net: add rte prefix to ether defines
[dpdk.git] / drivers / net / enic / enic_flow.c
index 0cf04ae..bdac1f7 100644 (file)
        rte_log(RTE_LOG_ ## level, enicpmd_logtype_flow, \
                fmt "\n", ##args)
 
+/*
+ * Common arguments passed to copy_item functions. Use this structure
+ * so we can easily add new arguments.
+ * item: Item specification.
+ * filter: Partially filled in NIC filter structure.
+ * inner_ofst: If zero, this is an outer header. If non-zero, this is
+ *   the offset into L5 where the header begins.
+ * l2_proto_off: offset to EtherType eth or vlan header.
+ * l3_proto_off: offset to next protocol field in IPv4 or 6 header.
+ */
+struct copy_item_args {
+       const struct rte_flow_item *item;
+       struct filter_v2 *filter;
+       uint8_t *inner_ofst;
+       uint8_t l2_proto_off;
+       uint8_t l3_proto_off;
+       struct enic *enic;
+};
+
+/* functions for copying items into enic filters */
+typedef int (enic_copy_item_fn)(struct copy_item_args *arg);
+
 /** Info about how to copy items into enic filters. */
 struct enic_items {
        /** Function for copying and validating an item. */
-       int (*copy_item)(const struct rte_flow_item *item,
-                        struct filter_v2 *enic_filter, u8 *inner_ofst);
+       enic_copy_item_fn *copy_item;
        /** List of valid previous items. */
        const enum rte_flow_item_type * const prev_items;
        /** True if it's OK for this item to be the first item. For some NIC
         * versions, it's invalid to start the stack above layer 3.
         */
        const u8 valid_start_item;
+       /* Inner packet version of copy_item. */
+       enic_copy_item_fn *inner_copy_item;
 };
 
 /** Filtering capabilities for various NIC and firmware versions. */
 struct enic_filter_cap {
        /** list of valid items and their handlers and attributes. */
        const struct enic_items *item_info;
+       /* Max type in the above list, used to detect unsupported types */
+       enum rte_flow_item_type max_item_type;
 };
 
 /* functions for copying flow actions into enic actions */
-typedef int (copy_action_fn)(const struct rte_flow_action actions[],
+typedef int (copy_action_fn)(struct enic *enic,
+                            const struct rte_flow_action actions[],
                             struct filter_action_v2 *enic_action);
 
-/* functions for copying items into enic filters */
-typedef int(enic_copy_item_fn)(const struct rte_flow_item *item,
-                         struct filter_v2 *enic_filter, u8 *inner_ofst);
-
 /** Action capabilities for various NICs. */
 struct enic_action_cap {
        /** list of valid actions */
        const enum rte_flow_action_type *actions;
        /** copy function for a particular NIC */
-       int (*copy_fn)(const struct rte_flow_action actions[],
-                      struct filter_action_v2 *enic_action);
+       copy_action_fn *copy_fn;
 };
 
 /* Forward declarations */
 static enic_copy_item_fn enic_copy_item_ipv4_v1;
 static enic_copy_item_fn enic_copy_item_udp_v1;
 static enic_copy_item_fn enic_copy_item_tcp_v1;
+static enic_copy_item_fn enic_copy_item_raw_v2;
 static enic_copy_item_fn enic_copy_item_eth_v2;
 static enic_copy_item_fn enic_copy_item_vlan_v2;
 static enic_copy_item_fn enic_copy_item_ipv4_v2;
@@ -70,8 +92,13 @@ static enic_copy_item_fn enic_copy_item_ipv6_v2;
 static enic_copy_item_fn enic_copy_item_udp_v2;
 static enic_copy_item_fn enic_copy_item_tcp_v2;
 static enic_copy_item_fn enic_copy_item_sctp_v2;
-static enic_copy_item_fn enic_copy_item_sctp_v2;
 static enic_copy_item_fn enic_copy_item_vxlan_v2;
+static enic_copy_item_fn enic_copy_item_inner_eth_v2;
+static enic_copy_item_fn enic_copy_item_inner_vlan_v2;
+static enic_copy_item_fn enic_copy_item_inner_ipv4_v2;
+static enic_copy_item_fn enic_copy_item_inner_ipv6_v2;
+static enic_copy_item_fn enic_copy_item_inner_udp_v2;
+static enic_copy_item_fn enic_copy_item_inner_tcp_v2;
 static copy_action_fn enic_copy_action_v1;
 static copy_action_fn enic_copy_action_v2;
 
@@ -86,6 +113,7 @@ static const struct enic_items enic_items_v1[] = {
                .prev_items = (const enum rte_flow_item_type[]) {
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = NULL,
        },
        [RTE_FLOW_ITEM_TYPE_UDP] = {
                .copy_item = enic_copy_item_udp_v1,
@@ -94,6 +122,7 @@ static const struct enic_items enic_items_v1[] = {
                               RTE_FLOW_ITEM_TYPE_IPV4,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = NULL,
        },
        [RTE_FLOW_ITEM_TYPE_TCP] = {
                .copy_item = enic_copy_item_tcp_v1,
@@ -102,6 +131,7 @@ static const struct enic_items enic_items_v1[] = {
                               RTE_FLOW_ITEM_TYPE_IPV4,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = NULL,
        },
 };
 
@@ -110,6 +140,15 @@ static const struct enic_items enic_items_v1[] = {
  * that layer 3 must be specified.
  */
 static const struct enic_items enic_items_v2[] = {
+       [RTE_FLOW_ITEM_TYPE_RAW] = {
+               .copy_item = enic_copy_item_raw_v2,
+               .valid_start_item = 0,
+               .prev_items = (const enum rte_flow_item_type[]) {
+                              RTE_FLOW_ITEM_TYPE_UDP,
+                              RTE_FLOW_ITEM_TYPE_END,
+               },
+               .inner_copy_item = NULL,
+       },
        [RTE_FLOW_ITEM_TYPE_ETH] = {
                .copy_item = enic_copy_item_eth_v2,
                .valid_start_item = 1,
@@ -117,6 +156,7 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_VXLAN,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_eth_v2,
        },
        [RTE_FLOW_ITEM_TYPE_VLAN] = {
                .copy_item = enic_copy_item_vlan_v2,
@@ -125,6 +165,7 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_ETH,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_vlan_v2,
        },
        [RTE_FLOW_ITEM_TYPE_IPV4] = {
                .copy_item = enic_copy_item_ipv4_v2,
@@ -134,6 +175,7 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_VLAN,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_ipv4_v2,
        },
        [RTE_FLOW_ITEM_TYPE_IPV6] = {
                .copy_item = enic_copy_item_ipv6_v2,
@@ -143,6 +185,7 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_VLAN,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_ipv6_v2,
        },
        [RTE_FLOW_ITEM_TYPE_UDP] = {
                .copy_item = enic_copy_item_udp_v2,
@@ -152,6 +195,7 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_IPV6,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_udp_v2,
        },
        [RTE_FLOW_ITEM_TYPE_TCP] = {
                .copy_item = enic_copy_item_tcp_v2,
@@ -161,6 +205,7 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_IPV6,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_tcp_v2,
        },
        [RTE_FLOW_ITEM_TYPE_SCTP] = {
                .copy_item = enic_copy_item_sctp_v2,
@@ -170,6 +215,7 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_IPV6,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = NULL,
        },
        [RTE_FLOW_ITEM_TYPE_VXLAN] = {
                .copy_item = enic_copy_item_vxlan_v2,
@@ -178,11 +224,21 @@ static const struct enic_items enic_items_v2[] = {
                               RTE_FLOW_ITEM_TYPE_UDP,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = NULL,
        },
 };
 
 /** NICs with Advanced filters enabled */
 static const struct enic_items enic_items_v3[] = {
+       [RTE_FLOW_ITEM_TYPE_RAW] = {
+               .copy_item = enic_copy_item_raw_v2,
+               .valid_start_item = 0,
+               .prev_items = (const enum rte_flow_item_type[]) {
+                              RTE_FLOW_ITEM_TYPE_UDP,
+                              RTE_FLOW_ITEM_TYPE_END,
+               },
+               .inner_copy_item = NULL,
+       },
        [RTE_FLOW_ITEM_TYPE_ETH] = {
                .copy_item = enic_copy_item_eth_v2,
                .valid_start_item = 1,
@@ -190,6 +246,7 @@ static const struct enic_items enic_items_v3[] = {
                               RTE_FLOW_ITEM_TYPE_VXLAN,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_eth_v2,
        },
        [RTE_FLOW_ITEM_TYPE_VLAN] = {
                .copy_item = enic_copy_item_vlan_v2,
@@ -198,6 +255,7 @@ static const struct enic_items enic_items_v3[] = {
                               RTE_FLOW_ITEM_TYPE_ETH,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_vlan_v2,
        },
        [RTE_FLOW_ITEM_TYPE_IPV4] = {
                .copy_item = enic_copy_item_ipv4_v2,
@@ -207,6 +265,7 @@ static const struct enic_items enic_items_v3[] = {
                               RTE_FLOW_ITEM_TYPE_VLAN,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_ipv4_v2,
        },
        [RTE_FLOW_ITEM_TYPE_IPV6] = {
                .copy_item = enic_copy_item_ipv6_v2,
@@ -216,6 +275,7 @@ static const struct enic_items enic_items_v3[] = {
                               RTE_FLOW_ITEM_TYPE_VLAN,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_ipv6_v2,
        },
        [RTE_FLOW_ITEM_TYPE_UDP] = {
                .copy_item = enic_copy_item_udp_v2,
@@ -225,6 +285,7 @@ static const struct enic_items enic_items_v3[] = {
                               RTE_FLOW_ITEM_TYPE_IPV6,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_udp_v2,
        },
        [RTE_FLOW_ITEM_TYPE_TCP] = {
                .copy_item = enic_copy_item_tcp_v2,
@@ -234,15 +295,17 @@ static const struct enic_items enic_items_v3[] = {
                               RTE_FLOW_ITEM_TYPE_IPV6,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = enic_copy_item_inner_tcp_v2,
        },
        [RTE_FLOW_ITEM_TYPE_SCTP] = {
                .copy_item = enic_copy_item_sctp_v2,
-               .valid_start_item = 1,
+               .valid_start_item = 0,
                .prev_items = (const enum rte_flow_item_type[]) {
                               RTE_FLOW_ITEM_TYPE_IPV4,
                               RTE_FLOW_ITEM_TYPE_IPV6,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = NULL,
        },
        [RTE_FLOW_ITEM_TYPE_VXLAN] = {
                .copy_item = enic_copy_item_vxlan_v2,
@@ -251,6 +314,7 @@ static const struct enic_items enic_items_v3[] = {
                               RTE_FLOW_ITEM_TYPE_UDP,
                               RTE_FLOW_ITEM_TYPE_END,
                },
+               .inner_copy_item = NULL,
        },
 };
 
@@ -258,12 +322,15 @@ static const struct enic_items enic_items_v3[] = {
 static const struct enic_filter_cap enic_filter_cap[] = {
        [FILTER_IPV4_5TUPLE] = {
                .item_info = enic_items_v1,
+               .max_item_type = RTE_FLOW_ITEM_TYPE_TCP,
        },
        [FILTER_USNIC_IP] = {
                .item_info = enic_items_v2,
+               .max_item_type = RTE_FLOW_ITEM_TYPE_VXLAN,
        },
        [FILTER_DPDK_1] = {
                .item_info = enic_items_v3,
+               .max_item_type = RTE_FLOW_ITEM_TYPE_VXLAN,
        },
 };
 
@@ -278,6 +345,8 @@ static const enum rte_flow_action_type enic_supported_actions_v2_id[] = {
        RTE_FLOW_ACTION_TYPE_QUEUE,
        RTE_FLOW_ACTION_TYPE_MARK,
        RTE_FLOW_ACTION_TYPE_FLAG,
+       RTE_FLOW_ACTION_TYPE_RSS,
+       RTE_FLOW_ACTION_TYPE_PASSTHRU,
        RTE_FLOW_ACTION_TYPE_END,
 };
 
@@ -286,6 +355,19 @@ static const enum rte_flow_action_type enic_supported_actions_v2_drop[] = {
        RTE_FLOW_ACTION_TYPE_MARK,
        RTE_FLOW_ACTION_TYPE_FLAG,
        RTE_FLOW_ACTION_TYPE_DROP,
+       RTE_FLOW_ACTION_TYPE_RSS,
+       RTE_FLOW_ACTION_TYPE_PASSTHRU,
+       RTE_FLOW_ACTION_TYPE_END,
+};
+
+static const enum rte_flow_action_type enic_supported_actions_v2_count[] = {
+       RTE_FLOW_ACTION_TYPE_QUEUE,
+       RTE_FLOW_ACTION_TYPE_MARK,
+       RTE_FLOW_ACTION_TYPE_FLAG,
+       RTE_FLOW_ACTION_TYPE_DROP,
+       RTE_FLOW_ACTION_TYPE_COUNT,
+       RTE_FLOW_ACTION_TYPE_RSS,
+       RTE_FLOW_ACTION_TYPE_PASSTHRU,
        RTE_FLOW_ACTION_TYPE_END,
 };
 
@@ -303,6 +385,10 @@ static const struct enic_action_cap enic_action_cap[] = {
                .actions = enic_supported_actions_v2_drop,
                .copy_fn = enic_copy_action_v2,
        },
+       [FILTER_ACTION_COUNTER_FLAG] = {
+               .actions = enic_supported_actions_v2_count,
+               .copy_fn = enic_copy_action_v2,
+       },
 };
 
 static int
@@ -317,20 +403,11 @@ mask_exact_match(const u8 *supported, const u8 *supplied,
        return 1;
 }
 
-/**
- * Copy IPv4 item into version 1 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Should always be 0 for version 1.
- */
 static int
-enic_copy_item_ipv4_v1(const struct rte_flow_item *item,
-                      struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_ipv4_v1(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_ipv4 *spec = item->spec;
        const struct rte_flow_item_ipv4 *mask = item->mask;
        struct filter_ipv4_5tuple *enic_5tup = &enic_filter->u.ipv4;
@@ -341,9 +418,6 @@ enic_copy_item_ipv4_v1(const struct rte_flow_item *item,
 
        FLOW_TRACE();
 
-       if (*inner_ofst)
-               return ENOTSUP;
-
        if (!mask)
                mask = &rte_flow_item_ipv4_mask;
 
@@ -367,20 +441,11 @@ enic_copy_item_ipv4_v1(const struct rte_flow_item *item,
        return 0;
 }
 
-/**
- * Copy UDP item into version 1 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Should always be 0 for version 1.
- */
 static int
-enic_copy_item_udp_v1(const struct rte_flow_item *item,
-                     struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_udp_v1(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_udp *spec = item->spec;
        const struct rte_flow_item_udp *mask = item->mask;
        struct filter_ipv4_5tuple *enic_5tup = &enic_filter->u.ipv4;
@@ -391,9 +456,6 @@ enic_copy_item_udp_v1(const struct rte_flow_item *item,
 
        FLOW_TRACE();
 
-       if (*inner_ofst)
-               return ENOTSUP;
-
        if (!mask)
                mask = &rte_flow_item_udp_mask;
 
@@ -418,20 +480,11 @@ enic_copy_item_udp_v1(const struct rte_flow_item *item,
        return 0;
 }
 
-/**
- * Copy TCP item into version 1 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Should always be 0 for version 1.
- */
 static int
-enic_copy_item_tcp_v1(const struct rte_flow_item *item,
-                     struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_tcp_v1(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_tcp *spec = item->spec;
        const struct rte_flow_item_tcp *mask = item->mask;
        struct filter_ipv4_5tuple *enic_5tup = &enic_filter->u.ipv4;
@@ -442,9 +495,6 @@ enic_copy_item_tcp_v1(const struct rte_flow_item *item,
 
        FLOW_TRACE();
 
-       if (*inner_ofst)
-               return ENOTSUP;
-
        if (!mask)
                mask = &rte_flow_item_tcp_mask;
 
@@ -469,23 +519,152 @@ enic_copy_item_tcp_v1(const struct rte_flow_item *item,
        return 0;
 }
 
-/**
- * Copy ETH item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   If zero, this is an outer header. If non-zero, this is the offset into L5
- *   where the header begins.
+/*
+ * The common 'copy' function for all inner packet patterns. Patterns are
+ * first appended to the L5 pattern buffer. Then, since the NIC filter
+ * API has no special support for inner packet matching at the moment,
+ * we set EtherType and IP proto as necessary.
  */
 static int
-enic_copy_item_eth_v2(const struct rte_flow_item *item,
-                     struct filter_v2 *enic_filter, u8 *inner_ofst)
+copy_inner_common(struct filter_generic_1 *gp, uint8_t *inner_ofst,
+                 const void *val, const void *mask, uint8_t val_size,
+                 uint8_t proto_off, uint16_t proto_val, uint8_t proto_size)
+{
+       uint8_t *l5_mask, *l5_val;
+       uint8_t start_off;
+
+       /* No space left in the L5 pattern buffer. */
+       start_off = *inner_ofst;
+       if ((start_off + val_size) > FILTER_GENERIC_1_KEY_LEN)
+               return ENOTSUP;
+       l5_mask = gp->layer[FILTER_GENERIC_1_L5].mask;
+       l5_val = gp->layer[FILTER_GENERIC_1_L5].val;
+       /* Copy the pattern into the L5 buffer. */
+       if (val) {
+               memcpy(l5_mask + start_off, mask, val_size);
+               memcpy(l5_val + start_off, val, val_size);
+       }
+       /* Set the protocol field in the previous header. */
+       if (proto_off) {
+               void *m, *v;
+
+               m = l5_mask + proto_off;
+               v = l5_val + proto_off;
+               if (proto_size == 1) {
+                       *(uint8_t *)m = 0xff;
+                       *(uint8_t *)v = (uint8_t)proto_val;
+               } else if (proto_size == 2) {
+                       *(uint16_t *)m = 0xffff;
+                       *(uint16_t *)v = proto_val;
+               }
+       }
+       /* All inner headers land in L5 buffer even if their spec is null. */
+       *inner_ofst += val_size;
+       return 0;
+}
+
+static int
+enic_copy_item_inner_eth_v2(struct copy_item_args *arg)
+{
+       const void *mask = arg->item->mask;
+       uint8_t *off = arg->inner_ofst;
+
+       FLOW_TRACE();
+       if (!mask)
+               mask = &rte_flow_item_eth_mask;
+       arg->l2_proto_off = *off + offsetof(struct rte_ether_hdr, ether_type);
+       return copy_inner_common(&arg->filter->u.generic_1, off,
+               arg->item->spec, mask, sizeof(struct rte_ether_hdr),
+               0 /* no previous protocol */, 0, 0);
+}
+
+static int
+enic_copy_item_inner_vlan_v2(struct copy_item_args *arg)
+{
+       const void *mask = arg->item->mask;
+       uint8_t *off = arg->inner_ofst;
+       uint8_t eth_type_off;
+
+       FLOW_TRACE();
+       if (!mask)
+               mask = &rte_flow_item_vlan_mask;
+       /* Append vlan header to L5 and set ether type = TPID */
+       eth_type_off = arg->l2_proto_off;
+       arg->l2_proto_off = *off + offsetof(struct rte_vlan_hdr, eth_proto);
+       return copy_inner_common(&arg->filter->u.generic_1, off,
+               arg->item->spec, mask, sizeof(struct rte_vlan_hdr),
+               eth_type_off, rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN), 2);
+}
+
+static int
+enic_copy_item_inner_ipv4_v2(struct copy_item_args *arg)
+{
+       const void *mask = arg->item->mask;
+       uint8_t *off = arg->inner_ofst;
+
+       FLOW_TRACE();
+       if (!mask)
+               mask = &rte_flow_item_ipv4_mask;
+       /* Append ipv4 header to L5 and set ether type = ipv4 */
+       arg->l3_proto_off = *off + offsetof(struct ipv4_hdr, next_proto_id);
+       return copy_inner_common(&arg->filter->u.generic_1, off,
+               arg->item->spec, mask, sizeof(struct ipv4_hdr),
+               arg->l2_proto_off, rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv4), 2);
+}
+
+static int
+enic_copy_item_inner_ipv6_v2(struct copy_item_args *arg)
+{
+       const void *mask = arg->item->mask;
+       uint8_t *off = arg->inner_ofst;
+
+       FLOW_TRACE();
+       if (!mask)
+               mask = &rte_flow_item_ipv6_mask;
+       /* Append ipv6 header to L5 and set ether type = ipv6 */
+       arg->l3_proto_off = *off + offsetof(struct ipv6_hdr, proto);
+       return copy_inner_common(&arg->filter->u.generic_1, off,
+               arg->item->spec, mask, sizeof(struct ipv6_hdr),
+               arg->l2_proto_off, rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv6), 2);
+}
+
+static int
+enic_copy_item_inner_udp_v2(struct copy_item_args *arg)
+{
+       const void *mask = arg->item->mask;
+       uint8_t *off = arg->inner_ofst;
+
+       FLOW_TRACE();
+       if (!mask)
+               mask = &rte_flow_item_udp_mask;
+       /* Append udp header to L5 and set ip proto = udp */
+       return copy_inner_common(&arg->filter->u.generic_1, off,
+               arg->item->spec, mask, sizeof(struct udp_hdr),
+               arg->l3_proto_off, IPPROTO_UDP, 1);
+}
+
+static int
+enic_copy_item_inner_tcp_v2(struct copy_item_args *arg)
+{
+       const void *mask = arg->item->mask;
+       uint8_t *off = arg->inner_ofst;
+
+       FLOW_TRACE();
+       if (!mask)
+               mask = &rte_flow_item_tcp_mask;
+       /* Append tcp header to L5 and set ip proto = tcp */
+       return copy_inner_common(&arg->filter->u.generic_1, off,
+               arg->item->spec, mask, sizeof(struct tcp_hdr),
+               arg->l3_proto_off, IPPROTO_TCP, 1);
+}
+
+static int
+enic_copy_item_eth_v2(struct copy_item_args *arg)
 {
-       struct ether_hdr enic_spec;
-       struct ether_hdr enic_mask;
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
+       struct rte_ether_hdr enic_spec;
+       struct rte_ether_hdr enic_mask;
        const struct rte_flow_item_eth *spec = item->spec;
        const struct rte_flow_item_eth *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
@@ -500,56 +679,35 @@ enic_copy_item_eth_v2(const struct rte_flow_item *item,
                mask = &rte_flow_item_eth_mask;
 
        memcpy(enic_spec.d_addr.addr_bytes, spec->dst.addr_bytes,
-              ETHER_ADDR_LEN);
+              RTE_ETHER_ADDR_LEN);
        memcpy(enic_spec.s_addr.addr_bytes, spec->src.addr_bytes,
-              ETHER_ADDR_LEN);
+              RTE_ETHER_ADDR_LEN);
 
        memcpy(enic_mask.d_addr.addr_bytes, mask->dst.addr_bytes,
-              ETHER_ADDR_LEN);
+              RTE_ETHER_ADDR_LEN);
        memcpy(enic_mask.s_addr.addr_bytes, mask->src.addr_bytes,
-              ETHER_ADDR_LEN);
+              RTE_ETHER_ADDR_LEN);
        enic_spec.ether_type = spec->type;
        enic_mask.ether_type = mask->type;
 
-       if (*inner_ofst == 0) {
-               /* outer header */
-               memcpy(gp->layer[FILTER_GENERIC_1_L2].mask, &enic_mask,
-                      sizeof(struct ether_hdr));
-               memcpy(gp->layer[FILTER_GENERIC_1_L2].val, &enic_spec,
-                      sizeof(struct ether_hdr));
-       } else {
-               /* inner header */
-               if ((*inner_ofst + sizeof(struct ether_hdr)) >
-                    FILTER_GENERIC_1_KEY_LEN)
-                       return ENOTSUP;
-               /* Offset into L5 where inner Ethernet header goes */
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].mask[*inner_ofst],
-                      &enic_mask, sizeof(struct ether_hdr));
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].val[*inner_ofst],
-                      &enic_spec, sizeof(struct ether_hdr));
-               *inner_ofst += sizeof(struct ether_hdr);
-       }
+       /* outer header */
+       memcpy(gp->layer[FILTER_GENERIC_1_L2].mask, &enic_mask,
+              sizeof(struct rte_ether_hdr));
+       memcpy(gp->layer[FILTER_GENERIC_1_L2].val, &enic_spec,
+              sizeof(struct rte_ether_hdr));
        return 0;
 }
 
-/**
- * Copy VLAN item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   If zero, this is an outer header. If non-zero, this is the offset into L5
- *   where the header begins.
- */
 static int
-enic_copy_item_vlan_v2(const struct rte_flow_item *item,
-                      struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_vlan_v2(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_vlan *spec = item->spec;
        const struct rte_flow_item_vlan *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
+       struct rte_ether_hdr *eth_mask;
+       struct rte_ether_hdr *eth_val;
 
        FLOW_TRACE();
 
@@ -560,99 +718,72 @@ enic_copy_item_vlan_v2(const struct rte_flow_item *item,
        if (!mask)
                mask = &rte_flow_item_vlan_mask;
 
-       if (*inner_ofst == 0) {
-               struct ether_hdr *eth_mask =
-                       (void *)gp->layer[FILTER_GENERIC_1_L2].mask;
-               struct ether_hdr *eth_val =
-                       (void *)gp->layer[FILTER_GENERIC_1_L2].val;
+       eth_mask = (void *)gp->layer[FILTER_GENERIC_1_L2].mask;
+       eth_val = (void *)gp->layer[FILTER_GENERIC_1_L2].val;
+       /* Outer TPID cannot be matched */
+       if (eth_mask->ether_type)
+               return ENOTSUP;
+       /*
+        * For recent models:
+        * When packet matching, the VIC always compares vlan-stripped
+        * L2, regardless of vlan stripping settings. So, the inner type
+        * from vlan becomes the ether type of the eth header.
+        *
+        * Older models w/o hardware vxlan parser have a different
+        * behavior when vlan stripping is disabled. In this case,
+        * vlan tag remains in the L2 buffer.
+        */
+       if (!arg->enic->vxlan && !arg->enic->ig_vlan_strip_en) {
+               struct rte_vlan_hdr *vlan;
 
-               /* Outer TPID cannot be matched */
-               if (eth_mask->ether_type)
-                       return ENOTSUP;
+               vlan = (struct rte_vlan_hdr *)(eth_mask + 1);
+               vlan->eth_proto = mask->inner_type;
+               vlan = (struct rte_vlan_hdr *)(eth_val + 1);
+               vlan->eth_proto = spec->inner_type;
+       } else {
                eth_mask->ether_type = mask->inner_type;
                eth_val->ether_type = spec->inner_type;
-
-               /* Outer header. Use the vlan mask/val fields */
-               gp->mask_vlan = mask->tci;
-               gp->val_vlan = spec->tci;
-       } else {
-               /* Inner header. Mask/Val start at *inner_ofst into L5 */
-               if ((*inner_ofst + sizeof(struct vlan_hdr)) >
-                    FILTER_GENERIC_1_KEY_LEN)
-                       return ENOTSUP;
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].mask[*inner_ofst],
-                      mask, sizeof(struct vlan_hdr));
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].val[*inner_ofst],
-                      spec, sizeof(struct vlan_hdr));
-               *inner_ofst += sizeof(struct vlan_hdr);
        }
+       /* For TCI, use the vlan mask/val fields (little endian). */
+       gp->mask_vlan = rte_be_to_cpu_16(mask->tci);
+       gp->val_vlan = rte_be_to_cpu_16(spec->tci);
        return 0;
 }
 
-/**
- * Copy IPv4 item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Must be 0. Don't support inner IPv4 filtering.
- */
 static int
-enic_copy_item_ipv4_v2(const struct rte_flow_item *item,
-                      struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_ipv4_v2(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_ipv4 *spec = item->spec;
        const struct rte_flow_item_ipv4 *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
 
        FLOW_TRACE();
 
-       if (*inner_ofst == 0) {
-               /* Match IPv4 */
-               gp->mask_flags |= FILTER_GENERIC_1_IPV4;
-               gp->val_flags |= FILTER_GENERIC_1_IPV4;
+       /* Match IPv4 */
+       gp->mask_flags |= FILTER_GENERIC_1_IPV4;
+       gp->val_flags |= FILTER_GENERIC_1_IPV4;
 
-               /* Match all if no spec */
-               if (!spec)
-                       return 0;
+       /* Match all if no spec */
+       if (!spec)
+               return 0;
 
-               if (!mask)
-                       mask = &rte_flow_item_ipv4_mask;
+       if (!mask)
+               mask = &rte_flow_item_ipv4_mask;
 
-               memcpy(gp->layer[FILTER_GENERIC_1_L3].mask, &mask->hdr,
-                      sizeof(struct ipv4_hdr));
-               memcpy(gp->layer[FILTER_GENERIC_1_L3].val, &spec->hdr,
-                      sizeof(struct ipv4_hdr));
-       } else {
-               /* Inner IPv4 header. Mask/Val start at *inner_ofst into L5 */
-               if ((*inner_ofst + sizeof(struct ipv4_hdr)) >
-                    FILTER_GENERIC_1_KEY_LEN)
-                       return ENOTSUP;
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].mask[*inner_ofst],
-                      mask, sizeof(struct ipv4_hdr));
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].val[*inner_ofst],
-                      spec, sizeof(struct ipv4_hdr));
-               *inner_ofst += sizeof(struct ipv4_hdr);
-       }
+       memcpy(gp->layer[FILTER_GENERIC_1_L3].mask, &mask->hdr,
+              sizeof(struct ipv4_hdr));
+       memcpy(gp->layer[FILTER_GENERIC_1_L3].val, &spec->hdr,
+              sizeof(struct ipv4_hdr));
        return 0;
 }
 
-/**
- * Copy IPv6 item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Must be 0. Don't support inner IPv6 filtering.
- */
 static int
-enic_copy_item_ipv6_v2(const struct rte_flow_item *item,
-                      struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_ipv6_v2(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_ipv6 *spec = item->spec;
        const struct rte_flow_item_ipv6 *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
@@ -670,39 +801,18 @@ enic_copy_item_ipv6_v2(const struct rte_flow_item *item,
        if (!mask)
                mask = &rte_flow_item_ipv6_mask;
 
-       if (*inner_ofst == 0) {
-               memcpy(gp->layer[FILTER_GENERIC_1_L3].mask, &mask->hdr,
-                      sizeof(struct ipv6_hdr));
-               memcpy(gp->layer[FILTER_GENERIC_1_L3].val, &spec->hdr,
-                      sizeof(struct ipv6_hdr));
-       } else {
-               /* Inner IPv6 header. Mask/Val start at *inner_ofst into L5 */
-               if ((*inner_ofst + sizeof(struct ipv6_hdr)) >
-                    FILTER_GENERIC_1_KEY_LEN)
-                       return ENOTSUP;
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].mask[*inner_ofst],
-                      mask, sizeof(struct ipv6_hdr));
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].val[*inner_ofst],
-                      spec, sizeof(struct ipv6_hdr));
-               *inner_ofst += sizeof(struct ipv6_hdr);
-       }
+       memcpy(gp->layer[FILTER_GENERIC_1_L3].mask, &mask->hdr,
+              sizeof(struct ipv6_hdr));
+       memcpy(gp->layer[FILTER_GENERIC_1_L3].val, &spec->hdr,
+              sizeof(struct ipv6_hdr));
        return 0;
 }
 
-/**
- * Copy UDP item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Must be 0. Don't support inner UDP filtering.
- */
 static int
-enic_copy_item_udp_v2(const struct rte_flow_item *item,
-                     struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_udp_v2(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_udp *spec = item->spec;
        const struct rte_flow_item_udp *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
@@ -720,39 +830,18 @@ enic_copy_item_udp_v2(const struct rte_flow_item *item,
        if (!mask)
                mask = &rte_flow_item_udp_mask;
 
-       if (*inner_ofst == 0) {
-               memcpy(gp->layer[FILTER_GENERIC_1_L4].mask, &mask->hdr,
-                      sizeof(struct udp_hdr));
-               memcpy(gp->layer[FILTER_GENERIC_1_L4].val, &spec->hdr,
-                      sizeof(struct udp_hdr));
-       } else {
-               /* Inner IPv6 header. Mask/Val start at *inner_ofst into L5 */
-               if ((*inner_ofst + sizeof(struct udp_hdr)) >
-                    FILTER_GENERIC_1_KEY_LEN)
-                       return ENOTSUP;
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].mask[*inner_ofst],
-                      mask, sizeof(struct udp_hdr));
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].val[*inner_ofst],
-                      spec, sizeof(struct udp_hdr));
-               *inner_ofst += sizeof(struct udp_hdr);
-       }
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].mask, &mask->hdr,
+              sizeof(struct udp_hdr));
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].val, &spec->hdr,
+              sizeof(struct udp_hdr));
        return 0;
 }
 
-/**
- * Copy TCP item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Must be 0. Don't support inner TCP filtering.
- */
 static int
-enic_copy_item_tcp_v2(const struct rte_flow_item *item,
-                     struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_tcp_v2(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_tcp *spec = item->spec;
        const struct rte_flow_item_tcp *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
@@ -770,47 +859,48 @@ enic_copy_item_tcp_v2(const struct rte_flow_item *item,
        if (!mask)
                return ENOTSUP;
 
-       if (*inner_ofst == 0) {
-               memcpy(gp->layer[FILTER_GENERIC_1_L4].mask, &mask->hdr,
-                      sizeof(struct tcp_hdr));
-               memcpy(gp->layer[FILTER_GENERIC_1_L4].val, &spec->hdr,
-                      sizeof(struct tcp_hdr));
-       } else {
-               /* Inner IPv6 header. Mask/Val start at *inner_ofst into L5 */
-               if ((*inner_ofst + sizeof(struct tcp_hdr)) >
-                    FILTER_GENERIC_1_KEY_LEN)
-                       return ENOTSUP;
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].mask[*inner_ofst],
-                      mask, sizeof(struct tcp_hdr));
-               memcpy(&gp->layer[FILTER_GENERIC_1_L5].val[*inner_ofst],
-                      spec, sizeof(struct tcp_hdr));
-               *inner_ofst += sizeof(struct tcp_hdr);
-       }
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].mask, &mask->hdr,
+              sizeof(struct tcp_hdr));
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].val, &spec->hdr,
+              sizeof(struct tcp_hdr));
        return 0;
 }
 
-/**
- * Copy SCTP item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Must be 0. Don't support inner SCTP filtering.
- */
 static int
-enic_copy_item_sctp_v2(const struct rte_flow_item *item,
-                      struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_sctp_v2(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
        const struct rte_flow_item_sctp *spec = item->spec;
        const struct rte_flow_item_sctp *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
+       uint8_t *ip_proto_mask = NULL;
+       uint8_t *ip_proto = NULL;
 
        FLOW_TRACE();
 
-       if (*inner_ofst)
-               return ENOTSUP;
+       /*
+        * The NIC filter API has no flags for "match sctp", so explicitly set
+        * the protocol number in the IP pattern.
+        */
+       if (gp->val_flags & FILTER_GENERIC_1_IPV4) {
+               struct ipv4_hdr *ip;
+               ip = (struct ipv4_hdr *)gp->layer[FILTER_GENERIC_1_L3].mask;
+               ip_proto_mask = &ip->next_proto_id;
+               ip = (struct ipv4_hdr *)gp->layer[FILTER_GENERIC_1_L3].val;
+               ip_proto = &ip->next_proto_id;
+       } else if (gp->val_flags & FILTER_GENERIC_1_IPV6) {
+               struct ipv6_hdr *ip;
+               ip = (struct ipv6_hdr *)gp->layer[FILTER_GENERIC_1_L3].mask;
+               ip_proto_mask = &ip->proto;
+               ip = (struct ipv6_hdr *)gp->layer[FILTER_GENERIC_1_L3].val;
+               ip_proto = &ip->proto;
+       } else {
+               /* Need IPv4/IPv6 pattern first */
+               return EINVAL;
+       }
+       *ip_proto = IPPROTO_SCTP;
+       *ip_proto_mask = 0xff;
 
        /* Match all if no spec */
        if (!spec)
@@ -826,29 +916,29 @@ enic_copy_item_sctp_v2(const struct rte_flow_item *item,
        return 0;
 }
 
-/**
- * Copy UDP item into version 2 NIC filter.
- *
- * @param item[in]
- *   Item specification.
- * @param enic_filter[out]
- *   Partially filled in NIC filter structure.
- * @param inner_ofst[in]
- *   Must be 0. VxLAN headers always start at the beginning of L5.
- */
 static int
-enic_copy_item_vxlan_v2(const struct rte_flow_item *item,
-                       struct filter_v2 *enic_filter, u8 *inner_ofst)
+enic_copy_item_vxlan_v2(struct copy_item_args *arg)
 {
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
+       uint8_t *inner_ofst = arg->inner_ofst;
        const struct rte_flow_item_vxlan *spec = item->spec;
        const struct rte_flow_item_vxlan *mask = item->mask;
        struct filter_generic_1 *gp = &enic_filter->u.generic_1;
+       struct udp_hdr *udp;
 
        FLOW_TRACE();
 
-       if (*inner_ofst)
-               return EINVAL;
-
+       /*
+        * The NIC filter API has no flags for "match vxlan". Set UDP port to
+        * avoid false positives.
+        */
+       gp->mask_flags |= FILTER_GENERIC_1_UDP;
+       gp->val_flags |= FILTER_GENERIC_1_UDP;
+       udp = (struct udp_hdr *)gp->layer[FILTER_GENERIC_1_L4].mask;
+       udp->dst_port = 0xffff;
+       udp = (struct udp_hdr *)gp->layer[FILTER_GENERIC_1_L4].val;
+       udp->dst_port = RTE_BE16(4789);
        /* Match all if no spec */
        if (!spec)
                return 0;
@@ -857,11 +947,60 @@ enic_copy_item_vxlan_v2(const struct rte_flow_item *item,
                mask = &rte_flow_item_vxlan_mask;
 
        memcpy(gp->layer[FILTER_GENERIC_1_L5].mask, mask,
-              sizeof(struct vxlan_hdr));
+              sizeof(struct rte_vxlan_hdr));
        memcpy(gp->layer[FILTER_GENERIC_1_L5].val, spec,
-              sizeof(struct vxlan_hdr));
+              sizeof(struct rte_vxlan_hdr));
+
+       *inner_ofst = sizeof(struct rte_vxlan_hdr);
+       return 0;
+}
+
+/*
+ * Copy raw item into version 2 NIC filter. Currently, raw pattern match is
+ * very limited. It is intended for matching UDP tunnel header (e.g. vxlan
+ * or geneve).
+ */
+static int
+enic_copy_item_raw_v2(struct copy_item_args *arg)
+{
+       const struct rte_flow_item *item = arg->item;
+       struct filter_v2 *enic_filter = arg->filter;
+       uint8_t *inner_ofst = arg->inner_ofst;
+       const struct rte_flow_item_raw *spec = item->spec;
+       const struct rte_flow_item_raw *mask = item->mask;
+       struct filter_generic_1 *gp = &enic_filter->u.generic_1;
+
+       FLOW_TRACE();
+
+       /* Cannot be used for inner packet */
+       if (*inner_ofst)
+               return EINVAL;
+       /* Need both spec and mask */
+       if (!spec || !mask)
+               return EINVAL;
+       /* Only supports relative with offset 0 */
+       if (!spec->relative || spec->offset != 0 || spec->search || spec->limit)
+               return EINVAL;
+       /* Need non-null pattern that fits within the NIC's filter pattern */
+       if (spec->length == 0 ||
+           spec->length + sizeof(struct udp_hdr) > FILTER_GENERIC_1_KEY_LEN ||
+           !spec->pattern || !mask->pattern)
+               return EINVAL;
+       /*
+        * Mask fields, including length, are often set to zero. Assume that
+        * means "same as spec" to avoid breaking existing apps. If length
+        * is not zero, then it should be >= spec length.
+        *
+        * No more pattern follows this, so append to the L4 layer instead of
+        * L5 to work with both recent and older VICs.
+        */
+       if (mask->length != 0 && mask->length < spec->length)
+               return EINVAL;
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].mask + sizeof(struct udp_hdr),
+              mask->pattern, spec->length);
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].val + sizeof(struct udp_hdr),
+              spec->pattern, spec->length);
 
-       *inner_ofst = sizeof(struct vxlan_hdr);
        return 0;
 }
 
@@ -896,6 +1035,36 @@ item_stacking_valid(enum rte_flow_item_type prev_item,
        return 0;
 }
 
+/*
+ * Fix up the L5 layer.. HW vxlan parsing removes vxlan header from L5.
+ * Instead it is in L4 following the UDP header. Append the vxlan
+ * pattern to L4 (udp) and shift any inner packet pattern in L5.
+ */
+static void
+fixup_l5_layer(struct enic *enic, struct filter_generic_1 *gp,
+              uint8_t inner_ofst)
+{
+       uint8_t layer[FILTER_GENERIC_1_KEY_LEN];
+       uint8_t inner;
+       uint8_t vxlan;
+
+       if (!(inner_ofst > 0 && enic->vxlan))
+               return;
+       FLOW_TRACE();
+       vxlan = sizeof(struct rte_vxlan_hdr);
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].mask + sizeof(struct udp_hdr),
+              gp->layer[FILTER_GENERIC_1_L5].mask, vxlan);
+       memcpy(gp->layer[FILTER_GENERIC_1_L4].val + sizeof(struct udp_hdr),
+              gp->layer[FILTER_GENERIC_1_L5].val, vxlan);
+       inner = inner_ofst - vxlan;
+       memset(layer, 0, sizeof(layer));
+       memcpy(layer, gp->layer[FILTER_GENERIC_1_L5].mask + vxlan, inner);
+       memcpy(gp->layer[FILTER_GENERIC_1_L5].mask, layer, sizeof(layer));
+       memset(layer, 0, sizeof(layer));
+       memcpy(layer, gp->layer[FILTER_GENERIC_1_L5].val + vxlan, inner);
+       memcpy(gp->layer[FILTER_GENERIC_1_L5].val, layer, sizeof(layer));
+}
+
 /**
  * Build the intenal enic filter structure from the provided pattern. The
  * pattern is validated as the items are copied.
@@ -909,7 +1078,8 @@ item_stacking_valid(enum rte_flow_item_type prev_item,
  */
 static int
 enic_copy_filter(const struct rte_flow_item pattern[],
-                const struct enic_items *items_info,
+                const struct enic_filter_cap *cap,
+                struct enic *enic,
                 struct filter_v2 *enic_filter,
                 struct rte_flow_error *error)
 {
@@ -918,13 +1088,17 @@ enic_copy_filter(const struct rte_flow_item pattern[],
        u8 inner_ofst = 0; /* If encapsulated, ofst into L5 */
        enum rte_flow_item_type prev_item;
        const struct enic_items *item_info;
-
+       struct copy_item_args args;
+       enic_copy_item_fn *copy_fn;
        u8 is_first_item = 1;
 
        FLOW_TRACE();
 
        prev_item = 0;
 
+       args.filter = enic_filter;
+       args.inner_ofst = &inner_ofst;
+       args.enic = enic;
        for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
                /* Get info about how to validate and copy the item. If NULL
                 * is returned the nic does not support the item.
@@ -932,18 +1106,31 @@ enic_copy_filter(const struct rte_flow_item pattern[],
                if (item->type == RTE_FLOW_ITEM_TYPE_VOID)
                        continue;
 
-               item_info = &items_info[item->type];
+               item_info = &cap->item_info[item->type];
+               if (item->type > cap->max_item_type ||
+                   item_info->copy_item == NULL ||
+                   (inner_ofst > 0 && item_info->inner_copy_item == NULL)) {
+                       rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ITEM,
+                               NULL, "Unsupported item.");
+                       return -rte_errno;
+               }
 
                /* check to see if item stacking is valid */
                if (!item_stacking_valid(prev_item, item_info, is_first_item))
                        goto stacking_error;
 
-               ret = item_info->copy_item(item, enic_filter, &inner_ofst);
+               args.item = item;
+               copy_fn = inner_ofst > 0 ? item_info->inner_copy_item :
+                       item_info->copy_item;
+               ret = copy_fn(&args);
                if (ret)
                        goto item_not_supported;
                prev_item = item->type;
                is_first_item = 0;
        }
+       fixup_l5_layer(enic, &enic_filter->u.generic_1, inner_ofst);
+
        return 0;
 
 item_not_supported:
@@ -967,7 +1154,8 @@ stacking_error:
  * @param error[out]
  */
 static int
-enic_copy_action_v1(const struct rte_flow_action actions[],
+enic_copy_action_v1(__rte_unused struct enic *enic,
+                   const struct rte_flow_action actions[],
                    struct filter_action_v2 *enic_action)
 {
        enum { FATE = 1, };
@@ -1013,11 +1201,13 @@ enic_copy_action_v1(const struct rte_flow_action actions[],
  * @param error[out]
  */
 static int
-enic_copy_action_v2(const struct rte_flow_action actions[],
+enic_copy_action_v2(struct enic *enic,
+                   const struct rte_flow_action actions[],
                    struct filter_action_v2 *enic_action)
 {
        enum { FATE = 1, MARK = 2, };
        uint32_t overlap = 0;
+       bool passthru = false;
 
        FLOW_TRACE();
 
@@ -1044,12 +1234,18 @@ enic_copy_action_v2(const struct rte_flow_action actions[],
                        if (overlap & MARK)
                                return ENOTSUP;
                        overlap |= MARK;
-                       /* ENIC_MAGIC_FILTER_ID is reserved and is the highest
-                        * in the range of allows mark ids.
+                       /*
+                        * Map mark ID (32-bit) to filter ID (16-bit):
+                        * - Reject values > 16 bits
+                        * - Filter ID 0 is reserved for filters that steer
+                        *   but not mark. So add 1 to the mark ID to avoid
+                        *   using 0.
+                        * - Filter ID (ENIC_MAGIC_FILTER_ID = 0xffff) is
+                        *   reserved for the "flag" action below.
                         */
-                       if (mark->id >= ENIC_MAGIC_FILTER_ID)
+                       if (mark->id >= ENIC_MAGIC_FILTER_ID - 1)
                                return EINVAL;
-                       enic_action->filter_id = mark->id;
+                       enic_action->filter_id = mark->id + 1;
                        enic_action->flags |= FILTER_ACTION_FILTER_ID_FLAG;
                        break;
                }
@@ -1057,6 +1253,7 @@ enic_copy_action_v2(const struct rte_flow_action actions[],
                        if (overlap & MARK)
                                return ENOTSUP;
                        overlap |= MARK;
+                       /* ENIC_MAGIC_FILTER_ID is reserved for flagging */
                        enic_action->filter_id = ENIC_MAGIC_FILTER_ID;
                        enic_action->flags |= FILTER_ACTION_FILTER_ID_FLAG;
                        break;
@@ -1068,6 +1265,54 @@ enic_copy_action_v2(const struct rte_flow_action actions[],
                        enic_action->flags |= FILTER_ACTION_DROP_FLAG;
                        break;
                }
+               case RTE_FLOW_ACTION_TYPE_COUNT: {
+                       enic_action->flags |= FILTER_ACTION_COUNTER_FLAG;
+                       break;
+               }
+               case RTE_FLOW_ACTION_TYPE_RSS: {
+                       const struct rte_flow_action_rss *rss =
+                               (const struct rte_flow_action_rss *)
+                               actions->conf;
+                       bool allow;
+                       uint16_t i;
+
+                       /*
+                        * Hardware does not support general RSS actions, but
+                        * we can still support the dummy one that is used to
+                        * "receive normally".
+                        */
+                       allow = rss->func == RTE_ETH_HASH_FUNCTION_DEFAULT &&
+                               rss->level == 0 &&
+                               (rss->types == 0 ||
+                                rss->types == enic->rss_hf) &&
+                               rss->queue_num == enic->rq_count &&
+                               rss->key_len == 0;
+                       /* Identity queue map is ok */
+                       for (i = 0; i < rss->queue_num; i++)
+                               allow = allow && (i == rss->queue[i]);
+                       if (!allow)
+                               return ENOTSUP;
+                       if (overlap & FATE)
+                               return ENOTSUP;
+                       /* Need MARK or FLAG */
+                       if (!(overlap & MARK))
+                               return ENOTSUP;
+                       overlap |= FATE;
+                       break;
+               }
+               case RTE_FLOW_ACTION_TYPE_PASSTHRU: {
+                       /*
+                        * Like RSS above, PASSTHRU + MARK may be used to
+                        * "mark and then receive normally". MARK usually comes
+                        * after PASSTHRU, so remember we have seen passthru
+                        * and check for mark later.
+                        */
+                       if (overlap & FATE)
+                               return ENOTSUP;
+                       overlap |= FATE;
+                       passthru = true;
+                       break;
+               }
                case RTE_FLOW_ACTION_TYPE_VOID:
                        continue;
                default:
@@ -1075,6 +1320,9 @@ enic_copy_action_v2(const struct rte_flow_action actions[],
                        break;
                }
        }
+       /* Only PASSTHRU + MARK is allowed */
+       if (passthru && !(overlap & MARK))
+               return ENOTSUP;
        if (!(overlap & FATE))
                return ENOTSUP;
        enic_action->type = FILTER_ACTION_V2;
@@ -1112,7 +1360,9 @@ enic_get_action_cap(struct enic *enic)
        uint8_t actions;
 
        actions = enic->filter_actions;
-       if (actions & FILTER_ACTION_DROP_FLAG)
+       if (actions & FILTER_ACTION_COUNTER_FLAG)
+               ea = &enic_action_cap[FILTER_ACTION_COUNTER_FLAG];
+       else if (actions & FILTER_ACTION_DROP_FLAG)
                ea = &enic_action_cap[FILTER_ACTION_DROP_FLAG];
        else if (actions & FILTER_ACTION_FILTER_ID_FLAG)
                ea = &enic_action_cap[FILTER_ACTION_FILTER_ID_FLAG];
@@ -1356,7 +1606,7 @@ enic_flow_parse(struct rte_eth_dev *dev,
                                   action, "Invalid action.");
                return -rte_errno;
        }
-       ret = enic_action_cap->copy_fn(actions, enic_action);
+       ret = enic_action_cap->copy_fn(enic, actions, enic_action);
        if (ret) {
                rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
                           NULL, "Unsupported action.");
@@ -1373,7 +1623,7 @@ enic_flow_parse(struct rte_eth_dev *dev,
                return -rte_errno;
        }
        enic_filter->type = enic->flow_filter_mode;
-       ret = enic_copy_filter(pattern, enic_filter_cap->item_info,
+       ret = enic_copy_filter(pattern, enic_filter_cap, enic,
                                       enic_filter, error);
        return ret;
 }
@@ -1395,8 +1645,10 @@ enic_flow_add_filter(struct enic *enic, struct filter_v2 *enic_filter,
                   struct rte_flow_error *error)
 {
        struct rte_flow *flow;
-       int ret;
-       u16 entry;
+       int err;
+       uint16_t entry;
+       int ctr_idx;
+       int last_max_flow_ctr;
 
        FLOW_TRACE();
 
@@ -1407,20 +1659,64 @@ enic_flow_add_filter(struct enic *enic, struct filter_v2 *enic_filter,
                return NULL;
        }
 
+       flow->counter_idx = -1;
+       last_max_flow_ctr = -1;
+       if (enic_action->flags & FILTER_ACTION_COUNTER_FLAG) {
+               if (!vnic_dev_counter_alloc(enic->vdev, (uint32_t *)&ctr_idx)) {
+                       rte_flow_error_set(error, ENOMEM,
+                                          RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+                                          NULL, "cannot allocate counter");
+                       goto unwind_flow_alloc;
+               }
+               flow->counter_idx = ctr_idx;
+               enic_action->counter_index = ctr_idx;
+
+               /* If index is the largest, increase the counter DMA size */
+               if (ctr_idx > enic->max_flow_counter) {
+                       err = vnic_dev_counter_dma_cfg(enic->vdev,
+                                                VNIC_FLOW_COUNTER_UPDATE_MSECS,
+                                                ctr_idx + 1);
+                       if (err) {
+                               rte_flow_error_set(error, -err,
+                                          RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+                                          NULL, "counter DMA config failed");
+                               goto unwind_ctr_alloc;
+                       }
+                       last_max_flow_ctr = enic->max_flow_counter;
+                       enic->max_flow_counter = ctr_idx;
+               }
+       }
+
        /* entry[in] is the queue id, entry[out] is the filter Id for delete */
        entry = enic_action->rq_idx;
-       ret = vnic_dev_classifier(enic->vdev, CLSF_ADD, &entry, enic_filter,
+       err = vnic_dev_classifier(enic->vdev, CLSF_ADD, &entry, enic_filter,
                                  enic_action);
-       if (!ret) {
-               flow->enic_filter_id = entry;
-               flow->enic_filter = *enic_filter;
-       } else {
-               rte_flow_error_set(error, ret, RTE_FLOW_ERROR_TYPE_HANDLE,
+       if (err) {
+               rte_flow_error_set(error, -err, RTE_FLOW_ERROR_TYPE_HANDLE,
                                   NULL, "vnic_dev_classifier error");
-               rte_free(flow);
-               return NULL;
+               goto unwind_ctr_dma_cfg;
        }
+
+       flow->enic_filter_id = entry;
+       flow->enic_filter = *enic_filter;
+
        return flow;
+
+/* unwind if there are errors */
+unwind_ctr_dma_cfg:
+       if (last_max_flow_ctr != -1) {
+               /* reduce counter DMA size */
+               vnic_dev_counter_dma_cfg(enic->vdev,
+                                        VNIC_FLOW_COUNTER_UPDATE_MSECS,
+                                        last_max_flow_ctr + 1);
+               enic->max_flow_counter = last_max_flow_ctr;
+       }
+unwind_ctr_alloc:
+       if (flow->counter_idx != -1)
+               vnic_dev_counter_free(enic->vdev, ctr_idx);
+unwind_flow_alloc:
+       rte_free(flow);
+       return NULL;
 }
 
 /**
@@ -1435,18 +1731,29 @@ enic_flow_add_filter(struct enic *enic, struct filter_v2 *enic_filter,
  * @param error[out]
  */
 static int
-enic_flow_del_filter(struct enic *enic, u16 filter_id,
+enic_flow_del_filter(struct enic *enic, struct rte_flow *flow,
                   struct rte_flow_error *error)
 {
-       int ret;
+       u16 filter_id;
+       int err;
 
        FLOW_TRACE();
 
-       ret = vnic_dev_classifier(enic->vdev, CLSF_DEL, &filter_id, NULL, NULL);
-       if (!ret)
-               rte_flow_error_set(error, ret, RTE_FLOW_ERROR_TYPE_HANDLE,
+       filter_id = flow->enic_filter_id;
+       err = vnic_dev_classifier(enic->vdev, CLSF_DEL, &filter_id, NULL, NULL);
+       if (err) {
+               rte_flow_error_set(error, -err, RTE_FLOW_ERROR_TYPE_HANDLE,
                                   NULL, "vnic_dev_classifier failed");
-       return ret;
+               return -err;
+       }
+
+       if (flow->counter_idx != -1) {
+               if (!vnic_dev_counter_free(enic->vdev, flow->counter_idx))
+                       dev_err(enic, "counter free failed, idx: %d\n",
+                               flow->counter_idx);
+               flow->counter_idx = -1;
+       }
+       return 0;
 }
 
 /*
@@ -1529,9 +1836,10 @@ enic_flow_destroy(struct rte_eth_dev *dev, struct rte_flow *flow,
        FLOW_TRACE();
 
        rte_spinlock_lock(&enic->flows_lock);
-       enic_flow_del_filter(enic, flow->enic_filter_id, error);
+       enic_flow_del_filter(enic, flow, error);
        LIST_REMOVE(flow, next);
        rte_spinlock_unlock(&enic->flows_lock);
+       rte_free(flow);
        return 0;
 }
 
@@ -1553,13 +1861,77 @@ enic_flow_flush(struct rte_eth_dev *dev, struct rte_flow_error *error)
 
        while (!LIST_EMPTY(&enic->flows)) {
                flow = LIST_FIRST(&enic->flows);
-               enic_flow_del_filter(enic, flow->enic_filter_id, error);
+               enic_flow_del_filter(enic, flow, error);
                LIST_REMOVE(flow, next);
+               rte_free(flow);
        }
        rte_spinlock_unlock(&enic->flows_lock);
        return 0;
 }
 
+static int
+enic_flow_query_count(struct rte_eth_dev *dev,
+                     struct rte_flow *flow, void *data,
+                     struct rte_flow_error *error)
+{
+       struct enic *enic = pmd_priv(dev);
+       struct rte_flow_query_count *query;
+       uint64_t packets, bytes;
+
+       FLOW_TRACE();
+
+       if (flow->counter_idx == -1) {
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                         NULL,
+                                         "flow does not have counter");
+       }
+       query = (struct rte_flow_query_count *)data;
+       if (!vnic_dev_counter_query(enic->vdev, flow->counter_idx,
+                                   !!query->reset, &packets, &bytes)) {
+               return rte_flow_error_set
+                       (error, EINVAL,
+                        RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                        NULL,
+                        "cannot read counter");
+       }
+       query->hits_set = 1;
+       query->bytes_set = 1;
+       query->hits = packets;
+       query->bytes = bytes;
+       return 0;
+}
+
+static int
+enic_flow_query(struct rte_eth_dev *dev,
+               struct rte_flow *flow,
+               const struct rte_flow_action *actions,
+               void *data,
+               struct rte_flow_error *error)
+{
+       int ret = 0;
+
+       FLOW_TRACE();
+
+       for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+               switch (actions->type) {
+               case RTE_FLOW_ACTION_TYPE_VOID:
+                       break;
+               case RTE_FLOW_ACTION_TYPE_COUNT:
+                       ret = enic_flow_query_count(dev, flow, data, error);
+                       break;
+               default:
+                       return rte_flow_error_set(error, ENOTSUP,
+                                                 RTE_FLOW_ERROR_TYPE_ACTION,
+                                                 actions,
+                                                 "action not supported");
+               }
+               if (ret < 0)
+                       return ret;
+       }
+       return 0;
+}
+
 /**
  * Flow callback registration.
  *
@@ -1570,4 +1942,5 @@ const struct rte_flow_ops enic_flow_ops = {
        .create = enic_flow_create,
        .destroy = enic_flow_destroy,
        .flush = enic_flow_flush,
+       .query = enic_flow_query,
 };