net/mlx5: fix RSS expansion with explicit next protocol
[dpdk.git] / drivers / net / mlx5 / mlx5_flow.c
index f657980..2cadf61 100644 (file)
@@ -164,105 +164,143 @@ mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
        return false;
 }
 
+/**
+ * Network Service Header (NSH) and its next protocol values
+ * are described in RFC-8393.
+ */
+static enum rte_flow_item_type
+mlx5_nsh_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
+{
+       enum rte_flow_item_type type;
+
+       switch (proto_mask & proto_spec) {
+       case RTE_VXLAN_GPE_TYPE_IPV4:
+               type = RTE_FLOW_ITEM_TYPE_IPV4;
+               break;
+       case RTE_VXLAN_GPE_TYPE_IPV6:
+               type = RTE_VXLAN_GPE_TYPE_IPV6;
+               break;
+       case RTE_VXLAN_GPE_TYPE_ETH:
+               type = RTE_FLOW_ITEM_TYPE_ETH;
+               break;
+       default:
+               type = RTE_FLOW_ITEM_TYPE_END;
+       }
+       return type;
+}
+
+static enum rte_flow_item_type
+mlx5_inet_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
+{
+       enum rte_flow_item_type type;
+
+       switch (proto_mask & proto_spec) {
+       case IPPROTO_UDP:
+               type = RTE_FLOW_ITEM_TYPE_UDP;
+               break;
+       case IPPROTO_TCP:
+               type = RTE_FLOW_ITEM_TYPE_TCP;
+               break;
+       case IPPROTO_IP:
+               type = RTE_FLOW_ITEM_TYPE_IPV4;
+               break;
+       case IPPROTO_IPV6:
+               type = RTE_FLOW_ITEM_TYPE_IPV6;
+               break;
+       default:
+               type = RTE_FLOW_ITEM_TYPE_END;
+       }
+       return type;
+}
+
+static enum rte_flow_item_type
+mlx5_ethertype_to_item_type(rte_be16_t type_spec,
+                           rte_be16_t type_mask, bool is_tunnel)
+{
+       enum rte_flow_item_type type;
+
+       switch (rte_be_to_cpu_16(type_spec & type_mask)) {
+       case RTE_ETHER_TYPE_TEB:
+               type = is_tunnel ?
+                      RTE_FLOW_ITEM_TYPE_ETH : RTE_FLOW_ITEM_TYPE_END;
+               break;
+       case RTE_ETHER_TYPE_VLAN:
+               type = !is_tunnel ?
+                      RTE_FLOW_ITEM_TYPE_VLAN : RTE_FLOW_ITEM_TYPE_END;
+               break;
+       case RTE_ETHER_TYPE_IPV4:
+               type = RTE_FLOW_ITEM_TYPE_IPV4;
+               break;
+       case RTE_ETHER_TYPE_IPV6:
+               type = RTE_FLOW_ITEM_TYPE_IPV6;
+               break;
+       default:
+               type = RTE_FLOW_ITEM_TYPE_END;
+       }
+       return type;
+}
+
 static enum rte_flow_item_type
 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
 {
-       enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
-       uint16_t ether_type = 0;
-       uint16_t ether_type_m;
-       uint8_t ip_next_proto = 0;
-       uint8_t ip_next_proto_m;
+#define MLX5_XSET_ITEM_MASK_SPEC(type, fld)                              \
+       do {                                                             \
+               const void *m = item->mask;                              \
+               const void *s = item->spec;                              \
+               mask = m ?                                               \
+                       ((const struct rte_flow_item_##type *)m)->fld :  \
+                       rte_flow_item_##type##_mask.fld;                 \
+               spec = ((const struct rte_flow_item_##type *)s)->fld;    \
+       } while (0)
+
+       enum rte_flow_item_type ret;
+       uint16_t spec, mask;
 
        if (item == NULL || item->spec == NULL)
-               return ret;
+               return RTE_FLOW_ITEM_TYPE_VOID;
        switch (item->type) {
        case RTE_FLOW_ITEM_TYPE_ETH:
-               if (item->mask)
-                       ether_type_m = ((const struct rte_flow_item_eth *)
-                                               (item->mask))->type;
-               else
-                       ether_type_m = rte_flow_item_eth_mask.type;
-               if (ether_type_m != RTE_BE16(0xFFFF))
-                       break;
-               ether_type = ((const struct rte_flow_item_eth *)
-                               (item->spec))->type;
-               if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV4;
-               else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV6;
-               else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
-                       ret = RTE_FLOW_ITEM_TYPE_VLAN;
-               else
-                       ret = RTE_FLOW_ITEM_TYPE_END;
+               MLX5_XSET_ITEM_MASK_SPEC(eth, type);
+               if (!mask)
+                       return RTE_FLOW_ITEM_TYPE_VOID;
+               ret = mlx5_ethertype_to_item_type(spec, mask, false);
                break;
        case RTE_FLOW_ITEM_TYPE_VLAN:
-               if (item->mask)
-                       ether_type_m = ((const struct rte_flow_item_vlan *)
-                                               (item->mask))->inner_type;
-               else
-                       ether_type_m = rte_flow_item_vlan_mask.inner_type;
-               if (ether_type_m != RTE_BE16(0xFFFF))
-                       break;
-               ether_type = ((const struct rte_flow_item_vlan *)
-                               (item->spec))->inner_type;
-               if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV4;
-               else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV6;
-               else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
-                       ret = RTE_FLOW_ITEM_TYPE_VLAN;
-               else
-                       ret = RTE_FLOW_ITEM_TYPE_END;
+               MLX5_XSET_ITEM_MASK_SPEC(vlan, inner_type);
+               if (!mask)
+                       return RTE_FLOW_ITEM_TYPE_VOID;
+               ret = mlx5_ethertype_to_item_type(spec, mask, false);
                break;
        case RTE_FLOW_ITEM_TYPE_IPV4:
-               if (item->mask)
-                       ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
-                                       (item->mask))->hdr.next_proto_id;
-               else
-                       ip_next_proto_m =
-                               rte_flow_item_ipv4_mask.hdr.next_proto_id;
-               if (ip_next_proto_m != 0xFF)
-                       break;
-               ip_next_proto = ((const struct rte_flow_item_ipv4 *)
-                               (item->spec))->hdr.next_proto_id;
-               if (ip_next_proto == IPPROTO_UDP)
-                       ret = RTE_FLOW_ITEM_TYPE_UDP;
-               else if (ip_next_proto == IPPROTO_TCP)
-                       ret = RTE_FLOW_ITEM_TYPE_TCP;
-               else if (ip_next_proto == IPPROTO_IP)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV4;
-               else if (ip_next_proto == IPPROTO_IPV6)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV6;
-               else
-                       ret = RTE_FLOW_ITEM_TYPE_END;
+               MLX5_XSET_ITEM_MASK_SPEC(ipv4, hdr.next_proto_id);
+               if (!mask)
+                       return RTE_FLOW_ITEM_TYPE_VOID;
+               ret = mlx5_inet_proto_to_item_type(spec, mask);
                break;
        case RTE_FLOW_ITEM_TYPE_IPV6:
-               if (item->mask)
-                       ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
-                                               (item->mask))->hdr.proto;
-               else
-                       ip_next_proto_m =
-                               rte_flow_item_ipv6_mask.hdr.proto;
-               if (ip_next_proto_m != 0xFF)
-                       break;
-               ip_next_proto = ((const struct rte_flow_item_ipv6 *)
-                               (item->spec))->hdr.proto;
-               if (ip_next_proto == IPPROTO_UDP)
-                       ret = RTE_FLOW_ITEM_TYPE_UDP;
-               else if (ip_next_proto == IPPROTO_TCP)
-                       ret = RTE_FLOW_ITEM_TYPE_TCP;
-               else if (ip_next_proto == IPPROTO_IP)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV4;
-               else if (ip_next_proto == IPPROTO_IPV6)
-                       ret = RTE_FLOW_ITEM_TYPE_IPV6;
-               else
-                       ret = RTE_FLOW_ITEM_TYPE_END;
+               MLX5_XSET_ITEM_MASK_SPEC(ipv6, hdr.proto);
+               if (!mask)
+                       return RTE_FLOW_ITEM_TYPE_VOID;
+               ret = mlx5_inet_proto_to_item_type(spec, mask);
+               break;
+       case RTE_FLOW_ITEM_TYPE_GENEVE:
+               MLX5_XSET_ITEM_MASK_SPEC(geneve, protocol);
+               ret = mlx5_ethertype_to_item_type(spec, mask, true);
+               break;
+       case RTE_FLOW_ITEM_TYPE_GRE:
+               MLX5_XSET_ITEM_MASK_SPEC(gre, protocol);
+               ret = mlx5_ethertype_to_item_type(spec, mask, true);
+               break;
+       case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
+               MLX5_XSET_ITEM_MASK_SPEC(vxlan_gpe, protocol);
+               ret = mlx5_nsh_proto_to_item_type(spec, mask);
                break;
        default:
                ret = RTE_FLOW_ITEM_TYPE_VOID;
                break;
        }
        return ret;
+#undef MLX5_XSET_ITEM_MASK_SPEC
 }
 
 static const int *
@@ -530,7 +568,8 @@ enum mlx5_expansion {
        MLX5_EXPANSION_IPV6_UDP,
        MLX5_EXPANSION_IPV6_TCP,
        MLX5_EXPANSION_IPV6_FRAG_EXT,
-       MLX5_EXPANSION_GTP
+       MLX5_EXPANSION_GTP,
+       MLX5_EXPANSION_GENEVE,
 };
 
 /** Supported expansion of items. */
@@ -574,6 +613,7 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
                                                  MLX5_EXPANSION_VXLAN_GPE,
                                                  MLX5_EXPANSION_MPLS,
+                                                 MLX5_EXPANSION_GENEVE,
                                                  MLX5_EXPANSION_GTP),
                .type = RTE_FLOW_ITEM_TYPE_UDP,
                .rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
@@ -598,6 +638,7 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
                                                  MLX5_EXPANSION_VXLAN_GPE,
                                                  MLX5_EXPANSION_MPLS,
+                                                 MLX5_EXPANSION_GENEVE,
                                                  MLX5_EXPANSION_GTP),
                .type = RTE_FLOW_ITEM_TYPE_UDP,
                .rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
@@ -628,7 +669,8 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
        },
        [MLX5_EXPANSION_GRE] = {
-               .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
+               .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
+                                                 MLX5_EXPANSION_IPV4,
                                                  MLX5_EXPANSION_IPV6,
                                                  MLX5_EXPANSION_GRE_KEY,
                                                  MLX5_EXPANSION_MPLS),
@@ -701,6 +743,12 @@ static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
                                                  MLX5_EXPANSION_IPV6),
                .type = RTE_FLOW_ITEM_TYPE_GTP,
        },
+       [MLX5_EXPANSION_GENEVE] = {
+               .next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
+                                                 MLX5_EXPANSION_IPV4,
+                                                 MLX5_EXPANSION_IPV6),
+               .type = RTE_FLOW_ITEM_TYPE_GENEVE,
+       },
 };
 
 static struct rte_flow_action_handle *
@@ -1173,7 +1221,7 @@ flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
 }
 
 /**
- * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
+ * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the device
  * flow.
  *
  * @param[in] dev
@@ -2975,7 +3023,7 @@ mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
        if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
                return rte_flow_error_set
                        (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
-                       "Geneve TLV opt length exceeeds the limit (31)");
+                       "Geneve TLV opt length exceeds the limit (31)");
        /* Check if class type and length masks are full. */
        if (full_mask.option_class != mask->option_class ||
            full_mask.option_type != mask->option_type ||
@@ -3924,7 +3972,7 @@ find_graph_root(uint32_t rss_level)
  *  subflow.
  *
  * @param[in] dev_flow
- *   Pointer the created preifx subflow.
+ *   Pointer the created prefix subflow.
  *
  * @return
  *   The layers get from prefix subflow.
@@ -4251,7 +4299,7 @@ flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
                [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
        };
 
-       /* Fill the register fileds in the flow. */
+       /* Fill the register fields in the flow. */
        ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
        if (ret < 0)
                return NULL;
@@ -4320,7 +4368,7 @@ flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
        /*
         * The copy Flows are not included in any list. There
         * ones are referenced from other Flows and can not
-        * be applied, removed, deleted in ardbitrary order
+        * be applied, removed, deleted in arbitrary order
         * by list traversing.
         */
        mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
@@ -4777,7 +4825,7 @@ flow_create_split_inner(struct rte_eth_dev *dev,
        /*
         * If dev_flow is as one of the suffix flow, some actions in suffix
         * flow may need some user defined item layer flags, and pass the
-        * Metadate rxq mark flag to suffix flow as well.
+        * Metadata rxq mark flag to suffix flow as well.
         */
        if (flow_split_info->prefix_layers)
                dev_flow->handle->layers = flow_split_info->prefix_layers;
@@ -5218,6 +5266,8 @@ exit:
  *   Pointer to the Q/RSS action.
  * @param[in] actions_n
  *   Number of original actions.
+ * @param[in] mtr_sfx
+ *   Check if it is in meter suffix table.
  * @param[out] error
  *   Perform verbose error reporting if not NULL.
  *
@@ -5230,7 +5280,8 @@ flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
                          struct rte_flow_action *split_actions,
                          const struct rte_flow_action *actions,
                          const struct rte_flow_action *qrss,
-                         int actions_n, struct rte_flow_error *error)
+                         int actions_n, int mtr_sfx,
+                         struct rte_flow_error *error)
 {
        struct mlx5_priv *priv = dev->data->dev_private;
        struct mlx5_rte_flow_action_set_tag *set_tag;
@@ -5245,15 +5296,15 @@ flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
         * - Add jump to mreg CP_TBL.
         * As a result, there will be one more action.
         */
-       ++actions_n;
        memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
+       /* Count MLX5_RTE_FLOW_ACTION_TYPE_TAG. */
+       ++actions_n;
        set_tag = (void *)(split_actions + actions_n);
        /*
-        * If tag action is not set to void(it means we are not the meter
-        * suffix flow), add the tag action. Since meter suffix flow already
-        * has the tag added.
+        * If we are not the meter suffix flow, add the tag action.
+        * Since meter suffix flow already has the tag added.
         */
-       if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
+       if (!mtr_sfx) {
                /*
                 * Allocate the new subflow ID. This one is unique within
                 * device and not shared with representors. Otherwise,
@@ -5286,6 +5337,12 @@ flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
                                MLX5_RTE_FLOW_ACTION_TYPE_TAG,
                        .conf = set_tag,
                };
+       } else {
+               /*
+                * If we are the suffix flow of meter, tag already exist.
+                * Set the QUEUE/RSS action to void.
+                */
+               split_actions[qrss_idx].type = RTE_FLOW_ACTION_TYPE_VOID;
        }
        /* JUMP action to jump to mreg copy table (CP_TBL). */
        jump = (void *)(set_tag + 1);
@@ -5317,7 +5374,7 @@ flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
  * @param[out] error
  *   Perform verbose error reporting if not NULL.
  * @param[in] encap_idx
- *   The encap action inndex.
+ *   The encap action index.
  *
  * @return
  *   0 on success, negative value otherwise
@@ -5740,17 +5797,6 @@ flow_create_split_metadata(struct rte_eth_dev *dev,
                                                  RTE_FLOW_ERROR_TYPE_ACTION,
                                                  NULL, "no memory to split "
                                                  "metadata flow");
-               /*
-                * If we are the suffix flow of meter, tag already exist.
-                * Set the tag action to void.
-                */
-               if (mtr_sfx)
-                       ext_actions[qrss - actions].type =
-                                               RTE_FLOW_ACTION_TYPE_VOID;
-               else
-                       ext_actions[qrss - actions].type =
-                                               (enum rte_flow_action_type)
-                                               MLX5_RTE_FLOW_ACTION_TYPE_TAG;
                /*
                 * Create the new actions list with removed Q/RSS action
                 * and appended set tag and jump to register copy table
@@ -5758,7 +5804,8 @@ flow_create_split_metadata(struct rte_eth_dev *dev,
                 * in advance, because it is needed for set tag action.
                 */
                qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
-                                                   qrss, actions_n, error);
+                                                   qrss, actions_n,
+                                                   mtr_sfx, error);
                if (!mtr_sfx && !qrss_id) {
                        ret = -rte_errno;
                        goto exit;
@@ -5849,6 +5896,7 @@ flow_create_split_metadata(struct rte_eth_dev *dev,
                /* Add suffix subflow to execute Q/RSS. */
                flow_split_info->prefix_layers = layers;
                flow_split_info->prefix_mark = 0;
+               flow_split_info->table_id = 0;
                ret = flow_create_split_inner(dev, flow, &dev_flow,
                                              &q_attr, mtr_sfx ? items :
                                              q_items, q_actions,
@@ -6063,7 +6111,6 @@ flow_create_split_meter(struct rte_eth_dev *dev,
                        goto exit;
                }
                /* Add the prefix subflow. */
-               flow_split_info->prefix_mark = 0;
                skip_scale_restore = flow_split_info->skip_scale;
                flow_split_info->skip_scale |=
                        1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
@@ -6096,7 +6143,7 @@ flow_create_split_meter(struct rte_eth_dev *dev,
                                 MLX5_FLOW_TABLE_LEVEL_METER;
                flow_split_info->prefix_layers =
                                flow_get_prefix_layer_flags(dev_flow);
-               flow_split_info->prefix_mark = dev_flow->handle->mark;
+               flow_split_info->prefix_mark |= dev_flow->handle->mark;
                flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
        }
        /* Add the prefix subflow. */
@@ -6248,7 +6295,7 @@ flow_create_split_sample(struct rte_eth_dev *dev,
                }
                flow_split_info->prefix_layers =
                                flow_get_prefix_layer_flags(dev_flow);
-               flow_split_info->prefix_mark = dev_flow->handle->mark;
+               flow_split_info->prefix_mark |= dev_flow->handle->mark;
                /* Suffix group level already be scaled with factor, set
                 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
                 * again in translation.
@@ -6852,7 +6899,7 @@ flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
  * @param type
  *   Flow type to be flushed.
  * @param active
- *   If flushing is called avtively.
+ *   If flushing is called actively.
  */
 void
 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
@@ -8499,7 +8546,7 @@ mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
  *   Perform verbose error reporting if not NULL. PMDs initialize this
  *   structure in case of error only.
  * @return
- *   0 on success, a nagative value otherwise.
+ *   0 on success, a negative value otherwise.
  */
 int
 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
@@ -8977,7 +9024,7 @@ mlx5_get_tof(const struct rte_flow_item *item,
 }
 
 /**
- * tunnel offload functionalilty is defined for DV environment only
+ * tunnel offload functionality is defined for DV environment only
  */
 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
 __extension__