net/mlx5: separate Rx hash queue creation
authorMichael Baum <michaelba@nvidia.com>
Thu, 3 Sep 2020 10:13:44 +0000 (10:13 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 18 Sep 2020 16:55:08 +0000 (18:55 +0200)
Separate Rx hash queue creation into both Verbs and DevX modules.

Signed-off-by: Michael Baum <michaelba@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
drivers/net/mlx5/linux/mlx5_verbs.c
drivers/net/mlx5/mlx5.h
drivers/net/mlx5/mlx5_devx.c
drivers/net/mlx5/mlx5_flow_dv.c
drivers/net/mlx5/mlx5_flow_verbs.c
drivers/net/mlx5/mlx5_rxq.c
drivers/net/mlx5/mlx5_rxtx.h

index 916c1f1..c281e86 100644 (file)
@@ -516,6 +516,156 @@ mlx5_ibv_ind_table_obj_destroy(struct mlx5_ind_table_obj *ind_tbl)
        claim_zero(mlx5_glue->destroy_rwq_ind_table(ind_tbl->ind_table));
 }
 
+/**
+ * Create an Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param rss_key
+ *   RSS key for the Rx hash queue.
+ * @param rss_key_len
+ *   RSS key length.
+ * @param hash_fields
+ *   Verbs protocol hash field to make the RSS on.
+ * @param queues
+ *   Queues entering in hash queue. In case of empty hash_fields only the
+ *   first queue index will be taken for the indirection table.
+ * @param queues_n
+ *   Number of queues.
+ * @param tunnel
+ *   Tunnel type.
+ *
+ * @return
+ *   The Verbs object initialized index, 0 otherwise and rte_errno is set.
+ */
+static uint32_t
+mlx5_ibv_hrxq_new(struct rte_eth_dev *dev,
+                 const uint8_t *rss_key, uint32_t rss_key_len,
+                 uint64_t hash_fields,
+                 const uint16_t *queues, uint32_t queues_n,
+                 int tunnel __rte_unused)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_hrxq *hrxq = NULL;
+       uint32_t hrxq_idx = 0;
+       struct ibv_qp *qp = NULL;
+       struct mlx5_ind_table_obj *ind_tbl;
+       int err;
+
+       queues_n = hash_fields ? queues_n : 1;
+       ind_tbl = mlx5_ind_table_obj_get(dev, queues, queues_n);
+       if (!ind_tbl)
+               ind_tbl = priv->obj_ops->ind_table_obj_new(dev, queues,
+                                                          queues_n);
+       if (!ind_tbl) {
+               rte_errno = ENOMEM;
+               return 0;
+       }
+#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
+       struct mlx5dv_qp_init_attr qp_init_attr;
+
+       memset(&qp_init_attr, 0, sizeof(qp_init_attr));
+       if (tunnel) {
+               qp_init_attr.comp_mask =
+                                      MLX5DV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS;
+               qp_init_attr.create_flags = MLX5DV_QP_CREATE_TUNNEL_OFFLOADS;
+       }
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+       if (dev->data->dev_conf.lpbk_mode) {
+               /* Allow packet sent from NIC loop back w/o source MAC check. */
+               qp_init_attr.comp_mask |=
+                               MLX5DV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS;
+               qp_init_attr.create_flags |=
+                               MLX5DV_QP_CREATE_TIR_ALLOW_SELF_LOOPBACK_UC;
+       }
+#endif
+       qp = mlx5_glue->dv_create_qp
+                       (priv->sh->ctx,
+                        &(struct ibv_qp_init_attr_ex){
+                               .qp_type = IBV_QPT_RAW_PACKET,
+                               .comp_mask =
+                                       IBV_QP_INIT_ATTR_PD |
+                                       IBV_QP_INIT_ATTR_IND_TABLE |
+                                       IBV_QP_INIT_ATTR_RX_HASH,
+                               .rx_hash_conf = (struct ibv_rx_hash_conf){
+                                       .rx_hash_function =
+                                               IBV_RX_HASH_FUNC_TOEPLITZ,
+                                       .rx_hash_key_len = rss_key_len,
+                                       .rx_hash_key =
+                                               (void *)(uintptr_t)rss_key,
+                                       .rx_hash_fields_mask = hash_fields,
+                               },
+                               .rwq_ind_tbl = ind_tbl->ind_table,
+                               .pd = priv->sh->pd,
+                         },
+                         &qp_init_attr);
+#else
+       qp = mlx5_glue->create_qp_ex
+                       (priv->sh->ctx,
+                        &(struct ibv_qp_init_attr_ex){
+                               .qp_type = IBV_QPT_RAW_PACKET,
+                               .comp_mask =
+                                       IBV_QP_INIT_ATTR_PD |
+                                       IBV_QP_INIT_ATTR_IND_TABLE |
+                                       IBV_QP_INIT_ATTR_RX_HASH,
+                               .rx_hash_conf = (struct ibv_rx_hash_conf){
+                                       .rx_hash_function =
+                                               IBV_RX_HASH_FUNC_TOEPLITZ,
+                                       .rx_hash_key_len = rss_key_len,
+                                       .rx_hash_key =
+                                               (void *)(uintptr_t)rss_key,
+                                       .rx_hash_fields_mask = hash_fields,
+                               },
+                               .rwq_ind_tbl = ind_tbl->ind_table,
+                               .pd = priv->sh->pd,
+                        });
+#endif
+       if (!qp) {
+               rte_errno = errno;
+               goto error;
+       }
+       hrxq = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_HRXQ], &hrxq_idx);
+       if (!hrxq)
+               goto error;
+       hrxq->ind_table = ind_tbl;
+       hrxq->qp = qp;
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+       hrxq->action = mlx5_glue->dv_create_flow_action_dest_ibv_qp(hrxq->qp);
+       if (!hrxq->action) {
+               rte_errno = errno;
+               goto error;
+       }
+#endif
+       hrxq->rss_key_len = rss_key_len;
+       hrxq->hash_fields = hash_fields;
+       memcpy(hrxq->rss_key, rss_key, rss_key_len);
+       rte_atomic32_inc(&hrxq->refcnt);
+       ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_HRXQ], &priv->hrxqs, hrxq_idx,
+                    hrxq, next);
+       return hrxq_idx;
+error:
+       err = rte_errno; /* Save rte_errno before cleanup. */
+       mlx5_ind_table_obj_release(dev, ind_tbl);
+       if (qp)
+               claim_zero(mlx5_glue->destroy_qp(qp));
+       if (hrxq)
+               mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_HRXQ], hrxq_idx);
+       rte_errno = err; /* Restore rte_errno. */
+       return 0;
+}
+
+/**
+ * Destroy a Verbs queue pair.
+ *
+ * @param hrxq
+ *   Hash Rx queue to release its qp.
+ */
+static void
+mlx5_ibv_qp_destroy(struct mlx5_hrxq *hrxq)
+{
+       claim_zero(mlx5_glue->destroy_qp(hrxq->qp));
+}
+
 struct mlx5_obj_ops ibv_obj_ops = {
        .rxq_obj_modify_vlan_strip = mlx5_rxq_obj_modify_wq_vlan_strip,
        .rxq_obj_new = mlx5_rxq_ibv_obj_new,
@@ -524,4 +674,6 @@ struct mlx5_obj_ops ibv_obj_ops = {
        .rxq_obj_release = mlx5_rxq_ibv_obj_release,
        .ind_table_obj_new = mlx5_ibv_ind_table_obj_new,
        .ind_table_obj_destroy = mlx5_ibv_ind_table_obj_destroy,
+       .hrxq_new = mlx5_ibv_hrxq_new,
+       .hrxq_destroy = mlx5_ibv_qp_destroy,
 };
index c151e64..9fc4639 100644 (file)
@@ -723,6 +723,24 @@ struct mlx5_ind_table_obj {
        uint16_t queues[]; /**< Queue list. */
 };
 
+/* Hash Rx queue. */
+struct mlx5_hrxq {
+       ILIST_ENTRY(uint32_t)next; /* Index to the next element. */
+       rte_atomic32_t refcnt; /* Reference counter. */
+       struct mlx5_ind_table_obj *ind_table; /* Indirection table. */
+       RTE_STD_C11
+       union {
+               void *qp; /* Verbs queue pair. */
+               struct mlx5_devx_obj *tir; /* DevX TIR object. */
+       };
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+       void *action; /* DV QP action pointer. */
+#endif
+       uint64_t hash_fields; /* Verbs Hash fields. */
+       uint32_t rss_key_len; /* Hash key length in bytes. */
+       uint8_t rss_key[]; /* Hash key. */
+};
+
 /* HW objects operations structure. */
 struct mlx5_obj_ops {
        int (*rxq_obj_modify_vlan_strip)(struct mlx5_rxq_obj *rxq_obj, int on);
@@ -734,6 +752,11 @@ struct mlx5_obj_ops {
                                                        const uint16_t *queues,
                                                        uint32_t queues_n);
        void (*ind_table_obj_destroy)(struct mlx5_ind_table_obj *ind_tbl);
+       uint32_t (*hrxq_new)(struct rte_eth_dev *dev, const uint8_t *rss_key,
+                            uint32_t rss_key_len, uint64_t hash_fields,
+                            const uint16_t *queues, uint32_t queues_n,
+                            int tunnel __rte_unused);
+       void (*hrxq_destroy)(struct mlx5_hrxq *hrxq);
 };
 
 struct mlx5_priv {
index aab5e50..b1b3037 100644 (file)
@@ -694,6 +694,159 @@ mlx5_devx_ind_table_obj_destroy(struct mlx5_ind_table_obj *ind_tbl)
        claim_zero(mlx5_devx_cmd_destroy(ind_tbl->rqt));
 }
 
+/**
+ * Create an Rx Hash queue.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param rss_key
+ *   RSS key for the Rx hash queue.
+ * @param rss_key_len
+ *   RSS key length.
+ * @param hash_fields
+ *   Verbs protocol hash field to make the RSS on.
+ * @param queues
+ *   Queues entering in hash queue. In case of empty hash_fields only the
+ *   first queue index will be taken for the indirection table.
+ * @param queues_n
+ *   Number of queues.
+ * @param tunnel
+ *   Tunnel type.
+ *
+ * @return
+ *   The DevX object initialized index, 0 otherwise and rte_errno is set.
+ */
+static uint32_t
+mlx5_devx_hrxq_new(struct rte_eth_dev *dev,
+                  const uint8_t *rss_key, uint32_t rss_key_len,
+                  uint64_t hash_fields,
+                  const uint16_t *queues, uint32_t queues_n,
+                  int tunnel __rte_unused)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_hrxq *hrxq = NULL;
+       uint32_t hrxq_idx = 0;
+       struct mlx5_ind_table_obj *ind_tbl;
+       struct mlx5_devx_obj *tir = NULL;
+       struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[queues[0]];
+       struct mlx5_rxq_ctrl *rxq_ctrl =
+               container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+       struct mlx5_devx_tir_attr tir_attr;
+       int err;
+       uint32_t i;
+       bool lro = true;
+
+       queues_n = hash_fields ? queues_n : 1;
+       ind_tbl = mlx5_ind_table_obj_get(dev, queues, queues_n);
+       if (!ind_tbl)
+               ind_tbl = priv->obj_ops->ind_table_obj_new(dev, queues,
+                                                          queues_n);
+       if (!ind_tbl) {
+               rte_errno = ENOMEM;
+               return 0;
+       }
+       /* Enable TIR LRO only if all the queues were configured for. */
+       for (i = 0; i < queues_n; ++i) {
+               if (!(*priv->rxqs)[queues[i]]->lro) {
+                       lro = false;
+                       break;
+               }
+       }
+       memset(&tir_attr, 0, sizeof(tir_attr));
+       tir_attr.disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
+       tir_attr.rx_hash_fn = MLX5_RX_HASH_FN_TOEPLITZ;
+       tir_attr.tunneled_offload_en = !!tunnel;
+       /* If needed, translate hash_fields bitmap to PRM format. */
+       if (hash_fields) {
+               struct mlx5_rx_hash_field_select *rx_hash_field_select = NULL;
+#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
+               rx_hash_field_select = hash_fields & IBV_RX_HASH_INNER ?
+                                      &tir_attr.rx_hash_field_selector_inner :
+                                      &tir_attr.rx_hash_field_selector_outer;
+#else
+               rx_hash_field_select = &tir_attr.rx_hash_field_selector_outer;
+#endif
+               /* 1 bit: 0: IPv4, 1: IPv6. */
+               rx_hash_field_select->l3_prot_type =
+                                       !!(hash_fields & MLX5_IPV6_IBV_RX_HASH);
+               /* 1 bit: 0: TCP, 1: UDP. */
+               rx_hash_field_select->l4_prot_type =
+                                        !!(hash_fields & MLX5_UDP_IBV_RX_HASH);
+               /* Bitmask which sets which fields to use in RX Hash. */
+               rx_hash_field_select->selected_fields =
+                       ((!!(hash_fields & MLX5_L3_SRC_IBV_RX_HASH)) <<
+                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_SRC_IP) |
+                       (!!(hash_fields & MLX5_L3_DST_IBV_RX_HASH)) <<
+                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_DST_IP |
+                       (!!(hash_fields & MLX5_L4_SRC_IBV_RX_HASH)) <<
+                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_L4_SPORT |
+                       (!!(hash_fields & MLX5_L4_DST_IBV_RX_HASH)) <<
+                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_L4_DPORT;
+       }
+       if (rxq_ctrl->type == MLX5_RXQ_TYPE_HAIRPIN)
+               tir_attr.transport_domain = priv->sh->td->id;
+       else
+               tir_attr.transport_domain = priv->sh->tdn;
+       memcpy(tir_attr.rx_hash_toeplitz_key, rss_key, MLX5_RSS_HASH_KEY_LEN);
+       tir_attr.indirect_table = ind_tbl->rqt->id;
+       if (dev->data->dev_conf.lpbk_mode)
+               tir_attr.self_lb_block = MLX5_TIRC_SELF_LB_BLOCK_BLOCK_UNICAST;
+       if (lro) {
+               tir_attr.lro_timeout_period_usecs = priv->config.lro.timeout;
+               tir_attr.lro_max_msg_sz = priv->max_lro_msg_size;
+               tir_attr.lro_enable_mask = MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
+                                          MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO;
+       }
+       tir = mlx5_devx_cmd_create_tir(priv->sh->ctx, &tir_attr);
+       if (!tir) {
+               DRV_LOG(ERR, "Port %u cannot create DevX TIR.",
+                       dev->data->port_id);
+               rte_errno = errno;
+               goto error;
+       }
+       hrxq = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_HRXQ], &hrxq_idx);
+       if (!hrxq)
+               goto error;
+       hrxq->ind_table = ind_tbl;
+       hrxq->tir = tir;
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+       hrxq->action = mlx5_glue->dv_create_flow_action_dest_devx_tir
+                                                              (hrxq->tir->obj);
+       if (!hrxq->action) {
+               rte_errno = errno;
+               goto error;
+       }
+#endif
+       hrxq->rss_key_len = rss_key_len;
+       hrxq->hash_fields = hash_fields;
+       memcpy(hrxq->rss_key, rss_key, rss_key_len);
+       rte_atomic32_inc(&hrxq->refcnt);
+       ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_HRXQ], &priv->hrxqs, hrxq_idx,
+                    hrxq, next);
+       return hrxq_idx;
+error:
+       err = rte_errno; /* Save rte_errno before cleanup. */
+       mlx5_ind_table_obj_release(dev, ind_tbl);
+       if (tir)
+               claim_zero(mlx5_devx_cmd_destroy(tir));
+       if (hrxq)
+               mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_HRXQ], hrxq_idx);
+       rte_errno = err; /* Restore rte_errno. */
+       return 0;
+}
+
+/**
+ * Destroy a DevX TIR object.
+ *
+ * @param hrxq
+ *   Hash Rx queue to release its tir.
+ */
+static void
+mlx5_devx_tir_destroy(struct mlx5_hrxq *hrxq)
+{
+       claim_zero(mlx5_devx_cmd_destroy(hrxq->tir));
+}
+
 struct mlx5_obj_ops devx_obj_ops = {
        .rxq_obj_modify_vlan_strip = mlx5_rxq_obj_modify_rq_vlan_strip,
        .rxq_obj_new = mlx5_rxq_devx_obj_new,
@@ -702,4 +855,6 @@ struct mlx5_obj_ops devx_obj_ops = {
        .rxq_obj_release = mlx5_rxq_devx_obj_release,
        .ind_table_obj_new = mlx5_devx_ind_table_obj_new,
        .ind_table_obj_destroy = mlx5_devx_ind_table_obj_destroy,
+       .hrxq_new = mlx5_devx_hrxq_new,
+       .hrxq_destroy = mlx5_devx_tir_destroy,
 };
index 58358ce..fa41486 100644 (file)
@@ -8949,14 +8949,14 @@ __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
                                                 rss_desc->queue,
                                                 rss_desc->queue_num);
                        if (!hrxq_idx) {
-                               hrxq_idx = mlx5_hrxq_new
+                               hrxq_idx = priv->obj_ops->hrxq_new
                                                (dev, rss_desc->key,
-                                               MLX5_RSS_HASH_KEY_LEN,
-                                               dev_flow->hash_fields,
-                                               rss_desc->queue,
-                                               rss_desc->queue_num,
-                                               !!(dh->layers &
-                                               MLX5_FLOW_LAYER_TUNNEL));
+                                                MLX5_RSS_HASH_KEY_LEN,
+                                                dev_flow->hash_fields,
+                                                rss_desc->queue,
+                                                rss_desc->queue_num,
+                                                !!(dh->layers &
+                                                MLX5_FLOW_LAYER_TUNNEL));
                        }
                        hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
                                              hrxq_idx);
index 80c549a..f8edae1 100644 (file)
@@ -1986,13 +1986,14 @@ flow_verbs_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
                                                 rss_desc->queue,
                                                 rss_desc->queue_num);
                        if (!hrxq_idx)
-                               hrxq_idx = mlx5_hrxq_new(dev, rss_desc->key,
-                                               MLX5_RSS_HASH_KEY_LEN,
-                                               dev_flow->hash_fields,
-                                               rss_desc->queue,
-                                               rss_desc->queue_num,
-                                               !!(handle->layers &
-                                               MLX5_FLOW_LAYER_TUNNEL));
+                               hrxq_idx = priv->obj_ops->hrxq_new
+                                               (dev, rss_desc->key,
+                                                MLX5_RSS_HASH_KEY_LEN,
+                                                dev_flow->hash_fields,
+                                                rss_desc->queue,
+                                                rss_desc->queue_num,
+                                                !!(handle->layers &
+                                                MLX5_FLOW_LAYER_TUNNEL));
                        hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
                                         hrxq_idx);
                        if (!hrxq) {
index aa39892..d84dfe1 100644 (file)
@@ -20,7 +20,6 @@
 #include <rte_eal_paging.h>
 
 #include <mlx5_glue.h>
-#include <mlx5_devx_cmds.h>
 #include <mlx5_malloc.h>
 
 #include "mlx5_defs.h"
@@ -28,7 +27,6 @@
 #include "mlx5_rxtx.h"
 #include "mlx5_utils.h"
 #include "mlx5_autoconf.h"
-#include "mlx5_flow.h"
 
 
 /* Default RSS hash key also used for ConnectX-3. */
@@ -1721,7 +1719,7 @@ mlx5_rxq_get_type(struct rte_eth_dev *dev, uint16_t idx)
  * @return
  *   An indirection table if found.
  */
-static struct mlx5_ind_table_obj *
+struct mlx5_ind_table_obj *
 mlx5_ind_table_obj_get(struct rte_eth_dev *dev, const uint16_t *queues,
                       uint32_t queues_n)
 {
@@ -1756,7 +1754,7 @@ mlx5_ind_table_obj_get(struct rte_eth_dev *dev, const uint16_t *queues,
  * @return
  *   1 while a reference on it exists, 0 when freed.
  */
-static int
+int
 mlx5_ind_table_obj_release(struct rte_eth_dev *dev,
                           struct mlx5_ind_table_obj *ind_tbl)
 {
@@ -1800,238 +1798,6 @@ mlx5_ind_table_obj_verify(struct rte_eth_dev *dev)
        return ret;
 }
 
-/**
- * Create an Rx Hash queue.
- *
- * @param dev
- *   Pointer to Ethernet device.
- * @param rss_key
- *   RSS key for the Rx hash queue.
- * @param rss_key_len
- *   RSS key length.
- * @param hash_fields
- *   Verbs protocol hash field to make the RSS on.
- * @param queues
- *   Queues entering in hash queue. In case of empty hash_fields only the
- *   first queue index will be taken for the indirection table.
- * @param queues_n
- *   Number of queues.
- * @param tunnel
- *   Tunnel type.
- *
- * @return
- *   The Verbs/DevX object initialised index, 0 otherwise and rte_errno is set.
- */
-uint32_t
-mlx5_hrxq_new(struct rte_eth_dev *dev,
-             const uint8_t *rss_key, uint32_t rss_key_len,
-             uint64_t hash_fields,
-             const uint16_t *queues, uint32_t queues_n,
-             int tunnel __rte_unused)
-{
-       struct mlx5_priv *priv = dev->data->dev_private;
-       struct mlx5_hrxq *hrxq = NULL;
-       uint32_t hrxq_idx = 0;
-       struct ibv_qp *qp = NULL;
-       struct mlx5_ind_table_obj *ind_tbl;
-       int err;
-       struct mlx5_devx_obj *tir = NULL;
-       struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[queues[0]];
-       struct mlx5_rxq_ctrl *rxq_ctrl =
-               container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
-
-       queues_n = hash_fields ? queues_n : 1;
-       ind_tbl = mlx5_ind_table_obj_get(dev, queues, queues_n);
-       if (!ind_tbl)
-               ind_tbl = priv->obj_ops->ind_table_obj_new(dev, queues,
-                                                          queues_n);
-       if (!ind_tbl) {
-               rte_errno = ENOMEM;
-               return 0;
-       }
-       if (ind_tbl->type == MLX5_IND_TBL_TYPE_IBV) {
-#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
-               struct mlx5dv_qp_init_attr qp_init_attr;
-
-               memset(&qp_init_attr, 0, sizeof(qp_init_attr));
-               if (tunnel) {
-                       qp_init_attr.comp_mask =
-                               MLX5DV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS;
-                       qp_init_attr.create_flags =
-                               MLX5DV_QP_CREATE_TUNNEL_OFFLOADS;
-               }
-#ifdef HAVE_IBV_FLOW_DV_SUPPORT
-               if (dev->data->dev_conf.lpbk_mode) {
-                       /*
-                        * Allow packet sent from NIC loop back
-                        * w/o source MAC check.
-                        */
-                       qp_init_attr.comp_mask |=
-                               MLX5DV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS;
-                       qp_init_attr.create_flags |=
-                               MLX5DV_QP_CREATE_TIR_ALLOW_SELF_LOOPBACK_UC;
-               }
-#endif
-               qp = mlx5_glue->dv_create_qp
-                       (priv->sh->ctx,
-                        &(struct ibv_qp_init_attr_ex){
-                               .qp_type = IBV_QPT_RAW_PACKET,
-                               .comp_mask =
-                                       IBV_QP_INIT_ATTR_PD |
-                                       IBV_QP_INIT_ATTR_IND_TABLE |
-                                       IBV_QP_INIT_ATTR_RX_HASH,
-                               .rx_hash_conf = (struct ibv_rx_hash_conf){
-                                       .rx_hash_function =
-                                               IBV_RX_HASH_FUNC_TOEPLITZ,
-                                       .rx_hash_key_len = rss_key_len,
-                                       .rx_hash_key =
-                                               (void *)(uintptr_t)rss_key,
-                                       .rx_hash_fields_mask = hash_fields,
-                               },
-                               .rwq_ind_tbl = ind_tbl->ind_table,
-                               .pd = priv->sh->pd,
-                         },
-                         &qp_init_attr);
-#else
-               qp = mlx5_glue->create_qp_ex
-                       (priv->sh->ctx,
-                        &(struct ibv_qp_init_attr_ex){
-                               .qp_type = IBV_QPT_RAW_PACKET,
-                               .comp_mask =
-                                       IBV_QP_INIT_ATTR_PD |
-                                       IBV_QP_INIT_ATTR_IND_TABLE |
-                                       IBV_QP_INIT_ATTR_RX_HASH,
-                               .rx_hash_conf = (struct ibv_rx_hash_conf){
-                                       .rx_hash_function =
-                                               IBV_RX_HASH_FUNC_TOEPLITZ,
-                                       .rx_hash_key_len = rss_key_len,
-                                       .rx_hash_key =
-                                               (void *)(uintptr_t)rss_key,
-                                       .rx_hash_fields_mask = hash_fields,
-                               },
-                               .rwq_ind_tbl = ind_tbl->ind_table,
-                               .pd = priv->sh->pd,
-                        });
-#endif
-               if (!qp) {
-                       rte_errno = errno;
-                       goto error;
-               }
-       } else { /* ind_tbl->type == MLX5_IND_TBL_TYPE_DEVX */
-               struct mlx5_devx_tir_attr tir_attr;
-               uint32_t i;
-               uint32_t lro = 1;
-
-               /* Enable TIR LRO only if all the queues were configured for. */
-               for (i = 0; i < queues_n; ++i) {
-                       if (!(*priv->rxqs)[queues[i]]->lro) {
-                               lro = 0;
-                               break;
-                       }
-               }
-               memset(&tir_attr, 0, sizeof(tir_attr));
-               tir_attr.disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
-               tir_attr.rx_hash_fn = MLX5_RX_HASH_FN_TOEPLITZ;
-               tir_attr.tunneled_offload_en = !!tunnel;
-               /* If needed, translate hash_fields bitmap to PRM format. */
-               if (hash_fields) {
-#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
-                       struct mlx5_rx_hash_field_select *rx_hash_field_select =
-                                       hash_fields & IBV_RX_HASH_INNER ?
-                                       &tir_attr.rx_hash_field_selector_inner :
-                                       &tir_attr.rx_hash_field_selector_outer;
-#else
-                       struct mlx5_rx_hash_field_select *rx_hash_field_select =
-                                       &tir_attr.rx_hash_field_selector_outer;
-#endif
-                       /* 1 bit: 0: IPv4, 1: IPv6. */
-                       rx_hash_field_select->l3_prot_type =
-                               !!(hash_fields & MLX5_IPV6_IBV_RX_HASH);
-                       /* 1 bit: 0: TCP, 1: UDP. */
-                       rx_hash_field_select->l4_prot_type =
-                               !!(hash_fields & MLX5_UDP_IBV_RX_HASH);
-                       /* Bitmask which sets which fields to use in RX Hash. */
-                       rx_hash_field_select->selected_fields =
-                       ((!!(hash_fields & MLX5_L3_SRC_IBV_RX_HASH)) <<
-                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_SRC_IP) |
-                       (!!(hash_fields & MLX5_L3_DST_IBV_RX_HASH)) <<
-                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_DST_IP |
-                       (!!(hash_fields & MLX5_L4_SRC_IBV_RX_HASH)) <<
-                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_L4_SPORT |
-                       (!!(hash_fields & MLX5_L4_DST_IBV_RX_HASH)) <<
-                        MLX5_RX_HASH_FIELD_SELECT_SELECTED_FIELDS_L4_DPORT;
-               }
-               if (rxq_ctrl->obj->type == MLX5_RXQ_OBJ_TYPE_DEVX_HAIRPIN)
-                       tir_attr.transport_domain = priv->sh->td->id;
-               else
-                       tir_attr.transport_domain = priv->sh->tdn;
-               memcpy(tir_attr.rx_hash_toeplitz_key, rss_key,
-                      MLX5_RSS_HASH_KEY_LEN);
-               tir_attr.indirect_table = ind_tbl->rqt->id;
-               if (dev->data->dev_conf.lpbk_mode)
-                       tir_attr.self_lb_block =
-                                       MLX5_TIRC_SELF_LB_BLOCK_BLOCK_UNICAST;
-               if (lro) {
-                       tir_attr.lro_timeout_period_usecs =
-                                       priv->config.lro.timeout;
-                       tir_attr.lro_max_msg_sz = priv->max_lro_msg_size;
-                       tir_attr.lro_enable_mask =
-                                       MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO |
-                                       MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO;
-               }
-               tir = mlx5_devx_cmd_create_tir(priv->sh->ctx, &tir_attr);
-               if (!tir) {
-                       DRV_LOG(ERR, "port %u cannot create DevX TIR",
-                               dev->data->port_id);
-                       rte_errno = errno;
-                       goto error;
-               }
-       }
-       hrxq = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_HRXQ], &hrxq_idx);
-       if (!hrxq)
-               goto error;
-       hrxq->ind_table = ind_tbl;
-       if (ind_tbl->type == MLX5_IND_TBL_TYPE_IBV) {
-               hrxq->qp = qp;
-#ifdef HAVE_IBV_FLOW_DV_SUPPORT
-               hrxq->action =
-                       mlx5_glue->dv_create_flow_action_dest_ibv_qp(hrxq->qp);
-               if (!hrxq->action) {
-                       rte_errno = errno;
-                       goto error;
-               }
-#endif
-       } else { /* ind_tbl->type == MLX5_IND_TBL_TYPE_DEVX */
-               hrxq->tir = tir;
-#ifdef HAVE_IBV_FLOW_DV_SUPPORT
-               hrxq->action = mlx5_glue->dv_create_flow_action_dest_devx_tir
-                                                       (hrxq->tir->obj);
-               if (!hrxq->action) {
-                       rte_errno = errno;
-                       goto error;
-               }
-#endif
-       }
-       hrxq->rss_key_len = rss_key_len;
-       hrxq->hash_fields = hash_fields;
-       memcpy(hrxq->rss_key, rss_key, rss_key_len);
-       rte_atomic32_inc(&hrxq->refcnt);
-       ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_HRXQ], &priv->hrxqs, hrxq_idx,
-                    hrxq, next);
-       return hrxq_idx;
-error:
-       err = rte_errno; /* Save rte_errno before cleanup. */
-       mlx5_ind_table_obj_release(dev, ind_tbl);
-       if (qp)
-               claim_zero(mlx5_glue->destroy_qp(qp));
-       else if (tir)
-               claim_zero(mlx5_devx_cmd_destroy(tir));
-       if (hrxq)
-               mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_HRXQ], hrxq_idx);
-       rte_errno = err; /* Restore rte_errno. */
-       return 0;
-}
-
 /**
  * Get an Rx Hash queue.
  *
@@ -2106,10 +1872,7 @@ mlx5_hrxq_release(struct rte_eth_dev *dev, uint32_t hrxq_idx)
 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
                mlx5_glue->destroy_flow_action(hrxq->action);
 #endif
-               if (hrxq->ind_table->type == MLX5_IND_TBL_TYPE_IBV)
-                       claim_zero(mlx5_glue->destroy_qp(hrxq->qp));
-               else /* hrxq->ind_table->type == MLX5_IND_TBL_TYPE_DEVX */
-                       claim_zero(mlx5_devx_cmd_destroy(hrxq->tir));
+               priv->obj_ops->hrxq_destroy(hrxq);
                mlx5_ind_table_obj_release(dev, hrxq->ind_table);
                ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_HRXQ], &priv->hrxqs,
                             hrxq_idx, hrxq, next);
index 7878c81..14a3535 100644 (file)
@@ -186,24 +186,6 @@ struct mlx5_rxq_ctrl {
        struct rte_eth_hairpin_conf hairpin_conf; /* Hairpin configuration. */
 };
 
-/* Hash Rx queue. */
-struct mlx5_hrxq {
-       ILIST_ENTRY(uint32_t)next; /* Index to the next element. */
-       rte_atomic32_t refcnt; /* Reference counter. */
-       struct mlx5_ind_table_obj *ind_table; /* Indirection table. */
-       RTE_STD_C11
-       union {
-               void *qp; /* Verbs queue pair. */
-               struct mlx5_devx_obj *tir; /* DevX TIR object. */
-       };
-#ifdef HAVE_IBV_FLOW_DV_SUPPORT
-       void *action; /* DV QP action pointer. */
-#endif
-       uint64_t hash_fields; /* Verbs Hash fields. */
-       uint32_t rss_key_len; /* Hash key length in bytes. */
-       uint8_t rss_key[]; /* Hash key. */
-};
-
 /* TX queue send local data. */
 __extension__
 struct mlx5_txq_local {
@@ -383,11 +365,11 @@ int mlx5_rxq_release(struct rte_eth_dev *dev, uint16_t idx);
 int mlx5_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);
-uint32_t mlx5_hrxq_new(struct rte_eth_dev *dev,
-                      const uint8_t *rss_key, uint32_t rss_key_len,
-                      uint64_t hash_fields,
-                      const uint16_t *queues, uint32_t queues_n,
-                      int tunnel __rte_unused);
+struct mlx5_ind_table_obj *mlx5_ind_table_obj_get(struct rte_eth_dev *dev,
+                                                 const uint16_t *queues,
+                                                 uint32_t queues_n);
+int mlx5_ind_table_obj_release(struct rte_eth_dev *dev,
+                              struct mlx5_ind_table_obj *ind_tbl);
 uint32_t mlx5_hrxq_get(struct rte_eth_dev *dev,
                       const uint8_t *rss_key, uint32_t rss_key_len,
                       uint64_t hash_fields,