net/mlx5: refactor flow structure
authorViacheslav Ovsiienko <viacheslavo@mellanox.com>
Thu, 7 Nov 2019 17:09:49 +0000 (17:09 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 11 Nov 2019 13:23:01 +0000 (14:23 +0100)
Some rte_flow fields which are local to subflows have been moved to
mlx5_flow structure. RSS attributes are grouped by mlx5_flow_rss structure.
tag_resource is moved to mlx5_flow_dv structure.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
drivers/net/mlx5/mlx5_flow.c
drivers/net/mlx5/mlx5_flow.h
drivers/net/mlx5/mlx5_flow_dv.c
drivers/net/mlx5/mlx5_flow_verbs.c

index bcefc1b..d7854be 100644 (file)
@@ -616,7 +616,7 @@ flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
        unsigned int i;
 
        for (i = 0; i != flow->rss.queue_num; ++i) {
-               int idx = (*flow->queue)[i];
+               int idx = (*flow->rss.queue)[i];
                struct mlx5_rxq_ctrl *rxq_ctrl =
                        container_of((*priv->rxqs)[idx],
                                     struct mlx5_rxq_ctrl, rxq);
@@ -680,7 +680,7 @@ flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
 
        assert(dev->data->dev_started);
        for (i = 0; i != flow->rss.queue_num; ++i) {
-               int idx = (*flow->queue)[i];
+               int idx = (*flow->rss.queue)[i];
                struct mlx5_rxq_ctrl *rxq_ctrl =
                        container_of((*priv->rxqs)[idx],
                                     struct mlx5_rxq_ctrl, rxq);
@@ -2833,13 +2833,20 @@ flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
                goto error_before_flow;
        }
        flow->drv_type = flow_get_drv_type(dev, attr);
-       flow->ingress = attr->ingress;
-       flow->transfer = attr->transfer;
        if (hairpin_id != 0)
                flow->hairpin_flow_id = hairpin_id;
        assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
               flow->drv_type < MLX5_FLOW_TYPE_MAX);
-       flow->queue = (void *)(flow + 1);
+       flow->rss.queue = (void *)(flow + 1);
+       if (rss) {
+               /*
+                * The following information is required by
+                * mlx5_flow_hashfields_adjust() in advance.
+                */
+               flow->rss.level = rss->level;
+               /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
+               flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
+       }
        LIST_INIT(&flow->dev_flows);
        if (rss && rss->types) {
                unsigned int graph_root;
@@ -2879,6 +2886,7 @@ flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
                if (!dev_flow)
                        goto error;
                dev_flow->flow = flow;
+               dev_flow->external = 0;
                LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
                ret = flow_drv_translate(dev, dev_flow, &attr_tx,
                                         items_tx.items,
index 492f65e..05fee83 100644 (file)
@@ -417,7 +417,6 @@ struct mlx5_flow_dv_push_vlan_action_resource {
 
 /* DV flows structure. */
 struct mlx5_flow_dv {
-       uint64_t hash_fields; /**< Fields that participate in the hash. */
        struct mlx5_hrxq *hrxq; /**< Hash Rx queues. */
        /* Flow DV api: */
        struct mlx5_flow_dv_matcher *matcher; /**< Cache to matcher. */
@@ -436,6 +435,8 @@ struct mlx5_flow_dv {
        /**< Structure for VF VLAN workaround. */
        struct mlx5_flow_dv_push_vlan_action_resource *push_vlan_res;
        /**< Pointer to push VLAN action resource in cache. */
+       struct mlx5_flow_dv_tag_resource *tag_resource;
+       /**< pointer to the tag action. */
 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
        void *actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS];
        /**< Action list. */
@@ -460,11 +461,18 @@ struct mlx5_flow_verbs {
        };
        struct ibv_flow *flow; /**< Verbs flow pointer. */
        struct mlx5_hrxq *hrxq; /**< Hash Rx queue object. */
-       uint64_t hash_fields; /**< Verbs hash Rx queue hash fields. */
        struct mlx5_vf_vlan vf_vlan;
        /**< Structure for VF VLAN workaround. */
 };
 
+struct mlx5_flow_rss {
+       uint32_t level;
+       uint32_t queue_num; /**< Number of entries in @p queue. */
+       uint64_t types; /**< Specific RSS hash types (see ETH_RSS_*). */
+       uint16_t (*queue)[]; /**< Destination queues to redirect traffic to. */
+       uint8_t key[MLX5_RSS_HASH_KEY_LEN]; /**< RSS hash key. */
+};
+
 /** Device flow structure. */
 struct mlx5_flow {
        LIST_ENTRY(mlx5_flow) next;
@@ -473,6 +481,10 @@ struct mlx5_flow {
        /**< Bit-fields of present layers, see MLX5_FLOW_LAYER_*. */
        uint64_t actions;
        /**< Bit-fields of detected actions, see MLX5_FLOW_ACTION_*. */
+       uint64_t hash_fields; /**< Verbs hash Rx queue hash fields. */
+       uint8_t ingress; /**< 1 if the flow is ingress. */
+       uint32_t group; /**< The group index. */
+       uint8_t transfer; /**< 1 if the flow is E-Switch flow. */
        union {
 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
                struct mlx5_flow_dv dv;
@@ -486,18 +498,11 @@ struct mlx5_flow {
 struct rte_flow {
        TAILQ_ENTRY(rte_flow) next; /**< Pointer to the next flow structure. */
        enum mlx5_flow_drv_type drv_type; /**< Driver type. */
+       struct mlx5_flow_rss rss; /**< RSS context. */
        struct mlx5_flow_counter *counter; /**< Holds flow counter. */
-       struct mlx5_flow_dv_tag_resource *tag_resource;
-       /**< pointer to the tag action. */
-       struct rte_flow_action_rss rss;/**< RSS context. */
-       uint8_t key[MLX5_RSS_HASH_KEY_LEN]; /**< RSS hash key. */
-       uint16_t (*queue)[]; /**< Destination queues to redirect traffic to. */
        LIST_HEAD(dev_flows, mlx5_flow) dev_flows;
        /**< Device flows that are part of the flow. */
        struct mlx5_fdir *fdir; /**< Pointer to associated FDIR if any. */
-       uint8_t ingress; /**< 1 if the flow is ingress. */
-       uint32_t group; /**< The group index. */
-       uint8_t transfer; /**< 1 if the flow is E-Switch flow. */
        uint32_t hairpin_flow_id; /**< The flow id used for hairpin. */
 };
 
index bbebeaf..18901a6 100644 (file)
@@ -1585,10 +1585,9 @@ flow_dv_encap_decap_resource_register
        struct mlx5_priv *priv = dev->data->dev_private;
        struct mlx5_ibv_shared *sh = priv->sh;
        struct mlx5_flow_dv_encap_decap_resource *cache_resource;
-       struct rte_flow *flow = dev_flow->flow;
        struct mlx5dv_dr_domain *domain;
 
-       resource->flags = flow->group ? 0 : 1;
+       resource->flags = dev_flow->group ? 0 : 1;
        if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
                domain = sh->fdb_domain;
        else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
@@ -2747,7 +2746,7 @@ flow_dv_modify_hdr_resource_register
        else
                ns = sh->rx_domain;
        resource->flags =
-               dev_flow->flow->group ? 0 : MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
+               dev_flow->group ? 0 : MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
        /* Lookup a matching resource from cache. */
        LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
                if (resource->ft_type == cache_resource->ft_type &&
@@ -4095,18 +4094,20 @@ flow_dv_prepare(const struct rte_flow_attr *attr __rte_unused,
                const struct rte_flow_action actions[] __rte_unused,
                struct rte_flow_error *error)
 {
-       uint32_t size = sizeof(struct mlx5_flow);
-       struct mlx5_flow *flow;
+       size_t size = sizeof(struct mlx5_flow);
+       struct mlx5_flow *dev_flow;
 
-       flow = rte_calloc(__func__, 1, size, 0);
-       if (!flow) {
+       dev_flow = rte_calloc(__func__, 1, size, 0);
+       if (!dev_flow) {
                rte_flow_error_set(error, ENOMEM,
                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
                                   "not enough memory to create flow");
                return NULL;
        }
-       flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
-       return flow;
+       dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param);
+       dev_flow->ingress = attr->ingress;
+       dev_flow->transfer = attr->transfer;
+       return dev_flow;
 }
 
 #ifndef NDEBUG
@@ -5487,7 +5488,7 @@ flow_dv_tag_resource_register
                                (void *)cache_resource,
                                rte_atomic32_read(&cache_resource->refcnt));
                        rte_atomic32_inc(&cache_resource->refcnt);
-                       dev_flow->flow->tag_resource = cache_resource;
+                       dev_flow->dv.tag_resource = cache_resource;
                        return 0;
                }
        }
@@ -5509,7 +5510,7 @@ flow_dv_tag_resource_register
        rte_atomic32_init(&cache_resource->refcnt);
        rte_atomic32_inc(&cache_resource->refcnt);
        LIST_INSERT_HEAD(&sh->tags, cache_resource, next);
-       dev_flow->flow->tag_resource = cache_resource;
+       dev_flow->dv.tag_resource = cache_resource;
        DRV_LOG(DEBUG, "new tag resource %p: refcnt %d++",
                (void *)cache_resource,
                rte_atomic32_read(&cache_resource->refcnt));
@@ -5689,7 +5690,7 @@ flow_dv_translate(struct rte_eth_dev *dev,
                                       &table, error);
        if (ret)
                return ret;
-       flow->group = table;
+       dev_flow->group = table;
        if (attr->transfer)
                res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
        if (priority == MLX5_FLOW_PRIO_RSVD)
@@ -5726,47 +5727,50 @@ flow_dv_translate(struct rte_eth_dev *dev,
                case RTE_FLOW_ACTION_TYPE_FLAG:
                        tag_resource.tag =
                                mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
-                       if (!flow->tag_resource)
+                       if (!dev_flow->dv.tag_resource)
                                if (flow_dv_tag_resource_register
                                    (dev, &tag_resource, dev_flow, error))
                                        return errno;
                        dev_flow->dv.actions[actions_n++] =
-                               flow->tag_resource->action;
+                               dev_flow->dv.tag_resource->action;
                        action_flags |= MLX5_FLOW_ACTION_FLAG;
                        break;
                case RTE_FLOW_ACTION_TYPE_MARK:
                        tag_resource.tag = mlx5_flow_mark_set
                              (((const struct rte_flow_action_mark *)
                               (actions->conf))->id);
-                       if (!flow->tag_resource)
+                       if (!dev_flow->dv.tag_resource)
                                if (flow_dv_tag_resource_register
                                    (dev, &tag_resource, dev_flow, error))
                                        return errno;
                        dev_flow->dv.actions[actions_n++] =
-                               flow->tag_resource->action;
+                               dev_flow->dv.tag_resource->action;
                        action_flags |= MLX5_FLOW_ACTION_MARK;
                        break;
                case RTE_FLOW_ACTION_TYPE_DROP:
                        action_flags |= MLX5_FLOW_ACTION_DROP;
                        break;
                case RTE_FLOW_ACTION_TYPE_QUEUE:
+                       assert(flow->rss.queue);
                        queue = actions->conf;
                        flow->rss.queue_num = 1;
-                       (*flow->queue)[0] = queue->index;
+                       (*flow->rss.queue)[0] = queue->index;
                        action_flags |= MLX5_FLOW_ACTION_QUEUE;
                        break;
                case RTE_FLOW_ACTION_TYPE_RSS:
+                       assert(flow->rss.queue);
                        rss = actions->conf;
-                       if (flow->queue)
-                               memcpy((*flow->queue), rss->queue,
+                       if (flow->rss.queue)
+                               memcpy((*flow->rss.queue), rss->queue,
                                       rss->queue_num * sizeof(uint16_t));
                        flow->rss.queue_num = rss->queue_num;
                        /* NULL RSS key indicates default RSS key. */
                        rss_key = !rss->key ? rss_hash_default_key : rss->key;
-                       memcpy(flow->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
-                       /* RSS type 0 indicates default RSS type ETH_RSS_IP. */
-                       flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
-                       flow->rss.level = rss->level;
+                       memcpy(flow->rss.key, rss_key, MLX5_RSS_HASH_KEY_LEN);
+                       /*
+                        * rss->level and rss.types should be set in advance
+                        * when expanding items for RSS.
+                        */
                        action_flags |= MLX5_FLOW_ACTION_RSS;
                        break;
                case RTE_FLOW_ACTION_TYPE_COUNT:
@@ -5777,7 +5781,7 @@ flow_dv_translate(struct rte_eth_dev *dev,
                        flow->counter = flow_dv_counter_alloc(dev,
                                                              count->shared,
                                                              count->id,
-                                                             flow->group);
+                                                             dev_flow->group);
                        if (flow->counter == NULL)
                                goto cnt_err;
                        dev_flow->dv.actions[actions_n++] =
@@ -6075,9 +6079,10 @@ cnt_err:
                        mlx5_flow_tunnel_ip_check(items, next_protocol,
                                                  &item_flags, &tunnel);
                        flow_dv_translate_item_ipv4(match_mask, match_value,
-                                                   items, tunnel, flow->group);
+                                                   items, tunnel,
+                                                   dev_flow->group);
                        matcher.priority = MLX5_PRIORITY_MAP_L3;
-                       dev_flow->dv.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel,
                                         MLX5_IPV4_LAYER_TYPES,
@@ -6102,9 +6107,10 @@ cnt_err:
                        mlx5_flow_tunnel_ip_check(items, next_protocol,
                                                  &item_flags, &tunnel);
                        flow_dv_translate_item_ipv6(match_mask, match_value,
-                                                   items, tunnel, flow->group);
+                                                   items, tunnel,
+                                                   dev_flow->group);
                        matcher.priority = MLX5_PRIORITY_MAP_L3;
-                       dev_flow->dv.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel,
                                         MLX5_IPV6_LAYER_TYPES,
@@ -6129,7 +6135,7 @@ cnt_err:
                        flow_dv_translate_item_tcp(match_mask, match_value,
                                                   items, tunnel);
                        matcher.priority = MLX5_PRIORITY_MAP_L4;
-                       dev_flow->dv.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel, ETH_RSS_TCP,
                                         IBV_RX_HASH_SRC_PORT_TCP |
@@ -6141,7 +6147,7 @@ cnt_err:
                        flow_dv_translate_item_udp(match_mask, match_value,
                                                   items, tunnel);
                        matcher.priority = MLX5_PRIORITY_MAP_L4;
-                       dev_flow->dv.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel, ETH_RSS_UDP,
                                         IBV_RX_HASH_SRC_PORT_UDP |
@@ -6237,7 +6243,7 @@ cnt_err:
        matcher.priority = mlx5_flow_adjust_priority(dev, priority,
                                                     matcher.priority);
        matcher.egress = attr->egress;
-       matcher.group = flow->group;
+       matcher.group = dev_flow->group;
        matcher.transfer = attr->transfer;
        if (flow_dv_matcher_register(dev, &matcher, dev_flow, error))
                return -rte_errno;
@@ -6271,7 +6277,7 @@ flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
                dv = &dev_flow->dv;
                n = dv->actions_n;
                if (dev_flow->actions & MLX5_FLOW_ACTION_DROP) {
-                       if (flow->transfer) {
+                       if (dev_flow->transfer) {
                                dv->actions[n++] = priv->sh->esw_drop_action;
                        } else {
                                dv->hrxq = mlx5_hrxq_drop_new(dev);
@@ -6289,15 +6295,18 @@ flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
                           (MLX5_FLOW_ACTION_QUEUE | MLX5_FLOW_ACTION_RSS)) {
                        struct mlx5_hrxq *hrxq;
 
-                       hrxq = mlx5_hrxq_get(dev, flow->key,
+                       assert(flow->rss.queue);
+                       hrxq = mlx5_hrxq_get(dev, flow->rss.key,
                                             MLX5_RSS_HASH_KEY_LEN,
-                                            dv->hash_fields,
-                                            (*flow->queue),
+                                            dev_flow->hash_fields,
+                                            (*flow->rss.queue),
                                             flow->rss.queue_num);
                        if (!hrxq) {
                                hrxq = mlx5_hrxq_new
-                                       (dev, flow->key, MLX5_RSS_HASH_KEY_LEN,
-                                        dv->hash_fields, (*flow->queue),
+                                       (dev, flow->rss.key,
+                                        MLX5_RSS_HASH_KEY_LEN,
+                                        dev_flow->hash_fields,
+                                        (*flow->rss.queue),
                                         flow->rss.queue_num,
                                         !!(dev_flow->layers &
                                            MLX5_FLOW_LAYER_TUNNEL));
@@ -6607,10 +6616,6 @@ flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
                flow_dv_counter_release(dev, flow->counter);
                flow->counter = NULL;
        }
-       if (flow->tag_resource) {
-               flow_dv_tag_release(dev, flow->tag_resource);
-               flow->tag_resource = NULL;
-       }
        while (!LIST_EMPTY(&flow->dev_flows)) {
                dev_flow = LIST_FIRST(&flow->dev_flows);
                LIST_REMOVE(dev_flow, next);
@@ -6626,6 +6631,8 @@ flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
                        flow_dv_port_id_action_resource_release(dev_flow);
                if (dev_flow->dv.push_vlan_res)
                        flow_dv_push_vlan_action_resource_release(dev_flow);
+               if (dev_flow->dv.tag_resource)
+                       flow_dv_tag_release(dev, dev_flow->dv.tag_resource);
                rte_free(dev_flow);
        }
 }
index 147436d..ea85fa1 100644 (file)
@@ -864,8 +864,8 @@ flow_verbs_translate_action_queue(struct mlx5_flow *dev_flow,
        const struct rte_flow_action_queue *queue = action->conf;
        struct rte_flow *flow = dev_flow->flow;
 
-       if (flow->queue)
-               (*flow->queue)[0] = queue->index;
+       if (flow->rss.queue)
+               (*flow->rss.queue)[0] = queue->index;
        flow->rss.queue_num = 1;
 }
 
@@ -889,16 +889,17 @@ flow_verbs_translate_action_rss(struct mlx5_flow *dev_flow,
        const uint8_t *rss_key;
        struct rte_flow *flow = dev_flow->flow;
 
-       if (flow->queue)
-               memcpy((*flow->queue), rss->queue,
+       if (flow->rss.queue)
+               memcpy((*flow->rss.queue), rss->queue,
                       rss->queue_num * sizeof(uint16_t));
        flow->rss.queue_num = rss->queue_num;
        /* NULL RSS key indicates default RSS key. */
        rss_key = !rss->key ? rss_hash_default_key : rss->key;
-       memcpy(flow->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
-       /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
-       flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
-       flow->rss.level = rss->level;
+       memcpy(flow->rss.key, rss_key, MLX5_RSS_HASH_KEY_LEN);
+       /*
+        * rss->level and rss.types should be set in advance when expanding
+        * items for RSS.
+        */
 }
 
 /**
@@ -1392,22 +1393,23 @@ flow_verbs_prepare(const struct rte_flow_attr *attr __rte_unused,
                   const struct rte_flow_action actions[],
                   struct rte_flow_error *error)
 {
-       uint32_t size = sizeof(struct mlx5_flow) + sizeof(struct ibv_flow_attr);
-       struct mlx5_flow *flow;
+       size_t size = sizeof(struct mlx5_flow) + sizeof(struct ibv_flow_attr);
+       struct mlx5_flow *dev_flow;
 
        size += flow_verbs_get_actions_size(actions);
        size += flow_verbs_get_items_size(items);
-       flow = rte_calloc(__func__, 1, size, 0);
-       if (!flow) {
+       dev_flow = rte_calloc(__func__, 1, size, 0);
+       if (!dev_flow) {
                rte_flow_error_set(error, ENOMEM,
                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
                                   "not enough memory to create flow");
                return NULL;
        }
-       flow->verbs.attr = (void *)(flow + 1);
-       flow->verbs.specs =
-               (uint8_t *)(flow + 1) + sizeof(struct ibv_flow_attr);
-       return flow;
+       dev_flow->verbs.attr = (void *)(dev_flow + 1);
+       dev_flow->verbs.specs = (void *)(dev_flow->verbs.attr + 1);
+       dev_flow->ingress = attr->ingress;
+       dev_flow->transfer = attr->transfer;
+       return dev_flow;
 }
 
 /**
@@ -1513,7 +1515,7 @@ flow_verbs_translate(struct rte_eth_dev *dev,
                        flow_verbs_translate_item_ipv4(dev_flow, items,
                                                       item_flags);
                        subpriority = MLX5_PRIORITY_MAP_L3;
-                       dev_flow->verbs.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel,
                                         MLX5_IPV4_LAYER_TYPES,
@@ -1525,7 +1527,7 @@ flow_verbs_translate(struct rte_eth_dev *dev,
                        flow_verbs_translate_item_ipv6(dev_flow, items,
                                                       item_flags);
                        subpriority = MLX5_PRIORITY_MAP_L3;
-                       dev_flow->verbs.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel,
                                         MLX5_IPV6_LAYER_TYPES,
@@ -1537,7 +1539,7 @@ flow_verbs_translate(struct rte_eth_dev *dev,
                        flow_verbs_translate_item_tcp(dev_flow, items,
                                                      item_flags);
                        subpriority = MLX5_PRIORITY_MAP_L4;
-                       dev_flow->verbs.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel, ETH_RSS_TCP,
                                         (IBV_RX_HASH_SRC_PORT_TCP |
@@ -1549,7 +1551,7 @@ flow_verbs_translate(struct rte_eth_dev *dev,
                        flow_verbs_translate_item_udp(dev_flow, items,
                                                      item_flags);
                        subpriority = MLX5_PRIORITY_MAP_L4;
-                       dev_flow->verbs.hash_fields |=
+                       dev_flow->hash_fields |=
                                mlx5_flow_hashfields_adjust
                                        (dev_flow, tunnel, ETH_RSS_UDP,
                                         (IBV_RX_HASH_SRC_PORT_UDP |
@@ -1694,16 +1696,17 @@ flow_verbs_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
                } else {
                        struct mlx5_hrxq *hrxq;
 
-                       hrxq = mlx5_hrxq_get(dev, flow->key,
+                       assert(flow->rss.queue);
+                       hrxq = mlx5_hrxq_get(dev, flow->rss.key,
                                             MLX5_RSS_HASH_KEY_LEN,
-                                            verbs->hash_fields,
-                                            (*flow->queue),
+                                            dev_flow->hash_fields,
+                                            (*flow->rss.queue),
                                             flow->rss.queue_num);
                        if (!hrxq)
-                               hrxq = mlx5_hrxq_new(dev, flow->key,
+                               hrxq = mlx5_hrxq_new(dev, flow->rss.key,
                                                     MLX5_RSS_HASH_KEY_LEN,
-                                                    verbs->hash_fields,
-                                                    (*flow->queue),
+                                                    dev_flow->hash_fields,
+                                                    (*flow->rss.queue),
                                                     flow->rss.queue_num,
                                                     !!(dev_flow->layers &
                                                       MLX5_FLOW_LAYER_TUNNEL));