[PATCH v2 6/6] net/mlx5: support queue/RSS action for external RxQ

[Michael Baum <michaelba@nvidia.com>]

Add support queue/RSS action for external RxQ.
In indirection table creation, the queue index will be taken from
mapping array.

This feature supports neither LRO nor Hairpin.

Signed-off-by: Michael Baum <michaelba@nvidia.com>
---
 doc/guides/nics/mlx5.rst               |   1 +
 doc/guides/rel_notes/release_22_03.rst |   1 +
 drivers/net/mlx5/mlx5.c                |   8 +-
 drivers/net/mlx5/mlx5_devx.c           |  30 +++++--
 drivers/net/mlx5/mlx5_flow.c           |  29 +++++--
 drivers/net/mlx5/mlx5_rx.h             |  30 +++++++
 drivers/net/mlx5/mlx5_rxq.c            | 116 +++++++++++++++++++++++--
 7 files changed, 189 insertions(+), 26 deletions(-)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 8956cd1dd8..34be031360 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -38,6 +38,7 @@ Features
 - Multiple TX and RX queues.
 - Shared Rx queue.
 - Rx queue delay drop.
+- Support steering for external Rx queue created outside the PMD.
 - Support for scattered TX frames.
 - Advanced support for scattered Rx frames with tunable buffer attributes.
 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
diff --git a/doc/guides/rel_notes/release_22_03.rst b/doc/guides/rel_notes/release_22_03.rst
index acd56e0a80..c093616d7f 100644
--- a/doc/guides/rel_notes/release_22_03.rst
+++ b/doc/guides/rel_notes/release_22_03.rst
@@ -123,6 +123,7 @@ New Features
   Updated the Mellanox mlx5 driver with new features and improvements, including:
 
   * Support ConnectX-7 capability to schedule traffic sending on timestamp
+  * Support steering for external Rx queue created outside the PMD.
 
 * **Updated Wangxun ngbe driver.**
 
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 415e0fe2f2..9760f52b46 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -1855,8 +1855,6 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 		close(priv->nl_socket_rdma);
 	if (priv->vmwa_context)
 		mlx5_vlan_vmwa_exit(priv->vmwa_context);
-	if (priv->ext_rxqs)
-		mlx5_free(priv->ext_rxqs);
 	ret = mlx5_hrxq_verify(dev);
 	if (ret)
 		DRV_LOG(WARNING, "port %u some hash Rx queue still remain",
@@ -1869,6 +1867,10 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 	if (ret)
 		DRV_LOG(WARNING, "port %u some Rx queue objects still remain",
 			dev->data->port_id);
+	ret = mlx5_ext_rxq_verify(dev);
+	if (ret)
+		DRV_LOG(WARNING, "Port %u some external RxQ still remain.",
+			dev->data->port_id);
 	ret = mlx5_rxq_verify(dev);
 	if (ret)
 		DRV_LOG(WARNING, "port %u some Rx queues still remain",
@@ -1887,6 +1889,8 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 			dev->data->port_id);
 	if (priv->hrxqs)
 		mlx5_list_destroy(priv->hrxqs);
+	if (priv->ext_rxqs)
+		mlx5_free(priv->ext_rxqs);
 	/*
 	 * Free the shared context in last turn, because the cleanup
 	 * routines above may use some shared fields, like
diff --git a/drivers/net/mlx5/mlx5_devx.c b/drivers/net/mlx5/mlx5_devx.c
index e4bc90a30e..8aa68d9658 100644
--- a/drivers/net/mlx5/mlx5_devx.c
+++ b/drivers/net/mlx5/mlx5_devx.c
@@ -580,13 +580,21 @@ mlx5_devx_ind_table_create_rqt_attr(struct rte_eth_dev *dev,
 		return rqt_attr;
 	}
 	for (i = 0; i != queues_n; ++i) {
-		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, queues[i]);
+		if (mlx5_is_external_rxq(dev, queues[i])) {
+			struct mlx5_external_rxq *ext_rxq =
+					mlx5_ext_rxq_get(dev, queues[i]);
 
-		MLX5_ASSERT(rxq != NULL);
-		if (rxq->ctrl->is_hairpin)
-			rqt_attr->rq_list[i] = rxq->ctrl->obj->rq->id;
-		else
-			rqt_attr->rq_list[i] = rxq->devx_rq.rq->id;
+			rqt_attr->rq_list[i] = ext_rxq->hw_id;
+		} else {
+			struct mlx5_rxq_priv *rxq =
+					mlx5_rxq_get(dev, queues[i]);
+
+			MLX5_ASSERT(rxq != NULL);
+			if (rxq->ctrl->is_hairpin)
+				rqt_attr->rq_list[i] = rxq->ctrl->obj->rq->id;
+			else
+				rqt_attr->rq_list[i] = rxq->devx_rq.rq->id;
+		}
 	}
 	MLX5_ASSERT(i > 0);
 	for (j = 0; i != rqt_n; ++j, ++i)
@@ -711,7 +719,13 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 	uint32_t i;
 
 	/* NULL queues designate drop queue. */
-	if (ind_tbl->queues != NULL) {
+	if (ind_tbl->queues == NULL) {
+		is_hairpin = priv->drop_queue.rxq->ctrl->is_hairpin;
+	} else if (mlx5_is_external_rxq(dev, ind_tbl->queues[0])) {
+		/* External RxQ supports neither Hairpin nor LRO. */
+		is_hairpin = false;
+		lro = false;
+	} else {
 		is_hairpin = mlx5_rxq_is_hairpin(dev, ind_tbl->queues[0]);
 		/* Enable TIR LRO only if all the queues were configured for. */
 		for (i = 0; i < ind_tbl->queues_n; ++i) {
@@ -723,8 +737,6 @@ mlx5_devx_tir_attr_set(struct rte_eth_dev *dev, const uint8_t *rss_key,
 				break;
 			}
 		}
-	} else {
-		is_hairpin = priv->drop_queue.rxq->ctrl->is_hairpin;
 	}
 	memset(tir_attr, 0, sizeof(*tir_attr));
 	tir_attr->disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 58f0aba294..96f3402418 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -1631,6 +1631,12 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
 					  "can't have 2 fate actions in"
 					  " same flow");
+	if (attr->egress)
+		return rte_flow_error_set(error, ENOTSUP,
+					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
+					  "queue action not supported for egress.");
+	if (mlx5_is_external_rxq(dev, queue->index))
+		return 0;
 	if (!priv->rxqs_n)
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
@@ -1645,11 +1651,6 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  &queue->index,
 					  "queue is not configured");
-	if (attr->egress)
-		return rte_flow_error_set(error, ENOTSUP,
-					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
-					  "queue action not supported for "
-					  "egress");
 	return 0;
 }
 
@@ -1664,7 +1665,7 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
  *   Size of the @p queues array.
  * @param[out] error
  *   On error, filled with a textual error description.
- * @param[out] queue
+ * @param[out] queue_idx
  *   On error, filled with an offending queue index in @p queues array.
  *
  * @return
@@ -1677,17 +1678,27 @@ mlx5_validate_rss_queues(struct rte_eth_dev *dev,
 {
 	const struct mlx5_priv *priv = dev->data->dev_private;
 	bool is_hairpin = false;
+	bool is_ext_rss = false;
 	uint32_t i;
 
 	for (i = 0; i != queues_n; ++i) {
-		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
-								   queues[i]);
+		struct mlx5_rxq_ctrl *rxq_ctrl;
 
+		if (mlx5_is_external_rxq(dev, queues[0])) {
+			is_ext_rss = true;
+			continue;
+		}
+		if (is_ext_rss) {
+			*error = "Combining external and regular RSS queues is not supported";
+			*queue_idx = i;
+			return -ENOTSUP;
+		}
 		if (queues[i] >= priv->rxqs_n) {
 			*error = "queue index out of range";
 			*queue_idx = i;
 			return -EINVAL;
 		}
+		rxq_ctrl = mlx5_rxq_ctrl_get(dev, queues[i]);
 		if (rxq_ctrl == NULL) {
 			*error =  "queue is not configured";
 			*queue_idx = i;
@@ -1782,7 +1793,7 @@ mlx5_validate_action_rss(struct rte_eth_dev *dev,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
 					  "L4 partial RSS requested but L4 RSS"
 					  " type not specified");
-	if (!priv->rxqs_n)
+	if (!priv->rxqs_n && priv->ext_rxqs == NULL)
 		return rte_flow_error_set(error, EINVAL,
 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
 					  NULL, "No Rx queues configured");
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index 754c526464..29652a8c9f 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -18,6 +18,7 @@
 
 #include "mlx5.h"
 #include "mlx5_autoconf.h"
+#include "rte_pmd_mlx5.h"
 
 /* Support tunnel matching. */
 #define MLX5_FLOW_TUNNEL 10
@@ -217,8 +218,14 @@ uint32_t mlx5_rxq_deref(struct rte_eth_dev *dev, uint16_t idx);
 struct mlx5_rxq_priv *mlx5_rxq_get(struct rte_eth_dev *dev, uint16_t idx);
 struct mlx5_rxq_ctrl *mlx5_rxq_ctrl_get(struct rte_eth_dev *dev, uint16_t idx);
 struct mlx5_rxq_data *mlx5_rxq_data_get(struct rte_eth_dev *dev, uint16_t idx);
+struct mlx5_external_rxq *mlx5_ext_rxq_ref(struct rte_eth_dev *dev,
+					   uint16_t idx);
+uint32_t mlx5_ext_rxq_deref(struct rte_eth_dev *dev, uint16_t idx);
+struct mlx5_external_rxq *mlx5_ext_rxq_get(struct rte_eth_dev *dev,
+					   uint16_t idx);
 int mlx5_rxq_release(struct rte_eth_dev *dev, uint16_t idx);
 int mlx5_rxq_verify(struct rte_eth_dev *dev);
+int mlx5_ext_rxq_verify(struct rte_eth_dev *dev);
 int rxq_alloc_elts(struct mlx5_rxq_ctrl *rxq_ctrl);
 int mlx5_ind_table_obj_verify(struct rte_eth_dev *dev);
 struct mlx5_ind_table_obj *mlx5_ind_table_obj_get(struct rte_eth_dev *dev,
@@ -638,4 +645,27 @@ mlx5_mprq_enabled(struct rte_eth_dev *dev)
 	return n == n_ibv;
 }
 
+/**
+ * Check whether given RxQ is external.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param queue_idx
+ *   Rx queue index.
+ *
+ * @return
+ *   True if is external RxQ, otherwise false.
+ */
+static __rte_always_inline bool
+mlx5_is_external_rxq(struct rte_eth_dev *dev, uint16_t queue_idx)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_external_rxq *rxq;
+
+	if (!priv->ext_rxqs || queue_idx < MLX5_EXTERNAL_RX_QUEUE_ID_MIN)
+		return false;
+	rxq = &priv->ext_rxqs[queue_idx - MLX5_EXTERNAL_RX_QUEUE_ID_MIN];
+	return !!__atomic_load_n(&rxq->refcnt, __ATOMIC_RELAXED);
+}
+
 #endif /* RTE_PMD_MLX5_RX_H_ */
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 145da2dbbb..22679755a4 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -2084,6 +2084,65 @@ mlx5_rxq_data_get(struct rte_eth_dev *dev, uint16_t idx)
 	return rxq == NULL ? NULL : &rxq->ctrl->rxq;
 }
 
+/**
+ * Increase an external Rx queue reference count.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param idx
+ *   External RX queue index.
+ *
+ * @return
+ *   A pointer to the queue if it exists, NULL otherwise.
+ */
+struct mlx5_external_rxq *
+mlx5_ext_rxq_ref(struct rte_eth_dev *dev, uint16_t idx)
+{
+	struct mlx5_external_rxq *rxq = mlx5_ext_rxq_get(dev, idx);
+
+	__atomic_fetch_add(&rxq->refcnt, 1, __ATOMIC_RELAXED);
+	return rxq;
+}
+
+/**
+ * Decrease an external Rx queue reference count.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param idx
+ *   External RX queue index.
+ *
+ * @return
+ *   Updated reference count.
+ */
+uint32_t
+mlx5_ext_rxq_deref(struct rte_eth_dev *dev, uint16_t idx)
+{
+	struct mlx5_external_rxq *rxq = mlx5_ext_rxq_get(dev, idx);
+
+	return __atomic_sub_fetch(&rxq->refcnt, 1, __ATOMIC_RELAXED);
+}
+
+/**
+ * Get an external Rx queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param idx
+ *   External Rx queue index.
+ *
+ * @return
+ *   A pointer to the queue if it exists, NULL otherwise.
+ */
+struct mlx5_external_rxq *
+mlx5_ext_rxq_get(struct rte_eth_dev *dev, uint16_t idx)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	MLX5_ASSERT(mlx5_is_external_rxq(dev, idx));
+	return &priv->ext_rxqs[idx - MLX5_EXTERNAL_RX_QUEUE_ID_MIN];
+}
+
 /**
  * Release a Rx queue.
  *
@@ -2167,6 +2226,37 @@ mlx5_rxq_verify(struct rte_eth_dev *dev)
 	return ret;
 }
 
+/**
+ * Verify the external Rx Queue list is empty.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ *
+ * @return
+ *   The number of object not released.
+ */
+int
+mlx5_ext_rxq_verify(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_external_rxq *rxq;
+	uint16_t i;
+	int ret = 0;
+
+	if (priv->ext_rxqs == NULL)
+		return 0;
+
+	for (i = MLX5_EXTERNAL_RX_QUEUE_ID_MIN; i <= UINT16_MAX ; ++i) {
+		rxq = mlx5_ext_rxq_get(dev, i);
+		if (rxq->refcnt < 2)
+			continue;
+		DRV_LOG(DEBUG, "Port %u external RxQ %u still referenced.",
+			dev->data->port_id, i);
+		++ret;
+	}
+	return ret;
+}
+
 /**
  * Check whether RxQ type is Hairpin.
  *
@@ -2182,8 +2272,11 @@ bool
 mlx5_rxq_is_hairpin(struct rte_eth_dev *dev, uint16_t idx)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
+	struct mlx5_rxq_ctrl *rxq_ctrl;
 
+	if (mlx5_is_external_rxq(dev, idx))
+		return false;
+	rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
 	return (idx < priv->rxqs_n && rxq_ctrl != NULL && rxq_ctrl->is_hairpin);
 }
 
@@ -2361,9 +2454,16 @@ mlx5_ind_table_obj_setup(struct rte_eth_dev *dev,
 
 	if (ref_qs)
 		for (i = 0; i != queues_n; ++i) {
-			if (mlx5_rxq_ref(dev, queues[i]) == NULL) {
-				ret = -rte_errno;
-				goto error;
+			if (mlx5_is_external_rxq(dev, queues[i])) {
+				if (mlx5_ext_rxq_ref(dev, queues[i]) == NULL) {
+					ret = -rte_errno;
+					goto error;
+				}
+			} else {
+				if (mlx5_rxq_ref(dev, queues[i]) == NULL) {
+					ret = -rte_errno;
+					goto error;
+				}
 			}
 		}
 	ret = priv->obj_ops.ind_table_new(dev, n, ind_tbl);
@@ -2374,8 +2474,12 @@ mlx5_ind_table_obj_setup(struct rte_eth_dev *dev,
 error:
 	if (ref_qs) {
 		err = rte_errno;
-		for (j = 0; j < i; j++)
-			mlx5_rxq_deref(dev, queues[j]);
+		for (j = 0; j < i; j++) {
+			if (mlx5_is_external_rxq(dev, queues[j]))
+				mlx5_ext_rxq_deref(dev, queues[j]);
+			else
+				mlx5_rxq_deref(dev, queues[j]);
+		}
 		rte_errno = err;
 	}
 	DRV_LOG(DEBUG, "Port %u cannot setup indirection table.",
-- 
2.25.1