Adjust flow priority based on the highest layer and the request priority.
- *
- * @param[in] dev
- *   Pointer to the Ethernet device structure.
- * @param[in] priority
- *   The rule base priority.
- * @param[in] subpriority
- *   The priority based on the items.
- *
- * @return
- *   The new priority.
- */
-uint32_t
-mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
-				   uint32_t subpriority)
-{
-	uint32_t res = 0;
-	struct mlx5_priv *priv = dev->data->dev_private;
-
-	switch (priv->config.flow_prio) {
-	case RTE_DIM(priority_map_3):
-		res = priority_map_3[priority][subpriority];
-		break;
-	case RTE_DIM(priority_map_5):
-		res = priority_map_5[priority][subpriority];
-		break;
-	}
-	return  res;
-}
-
 /**
  * Get Verbs flow counter by index.
  *
@@ -2105,4 +2047,5 @@ const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
 	.destroy = flow_verbs_destroy,
 	.query = flow_verbs_query,
 	.sync_domain = flow_verbs_sync_domain,
+	.discover_priorities = flow_verbs_discover_priorities,
 };
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v2 4/5] net/mlx5: create drop queue using DevX
       [not found] ` <20211015161822.3099818-1-dkozlyuk@nvidia.com>
  2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 3/5] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
@ 2021-10-15 16:18   ` Dmitry Kozlyuk
  2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 5/5] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
       [not found]   ` <20211019123722.3414694-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-15 16:18 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Drop queue creation and destruction were not implemented for DevX
flow engine and Verbs engine methods were used as a workaround.
Implement these methods for DevX so that there is a valid queue ID
that can be used regardless of queue configuration via API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c |   4 -
 drivers/net/mlx5/mlx5_devx.c     | 211 ++++++++++++++++++++++++++-----
 2 files changed, 180 insertions(+), 35 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 8ee7ada51b..985f0bd489 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1790,10 +1790,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	}
 	if (config->devx && config->dv_flow_en && config->dest_tir) {
 		priv->obj_ops = devx_obj_ops;
-		priv->obj_ops.drop_action_create =
-						ibv_obj_ops.drop_action_create;
-		priv->obj_ops.drop_action_destroy =
-						ibv_obj_ops.drop_action_destroy;
 #ifndef HAVE_MLX5DV_DEVX_UAR_OFFSET
 		priv->obj_ops.txq_obj_modify = ibv_obj_ops.txq_obj_modify;
 #else
diff --git a/drivers/net/mlx5/mlx5_devx.c b/drivers/net/mlx5/mlx5_devx.c
index a1db53577a..1e62108c94 100644
--- a/drivers/net/mlx5/mlx5_devx.c
+++ b/drivers/net/mlx5/mlx5_devx.c
@@ -226,17 +226,17 @@ mlx5_rx_devx_get_event(struct mlx5_rxq_obj *rxq_obj)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	struct mlx5_devx_create_rq_attr rq_attr = { 0 };
@@ -289,20 +289,20 @@ mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_devx_cq *cq_obj = 0;
 	struct mlx5_devx_cq_attr cq_attr = { 0 };
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_dev_ctx_shared *sh = priv->sh;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	unsigned int cqe_n = mlx5_rxq_cqe_num(rxq_data);
@@ -497,13 +497,13 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
 		tmpl->fd = mlx5_os_get_devx_channel_fd(tmpl->devx_channel);
 	}
 	/* Create CQ using DevX API. */
-	ret = mlx5_rxq_create_devx_cq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Failed to create CQ.");
 		goto error;
 	}
 	/* Create RQ using DevX API. */
-	ret = mlx5_rxq_create_devx_rq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Port %u Rx queue %u RQ creation failure.",
 			dev->data->port_id, idx);
@@ -536,6 +536,11 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
  *   Pointer to Ethernet device.
  * @param log_n
  *   Log of number of queues in the array.
+ * @param queues
+ *   List of RX queue indices or NULL, in which case
+ *   the attribute will be filled by drop queue ID.
+ * @param queues_n
+ *   Size of @p queues array or 0 if it is NULL.
  * @param ind_tbl
  *   DevX indirection table object.
  *
@@ -563,6 +568,11 @@ mlx5_devx_ind_table_create_rqt_attr(struct rte_eth_dev *dev,
 	}
 	rqt_attr->rqt_max_size = priv->config.ind_table_max_size;
 	rqt_attr->rqt_actual_size = rqt_n;
+	if (queues == NULL) {
+		for (i = 0; i < rqt_n; i++)
+			rqt_attr->rq_list[i] = priv->drop_queue.rxq->rq->id;
+		return rqt_attr;
+	}
 	for (i = 0; i != queues_n; ++i) {
 		struct mlx5_rxq_data *rxq = (*priv->rxqs)[queues[i]];
 		struct mlx5_rxq_ctrl *rxq_ctrl =
@@ -595,11 +605,12 @@ mlx5_devx_ind_table_new(struct rte_eth_dev *dev, const unsigned int log_n,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_devx_rqt_attr *rqt_attr = NULL;
+	const uint16_t *queues = dev->data->dev_started ? ind_tbl->queues :
+							  NULL;
 
 	MLX5_ASSERT(ind_tbl);
-	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n,
-							ind_tbl->queues,
-							ind_tbl->queues_n);
+	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n, queues,
+						       ind_tbl->queues_n);
 	if (!rqt_attr)
 		return -rte_errno;
 	ind_tbl->rqt = mlx5_devx_cmd_create_rqt(priv->sh->ctx, rqt_attr);
@@ -670,7 +681,8 @@ mlx5_devx_ind_table_destroy(struct mlx5_ind_table_obj *ind_tbl)
  * @param[in] hash_fields
  *   Verbs protocol hash field to make the RSS on.
  * @param[in] ind_tbl
- *   Indirection table for TIR.
+ *   Indirection table for TIR. If table queues array is NULL,
+ *   a TIR for drop queue is assumed.
  * @param[in] tunnel
  *   Tunnel type.
  * @param[out] tir_attr
@@ -686,19 +698,27 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 		       int tunnel, struct mlx5_devx_tir_attr *tir_attr)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[ind_tbl->queues[0]];
-	struct mlx5_rxq_ctrl *rxq_ctrl =
-		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
-	enum mlx5_rxq_type rxq_obj_type = rxq_ctrl->type;
+	enum mlx5_rxq_type rxq_obj_type;
 	bool lro = true;
 	uint32_t i;
 
-	/* Enable TIR LRO only if all the queues were configured for. */
-	for (i = 0; i < ind_tbl->queues_n; ++i) {
-		if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
-			lro = false;
-			break;
+	/* NULL queues designate drop queue. */
+	if (ind_tbl->queues != NULL) {
+		struct mlx5_rxq_data *rxq_data =
+					(*priv->rxqs)[ind_tbl->queues[0]];
+		struct mlx5_rxq_ctrl *rxq_ctrl =
+			container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+		rxq_obj_type = rxq_ctrl->type;
+
+		/* Enable TIR LRO only if all the queues were configured for. */
+		for (i = 0; i < ind_tbl->queues_n; ++i) {
+			if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
+				lro = false;
+				break;
+			}
 		}
+	} else {
+		rxq_obj_type = priv->drop_queue.rxq->rxq_ctrl->type;
 	}
 	memset(tir_attr, 0, sizeof(*tir_attr));
 	tir_attr->disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
@@ -857,7 +877,7 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
 }
 
 /**
- * Create a DevX drop action for Rx Hash queue.
+ * Create a DevX drop Rx queue.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -866,14 +886,99 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+mlx5_rxq_devx_obj_drop_create(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	int socket_id = dev->device->numa_node;
+	struct mlx5_rxq_ctrl *rxq_ctrl;
+	struct mlx5_rxq_data *rxq_data;
+	struct mlx5_rxq_obj *rxq = NULL;
+	int ret;
+
+	/*
+	 * Initialize dummy control structures.
+	 * They are required to hold pointers for cleanup
+	 * and are only accessible via drop queue DevX objects.
+	 */
+	rxq_ctrl = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq_ctrl),
+			       0, socket_id);
+	if (rxq_ctrl == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue control",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq), 0, socket_id);
+	if (rxq == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue object",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq->rxq_ctrl = rxq_ctrl;
+	rxq_ctrl->type = MLX5_RXQ_TYPE_STANDARD;
+	rxq_ctrl->priv = priv;
+	rxq_ctrl->obj = rxq;
+	rxq_data = &rxq_ctrl->rxq;
+	/* Create CQ using DevX API. */
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue CQ creation failed.",
+			dev->data->port_id);
+		goto error;
+	}
+	/* Create RQ using DevX API. */
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue RQ creation failed.",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	/* Change queue state to ready. */
+	ret = mlx5_devx_modify_rq(rxq, MLX5_RXQ_MOD_RST2RDY);
+	if (ret != 0)
+		goto error;
+	/* Initialize drop queue. */
+	priv->drop_queue.rxq = rxq;
+	return 0;
+error:
+	ret = rte_errno; /* Save rte_errno before cleanup. */
+	if (rxq != NULL) {
+		if (rxq->rq_obj.rq != NULL)
+			mlx5_devx_rq_destroy(&rxq->rq_obj);
+		if (rxq->cq_obj.cq != NULL)
+			mlx5_devx_cq_destroy(&rxq->cq_obj);
+		if (rxq->devx_channel)
+			mlx5_os_devx_destroy_event_channel
+							(rxq->devx_channel);
+		mlx5_free(rxq);
+	}
+	if (rxq_ctrl != NULL)
+		mlx5_free(rxq_ctrl);
+	rte_errno = ret; /* Restore rte_errno. */
 	return -rte_errno;
 }
 
+/**
+ * Release drop Rx queue resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ */
+static void
+mlx5_rxq_devx_obj_drop_release(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_obj *rxq = priv->drop_queue.rxq;
+	struct mlx5_rxq_ctrl *rxq_ctrl = rxq->rxq_ctrl;
+
+	mlx5_rxq_devx_obj_release(rxq);
+	mlx5_free(rxq);
+	mlx5_free(rxq_ctrl);
+	priv->drop_queue.rxq = NULL;
+}
+
 /**
  * Release a drop hash Rx queue.
  *
@@ -883,9 +988,53 @@ mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
 static void
 mlx5_devx_drop_action_destroy(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+
+	if (hrxq->tir != NULL)
+		mlx5_devx_tir_destroy(hrxq);
+	if (hrxq->ind_table->ind_table != NULL)
+		mlx5_devx_ind_table_destroy(hrxq->ind_table);
+	if (priv->drop_queue.rxq->rq != NULL)
+		mlx5_rxq_devx_obj_drop_release(dev);
+}
+
+/**
+ * Create a DevX drop action for Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+	int ret;
+
+	ret = mlx5_rxq_devx_obj_drop_create(dev);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop RX queue");
+		return ret;
+	}
+	/* hrxq->ind_table queues are NULL, drop RX queue ID will be used */
+	ret = mlx5_devx_ind_table_new(dev, 0, hrxq->ind_table);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue indirection table");
+		goto error;
+	}
+	ret = mlx5_devx_hrxq_new(dev, hrxq, /* tunnel */ false);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue");
+		goto error;
+	}
+	return 0;
+error:
+	mlx5_devx_drop_action_destroy(dev);
+	return ret;
 }
 
 /**
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v2 5/5] net/mlx5: preserve indirect actions on restart
       [not found] ` <20211015161822.3099818-1-dkozlyuk@nvidia.com>
  2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 3/5] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
  2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 4/5] net/mlx5: create drop queue " Dmitry Kozlyuk
@ 2021-10-15 16:18   ` Dmitry Kozlyuk
       [not found]   ` <20211019123722.3414694-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-15 16:18 UTC (permalink / raw)
  To: dev; +Cc: bingz, stable, Matan Azrad, Viacheslav Ovsiienko

MLX5 PMD uses reference counting to manage RX queue resources.
After port stop shared RSS actions kept references to RX queues,
preventing resource release. As a result, internal PMD mempool
for such queues had been exhausted after a number of port restarts.
Diagnostic message from rte_eth_dev_start():

    Rx queue allocation failed: Cannot allocate memory

Dereference RX queues used by indirect actions on port stop (detach)
and restore references on port start (attach) in order to allow RX queue
resource release, but keep indirect RSS across the port restart.
Replace queue IDs in HW by drop queue ID on detach and restore actual
queue IDs on attach.

When the port is stopped, create indirect RSS in the detached state.
As a result, MLX5 PMD is able to keep all its indirect actions
across port restart. Advertise this capability.

Fixes: 4b61b8774be9 ("ethdev: introduce indirect flow action")
Cc: bingz@nvidia.com
Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/mlx5_ethdev.c  |   1 +
 drivers/net/mlx5/mlx5_flow.c    | 194 ++++++++++++++++++++++++++++----
 drivers/net/mlx5/mlx5_flow.h    |   2 +
 drivers/net/mlx5/mlx5_rx.h      |   4 +
 drivers/net/mlx5/mlx5_rxq.c     |  99 ++++++++++++++--
 drivers/net/mlx5/mlx5_trigger.c |  10 ++
 6 files changed, 276 insertions(+), 34 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 82e2284d98..419fec3e4e 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -321,6 +321,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
 				 info->rx_queue_offload_capa);
 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
+	info->dev_capa = RTE_ETH_DEV_CAPA_FLOW_SHARED_OBJECT_KEEP;
 	info->if_index = mlx5_ifindex(dev);
 	info->reta_size = priv->reta_idx_n ?
 		priv->reta_idx_n : config->ind_table_max_size;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index bfc3e20c9a..c10b911259 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1560,6 +1560,58 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 	return 0;
 }
 
+/**
+ * Validate queue numbers for device RSS.
+ *
+ * @param[in] dev
+ *   Configured device.
+ * @param[in] queues
+ *   Array of queue numbers.
+ * @param[in] queues_n
+ *   Size of the @p queues array.
+ * @param[out] error
+ *   On error, filled with a textual error description.
+ * @param[out] queue
+ *   On error, filled with an offending queue index in @p queues array.
+ *
+ * @return
+ *   0 on success, a negative errno code on error.
+ */
+static int
+mlx5_validate_rss_queues(const struct rte_eth_dev *dev,
+			 const uint16_t *queues, uint32_t queues_n,
+			 const char **error, uint32_t *queue_idx)
+{
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
+	uint32_t i;
+
+	for (i = 0; i != queues_n; ++i) {
+		struct mlx5_rxq_ctrl *rxq_ctrl;
+
+		if (queues[i] >= priv->rxqs_n) {
+			*error = "queue index out of range";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		if (!(*priv->rxqs)[queues[i]]) {
+			*error =  "queue is not configured";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		rxq_ctrl = container_of((*priv->rxqs)[queues[i]],
+					struct mlx5_rxq_ctrl, rxq);
+		if (i == 0)
+			rxq_type = rxq_ctrl->type;
+		if (rxq_type != rxq_ctrl->type) {
+			*error = "combining hairpin and regular RSS queues is not supported";
+			*queue_idx = i;
+			return -ENOTSUP;
+		}
+	}
+	return 0;
+}
+
 /*
  * Validate the rss action.
  *
@@ -1580,8 +1632,9 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_rss *rss = action->conf;
-	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
-	unsigned int i;
+	int ret;
+	const char *message;
+	uint32_t queue_idx;
 
 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
@@ -1645,27 +1698,12 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  NULL, "No queues configured");
-	for (i = 0; i != rss->queue_num; ++i) {
-		struct mlx5_rxq_ctrl *rxq_ctrl;
-
-		if (rss->queue[i] >= priv->rxqs_n)
-			return rte_flow_error_set
-				(error, EINVAL,
-				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue index out of range");
-		if (!(*priv->rxqs)[rss->queue[i]])
-			return rte_flow_error_set
-				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue is not configured");
-		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
-					struct mlx5_rxq_ctrl, rxq);
-		if (i == 0)
-			rxq_type = rxq_ctrl->type;
-		if (rxq_type != rxq_ctrl->type)
-			return rte_flow_error_set
-				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i],
-				 "combining hairpin and regular RSS queues is not supported");
+	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
+				       &message, &queue_idx);
+	if (ret != 0) {
+		return rte_flow_error_set(error, -ret,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &rss->queue[queue_idx], message);
 	}
 	return 0;
 }
@@ -8547,6 +8585,116 @@ mlx5_action_handle_flush(struct rte_eth_dev *dev)
 	return ret;
 }
 
+/**
+ * Validate existing indirect actions against current device configuration
+ * and attach them to device resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_attach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+		const char *message;
+		uint32_t queue_idx;
+
+		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
+					       ind_tbl->queues_n,
+					       &message, &queue_idx);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
+				dev->data->port_id, ind_tbl->queues[queue_idx],
+				message);
+			break;
+		}
+	}
+	if (ret != 0)
+		return ret;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not attach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not detach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
+/**
+ * Detach indirect actions of the device from its resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_detach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not detach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not attach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
 #ifndef HAVE_MLX5DV_DR
 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
 #else
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 8f94125f26..6bc7946cc3 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1574,6 +1574,8 @@ void mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
 		struct mlx5_flow_meter_policy *mtr_policy);
 int mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev);
 int mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev);
+int mlx5_action_handle_attach(struct rte_eth_dev *dev);
+int mlx5_action_handle_detach(struct rte_eth_dev *dev);
 int mlx5_action_handle_flush(struct rte_eth_dev *dev);
 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id);
 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh);
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 2b7ad3e48b..d44c8078de 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -222,6 +222,10 @@ int mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 			      struct mlx5_ind_table_obj *ind_tbl,
 			      uint16_t *queues, const uint32_t queues_n,
 			      bool standalone);
+int mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
+int mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
 struct mlx5_list_entry *mlx5_hrxq_create_cb(void *tool_ctx, void *cb_ctx);
 int mlx5_hrxq_match_cb(void *tool_ctx, struct mlx5_list_entry *entry,
 		       void *cb_ctx);
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index b68443bed5..fd2b5779ff 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -2015,6 +2015,26 @@ mlx5_ind_table_obj_new(struct rte_eth_dev *dev, const uint16_t *queues,
 	return ind_tbl;
 }
 
+static int
+mlx5_ind_table_obj_check_standalone(struct rte_eth_dev *dev __rte_unused,
+				    struct mlx5_ind_table_obj *ind_tbl)
+{
+	uint32_t refcnt;
+
+	refcnt = __atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED);
+	if (refcnt <= 1)
+		return 0;
+	/*
+	 * Modification of indirection tables having more than 1
+	 * reference is unsupported.
+	 */
+	DRV_LOG(DEBUG,
+		"Port %u cannot modify indirection table %p (refcnt %u > 1).",
+		dev->data->port_id, (void *)ind_tbl, refcnt);
+	rte_errno = EINVAL;
+	return -rte_errno;
+}
+
 /**
  * Modify an indirection table.
  *
@@ -2047,18 +2067,8 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 
 	MLX5_ASSERT(standalone);
 	RTE_SET_USED(standalone);
-	if (__atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED) > 1) {
-		/*
-		 * Modification of indirection ntables having more than 1
-		 * reference unsupported. Intended for standalone indirection
-		 * tables only.
-		 */
-		DRV_LOG(DEBUG,
-			"Port %u cannot modify indirection table (refcnt> 1).",
-			dev->data->port_id);
-		rte_errno = EINVAL;
+	if (mlx5_ind_table_obj_check_standalone(dev, ind_tbl) < 0)
 		return -rte_errno;
-	}
 	for (i = 0; i != queues_n; ++i) {
 		if (!mlx5_rxq_get(dev, queues[i])) {
 			ret = -rte_errno;
@@ -2084,6 +2094,73 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 	return ret;
 }
 
+/**
+ * Attach an indirection table to its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to attach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_modify(dev, ind_tbl, ind_tbl->queues,
+					ind_tbl->queues_n, true);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_get(dev, ind_tbl->queues[i]);
+	return 0;
+}
+
+/**
+ * Detach an indirection table from its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to detach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	const unsigned int n = rte_is_power_of_2(ind_tbl->queues_n) ?
+			       log2above(ind_tbl->queues_n) :
+			       log2above(priv->config.ind_table_max_size);
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_check_standalone(dev, ind_tbl);
+	if (ret != 0)
+		return ret;
+	MLX5_ASSERT(priv->obj_ops.ind_table_modify);
+	ret = priv->obj_ops.ind_table_modify(dev, n, NULL, 0, ind_tbl);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_release(dev, ind_tbl->queues[i]);
+	return ret;
+}
+
 int
 mlx5_hrxq_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
 		   void *cb_ctx)
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 54173bfacb..c3adf5082e 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -14,6 +14,7 @@
 #include <mlx5_malloc.h>
 
 #include "mlx5.h"
+#include "mlx5_flow.h"
 #include "mlx5_mr.h"
 #include "mlx5_rx.h"
 #include "mlx5_tx.h"
@@ -1113,6 +1114,14 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 	mlx5_rxq_timestamp_set(dev);
 	/* Set a mask and offset of scheduling on timestamp into Tx queues. */
 	mlx5_txq_dynf_timestamp_set(dev);
+	/* Attach indirection table objects detached on port stop. */
+	ret = mlx5_action_handle_attach(dev);
+	if (ret) {
+		DRV_LOG(ERR,
+			"port %u failed to attach indirect actions: %s",
+			dev->data->port_id, rte_strerror(rte_errno));
+		goto error;
+	}
 	/*
 	 * In non-cached mode, it only needs to start the default mreg copy
 	 * action and no flow created by application exists anymore.
@@ -1185,6 +1194,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	/* All RX queue flags will be cleared in the flush interface. */
 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, true);
 	mlx5_flow_meter_rxq_flush(dev);
+	mlx5_action_handle_detach(dev);
 	mlx5_rx_intr_vec_disable(dev);
 	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
 	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v3 4/6] net/mlx5: discover max flow priority using DevX
       [not found]   ` <20211019123722.3414694-1-dkozlyuk@nvidia.com>
@ 2021-10-19 12:37     ` Dmitry Kozlyuk
  2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
                       ` (2 subsequent siblings)
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-19 12:37 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Maximum available flow priority was discovered using Verbs API
regardless of the selected flow engine. This required some Verbs
objects to be initialized in order to use DevX engine. Make priority
discovery an engine method and implement it for DevX using its API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c   |   1 -
 drivers/net/mlx5/mlx5_flow.c       |  98 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow.h       |   4 ++
 drivers/net/mlx5/mlx5_flow_dv.c    | 103 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  77 +++------------------
 5 files changed, 215 insertions(+), 68 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 3746057673..8ee7ada51b 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1830,7 +1830,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	priv->drop_queue.hrxq = mlx5_drop_action_create(eth_dev);
 	if (!priv->drop_queue.hrxq)
 		goto error;
-	/* Supported Verbs flow priority number detection. */
 	err = mlx5_flow_discover_priorities(eth_dev);
 	if (err < 0) {
 		err = -err;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index c914a7120c..bfc3e20c9a 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -9416,3 +9416,101 @@ mlx5_dbg__print_pattern(const struct rte_flow_item *item)
 	}
 	printf("END\n");
 }
+
+/* Map of Verbs to Flow priority with 8 Verbs priorities. */
+static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
+};
+
+/* Map of Verbs to Flow priority with 16 Verbs priorities. */
+static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
+	{ 9, 10, 11 }, { 12, 13, 14 },
+};
+
+/**
+ * Discover the number of available flow priorities.
+ *
+ * @param dev
+ *   Ethernet device.
+ *
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+int
+mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+{
+	static const uint16_t vprio[] = {8, 16};
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	const struct mlx5_flow_driver_ops *fops;
+	enum mlx5_flow_drv_type type;
+	int ret;
+
+	type = mlx5_flow_os_get_type();
+	if (type == MLX5_FLOW_TYPE_MAX) {
+		type = MLX5_FLOW_TYPE_VERBS;
+		if (priv->config.devx && priv->config.dv_flow_en)
+			type = MLX5_FLOW_TYPE_DV;
+	}
+	fops = flow_get_drv_ops(type);
+	if (fops->discover_priorities == NULL) {
+		DRV_LOG(ERR, "Priority discovery not supported");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
+	if (ret < 0)
+		return ret;
+	switch (ret) {
+	case 8:
+		ret = RTE_DIM(priority_map_3);
+		break;
+	case 16:
+		ret = RTE_DIM(priority_map_5);
+		break;
+	default:
+		rte_errno = ENOTSUP;
+		DRV_LOG(ERR,
+			"port %u maximum priority: %d expected 8/16",
+			dev->data->port_id, ret);
+		return -rte_errno;
+	}
+	DRV_LOG(INFO, "port %u supported flow priorities:"
+		" 0-%d for ingress or egress root table,"
+		" 0-%d for non-root table or transfer root table.",
+		dev->data->port_id, ret - 2,
+		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
+	return ret;
+}
+
+/**
+ * Adjust flow priority based on the highest layer and the request priority.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] priority
+ *   The rule base priority.
+ * @param[in] subpriority
+ *   The priority based on the items.
+ *
+ * @return
+ *   The new priority.
+ */
+uint32_t
+mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
+			  uint32_t subpriority)
+{
+	uint32_t res = 0;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	switch (priv->config.flow_prio) {
+	case RTE_DIM(priority_map_3):
+		res = priority_map_3[priority][subpriority];
+		break;
+	case RTE_DIM(priority_map_5):
+		res = priority_map_5[priority][subpriority];
+		break;
+	}
+	return  res;
+}
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 5c68d4f7d7..8f94125f26 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1226,6 +1226,9 @@ typedef int (*mlx5_flow_create_def_policy_t)
 			(struct rte_eth_dev *dev);
 typedef void (*mlx5_flow_destroy_def_policy_t)
 			(struct rte_eth_dev *dev);
+typedef int (*mlx5_flow_discover_priorities_t)
+			(struct rte_eth_dev *dev,
+			 const uint16_t *vprio, int vprio_n);
 
 struct mlx5_flow_driver_ops {
 	mlx5_flow_validate_t validate;
@@ -1260,6 +1263,7 @@ struct mlx5_flow_driver_ops {
 	mlx5_flow_action_update_t action_update;
 	mlx5_flow_action_query_t action_query;
 	mlx5_flow_sync_domain_t sync_domain;
+	mlx5_flow_discover_priorities_t discover_priorities;
 };
 
 /* mlx5_flow.c */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index c6370cd1d6..155745748f 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -17978,6 +17978,108 @@ flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
 	return 0;
 }
 
+/**
+ * Discover the number of available flow priorities
+ * by trying to create a flow with the highest priority value
+ * for each possible number.
+ *
+ * @param[in] dev
+ *   Ethernet device.
+ * @param[in] vprio
+ *   List of possible number of available priorities.
+ * @param[in] vprio_n
+ *   Size of @p vprio array.
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+static int
+flow_dv_discover_priorities(struct rte_eth_dev *dev,
+			    const uint16_t *vprio, int vprio_n)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+	struct mlx5_flow_dv_matcher matcher = {
+		.mask = {
+			.size = sizeof(matcher.mask.buf),
+		},
+	};
+	union mlx5_flow_tbl_key tbl_key;
+	struct mlx5_flow flow;
+	void *action;
+	struct rte_flow_error error;
+	uint8_t misc_mask;
+	int i, err, ret = -ENOTSUP;
+
+	/*
+	 * Prepare a flow with a catch-all pattern and a drop action.
+	 * Use drop queue, because shared drop action may be unavailable.
+	 */
+	action = priv->drop_queue.hrxq->action;
+	if (action == NULL) {
+		DRV_LOG(ERR, "Priority discovery requires a drop action");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	memset(&flow, 0, sizeof(flow));
+	flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
+	if (flow.handle == NULL) {
+		DRV_LOG(ERR, "Cannot create flow handle");
+		rte_errno = ENOMEM;
+		return -rte_errno;
+	}
+	flow.ingress = true;
+	flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
+	flow.dv.actions[0] = action;
+	flow.dv.actions_n = 1;
+	memset(&eth, 0, sizeof(eth));
+	flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
+				   &item, /* inner */ false, /* group */ 0);
+	matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
+	for (i = 0; i < vprio_n; i++) {
+		/* Configure the next proposed maximum priority. */
+		matcher.priority = vprio[i] - 1;
+		memset(&tbl_key, 0, sizeof(tbl_key));
+		err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
+					       /* tunnel */ NULL,
+					       /* group */ 0,
+					       &error);
+		if (err != 0) {
+			/* This action is pure SW and must always succeed. */
+			DRV_LOG(ERR, "Cannot register matcher");
+			ret = -rte_errno;
+			break;
+		}
+		/* Try to apply the flow to HW. */
+		misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
+		__flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
+		err = mlx5_flow_os_create_flow
+				(flow.handle->dvh.matcher->matcher_object,
+				 (void *)&flow.dv.value, flow.dv.actions_n,
+				 flow.dv.actions, &flow.handle->drv_flow);
+		if (err == 0) {
+			claim_zero(mlx5_flow_os_destroy_flow
+						(flow.handle->drv_flow));
+			flow.handle->drv_flow = NULL;
+		}
+		claim_zero(flow_dv_matcher_release(dev, flow.handle));
+		if (err != 0)
+			break;
+		ret = vprio[i];
+	}
+	mlx5_ipool_free(pool, flow.handle_idx);
+	/* Set rte_errno if no expected priority value matched. */
+	if (ret < 0)
+		rte_errno = -ret;
+	return ret;
+}
+
 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.validate = flow_dv_validate,
 	.prepare = flow_dv_prepare,
@@ -18011,6 +18113,7 @@ const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.action_update = flow_dv_action_update,
 	.action_query = flow_dv_action_query,
 	.sync_domain = flow_dv_sync_domain,
+	.discover_priorities = flow_dv_discover_priorities,
 };
 
 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
diff --git a/drivers/net/mlx5/mlx5_flow_verbs.c b/drivers/net/mlx5/mlx5_flow_verbs.c
index b93fd4d2c9..72b9db6c7f 100644
--- a/drivers/net/mlx5/mlx5_flow_verbs.c
+++ b/drivers/net/mlx5/mlx5_flow_verbs.c
@@ -28,17 +28,6 @@
 #define VERBS_SPEC_INNER(item_flags) \
 	(!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
 
-/* Map of Verbs to Flow priority with 8 Verbs priorities. */
-static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
-};
-
-/* Map of Verbs to Flow priority with 16 Verbs priorities. */
-static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
-	{ 9, 10, 11 }, { 12, 13, 14 },
-};
-
 /* Verbs specification header. */
 struct ibv_spec_header {
 	enum ibv_flow_spec_type type;
@@ -50,13 +39,17 @@ struct ibv_spec_header {
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
- *
+ * @param[in] vprio
+ *   Expected result variants.
+ * @param[in] vprio_n
+ *   Number of entries in @p vprio array.
  * @return
- *   number of supported flow priority on success, a negative errno
+ *   Number of supported flow priority on success, a negative errno
  *   value otherwise and rte_errno is set.
  */
-int
-mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+static int
+flow_verbs_discover_priorities(struct rte_eth_dev *dev,
+			       const uint16_t *vprio, int vprio_n)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct {
@@ -79,7 +72,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	};
 	struct ibv_flow *flow;
 	struct mlx5_hrxq *drop = priv->drop_queue.hrxq;
-	uint16_t vprio[] = { 8, 16 };
 	int i;
 	int priority = 0;
 
@@ -87,7 +79,7 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		rte_errno = ENOTSUP;
 		return -rte_errno;
 	}
-	for (i = 0; i != RTE_DIM(vprio); i++) {
+	for (i = 0; i != vprio_n; i++) {
 		flow_attr.attr.priority = vprio[i] - 1;
 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 		if (!flow)
@@ -95,59 +87,9 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		claim_zero(mlx5_glue->destroy_flow(flow));
 		priority = vprio[i];
 	}
-	switch (priority) {
-	case 8:
-		priority = RTE_DIM(priority_map_3);
-		break;
-	case 16:
-		priority = RTE_DIM(priority_map_5);
-		break;
-	default:
-		rte_errno = ENOTSUP;
-		DRV_LOG(ERR,
-			"port %u verbs maximum priority: %d expected 8/16",
-			dev->data->port_id, priority);
-		return -rte_errno;
-	}
-	DRV_LOG(INFO, "port %u supported flow priorities:"
-		" 0-%d for ingress or egress root table,"
-		" 0-%d for non-root table or transfer root table.",
-		dev->data->port_id, priority - 2,
-		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
 	return priority;
 }
 
-/**
- * Adjust flow priority based on the highest layer and the request priority.
- *
- * @param[in] dev
- *   Pointer to the Ethernet device structure.
- * @param[in] priority
- *   The rule base priority.
- * @param[in] subpriority
- *   The priority based on the items.
- *
- * @return
- *   The new priority.
- */
-uint32_t
-mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
-				   uint32_t subpriority)
-{
-	uint32_t res = 0;
-	struct mlx5_priv *priv = dev->data->dev_private;
-
-	switch (priv->config.flow_prio) {
-	case RTE_DIM(priority_map_3):
-		res = priority_map_3[priority][subpriority];
-		break;
-	case RTE_DIM(priority_map_5):
-		res = priority_map_5[priority][subpriority];
-		break;
-	}
-	return  res;
-}
-
 /**
  * Get Verbs flow counter by index.
  *
@@ -2105,4 +2047,5 @@ const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
 	.destroy = flow_verbs_destroy,
 	.query = flow_verbs_query,
 	.sync_domain = flow_verbs_sync_domain,
+	.discover_priorities = flow_verbs_discover_priorities,
 };
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v3 5/6] net/mlx5: create drop queue using DevX
       [not found]   ` <20211019123722.3414694-1-dkozlyuk@nvidia.com>
  2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
@ 2021-10-19 12:37     ` Dmitry Kozlyuk
  2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
       [not found]     ` <20211021063503.3632732-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-19 12:37 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Drop queue creation and destruction were not implemented for DevX
flow engine and Verbs engine methods were used as a workaround.
Implement these methods for DevX so that there is a valid queue ID
that can be used regardless of queue configuration via API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c |   4 -
 drivers/net/mlx5/mlx5_devx.c     | 211 ++++++++++++++++++++++++++-----
 2 files changed, 180 insertions(+), 35 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 8ee7ada51b..985f0bd489 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1790,10 +1790,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	}
 	if (config->devx && config->dv_flow_en && config->dest_tir) {
 		priv->obj_ops = devx_obj_ops;
-		priv->obj_ops.drop_action_create =
-						ibv_obj_ops.drop_action_create;
-		priv->obj_ops.drop_action_destroy =
-						ibv_obj_ops.drop_action_destroy;
 #ifndef HAVE_MLX5DV_DEVX_UAR_OFFSET
 		priv->obj_ops.txq_obj_modify = ibv_obj_ops.txq_obj_modify;
 #else
diff --git a/drivers/net/mlx5/mlx5_devx.c b/drivers/net/mlx5/mlx5_devx.c
index a1db53577a..1e62108c94 100644
--- a/drivers/net/mlx5/mlx5_devx.c
+++ b/drivers/net/mlx5/mlx5_devx.c
@@ -226,17 +226,17 @@ mlx5_rx_devx_get_event(struct mlx5_rxq_obj *rxq_obj)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	struct mlx5_devx_create_rq_attr rq_attr = { 0 };
@@ -289,20 +289,20 @@ mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_devx_cq *cq_obj = 0;
 	struct mlx5_devx_cq_attr cq_attr = { 0 };
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_dev_ctx_shared *sh = priv->sh;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	unsigned int cqe_n = mlx5_rxq_cqe_num(rxq_data);
@@ -497,13 +497,13 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
 		tmpl->fd = mlx5_os_get_devx_channel_fd(tmpl->devx_channel);
 	}
 	/* Create CQ using DevX API. */
-	ret = mlx5_rxq_create_devx_cq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Failed to create CQ.");
 		goto error;
 	}
 	/* Create RQ using DevX API. */
-	ret = mlx5_rxq_create_devx_rq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Port %u Rx queue %u RQ creation failure.",
 			dev->data->port_id, idx);
@@ -536,6 +536,11 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
  *   Pointer to Ethernet device.
  * @param log_n
  *   Log of number of queues in the array.
+ * @param queues
+ *   List of RX queue indices or NULL, in which case
+ *   the attribute will be filled by drop queue ID.
+ * @param queues_n
+ *   Size of @p queues array or 0 if it is NULL.
  * @param ind_tbl
  *   DevX indirection table object.
  *
@@ -563,6 +568,11 @@ mlx5_devx_ind_table_create_rqt_attr(struct rte_eth_dev *dev,
 	}
 	rqt_attr->rqt_max_size = priv->config.ind_table_max_size;
 	rqt_attr->rqt_actual_size = rqt_n;
+	if (queues == NULL) {
+		for (i = 0; i < rqt_n; i++)
+			rqt_attr->rq_list[i] = priv->drop_queue.rxq->rq->id;
+		return rqt_attr;
+	}
 	for (i = 0; i != queues_n; ++i) {
 		struct mlx5_rxq_data *rxq = (*priv->rxqs)[queues[i]];
 		struct mlx5_rxq_ctrl *rxq_ctrl =
@@ -595,11 +605,12 @@ mlx5_devx_ind_table_new(struct rte_eth_dev *dev, const unsigned int log_n,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_devx_rqt_attr *rqt_attr = NULL;
+	const uint16_t *queues = dev->data->dev_started ? ind_tbl->queues :
+							  NULL;
 
 	MLX5_ASSERT(ind_tbl);
-	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n,
-							ind_tbl->queues,
-							ind_tbl->queues_n);
+	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n, queues,
+						       ind_tbl->queues_n);
 	if (!rqt_attr)
 		return -rte_errno;
 	ind_tbl->rqt = mlx5_devx_cmd_create_rqt(priv->sh->ctx, rqt_attr);
@@ -670,7 +681,8 @@ mlx5_devx_ind_table_destroy(struct mlx5_ind_table_obj *ind_tbl)
  * @param[in] hash_fields
  *   Verbs protocol hash field to make the RSS on.
  * @param[in] ind_tbl
- *   Indirection table for TIR.
+ *   Indirection table for TIR. If table queues array is NULL,
+ *   a TIR for drop queue is assumed.
  * @param[in] tunnel
  *   Tunnel type.
  * @param[out] tir_attr
@@ -686,19 +698,27 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 		       int tunnel, struct mlx5_devx_tir_attr *tir_attr)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[ind_tbl->queues[0]];
-	struct mlx5_rxq_ctrl *rxq_ctrl =
-		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
-	enum mlx5_rxq_type rxq_obj_type = rxq_ctrl->type;
+	enum mlx5_rxq_type rxq_obj_type;
 	bool lro = true;
 	uint32_t i;
 
-	/* Enable TIR LRO only if all the queues were configured for. */
-	for (i = 0; i < ind_tbl->queues_n; ++i) {
-		if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
-			lro = false;
-			break;
+	/* NULL queues designate drop queue. */
+	if (ind_tbl->queues != NULL) {
+		struct mlx5_rxq_data *rxq_data =
+					(*priv->rxqs)[ind_tbl->queues[0]];
+		struct mlx5_rxq_ctrl *rxq_ctrl =
+			container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+		rxq_obj_type = rxq_ctrl->type;
+
+		/* Enable TIR LRO only if all the queues were configured for. */
+		for (i = 0; i < ind_tbl->queues_n; ++i) {
+			if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
+				lro = false;
+				break;
+			}
 		}
+	} else {
+		rxq_obj_type = priv->drop_queue.rxq->rxq_ctrl->type;
 	}
 	memset(tir_attr, 0, sizeof(*tir_attr));
 	tir_attr->disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
@@ -857,7 +877,7 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
 }
 
 /**
- * Create a DevX drop action for Rx Hash queue.
+ * Create a DevX drop Rx queue.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -866,14 +886,99 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+mlx5_rxq_devx_obj_drop_create(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	int socket_id = dev->device->numa_node;
+	struct mlx5_rxq_ctrl *rxq_ctrl;
+	struct mlx5_rxq_data *rxq_data;
+	struct mlx5_rxq_obj *rxq = NULL;
+	int ret;
+
+	/*
+	 * Initialize dummy control structures.
+	 * They are required to hold pointers for cleanup
+	 * and are only accessible via drop queue DevX objects.
+	 */
+	rxq_ctrl = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq_ctrl),
+			       0, socket_id);
+	if (rxq_ctrl == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue control",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq), 0, socket_id);
+	if (rxq == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue object",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq->rxq_ctrl = rxq_ctrl;
+	rxq_ctrl->type = MLX5_RXQ_TYPE_STANDARD;
+	rxq_ctrl->priv = priv;
+	rxq_ctrl->obj = rxq;
+	rxq_data = &rxq_ctrl->rxq;
+	/* Create CQ using DevX API. */
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue CQ creation failed.",
+			dev->data->port_id);
+		goto error;
+	}
+	/* Create RQ using DevX API. */
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue RQ creation failed.",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	/* Change queue state to ready. */
+	ret = mlx5_devx_modify_rq(rxq, MLX5_RXQ_MOD_RST2RDY);
+	if (ret != 0)
+		goto error;
+	/* Initialize drop queue. */
+	priv->drop_queue.rxq = rxq;
+	return 0;
+error:
+	ret = rte_errno; /* Save rte_errno before cleanup. */
+	if (rxq != NULL) {
+		if (rxq->rq_obj.rq != NULL)
+			mlx5_devx_rq_destroy(&rxq->rq_obj);
+		if (rxq->cq_obj.cq != NULL)
+			mlx5_devx_cq_destroy(&rxq->cq_obj);
+		if (rxq->devx_channel)
+			mlx5_os_devx_destroy_event_channel
+							(rxq->devx_channel);
+		mlx5_free(rxq);
+	}
+	if (rxq_ctrl != NULL)
+		mlx5_free(rxq_ctrl);
+	rte_errno = ret; /* Restore rte_errno. */
 	return -rte_errno;
 }
 
+/**
+ * Release drop Rx queue resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ */
+static void
+mlx5_rxq_devx_obj_drop_release(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_obj *rxq = priv->drop_queue.rxq;
+	struct mlx5_rxq_ctrl *rxq_ctrl = rxq->rxq_ctrl;
+
+	mlx5_rxq_devx_obj_release(rxq);
+	mlx5_free(rxq);
+	mlx5_free(rxq_ctrl);
+	priv->drop_queue.rxq = NULL;
+}
+
 /**
  * Release a drop hash Rx queue.
  *
@@ -883,9 +988,53 @@ mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
 static void
 mlx5_devx_drop_action_destroy(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+
+	if (hrxq->tir != NULL)
+		mlx5_devx_tir_destroy(hrxq);
+	if (hrxq->ind_table->ind_table != NULL)
+		mlx5_devx_ind_table_destroy(hrxq->ind_table);
+	if (priv->drop_queue.rxq->rq != NULL)
+		mlx5_rxq_devx_obj_drop_release(dev);
+}
+
+/**
+ * Create a DevX drop action for Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+	int ret;
+
+	ret = mlx5_rxq_devx_obj_drop_create(dev);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop RX queue");
+		return ret;
+	}
+	/* hrxq->ind_table queues are NULL, drop RX queue ID will be used */
+	ret = mlx5_devx_ind_table_new(dev, 0, hrxq->ind_table);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue indirection table");
+		goto error;
+	}
+	ret = mlx5_devx_hrxq_new(dev, hrxq, /* tunnel */ false);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue");
+		goto error;
+	}
+	return 0;
+error:
+	mlx5_devx_drop_action_destroy(dev);
+	return ret;
 }
 
 /**
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v3 6/6] net/mlx5: preserve indirect actions on restart
       [not found]   ` <20211019123722.3414694-1-dkozlyuk@nvidia.com>
  2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
  2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
@ 2021-10-19 12:37     ` Dmitry Kozlyuk
       [not found]     ` <20211021063503.3632732-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-19 12:37 UTC (permalink / raw)
  To: dev; +Cc: bingz, stable, Matan Azrad, Viacheslav Ovsiienko

MLX5 PMD uses reference counting to manage RX queue resources.
After port stop shared RSS actions kept references to RX queues,
preventing resource release. As a result, internal PMD mempool
for such queues had been exhausted after a number of port restarts.
Diagnostic message from rte_eth_dev_start():

    Rx queue allocation failed: Cannot allocate memory

Dereference RX queues used by indirect actions on port stop (detach)
and restore references on port start (attach) in order to allow RX queue
resource release, but keep indirect RSS across the port restart.
Replace queue IDs in HW by drop queue ID on detach and restore actual
queue IDs on attach.

When the port is stopped, create indirect RSS in the detached state.
As a result, MLX5 PMD is able to keep all its indirect actions
across port restart. Advertise this capability.

Fixes: 4b61b8774be9 ("ethdev: introduce indirect flow action")
Cc: bingz@nvidia.com
Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/mlx5_ethdev.c  |   1 +
 drivers/net/mlx5/mlx5_flow.c    | 194 ++++++++++++++++++++++++++++----
 drivers/net/mlx5/mlx5_flow.h    |   2 +
 drivers/net/mlx5/mlx5_rx.h      |   4 +
 drivers/net/mlx5/mlx5_rxq.c     |  99 ++++++++++++++--
 drivers/net/mlx5/mlx5_trigger.c |  10 ++
 6 files changed, 276 insertions(+), 34 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 82e2284d98..419fec3e4e 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -321,6 +321,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
 				 info->rx_queue_offload_capa);
 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
+	info->dev_capa = RTE_ETH_DEV_CAPA_FLOW_SHARED_OBJECT_KEEP;
 	info->if_index = mlx5_ifindex(dev);
 	info->reta_size = priv->reta_idx_n ?
 		priv->reta_idx_n : config->ind_table_max_size;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index bfc3e20c9a..c10b911259 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1560,6 +1560,58 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 	return 0;
 }
 
+/**
+ * Validate queue numbers for device RSS.
+ *
+ * @param[in] dev
+ *   Configured device.
+ * @param[in] queues
+ *   Array of queue numbers.
+ * @param[in] queues_n
+ *   Size of the @p queues array.
+ * @param[out] error
+ *   On error, filled with a textual error description.
+ * @param[out] queue
+ *   On error, filled with an offending queue index in @p queues array.
+ *
+ * @return
+ *   0 on success, a negative errno code on error.
+ */
+static int
+mlx5_validate_rss_queues(const struct rte_eth_dev *dev,
+			 const uint16_t *queues, uint32_t queues_n,
+			 const char **error, uint32_t *queue_idx)
+{
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
+	uint32_t i;
+
+	for (i = 0; i != queues_n; ++i) {
+		struct mlx5_rxq_ctrl *rxq_ctrl;
+
+		if (queues[i] >= priv->rxqs_n) {
+			*error = "queue index out of range";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		if (!(*priv->rxqs)[queues[i]]) {
+			*error =  "queue is not configured";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		rxq_ctrl = container_of((*priv->rxqs)[queues[i]],
+					struct mlx5_rxq_ctrl, rxq);
+		if (i == 0)
+			rxq_type = rxq_ctrl->type;
+		if (rxq_type != rxq_ctrl->type) {
+			*error = "combining hairpin and regular RSS queues is not supported";
+			*queue_idx = i;
+			return -ENOTSUP;
+		}
+	}
+	return 0;
+}
+
 /*
  * Validate the rss action.
  *
@@ -1580,8 +1632,9 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_rss *rss = action->conf;
-	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
-	unsigned int i;
+	int ret;
+	const char *message;
+	uint32_t queue_idx;
 
 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
@@ -1645,27 +1698,12 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  NULL, "No queues configured");
-	for (i = 0; i != rss->queue_num; ++i) {
-		struct mlx5_rxq_ctrl *rxq_ctrl;
-
-		if (rss->queue[i] >= priv->rxqs_n)
-			return rte_flow_error_set
-				(error, EINVAL,
-				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue index out of range");
-		if (!(*priv->rxqs)[rss->queue[i]])
-			return rte_flow_error_set
-				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue is not configured");
-		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
-					struct mlx5_rxq_ctrl, rxq);
-		if (i == 0)
-			rxq_type = rxq_ctrl->type;
-		if (rxq_type != rxq_ctrl->type)
-			return rte_flow_error_set
-				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i],
-				 "combining hairpin and regular RSS queues is not supported");
+	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
+				       &message, &queue_idx);
+	if (ret != 0) {
+		return rte_flow_error_set(error, -ret,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &rss->queue[queue_idx], message);
 	}
 	return 0;
 }
@@ -8547,6 +8585,116 @@ mlx5_action_handle_flush(struct rte_eth_dev *dev)
 	return ret;
 }
 
+/**
+ * Validate existing indirect actions against current device configuration
+ * and attach them to device resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_attach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+		const char *message;
+		uint32_t queue_idx;
+
+		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
+					       ind_tbl->queues_n,
+					       &message, &queue_idx);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
+				dev->data->port_id, ind_tbl->queues[queue_idx],
+				message);
+			break;
+		}
+	}
+	if (ret != 0)
+		return ret;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not attach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not detach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
+/**
+ * Detach indirect actions of the device from its resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_detach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not detach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not attach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
 #ifndef HAVE_MLX5DV_DR
 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
 #else
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 8f94125f26..6bc7946cc3 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1574,6 +1574,8 @@ void mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
 		struct mlx5_flow_meter_policy *mtr_policy);
 int mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev);
 int mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev);
+int mlx5_action_handle_attach(struct rte_eth_dev *dev);
+int mlx5_action_handle_detach(struct rte_eth_dev *dev);
 int mlx5_action_handle_flush(struct rte_eth_dev *dev);
 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id);
 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh);
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 2b7ad3e48b..d44c8078de 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -222,6 +222,10 @@ int mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 			      struct mlx5_ind_table_obj *ind_tbl,
 			      uint16_t *queues, const uint32_t queues_n,
 			      bool standalone);
+int mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
+int mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
 struct mlx5_list_entry *mlx5_hrxq_create_cb(void *tool_ctx, void *cb_ctx);
 int mlx5_hrxq_match_cb(void *tool_ctx, struct mlx5_list_entry *entry,
 		       void *cb_ctx);
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index b68443bed5..fd2b5779ff 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -2015,6 +2015,26 @@ mlx5_ind_table_obj_new(struct rte_eth_dev *dev, const uint16_t *queues,
 	return ind_tbl;
 }
 
+static int
+mlx5_ind_table_obj_check_standalone(struct rte_eth_dev *dev __rte_unused,
+				    struct mlx5_ind_table_obj *ind_tbl)
+{
+	uint32_t refcnt;
+
+	refcnt = __atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED);
+	if (refcnt <= 1)
+		return 0;
+	/*
+	 * Modification of indirection tables having more than 1
+	 * reference is unsupported.
+	 */
+	DRV_LOG(DEBUG,
+		"Port %u cannot modify indirection table %p (refcnt %u > 1).",
+		dev->data->port_id, (void *)ind_tbl, refcnt);
+	rte_errno = EINVAL;
+	return -rte_errno;
+}
+
 /**
  * Modify an indirection table.
  *
@@ -2047,18 +2067,8 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 
 	MLX5_ASSERT(standalone);
 	RTE_SET_USED(standalone);
-	if (__atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED) > 1) {
-		/*
-		 * Modification of indirection ntables having more than 1
-		 * reference unsupported. Intended for standalone indirection
-		 * tables only.
-		 */
-		DRV_LOG(DEBUG,
-			"Port %u cannot modify indirection table (refcnt> 1).",
-			dev->data->port_id);
-		rte_errno = EINVAL;
+	if (mlx5_ind_table_obj_check_standalone(dev, ind_tbl) < 0)
 		return -rte_errno;
-	}
 	for (i = 0; i != queues_n; ++i) {
 		if (!mlx5_rxq_get(dev, queues[i])) {
 			ret = -rte_errno;
@@ -2084,6 +2094,73 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 	return ret;
 }
 
+/**
+ * Attach an indirection table to its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to attach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_modify(dev, ind_tbl, ind_tbl->queues,
+					ind_tbl->queues_n, true);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_get(dev, ind_tbl->queues[i]);
+	return 0;
+}
+
+/**
+ * Detach an indirection table from its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to detach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	const unsigned int n = rte_is_power_of_2(ind_tbl->queues_n) ?
+			       log2above(ind_tbl->queues_n) :
+			       log2above(priv->config.ind_table_max_size);
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_check_standalone(dev, ind_tbl);
+	if (ret != 0)
+		return ret;
+	MLX5_ASSERT(priv->obj_ops.ind_table_modify);
+	ret = priv->obj_ops.ind_table_modify(dev, n, NULL, 0, ind_tbl);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_release(dev, ind_tbl->queues[i]);
+	return ret;
+}
+
 int
 mlx5_hrxq_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
 		   void *cb_ctx)
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 54173bfacb..c3adf5082e 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -14,6 +14,7 @@
 #include <mlx5_malloc.h>
 
 #include "mlx5.h"
+#include "mlx5_flow.h"
 #include "mlx5_mr.h"
 #include "mlx5_rx.h"
 #include "mlx5_tx.h"
@@ -1113,6 +1114,14 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 	mlx5_rxq_timestamp_set(dev);
 	/* Set a mask and offset of scheduling on timestamp into Tx queues. */
 	mlx5_txq_dynf_timestamp_set(dev);
+	/* Attach indirection table objects detached on port stop. */
+	ret = mlx5_action_handle_attach(dev);
+	if (ret) {
+		DRV_LOG(ERR,
+			"port %u failed to attach indirect actions: %s",
+			dev->data->port_id, rte_strerror(rte_errno));
+		goto error;
+	}
 	/*
 	 * In non-cached mode, it only needs to start the default mreg copy
 	 * action and no flow created by application exists anymore.
@@ -1185,6 +1194,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	/* All RX queue flags will be cleared in the flush interface. */
 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, true);
 	mlx5_flow_meter_rxq_flush(dev);
+	mlx5_action_handle_detach(dev);
 	mlx5_rx_intr_vec_disable(dev);
 	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
 	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v4 4/6] net/mlx5: discover max flow priority using DevX
       [not found]     ` <20211021063503.3632732-1-dkozlyuk@nvidia.com>
@ 2021-10-21  6:35       ` Dmitry Kozlyuk
  2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
                         ` (2 subsequent siblings)
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-21  6:35 UTC (permalink / raw)
  To: dev; +Cc: Ori Kam, stable, Matan Azrad, Viacheslav Ovsiienko

Maximum available flow priority was discovered using Verbs API
regardless of the selected flow engine. This required some Verbs
objects to be initialized in order to use DevX engine. Make priority
discovery an engine method and implement it for DevX using its API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c   |   1 -
 drivers/net/mlx5/mlx5_flow.c       |  98 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow.h       |   4 ++
 drivers/net/mlx5/mlx5_flow_dv.c    | 103 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  77 +++------------------
 5 files changed, 215 insertions(+), 68 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 26a8d75b99..60d2c398db 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1849,7 +1849,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	priv->drop_queue.hrxq = mlx5_drop_action_create(eth_dev);
 	if (!priv->drop_queue.hrxq)
 		goto error;
-	/* Supported Verbs flow priority number detection. */
 	err = mlx5_flow_discover_priorities(eth_dev);
 	if (err < 0) {
 		err = -err;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index b4d0b7b5ef..2768244c2e 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -9480,3 +9480,101 @@ mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
 		return node;
 	}
 }
+
+/* Map of Verbs to Flow priority with 8 Verbs priorities. */
+static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
+};
+
+/* Map of Verbs to Flow priority with 16 Verbs priorities. */
+static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
+	{ 9, 10, 11 }, { 12, 13, 14 },
+};
+
+/**
+ * Discover the number of available flow priorities.
+ *
+ * @param dev
+ *   Ethernet device.
+ *
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+int
+mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+{
+	static const uint16_t vprio[] = {8, 16};
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	const struct mlx5_flow_driver_ops *fops;
+	enum mlx5_flow_drv_type type;
+	int ret;
+
+	type = mlx5_flow_os_get_type();
+	if (type == MLX5_FLOW_TYPE_MAX) {
+		type = MLX5_FLOW_TYPE_VERBS;
+		if (priv->config.devx && priv->config.dv_flow_en)
+			type = MLX5_FLOW_TYPE_DV;
+	}
+	fops = flow_get_drv_ops(type);
+	if (fops->discover_priorities == NULL) {
+		DRV_LOG(ERR, "Priority discovery not supported");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
+	if (ret < 0)
+		return ret;
+	switch (ret) {
+	case 8:
+		ret = RTE_DIM(priority_map_3);
+		break;
+	case 16:
+		ret = RTE_DIM(priority_map_5);
+		break;
+	default:
+		rte_errno = ENOTSUP;
+		DRV_LOG(ERR,
+			"port %u maximum priority: %d expected 8/16",
+			dev->data->port_id, ret);
+		return -rte_errno;
+	}
+	DRV_LOG(INFO, "port %u supported flow priorities:"
+		" 0-%d for ingress or egress root table,"
+		" 0-%d for non-root table or transfer root table.",
+		dev->data->port_id, ret - 2,
+		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
+	return ret;
+}
+
+/**
+ * Adjust flow priority based on the highest layer and the request priority.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] priority
+ *   The rule base priority.
+ * @param[in] subpriority
+ *   The priority based on the items.
+ *
+ * @return
+ *   The new priority.
+ */
+uint32_t
+mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
+			  uint32_t subpriority)
+{
+	uint32_t res = 0;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	switch (priv->config.flow_prio) {
+	case RTE_DIM(priority_map_3):
+		res = priority_map_3[priority][subpriority];
+		break;
+	case RTE_DIM(priority_map_5):
+		res = priority_map_5[priority][subpriority];
+		break;
+	}
+	return  res;
+}
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 5c68d4f7d7..8f94125f26 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1226,6 +1226,9 @@ typedef int (*mlx5_flow_create_def_policy_t)
 			(struct rte_eth_dev *dev);
 typedef void (*mlx5_flow_destroy_def_policy_t)
 			(struct rte_eth_dev *dev);
+typedef int (*mlx5_flow_discover_priorities_t)
+			(struct rte_eth_dev *dev,
+			 const uint16_t *vprio, int vprio_n);
 
 struct mlx5_flow_driver_ops {
 	mlx5_flow_validate_t validate;
@@ -1260,6 +1263,7 @@ struct mlx5_flow_driver_ops {
 	mlx5_flow_action_update_t action_update;
 	mlx5_flow_action_query_t action_query;
 	mlx5_flow_sync_domain_t sync_domain;
+	mlx5_flow_discover_priorities_t discover_priorities;
 };
 
 /* mlx5_flow.c */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index e31d4d8468..5163f518d7 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -17883,6 +17883,108 @@ flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
 	return 0;
 }
 
+/**
+ * Discover the number of available flow priorities
+ * by trying to create a flow with the highest priority value
+ * for each possible number.
+ *
+ * @param[in] dev
+ *   Ethernet device.
+ * @param[in] vprio
+ *   List of possible number of available priorities.
+ * @param[in] vprio_n
+ *   Size of @p vprio array.
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+static int
+flow_dv_discover_priorities(struct rte_eth_dev *dev,
+			    const uint16_t *vprio, int vprio_n)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+	struct mlx5_flow_dv_matcher matcher = {
+		.mask = {
+			.size = sizeof(matcher.mask.buf),
+		},
+	};
+	union mlx5_flow_tbl_key tbl_key;
+	struct mlx5_flow flow;
+	void *action;
+	struct rte_flow_error error;
+	uint8_t misc_mask;
+	int i, err, ret = -ENOTSUP;
+
+	/*
+	 * Prepare a flow with a catch-all pattern and a drop action.
+	 * Use drop queue, because shared drop action may be unavailable.
+	 */
+	action = priv->drop_queue.hrxq->action;
+	if (action == NULL) {
+		DRV_LOG(ERR, "Priority discovery requires a drop action");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	memset(&flow, 0, sizeof(flow));
+	flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
+	if (flow.handle == NULL) {
+		DRV_LOG(ERR, "Cannot create flow handle");
+		rte_errno = ENOMEM;
+		return -rte_errno;
+	}
+	flow.ingress = true;
+	flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
+	flow.dv.actions[0] = action;
+	flow.dv.actions_n = 1;
+	memset(&eth, 0, sizeof(eth));
+	flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
+				   &item, /* inner */ false, /* group */ 0);
+	matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
+	for (i = 0; i < vprio_n; i++) {
+		/* Configure the next proposed maximum priority. */
+		matcher.priority = vprio[i] - 1;
+		memset(&tbl_key, 0, sizeof(tbl_key));
+		err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
+					       /* tunnel */ NULL,
+					       /* group */ 0,
+					       &error);
+		if (err != 0) {
+			/* This action is pure SW and must always succeed. */
+			DRV_LOG(ERR, "Cannot register matcher");
+			ret = -rte_errno;
+			break;
+		}
+		/* Try to apply the flow to HW. */
+		misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
+		__flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
+		err = mlx5_flow_os_create_flow
+				(flow.handle->dvh.matcher->matcher_object,
+				 (void *)&flow.dv.value, flow.dv.actions_n,
+				 flow.dv.actions, &flow.handle->drv_flow);
+		if (err == 0) {
+			claim_zero(mlx5_flow_os_destroy_flow
+						(flow.handle->drv_flow));
+			flow.handle->drv_flow = NULL;
+		}
+		claim_zero(flow_dv_matcher_release(dev, flow.handle));
+		if (err != 0)
+			break;
+		ret = vprio[i];
+	}
+	mlx5_ipool_free(pool, flow.handle_idx);
+	/* Set rte_errno if no expected priority value matched. */
+	if (ret < 0)
+		rte_errno = -ret;
+	return ret;
+}
+
 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.validate = flow_dv_validate,
 	.prepare = flow_dv_prepare,
@@ -17916,6 +18018,7 @@ const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.action_update = flow_dv_action_update,
 	.action_query = flow_dv_action_query,
 	.sync_domain = flow_dv_sync_domain,
+	.discover_priorities = flow_dv_discover_priorities,
 };
 
 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
diff --git a/drivers/net/mlx5/mlx5_flow_verbs.c b/drivers/net/mlx5/mlx5_flow_verbs.c
index 1627c3905f..c8cf7ef29c 100644
--- a/drivers/net/mlx5/mlx5_flow_verbs.c
+++ b/drivers/net/mlx5/mlx5_flow_verbs.c
@@ -28,17 +28,6 @@
 #define VERBS_SPEC_INNER(item_flags) \
 	(!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
 
-/* Map of Verbs to Flow priority with 8 Verbs priorities. */
-static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
-};
-
-/* Map of Verbs to Flow priority with 16 Verbs priorities. */
-static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
-	{ 9, 10, 11 }, { 12, 13, 14 },
-};
-
 /* Verbs specification header. */
 struct ibv_spec_header {
 	enum ibv_flow_spec_type type;
@@ -50,13 +39,17 @@ struct ibv_spec_header {
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
- *
+ * @param[in] vprio
+ *   Expected result variants.
+ * @param[in] vprio_n
+ *   Number of entries in @p vprio array.
  * @return
- *   number of supported flow priority on success, a negative errno
+ *   Number of supported flow priority on success, a negative errno
  *   value otherwise and rte_errno is set.
  */
-int
-mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+static int
+flow_verbs_discover_priorities(struct rte_eth_dev *dev,
+			       const uint16_t *vprio, int vprio_n)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct {
@@ -79,7 +72,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	};
 	struct ibv_flow *flow;
 	struct mlx5_hrxq *drop = priv->drop_queue.hrxq;
-	uint16_t vprio[] = { 8, 16 };
 	int i;
 	int priority = 0;
 
@@ -87,7 +79,7 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		rte_errno = ENOTSUP;
 		return -rte_errno;
 	}
-	for (i = 0; i != RTE_DIM(vprio); i++) {
+	for (i = 0; i != vprio_n; i++) {
 		flow_attr.attr.priority = vprio[i] - 1;
 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 		if (!flow)
@@ -95,59 +87,9 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		claim_zero(mlx5_glue->destroy_flow(flow));
 		priority = vprio[i];
 	}
-	switch (priority) {
-	case 8:
-		priority = RTE_DIM(priority_map_3);
-		break;
-	case 16:
-		priority = RTE_DIM(priority_map_5);
-		break;
-	default:
-		rte_errno = ENOTSUP;
-		DRV_LOG(ERR,
-			"port %u verbs maximum priority: %d expected 8/16",
-			dev->data->port_id, priority);
-		return -rte_errno;
-	}
-	DRV_LOG(INFO, "port %u supported flow priorities:"
-		" 0-%d for ingress or egress root table,"
-		" 0-%d for non-root table or transfer root table.",
-		dev->data->port_id, priority - 2,
-		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
 	return priority;
 }
 
-/**
- * Adjust flow priority based on the highest layer and the request priority.
- *
- * @param[in] dev
- *   Pointer to the Ethernet device structure.
- * @param[in] priority
- *   The rule base priority.
- * @param[in] subpriority
- *   The priority based on the items.
- *
- * @return
- *   The new priority.
- */
-uint32_t
-mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
-				   uint32_t subpriority)
-{
-	uint32_t res = 0;
-	struct mlx5_priv *priv = dev->data->dev_private;
-
-	switch (priv->config.flow_prio) {
-	case RTE_DIM(priority_map_3):
-		res = priority_map_3[priority][subpriority];
-		break;
-	case RTE_DIM(priority_map_5):
-		res = priority_map_5[priority][subpriority];
-		break;
-	}
-	return  res;
-}
-
 /**
  * Get Verbs flow counter by index.
  *
@@ -2087,4 +2029,5 @@ const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
 	.destroy = flow_verbs_destroy,
 	.query = flow_verbs_query,
 	.sync_domain = flow_verbs_sync_domain,
+	.discover_priorities = flow_verbs_discover_priorities,
 };
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v4 5/6] net/mlx5: create drop queue using DevX
       [not found]     ` <20211021063503.3632732-1-dkozlyuk@nvidia.com>
  2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
@ 2021-10-21  6:35       ` Dmitry Kozlyuk
  2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
       [not found]       ` <20211102135415.944050-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-21  6:35 UTC (permalink / raw)
  To: dev; +Cc: Ori Kam, stable, Matan Azrad, Viacheslav Ovsiienko

Drop queue creation and destruction were not implemented for DevX
flow engine and Verbs engine methods were used as a workaround.
Implement these methods for DevX so that there is a valid queue ID
that can be used regardless of queue configuration via API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c |   4 -
 drivers/net/mlx5/mlx5_devx.c     | 211 ++++++++++++++++++++++++++-----
 2 files changed, 180 insertions(+), 35 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 60d2c398db..a8d4b9ae88 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1809,10 +1809,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	}
 	if (config->devx && config->dv_flow_en && config->dest_tir) {
 		priv->obj_ops = devx_obj_ops;
-		priv->obj_ops.drop_action_create =
-						ibv_obj_ops.drop_action_create;
-		priv->obj_ops.drop_action_destroy =
-						ibv_obj_ops.drop_action_destroy;
 #ifndef HAVE_MLX5DV_DEVX_UAR_OFFSET
 		priv->obj_ops.txq_obj_modify = ibv_obj_ops.txq_obj_modify;
 #else
diff --git a/drivers/net/mlx5/mlx5_devx.c b/drivers/net/mlx5/mlx5_devx.c
index a1db53577a..1e62108c94 100644
--- a/drivers/net/mlx5/mlx5_devx.c
+++ b/drivers/net/mlx5/mlx5_devx.c
@@ -226,17 +226,17 @@ mlx5_rx_devx_get_event(struct mlx5_rxq_obj *rxq_obj)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	struct mlx5_devx_create_rq_attr rq_attr = { 0 };
@@ -289,20 +289,20 @@ mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_devx_cq *cq_obj = 0;
 	struct mlx5_devx_cq_attr cq_attr = { 0 };
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_dev_ctx_shared *sh = priv->sh;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	unsigned int cqe_n = mlx5_rxq_cqe_num(rxq_data);
@@ -497,13 +497,13 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
 		tmpl->fd = mlx5_os_get_devx_channel_fd(tmpl->devx_channel);
 	}
 	/* Create CQ using DevX API. */
-	ret = mlx5_rxq_create_devx_cq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Failed to create CQ.");
 		goto error;
 	}
 	/* Create RQ using DevX API. */
-	ret = mlx5_rxq_create_devx_rq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Port %u Rx queue %u RQ creation failure.",
 			dev->data->port_id, idx);
@@ -536,6 +536,11 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
  *   Pointer to Ethernet device.
  * @param log_n
  *   Log of number of queues in the array.
+ * @param queues
+ *   List of RX queue indices or NULL, in which case
+ *   the attribute will be filled by drop queue ID.
+ * @param queues_n
+ *   Size of @p queues array or 0 if it is NULL.
  * @param ind_tbl
  *   DevX indirection table object.
  *
@@ -563,6 +568,11 @@ mlx5_devx_ind_table_create_rqt_attr(struct rte_eth_dev *dev,
 	}
 	rqt_attr->rqt_max_size = priv->config.ind_table_max_size;
 	rqt_attr->rqt_actual_size = rqt_n;
+	if (queues == NULL) {
+		for (i = 0; i < rqt_n; i++)
+			rqt_attr->rq_list[i] = priv->drop_queue.rxq->rq->id;
+		return rqt_attr;
+	}
 	for (i = 0; i != queues_n; ++i) {
 		struct mlx5_rxq_data *rxq = (*priv->rxqs)[queues[i]];
 		struct mlx5_rxq_ctrl *rxq_ctrl =
@@ -595,11 +605,12 @@ mlx5_devx_ind_table_new(struct rte_eth_dev *dev, const unsigned int log_n,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_devx_rqt_attr *rqt_attr = NULL;
+	const uint16_t *queues = dev->data->dev_started ? ind_tbl->queues :
+							  NULL;
 
 	MLX5_ASSERT(ind_tbl);
-	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n,
-							ind_tbl->queues,
-							ind_tbl->queues_n);
+	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n, queues,
+						       ind_tbl->queues_n);
 	if (!rqt_attr)
 		return -rte_errno;
 	ind_tbl->rqt = mlx5_devx_cmd_create_rqt(priv->sh->ctx, rqt_attr);
@@ -670,7 +681,8 @@ mlx5_devx_ind_table_destroy(struct mlx5_ind_table_obj *ind_tbl)
  * @param[in] hash_fields
  *   Verbs protocol hash field to make the RSS on.
  * @param[in] ind_tbl
- *   Indirection table for TIR.
+ *   Indirection table for TIR. If table queues array is NULL,
+ *   a TIR for drop queue is assumed.
  * @param[in] tunnel
  *   Tunnel type.
  * @param[out] tir_attr
@@ -686,19 +698,27 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 		       int tunnel, struct mlx5_devx_tir_attr *tir_attr)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[ind_tbl->queues[0]];
-	struct mlx5_rxq_ctrl *rxq_ctrl =
-		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
-	enum mlx5_rxq_type rxq_obj_type = rxq_ctrl->type;
+	enum mlx5_rxq_type rxq_obj_type;
 	bool lro = true;
 	uint32_t i;
 
-	/* Enable TIR LRO only if all the queues were configured for. */
-	for (i = 0; i < ind_tbl->queues_n; ++i) {
-		if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
-			lro = false;
-			break;
+	/* NULL queues designate drop queue. */
+	if (ind_tbl->queues != NULL) {
+		struct mlx5_rxq_data *rxq_data =
+					(*priv->rxqs)[ind_tbl->queues[0]];
+		struct mlx5_rxq_ctrl *rxq_ctrl =
+			container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+		rxq_obj_type = rxq_ctrl->type;
+
+		/* Enable TIR LRO only if all the queues were configured for. */
+		for (i = 0; i < ind_tbl->queues_n; ++i) {
+			if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
+				lro = false;
+				break;
+			}
 		}
+	} else {
+		rxq_obj_type = priv->drop_queue.rxq->rxq_ctrl->type;
 	}
 	memset(tir_attr, 0, sizeof(*tir_attr));
 	tir_attr->disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
@@ -857,7 +877,7 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
 }
 
 /**
- * Create a DevX drop action for Rx Hash queue.
+ * Create a DevX drop Rx queue.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -866,14 +886,99 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+mlx5_rxq_devx_obj_drop_create(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	int socket_id = dev->device->numa_node;
+	struct mlx5_rxq_ctrl *rxq_ctrl;
+	struct mlx5_rxq_data *rxq_data;
+	struct mlx5_rxq_obj *rxq = NULL;
+	int ret;
+
+	/*
+	 * Initialize dummy control structures.
+	 * They are required to hold pointers for cleanup
+	 * and are only accessible via drop queue DevX objects.
+	 */
+	rxq_ctrl = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq_ctrl),
+			       0, socket_id);
+	if (rxq_ctrl == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue control",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq), 0, socket_id);
+	if (rxq == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue object",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq->rxq_ctrl = rxq_ctrl;
+	rxq_ctrl->type = MLX5_RXQ_TYPE_STANDARD;
+	rxq_ctrl->priv = priv;
+	rxq_ctrl->obj = rxq;
+	rxq_data = &rxq_ctrl->rxq;
+	/* Create CQ using DevX API. */
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue CQ creation failed.",
+			dev->data->port_id);
+		goto error;
+	}
+	/* Create RQ using DevX API. */
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue RQ creation failed.",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	/* Change queue state to ready. */
+	ret = mlx5_devx_modify_rq(rxq, MLX5_RXQ_MOD_RST2RDY);
+	if (ret != 0)
+		goto error;
+	/* Initialize drop queue. */
+	priv->drop_queue.rxq = rxq;
+	return 0;
+error:
+	ret = rte_errno; /* Save rte_errno before cleanup. */
+	if (rxq != NULL) {
+		if (rxq->rq_obj.rq != NULL)
+			mlx5_devx_rq_destroy(&rxq->rq_obj);
+		if (rxq->cq_obj.cq != NULL)
+			mlx5_devx_cq_destroy(&rxq->cq_obj);
+		if (rxq->devx_channel)
+			mlx5_os_devx_destroy_event_channel
+							(rxq->devx_channel);
+		mlx5_free(rxq);
+	}
+	if (rxq_ctrl != NULL)
+		mlx5_free(rxq_ctrl);
+	rte_errno = ret; /* Restore rte_errno. */
 	return -rte_errno;
 }
 
+/**
+ * Release drop Rx queue resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ */
+static void
+mlx5_rxq_devx_obj_drop_release(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_obj *rxq = priv->drop_queue.rxq;
+	struct mlx5_rxq_ctrl *rxq_ctrl = rxq->rxq_ctrl;
+
+	mlx5_rxq_devx_obj_release(rxq);
+	mlx5_free(rxq);
+	mlx5_free(rxq_ctrl);
+	priv->drop_queue.rxq = NULL;
+}
+
 /**
  * Release a drop hash Rx queue.
  *
@@ -883,9 +988,53 @@ mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
 static void
 mlx5_devx_drop_action_destroy(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+
+	if (hrxq->tir != NULL)
+		mlx5_devx_tir_destroy(hrxq);
+	if (hrxq->ind_table->ind_table != NULL)
+		mlx5_devx_ind_table_destroy(hrxq->ind_table);
+	if (priv->drop_queue.rxq->rq != NULL)
+		mlx5_rxq_devx_obj_drop_release(dev);
+}
+
+/**
+ * Create a DevX drop action for Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+	int ret;
+
+	ret = mlx5_rxq_devx_obj_drop_create(dev);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop RX queue");
+		return ret;
+	}
+	/* hrxq->ind_table queues are NULL, drop RX queue ID will be used */
+	ret = mlx5_devx_ind_table_new(dev, 0, hrxq->ind_table);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue indirection table");
+		goto error;
+	}
+	ret = mlx5_devx_hrxq_new(dev, hrxq, /* tunnel */ false);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue");
+		goto error;
+	}
+	return 0;
+error:
+	mlx5_devx_drop_action_destroy(dev);
+	return ret;
 }
 
 /**
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v4 6/6] net/mlx5: preserve indirect actions on restart
       [not found]     ` <20211021063503.3632732-1-dkozlyuk@nvidia.com>
  2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
  2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
@ 2021-10-21  6:35       ` Dmitry Kozlyuk
       [not found]       ` <20211102135415.944050-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-10-21  6:35 UTC (permalink / raw)
  To: dev; +Cc: Ori Kam, bingz, stable, Matan Azrad, Viacheslav Ovsiienko

MLX5 PMD uses reference counting to manage RX queue resources.
After port stop shared RSS actions kept references to RX queues,
preventing resource release. As a result, internal PMD mempool
for such queues had been exhausted after a number of port restarts.
Diagnostic message from rte_eth_dev_start():

    Rx queue allocation failed: Cannot allocate memory

Dereference RX queues used by indirect actions on port stop (detach)
and restore references on port start (attach) in order to allow RX queue
resource release, but keep indirect RSS across the port restart.
Replace queue IDs in HW by drop queue ID on detach and restore actual
queue IDs on attach.

When the port is stopped, create indirect RSS in the detached state.
As a result, MLX5 PMD is able to keep all its indirect actions
across port restart. Advertise this capability.

Fixes: 4b61b8774be9 ("ethdev: introduce indirect flow action")
Cc: bingz@nvidia.com
Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/mlx5_ethdev.c  |   1 +
 drivers/net/mlx5/mlx5_flow.c    | 194 ++++++++++++++++++++++++++++----
 drivers/net/mlx5/mlx5_flow.h    |   2 +
 drivers/net/mlx5/mlx5_rx.h      |   4 +
 drivers/net/mlx5/mlx5_rxq.c     |  99 ++++++++++++++--
 drivers/net/mlx5/mlx5_trigger.c |  10 ++
 6 files changed, 276 insertions(+), 34 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 82e2284d98..419fec3e4e 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -321,6 +321,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
 				 info->rx_queue_offload_capa);
 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
+	info->dev_capa = RTE_ETH_DEV_CAPA_FLOW_SHARED_OBJECT_KEEP;
 	info->if_index = mlx5_ifindex(dev);
 	info->reta_size = priv->reta_idx_n ?
 		priv->reta_idx_n : config->ind_table_max_size;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 2768244c2e..df1e927534 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1583,6 +1583,58 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 	return 0;
 }
 
+/**
+ * Validate queue numbers for device RSS.
+ *
+ * @param[in] dev
+ *   Configured device.
+ * @param[in] queues
+ *   Array of queue numbers.
+ * @param[in] queues_n
+ *   Size of the @p queues array.
+ * @param[out] error
+ *   On error, filled with a textual error description.
+ * @param[out] queue
+ *   On error, filled with an offending queue index in @p queues array.
+ *
+ * @return
+ *   0 on success, a negative errno code on error.
+ */
+static int
+mlx5_validate_rss_queues(const struct rte_eth_dev *dev,
+			 const uint16_t *queues, uint32_t queues_n,
+			 const char **error, uint32_t *queue_idx)
+{
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
+	uint32_t i;
+
+	for (i = 0; i != queues_n; ++i) {
+		struct mlx5_rxq_ctrl *rxq_ctrl;
+
+		if (queues[i] >= priv->rxqs_n) {
+			*error = "queue index out of range";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		if (!(*priv->rxqs)[queues[i]]) {
+			*error =  "queue is not configured";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		rxq_ctrl = container_of((*priv->rxqs)[queues[i]],
+					struct mlx5_rxq_ctrl, rxq);
+		if (i == 0)
+			rxq_type = rxq_ctrl->type;
+		if (rxq_type != rxq_ctrl->type) {
+			*error = "combining hairpin and regular RSS queues is not supported";
+			*queue_idx = i;
+			return -ENOTSUP;
+		}
+	}
+	return 0;
+}
+
 /*
  * Validate the rss action.
  *
@@ -1603,8 +1655,9 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_rss *rss = action->conf;
-	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
-	unsigned int i;
+	int ret;
+	const char *message;
+	uint32_t queue_idx;
 
 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
@@ -1668,27 +1721,12 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  NULL, "No queues configured");
-	for (i = 0; i != rss->queue_num; ++i) {
-		struct mlx5_rxq_ctrl *rxq_ctrl;
-
-		if (rss->queue[i] >= priv->rxqs_n)
-			return rte_flow_error_set
-				(error, EINVAL,
-				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue index out of range");
-		if (!(*priv->rxqs)[rss->queue[i]])
-			return rte_flow_error_set
-				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue is not configured");
-		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
-					struct mlx5_rxq_ctrl, rxq);
-		if (i == 0)
-			rxq_type = rxq_ctrl->type;
-		if (rxq_type != rxq_ctrl->type)
-			return rte_flow_error_set
-				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i],
-				 "combining hairpin and regular RSS queues is not supported");
+	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
+				       &message, &queue_idx);
+	if (ret != 0) {
+		return rte_flow_error_set(error, -ret,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &rss->queue[queue_idx], message);
 	}
 	return 0;
 }
@@ -8570,6 +8608,116 @@ mlx5_action_handle_flush(struct rte_eth_dev *dev)
 	return ret;
 }
 
+/**
+ * Validate existing indirect actions against current device configuration
+ * and attach them to device resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_attach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+		const char *message;
+		uint32_t queue_idx;
+
+		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
+					       ind_tbl->queues_n,
+					       &message, &queue_idx);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
+				dev->data->port_id, ind_tbl->queues[queue_idx],
+				message);
+			break;
+		}
+	}
+	if (ret != 0)
+		return ret;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not attach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not detach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
+/**
+ * Detach indirect actions of the device from its resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_detach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not detach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not attach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
 #ifndef HAVE_MLX5DV_DR
 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
 #else
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 8f94125f26..6bc7946cc3 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1574,6 +1574,8 @@ void mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
 		struct mlx5_flow_meter_policy *mtr_policy);
 int mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev);
 int mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev);
+int mlx5_action_handle_attach(struct rte_eth_dev *dev);
+int mlx5_action_handle_detach(struct rte_eth_dev *dev);
 int mlx5_action_handle_flush(struct rte_eth_dev *dev);
 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id);
 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh);
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index a90cb497d1..6d010059f1 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -222,6 +222,10 @@ int mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 			      struct mlx5_ind_table_obj *ind_tbl,
 			      uint16_t *queues, const uint32_t queues_n,
 			      bool standalone);
+int mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
+int mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
 struct mlx5_list_entry *mlx5_hrxq_create_cb(void *tool_ctx, void *cb_ctx);
 int mlx5_hrxq_match_cb(void *tool_ctx, struct mlx5_list_entry *entry,
 		       void *cb_ctx);
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 60673d014d..47124f6e81 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -2028,6 +2028,26 @@ mlx5_ind_table_obj_new(struct rte_eth_dev *dev, const uint16_t *queues,
 	return ind_tbl;
 }
 
+static int
+mlx5_ind_table_obj_check_standalone(struct rte_eth_dev *dev __rte_unused,
+				    struct mlx5_ind_table_obj *ind_tbl)
+{
+	uint32_t refcnt;
+
+	refcnt = __atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED);
+	if (refcnt <= 1)
+		return 0;
+	/*
+	 * Modification of indirection tables having more than 1
+	 * reference is unsupported.
+	 */
+	DRV_LOG(DEBUG,
+		"Port %u cannot modify indirection table %p (refcnt %u > 1).",
+		dev->data->port_id, (void *)ind_tbl, refcnt);
+	rte_errno = EINVAL;
+	return -rte_errno;
+}
+
 /**
  * Modify an indirection table.
  *
@@ -2060,18 +2080,8 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 
 	MLX5_ASSERT(standalone);
 	RTE_SET_USED(standalone);
-	if (__atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED) > 1) {
-		/*
-		 * Modification of indirection ntables having more than 1
-		 * reference unsupported. Intended for standalone indirection
-		 * tables only.
-		 */
-		DRV_LOG(DEBUG,
-			"Port %u cannot modify indirection table (refcnt> 1).",
-			dev->data->port_id);
-		rte_errno = EINVAL;
+	if (mlx5_ind_table_obj_check_standalone(dev, ind_tbl) < 0)
 		return -rte_errno;
-	}
 	for (i = 0; i != queues_n; ++i) {
 		if (!mlx5_rxq_get(dev, queues[i])) {
 			ret = -rte_errno;
@@ -2097,6 +2107,73 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 	return ret;
 }
 
+/**
+ * Attach an indirection table to its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to attach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_modify(dev, ind_tbl, ind_tbl->queues,
+					ind_tbl->queues_n, true);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_get(dev, ind_tbl->queues[i]);
+	return 0;
+}
+
+/**
+ * Detach an indirection table from its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to detach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	const unsigned int n = rte_is_power_of_2(ind_tbl->queues_n) ?
+			       log2above(ind_tbl->queues_n) :
+			       log2above(priv->config.ind_table_max_size);
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_check_standalone(dev, ind_tbl);
+	if (ret != 0)
+		return ret;
+	MLX5_ASSERT(priv->obj_ops.ind_table_modify);
+	ret = priv->obj_ops.ind_table_modify(dev, n, NULL, 0, ind_tbl);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_release(dev, ind_tbl->queues[i]);
+	return ret;
+}
+
 int
 mlx5_hrxq_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
 		   void *cb_ctx)
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 3cbf5816a1..6295c6b3e9 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -14,6 +14,7 @@
 #include <mlx5_malloc.h>
 
 #include "mlx5.h"
+#include "mlx5_flow.h"
 #include "mlx5_mr.h"
 #include "mlx5_rx.h"
 #include "mlx5_tx.h"
@@ -1161,6 +1162,14 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 	mlx5_rxq_timestamp_set(dev);
 	/* Set a mask and offset of scheduling on timestamp into Tx queues. */
 	mlx5_txq_dynf_timestamp_set(dev);
+	/* Attach indirection table objects detached on port stop. */
+	ret = mlx5_action_handle_attach(dev);
+	if (ret) {
+		DRV_LOG(ERR,
+			"port %u failed to attach indirect actions: %s",
+			dev->data->port_id, rte_strerror(rte_errno));
+		goto error;
+	}
 	/*
 	 * In non-cached mode, it only needs to start the default mreg copy
 	 * action and no flow created by application exists anymore.
@@ -1238,6 +1247,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	/* All RX queue flags will be cleared in the flush interface. */
 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, true);
 	mlx5_flow_meter_rxq_flush(dev);
+	mlx5_action_handle_detach(dev);
 	mlx5_rx_intr_vec_disable(dev);
 	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
 	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v5 4/6] net/mlx5: discover max flow priority using DevX
       [not found]       ` <20211102135415.944050-1-dkozlyuk@nvidia.com>
@ 2021-11-02 13:54         ` Dmitry Kozlyuk
  2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
                           ` (2 subsequent siblings)
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-11-02 13:54 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Maximum available flow priority was discovered using Verbs API
regardless of the selected flow engine. This required some Verbs
objects to be initialized in order to use DevX engine. Make priority
discovery an engine method and implement it for DevX using its API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c   |   1 -
 drivers/net/mlx5/mlx5_flow.c       |  98 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow.h       |   4 ++
 drivers/net/mlx5/mlx5_flow_dv.c    | 103 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  74 +++------------------
 5 files changed, 216 insertions(+), 64 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 72bbb665cf..34546635c4 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1720,7 +1720,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	priv->drop_queue.hrxq = mlx5_drop_action_create(eth_dev);
 	if (!priv->drop_queue.hrxq)
 		goto error;
-	/* Supported Verbs flow priority number detection. */
 	err = mlx5_flow_discover_priorities(eth_dev);
 	if (err < 0) {
 		err = -err;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 5d19ef1e82..850eb353fd 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -9570,3 +9570,101 @@ mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
 		return node;
 	}
 }
+
+/* Map of Verbs to Flow priority with 8 Verbs priorities. */
+static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
+};
+
+/* Map of Verbs to Flow priority with 16 Verbs priorities. */
+static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
+	{ 9, 10, 11 }, { 12, 13, 14 },
+};
+
+/**
+ * Discover the number of available flow priorities.
+ *
+ * @param dev
+ *   Ethernet device.
+ *
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+int
+mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+{
+	static const uint16_t vprio[] = {8, 16};
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	const struct mlx5_flow_driver_ops *fops;
+	enum mlx5_flow_drv_type type;
+	int ret;
+
+	type = mlx5_flow_os_get_type();
+	if (type == MLX5_FLOW_TYPE_MAX) {
+		type = MLX5_FLOW_TYPE_VERBS;
+		if (priv->sh->devx && priv->config.dv_flow_en)
+			type = MLX5_FLOW_TYPE_DV;
+	}
+	fops = flow_get_drv_ops(type);
+	if (fops->discover_priorities == NULL) {
+		DRV_LOG(ERR, "Priority discovery not supported");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
+	if (ret < 0)
+		return ret;
+	switch (ret) {
+	case 8:
+		ret = RTE_DIM(priority_map_3);
+		break;
+	case 16:
+		ret = RTE_DIM(priority_map_5);
+		break;
+	default:
+		rte_errno = ENOTSUP;
+		DRV_LOG(ERR,
+			"port %u maximum priority: %d expected 8/16",
+			dev->data->port_id, ret);
+		return -rte_errno;
+	}
+	DRV_LOG(INFO, "port %u supported flow priorities:"
+		" 0-%d for ingress or egress root table,"
+		" 0-%d for non-root table or transfer root table.",
+		dev->data->port_id, ret - 2,
+		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
+	return ret;
+}
+
+/**
+ * Adjust flow priority based on the highest layer and the request priority.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] priority
+ *   The rule base priority.
+ * @param[in] subpriority
+ *   The priority based on the items.
+ *
+ * @return
+ *   The new priority.
+ */
+uint32_t
+mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
+			  uint32_t subpriority)
+{
+	uint32_t res = 0;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	switch (priv->config.flow_prio) {
+	case RTE_DIM(priority_map_3):
+		res = priority_map_3[priority][subpriority];
+		break;
+	case RTE_DIM(priority_map_5):
+		res = priority_map_5[priority][subpriority];
+		break;
+	}
+	return  res;
+}
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 4a16f30fb7..2c9d3759b8 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1229,6 +1229,9 @@ typedef int (*mlx5_flow_create_def_policy_t)
 			(struct rte_eth_dev *dev);
 typedef void (*mlx5_flow_destroy_def_policy_t)
 			(struct rte_eth_dev *dev);
+typedef int (*mlx5_flow_discover_priorities_t)
+			(struct rte_eth_dev *dev,
+			 const uint16_t *vprio, int vprio_n);
 
 struct mlx5_flow_driver_ops {
 	mlx5_flow_validate_t validate;
@@ -1263,6 +1266,7 @@ struct mlx5_flow_driver_ops {
 	mlx5_flow_action_update_t action_update;
 	mlx5_flow_action_query_t action_query;
 	mlx5_flow_sync_domain_t sync_domain;
+	mlx5_flow_discover_priorities_t discover_priorities;
 };
 
 /* mlx5_flow.c */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 9cba22ca2d..3d59c72550 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -17909,6 +17909,108 @@ flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
 	return 0;
 }
 
+/**
+ * Discover the number of available flow priorities
+ * by trying to create a flow with the highest priority value
+ * for each possible number.
+ *
+ * @param[in] dev
+ *   Ethernet device.
+ * @param[in] vprio
+ *   List of possible number of available priorities.
+ * @param[in] vprio_n
+ *   Size of @p vprio array.
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+static int
+flow_dv_discover_priorities(struct rte_eth_dev *dev,
+			    const uint16_t *vprio, int vprio_n)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+	struct mlx5_flow_dv_matcher matcher = {
+		.mask = {
+			.size = sizeof(matcher.mask.buf),
+		},
+	};
+	union mlx5_flow_tbl_key tbl_key;
+	struct mlx5_flow flow;
+	void *action;
+	struct rte_flow_error error;
+	uint8_t misc_mask;
+	int i, err, ret = -ENOTSUP;
+
+	/*
+	 * Prepare a flow with a catch-all pattern and a drop action.
+	 * Use drop queue, because shared drop action may be unavailable.
+	 */
+	action = priv->drop_queue.hrxq->action;
+	if (action == NULL) {
+		DRV_LOG(ERR, "Priority discovery requires a drop action");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	memset(&flow, 0, sizeof(flow));
+	flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
+	if (flow.handle == NULL) {
+		DRV_LOG(ERR, "Cannot create flow handle");
+		rte_errno = ENOMEM;
+		return -rte_errno;
+	}
+	flow.ingress = true;
+	flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
+	flow.dv.actions[0] = action;
+	flow.dv.actions_n = 1;
+	memset(&eth, 0, sizeof(eth));
+	flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
+				   &item, /* inner */ false, /* group */ 0);
+	matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
+	for (i = 0; i < vprio_n; i++) {
+		/* Configure the next proposed maximum priority. */
+		matcher.priority = vprio[i] - 1;
+		memset(&tbl_key, 0, sizeof(tbl_key));
+		err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
+					       /* tunnel */ NULL,
+					       /* group */ 0,
+					       &error);
+		if (err != 0) {
+			/* This action is pure SW and must always succeed. */
+			DRV_LOG(ERR, "Cannot register matcher");
+			ret = -rte_errno;
+			break;
+		}
+		/* Try to apply the flow to HW. */
+		misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
+		__flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
+		err = mlx5_flow_os_create_flow
+				(flow.handle->dvh.matcher->matcher_object,
+				 (void *)&flow.dv.value, flow.dv.actions_n,
+				 flow.dv.actions, &flow.handle->drv_flow);
+		if (err == 0) {
+			claim_zero(mlx5_flow_os_destroy_flow
+						(flow.handle->drv_flow));
+			flow.handle->drv_flow = NULL;
+		}
+		claim_zero(flow_dv_matcher_release(dev, flow.handle));
+		if (err != 0)
+			break;
+		ret = vprio[i];
+	}
+	mlx5_ipool_free(pool, flow.handle_idx);
+	/* Set rte_errno if no expected priority value matched. */
+	if (ret < 0)
+		rte_errno = -ret;
+	return ret;
+}
+
 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.validate = flow_dv_validate,
 	.prepare = flow_dv_prepare,
@@ -17942,6 +18044,7 @@ const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.action_update = flow_dv_action_update,
 	.action_query = flow_dv_action_query,
 	.sync_domain = flow_dv_sync_domain,
+	.discover_priorities = flow_dv_discover_priorities,
 };
 
 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
diff --git a/drivers/net/mlx5/mlx5_flow_verbs.c b/drivers/net/mlx5/mlx5_flow_verbs.c
index 176d867202..92dc9903f3 100644
--- a/drivers/net/mlx5/mlx5_flow_verbs.c
+++ b/drivers/net/mlx5/mlx5_flow_verbs.c
@@ -28,17 +28,6 @@
 #define VERBS_SPEC_INNER(item_flags) \
 	(!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
 
-/* Map of Verbs to Flow priority with 8 Verbs priorities. */
-static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
-};
-
-/* Map of Verbs to Flow priority with 16 Verbs priorities. */
-static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
-	{ 9, 10, 11 }, { 12, 13, 14 },
-};
-
 /* Verbs specification header. */
 struct ibv_spec_header {
 	enum ibv_flow_spec_type type;
@@ -50,13 +39,17 @@ struct ibv_spec_header {
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
- *
+ * @param[in] vprio
+ *   Expected result variants.
+ * @param[in] vprio_n
+ *   Number of entries in @p vprio array.
  * @return
- *   number of supported flow priority on success, a negative errno
+ *   Number of supported flow priority on success, a negative errno
  *   value otherwise and rte_errno is set.
  */
-int
-mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+static int
+flow_verbs_discover_priorities(struct rte_eth_dev *dev,
+			       const uint16_t *vprio, int vprio_n)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct {
@@ -79,20 +72,19 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	};
 	struct ibv_flow *flow;
 	struct mlx5_hrxq *drop = priv->drop_queue.hrxq;
-	uint16_t vprio[] = { 8, 16 };
 	int i;
 	int priority = 0;
 
 #if defined(HAVE_MLX5DV_DR_DEVX_PORT) || defined(HAVE_MLX5DV_DR_DEVX_PORT_V35)
 	/* If DevX supported, driver must support 16 verbs flow priorities. */
-	priority = RTE_DIM(priority_map_5);
+	priority = 16;
 	goto out;
 #endif
 	if (!drop->qp) {
 		rte_errno = ENOTSUP;
 		return -rte_errno;
 	}
-	for (i = 0; i != RTE_DIM(vprio); i++) {
+	for (i = 0; i != vprio_n; i++) {
 		flow_attr.attr.priority = vprio[i] - 1;
 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 		if (!flow)
@@ -100,20 +92,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		claim_zero(mlx5_glue->destroy_flow(flow));
 		priority = vprio[i];
 	}
-	switch (priority) {
-	case 8:
-		priority = RTE_DIM(priority_map_3);
-		break;
-	case 16:
-		priority = RTE_DIM(priority_map_5);
-		break;
-	default:
-		rte_errno = ENOTSUP;
-		DRV_LOG(ERR,
-			"port %u verbs maximum priority: %d expected 8/16",
-			dev->data->port_id, priority);
-		return -rte_errno;
-	}
 #if defined(HAVE_MLX5DV_DR_DEVX_PORT) || defined(HAVE_MLX5DV_DR_DEVX_PORT_V35)
 out:
 #endif
@@ -125,37 +103,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	return priority;
 }
 
-/**
- * Adjust flow priority based on the highest layer and the request priority.
- *
- * @param[in] dev
- *   Pointer to the Ethernet device structure.
- * @param[in] priority
- *   The rule base priority.
- * @param[in] subpriority
- *   The priority based on the items.
- *
- * @return
- *   The new priority.
- */
-uint32_t
-mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
-				   uint32_t subpriority)
-{
-	uint32_t res = 0;
-	struct mlx5_priv *priv = dev->data->dev_private;
-
-	switch (priv->config.flow_prio) {
-	case RTE_DIM(priority_map_3):
-		res = priority_map_3[priority][subpriority];
-		break;
-	case RTE_DIM(priority_map_5):
-		res = priority_map_5[priority][subpriority];
-		break;
-	}
-	return  res;
-}
-
 /**
  * Get Verbs flow counter by index.
  *
@@ -2095,4 +2042,5 @@ const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
 	.destroy = flow_verbs_destroy,
 	.query = flow_verbs_query,
 	.sync_domain = flow_verbs_sync_domain,
+	.discover_priorities = flow_verbs_discover_priorities,
 };
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v5 5/6] net/mlx5: create drop queue using DevX
       [not found]       ` <20211102135415.944050-1-dkozlyuk@nvidia.com>
  2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
@ 2021-11-02 13:54         ` Dmitry Kozlyuk
  2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
       [not found]         ` <20211102170135.959380-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-11-02 13:54 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Drop queue creation and destruction were not implemented for DevX
flow engine and Verbs engine methods were used as a workaround.
Implement these methods for DevX so that there is a valid queue ID
that can be used regardless of queue configuration via API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c |   4 -
 drivers/net/mlx5/mlx5_devx.c     | 211 ++++++++++++++++++++++++++-----
 2 files changed, 180 insertions(+), 35 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 34546635c4..091763b745 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1685,10 +1685,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	}
 	if (sh->devx && config->dv_flow_en && config->dest_tir) {
 		priv->obj_ops = devx_obj_ops;
-		priv->obj_ops.drop_action_create =
-						ibv_obj_ops.drop_action_create;
-		priv->obj_ops.drop_action_destroy =
-						ibv_obj_ops.drop_action_destroy;
 		mlx5_queue_counter_id_prepare(eth_dev);
 		priv->obj_ops.lb_dummy_queue_create =
 					mlx5_rxq_ibv_obj_dummy_lb_create;
diff --git a/drivers/net/mlx5/mlx5_devx.c b/drivers/net/mlx5/mlx5_devx.c
index 7ed774e804..424f77be79 100644
--- a/drivers/net/mlx5/mlx5_devx.c
+++ b/drivers/net/mlx5/mlx5_devx.c
@@ -226,18 +226,18 @@ mlx5_rx_devx_get_event(struct mlx5_rxq_obj *rxq_obj)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_common_device *cdev = priv->sh->cdev;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	struct mlx5_devx_create_rq_attr rq_attr = { 0 };
@@ -290,20 +290,20 @@ mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_devx_cq *cq_obj = 0;
 	struct mlx5_devx_cq_attr cq_attr = { 0 };
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_dev_ctx_shared *sh = priv->sh;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	unsigned int cqe_n = mlx5_rxq_cqe_num(rxq_data);
@@ -498,13 +498,13 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
 		tmpl->fd = mlx5_os_get_devx_channel_fd(tmpl->devx_channel);
 	}
 	/* Create CQ using DevX API. */
-	ret = mlx5_rxq_create_devx_cq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Failed to create CQ.");
 		goto error;
 	}
 	/* Create RQ using DevX API. */
-	ret = mlx5_rxq_create_devx_rq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Port %u Rx queue %u RQ creation failure.",
 			dev->data->port_id, idx);
@@ -537,6 +537,11 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
  *   Pointer to Ethernet device.
  * @param log_n
  *   Log of number of queues in the array.
+ * @param queues
+ *   List of RX queue indices or NULL, in which case
+ *   the attribute will be filled by drop queue ID.
+ * @param queues_n
+ *   Size of @p queues array or 0 if it is NULL.
  * @param ind_tbl
  *   DevX indirection table object.
  *
@@ -564,6 +569,11 @@ mlx5_devx_ind_table_create_rqt_attr(struct rte_eth_dev *dev,
 	}
 	rqt_attr->rqt_max_size = priv->config.ind_table_max_size;
 	rqt_attr->rqt_actual_size = rqt_n;
+	if (queues == NULL) {
+		for (i = 0; i < rqt_n; i++)
+			rqt_attr->rq_list[i] = priv->drop_queue.rxq->rq->id;
+		return rqt_attr;
+	}
 	for (i = 0; i != queues_n; ++i) {
 		struct mlx5_rxq_data *rxq = (*priv->rxqs)[queues[i]];
 		struct mlx5_rxq_ctrl *rxq_ctrl =
@@ -596,11 +606,12 @@ mlx5_devx_ind_table_new(struct rte_eth_dev *dev, const unsigned int log_n,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_devx_rqt_attr *rqt_attr = NULL;
+	const uint16_t *queues = dev->data->dev_started ? ind_tbl->queues :
+							  NULL;
 
 	MLX5_ASSERT(ind_tbl);
-	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n,
-							ind_tbl->queues,
-							ind_tbl->queues_n);
+	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n, queues,
+						       ind_tbl->queues_n);
 	if (!rqt_attr)
 		return -rte_errno;
 	ind_tbl->rqt = mlx5_devx_cmd_create_rqt(priv->sh->cdev->ctx, rqt_attr);
@@ -671,7 +682,8 @@ mlx5_devx_ind_table_destroy(struct mlx5_ind_table_obj *ind_tbl)
  * @param[in] hash_fields
  *   Verbs protocol hash field to make the RSS on.
  * @param[in] ind_tbl
- *   Indirection table for TIR.
+ *   Indirection table for TIR. If table queues array is NULL,
+ *   a TIR for drop queue is assumed.
  * @param[in] tunnel
  *   Tunnel type.
  * @param[out] tir_attr
@@ -687,19 +699,27 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 		       int tunnel, struct mlx5_devx_tir_attr *tir_attr)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[ind_tbl->queues[0]];
-	struct mlx5_rxq_ctrl *rxq_ctrl =
-		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
-	enum mlx5_rxq_type rxq_obj_type = rxq_ctrl->type;
+	enum mlx5_rxq_type rxq_obj_type;
 	bool lro = true;
 	uint32_t i;
 
-	/* Enable TIR LRO only if all the queues were configured for. */
-	for (i = 0; i < ind_tbl->queues_n; ++i) {
-		if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
-			lro = false;
-			break;
+	/* NULL queues designate drop queue. */
+	if (ind_tbl->queues != NULL) {
+		struct mlx5_rxq_data *rxq_data =
+					(*priv->rxqs)[ind_tbl->queues[0]];
+		struct mlx5_rxq_ctrl *rxq_ctrl =
+			container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+		rxq_obj_type = rxq_ctrl->type;
+
+		/* Enable TIR LRO only if all the queues were configured for. */
+		for (i = 0; i < ind_tbl->queues_n; ++i) {
+			if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
+				lro = false;
+				break;
+			}
 		}
+	} else {
+		rxq_obj_type = priv->drop_queue.rxq->rxq_ctrl->type;
 	}
 	memset(tir_attr, 0, sizeof(*tir_attr));
 	tir_attr->disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
@@ -858,7 +878,7 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
 }
 
 /**
- * Create a DevX drop action for Rx Hash queue.
+ * Create a DevX drop Rx queue.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -867,14 +887,99 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+mlx5_rxq_devx_obj_drop_create(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	int socket_id = dev->device->numa_node;
+	struct mlx5_rxq_ctrl *rxq_ctrl;
+	struct mlx5_rxq_data *rxq_data;
+	struct mlx5_rxq_obj *rxq = NULL;
+	int ret;
+
+	/*
+	 * Initialize dummy control structures.
+	 * They are required to hold pointers for cleanup
+	 * and are only accessible via drop queue DevX objects.
+	 */
+	rxq_ctrl = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq_ctrl),
+			       0, socket_id);
+	if (rxq_ctrl == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue control",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq), 0, socket_id);
+	if (rxq == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue object",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq->rxq_ctrl = rxq_ctrl;
+	rxq_ctrl->type = MLX5_RXQ_TYPE_STANDARD;
+	rxq_ctrl->priv = priv;
+	rxq_ctrl->obj = rxq;
+	rxq_data = &rxq_ctrl->rxq;
+	/* Create CQ using DevX API. */
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue CQ creation failed.",
+			dev->data->port_id);
+		goto error;
+	}
+	/* Create RQ using DevX API. */
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue RQ creation failed.",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	/* Change queue state to ready. */
+	ret = mlx5_devx_modify_rq(rxq, MLX5_RXQ_MOD_RST2RDY);
+	if (ret != 0)
+		goto error;
+	/* Initialize drop queue. */
+	priv->drop_queue.rxq = rxq;
+	return 0;
+error:
+	ret = rte_errno; /* Save rte_errno before cleanup. */
+	if (rxq != NULL) {
+		if (rxq->rq_obj.rq != NULL)
+			mlx5_devx_rq_destroy(&rxq->rq_obj);
+		if (rxq->cq_obj.cq != NULL)
+			mlx5_devx_cq_destroy(&rxq->cq_obj);
+		if (rxq->devx_channel)
+			mlx5_os_devx_destroy_event_channel
+							(rxq->devx_channel);
+		mlx5_free(rxq);
+	}
+	if (rxq_ctrl != NULL)
+		mlx5_free(rxq_ctrl);
+	rte_errno = ret; /* Restore rte_errno. */
 	return -rte_errno;
 }
 
+/**
+ * Release drop Rx queue resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ */
+static void
+mlx5_rxq_devx_obj_drop_release(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_obj *rxq = priv->drop_queue.rxq;
+	struct mlx5_rxq_ctrl *rxq_ctrl = rxq->rxq_ctrl;
+
+	mlx5_rxq_devx_obj_release(rxq);
+	mlx5_free(rxq);
+	mlx5_free(rxq_ctrl);
+	priv->drop_queue.rxq = NULL;
+}
+
 /**
  * Release a drop hash Rx queue.
  *
@@ -884,9 +989,53 @@ mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
 static void
 mlx5_devx_drop_action_destroy(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+
+	if (hrxq->tir != NULL)
+		mlx5_devx_tir_destroy(hrxq);
+	if (hrxq->ind_table->ind_table != NULL)
+		mlx5_devx_ind_table_destroy(hrxq->ind_table);
+	if (priv->drop_queue.rxq->rq != NULL)
+		mlx5_rxq_devx_obj_drop_release(dev);
+}
+
+/**
+ * Create a DevX drop action for Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+	int ret;
+
+	ret = mlx5_rxq_devx_obj_drop_create(dev);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop RX queue");
+		return ret;
+	}
+	/* hrxq->ind_table queues are NULL, drop RX queue ID will be used */
+	ret = mlx5_devx_ind_table_new(dev, 0, hrxq->ind_table);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue indirection table");
+		goto error;
+	}
+	ret = mlx5_devx_hrxq_new(dev, hrxq, /* tunnel */ false);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue");
+		goto error;
+	}
+	return 0;
+error:
+	mlx5_devx_drop_action_destroy(dev);
+	return ret;
 }
 
 /**
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v5 6/6] net/mlx5: preserve indirect actions on restart
       [not found]       ` <20211102135415.944050-1-dkozlyuk@nvidia.com>
  2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
  2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
@ 2021-11-02 13:54         ` Dmitry Kozlyuk
       [not found]         ` <20211102170135.959380-1-dkozlyuk@nvidia.com>
  3 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-11-02 13:54 UTC (permalink / raw)
  To: dev; +Cc: bingz, stable, Matan Azrad, Viacheslav Ovsiienko

MLX5 PMD uses reference counting to manage RX queue resources.
After port stop shared RSS actions kept references to RX queues,
preventing resource release. As a result, internal PMD mempool
for such queues had been exhausted after a number of port restarts.
Diagnostic message from rte_eth_dev_start():

    Rx queue allocation failed: Cannot allocate memory

Dereference RX queues used by indirect actions on port stop (detach)
and restore references on port start (attach) in order to allow RX queue
resource release, but keep indirect RSS across the port restart.
Replace queue IDs in HW by drop queue ID on detach and restore actual
queue IDs on attach.

When the port is stopped, create indirect RSS in the detached state.
As a result, MLX5 PMD is able to keep all its indirect actions
across port restart. Advertise this capability.

Fixes: 4b61b8774be9 ("ethdev: introduce indirect flow action")
Cc: bingz@nvidia.com
Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/mlx5_ethdev.c  |   1 +
 drivers/net/mlx5/mlx5_flow.c    | 194 ++++++++++++++++++++++++++++----
 drivers/net/mlx5/mlx5_flow.h    |   2 +
 drivers/net/mlx5/mlx5_rx.h      |   4 +
 drivers/net/mlx5/mlx5_rxq.c     |  99 ++++++++++++++--
 drivers/net/mlx5/mlx5_trigger.c |  10 ++
 6 files changed, 276 insertions(+), 34 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index f2b78c3cc6..81fa8845bb 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -321,6 +321,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
 				 info->rx_queue_offload_capa);
 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
+	info->dev_capa = RTE_ETH_DEV_CAPA_FLOW_SHARED_OBJECT_KEEP;
 	info->if_index = mlx5_ifindex(dev);
 	info->reta_size = priv->reta_idx_n ?
 		priv->reta_idx_n : config->ind_table_max_size;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 850eb353fd..671a5a34d9 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1588,6 +1588,58 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 	return 0;
 }
 
+/**
+ * Validate queue numbers for device RSS.
+ *
+ * @param[in] dev
+ *   Configured device.
+ * @param[in] queues
+ *   Array of queue numbers.
+ * @param[in] queues_n
+ *   Size of the @p queues array.
+ * @param[out] error
+ *   On error, filled with a textual error description.
+ * @param[out] queue
+ *   On error, filled with an offending queue index in @p queues array.
+ *
+ * @return
+ *   0 on success, a negative errno code on error.
+ */
+static int
+mlx5_validate_rss_queues(const struct rte_eth_dev *dev,
+			 const uint16_t *queues, uint32_t queues_n,
+			 const char **error, uint32_t *queue_idx)
+{
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
+	uint32_t i;
+
+	for (i = 0; i != queues_n; ++i) {
+		struct mlx5_rxq_ctrl *rxq_ctrl;
+
+		if (queues[i] >= priv->rxqs_n) {
+			*error = "queue index out of range";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		if (!(*priv->rxqs)[queues[i]]) {
+			*error =  "queue is not configured";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		rxq_ctrl = container_of((*priv->rxqs)[queues[i]],
+					struct mlx5_rxq_ctrl, rxq);
+		if (i == 0)
+			rxq_type = rxq_ctrl->type;
+		if (rxq_type != rxq_ctrl->type) {
+			*error = "combining hairpin and regular RSS queues is not supported";
+			*queue_idx = i;
+			return -ENOTSUP;
+		}
+	}
+	return 0;
+}
+
 /*
  * Validate the rss action.
  *
@@ -1608,8 +1660,9 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_rss *rss = action->conf;
-	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
-	unsigned int i;
+	int ret;
+	const char *message;
+	uint32_t queue_idx;
 
 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
@@ -1673,27 +1726,12 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  NULL, "No queues configured");
-	for (i = 0; i != rss->queue_num; ++i) {
-		struct mlx5_rxq_ctrl *rxq_ctrl;
-
-		if (rss->queue[i] >= priv->rxqs_n)
-			return rte_flow_error_set
-				(error, EINVAL,
-				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue index out of range");
-		if (!(*priv->rxqs)[rss->queue[i]])
-			return rte_flow_error_set
-				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue is not configured");
-		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
-					struct mlx5_rxq_ctrl, rxq);
-		if (i == 0)
-			rxq_type = rxq_ctrl->type;
-		if (rxq_type != rxq_ctrl->type)
-			return rte_flow_error_set
-				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i],
-				 "combining hairpin and regular RSS queues is not supported");
+	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
+				       &message, &queue_idx);
+	if (ret != 0) {
+		return rte_flow_error_set(error, -ret,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &rss->queue[queue_idx], message);
 	}
 	return 0;
 }
@@ -8660,6 +8698,116 @@ mlx5_action_handle_flush(struct rte_eth_dev *dev)
 	return ret;
 }
 
+/**
+ * Validate existing indirect actions against current device configuration
+ * and attach them to device resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_attach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+		const char *message;
+		uint32_t queue_idx;
+
+		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
+					       ind_tbl->queues_n,
+					       &message, &queue_idx);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
+				dev->data->port_id, ind_tbl->queues[queue_idx],
+				message);
+			break;
+		}
+	}
+	if (ret != 0)
+		return ret;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not attach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not detach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
+/**
+ * Detach indirect actions of the device from its resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_detach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not detach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not attach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
 #ifndef HAVE_MLX5DV_DR
 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
 #else
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 2c9d3759b8..6eb254f115 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1577,6 +1577,8 @@ void mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
 		struct mlx5_flow_meter_policy *mtr_policy);
 int mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev);
 int mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev);
+int mlx5_action_handle_attach(struct rte_eth_dev *dev);
+int mlx5_action_handle_detach(struct rte_eth_dev *dev);
 int mlx5_action_handle_flush(struct rte_eth_dev *dev);
 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id);
 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh);
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 4952fe1455..69b1263339 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -211,6 +211,10 @@ int mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 			      struct mlx5_ind_table_obj *ind_tbl,
 			      uint16_t *queues, const uint32_t queues_n,
 			      bool standalone);
+int mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
+int mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
 struct mlx5_list_entry *mlx5_hrxq_create_cb(void *tool_ctx, void *cb_ctx);
 int mlx5_hrxq_match_cb(void *tool_ctx, struct mlx5_list_entry *entry,
 		       void *cb_ctx);
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 4f02fe02b9..9220bb2c15 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -2032,6 +2032,26 @@ mlx5_ind_table_obj_new(struct rte_eth_dev *dev, const uint16_t *queues,
 	return ind_tbl;
 }
 
+static int
+mlx5_ind_table_obj_check_standalone(struct rte_eth_dev *dev __rte_unused,
+				    struct mlx5_ind_table_obj *ind_tbl)
+{
+	uint32_t refcnt;
+
+	refcnt = __atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED);
+	if (refcnt <= 1)
+		return 0;
+	/*
+	 * Modification of indirection tables having more than 1
+	 * reference is unsupported.
+	 */
+	DRV_LOG(DEBUG,
+		"Port %u cannot modify indirection table %p (refcnt %u > 1).",
+		dev->data->port_id, (void *)ind_tbl, refcnt);
+	rte_errno = EINVAL;
+	return -rte_errno;
+}
+
 /**
  * Modify an indirection table.
  *
@@ -2064,18 +2084,8 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 
 	MLX5_ASSERT(standalone);
 	RTE_SET_USED(standalone);
-	if (__atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED) > 1) {
-		/*
-		 * Modification of indirection ntables having more than 1
-		 * reference unsupported. Intended for standalone indirection
-		 * tables only.
-		 */
-		DRV_LOG(DEBUG,
-			"Port %u cannot modify indirection table (refcnt> 1).",
-			dev->data->port_id);
-		rte_errno = EINVAL;
+	if (mlx5_ind_table_obj_check_standalone(dev, ind_tbl) < 0)
 		return -rte_errno;
-	}
 	for (i = 0; i != queues_n; ++i) {
 		if (!mlx5_rxq_get(dev, queues[i])) {
 			ret = -rte_errno;
@@ -2101,6 +2111,73 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 	return ret;
 }
 
+/**
+ * Attach an indirection table to its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to attach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_modify(dev, ind_tbl, ind_tbl->queues,
+					ind_tbl->queues_n, true);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_get(dev, ind_tbl->queues[i]);
+	return 0;
+}
+
+/**
+ * Detach an indirection table from its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to detach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	const unsigned int n = rte_is_power_of_2(ind_tbl->queues_n) ?
+			       log2above(ind_tbl->queues_n) :
+			       log2above(priv->config.ind_table_max_size);
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_check_standalone(dev, ind_tbl);
+	if (ret != 0)
+		return ret;
+	MLX5_ASSERT(priv->obj_ops.ind_table_modify);
+	ret = priv->obj_ops.ind_table_modify(dev, n, NULL, 0, ind_tbl);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_release(dev, ind_tbl->queues[i]);
+	return ret;
+}
+
 int
 mlx5_hrxq_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
 		   void *cb_ctx)
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index d916c8addc..ebeeae279e 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -14,6 +14,7 @@
 #include <mlx5_malloc.h>
 
 #include "mlx5.h"
+#include "mlx5_flow.h"
 #include "mlx5_rx.h"
 #include "mlx5_tx.h"
 #include "mlx5_utils.h"
@@ -1162,6 +1163,14 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 	mlx5_rxq_timestamp_set(dev);
 	/* Set a mask and offset of scheduling on timestamp into Tx queues. */
 	mlx5_txq_dynf_timestamp_set(dev);
+	/* Attach indirection table objects detached on port stop. */
+	ret = mlx5_action_handle_attach(dev);
+	if (ret) {
+		DRV_LOG(ERR,
+			"port %u failed to attach indirect actions: %s",
+			dev->data->port_id, rte_strerror(rte_errno));
+		goto error;
+	}
 	/*
 	 * In non-cached mode, it only needs to start the default mreg copy
 	 * action and no flow created by application exists anymore.
@@ -1239,6 +1248,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	/* All RX queue flags will be cleared in the flush interface. */
 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, true);
 	mlx5_flow_meter_rxq_flush(dev);
+	mlx5_action_handle_detach(dev);
 	mlx5_rx_intr_vec_disable(dev);
 	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
 	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v6 4/6] net/mlx5: discover max flow priority using DevX
       [not found]         ` <20211102170135.959380-1-dkozlyuk@nvidia.com>
@ 2021-11-02 17:01           ` Dmitry Kozlyuk
  2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
  2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
  2 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-11-02 17:01 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Maximum available flow priority was discovered using Verbs API
regardless of the selected flow engine. This required some Verbs
objects to be initialized in order to use DevX engine. Make priority
discovery an engine method and implement it for DevX using its API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/mlx5_flow.c       |  98 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow.h       |   4 ++
 drivers/net/mlx5/mlx5_flow_dv.c    | 103 +++++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  74 +++------------------
 4 files changed, 216 insertions(+), 63 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 2385a0b550..3d8dd974ce 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -9700,3 +9700,101 @@ mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
 	}
 	return node;
 }
+
+/* Map of Verbs to Flow priority with 8 Verbs priorities. */
+static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
+};
+
+/* Map of Verbs to Flow priority with 16 Verbs priorities. */
+static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
+	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
+	{ 9, 10, 11 }, { 12, 13, 14 },
+};
+
+/**
+ * Discover the number of available flow priorities.
+ *
+ * @param dev
+ *   Ethernet device.
+ *
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+int
+mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+{
+	static const uint16_t vprio[] = {8, 16};
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	const struct mlx5_flow_driver_ops *fops;
+	enum mlx5_flow_drv_type type;
+	int ret;
+
+	type = mlx5_flow_os_get_type();
+	if (type == MLX5_FLOW_TYPE_MAX) {
+		type = MLX5_FLOW_TYPE_VERBS;
+		if (priv->sh->devx && priv->config.dv_flow_en)
+			type = MLX5_FLOW_TYPE_DV;
+	}
+	fops = flow_get_drv_ops(type);
+	if (fops->discover_priorities == NULL) {
+		DRV_LOG(ERR, "Priority discovery not supported");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
+	if (ret < 0)
+		return ret;
+	switch (ret) {
+	case 8:
+		ret = RTE_DIM(priority_map_3);
+		break;
+	case 16:
+		ret = RTE_DIM(priority_map_5);
+		break;
+	default:
+		rte_errno = ENOTSUP;
+		DRV_LOG(ERR,
+			"port %u maximum priority: %d expected 8/16",
+			dev->data->port_id, ret);
+		return -rte_errno;
+	}
+	DRV_LOG(INFO, "port %u supported flow priorities:"
+		" 0-%d for ingress or egress root table,"
+		" 0-%d for non-root table or transfer root table.",
+		dev->data->port_id, ret - 2,
+		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
+	return ret;
+}
+
+/**
+ * Adjust flow priority based on the highest layer and the request priority.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] priority
+ *   The rule base priority.
+ * @param[in] subpriority
+ *   The priority based on the items.
+ *
+ * @return
+ *   The new priority.
+ */
+uint32_t
+mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
+			  uint32_t subpriority)
+{
+	uint32_t res = 0;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	switch (priv->sh->flow_max_priority) {
+	case RTE_DIM(priority_map_3):
+		res = priority_map_3[priority][subpriority];
+		break;
+	case RTE_DIM(priority_map_5):
+		res = priority_map_5[priority][subpriority];
+		break;
+	}
+	return  res;
+}
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 5509c28f01..8b83fa6f67 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1232,6 +1232,9 @@ typedef int (*mlx5_flow_create_def_policy_t)
 			(struct rte_eth_dev *dev);
 typedef void (*mlx5_flow_destroy_def_policy_t)
 			(struct rte_eth_dev *dev);
+typedef int (*mlx5_flow_discover_priorities_t)
+			(struct rte_eth_dev *dev,
+			 const uint16_t *vprio, int vprio_n);
 
 struct mlx5_flow_driver_ops {
 	mlx5_flow_validate_t validate;
@@ -1266,6 +1269,7 @@ struct mlx5_flow_driver_ops {
 	mlx5_flow_action_update_t action_update;
 	mlx5_flow_action_query_t action_query;
 	mlx5_flow_sync_domain_t sync_domain;
+	mlx5_flow_discover_priorities_t discover_priorities;
 };
 
 /* mlx5_flow.c */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 8962d26c75..aaf96fc297 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -17932,6 +17932,108 @@ flow_dv_sync_domain(struct rte_eth_dev *dev, uint32_t domains, uint32_t flags)
 	return 0;
 }
 
+/**
+ * Discover the number of available flow priorities
+ * by trying to create a flow with the highest priority value
+ * for each possible number.
+ *
+ * @param[in] dev
+ *   Ethernet device.
+ * @param[in] vprio
+ *   List of possible number of available priorities.
+ * @param[in] vprio_n
+ *   Size of @p vprio array.
+ * @return
+ *   On success, number of available flow priorities.
+ *   On failure, a negative errno-style code and rte_errno is set.
+ */
+static int
+flow_dv_discover_priorities(struct rte_eth_dev *dev,
+			    const uint16_t *vprio, int vprio_n)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *pool = priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW];
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+	struct mlx5_flow_dv_matcher matcher = {
+		.mask = {
+			.size = sizeof(matcher.mask.buf),
+		},
+	};
+	union mlx5_flow_tbl_key tbl_key;
+	struct mlx5_flow flow;
+	void *action;
+	struct rte_flow_error error;
+	uint8_t misc_mask;
+	int i, err, ret = -ENOTSUP;
+
+	/*
+	 * Prepare a flow with a catch-all pattern and a drop action.
+	 * Use drop queue, because shared drop action may be unavailable.
+	 */
+	action = priv->drop_queue.hrxq->action;
+	if (action == NULL) {
+		DRV_LOG(ERR, "Priority discovery requires a drop action");
+		rte_errno = ENOTSUP;
+		return -rte_errno;
+	}
+	memset(&flow, 0, sizeof(flow));
+	flow.handle = mlx5_ipool_zmalloc(pool, &flow.handle_idx);
+	if (flow.handle == NULL) {
+		DRV_LOG(ERR, "Cannot create flow handle");
+		rte_errno = ENOMEM;
+		return -rte_errno;
+	}
+	flow.ingress = true;
+	flow.dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
+	flow.dv.actions[0] = action;
+	flow.dv.actions_n = 1;
+	memset(&eth, 0, sizeof(eth));
+	flow_dv_translate_item_eth(matcher.mask.buf, flow.dv.value.buf,
+				   &item, /* inner */ false, /* group */ 0);
+	matcher.crc = rte_raw_cksum(matcher.mask.buf, matcher.mask.size);
+	for (i = 0; i < vprio_n; i++) {
+		/* Configure the next proposed maximum priority. */
+		matcher.priority = vprio[i] - 1;
+		memset(&tbl_key, 0, sizeof(tbl_key));
+		err = flow_dv_matcher_register(dev, &matcher, &tbl_key, &flow,
+					       /* tunnel */ NULL,
+					       /* group */ 0,
+					       &error);
+		if (err != 0) {
+			/* This action is pure SW and must always succeed. */
+			DRV_LOG(ERR, "Cannot register matcher");
+			ret = -rte_errno;
+			break;
+		}
+		/* Try to apply the flow to HW. */
+		misc_mask = flow_dv_matcher_enable(flow.dv.value.buf);
+		__flow_dv_adjust_buf_size(&flow.dv.value.size, misc_mask);
+		err = mlx5_flow_os_create_flow
+				(flow.handle->dvh.matcher->matcher_object,
+				 (void *)&flow.dv.value, flow.dv.actions_n,
+				 flow.dv.actions, &flow.handle->drv_flow);
+		if (err == 0) {
+			claim_zero(mlx5_flow_os_destroy_flow
+						(flow.handle->drv_flow));
+			flow.handle->drv_flow = NULL;
+		}
+		claim_zero(flow_dv_matcher_release(dev, flow.handle));
+		if (err != 0)
+			break;
+		ret = vprio[i];
+	}
+	mlx5_ipool_free(pool, flow.handle_idx);
+	/* Set rte_errno if no expected priority value matched. */
+	if (ret < 0)
+		rte_errno = -ret;
+	return ret;
+}
+
 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.validate = flow_dv_validate,
 	.prepare = flow_dv_prepare,
@@ -17965,6 +18067,7 @@ const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
 	.action_update = flow_dv_action_update,
 	.action_query = flow_dv_action_query,
 	.sync_domain = flow_dv_sync_domain,
+	.discover_priorities = flow_dv_discover_priorities,
 };
 
 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
diff --git a/drivers/net/mlx5/mlx5_flow_verbs.c b/drivers/net/mlx5/mlx5_flow_verbs.c
index 0a89a136a2..29cd694752 100644
--- a/drivers/net/mlx5/mlx5_flow_verbs.c
+++ b/drivers/net/mlx5/mlx5_flow_verbs.c
@@ -28,17 +28,6 @@
 #define VERBS_SPEC_INNER(item_flags) \
 	(!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
 
-/* Map of Verbs to Flow priority with 8 Verbs priorities. */
-static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
-};
-
-/* Map of Verbs to Flow priority with 16 Verbs priorities. */
-static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
-	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
-	{ 9, 10, 11 }, { 12, 13, 14 },
-};
-
 /* Verbs specification header. */
 struct ibv_spec_header {
 	enum ibv_flow_spec_type type;
@@ -50,13 +39,17 @@ struct ibv_spec_header {
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
- *
+ * @param[in] vprio
+ *   Expected result variants.
+ * @param[in] vprio_n
+ *   Number of entries in @p vprio array.
  * @return
- *   number of supported flow priority on success, a negative errno
+ *   Number of supported flow priority on success, a negative errno
  *   value otherwise and rte_errno is set.
  */
-int
-mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
+static int
+flow_verbs_discover_priorities(struct rte_eth_dev *dev,
+			       const uint16_t *vprio, int vprio_n)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct {
@@ -79,20 +72,19 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	};
 	struct ibv_flow *flow;
 	struct mlx5_hrxq *drop = priv->drop_queue.hrxq;
-	uint16_t vprio[] = { 8, 16 };
 	int i;
 	int priority = 0;
 
 #if defined(HAVE_MLX5DV_DR_DEVX_PORT) || defined(HAVE_MLX5DV_DR_DEVX_PORT_V35)
 	/* If DevX supported, driver must support 16 verbs flow priorities. */
-	priority = RTE_DIM(priority_map_5);
+	priority = 16;
 	goto out;
 #endif
 	if (!drop->qp) {
 		rte_errno = ENOTSUP;
 		return -rte_errno;
 	}
-	for (i = 0; i != RTE_DIM(vprio); i++) {
+	for (i = 0; i != vprio_n; i++) {
 		flow_attr.attr.priority = vprio[i] - 1;
 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
 		if (!flow)
@@ -100,20 +92,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 		claim_zero(mlx5_glue->destroy_flow(flow));
 		priority = vprio[i];
 	}
-	switch (priority) {
-	case 8:
-		priority = RTE_DIM(priority_map_3);
-		break;
-	case 16:
-		priority = RTE_DIM(priority_map_5);
-		break;
-	default:
-		rte_errno = ENOTSUP;
-		DRV_LOG(ERR,
-			"port %u verbs maximum priority: %d expected 8/16",
-			dev->data->port_id, priority);
-		return -rte_errno;
-	}
 #if defined(HAVE_MLX5DV_DR_DEVX_PORT) || defined(HAVE_MLX5DV_DR_DEVX_PORT_V35)
 out:
 #endif
@@ -125,37 +103,6 @@ mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
 	return priority;
 }
 
-/**
- * Adjust flow priority based on the highest layer and the request priority.
- *
- * @param[in] dev
- *   Pointer to the Ethernet device structure.
- * @param[in] priority
- *   The rule base priority.
- * @param[in] subpriority
- *   The priority based on the items.
- *
- * @return
- *   The new priority.
- */
-uint32_t
-mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
-				   uint32_t subpriority)
-{
-	uint32_t res = 0;
-	struct mlx5_priv *priv = dev->data->dev_private;
-
-	switch (priv->sh->flow_max_priority) {
-	case RTE_DIM(priority_map_3):
-		res = priority_map_3[priority][subpriority];
-		break;
-	case RTE_DIM(priority_map_5):
-		res = priority_map_5[priority][subpriority];
-		break;
-	}
-	return  res;
-}
-
 /**
  * Get Verbs flow counter by index.
  *
@@ -2095,4 +2042,5 @@ const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
 	.destroy = flow_verbs_destroy,
 	.query = flow_verbs_query,
 	.sync_domain = flow_verbs_sync_domain,
+	.discover_priorities = flow_verbs_discover_priorities,
 };
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v6 5/6] net/mlx5: create drop queue using DevX
       [not found]         ` <20211102170135.959380-1-dkozlyuk@nvidia.com>
  2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
@ 2021-11-02 17:01           ` Dmitry Kozlyuk
  2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
  2 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-11-02 17:01 UTC (permalink / raw)
  To: dev; +Cc: stable, Matan Azrad, Viacheslav Ovsiienko

Drop queue creation and destruction were not implemented for DevX
flow engine and Verbs engine methods were used as a workaround.
Implement these methods for DevX so that there is a valid queue ID
that can be used regardless of queue configuration via API.

Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c |   4 -
 drivers/net/mlx5/mlx5_devx.c     | 211 ++++++++++++++++++++++++++-----
 2 files changed, 180 insertions(+), 35 deletions(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index f31f1e96c6..dd4fc0c716 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1690,10 +1690,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 	}
 	if (sh->devx && config->dv_flow_en && config->dest_tir) {
 		priv->obj_ops = devx_obj_ops;
-		priv->obj_ops.drop_action_create =
-						ibv_obj_ops.drop_action_create;
-		priv->obj_ops.drop_action_destroy =
-						ibv_obj_ops.drop_action_destroy;
 		mlx5_queue_counter_id_prepare(eth_dev);
 		priv->obj_ops.lb_dummy_queue_create =
 					mlx5_rxq_ibv_obj_dummy_lb_create;
diff --git a/drivers/net/mlx5/mlx5_devx.c b/drivers/net/mlx5/mlx5_devx.c
index 7ed774e804..424f77be79 100644
--- a/drivers/net/mlx5/mlx5_devx.c
+++ b/drivers/net/mlx5/mlx5_devx.c
@@ -226,18 +226,18 @@ mlx5_rx_devx_get_event(struct mlx5_rxq_obj *rxq_obj)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_common_device *cdev = priv->sh->cdev;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	struct mlx5_devx_create_rq_attr rq_attr = { 0 };
@@ -290,20 +290,20 @@ mlx5_rxq_create_devx_rq_resources(struct rte_eth_dev *dev, uint16_t idx)
  *
  * @param dev
  *   Pointer to Ethernet device.
- * @param idx
- *   Queue index in DPDK Rx queue array.
+ * @param rxq_data
+ *   RX queue data.
  *
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev, uint16_t idx)
+mlx5_rxq_create_devx_cq_resources(struct rte_eth_dev *dev,
+				  struct mlx5_rxq_data *rxq_data)
 {
 	struct mlx5_devx_cq *cq_obj = 0;
 	struct mlx5_devx_cq_attr cq_attr = { 0 };
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_dev_ctx_shared *sh = priv->sh;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
 	struct mlx5_rxq_ctrl *rxq_ctrl =
 		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
 	unsigned int cqe_n = mlx5_rxq_cqe_num(rxq_data);
@@ -498,13 +498,13 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
 		tmpl->fd = mlx5_os_get_devx_channel_fd(tmpl->devx_channel);
 	}
 	/* Create CQ using DevX API. */
-	ret = mlx5_rxq_create_devx_cq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Failed to create CQ.");
 		goto error;
 	}
 	/* Create RQ using DevX API. */
-	ret = mlx5_rxq_create_devx_rq_resources(dev, idx);
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
 	if (ret) {
 		DRV_LOG(ERR, "Port %u Rx queue %u RQ creation failure.",
 			dev->data->port_id, idx);
@@ -537,6 +537,11 @@ mlx5_rxq_devx_obj_new(struct rte_eth_dev *dev, uint16_t idx)
  *   Pointer to Ethernet device.
  * @param log_n
  *   Log of number of queues in the array.
+ * @param queues
+ *   List of RX queue indices or NULL, in which case
+ *   the attribute will be filled by drop queue ID.
+ * @param queues_n
+ *   Size of @p queues array or 0 if it is NULL.
  * @param ind_tbl
  *   DevX indirection table object.
  *
@@ -564,6 +569,11 @@ mlx5_devx_ind_table_create_rqt_attr(struct rte_eth_dev *dev,
 	}
 	rqt_attr->rqt_max_size = priv->config.ind_table_max_size;
 	rqt_attr->rqt_actual_size = rqt_n;
+	if (queues == NULL) {
+		for (i = 0; i < rqt_n; i++)
+			rqt_attr->rq_list[i] = priv->drop_queue.rxq->rq->id;
+		return rqt_attr;
+	}
 	for (i = 0; i != queues_n; ++i) {
 		struct mlx5_rxq_data *rxq = (*priv->rxqs)[queues[i]];
 		struct mlx5_rxq_ctrl *rxq_ctrl =
@@ -596,11 +606,12 @@ mlx5_devx_ind_table_new(struct rte_eth_dev *dev, const unsigned int log_n,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	struct mlx5_devx_rqt_attr *rqt_attr = NULL;
+	const uint16_t *queues = dev->data->dev_started ? ind_tbl->queues :
+							  NULL;
 
 	MLX5_ASSERT(ind_tbl);
-	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n,
-							ind_tbl->queues,
-							ind_tbl->queues_n);
+	rqt_attr = mlx5_devx_ind_table_create_rqt_attr(dev, log_n, queues,
+						       ind_tbl->queues_n);
 	if (!rqt_attr)
 		return -rte_errno;
 	ind_tbl->rqt = mlx5_devx_cmd_create_rqt(priv->sh->cdev->ctx, rqt_attr);
@@ -671,7 +682,8 @@ mlx5_devx_ind_table_destroy(struct mlx5_ind_table_obj *ind_tbl)
  * @param[in] hash_fields
  *   Verbs protocol hash field to make the RSS on.
  * @param[in] ind_tbl
- *   Indirection table for TIR.
+ *   Indirection table for TIR. If table queues array is NULL,
+ *   a TIR for drop queue is assumed.
  * @param[in] tunnel
  *   Tunnel type.
  * @param[out] tir_attr
@@ -687,19 +699,27 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 		       int tunnel, struct mlx5_devx_tir_attr *tir_attr)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[ind_tbl->queues[0]];
-	struct mlx5_rxq_ctrl *rxq_ctrl =
-		container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
-	enum mlx5_rxq_type rxq_obj_type = rxq_ctrl->type;
+	enum mlx5_rxq_type rxq_obj_type;
 	bool lro = true;
 	uint32_t i;
 
-	/* Enable TIR LRO only if all the queues were configured for. */
-	for (i = 0; i < ind_tbl->queues_n; ++i) {
-		if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
-			lro = false;
-			break;
+	/* NULL queues designate drop queue. */
+	if (ind_tbl->queues != NULL) {
+		struct mlx5_rxq_data *rxq_data =
+					(*priv->rxqs)[ind_tbl->queues[0]];
+		struct mlx5_rxq_ctrl *rxq_ctrl =
+			container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+		rxq_obj_type = rxq_ctrl->type;
+
+		/* Enable TIR LRO only if all the queues were configured for. */
+		for (i = 0; i < ind_tbl->queues_n; ++i) {
+			if (!(*priv->rxqs)[ind_tbl->queues[i]]->lro) {
+				lro = false;
+				break;
+			}
 		}
+	} else {
+		rxq_obj_type = priv->drop_queue.rxq->rxq_ctrl->type;
 	}
 	memset(tir_attr, 0, sizeof(*tir_attr));
 	tir_attr->disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
@@ -858,7 +878,7 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
 }
 
 /**
- * Create a DevX drop action for Rx Hash queue.
+ * Create a DevX drop Rx queue.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -867,14 +887,99 @@ mlx5_devx_hrxq_modify(struct rte_eth_dev *dev, struct mlx5_hrxq *hrxq,
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+mlx5_rxq_devx_obj_drop_create(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	int socket_id = dev->device->numa_node;
+	struct mlx5_rxq_ctrl *rxq_ctrl;
+	struct mlx5_rxq_data *rxq_data;
+	struct mlx5_rxq_obj *rxq = NULL;
+	int ret;
+
+	/*
+	 * Initialize dummy control structures.
+	 * They are required to hold pointers for cleanup
+	 * and are only accessible via drop queue DevX objects.
+	 */
+	rxq_ctrl = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq_ctrl),
+			       0, socket_id);
+	if (rxq_ctrl == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue control",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*rxq), 0, socket_id);
+	if (rxq == NULL) {
+		DRV_LOG(ERR, "Port %u could not allocate drop queue object",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	rxq->rxq_ctrl = rxq_ctrl;
+	rxq_ctrl->type = MLX5_RXQ_TYPE_STANDARD;
+	rxq_ctrl->priv = priv;
+	rxq_ctrl->obj = rxq;
+	rxq_data = &rxq_ctrl->rxq;
+	/* Create CQ using DevX API. */
+	ret = mlx5_rxq_create_devx_cq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue CQ creation failed.",
+			dev->data->port_id);
+		goto error;
+	}
+	/* Create RQ using DevX API. */
+	ret = mlx5_rxq_create_devx_rq_resources(dev, rxq_data);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u drop queue RQ creation failed.",
+			dev->data->port_id);
+		rte_errno = ENOMEM;
+		goto error;
+	}
+	/* Change queue state to ready. */
+	ret = mlx5_devx_modify_rq(rxq, MLX5_RXQ_MOD_RST2RDY);
+	if (ret != 0)
+		goto error;
+	/* Initialize drop queue. */
+	priv->drop_queue.rxq = rxq;
+	return 0;
+error:
+	ret = rte_errno; /* Save rte_errno before cleanup. */
+	if (rxq != NULL) {
+		if (rxq->rq_obj.rq != NULL)
+			mlx5_devx_rq_destroy(&rxq->rq_obj);
+		if (rxq->cq_obj.cq != NULL)
+			mlx5_devx_cq_destroy(&rxq->cq_obj);
+		if (rxq->devx_channel)
+			mlx5_os_devx_destroy_event_channel
+							(rxq->devx_channel);
+		mlx5_free(rxq);
+	}
+	if (rxq_ctrl != NULL)
+		mlx5_free(rxq_ctrl);
+	rte_errno = ret; /* Restore rte_errno. */
 	return -rte_errno;
 }
 
+/**
+ * Release drop Rx queue resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ */
+static void
+mlx5_rxq_devx_obj_drop_release(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_obj *rxq = priv->drop_queue.rxq;
+	struct mlx5_rxq_ctrl *rxq_ctrl = rxq->rxq_ctrl;
+
+	mlx5_rxq_devx_obj_release(rxq);
+	mlx5_free(rxq);
+	mlx5_free(rxq_ctrl);
+	priv->drop_queue.rxq = NULL;
+}
+
 /**
  * Release a drop hash Rx queue.
  *
@@ -884,9 +989,53 @@ mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
 static void
 mlx5_devx_drop_action_destroy(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	DRV_LOG(ERR, "DevX drop action is not supported yet.");
-	rte_errno = ENOTSUP;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+
+	if (hrxq->tir != NULL)
+		mlx5_devx_tir_destroy(hrxq);
+	if (hrxq->ind_table->ind_table != NULL)
+		mlx5_devx_ind_table_destroy(hrxq->ind_table);
+	if (priv->drop_queue.rxq->rq != NULL)
+		mlx5_rxq_devx_obj_drop_release(dev);
+}
+
+/**
+ * Create a DevX drop action for Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_devx_drop_action_create(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_hrxq *hrxq = priv->drop_queue.hrxq;
+	int ret;
+
+	ret = mlx5_rxq_devx_obj_drop_create(dev);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop RX queue");
+		return ret;
+	}
+	/* hrxq->ind_table queues are NULL, drop RX queue ID will be used */
+	ret = mlx5_devx_ind_table_new(dev, 0, hrxq->ind_table);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue indirection table");
+		goto error;
+	}
+	ret = mlx5_devx_hrxq_new(dev, hrxq, /* tunnel */ false);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Cannot create drop hash RX queue");
+		goto error;
+	}
+	return 0;
+error:
+	mlx5_devx_drop_action_destroy(dev);
+	return ret;
 }
 
 /**
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [dpdk-stable] [PATCH v6 6/6] net/mlx5: preserve indirect actions on restart
       [not found]         ` <20211102170135.959380-1-dkozlyuk@nvidia.com>
  2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
  2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
@ 2021-11-02 17:01           ` Dmitry Kozlyuk
  2 siblings, 0 replies; 18+ messages in thread
From: Dmitry Kozlyuk @ 2021-11-02 17:01 UTC (permalink / raw)
  To: dev; +Cc: bingz, stable, Matan Azrad, Viacheslav Ovsiienko

MLX5 PMD uses reference counting to manage RX queue resources.
After port stop shared RSS actions kept references to RX queues,
preventing resource release. As a result, internal PMD mempool
for such queues had been exhausted after a number of port restarts.
Diagnostic message from rte_eth_dev_start():

    Rx queue allocation failed: Cannot allocate memory

Dereference RX queues used by indirect actions on port stop (detach)
and restore references on port start (attach) in order to allow RX queue
resource release, but keep indirect RSS across the port restart.
Replace queue IDs in HW by drop queue ID on detach and restore actual
queue IDs on attach.

When the port is stopped, create indirect RSS in the detached state.
As a result, MLX5 PMD is able to keep all its indirect actions
across port restart. Advertise this capability.

Fixes: 4b61b8774be9 ("ethdev: introduce indirect flow action")
Cc: bingz@nvidia.com
Cc: stable@dpdk.org

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
 drivers/net/mlx5/mlx5_ethdev.c  |   1 +
 drivers/net/mlx5/mlx5_flow.c    | 194 ++++++++++++++++++++++++++++----
 drivers/net/mlx5/mlx5_flow.h    |   2 +
 drivers/net/mlx5/mlx5_rx.h      |   4 +
 drivers/net/mlx5/mlx5_rxq.c     |  99 ++++++++++++++--
 drivers/net/mlx5/mlx5_trigger.c |  10 ++
 6 files changed, 276 insertions(+), 34 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index f2b78c3cc6..81fa8845bb 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -321,6 +321,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
 				 info->rx_queue_offload_capa);
 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
+	info->dev_capa = RTE_ETH_DEV_CAPA_FLOW_SHARED_OBJECT_KEEP;
 	info->if_index = mlx5_ifindex(dev);
 	info->reta_size = priv->reta_idx_n ?
 		priv->reta_idx_n : config->ind_table_max_size;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 3d8dd974ce..9904bc5863 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1594,6 +1594,58 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 	return 0;
 }
 
+/**
+ * Validate queue numbers for device RSS.
+ *
+ * @param[in] dev
+ *   Configured device.
+ * @param[in] queues
+ *   Array of queue numbers.
+ * @param[in] queues_n
+ *   Size of the @p queues array.
+ * @param[out] error
+ *   On error, filled with a textual error description.
+ * @param[out] queue
+ *   On error, filled with an offending queue index in @p queues array.
+ *
+ * @return
+ *   0 on success, a negative errno code on error.
+ */
+static int
+mlx5_validate_rss_queues(const struct rte_eth_dev *dev,
+			 const uint16_t *queues, uint32_t queues_n,
+			 const char **error, uint32_t *queue_idx)
+{
+	const struct mlx5_priv *priv = dev->data->dev_private;
+	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
+	uint32_t i;
+
+	for (i = 0; i != queues_n; ++i) {
+		struct mlx5_rxq_ctrl *rxq_ctrl;
+
+		if (queues[i] >= priv->rxqs_n) {
+			*error = "queue index out of range";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		if (!(*priv->rxqs)[queues[i]]) {
+			*error =  "queue is not configured";
+			*queue_idx = i;
+			return -EINVAL;
+		}
+		rxq_ctrl = container_of((*priv->rxqs)[queues[i]],
+					struct mlx5_rxq_ctrl, rxq);
+		if (i == 0)
+			rxq_type = rxq_ctrl->type;
+		if (rxq_type != rxq_ctrl->type) {
+			*error = "combining hairpin and regular RSS queues is not supported";
+			*queue_idx = i;
+			return -ENOTSUP;
+		}
+	}
+	return 0;
+}
+
 /*
  * Validate the rss action.
  *
@@ -1614,8 +1666,9 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_rss *rss = action->conf;
-	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
-	unsigned int i;
+	int ret;
+	const char *message;
+	uint32_t queue_idx;
 
 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
@@ -1679,27 +1732,12 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  NULL, "No queues configured");
-	for (i = 0; i != rss->queue_num; ++i) {
-		struct mlx5_rxq_ctrl *rxq_ctrl;
-
-		if (rss->queue[i] >= priv->rxqs_n)
-			return rte_flow_error_set
-				(error, EINVAL,
-				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue index out of range");
-		if (!(*priv->rxqs)[rss->queue[i]])
-			return rte_flow_error_set
-				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i], "queue is not configured");
-		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
-					struct mlx5_rxq_ctrl, rxq);
-		if (i == 0)
-			rxq_type = rxq_ctrl->type;
-		if (rxq_type != rxq_ctrl->type)
-			return rte_flow_error_set
-				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-				 &rss->queue[i],
-				 "combining hairpin and regular RSS queues is not supported");
+	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
+				       &message, &queue_idx);
+	if (ret != 0) {
+		return rte_flow_error_set(error, -ret,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &rss->queue[queue_idx], message);
 	}
 	return 0;
 }
@@ -8786,6 +8824,116 @@ mlx5_action_handle_flush(struct rte_eth_dev *dev)
 	return ret;
 }
 
+/**
+ * Validate existing indirect actions against current device configuration
+ * and attach them to device resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_attach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+		const char *message;
+		uint32_t queue_idx;
+
+		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
+					       ind_tbl->queues_n,
+					       &message, &queue_idx);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
+				dev->data->port_id, ind_tbl->queues[queue_idx],
+				message);
+			break;
+		}
+	}
+	if (ret != 0)
+		return ret;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not attach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not detach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
+/**
+ * Detach indirect actions of the device from its resources.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_action_handle_detach(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_indexed_pool *ipool =
+			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
+	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
+	int ret = 0;
+	uint32_t idx;
+
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
+		if (ret != 0) {
+			DRV_LOG(ERR, "Port %u could not detach "
+				"indirection table obj %p",
+				dev->data->port_id, (void *)ind_tbl);
+			goto error;
+		}
+	}
+	return 0;
+error:
+	shared_rss_last = shared_rss;
+	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
+		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
+
+		if (shared_rss == shared_rss_last)
+			break;
+		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
+			DRV_LOG(CRIT, "Port %u could not attach "
+				"indirection table obj %p on rollback",
+				dev->data->port_id, (void *)ind_tbl);
+	}
+	return ret;
+}
+
 #ifndef HAVE_MLX5DV_DR
 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
 #else
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index 8b83fa6f67..8fbc37feb7 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1582,6 +1582,8 @@ void mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
 		struct mlx5_flow_meter_policy *mtr_policy);
 int mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev);
 int mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev);
+int mlx5_action_handle_attach(struct rte_eth_dev *dev);
+int mlx5_action_handle_detach(struct rte_eth_dev *dev);
 int mlx5_action_handle_flush(struct rte_eth_dev *dev);
 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id);
 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh);
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 4952fe1455..69b1263339 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -211,6 +211,10 @@ int mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 			      struct mlx5_ind_table_obj *ind_tbl,
 			      uint16_t *queues, const uint32_t queues_n,
 			      bool standalone);
+int mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
+int mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			      struct mlx5_ind_table_obj *ind_tbl);
 struct mlx5_list_entry *mlx5_hrxq_create_cb(void *tool_ctx, void *cb_ctx);
 int mlx5_hrxq_match_cb(void *tool_ctx, struct mlx5_list_entry *entry,
 		       void *cb_ctx);
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 4f02fe02b9..9220bb2c15 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -2032,6 +2032,26 @@ mlx5_ind_table_obj_new(struct rte_eth_dev *dev, const uint16_t *queues,
 	return ind_tbl;
 }
 
+static int
+mlx5_ind_table_obj_check_standalone(struct rte_eth_dev *dev __rte_unused,
+				    struct mlx5_ind_table_obj *ind_tbl)
+{
+	uint32_t refcnt;
+
+	refcnt = __atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED);
+	if (refcnt <= 1)
+		return 0;
+	/*
+	 * Modification of indirection tables having more than 1
+	 * reference is unsupported.
+	 */
+	DRV_LOG(DEBUG,
+		"Port %u cannot modify indirection table %p (refcnt %u > 1).",
+		dev->data->port_id, (void *)ind_tbl, refcnt);
+	rte_errno = EINVAL;
+	return -rte_errno;
+}
+
 /**
  * Modify an indirection table.
  *
@@ -2064,18 +2084,8 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 
 	MLX5_ASSERT(standalone);
 	RTE_SET_USED(standalone);
-	if (__atomic_load_n(&ind_tbl->refcnt, __ATOMIC_RELAXED) > 1) {
-		/*
-		 * Modification of indirection ntables having more than 1
-		 * reference unsupported. Intended for standalone indirection
-		 * tables only.
-		 */
-		DRV_LOG(DEBUG,
-			"Port %u cannot modify indirection table (refcnt> 1).",
-			dev->data->port_id);
-		rte_errno = EINVAL;
+	if (mlx5_ind_table_obj_check_standalone(dev, ind_tbl) < 0)
 		return -rte_errno;
-	}
 	for (i = 0; i != queues_n; ++i) {
 		if (!mlx5_rxq_get(dev, queues[i])) {
 			ret = -rte_errno;
@@ -2101,6 +2111,73 @@ mlx5_ind_table_obj_modify(struct rte_eth_dev *dev,
 	return ret;
 }
 
+/**
+ * Attach an indirection table to its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to attach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_attach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_modify(dev, ind_tbl, ind_tbl->queues,
+					ind_tbl->queues_n, true);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_get(dev, ind_tbl->queues[i]);
+	return 0;
+}
+
+/**
+ * Detach an indirection table from its queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param ind_table
+ *   Indirection table to detach.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_ind_table_obj_detach(struct rte_eth_dev *dev,
+			  struct mlx5_ind_table_obj *ind_tbl)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	const unsigned int n = rte_is_power_of_2(ind_tbl->queues_n) ?
+			       log2above(ind_tbl->queues_n) :
+			       log2above(priv->config.ind_table_max_size);
+	unsigned int i;
+	int ret;
+
+	ret = mlx5_ind_table_obj_check_standalone(dev, ind_tbl);
+	if (ret != 0)
+		return ret;
+	MLX5_ASSERT(priv->obj_ops.ind_table_modify);
+	ret = priv->obj_ops.ind_table_modify(dev, n, NULL, 0, ind_tbl);
+	if (ret != 0) {
+		DRV_LOG(ERR, "Port %u could not modify indirect table obj %p",
+			dev->data->port_id, (void *)ind_tbl);
+		return ret;
+	}
+	for (i = 0; i < ind_tbl->queues_n; i++)
+		mlx5_rxq_release(dev, ind_tbl->queues[i]);
+	return ret;
+}
+
 int
 mlx5_hrxq_match_cb(void *tool_ctx __rte_unused, struct mlx5_list_entry *entry,
 		   void *cb_ctx)
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index d916c8addc..ebeeae279e 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -14,6 +14,7 @@
 #include <mlx5_malloc.h>
 
 #include "mlx5.h"
+#include "mlx5_flow.h"
 #include "mlx5_rx.h"
 #include "mlx5_tx.h"
 #include "mlx5_utils.h"
@@ -1162,6 +1163,14 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 	mlx5_rxq_timestamp_set(dev);
 	/* Set a mask and offset of scheduling on timestamp into Tx queues. */
 	mlx5_txq_dynf_timestamp_set(dev);
+	/* Attach indirection table objects detached on port stop. */
+	ret = mlx5_action_handle_attach(dev);
+	if (ret) {
+		DRV_LOG(ERR,
+			"port %u failed to attach indirect actions: %s",
+			dev->data->port_id, rte_strerror(rte_errno));
+		goto error;
+	}
 	/*
 	 * In non-cached mode, it only needs to start the default mreg copy
 	 * action and no flow created by application exists anymore.
@@ -1239,6 +1248,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	/* All RX queue flags will be cleared in the flush interface. */
 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, true);
 	mlx5_flow_meter_rxq_flush(dev);
+	mlx5_action_handle_detach(dev);
 	mlx5_rx_intr_vec_disable(dev);
 	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
 	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
-- 
2.25.1


^ permalink raw reply	[flat|nested] 18+ messages in thread

end of thread, other threads:[~2021-11-02 17:02 UTC | newest]

Thread overview: 18+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <20211005005216.2427489-1-dkozlyuk@nvidia.com>
2021-10-05  0:52 ` [dpdk-stable] [PATCH 3/5] net/mlx5: discover max flow priority using DevX dkozlyuk
2021-10-05  0:52 ` [dpdk-stable] [PATCH 4/5] net/mlx5: create drop queue " dkozlyuk
2021-10-05  0:52 ` [dpdk-stable] [PATCH 5/5] net/mlx5: preserve indirect actions on restart dkozlyuk
     [not found] ` <20211015161822.3099818-1-dkozlyuk@nvidia.com>
2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 3/5] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 4/5] net/mlx5: create drop queue " Dmitry Kozlyuk
2021-10-15 16:18   ` [dpdk-stable] [PATCH v2 5/5] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
     [not found]   ` <20211019123722.3414694-1-dkozlyuk@nvidia.com>
2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
2021-10-19 12:37     ` [dpdk-stable] [PATCH v3 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
     [not found]     ` <20211021063503.3632732-1-dkozlyuk@nvidia.com>
2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
2021-10-21  6:35       ` [dpdk-stable] [PATCH v4 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
     [not found]       ` <20211102135415.944050-1-dkozlyuk@nvidia.com>
2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
2021-11-02 13:54         ` [dpdk-stable] [PATCH v5 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk
     [not found]         ` <20211102170135.959380-1-dkozlyuk@nvidia.com>
2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 4/6] net/mlx5: discover max flow priority using DevX Dmitry Kozlyuk
2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 5/6] net/mlx5: create drop queue " Dmitry Kozlyuk
2021-11-02 17:01           ` [dpdk-stable] [PATCH v6 6/6] net/mlx5: preserve indirect actions on restart Dmitry Kozlyuk

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).