[dpdk-dev] [RFC v1] net/mlx5: support multiple flow priorities

[Xueming Li <xuemingl@mellanox.com>]

This patch supports multiple rte_flow priorities on NICs that support 16
or more Verbs priorities. Control flow always consume 3 Verbs priorities
even in isolated mode.

For NICs that support 16 or more Verbs priorities, each rte_flow
priority consumes 6 Verbs priorities, higher 3 for tunnels, lower 3 for
non-tunnels.

NICs that have 8 Verbs priorities only support 1 user prioirty:
	User tunnel Verbs flows:	0 - 2
	User non-tunnel Verbs flows:	2 - 4
	control Verbs flows:		5 - 8

Signed-off-by: Xueming Li <xuemingl@mellanox.com>
---
 drivers/net/mlx5/mlx5.c      |  8 +++++-
 drivers/net/mlx5/mlx5.h      | 10 +++++--
 drivers/net/mlx5/mlx5_flow.c | 67 ++++++++++++++++++++++++++------------------
 3 files changed, 54 insertions(+), 31 deletions(-)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index c933e274f..6897cc546 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -1189,12 +1189,18 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 		/* Supported Verbs flow priority number detection. */
 		if (verb_priorities == 0)
 			verb_priorities = mlx5_get_max_verbs_prio(eth_dev);
-		if (verb_priorities < MLX5_VERBS_FLOW_PRIO_8) {
+		if (verb_priorities < MLX5_VERBS_FLOW_MIN_PRIOS) {
 			DRV_LOG(ERR, "port %u wrong Verbs flow priorities: %u",
 				eth_dev->data->port_id, verb_priorities);
 			goto port_error;
 		}
 		priv->config.max_verbs_prio = verb_priorities;
+		if (verb_priorities == MLX5_VERBS_FLOW_MIN_PRIOS)
+			priv->config.ctrl_flow_prio = 1;
+		else
+			priv->config.ctrl_flow_prio =
+				(verb_priorities - MLX5_VERBS_FLOW_CTRL_PRIOS) /
+				MLX5_VERBS_FLOW_USER_PRIOS;
 		/*
 		 * Once the device is added to the list of memory event
 		 * callback, its global MR cache table cannot be expanded
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 997b04a33..514501cc3 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -115,6 +115,7 @@ struct mlx5_dev_config {
 		/* Rx queue count threshold to enable MPRQ. */
 	} mprq; /* Configurations for Multi-Packet RQ. */
 	unsigned int max_verbs_prio; /* Number of Verb flow priorities. */
+	unsigned int ctrl_flow_prio; /* Control flow priority. */
 	unsigned int tso_max_payload_sz; /* Maximum TCP payload for TSO. */
 	unsigned int ind_table_max_size; /* Maximum indirection table size. */
 	int txq_inline; /* Maximum packet size for inlining. */
@@ -131,8 +132,13 @@ enum mlx5_verbs_alloc_type {
 	MLX5_VERBS_ALLOC_TYPE_RX_QUEUE,
 };
 
-/* 8 Verbs priorities. */
-#define MLX5_VERBS_FLOW_PRIO_8 8
+/* Minimal hardware Verbs priorities. */
+#define MLX5_VERBS_FLOW_MIN_PRIOS 8
+/* Verbs priorities used by control flow. */
+#define MLX5_VERBS_FLOW_CTRL_PRIOS 3
+/* Verbs priorities used by each user flow. */
+#define MLX5_VERBS_FLOW_USER_PRIOS 6
+
 
 /**
  * Verbs allocator needs a context to know in the callback which kind of
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 994be05be..0e4b6f179 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -31,9 +31,6 @@
 #include "mlx5_prm.h"
 #include "mlx5_glue.h"
 
-/* Flow priority for control plane flows. */
-#define MLX5_CTRL_FLOW_PRIORITY 1
-
 /* Internet Protocol versions. */
 #define MLX5_IPV4 4
 #define MLX5_IPV6 6
@@ -619,6 +616,8 @@ mlx5_flow_item_validate(const struct rte_flow_item *item,
 /**
  * Extract attribute to the parser.
  *
+ * @param dev
+ *   Pointer to Ethernet device.
  * @param[in] attr
  *   Flow rule attributes.
  * @param[out] error
@@ -628,9 +627,12 @@ mlx5_flow_item_validate(const struct rte_flow_item *item,
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 static int
-mlx5_flow_convert_attributes(const struct rte_flow_attr *attr,
+mlx5_flow_convert_attributes(struct rte_eth_dev *dev,
+			     const struct rte_flow_attr *attr,
 			     struct rte_flow_error *error)
 {
+	struct priv *priv = dev->data->dev_private;
+
 	if (attr->group) {
 		rte_flow_error_set(error, ENOTSUP,
 				   RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
@@ -638,11 +640,11 @@ mlx5_flow_convert_attributes(const struct rte_flow_attr *attr,
 				   "groups are not supported");
 		return -rte_errno;
 	}
-	if (attr->priority && attr->priority != MLX5_CTRL_FLOW_PRIORITY) {
+	if (attr->priority > priv->config.ctrl_flow_prio) {
 		rte_flow_error_set(error, ENOTSUP,
 				   RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
 				   NULL,
-				   "priorities are not supported");
+				   "priority not supported");
 		return -rte_errno;
 	}
 	if (attr->egress) {
@@ -1039,23 +1041,33 @@ mlx5_flow_update_priority(struct rte_eth_dev *dev,
 	unsigned int i;
 	uint16_t priority;
 
-	/*			8 priorities	>= 16 priorities
-	 * Control flow:	4-7		8-15
-	 * User normal flow:	1-3		4-7
-	 * User tunnel flow:	0-2		0-3
-	 */
-	priority = attr->priority * MLX5_VERBS_FLOW_PRIO_8;
-	if (priv->config.max_verbs_prio == MLX5_VERBS_FLOW_PRIO_8)
-		priority /= 2;
-	/*
-	 * Lower non-tunnel flow Verbs priority 1 if only support 8 Verbs
-	 * priorities, lower 4 otherwise.
-	 */
-	if (!parser->inner) {
-		if (priv->config.max_verbs_prio == MLX5_VERBS_FLOW_PRIO_8)
-			priority += 1;
+	if (priv->config.max_verbs_prio == MLX5_VERBS_FLOW_MIN_PRIOS) {
+		/*
+		 * Verbs priority mapping:
+		 *    user tunnel flows:	0 - 2
+		 *    user non-tunnel flows:	2 - 4
+		 *    control flows:		5 - 8
+		 */
+		if (attr->priority >= priv->config.ctrl_flow_prio)
+			priority = MLX5_VERBS_FLOW_MIN_PRIOS -
+				   MLX5_VERBS_FLOW_CTRL_PRIOS;
+		else if (!parser->inner)
+			priority = (MLX5_VERBS_FLOW_MIN_PRIOS -
+				     MLX5_VERBS_FLOW_CTRL_PRIOS -
+				     MLX5_VERBS_FLOW_USER_PRIOS / 2);
 		else
-			priority += MLX5_VERBS_FLOW_PRIO_8 / 2;
+			priority = 0;
+	} else {
+		/*
+		 * Verbs priority mapping:
+		 *    user flows * 6, control flows 3.
+		 * User flow mapping:
+		 *    tunnel 3, non-tunnel 3.
+		 */
+		priority = attr->priority * MLX5_VERBS_FLOW_USER_PRIOS;
+		if (attr->priority != priv->config.ctrl_flow_prio &&
+		    !parser->inner)
+			priority += MLX5_VERBS_FLOW_USER_PRIOS / 2;
 	}
 	if (parser->drop) {
 		parser->queue[HASH_RXQ_ETH].ibv_attr->priority = priority +
@@ -1285,7 +1297,7 @@ mlx5_flow_convert(struct rte_eth_dev *dev,
 		.layer = HASH_RXQ_ETH,
 		.mark_id = MLX5_FLOW_MARK_DEFAULT,
 	};
-	ret = mlx5_flow_convert_attributes(attr, error);
+	ret = mlx5_flow_convert_attributes(dev, attr, error);
 	if (ret)
 		return ret;
 	ret = mlx5_flow_convert_actions(dev, actions, error, parser);
@@ -3019,7 +3031,7 @@ mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
 	struct priv *priv = dev->data->dev_private;
 	const struct rte_flow_attr attr = {
 		.ingress = 1,
-		.priority = MLX5_CTRL_FLOW_PRIORITY,
+		.priority = priv->config.ctrl_flow_prio,
 	};
 	struct rte_flow_item items[] = {
 		{
@@ -3739,7 +3751,7 @@ unsigned int
 mlx5_get_max_verbs_prio(struct rte_eth_dev *dev)
 {
 	struct priv *priv = dev->data->dev_private;
-	unsigned int verb_priorities = MLX5_VERBS_FLOW_PRIO_8;
+	unsigned int verb_priorities = MLX5_VERBS_FLOW_MIN_PRIOS;
 	struct {
 		struct ibv_flow_attr attr;
 		struct ibv_flow_spec_eth eth;
@@ -3773,8 +3785,7 @@ mlx5_get_max_verbs_prio(struct rte_eth_dev *dev)
 			break;
 		}
 	} while (1);
-	DRV_LOG(DEBUG, "port %u Verbs flow priorities: %d,"
-		" user flow priorities: %d",
-		dev->data->port_id, verb_priorities, MLX5_CTRL_FLOW_PRIORITY);
+	DRV_LOG(DEBUG, "port %u Verbs flow priorities: %d",
+		dev->data->port_id, verb_priorities);
 	return verb_priorities;
 }
-- 
2.13.3