net/mlx5: add flow VXLAN item
authorNelio Laranjeiro <nelio.laranjeiro@6wind.com>
Thu, 12 Jul 2018 09:31:03 +0000 (11:31 +0200)
committerShahaf Shuler <shahafs@mellanox.com>
Thu, 12 Jul 2018 10:12:24 +0000 (12:12 +0200)
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
drivers/net/mlx5/mlx5_flow.c
drivers/net/mlx5/mlx5_rxtx.h

index eecc513..54d85af 100644 (file)
@@ -51,6 +51,9 @@ extern const struct eth_dev_ops mlx5_dev_ops_isolate;
 #define MLX5_FLOW_LAYER_INNER_L4_TCP (1u << 10)
 #define MLX5_FLOW_LAYER_INNER_VLAN (1u << 11)
 
+/* Pattern tunnel Layer bits. */
+#define MLX5_FLOW_LAYER_VXLAN (1u << 12)
+
 /* Outer Masks. */
 #define MLX5_FLOW_LAYER_OUTER_L3 \
        (MLX5_FLOW_LAYER_OUTER_L3_IPV4 | MLX5_FLOW_LAYER_OUTER_L3_IPV6)
@@ -61,7 +64,7 @@ extern const struct eth_dev_ops mlx5_dev_ops_isolate;
         MLX5_FLOW_LAYER_OUTER_L4)
 
 /* Tunnel Masks. */
-#define MLX5_FLOW_LAYER_TUNNEL 0
+#define MLX5_FLOW_LAYER_TUNNEL MLX5_FLOW_LAYER_VXLAN
 
 /* Inner Masks. */
 #define MLX5_FLOW_LAYER_INNER_L3 \
@@ -98,6 +101,7 @@ enum mlx5_expansion {
        MLX5_EXPANSION_OUTER_IPV6,
        MLX5_EXPANSION_OUTER_IPV6_UDP,
        MLX5_EXPANSION_OUTER_IPV6_TCP,
+       MLX5_EXPANSION_VXLAN,
        MLX5_EXPANSION_ETH,
        MLX5_EXPANSION_IPV4,
        MLX5_EXPANSION_IPV4_UDP,
@@ -136,6 +140,7 @@ static const struct rte_flow_expand_node mlx5_support_expansion[] = {
                        ETH_RSS_NONFRAG_IPV4_OTHER,
        },
        [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
+               .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN),
                .type = RTE_FLOW_ITEM_TYPE_UDP,
                .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
        },
@@ -152,6 +157,7 @@ static const struct rte_flow_expand_node mlx5_support_expansion[] = {
                        ETH_RSS_NONFRAG_IPV6_OTHER,
        },
        [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
+               .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN),
                .type = RTE_FLOW_ITEM_TYPE_UDP,
                .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
        },
@@ -159,6 +165,10 @@ static const struct rte_flow_expand_node mlx5_support_expansion[] = {
                .type = RTE_FLOW_ITEM_TYPE_TCP,
                .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
        },
+       [MLX5_EXPANSION_VXLAN] = {
+               .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
+               .type = RTE_FLOW_ITEM_TYPE_VXLAN,
+       },
        [MLX5_EXPANSION_ETH] = {
                .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
                                                 MLX5_EXPANSION_IPV6),
@@ -226,6 +236,8 @@ struct rte_flow {
        struct mlx5_flow_verbs *cur_verbs;
        /**< Current Verbs flow structure being filled. */
        struct rte_flow_action_rss rss;/**< RSS context. */
+       uint32_t tunnel_ptype;
+       /**< Store tunnel packet type data to store in Rx queue. */
        uint8_t key[MLX5_RSS_HASH_KEY_LEN]; /**< RSS hash key. */
        uint16_t (*queue)[]; /**< Destination queues to redirect traffic to. */
 };
@@ -1161,6 +1173,103 @@ mlx5_flow_item_tcp(const struct rte_flow_item *item, struct rte_flow *flow,
        return size;
 }
 
+/**
+ * Convert the @p item into a Verbs specification after ensuring the NIC
+ * will understand and process it correctly.
+ * If the necessary size for the conversion is greater than the @p flow_size,
+ * nothing is written in @p flow, the validation is still performed.
+ *
+ * @param[in] item
+ *   Item specification.
+ * @param[in, out] flow
+ *   Pointer to flow structure.
+ * @param[in] flow_size
+ *   Size in bytes of the available space in @p flow, if too small, nothing is
+ *   written.
+ * @param[out] error
+ *   Pointer to error structure.
+ *
+ * @return
+ *   On success the number of bytes consumed/necessary, if the returned value
+ *   is lesser or equal to @p flow_size, the @p item has fully been converted,
+ *   otherwise another call with this returned memory size should be done.
+ *   On error, a negative errno value is returned and rte_errno is set.
+ */
+static int
+mlx5_flow_item_vxlan(const struct rte_flow_item *item, struct rte_flow *flow,
+                    const size_t flow_size, struct rte_flow_error *error)
+{
+       const struct rte_flow_item_vxlan *spec = item->spec;
+       const struct rte_flow_item_vxlan *mask = item->mask;
+       unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
+       struct ibv_flow_spec_tunnel vxlan = {
+               .type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
+               .size = size,
+       };
+       int ret;
+       union vni {
+               uint32_t vlan_id;
+               uint8_t vni[4];
+       } id = { .vlan_id = 0, };
+
+       if (flow->layers & MLX5_FLOW_LAYER_TUNNEL)
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ITEM,
+                                         item,
+                                         "a tunnel is already present");
+       /*
+        * Verify only UDPv4 is present as defined in
+        * https://tools.ietf.org/html/rfc7348
+        */
+       if (!(flow->layers & MLX5_FLOW_LAYER_OUTER_L4_UDP))
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ITEM,
+                                         item,
+                                         "no outer UDP layer found");
+       if (!mask)
+               mask = &rte_flow_item_vxlan_mask;
+       ret = mlx5_flow_item_acceptable
+               (item, (const uint8_t *)mask,
+                (const uint8_t *)&rte_flow_item_vxlan_mask,
+                sizeof(struct rte_flow_item_vxlan), error);
+       if (ret < 0)
+               return ret;
+       if (spec) {
+               memcpy(&id.vni[1], spec->vni, 3);
+               vxlan.val.tunnel_id = id.vlan_id;
+               memcpy(&id.vni[1], mask->vni, 3);
+               vxlan.mask.tunnel_id = id.vlan_id;
+               /* Remove unwanted bits from values. */
+               vxlan.val.tunnel_id &= vxlan.mask.tunnel_id;
+       }
+       /*
+        * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
+        * only this layer is defined in the Verbs specification it is
+        * interpreted as wildcard and all packets will match this
+        * rule, if it follows a full stack layer (ex: eth / ipv4 /
+        * udp), all packets matching the layers before will also
+        * match this rule.  To avoid such situation, VNI 0 is
+        * currently refused.
+        */
+       if (!vxlan.val.tunnel_id)
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ITEM,
+                                         item,
+                                         "VXLAN vni cannot be 0");
+       if (!(flow->layers & MLX5_FLOW_LAYER_OUTER))
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ITEM,
+                                         item,
+                                         "VXLAN tunnel must be fully defined");
+       if (size <= flow_size) {
+               mlx5_flow_spec_verbs_add(flow, &vxlan, size);
+               flow->cur_verbs->attr->priority = MLX5_PRIORITY_MAP_L2;
+       }
+       flow->layers |= MLX5_FLOW_LAYER_VXLAN;
+       flow->tunnel_ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP;
+       return size;
+}
+
 /**
  * Convert the @p pattern into a Verbs specifications after ensuring the NIC
  * will understand and process it correctly.
@@ -1218,6 +1327,10 @@ mlx5_flow_items(const struct rte_flow_item pattern[],
                case RTE_FLOW_ITEM_TYPE_TCP:
                        ret = mlx5_flow_item_tcp(pattern, flow, remain, error);
                        break;
+               case RTE_FLOW_ITEM_TYPE_VXLAN:
+                       ret = mlx5_flow_item_vxlan(pattern, flow, remain,
+                                                  error);
+                       break;
                default:
                        return rte_flow_error_set(error, ENOTSUP,
                                                  RTE_FLOW_ERROR_TYPE_ITEM,
@@ -1833,7 +1946,7 @@ mlx5_flow_merge(struct rte_eth_dev *dev, struct rte_flow *flow,
 }
 
 /**
- * Mark the Rx queues mark flag if the flow has a mark or flag modifier.
+ * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the flow.
  *
  * @param[in] dev
  *   Pointer to Ethernet device.
@@ -1841,28 +1954,34 @@ mlx5_flow_merge(struct rte_eth_dev *dev, struct rte_flow *flow,
  *   Pointer to flow structure.
  */
 static void
-mlx5_flow_rxq_mark_set(struct rte_eth_dev *dev, struct rte_flow *flow)
+mlx5_flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
 {
        struct priv *priv = dev->data->dev_private;
+       const int mark = !!(flow->modifier &
+                           (MLX5_FLOW_MOD_FLAG | MLX5_FLOW_MOD_MARK));
+       const int tunnel = !!(flow->layers & MLX5_FLOW_LAYER_TUNNEL);
+       unsigned int i;
 
-       if (flow->modifier & (MLX5_FLOW_MOD_FLAG | MLX5_FLOW_MOD_MARK)) {
-               unsigned int i;
-
-               for (i = 0; i != flow->rss.queue_num; ++i) {
-                       int idx = (*flow->queue)[i];
-                       struct mlx5_rxq_ctrl *rxq_ctrl =
-                               container_of((*priv->rxqs)[idx],
-                                            struct mlx5_rxq_ctrl, rxq);
+       for (i = 0; i != flow->rss.queue_num; ++i) {
+               int idx = (*flow->queue)[i];
+               struct mlx5_rxq_ctrl *rxq_ctrl =
+                       container_of((*priv->rxqs)[idx],
+                                    struct mlx5_rxq_ctrl, rxq);
 
+               if (mark) {
                        rxq_ctrl->rxq.mark = 1;
                        rxq_ctrl->flow_mark_n++;
                }
+               if (tunnel) {
+                       rxq_ctrl->rxq.tunnel = flow->tunnel_ptype;
+                       rxq_ctrl->flow_vxlan_n++;
+               }
        }
 }
 
 /**
- * Clear the Rx queue mark associated with the @p flow if no other flow uses
- * it with a mark request.
+ * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
+ * @p flow if no other flow uses it with the same kind of request.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -1870,33 +1989,41 @@ mlx5_flow_rxq_mark_set(struct rte_eth_dev *dev, struct rte_flow *flow)
  *   Pointer to the flow.
  */
 static void
-mlx5_flow_rxq_mark_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
+mlx5_flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
 {
        struct priv *priv = dev->data->dev_private;
+       const int mark = !!(flow->modifier &
+                           (MLX5_FLOW_MOD_FLAG | MLX5_FLOW_MOD_MARK));
+       const int tunnel = !!(flow->layers & MLX5_FLOW_LAYER_TUNNEL);
+       unsigned int i;
 
-       if (flow->modifier & (MLX5_FLOW_MOD_FLAG | MLX5_FLOW_MOD_MARK)) {
-               unsigned int i;
-
-               for (i = 0; i != flow->rss.queue_num; ++i) {
-                       int idx = (*flow->queue)[i];
-                       struct mlx5_rxq_ctrl *rxq_ctrl =
-                               container_of((*priv->rxqs)[idx],
-                                            struct mlx5_rxq_ctrl, rxq);
+       assert(dev->data->dev_started);
+       for (i = 0; i != flow->rss.queue_num; ++i) {
+               int idx = (*flow->queue)[i];
+               struct mlx5_rxq_ctrl *rxq_ctrl =
+                       container_of((*priv->rxqs)[idx],
+                                    struct mlx5_rxq_ctrl, rxq);
 
+               if (mark) {
                        rxq_ctrl->flow_mark_n--;
                        rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
                }
+               if (tunnel) {
+                       rxq_ctrl->flow_vxlan_n++;
+                       if (!rxq_ctrl->flow_vxlan_n)
+                               rxq_ctrl->rxq.tunnel = 0;
+               }
        }
 }
 
 /**
- * Clear the mark bit in all Rx queues.
+ * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
  *
  * @param dev
  *   Pointer to Ethernet device.
  */
 static void
-mlx5_flow_rxq_mark_clear(struct rte_eth_dev *dev)
+mlx5_flow_rxq_flags_clear(struct rte_eth_dev *dev)
 {
        struct priv *priv = dev->data->dev_private;
        unsigned int i;
@@ -1911,6 +2038,8 @@ mlx5_flow_rxq_mark_clear(struct rte_eth_dev *dev)
                                        struct mlx5_rxq_ctrl, rxq);
                rxq_ctrl->flow_mark_n = 0;
                rxq_ctrl->rxq.mark = 0;
+               rxq_ctrl->flow_vxlan_n = 0;
+               rxq_ctrl->rxq.tunnel = 0;
                ++idx;
        }
 }
@@ -2106,7 +2235,7 @@ mlx5_flow_list_create(struct rte_eth_dev *dev,
                }
        }
        TAILQ_INSERT_TAIL(list, flow, next);
-       mlx5_flow_rxq_mark_set(dev, flow);
+       mlx5_flow_rxq_flags_set(dev, flow);
        return flow;
 }
 
@@ -2144,7 +2273,12 @@ mlx5_flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
 {
        mlx5_flow_remove(dev, flow);
        TAILQ_REMOVE(list, flow, next);
-       mlx5_flow_rxq_mark_trim(dev, flow);
+       /*
+        * Update RX queue flags only if port is started, otherwise it is
+        * already clean.
+        */
+       if (dev->data->dev_started)
+               mlx5_flow_rxq_flags_trim(dev, flow);
        rte_free(flow);
 }
 
@@ -2182,7 +2316,7 @@ mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
 
        TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
                mlx5_flow_remove(dev, flow);
-       mlx5_flow_rxq_mark_clear(dev);
+       mlx5_flow_rxq_flags_clear(dev);
 }
 
 /**
@@ -2207,7 +2341,7 @@ mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
                ret = mlx5_flow_apply(dev, flow, &error);
                if (ret < 0)
                        goto error;
-               mlx5_flow_rxq_mark_set(dev, flow);
+               mlx5_flow_rxq_flags_set(dev, flow);
        }
        return 0;
 error:
index 808118e..ae9b564 100644 (file)
@@ -139,6 +139,7 @@ struct mlx5_rxq_ctrl {
        unsigned int irq:1; /* Whether IRQ is enabled. */
        uint16_t idx; /* Queue index. */
        uint32_t flow_mark_n; /* Number of Mark/Flag flows using this Queue. */
+       uint32_t flow_vxlan_n; /* Number of VXLAN flows using this queue. */
 };
 
 /* Indirection table. */