net/sfc: support SW stats groups
[dpdk.git] / drivers / net / mlx5 / mlx5_flow.c
index 2bbb5f5..c914a71 100644 (file)
@@ -100,10 +100,25 @@ struct mlx5_flow_expand_node {
         * RSS types bit-field associated with this node
         * (see ETH_RSS_* definitions).
         */
-       uint8_t optional;
-       /**< optional expand field. Default 0 to expand, 1 not go deeper. */
+       uint64_t node_flags;
+       /**<
+        *  Bit-fields that define how the node is used in the expansion.
+        * (see MLX5_EXPANSION_NODE_* definitions).
+        */
 };
 
+/* Optional expand field. The expansion alg will not go deeper. */
+#define MLX5_EXPANSION_NODE_OPTIONAL (UINT64_C(1) << 0)
+
+/* The node is not added implicitly as expansion to the flow pattern.
+ * If the node type does not match the flow pattern item type, the
+ * expansion alg will go deeper to its next items.
+ * In the current implementation, the list of next nodes indexes can
+ * have up to one node with this flag set and it has to be the last
+ * node index (before the list terminator).
+ */
+#define MLX5_EXPANSION_NODE_EXPLICIT (UINT64_C(1) << 1)
+
 /** Object returned by mlx5_flow_expand_rss(). */
 struct mlx5_flow_expand_rss {
        uint32_t entries;
@@ -132,6 +147,10 @@ mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
        case RTE_FLOW_ITEM_TYPE_GRE:
        case RTE_FLOW_ITEM_TYPE_GENEVE:
        case RTE_FLOW_ITEM_TYPE_MPLS:
+       case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
+       case RTE_FLOW_ITEM_TYPE_GRE_KEY:
+       case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
+       case RTE_FLOW_ITEM_TYPE_GTP:
                return true;
        default:
                break;
@@ -240,6 +259,26 @@ mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
        return ret;
 }
 
+static const int *
+mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
+               const int *next_node)
+{
+       const struct mlx5_flow_expand_node *node = NULL;
+       const int *next = next_node;
+
+       while (next && *next) {
+               /*
+                * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
+                * flag set, because they were not found in the flow pattern.
+                */
+               node = &graph[*next];
+               if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
+                       break;
+               next = node->next;
+       }
+       return next;
+}
+
 #define MLX5_RSS_EXP_ELT_N 16
 
 /**
@@ -304,10 +343,17 @@ mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
                        continue;
                }
                last_item = item;
-               for (i = 0; node->next && node->next[i]; ++i) {
+               i = 0;
+               while (node->next && node->next[i]) {
                        next = &graph[node->next[i]];
                        if (next->type == item->type)
                                break;
+                       if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
+                               node = next;
+                               i = 0;
+                       } else {
+                               ++i;
+                       }
                }
                if (next)
                        node = next;
@@ -365,7 +411,8 @@ mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
                }
        }
        memset(flow_items, 0, sizeof(flow_items));
-       next_node = node->next;
+       next_node = mlx5_flow_expand_rss_skip_explicit(graph,
+                       node->next);
        stack[stack_pos] = next_node;
        node = next_node ? &graph[*next_node] : NULL;
        while (node) {
@@ -400,8 +447,10 @@ mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
                        addr = (void *)(((uintptr_t)addr) + n);
                }
                /* Go deeper. */
-               if (!node->optional && node->next) {
-                       next_node = node->next;
+               if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
+                               node->next) {
+                       next_node = mlx5_flow_expand_rss_skip_explicit(graph,
+                                       node->next);
                        if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
                                rte_errno = E2BIG;
                                return -rte_errno;
@@ -409,15 +458,27 @@ mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
                        stack[stack_pos] = next_node;
                } else if (*(next_node + 1)) {
                        /* Follow up with the next possibility. */
+                       next_node = mlx5_flow_expand_rss_skip_explicit(graph,
+                                       ++next_node);
+               } else if (!stack_pos) {
+                       /*
+                        * Completing the traverse over the different paths.
+                        * The next_node is advanced to the terminator.
+                        */
                        ++next_node;
                } else {
                        /* Move to the next path. */
-                       if (stack_pos)
+                       while (stack_pos) {
                                next_node = stack[--stack_pos];
-                       next_node++;
+                               next_node++;
+                               if (*next_node)
+                                       break;
+                       }
+                       next_node = mlx5_flow_expand_rss_skip_explicit(graph,
+                                       next_node);
                        stack[stack_pos] = next_node;
                }
-               node = *next_node ? &graph[*next_node] : NULL;
+               node = next_node && *next_node ? &graph[*next_node] : NULL;
        };
        return lsize;
 }
@@ -425,10 +486,7 @@ mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
 enum mlx5_expansion {
        MLX5_EXPANSION_ROOT,
        MLX5_EXPANSION_ROOT_OUTER,
-       MLX5_EXPANSION_ROOT_ETH_VLAN,
-       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
        MLX5_EXPANSION_OUTER_ETH,
-       MLX5_EXPANSION_OUTER_ETH_VLAN,
        MLX5_EXPANSION_OUTER_VLAN,
        MLX5_EXPANSION_OUTER_IPV4,
        MLX5_EXPANSION_OUTER_IPV4_UDP,
@@ -443,7 +501,6 @@ enum mlx5_expansion {
        MLX5_EXPANSION_GRE_KEY,
        MLX5_EXPANSION_MPLS,
        MLX5_EXPANSION_ETH,
-       MLX5_EXPANSION_ETH_VLAN,
        MLX5_EXPANSION_VLAN,
        MLX5_EXPANSION_IPV4,
        MLX5_EXPANSION_IPV4_UDP,
@@ -451,6 +508,8 @@ enum mlx5_expansion {
        MLX5_EXPANSION_IPV6,
        MLX5_EXPANSION_IPV6_UDP,
        MLX5_EXPANSION_IPV6_TCP,
+       MLX5_EXPANSION_IPV6_FRAG_EXT,
+       MLX5_EXPANSION_GTP
 };
 
 /** Supported expansion of items. */
@@ -467,22 +526,7 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                                                  MLX5_EXPANSION_OUTER_IPV6),
                .type = RTE_FLOW_ITEM_TYPE_END,
        },
-       [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
-               .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
-               .type = RTE_FLOW_ITEM_TYPE_END,
-       },
-       [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
-               .next = MLX5_FLOW_EXPAND_RSS_NEXT
-                                               (MLX5_EXPANSION_OUTER_ETH_VLAN),
-               .type = RTE_FLOW_ITEM_TYPE_END,
-       },
        [MLX5_EXPANSION_OUTER_ETH] = {
-               .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
-                                                 MLX5_EXPANSION_OUTER_IPV6),
-               .type = RTE_FLOW_ITEM_TYPE_ETH,
-               .rss_types = 0,
-       },
-       [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
                .type = RTE_FLOW_ITEM_TYPE_ETH,
                .rss_types = 0,
@@ -491,6 +535,7 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
                                                  MLX5_EXPANSION_OUTER_IPV6),
                .type = RTE_FLOW_ITEM_TYPE_VLAN,
+               .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
        },
        [MLX5_EXPANSION_OUTER_IPV4] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT
@@ -507,7 +552,8 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
        [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
                                                  MLX5_EXPANSION_VXLAN_GPE,
-                                                 MLX5_EXPANSION_MPLS),
+                                                 MLX5_EXPANSION_MPLS,
+                                                 MLX5_EXPANSION_GTP),
                .type = RTE_FLOW_ITEM_TYPE_UDP,
                .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
        },
@@ -530,7 +576,8 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
        [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
                                                  MLX5_EXPANSION_VXLAN_GPE,
-                                                 MLX5_EXPANSION_MPLS),
+                                                 MLX5_EXPANSION_MPLS,
+                                                 MLX5_EXPANSION_GTP),
                .type = RTE_FLOW_ITEM_TYPE_UDP,
                .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
        },
@@ -562,7 +609,7 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                                                  MLX5_EXPANSION_IPV6,
                                                  MLX5_EXPANSION_MPLS),
                .type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
-               .optional = 1,
+               .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
        },
        [MLX5_EXPANSION_NVGRE] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
@@ -573,13 +620,9 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                                                  MLX5_EXPANSION_IPV6,
                                                  MLX5_EXPANSION_ETH),
                .type = RTE_FLOW_ITEM_TYPE_MPLS,
+               .node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
        },
        [MLX5_EXPANSION_ETH] = {
-               .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
-                                                 MLX5_EXPANSION_IPV6),
-               .type = RTE_FLOW_ITEM_TYPE_ETH,
-       },
-       [MLX5_EXPANSION_ETH_VLAN] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
                .type = RTE_FLOW_ITEM_TYPE_ETH,
        },
@@ -587,6 +630,7 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
                                                  MLX5_EXPANSION_IPV6),
                .type = RTE_FLOW_ITEM_TYPE_VLAN,
+               .node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
        },
        [MLX5_EXPANSION_IPV4] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
@@ -605,7 +649,8 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
        },
        [MLX5_EXPANSION_IPV6] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
-                                                 MLX5_EXPANSION_IPV6_TCP),
+                                                 MLX5_EXPANSION_IPV6_TCP,
+                                                 MLX5_EXPANSION_IPV6_FRAG_EXT),
                .type = RTE_FLOW_ITEM_TYPE_IPV6,
                .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
                        ETH_RSS_NONFRAG_IPV6_OTHER,
@@ -618,6 +663,14 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                .type = RTE_FLOW_ITEM_TYPE_TCP,
                .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
        },
+       [MLX5_EXPANSION_IPV6_FRAG_EXT] = {
+               .type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
+       },
+       [MLX5_EXPANSION_GTP] = {
+               .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
+                                                 MLX5_EXPANSION_IPV6),
+               .type = RTE_FLOW_ITEM_TYPE_GTP,
+       },
 };
 
 static struct rte_flow_action_handle *
@@ -1313,9 +1366,7 @@ mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
                        data->dynf_meta = 1;
                        data->flow_meta_mask = rte_flow_dynf_metadata_mask;
                        data->flow_meta_offset = rte_flow_dynf_metadata_offs;
-                       data->flow_meta_port_mask = (uint32_t)~0;
-                       if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
-                               data->flow_meta_port_mask >>= 16;
+                       data->flow_meta_port_mask = priv->sh->dv_meta_mask;
                }
        }
 }
@@ -1984,6 +2035,10 @@ mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
                return rte_flow_error_set(error, EINVAL,
                                          RTE_FLOW_ERROR_TYPE_ITEM, item,
                                          "L2 layer should not follow VLAN");
+       if (item_flags & MLX5_FLOW_LAYER_GTP)
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ITEM, item,
+                                         "L2 layer should not follow GTP");
        if (!mask)
                mask = &rte_flow_item_eth_mask;
        ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
@@ -2412,6 +2467,8 @@ mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
  *
  * @param[in] dev
  *   Pointer to the Ethernet device structure.
+ * @param[in] udp_dport
+ *   UDP destination port
  * @param[in] item
  *   Item specification.
  * @param[in] item_flags
@@ -2426,6 +2483,7 @@ mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
  */
 int
 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
+                             uint16_t udp_dport,
                              const struct rte_flow_item *item,
                              uint64_t item_flags,
                              const struct rte_flow_attr *attr,
@@ -2461,12 +2519,18 @@ mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
                                          "no outer UDP layer found");
        if (!mask)
                mask = &rte_flow_item_vxlan_mask;
-       /* FDB domain & NIC domain non-zero group */
-       if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
-               valid_mask = &nic_mask;
-       /* Group zero in NIC domain */
-       if (!attr->group && !attr->transfer && priv->sh->tunnel_header_0_1)
-               valid_mask = &nic_mask;
+
+       if (priv->sh->steering_format_version !=
+           MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
+           !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
+               /* FDB domain & NIC domain non-zero group */
+               if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
+                       valid_mask = &nic_mask;
+               /* Group zero in NIC domain */
+               if (!attr->group && !attr->transfer &&
+                   priv->sh->tunnel_header_0_1)
+                       valid_mask = &nic_mask;
+       }
        ret = mlx5_flow_item_acceptable
                (item, (const uint8_t *)mask,
                 (const uint8_t *)valid_mask,
@@ -3745,20 +3809,8 @@ flow_get_shared_rss_action(struct rte_eth_dev *dev,
 }
 
 static unsigned int
-find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
+find_graph_root(uint32_t rss_level)
 {
-       const struct rte_flow_item *item;
-       unsigned int has_vlan = 0;
-
-       for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
-               if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
-                       has_vlan = 1;
-                       break;
-               }
-       }
-       if (has_vlan)
-               return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
-                                      MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
        return rss_level < 2 ? MLX5_EXPANSION_ROOT :
                               MLX5_EXPANSION_ROOT_OUTER;
 }
@@ -4056,28 +4108,27 @@ flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
                  uint32_t flow_idx);
 
 int
-flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
-                     struct mlx5_hlist_entry *entry,
-                     uint64_t key, void *cb_ctx __rte_unused)
+flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
+                     struct mlx5_list_entry *entry, void *cb_ctx)
 {
+       struct mlx5_flow_cb_ctx *ctx = cb_ctx;
        struct mlx5_flow_mreg_copy_resource *mcp_res =
-               container_of(entry, typeof(*mcp_res), hlist_ent);
+                              container_of(entry, typeof(*mcp_res), hlist_ent);
 
-       return mcp_res->mark_id != key;
+       return mcp_res->mark_id != *(uint32_t *)(ctx->data);
 }
 
-struct mlx5_hlist_entry *
-flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
-                      void *cb_ctx)
+struct mlx5_list_entry *
+flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
 {
-       struct rte_eth_dev *dev = list->ctx;
+       struct rte_eth_dev *dev = tool_ctx;
        struct mlx5_priv *priv = dev->data->dev_private;
        struct mlx5_flow_cb_ctx *ctx = cb_ctx;
        struct mlx5_flow_mreg_copy_resource *mcp_res;
        struct rte_flow_error *error = ctx->error;
        uint32_t idx = 0;
        int ret;
-       uint32_t mark_id = key;
+       uint32_t mark_id = *(uint32_t *)(ctx->data);
        struct rte_flow_attr attr = {
                .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
                .ingress = 1,
@@ -4183,6 +4234,36 @@ flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
        return &mcp_res->hlist_ent;
 }
 
+struct mlx5_list_entry *
+flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
+                     void *cb_ctx __rte_unused)
+{
+       struct rte_eth_dev *dev = tool_ctx;
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_flow_mreg_copy_resource *mcp_res;
+       uint32_t idx = 0;
+
+       mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
+       if (!mcp_res) {
+               rte_errno = ENOMEM;
+               return NULL;
+       }
+       memcpy(mcp_res, oentry, sizeof(*mcp_res));
+       mcp_res->idx = idx;
+       return &mcp_res->hlist_ent;
+}
+
+void
+flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
+{
+       struct mlx5_flow_mreg_copy_resource *mcp_res =
+                              container_of(entry, typeof(*mcp_res), hlist_ent);
+       struct rte_eth_dev *dev = tool_ctx;
+       struct mlx5_priv *priv = dev->data->dev_private;
+
+       mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
+}
+
 /**
  * Add a flow of copying flow metadata registers in RX_CP_TBL.
  *
@@ -4213,10 +4294,11 @@ flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
                          struct rte_flow_error *error)
 {
        struct mlx5_priv *priv = dev->data->dev_private;
-       struct mlx5_hlist_entry *entry;
+       struct mlx5_list_entry *entry;
        struct mlx5_flow_cb_ctx ctx = {
                .dev = dev,
                .error = error,
+               .data = &mark_id,
        };
 
        /* Check if already registered. */
@@ -4229,11 +4311,11 @@ flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
 }
 
 void
-flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
+flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
 {
        struct mlx5_flow_mreg_copy_resource *mcp_res =
-               container_of(entry, typeof(*mcp_res), hlist_ent);
-       struct rte_eth_dev *dev = list->ctx;
+                              container_of(entry, typeof(*mcp_res), hlist_ent);
+       struct rte_eth_dev *dev = tool_ctx;
        struct mlx5_priv *priv = dev->data->dev_private;
 
        MLX5_ASSERT(mcp_res->rix_flow);
@@ -4279,14 +4361,17 @@ flow_mreg_del_copy_action(struct rte_eth_dev *dev,
 static void
 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
 {
-       struct mlx5_hlist_entry *entry;
+       struct mlx5_list_entry *entry;
        struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_flow_cb_ctx ctx;
+       uint32_t mark_id;
 
        /* Check if default flow is registered. */
        if (!priv->mreg_cp_tbl)
                return;
-       entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
-                                 MLX5_DEFAULT_COPY_ID, NULL);
+       mark_id = MLX5_DEFAULT_COPY_ID;
+       ctx.data = &mark_id;
+       entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
        if (!entry)
                return;
        mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
@@ -4312,6 +4397,8 @@ flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
 {
        struct mlx5_priv *priv = dev->data->dev_private;
        struct mlx5_flow_mreg_copy_resource *mcp_res;
+       struct mlx5_flow_cb_ctx ctx;
+       uint32_t mark_id;
 
        /* Check whether extensive metadata feature is engaged. */
        if (!priv->config.dv_flow_en ||
@@ -4323,9 +4410,11 @@ flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
         * Add default mreg copy flow may be called multiple time, but
         * only be called once in stop. Avoid register it twice.
         */
-       if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
+       mark_id = MLX5_DEFAULT_COPY_ID;
+       ctx.data = &mark_id;
+       if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
                return 0;
-       mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
+       mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
        if (!mcp_res)
                return -rte_errno;
        return 0;
@@ -4642,7 +4731,7 @@ get_meter_sub_policy(struct rte_eth_dev *dev,
                struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
                uint32_t i;
 
-               /**
+               /*
                 * This is a tmp dev_flow,
                 * no need to register any matcher for it in translate.
                 */
@@ -4650,18 +4739,19 @@ get_meter_sub_policy(struct rte_eth_dev *dev,
                for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
                        struct mlx5_flow dev_flow = {0};
                        struct mlx5_flow_handle dev_handle = { {0} };
+                       uint8_t fate = final_policy->act_cnt[i].fate_action;
 
-                       if (final_policy->is_rss) {
-                               const void *rss_act =
+                       if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
+                               const struct rte_flow_action_rss *rss_act =
                                        final_policy->act_cnt[i].rss->conf;
                                struct rte_flow_action rss_actions[2] = {
                                        [0] = {
                                        .type = RTE_FLOW_ACTION_TYPE_RSS,
-                                       .conf = rss_act
+                                       .conf = rss_act,
                                        },
                                        [1] = {
                                        .type = RTE_FLOW_ACTION_TYPE_END,
-                                       .conf = NULL
+                                       .conf = NULL,
                                        }
                                };
 
@@ -4686,9 +4776,13 @@ get_meter_sub_policy(struct rte_eth_dev *dev,
                                                rss_desc_v[i].hash_fields ?
                                                rss_desc_v[i].queue_num : 1;
                                rss_desc_v[i].tunnel =
-                                       !!(dev_flow.handle->layers &
-                                       MLX5_FLOW_LAYER_TUNNEL);
-                       } else {
+                                               !!(dev_flow.handle->layers &
+                                                  MLX5_FLOW_LAYER_TUNNEL);
+                               /* Use the RSS queues in the containers. */
+                               rss_desc_v[i].queue =
+                                       (uint16_t *)(uintptr_t)rss_act->queue;
+                               rss_desc[i] = &rss_desc_v[i];
+                       } else if (fate == MLX5_FLOW_FATE_QUEUE) {
                                /* This is queue action. */
                                rss_desc_v[i] = wks->rss_desc;
                                rss_desc_v[i].key_len = 0;
@@ -4696,24 +4790,24 @@ get_meter_sub_policy(struct rte_eth_dev *dev,
                                rss_desc_v[i].queue =
                                        &final_policy->act_cnt[i].queue;
                                rss_desc_v[i].queue_num = 1;
+                               rss_desc[i] = &rss_desc_v[i];
+                       } else {
+                               rss_desc[i] = NULL;
                        }
-                       rss_desc[i] = &rss_desc_v[i];
                }
                sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
                                                flow, policy, rss_desc);
        } else {
                enum mlx5_meter_domain mtr_domain =
                        attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
-                               attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
-                                       MLX5_MTR_DOMAIN_INGRESS;
+                               (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
+                                               MLX5_MTR_DOMAIN_INGRESS);
                sub_policy = policy->sub_policys[mtr_domain][0];
        }
-       if (!sub_policy) {
+       if (!sub_policy)
                rte_flow_error_set(error, EINVAL,
-                       RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
-                       "Failed to get meter sub-policy.");
-               goto exit;
-       }
+                                  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                                  "Failed to get meter sub-policy.");
 exit:
        return sub_policy;
 }
@@ -4911,8 +5005,8 @@ flow_meter_split_prep(struct rte_eth_dev *dev,
                } else {
                        enum mlx5_meter_domain mtr_domain =
                        attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
-                               attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
-                                       MLX5_MTR_DOMAIN_INGRESS;
+                               (attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
+                                               MLX5_MTR_DOMAIN_INGRESS);
 
                        sub_policy =
                        &priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
@@ -4928,8 +5022,8 @@ flow_meter_split_prep(struct rte_eth_dev *dev,
        actions_pre++;
        if (!tag_action)
                return rte_flow_error_set(error, ENOMEM,
-                                       RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
-                                       "No tag action space.");
+                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                         NULL, "No tag action space.");
        if (!mtr_flow_id) {
                tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
                goto exit;
@@ -5201,6 +5295,7 @@ flow_check_match_action(const struct rte_flow_action actions[],
                        int *modify_after_mirror)
 {
        const struct rte_flow_action_sample *sample;
+       const struct rte_flow_action_raw_decap *decap;
        int actions_n = 0;
        uint32_t ratio = 0;
        int sub_type = 0;
@@ -5253,12 +5348,29 @@ flow_check_match_action(const struct rte_flow_action actions[],
                case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
                case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
                case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
-               case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
                case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
                case RTE_FLOW_ACTION_TYPE_METER:
                        if (fdb_mirror)
                                *modify_after_mirror = 1;
                        break;
+               case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
+                       decap = actions->conf;
+                       while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
+                               ;
+                       actions_n++;
+                       if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
+                               const struct rte_flow_action_raw_encap *encap =
+                                                               actions->conf;
+                               if (decap->size <=
+                                       MLX5_ENCAPSULATION_DECISION_SIZE &&
+                                   encap->size >
+                                       MLX5_ENCAPSULATION_DECISION_SIZE)
+                                       /* L3 encap. */
+                                       break;
+                       }
+                       if (fdb_mirror)
+                               *modify_after_mirror = 1;
+                       break;
                default:
                        break;
                }
@@ -6192,7 +6304,7 @@ flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
        int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
        union {
                struct mlx5_flow_expand_rss buf;
-               uint8_t buffer[2048];
+               uint8_t buffer[4096];
        } expand_buffer;
        union {
                struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
@@ -6283,7 +6395,7 @@ flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
        if (rss && rss->types) {
                unsigned int graph_root;
 
-               graph_root = find_graph_root(items, rss->level);
+               graph_root = find_graph_root(rss->level);
                ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
                                           items, rss->types,
                                           mlx5_support_expansion, graph_root);
@@ -7152,14 +7264,14 @@ mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
                        struct rte_flow_attr *attr,
                        bool *is_rss,
                        uint8_t *domain_bitmap,
-                       bool *is_def_policy,
+                       uint8_t *policy_mode,
                        struct rte_mtr_error *error)
 {
        const struct mlx5_flow_driver_ops *fops;
 
        fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
-       return fops->validate_mtr_acts(dev, actions, attr,
-                       is_rss, domain_bitmap, is_def_policy, error);
+       return fops->validate_mtr_acts(dev, actions, attr, is_rss,
+                                      domain_bitmap, policy_mode, error);
 }
 
 /**
@@ -8622,7 +8734,7 @@ tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
 {
        struct mlx5_priv *priv = dev->data->dev_private;
        struct mlx5_dev_ctx_shared *sh = priv->sh;
-       struct mlx5_hlist_entry *he;
+       struct mlx5_list_entry *he;
        union tunnel_offload_mark mbits = { .val = mark };
        union mlx5_flow_tbl_key table_key = {
                {
@@ -8634,16 +8746,20 @@ tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
                        .is_egress = 0,
                }
        };
-       he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
+       struct mlx5_flow_cb_ctx ctx = {
+               .data = &table_key.v64,
+       };
+
+       he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
        return he ?
               container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
 }
 
 static void
-mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
-                                  struct mlx5_hlist_entry *entry)
+mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
+                                  struct mlx5_list_entry *entry)
 {
-       struct mlx5_dev_ctx_shared *sh = list->ctx;
+       struct mlx5_dev_ctx_shared *sh = tool_ctx;
        struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
 
        mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
@@ -8652,26 +8768,26 @@ mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
 }
 
 static int
-mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
-                                 struct mlx5_hlist_entry *entry,
-                                 uint64_t key, void *cb_ctx __rte_unused)
+mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
+                                 struct mlx5_list_entry *entry, void *cb_ctx)
 {
+       struct mlx5_flow_cb_ctx *ctx = cb_ctx;
        union tunnel_tbl_key tbl = {
-               .val = key,
+               .val = *(uint64_t *)(ctx->data),
        };
        struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
 
        return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
 }
 
-static struct mlx5_hlist_entry *
-mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
-                                  void *ctx __rte_unused)
+static struct mlx5_list_entry *
+mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
 {
-       struct mlx5_dev_ctx_shared *sh = list->ctx;
+       struct mlx5_dev_ctx_shared *sh = tool_ctx;
+       struct mlx5_flow_cb_ctx *ctx = cb_ctx;
        struct tunnel_tbl_entry *tte;
        union tunnel_tbl_key tbl = {
-               .val = key,
+               .val = *(uint64_t *)(ctx->data),
        };
 
        tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
@@ -8700,13 +8816,36 @@ err:
        return NULL;
 }
 
+static struct mlx5_list_entry *
+mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
+                                 struct mlx5_list_entry *oentry,
+                                 void *cb_ctx __rte_unused)
+{
+       struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
+                                                  0, SOCKET_ID_ANY);
+
+       if (!tte)
+               return NULL;
+       memcpy(tte, oentry, sizeof(*tte));
+       return &tte->hash;
+}
+
+static void
+mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
+                                      struct mlx5_list_entry *entry)
+{
+       struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
+
+       mlx5_free(tte);
+}
+
 static uint32_t
 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
                                const struct mlx5_flow_tunnel *tunnel,
                                uint32_t group, uint32_t *table,
                                struct rte_flow_error *error)
 {
-       struct mlx5_hlist_entry *he;
+       struct mlx5_list_entry *he;
        struct tunnel_tbl_entry *tte;
        union tunnel_tbl_key key = {
                .tunnel_id = tunnel ? tunnel->tunnel_id : 0,
@@ -8714,9 +8853,12 @@ tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
        };
        struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
        struct mlx5_hlist *group_hash;
+       struct mlx5_flow_cb_ctx ctx = {
+               .data = &key.val,
+       };
 
        group_hash = tunnel ? tunnel->groups : thub->groups;
-       he = mlx5_hlist_register(group_hash, key.val, NULL);
+       he = mlx5_hlist_register(group_hash, key.val, &ctx);
        if (!he)
                return rte_flow_error_set(error, EINVAL,
                                          RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
@@ -8830,15 +8972,17 @@ mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
                DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
                return NULL;
        }
-       tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
+       tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
+                                          priv->sh,
                                           mlx5_flow_tunnel_grp2tbl_create_cb,
                                           mlx5_flow_tunnel_grp2tbl_match_cb,
-                                          mlx5_flow_tunnel_grp2tbl_remove_cb);
+                                          mlx5_flow_tunnel_grp2tbl_remove_cb,
+                                          mlx5_flow_tunnel_grp2tbl_clone_cb,
+                                       mlx5_flow_tunnel_grp2tbl_clone_free_cb);
        if (!tunnel->groups) {
                mlx5_ipool_free(ipool, id);
                return NULL;
        }
-       tunnel->groups->ctx = priv->sh;
        /* initiate new PMD tunnel */
        memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
        tunnel->tunnel_id = id;
@@ -8937,16 +9081,17 @@ int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
                return -ENOMEM;
        LIST_INIT(&thub->tunnels);
        rte_spinlock_init(&thub->sl);
-       thub->groups = mlx5_hlist_create("flow groups",
-                                        rte_align32pow2(MLX5_MAX_TABLES), 0,
-                                        0, mlx5_flow_tunnel_grp2tbl_create_cb,
+       thub->groups = mlx5_hlist_create("flow groups", 64,
+                                        false, true, sh,
+                                        mlx5_flow_tunnel_grp2tbl_create_cb,
                                         mlx5_flow_tunnel_grp2tbl_match_cb,
-                                        mlx5_flow_tunnel_grp2tbl_remove_cb);
+                                        mlx5_flow_tunnel_grp2tbl_remove_cb,
+                                        mlx5_flow_tunnel_grp2tbl_clone_cb,
+                                       mlx5_flow_tunnel_grp2tbl_clone_free_cb);
        if (!thub->groups) {
                err = -rte_errno;
                goto err;
        }
-       thub->groups->ctx = sh;
        sh->tunnel_hub = thub;
 
        return 0;