net/mlx5: add VXLAN encap action to Direct Verbs
authorDekel Peled <dekelp@mellanox.com>
Thu, 1 Nov 2018 09:37:28 +0000 (11:37 +0200)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 5 Nov 2018 14:01:25 +0000 (15:01 +0100)
This patch implements the VXLAN encap action in DV flow for MLX5 PMD.

Signed-off-by: Dekel Peled <dekelp@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
drivers/net/mlx5/mlx5_flow.h
drivers/net/mlx5/mlx5_flow_dv.c

index 838cf75..cb57b1a 100644 (file)
@@ -92,6 +92,7 @@
 #define MLX5_FLOW_ACTION_DEC_TTL (1u << 19)
 #define MLX5_FLOW_ACTION_SET_MAC_SRC (1u << 20)
 #define MLX5_FLOW_ACTION_SET_MAC_DST (1u << 21)
+#define MLX5_FLOW_ACTION_VXLAN_ENCAP (1u << 22)
 
 #define MLX5_FLOW_FATE_ACTIONS \
        (MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE | MLX5_FLOW_ACTION_RSS)
@@ -181,6 +182,8 @@ struct mlx5_flow_dv {
 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
        struct mlx5dv_flow_action_attr actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS];
        /**< Action list. */
+       struct ibv_flow_action *encap_decap_verbs_action;
+       /**< Verbs encap/decap object. */
 #endif
        int actions_n; /**< number of actions. */
 };
index 0e1f715..349b6b5 100644 (file)
@@ -34,6 +34,8 @@
 
 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
 
+#define MLX5_ENCAP_MAX_LEN 132
+
 /**
  * Validate META item.
  *
@@ -95,6 +97,305 @@ flow_dv_validate_item_meta(struct rte_eth_dev *dev,
        return 0;
 }
 
+/**
+ * Validate the L2 encap action.
+ *
+ * @param[in] action_flags
+ *   Holds the actions detected until now.
+ * @param[in] action
+ *   Pointer to the encap action.
+ * @param[in] attr
+ *   Pointer to flow attributes
+ * @param[out] error
+ *   Pointer to error structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+flow_dv_validate_action_l2_encap(uint64_t action_flags,
+                                const struct rte_flow_action *action,
+                                const struct rte_flow_attr *attr,
+                                struct rte_flow_error *error)
+{
+       if (!(action->conf))
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ACTION, action,
+                                         "configuration cannot be null");
+       if (action_flags & MLX5_FLOW_ACTION_DROP)
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ACTION, NULL,
+                                         "can't drop and encap in same flow");
+       if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP)
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ACTION, NULL,
+                                         "can only have a single encap"
+                                         " action in a flow");
+       if (attr->ingress)
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
+                                         NULL,
+                                         "encap action not supported for "
+                                         "ingress");
+       return 0;
+}
+
+/**
+ * Get the size of specific rte_flow_item_type
+ *
+ * @param[in] item_type
+ *   Tested rte_flow_item_type.
+ *
+ * @return
+ *   sizeof struct item_type, 0 if void or irrelevant.
+ */
+static size_t
+flow_dv_get_item_len(const enum rte_flow_item_type item_type)
+{
+       size_t retval;
+
+       switch (item_type) {
+       case RTE_FLOW_ITEM_TYPE_ETH:
+               retval = sizeof(struct rte_flow_item_eth);
+               break;
+       case RTE_FLOW_ITEM_TYPE_VLAN:
+               retval = sizeof(struct rte_flow_item_vlan);
+               break;
+       case RTE_FLOW_ITEM_TYPE_IPV4:
+               retval = sizeof(struct rte_flow_item_ipv4);
+               break;
+       case RTE_FLOW_ITEM_TYPE_IPV6:
+               retval = sizeof(struct rte_flow_item_ipv6);
+               break;
+       case RTE_FLOW_ITEM_TYPE_UDP:
+               retval = sizeof(struct rte_flow_item_udp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_TCP:
+               retval = sizeof(struct rte_flow_item_tcp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_VXLAN:
+               retval = sizeof(struct rte_flow_item_vxlan);
+               break;
+       case RTE_FLOW_ITEM_TYPE_GRE:
+               retval = sizeof(struct rte_flow_item_gre);
+               break;
+       case RTE_FLOW_ITEM_TYPE_NVGRE:
+               retval = sizeof(struct rte_flow_item_nvgre);
+               break;
+       case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
+               retval = sizeof(struct rte_flow_item_vxlan_gpe);
+               break;
+       case RTE_FLOW_ITEM_TYPE_MPLS:
+               retval = sizeof(struct rte_flow_item_mpls);
+               break;
+       case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
+       default:
+               retval = 0;
+               break;
+       }
+       return retval;
+}
+
+#define MLX5_ENCAP_IPV4_VERSION                0x40
+#define MLX5_ENCAP_IPV4_IHL_MIN                0x05
+#define MLX5_ENCAP_IPV4_TTL_DEF                0x40
+#define MLX5_ENCAP_IPV6_VTC_FLOW       0x60000000
+#define MLX5_ENCAP_IPV6_HOP_LIMIT      0xff
+#define MLX5_ENCAP_VXLAN_FLAGS         0x08000000
+#define MLX5_ENCAP_VXLAN_GPE_FLAGS     0x04
+
+/**
+ * Convert the encap action data from list of rte_flow_item to raw buffer
+ *
+ * @param[in] items
+ *   Pointer to rte_flow_item objects list.
+ * @param[out] buf
+ *   Pointer to the output buffer.
+ * @param[out] size
+ *   Pointer to the output buffer size.
+ * @param[out] error
+ *   Pointer to the error structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
+                          size_t *size, struct rte_flow_error *error)
+{
+       struct ether_hdr *eth = NULL;
+       struct vlan_hdr *vlan = NULL;
+       struct ipv4_hdr *ipv4 = NULL;
+       struct ipv6_hdr *ipv6 = NULL;
+       struct udp_hdr *udp = NULL;
+       struct vxlan_hdr *vxlan = NULL;
+       struct vxlan_gpe_hdr *vxlan_gpe = NULL;
+       size_t len;
+       size_t temp_size = 0;
+
+       if (!items)
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ACTION,
+                                         NULL, "invalid empty data");
+       for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
+               len = flow_dv_get_item_len(items->type);
+               if (len + temp_size > MLX5_ENCAP_MAX_LEN)
+                       return rte_flow_error_set(error, EINVAL,
+                                                 RTE_FLOW_ERROR_TYPE_ACTION,
+                                                 (void *)items->type,
+                                                 "items total size is too big"
+                                                 " for encap action");
+               rte_memcpy((void *)&buf[temp_size], items->spec, len);
+               switch (items->type) {
+               case RTE_FLOW_ITEM_TYPE_ETH:
+                       eth = (struct ether_hdr *)&buf[temp_size];
+                       break;
+               case RTE_FLOW_ITEM_TYPE_VLAN:
+                       vlan = (struct vlan_hdr *)&buf[temp_size];
+                       if (!eth)
+                               return rte_flow_error_set(error, EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               (void *)items->type,
+                                               "eth header not found");
+                       if (!eth->ether_type)
+                               eth->ether_type = RTE_BE16(ETHER_TYPE_VLAN);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_IPV4:
+                       ipv4 = (struct ipv4_hdr *)&buf[temp_size];
+                       if (!vlan && !eth)
+                               return rte_flow_error_set(error, EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               (void *)items->type,
+                                               "neither eth nor vlan"
+                                               " header found");
+                       if (vlan && !vlan->eth_proto)
+                               vlan->eth_proto = RTE_BE16(ETHER_TYPE_IPv4);
+                       else if (eth && !eth->ether_type)
+                               eth->ether_type = RTE_BE16(ETHER_TYPE_IPv4);
+                       if (!ipv4->version_ihl)
+                               ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
+                                                   MLX5_ENCAP_IPV4_IHL_MIN;
+                       if (!ipv4->time_to_live)
+                               ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
+                       break;
+               case RTE_FLOW_ITEM_TYPE_IPV6:
+                       ipv6 = (struct ipv6_hdr *)&buf[temp_size];
+                       if (!vlan && !eth)
+                               return rte_flow_error_set(error, EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               (void *)items->type,
+                                               "neither eth nor vlan"
+                                               " header found");
+                       if (vlan && !vlan->eth_proto)
+                               vlan->eth_proto = RTE_BE16(ETHER_TYPE_IPv6);
+                       else if (eth && !eth->ether_type)
+                               eth->ether_type = RTE_BE16(ETHER_TYPE_IPv6);
+                       if (!ipv6->vtc_flow)
+                               ipv6->vtc_flow =
+                                       RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
+                       if (!ipv6->hop_limits)
+                               ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
+                       break;
+               case RTE_FLOW_ITEM_TYPE_UDP:
+                       udp = (struct udp_hdr *)&buf[temp_size];
+                       if (!ipv4 && !ipv6)
+                               return rte_flow_error_set(error, EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               (void *)items->type,
+                                               "ip header not found");
+                       if (ipv4 && !ipv4->next_proto_id)
+                               ipv4->next_proto_id = IPPROTO_UDP;
+                       else if (ipv6 && !ipv6->proto)
+                               ipv6->proto = IPPROTO_UDP;
+                       break;
+               case RTE_FLOW_ITEM_TYPE_VXLAN:
+                       vxlan = (struct vxlan_hdr *)&buf[temp_size];
+                       if (!udp)
+                               return rte_flow_error_set(error, EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               (void *)items->type,
+                                               "udp header not found");
+                       if (!udp->dst_port)
+                               udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
+                       if (!vxlan->vx_flags)
+                               vxlan->vx_flags =
+                                       RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
+                       vxlan_gpe = (struct vxlan_gpe_hdr *)&buf[temp_size];
+                       if (!udp)
+                               return rte_flow_error_set(error, EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               (void *)items->type,
+                                               "udp header not found");
+                       if (!vxlan_gpe->proto)
+                               return rte_flow_error_set(error, EINVAL,
+                                               RTE_FLOW_ERROR_TYPE_ACTION,
+                                               (void *)items->type,
+                                               "next protocol not found");
+                       if (!udp->dst_port)
+                               udp->dst_port =
+                                       RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
+                       if (!vxlan_gpe->vx_flags)
+                               vxlan_gpe->vx_flags =
+                                               MLX5_ENCAP_VXLAN_GPE_FLAGS;
+                       break;
+               case RTE_FLOW_ITEM_TYPE_VOID:
+                       break;
+               default:
+                       return rte_flow_error_set(error, EINVAL,
+                                                 RTE_FLOW_ERROR_TYPE_ACTION,
+                                                 (void *)items->type,
+                                                 "unsupported item type");
+                       break;
+               }
+               temp_size += len;
+       }
+       *size = temp_size;
+       return 0;
+}
+
+/**
+ * Convert L2 encap action to DV specification.
+ *
+ * @param[in] dev
+ *   Pointer to rte_eth_dev structure.
+ * @param[in] action
+ *   Pointer to action structure.
+ * @param[out] error
+ *   Pointer to the error structure.
+ *
+ * @return
+ *   Pointer to action on success, NULL otherwise and rte_errno is set.
+ */
+static struct ibv_flow_action *
+flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
+                              const struct rte_flow_action *action,
+                              struct rte_flow_error *error)
+{
+       struct ibv_flow_action *verbs_action = NULL;
+       const struct rte_flow_item *encap_data;
+       struct priv *priv = dev->data->dev_private;
+       uint8_t buf[MLX5_ENCAP_MAX_LEN];
+       size_t size = 0;
+       int convert_result = 0;
+
+       encap_data = ((const struct rte_flow_action_vxlan_encap *)
+                                               action->conf)->definition;
+       convert_result = flow_dv_convert_encap_data(encap_data, buf,
+                                                   &size, error);
+       if (convert_result)
+               return NULL;
+       verbs_action = mlx5_glue->dv_create_flow_action_packet_reformat
+               (priv->ctx, size, buf,
+                MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
+                MLX5DV_FLOW_TABLE_TYPE_NIC_TX);
+       if (!verbs_action)
+               rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
+                                  NULL, "cannot create L2 encap action");
+       return verbs_action;
+}
+
 /**
  * Verify the @p attributes will be correctly understood by the NIC and store
  * them in the @p flow if everything is correct.
@@ -339,6 +640,16 @@ flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
                        action_flags |= MLX5_FLOW_ACTION_COUNT;
                        ++actions_n;
                        break;
+               case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
+                       ret = flow_dv_validate_action_l2_encap(action_flags,
+                                                              actions, attr,
+                                                              error);
+                       if (ret < 0)
+                               return ret;
+                       action_flags |= MLX5_FLOW_ACTION_VXLAN_ENCAP;
+                       ++actions_n;
+                       break;
+
                default:
                        return rte_flow_error_set(error, ENOTSUP,
                                                  RTE_FLOW_ERROR_TYPE_ACTION,
@@ -1045,14 +1356,23 @@ flow_dv_create_item(void *matcher, void *key,
 /**
  * Store the requested actions in an array.
  *
+ * @param[in] dev
+ *   Pointer to rte_eth_dev structure.
  * @param[in] action
  *   Flow action to translate.
  * @param[in, out] dev_flow
  *   Pointer to the mlx5_flow.
+ * @param[out] error
+ *   Pointer to the error structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
-static void
-flow_dv_create_action(const struct rte_flow_action *action,
-                     struct mlx5_flow *dev_flow)
+static int
+flow_dv_create_action(struct rte_eth_dev *dev,
+                     const struct rte_flow_action *action,
+                     struct mlx5_flow *dev_flow,
+                     struct rte_flow_error *error)
 {
        const struct rte_flow_action_queue *queue;
        const struct rte_flow_action_rss *rss;
@@ -1100,10 +1420,24 @@ flow_dv_create_action(const struct rte_flow_action *action,
                /* Added to array only in apply since we need the QP */
                flow->actions |= MLX5_FLOW_ACTION_RSS;
                break;
+       case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
+               dev_flow->dv.actions[actions_n].type =
+                       MLX5DV_FLOW_ACTION_IBV_FLOW_ACTION;
+               dev_flow->dv.actions[actions_n].action =
+                               flow_dv_create_action_l2_encap(dev, action,
+                                                              error);
+               if (!(dev_flow->dv.actions[actions_n].action))
+                       return -rte_errno;
+               dev_flow->dv.encap_decap_verbs_action =
+                       dev_flow->dv.actions[actions_n].action;
+               flow->actions |= MLX5_FLOW_ACTION_VXLAN_ENCAP;
+               actions_n++;
+               break;
        default:
                break;
        }
        dev_flow->dv.actions_n = actions_n;
+       return 0;
 }
 
 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
@@ -1219,7 +1553,6 @@ flow_dv_matcher_register(struct rte_eth_dev *dev,
        return 0;
 }
 
-
 /**
  * Fill the flow with DV spec.
  *
@@ -1274,7 +1607,8 @@ flow_dv_translate(struct rte_eth_dev *dev,
        if (flow_dv_matcher_register(dev, &matcher, dev_flow, error))
                return -rte_errno;
        for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
-               flow_dv_create_action(actions, dev_flow);
+               if (flow_dv_create_action(dev, actions, dev_flow, error))
+                       return -rte_errno;
        return 0;
 }
 
@@ -1459,6 +1793,11 @@ flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
                LIST_REMOVE(dev_flow, next);
                if (dev_flow->dv.matcher)
                        flow_dv_matcher_release(dev, dev_flow);
+               if (dev_flow->dv.encap_decap_verbs_action) {
+                       claim_zero(mlx5_glue->destroy_flow_action
+                               (dev_flow->dv.encap_decap_verbs_action));
+                       dev_flow->dv.encap_decap_verbs_action = NULL;
+               }
                rte_free(dev_flow);
        }
 }