net/sfc: fix outer rule and encap rollback on error
[dpdk.git] / drivers / net / sfc / sfc_mae.c
index 42200c3..0270c91 100644 (file)
@@ -1,6 +1,6 @@
 /* SPDX-License-Identifier: BSD-3-Clause
  *
- * Copyright(c) 2019-2020 Xilinx, Inc.
+ * Copyright(c) 2019-2021 Xilinx, Inc.
  * Copyright(c) 2019 Solarflare Communications Inc.
  *
  * This software was jointly developed between OKTET Labs (under contract
 
 #include <stdbool.h>
 
+#include <rte_bitops.h>
 #include <rte_common.h>
+#include <rte_vxlan.h>
 
 #include "efx.h"
 
 #include "sfc.h"
 #include "sfc_log.h"
+#include "sfc_switch.h"
+
+static int
+sfc_mae_assign_entity_mport(struct sfc_adapter *sa,
+                           efx_mport_sel_t *mportp)
+{
+       const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
+
+       return efx_mae_mport_by_pcie_function(encp->enc_pf, encp->enc_vf,
+                                             mportp);
+}
 
 int
 sfc_mae_attach(struct sfc_adapter *sa)
 {
-       const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
-       struct sfc_mae *mae = &sa->mae;
-       efx_mae_limits_t limits;
-       int rc;
+       struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
+       struct sfc_mae_switch_port_request switch_port_request = {0};
+       const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
+       efx_mport_sel_t entity_mport;
+       struct sfc_mae *mae = &sa->mae;
+       struct sfc_mae_bounce_eh *bounce_eh = &mae->bounce_eh;
+       efx_mae_limits_t limits;
+       int rc;
+
+       sfc_log_init(sa, "entry");
+
+       if (!encp->enc_mae_supported) {
+               mae->status = SFC_MAE_STATUS_UNSUPPORTED;
+               return 0;
+       }
+
+       sfc_log_init(sa, "init MAE");
+       rc = efx_mae_init(sa->nic);
+       if (rc != 0)
+               goto fail_mae_init;
+
+       sfc_log_init(sa, "get MAE limits");
+       rc = efx_mae_get_limits(sa->nic, &limits);
+       if (rc != 0)
+               goto fail_mae_get_limits;
+
+       sfc_log_init(sa, "assign entity MPORT");
+       rc = sfc_mae_assign_entity_mport(sa, &entity_mport);
+       if (rc != 0)
+               goto fail_mae_assign_entity_mport;
+
+       sfc_log_init(sa, "assign RTE switch domain");
+       rc = sfc_mae_assign_switch_domain(sa, &mae->switch_domain_id);
+       if (rc != 0)
+               goto fail_mae_assign_switch_domain;
+
+       sfc_log_init(sa, "assign RTE switch port");
+       switch_port_request.type = SFC_MAE_SWITCH_PORT_INDEPENDENT;
+       switch_port_request.entity_mportp = &entity_mport;
+       /*
+        * As of now, the driver does not support representors, so
+        * RTE ethdev MPORT simply matches that of the entity.
+        */
+       switch_port_request.ethdev_mportp = &entity_mport;
+       switch_port_request.ethdev_port_id = sas->port_id;
+       rc = sfc_mae_assign_switch_port(mae->switch_domain_id,
+                                       &switch_port_request,
+                                       &mae->switch_port_id);
+       if (rc != 0)
+               goto fail_mae_assign_switch_port;
+
+       sfc_log_init(sa, "allocate encap. header bounce buffer");
+       bounce_eh->buf_size = limits.eml_encap_header_size_limit;
+       bounce_eh->buf = rte_malloc("sfc_mae_bounce_eh",
+                                   bounce_eh->buf_size, 0);
+       if (bounce_eh->buf == NULL)
+               goto fail_mae_alloc_bounce_eh;
+
+       mae->status = SFC_MAE_STATUS_SUPPORTED;
+       mae->nb_outer_rule_prios_max = limits.eml_max_n_outer_prios;
+       mae->nb_action_rule_prios_max = limits.eml_max_n_action_prios;
+       mae->encap_types_supported = limits.eml_encap_types_supported;
+       TAILQ_INIT(&mae->outer_rules);
+       TAILQ_INIT(&mae->encap_headers);
+       TAILQ_INIT(&mae->action_sets);
+
+       sfc_log_init(sa, "done");
+
+       return 0;
+
+fail_mae_alloc_bounce_eh:
+fail_mae_assign_switch_port:
+fail_mae_assign_switch_domain:
+fail_mae_assign_entity_mport:
+fail_mae_get_limits:
+       efx_mae_fini(sa->nic);
+
+fail_mae_init:
+       sfc_log_init(sa, "failed %d", rc);
+
+       return rc;
+}
+
+void
+sfc_mae_detach(struct sfc_adapter *sa)
+{
+       struct sfc_mae *mae = &sa->mae;
+       enum sfc_mae_status status_prev = mae->status;
+
+       sfc_log_init(sa, "entry");
+
+       mae->nb_action_rule_prios_max = 0;
+       mae->status = SFC_MAE_STATUS_UNKNOWN;
+
+       if (status_prev != SFC_MAE_STATUS_SUPPORTED)
+               return;
+
+       rte_free(mae->bounce_eh.buf);
+
+       efx_mae_fini(sa->nic);
+
+       sfc_log_init(sa, "done");
+}
+
+static struct sfc_mae_outer_rule *
+sfc_mae_outer_rule_attach(struct sfc_adapter *sa,
+                         const efx_mae_match_spec_t *match_spec,
+                         efx_tunnel_protocol_t encap_type)
+{
+       struct sfc_mae_outer_rule *rule;
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       TAILQ_FOREACH(rule, &mae->outer_rules, entries) {
+               if (efx_mae_match_specs_equal(rule->match_spec, match_spec) &&
+                   rule->encap_type == encap_type) {
+                       ++(rule->refcnt);
+                       return rule;
+               }
+       }
+
+       return NULL;
+}
+
+static int
+sfc_mae_outer_rule_add(struct sfc_adapter *sa,
+                      efx_mae_match_spec_t *match_spec,
+                      efx_tunnel_protocol_t encap_type,
+                      struct sfc_mae_outer_rule **rulep)
+{
+       struct sfc_mae_outer_rule *rule;
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       rule = rte_zmalloc("sfc_mae_outer_rule", sizeof(*rule), 0);
+       if (rule == NULL)
+               return ENOMEM;
+
+       rule->refcnt = 1;
+       rule->match_spec = match_spec;
+       rule->encap_type = encap_type;
+
+       rule->fw_rsrc.rule_id.id = EFX_MAE_RSRC_ID_INVALID;
+
+       TAILQ_INSERT_TAIL(&mae->outer_rules, rule, entries);
+
+       *rulep = rule;
+
+       return 0;
+}
+
+static void
+sfc_mae_outer_rule_del(struct sfc_adapter *sa,
+                      struct sfc_mae_outer_rule *rule)
+{
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+       SFC_ASSERT(rule->refcnt != 0);
+
+       --(rule->refcnt);
+
+       if (rule->refcnt != 0)
+               return;
+
+       SFC_ASSERT(rule->fw_rsrc.rule_id.id == EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(rule->fw_rsrc.refcnt == 0);
+
+       efx_mae_match_spec_fini(sa->nic, rule->match_spec);
+
+       TAILQ_REMOVE(&mae->outer_rules, rule, entries);
+       rte_free(rule);
+}
+
+static int
+sfc_mae_outer_rule_enable(struct sfc_adapter *sa,
+                         struct sfc_mae_outer_rule *rule,
+                         efx_mae_match_spec_t *match_spec_action)
+{
+       struct sfc_mae_fw_rsrc *fw_rsrc = &rule->fw_rsrc;
+       int rc;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       if (fw_rsrc->refcnt == 0) {
+               SFC_ASSERT(fw_rsrc->rule_id.id == EFX_MAE_RSRC_ID_INVALID);
+               SFC_ASSERT(rule->match_spec != NULL);
+
+               rc = efx_mae_outer_rule_insert(sa->nic, rule->match_spec,
+                                              rule->encap_type,
+                                              &fw_rsrc->rule_id);
+               if (rc != 0)
+                       return rc;
+       }
+
+       rc = efx_mae_match_spec_outer_rule_id_set(match_spec_action,
+                                                 &fw_rsrc->rule_id);
+       if (rc != 0) {
+               if (fw_rsrc->refcnt == 0) {
+                       (void)efx_mae_outer_rule_remove(sa->nic,
+                                                       &fw_rsrc->rule_id);
+                       fw_rsrc->rule_id.id = EFX_MAE_RSRC_ID_INVALID;
+               }
+               return rc;
+       }
+
+       ++(fw_rsrc->refcnt);
+
+       return 0;
+}
+
+static int
+sfc_mae_outer_rule_disable(struct sfc_adapter *sa,
+                          struct sfc_mae_outer_rule *rule)
+{
+       struct sfc_mae_fw_rsrc *fw_rsrc = &rule->fw_rsrc;
+       int rc;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+       SFC_ASSERT(fw_rsrc->rule_id.id != EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(fw_rsrc->refcnt != 0);
+
+       if (fw_rsrc->refcnt == 1) {
+               rc = efx_mae_outer_rule_remove(sa->nic, &fw_rsrc->rule_id);
+               if (rc != 0)
+                       return rc;
+
+               fw_rsrc->rule_id.id = EFX_MAE_RSRC_ID_INVALID;
+       }
+
+       --(fw_rsrc->refcnt);
+
+       return 0;
+}
+
+static struct sfc_mae_encap_header *
+sfc_mae_encap_header_attach(struct sfc_adapter *sa,
+                           const struct sfc_mae_bounce_eh *bounce_eh)
+{
+       struct sfc_mae_encap_header *encap_header;
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       TAILQ_FOREACH(encap_header, &mae->encap_headers, entries) {
+               if (encap_header->size == bounce_eh->size &&
+                   memcmp(encap_header->buf, bounce_eh->buf,
+                          bounce_eh->size) == 0) {
+                       ++(encap_header->refcnt);
+                       return encap_header;
+               }
+       }
+
+       return NULL;
+}
+
+static int
+sfc_mae_encap_header_add(struct sfc_adapter *sa,
+                        const struct sfc_mae_bounce_eh *bounce_eh,
+                        struct sfc_mae_encap_header **encap_headerp)
+{
+       struct sfc_mae_encap_header *encap_header;
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       encap_header = rte_zmalloc("sfc_mae_encap_header",
+                                  sizeof(*encap_header), 0);
+       if (encap_header == NULL)
+               return ENOMEM;
+
+       encap_header->size = bounce_eh->size;
+
+       encap_header->buf = rte_malloc("sfc_mae_encap_header_buf",
+                                      encap_header->size, 0);
+       if (encap_header->buf == NULL) {
+               rte_free(encap_header);
+               return ENOMEM;
+       }
+
+       rte_memcpy(encap_header->buf, bounce_eh->buf, bounce_eh->size);
+
+       encap_header->refcnt = 1;
+       encap_header->type = bounce_eh->type;
+       encap_header->fw_rsrc.eh_id.id = EFX_MAE_RSRC_ID_INVALID;
+
+       TAILQ_INSERT_TAIL(&mae->encap_headers, encap_header, entries);
+
+       *encap_headerp = encap_header;
+
+       return 0;
+}
+
+static void
+sfc_mae_encap_header_del(struct sfc_adapter *sa,
+                      struct sfc_mae_encap_header *encap_header)
+{
+       struct sfc_mae *mae = &sa->mae;
+
+       if (encap_header == NULL)
+               return;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+       SFC_ASSERT(encap_header->refcnt != 0);
+
+       --(encap_header->refcnt);
+
+       if (encap_header->refcnt != 0)
+               return;
+
+       SFC_ASSERT(encap_header->fw_rsrc.eh_id.id == EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(encap_header->fw_rsrc.refcnt == 0);
+
+       TAILQ_REMOVE(&mae->encap_headers, encap_header, entries);
+       rte_free(encap_header->buf);
+       rte_free(encap_header);
+}
+
+static int
+sfc_mae_encap_header_enable(struct sfc_adapter *sa,
+                           struct sfc_mae_encap_header *encap_header,
+                           efx_mae_actions_t *action_set_spec)
+{
+       struct sfc_mae_fw_rsrc *fw_rsrc;
+       int rc;
+
+       if (encap_header == NULL)
+               return 0;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       fw_rsrc = &encap_header->fw_rsrc;
+
+       if (fw_rsrc->refcnt == 0) {
+               SFC_ASSERT(fw_rsrc->eh_id.id == EFX_MAE_RSRC_ID_INVALID);
+               SFC_ASSERT(encap_header->buf != NULL);
+               SFC_ASSERT(encap_header->size != 0);
+
+               rc = efx_mae_encap_header_alloc(sa->nic, encap_header->type,
+                                               encap_header->buf,
+                                               encap_header->size,
+                                               &fw_rsrc->eh_id);
+               if (rc != 0)
+                       return rc;
+       }
+
+       rc = efx_mae_action_set_fill_in_eh_id(action_set_spec,
+                                             &fw_rsrc->eh_id);
+       if (rc != 0) {
+               if (fw_rsrc->refcnt == 0) {
+                       (void)efx_mae_encap_header_free(sa->nic,
+                                                       &fw_rsrc->eh_id);
+                       fw_rsrc->eh_id.id = EFX_MAE_RSRC_ID_INVALID;
+               }
+               return rc;
+       }
+
+       ++(fw_rsrc->refcnt);
+
+       return 0;
+}
+
+static int
+sfc_mae_encap_header_disable(struct sfc_adapter *sa,
+                            struct sfc_mae_encap_header *encap_header)
+{
+       struct sfc_mae_fw_rsrc *fw_rsrc;
+       int rc;
+
+       if (encap_header == NULL)
+               return 0;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       fw_rsrc = &encap_header->fw_rsrc;
+
+       SFC_ASSERT(fw_rsrc->eh_id.id != EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(fw_rsrc->refcnt != 0);
+
+       if (fw_rsrc->refcnt == 1) {
+               rc = efx_mae_encap_header_free(sa->nic, &fw_rsrc->eh_id);
+               if (rc != 0)
+                       return rc;
+
+               fw_rsrc->eh_id.id = EFX_MAE_RSRC_ID_INVALID;
+       }
+
+       --(fw_rsrc->refcnt);
+
+       return 0;
+}
+
+static struct sfc_mae_action_set *
+sfc_mae_action_set_attach(struct sfc_adapter *sa,
+                         const struct sfc_mae_encap_header *encap_header,
+                         const efx_mae_actions_t *spec)
+{
+       struct sfc_mae_action_set *action_set;
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       TAILQ_FOREACH(action_set, &mae->action_sets, entries) {
+               if (action_set->encap_header == encap_header &&
+                   efx_mae_action_set_specs_equal(action_set->spec, spec)) {
+                       ++(action_set->refcnt);
+                       return action_set;
+               }
+       }
+
+       return NULL;
+}
+
+static int
+sfc_mae_action_set_add(struct sfc_adapter *sa,
+                      efx_mae_actions_t *spec,
+                      struct sfc_mae_encap_header *encap_header,
+                      struct sfc_mae_action_set **action_setp)
+{
+       struct sfc_mae_action_set *action_set;
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       action_set = rte_zmalloc("sfc_mae_action_set", sizeof(*action_set), 0);
+       if (action_set == NULL)
+               return ENOMEM;
+
+       action_set->refcnt = 1;
+       action_set->spec = spec;
+       action_set->encap_header = encap_header;
+
+       action_set->fw_rsrc.aset_id.id = EFX_MAE_RSRC_ID_INVALID;
+
+       TAILQ_INSERT_TAIL(&mae->action_sets, action_set, entries);
+
+       *action_setp = action_set;
+
+       return 0;
+}
+
+static void
+sfc_mae_action_set_del(struct sfc_adapter *sa,
+                      struct sfc_mae_action_set *action_set)
+{
+       struct sfc_mae *mae = &sa->mae;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+       SFC_ASSERT(action_set->refcnt != 0);
+
+       --(action_set->refcnt);
+
+       if (action_set->refcnt != 0)
+               return;
+
+       SFC_ASSERT(action_set->fw_rsrc.aset_id.id == EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(action_set->fw_rsrc.refcnt == 0);
+
+       efx_mae_action_set_spec_fini(sa->nic, action_set->spec);
+       sfc_mae_encap_header_del(sa, action_set->encap_header);
+       TAILQ_REMOVE(&mae->action_sets, action_set, entries);
+       rte_free(action_set);
+}
+
+static int
+sfc_mae_action_set_enable(struct sfc_adapter *sa,
+                         struct sfc_mae_action_set *action_set)
+{
+       struct sfc_mae_encap_header *encap_header = action_set->encap_header;
+       struct sfc_mae_fw_rsrc *fw_rsrc = &action_set->fw_rsrc;
+       int rc;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+       if (fw_rsrc->refcnt == 0) {
+               SFC_ASSERT(fw_rsrc->aset_id.id == EFX_MAE_RSRC_ID_INVALID);
+               SFC_ASSERT(action_set->spec != NULL);
+
+               rc = sfc_mae_encap_header_enable(sa, encap_header,
+                                                action_set->spec);
+               if (rc != 0)
+                       return rc;
+
+               rc = efx_mae_action_set_alloc(sa->nic, action_set->spec,
+                                             &fw_rsrc->aset_id);
+               if (rc != 0) {
+                       (void)sfc_mae_encap_header_disable(sa, encap_header);
+
+                       return rc;
+               }
+       }
+
+       ++(fw_rsrc->refcnt);
+
+       return 0;
+}
+
+static int
+sfc_mae_action_set_disable(struct sfc_adapter *sa,
+                          struct sfc_mae_action_set *action_set)
+{
+       struct sfc_mae_fw_rsrc *fw_rsrc = &action_set->fw_rsrc;
+       int rc;
+
+       SFC_ASSERT(sfc_adapter_is_locked(sa));
+       SFC_ASSERT(fw_rsrc->aset_id.id != EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(fw_rsrc->refcnt != 0);
+
+       if (fw_rsrc->refcnt == 1) {
+               rc = efx_mae_action_set_free(sa->nic, &fw_rsrc->aset_id);
+               if (rc != 0)
+                       return rc;
+
+               fw_rsrc->aset_id.id = EFX_MAE_RSRC_ID_INVALID;
+
+               rc = sfc_mae_encap_header_disable(sa, action_set->encap_header);
+               if (rc != 0)
+                       return rc;
+       }
+
+       --(fw_rsrc->refcnt);
+
+       return 0;
+}
+
+void
+sfc_mae_flow_cleanup(struct sfc_adapter *sa,
+                    struct rte_flow *flow)
+{
+       struct sfc_flow_spec *spec;
+       struct sfc_flow_spec_mae *spec_mae;
+
+       if (flow == NULL)
+               return;
+
+       spec = &flow->spec;
+
+       if (spec == NULL)
+               return;
+
+       spec_mae = &spec->mae;
+
+       SFC_ASSERT(spec_mae->rule_id.id == EFX_MAE_RSRC_ID_INVALID);
+
+       if (spec_mae->outer_rule != NULL)
+               sfc_mae_outer_rule_del(sa, spec_mae->outer_rule);
+
+       if (spec_mae->action_set != NULL)
+               sfc_mae_action_set_del(sa, spec_mae->action_set);
+
+       if (spec_mae->match_spec != NULL)
+               efx_mae_match_spec_fini(sa->nic, spec_mae->match_spec);
+}
+
+static int
+sfc_mae_set_ethertypes(struct sfc_mae_parse_ctx *ctx)
+{
+       struct sfc_mae_pattern_data *pdata = &ctx->pattern_data;
+       const efx_mae_field_id_t *fremap = ctx->field_ids_remap;
+       const efx_mae_field_id_t field_ids[] = {
+               EFX_MAE_FIELD_VLAN0_PROTO_BE,
+               EFX_MAE_FIELD_VLAN1_PROTO_BE,
+       };
+       const struct sfc_mae_ethertype *et;
+       unsigned int i;
+       int rc;
+
+       /*
+        * In accordance with RTE flow API convention, the innermost L2
+        * item's "type" ("inner_type") is a L3 EtherType. If there is
+        * no L3 item, it's 0x0000/0x0000.
+        */
+       et = &pdata->ethertypes[pdata->nb_vlan_tags];
+       rc = efx_mae_match_spec_field_set(ctx->match_spec,
+                                         fremap[EFX_MAE_FIELD_ETHER_TYPE_BE],
+                                         sizeof(et->value),
+                                         (const uint8_t *)&et->value,
+                                         sizeof(et->mask),
+                                         (const uint8_t *)&et->mask);
+       if (rc != 0)
+               return rc;
+
+       /*
+        * sfc_mae_rule_parse_item_vlan() has already made sure
+        * that pdata->nb_vlan_tags does not exceed this figure.
+        */
+       RTE_BUILD_BUG_ON(SFC_MAE_MATCH_VLAN_MAX_NTAGS != 2);
+
+       for (i = 0; i < pdata->nb_vlan_tags; ++i) {
+               et = &pdata->ethertypes[i];
+
+               rc = efx_mae_match_spec_field_set(ctx->match_spec,
+                                                 fremap[field_ids[i]],
+                                                 sizeof(et->value),
+                                                 (const uint8_t *)&et->value,
+                                                 sizeof(et->mask),
+                                                 (const uint8_t *)&et->mask);
+               if (rc != 0)
+                       return rc;
+       }
+
+       return 0;
+}
+
+static int
+sfc_mae_rule_process_pattern_data(struct sfc_mae_parse_ctx *ctx,
+                                 struct rte_flow_error *error)
+{
+       const efx_mae_field_id_t *fremap = ctx->field_ids_remap;
+       struct sfc_mae_pattern_data *pdata = &ctx->pattern_data;
+       struct sfc_mae_ethertype *ethertypes = pdata->ethertypes;
+       const rte_be16_t supported_tpids[] = {
+               /* VLAN standard TPID (always the first element) */
+               RTE_BE16(RTE_ETHER_TYPE_VLAN),
+
+               /* Double-tagging TPIDs */
+               RTE_BE16(RTE_ETHER_TYPE_QINQ),
+               RTE_BE16(RTE_ETHER_TYPE_QINQ1),
+               RTE_BE16(RTE_ETHER_TYPE_QINQ2),
+               RTE_BE16(RTE_ETHER_TYPE_QINQ3),
+       };
+       unsigned int nb_supported_tpids = RTE_DIM(supported_tpids);
+       unsigned int ethertype_idx;
+       const uint8_t *valuep;
+       const uint8_t *maskp;
+       int rc;
+
+       if (pdata->innermost_ethertype_restriction.mask != 0 &&
+           pdata->nb_vlan_tags < SFC_MAE_MATCH_VLAN_MAX_NTAGS) {
+               /*
+                * If a single item VLAN is followed by a L3 item, value
+                * of "type" in item ETH can't be a double-tagging TPID.
+                */
+               nb_supported_tpids = 1;
+       }
+
+       /*
+        * sfc_mae_rule_parse_item_vlan() has already made sure
+        * that pdata->nb_vlan_tags does not exceed this figure.
+        */
+       RTE_BUILD_BUG_ON(SFC_MAE_MATCH_VLAN_MAX_NTAGS != 2);
+
+       for (ethertype_idx = 0;
+            ethertype_idx < pdata->nb_vlan_tags; ++ethertype_idx) {
+               unsigned int tpid_idx;
+
+               /* Exact match is supported only. */
+               if (ethertypes[ethertype_idx].mask != RTE_BE16(0xffff)) {
+                       rc = EINVAL;
+                       goto fail;
+               }
+
+               for (tpid_idx = pdata->nb_vlan_tags - ethertype_idx - 1;
+                    tpid_idx < nb_supported_tpids; ++tpid_idx) {
+                       if (ethertypes[ethertype_idx].value ==
+                           supported_tpids[tpid_idx])
+                               break;
+               }
+
+               if (tpid_idx == nb_supported_tpids) {
+                       rc = EINVAL;
+                       goto fail;
+               }
+
+               nb_supported_tpids = 1;
+       }
+
+       if (pdata->innermost_ethertype_restriction.mask == RTE_BE16(0xffff)) {
+               struct sfc_mae_ethertype *et = &ethertypes[ethertype_idx];
+
+               if (et->mask == 0) {
+                       et->mask = RTE_BE16(0xffff);
+                       et->value =
+                           pdata->innermost_ethertype_restriction.value;
+               } else if (et->mask != RTE_BE16(0xffff) ||
+                          et->value !=
+                          pdata->innermost_ethertype_restriction.value) {
+                       rc = EINVAL;
+                       goto fail;
+               }
+       }
+
+       /*
+        * Now, when the number of VLAN tags is known, set fields
+        * ETHER_TYPE, VLAN0_PROTO and VLAN1_PROTO so that the first
+        * one is either a valid L3 EtherType (or 0x0000/0x0000),
+        * and the last two are valid TPIDs (or 0x0000/0x0000).
+        */
+       rc = sfc_mae_set_ethertypes(ctx);
+       if (rc != 0)
+               goto fail;
+
+       if (pdata->l3_next_proto_restriction_mask == 0xff) {
+               if (pdata->l3_next_proto_mask == 0) {
+                       pdata->l3_next_proto_mask = 0xff;
+                       pdata->l3_next_proto_value =
+                           pdata->l3_next_proto_restriction_value;
+               } else if (pdata->l3_next_proto_mask != 0xff ||
+                          pdata->l3_next_proto_value !=
+                          pdata->l3_next_proto_restriction_value) {
+                       rc = EINVAL;
+                       goto fail;
+               }
+       }
+
+       valuep = (const uint8_t *)&pdata->l3_next_proto_value;
+       maskp = (const uint8_t *)&pdata->l3_next_proto_mask;
+       rc = efx_mae_match_spec_field_set(ctx->match_spec,
+                                         fremap[EFX_MAE_FIELD_IP_PROTO],
+                                         sizeof(pdata->l3_next_proto_value),
+                                         valuep,
+                                         sizeof(pdata->l3_next_proto_mask),
+                                         maskp);
+       if (rc != 0)
+               goto fail;
+
+       return 0;
+
+fail:
+       return rte_flow_error_set(error, rc, RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+                                 "Failed to process pattern data");
+}
+
+static int
+sfc_mae_rule_parse_item_port_id(const struct rte_flow_item *item,
+                               struct sfc_flow_parse_ctx *ctx,
+                               struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       const struct rte_flow_item_port_id supp_mask = {
+               .id = 0xffffffff,
+       };
+       const void *def_mask = &rte_flow_item_port_id_mask;
+       const struct rte_flow_item_port_id *spec = NULL;
+       const struct rte_flow_item_port_id *mask = NULL;
+       efx_mport_sel_t mport_sel;
+       int rc;
+
+       if (ctx_mae->match_mport_set) {
+               return rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Can't handle multiple traffic source items");
+       }
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask, def_mask,
+                                sizeof(struct rte_flow_item_port_id), error);
+       if (rc != 0)
+               return rc;
+
+       if (mask->id != supp_mask.id) {
+               return rte_flow_error_set(error, EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Bad mask in the PORT_ID pattern item");
+       }
+
+       /* If "spec" is not set, could be any port ID */
+       if (spec == NULL)
+               return 0;
+
+       if (spec->id > UINT16_MAX) {
+               return rte_flow_error_set(error, EOVERFLOW,
+                                         RTE_FLOW_ERROR_TYPE_ITEM, item,
+                                         "The port ID is too large");
+       }
+
+       rc = sfc_mae_switch_port_by_ethdev(ctx_mae->sa->mae.switch_domain_id,
+                                          spec->id, &mport_sel);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Can't find RTE ethdev by the port ID");
+       }
+
+       rc = efx_mae_match_spec_mport_set(ctx_mae->match_spec,
+                                         &mport_sel, NULL);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Failed to set MPORT for the port ID");
+       }
+
+       ctx_mae->match_mport_set = B_TRUE;
+
+       return 0;
+}
+
+static int
+sfc_mae_rule_parse_item_phy_port(const struct rte_flow_item *item,
+                                struct sfc_flow_parse_ctx *ctx,
+                                struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       const struct rte_flow_item_phy_port supp_mask = {
+               .index = 0xffffffff,
+       };
+       const void *def_mask = &rte_flow_item_phy_port_mask;
+       const struct rte_flow_item_phy_port *spec = NULL;
+       const struct rte_flow_item_phy_port *mask = NULL;
+       efx_mport_sel_t mport_v;
+       int rc;
+
+       if (ctx_mae->match_mport_set) {
+               return rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Can't handle multiple traffic source items");
+       }
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask, def_mask,
+                                sizeof(struct rte_flow_item_phy_port), error);
+       if (rc != 0)
+               return rc;
+
+       if (mask->index != supp_mask.index) {
+               return rte_flow_error_set(error, EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Bad mask in the PHY_PORT pattern item");
+       }
+
+       /* If "spec" is not set, could be any physical port */
+       if (spec == NULL)
+               return 0;
+
+       rc = efx_mae_mport_by_phy_port(spec->index, &mport_v);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Failed to convert the PHY_PORT index");
+       }
+
+       rc = efx_mae_match_spec_mport_set(ctx_mae->match_spec, &mport_v, NULL);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Failed to set MPORT for the PHY_PORT");
+       }
+
+       ctx_mae->match_mport_set = B_TRUE;
+
+       return 0;
+}
+
+static int
+sfc_mae_rule_parse_item_pf(const struct rte_flow_item *item,
+                          struct sfc_flow_parse_ctx *ctx,
+                          struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       const efx_nic_cfg_t *encp = efx_nic_cfg_get(ctx_mae->sa->nic);
+       efx_mport_sel_t mport_v;
+       int rc;
+
+       if (ctx_mae->match_mport_set) {
+               return rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Can't handle multiple traffic source items");
+       }
+
+       rc = efx_mae_mport_by_pcie_function(encp->enc_pf, EFX_PCI_VF_INVALID,
+                                           &mport_v);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Failed to convert the PF ID");
+       }
+
+       rc = efx_mae_match_spec_mport_set(ctx_mae->match_spec, &mport_v, NULL);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Failed to set MPORT for the PF");
+       }
+
+       ctx_mae->match_mport_set = B_TRUE;
+
+       return 0;
+}
+
+static int
+sfc_mae_rule_parse_item_vf(const struct rte_flow_item *item,
+                          struct sfc_flow_parse_ctx *ctx,
+                          struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       const efx_nic_cfg_t *encp = efx_nic_cfg_get(ctx_mae->sa->nic);
+       const struct rte_flow_item_vf supp_mask = {
+               .id = 0xffffffff,
+       };
+       const void *def_mask = &rte_flow_item_vf_mask;
+       const struct rte_flow_item_vf *spec = NULL;
+       const struct rte_flow_item_vf *mask = NULL;
+       efx_mport_sel_t mport_v;
+       int rc;
+
+       if (ctx_mae->match_mport_set) {
+               return rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Can't handle multiple traffic source items");
+       }
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask, def_mask,
+                                sizeof(struct rte_flow_item_vf), error);
+       if (rc != 0)
+               return rc;
+
+       if (mask->id != supp_mask.id) {
+               return rte_flow_error_set(error, EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Bad mask in the VF pattern item");
+       }
+
+       /*
+        * If "spec" is not set, the item requests any VF related to the
+        * PF of the current DPDK port (but not the PF itself).
+        * Reject this match criterion as unsupported.
+        */
+       if (spec == NULL) {
+               return rte_flow_error_set(error, EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Bad spec in the VF pattern item");
+       }
+
+       rc = efx_mae_mport_by_pcie_function(encp->enc_pf, spec->id, &mport_v);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Failed to convert the PF + VF IDs");
+       }
+
+       rc = efx_mae_match_spec_mport_set(ctx_mae->match_spec, &mport_v, NULL);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Failed to set MPORT for the PF + VF");
+       }
+
+       ctx_mae->match_mport_set = B_TRUE;
+
+       return 0;
+}
+
+/*
+ * Having this field ID in a field locator means that this
+ * locator cannot be used to actually set the field at the
+ * time when the corresponding item gets encountered. Such
+ * fields get stashed in the parsing context instead. This
+ * is required to resolve dependencies between the stashed
+ * fields. See sfc_mae_rule_process_pattern_data().
+ */
+#define SFC_MAE_FIELD_HANDLING_DEFERRED        EFX_MAE_FIELD_NIDS
+
+struct sfc_mae_field_locator {
+       efx_mae_field_id_t              field_id;
+       size_t                          size;
+       /* Field offset in the corresponding rte_flow_item_ struct */
+       size_t                          ofst;
+};
+
+static void
+sfc_mae_item_build_supp_mask(const struct sfc_mae_field_locator *field_locators,
+                            unsigned int nb_field_locators, void *mask_ptr,
+                            size_t mask_size)
+{
+       unsigned int i;
+
+       memset(mask_ptr, 0, mask_size);
+
+       for (i = 0; i < nb_field_locators; ++i) {
+               const struct sfc_mae_field_locator *fl = &field_locators[i];
+
+               SFC_ASSERT(fl->ofst + fl->size <= mask_size);
+               memset(RTE_PTR_ADD(mask_ptr, fl->ofst), 0xff, fl->size);
+       }
+}
+
+static int
+sfc_mae_parse_item(const struct sfc_mae_field_locator *field_locators,
+                  unsigned int nb_field_locators, const uint8_t *spec,
+                  const uint8_t *mask, struct sfc_mae_parse_ctx *ctx,
+                  struct rte_flow_error *error)
+{
+       const efx_mae_field_id_t *fremap = ctx->field_ids_remap;
+       unsigned int i;
+       int rc = 0;
+
+       for (i = 0; i < nb_field_locators; ++i) {
+               const struct sfc_mae_field_locator *fl = &field_locators[i];
+
+               if (fl->field_id == SFC_MAE_FIELD_HANDLING_DEFERRED)
+                       continue;
+
+               rc = efx_mae_match_spec_field_set(ctx->match_spec,
+                                                 fremap[fl->field_id],
+                                                 fl->size, spec + fl->ofst,
+                                                 fl->size, mask + fl->ofst);
+               if (rc != 0)
+                       break;
+       }
+
+       if (rc != 0) {
+               rc = rte_flow_error_set(error, rc, RTE_FLOW_ERROR_TYPE_ITEM,
+                               NULL, "Failed to process item fields");
+       }
+
+       return rc;
+}
+
+static const struct sfc_mae_field_locator flocs_eth[] = {
+       {
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_eth, type),
+               offsetof(struct rte_flow_item_eth, type),
+       },
+       {
+               EFX_MAE_FIELD_ETH_DADDR_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_eth, dst),
+               offsetof(struct rte_flow_item_eth, dst),
+       },
+       {
+               EFX_MAE_FIELD_ETH_SADDR_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_eth, src),
+               offsetof(struct rte_flow_item_eth, src),
+       },
+};
+
+static int
+sfc_mae_rule_parse_item_eth(const struct rte_flow_item *item,
+                           struct sfc_flow_parse_ctx *ctx,
+                           struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       struct rte_flow_item_eth supp_mask;
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       int rc;
+
+       sfc_mae_item_build_supp_mask(flocs_eth, RTE_DIM(flocs_eth),
+                                    &supp_mask, sizeof(supp_mask));
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                &rte_flow_item_eth_mask,
+                                sizeof(struct rte_flow_item_eth), error);
+       if (rc != 0)
+               return rc;
+
+       if (spec != NULL) {
+               struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+               struct sfc_mae_ethertype *ethertypes = pdata->ethertypes;
+               const struct rte_flow_item_eth *item_spec;
+               const struct rte_flow_item_eth *item_mask;
+
+               item_spec = (const struct rte_flow_item_eth *)spec;
+               item_mask = (const struct rte_flow_item_eth *)mask;
+
+               ethertypes[0].value = item_spec->type;
+               ethertypes[0].mask = item_mask->type;
+       } else {
+               /*
+                * The specification is empty. This is wrong in the case
+                * when there are more network patterns in line. Other
+                * than that, any Ethernet can match. All of that is
+                * checked at the end of parsing.
+                */
+               return 0;
+       }
+
+       return sfc_mae_parse_item(flocs_eth, RTE_DIM(flocs_eth), spec, mask,
+                                 ctx_mae, error);
+}
+
+static const struct sfc_mae_field_locator flocs_vlan[] = {
+       /* Outermost tag */
+       {
+               EFX_MAE_FIELD_VLAN0_TCI_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, tci),
+               offsetof(struct rte_flow_item_vlan, tci),
+       },
+       {
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, inner_type),
+               offsetof(struct rte_flow_item_vlan, inner_type),
+       },
+
+       /* Innermost tag */
+       {
+               EFX_MAE_FIELD_VLAN1_TCI_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, tci),
+               offsetof(struct rte_flow_item_vlan, tci),
+       },
+       {
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_vlan, inner_type),
+               offsetof(struct rte_flow_item_vlan, inner_type),
+       },
+};
+
+static int
+sfc_mae_rule_parse_item_vlan(const struct rte_flow_item *item,
+                            struct sfc_flow_parse_ctx *ctx,
+                            struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+       const struct sfc_mae_field_locator *flocs;
+       struct rte_flow_item_vlan supp_mask;
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       unsigned int nb_flocs;
+       int rc;
+
+       RTE_BUILD_BUG_ON(SFC_MAE_MATCH_VLAN_MAX_NTAGS != 2);
+
+       if (pdata->nb_vlan_tags == SFC_MAE_MATCH_VLAN_MAX_NTAGS) {
+               return rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ITEM, item,
+                               "Can't match that many VLAN tags");
+       }
+
+       nb_flocs = RTE_DIM(flocs_vlan) / SFC_MAE_MATCH_VLAN_MAX_NTAGS;
+       flocs = flocs_vlan + pdata->nb_vlan_tags * nb_flocs;
+
+       /* If parsing fails, this can remain incremented. */
+       ++pdata->nb_vlan_tags;
+
+       sfc_mae_item_build_supp_mask(flocs, nb_flocs,
+                                    &supp_mask, sizeof(supp_mask));
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                &rte_flow_item_vlan_mask,
+                                sizeof(struct rte_flow_item_vlan), error);
+       if (rc != 0)
+               return rc;
+
+       if (spec != NULL) {
+               struct sfc_mae_ethertype *ethertypes = pdata->ethertypes;
+               const struct rte_flow_item_vlan *item_spec;
+               const struct rte_flow_item_vlan *item_mask;
+
+               item_spec = (const struct rte_flow_item_vlan *)spec;
+               item_mask = (const struct rte_flow_item_vlan *)mask;
+
+               ethertypes[pdata->nb_vlan_tags].value = item_spec->inner_type;
+               ethertypes[pdata->nb_vlan_tags].mask = item_mask->inner_type;
+       } else {
+               /*
+                * The specification is empty. This is wrong in the case
+                * when there are more network patterns in line. Other
+                * than that, any Ethernet can match. All of that is
+                * checked at the end of parsing.
+                */
+               return 0;
+       }
+
+       return sfc_mae_parse_item(flocs, nb_flocs, spec, mask, ctx_mae, error);
+}
+
+static const struct sfc_mae_field_locator flocs_ipv4[] = {
+       {
+               EFX_MAE_FIELD_SRC_IP4_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv4, hdr.src_addr),
+               offsetof(struct rte_flow_item_ipv4, hdr.src_addr),
+       },
+       {
+               EFX_MAE_FIELD_DST_IP4_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv4, hdr.dst_addr),
+               offsetof(struct rte_flow_item_ipv4, hdr.dst_addr),
+       },
+       {
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv4, hdr.next_proto_id),
+               offsetof(struct rte_flow_item_ipv4, hdr.next_proto_id),
+       },
+       {
+               EFX_MAE_FIELD_IP_TOS,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv4,
+                                hdr.type_of_service),
+               offsetof(struct rte_flow_item_ipv4, hdr.type_of_service),
+       },
+       {
+               EFX_MAE_FIELD_IP_TTL,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv4, hdr.time_to_live),
+               offsetof(struct rte_flow_item_ipv4, hdr.time_to_live),
+       },
+};
+
+static int
+sfc_mae_rule_parse_item_ipv4(const struct rte_flow_item *item,
+                            struct sfc_flow_parse_ctx *ctx,
+                            struct rte_flow_error *error)
+{
+       rte_be16_t ethertype_ipv4_be = RTE_BE16(RTE_ETHER_TYPE_IPV4);
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+       struct rte_flow_item_ipv4 supp_mask;
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       int rc;
+
+       sfc_mae_item_build_supp_mask(flocs_ipv4, RTE_DIM(flocs_ipv4),
+                                    &supp_mask, sizeof(supp_mask));
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                &rte_flow_item_ipv4_mask,
+                                sizeof(struct rte_flow_item_ipv4), error);
+       if (rc != 0)
+               return rc;
+
+       pdata->innermost_ethertype_restriction.value = ethertype_ipv4_be;
+       pdata->innermost_ethertype_restriction.mask = RTE_BE16(0xffff);
+
+       if (spec != NULL) {
+               const struct rte_flow_item_ipv4 *item_spec;
+               const struct rte_flow_item_ipv4 *item_mask;
+
+               item_spec = (const struct rte_flow_item_ipv4 *)spec;
+               item_mask = (const struct rte_flow_item_ipv4 *)mask;
+
+               pdata->l3_next_proto_value = item_spec->hdr.next_proto_id;
+               pdata->l3_next_proto_mask = item_mask->hdr.next_proto_id;
+       } else {
+               return 0;
+       }
+
+       return sfc_mae_parse_item(flocs_ipv4, RTE_DIM(flocs_ipv4), spec, mask,
+                                 ctx_mae, error);
+}
+
+static const struct sfc_mae_field_locator flocs_ipv6[] = {
+       {
+               EFX_MAE_FIELD_SRC_IP6_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv6, hdr.src_addr),
+               offsetof(struct rte_flow_item_ipv6, hdr.src_addr),
+       },
+       {
+               EFX_MAE_FIELD_DST_IP6_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv6, hdr.dst_addr),
+               offsetof(struct rte_flow_item_ipv6, hdr.dst_addr),
+       },
+       {
+               /*
+                * This locator is used only for building supported fields mask.
+                * The field is handled by sfc_mae_rule_process_pattern_data().
+                */
+               SFC_MAE_FIELD_HANDLING_DEFERRED,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv6, hdr.proto),
+               offsetof(struct rte_flow_item_ipv6, hdr.proto),
+       },
+       {
+               EFX_MAE_FIELD_IP_TTL,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_ipv6, hdr.hop_limits),
+               offsetof(struct rte_flow_item_ipv6, hdr.hop_limits),
+       },
+};
+
+static int
+sfc_mae_rule_parse_item_ipv6(const struct rte_flow_item *item,
+                            struct sfc_flow_parse_ctx *ctx,
+                            struct rte_flow_error *error)
+{
+       rte_be16_t ethertype_ipv6_be = RTE_BE16(RTE_ETHER_TYPE_IPV6);
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       const efx_mae_field_id_t *fremap = ctx_mae->field_ids_remap;
+       struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+       struct rte_flow_item_ipv6 supp_mask;
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       rte_be32_t vtc_flow_be;
+       uint32_t vtc_flow;
+       uint8_t tc_value;
+       uint8_t tc_mask;
+       int rc;
+
+       sfc_mae_item_build_supp_mask(flocs_ipv6, RTE_DIM(flocs_ipv6),
+                                    &supp_mask, sizeof(supp_mask));
+
+       vtc_flow_be = RTE_BE32(RTE_IPV6_HDR_TC_MASK);
+       memcpy(&supp_mask, &vtc_flow_be, sizeof(vtc_flow_be));
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                &rte_flow_item_ipv6_mask,
+                                sizeof(struct rte_flow_item_ipv6), error);
+       if (rc != 0)
+               return rc;
+
+       pdata->innermost_ethertype_restriction.value = ethertype_ipv6_be;
+       pdata->innermost_ethertype_restriction.mask = RTE_BE16(0xffff);
+
+       if (spec != NULL) {
+               const struct rte_flow_item_ipv6 *item_spec;
+               const struct rte_flow_item_ipv6 *item_mask;
+
+               item_spec = (const struct rte_flow_item_ipv6 *)spec;
+               item_mask = (const struct rte_flow_item_ipv6 *)mask;
+
+               pdata->l3_next_proto_value = item_spec->hdr.proto;
+               pdata->l3_next_proto_mask = item_mask->hdr.proto;
+       } else {
+               return 0;
+       }
+
+       rc = sfc_mae_parse_item(flocs_ipv6, RTE_DIM(flocs_ipv6), spec, mask,
+                               ctx_mae, error);
+       if (rc != 0)
+               return rc;
+
+       memcpy(&vtc_flow_be, spec, sizeof(vtc_flow_be));
+       vtc_flow = rte_be_to_cpu_32(vtc_flow_be);
+       tc_value = (vtc_flow & RTE_IPV6_HDR_TC_MASK) >> RTE_IPV6_HDR_TC_SHIFT;
+
+       memcpy(&vtc_flow_be, mask, sizeof(vtc_flow_be));
+       vtc_flow = rte_be_to_cpu_32(vtc_flow_be);
+       tc_mask = (vtc_flow & RTE_IPV6_HDR_TC_MASK) >> RTE_IPV6_HDR_TC_SHIFT;
+
+       rc = efx_mae_match_spec_field_set(ctx_mae->match_spec,
+                                         fremap[EFX_MAE_FIELD_IP_TOS],
+                                         sizeof(tc_value), &tc_value,
+                                         sizeof(tc_mask), &tc_mask);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc, RTE_FLOW_ERROR_TYPE_ITEM,
+                               NULL, "Failed to process item fields");
+       }
+
+       return 0;
+}
+
+static const struct sfc_mae_field_locator flocs_tcp[] = {
+       {
+               EFX_MAE_FIELD_L4_SPORT_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_tcp, hdr.src_port),
+               offsetof(struct rte_flow_item_tcp, hdr.src_port),
+       },
+       {
+               EFX_MAE_FIELD_L4_DPORT_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_tcp, hdr.dst_port),
+               offsetof(struct rte_flow_item_tcp, hdr.dst_port),
+       },
+       {
+               EFX_MAE_FIELD_TCP_FLAGS_BE,
+               /*
+                * The values have been picked intentionally since the
+                * target MAE field is oversize (16 bit). This mapping
+                * relies on the fact that the MAE field is big-endian.
+                */
+               RTE_SIZEOF_FIELD(struct rte_flow_item_tcp, hdr.data_off) +
+               RTE_SIZEOF_FIELD(struct rte_flow_item_tcp, hdr.tcp_flags),
+               offsetof(struct rte_flow_item_tcp, hdr.data_off),
+       },
+};
+
+static int
+sfc_mae_rule_parse_item_tcp(const struct rte_flow_item *item,
+                           struct sfc_flow_parse_ctx *ctx,
+                           struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+       struct rte_flow_item_tcp supp_mask;
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       int rc;
+
+       /*
+        * When encountered among outermost items, item TCP is invalid.
+        * Check which match specification is being constructed now.
+        */
+       if (ctx_mae->match_spec != ctx_mae->match_spec_action) {
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ITEM, item,
+                                         "TCP in outer frame is invalid");
+       }
+
+       sfc_mae_item_build_supp_mask(flocs_tcp, RTE_DIM(flocs_tcp),
+                                    &supp_mask, sizeof(supp_mask));
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                &rte_flow_item_tcp_mask,
+                                sizeof(struct rte_flow_item_tcp), error);
+       if (rc != 0)
+               return rc;
+
+       pdata->l3_next_proto_restriction_value = IPPROTO_TCP;
+       pdata->l3_next_proto_restriction_mask = 0xff;
+
+       if (spec == NULL)
+               return 0;
+
+       return sfc_mae_parse_item(flocs_tcp, RTE_DIM(flocs_tcp), spec, mask,
+                                 ctx_mae, error);
+}
+
+static const struct sfc_mae_field_locator flocs_udp[] = {
+       {
+               EFX_MAE_FIELD_L4_SPORT_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_udp, hdr.src_port),
+               offsetof(struct rte_flow_item_udp, hdr.src_port),
+       },
+       {
+               EFX_MAE_FIELD_L4_DPORT_BE,
+               RTE_SIZEOF_FIELD(struct rte_flow_item_udp, hdr.dst_port),
+               offsetof(struct rte_flow_item_udp, hdr.dst_port),
+       },
+};
+
+static int
+sfc_mae_rule_parse_item_udp(const struct rte_flow_item *item,
+                           struct sfc_flow_parse_ctx *ctx,
+                           struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       struct sfc_mae_pattern_data *pdata = &ctx_mae->pattern_data;
+       struct rte_flow_item_udp supp_mask;
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       int rc;
+
+       sfc_mae_item_build_supp_mask(flocs_udp, RTE_DIM(flocs_udp),
+                                    &supp_mask, sizeof(supp_mask));
+
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                &rte_flow_item_udp_mask,
+                                sizeof(struct rte_flow_item_udp), error);
+       if (rc != 0)
+               return rc;
+
+       pdata->l3_next_proto_restriction_value = IPPROTO_UDP;
+       pdata->l3_next_proto_restriction_mask = 0xff;
+
+       if (spec == NULL)
+               return 0;
+
+       return sfc_mae_parse_item(flocs_udp, RTE_DIM(flocs_udp), spec, mask,
+                                 ctx_mae, error);
+}
+
+static const struct sfc_mae_field_locator flocs_tunnel[] = {
+       {
+               /*
+                * The size and offset values are relevant
+                * for Geneve and NVGRE, too.
+                */
+               .size = RTE_SIZEOF_FIELD(struct rte_flow_item_vxlan, vni),
+               .ofst = offsetof(struct rte_flow_item_vxlan, vni),
+       },
+};
+
+/*
+ * An auxiliary registry which allows using non-encap. field IDs
+ * directly when building a match specification of type ACTION.
+ *
+ * See sfc_mae_rule_parse_pattern() and sfc_mae_rule_parse_item_tunnel().
+ */
+static const efx_mae_field_id_t field_ids_no_remap[] = {
+#define FIELD_ID_NO_REMAP(_field) \
+       [EFX_MAE_FIELD_##_field] = EFX_MAE_FIELD_##_field
+
+       FIELD_ID_NO_REMAP(ETHER_TYPE_BE),
+       FIELD_ID_NO_REMAP(ETH_SADDR_BE),
+       FIELD_ID_NO_REMAP(ETH_DADDR_BE),
+       FIELD_ID_NO_REMAP(VLAN0_TCI_BE),
+       FIELD_ID_NO_REMAP(VLAN0_PROTO_BE),
+       FIELD_ID_NO_REMAP(VLAN1_TCI_BE),
+       FIELD_ID_NO_REMAP(VLAN1_PROTO_BE),
+       FIELD_ID_NO_REMAP(SRC_IP4_BE),
+       FIELD_ID_NO_REMAP(DST_IP4_BE),
+       FIELD_ID_NO_REMAP(IP_PROTO),
+       FIELD_ID_NO_REMAP(IP_TOS),
+       FIELD_ID_NO_REMAP(IP_TTL),
+       FIELD_ID_NO_REMAP(SRC_IP6_BE),
+       FIELD_ID_NO_REMAP(DST_IP6_BE),
+       FIELD_ID_NO_REMAP(L4_SPORT_BE),
+       FIELD_ID_NO_REMAP(L4_DPORT_BE),
+       FIELD_ID_NO_REMAP(TCP_FLAGS_BE),
+
+#undef FIELD_ID_NO_REMAP
+};
+
+/*
+ * An auxiliary registry which allows using "ENC" field IDs
+ * when building a match specification of type OUTER.
+ *
+ * See sfc_mae_rule_encap_parse_init().
+ */
+static const efx_mae_field_id_t field_ids_remap_to_encap[] = {
+#define FIELD_ID_REMAP_TO_ENCAP(_field) \
+       [EFX_MAE_FIELD_##_field] = EFX_MAE_FIELD_ENC_##_field
+
+       FIELD_ID_REMAP_TO_ENCAP(ETHER_TYPE_BE),
+       FIELD_ID_REMAP_TO_ENCAP(ETH_SADDR_BE),
+       FIELD_ID_REMAP_TO_ENCAP(ETH_DADDR_BE),
+       FIELD_ID_REMAP_TO_ENCAP(VLAN0_TCI_BE),
+       FIELD_ID_REMAP_TO_ENCAP(VLAN0_PROTO_BE),
+       FIELD_ID_REMAP_TO_ENCAP(VLAN1_TCI_BE),
+       FIELD_ID_REMAP_TO_ENCAP(VLAN1_PROTO_BE),
+       FIELD_ID_REMAP_TO_ENCAP(SRC_IP4_BE),
+       FIELD_ID_REMAP_TO_ENCAP(DST_IP4_BE),
+       FIELD_ID_REMAP_TO_ENCAP(IP_PROTO),
+       FIELD_ID_REMAP_TO_ENCAP(IP_TOS),
+       FIELD_ID_REMAP_TO_ENCAP(IP_TTL),
+       FIELD_ID_REMAP_TO_ENCAP(SRC_IP6_BE),
+       FIELD_ID_REMAP_TO_ENCAP(DST_IP6_BE),
+       FIELD_ID_REMAP_TO_ENCAP(L4_SPORT_BE),
+       FIELD_ID_REMAP_TO_ENCAP(L4_DPORT_BE),
+
+#undef FIELD_ID_REMAP_TO_ENCAP
+};
+
+static int
+sfc_mae_rule_parse_item_tunnel(const struct rte_flow_item *item,
+                              struct sfc_flow_parse_ctx *ctx,
+                              struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx *ctx_mae = ctx->mae;
+       uint8_t vnet_id_v[sizeof(uint32_t)] = {0};
+       uint8_t vnet_id_m[sizeof(uint32_t)] = {0};
+       const struct rte_flow_item_vxlan *vxp;
+       uint8_t supp_mask[sizeof(uint64_t)];
+       const uint8_t *spec = NULL;
+       const uint8_t *mask = NULL;
+       int rc;
+
+       /*
+        * We're about to start processing inner frame items.
+        * Process pattern data that has been deferred so far
+        * and reset pattern data storage.
+        */
+       rc = sfc_mae_rule_process_pattern_data(ctx_mae, error);
+       if (rc != 0)
+               return rc;
+
+       memset(&ctx_mae->pattern_data, 0, sizeof(ctx_mae->pattern_data));
+
+       sfc_mae_item_build_supp_mask(flocs_tunnel, RTE_DIM(flocs_tunnel),
+                                    &supp_mask, sizeof(supp_mask));
+
+       /*
+        * This tunnel item was preliminarily detected by
+        * sfc_mae_rule_encap_parse_init(). Default mask
+        * was also picked by that helper. Use it here.
+        */
+       rc = sfc_flow_parse_init(item,
+                                (const void **)&spec, (const void **)&mask,
+                                (const void *)&supp_mask,
+                                ctx_mae->tunnel_def_mask,
+                                ctx_mae->tunnel_def_mask_size,  error);
+       if (rc != 0)
+               return rc;
+
+       /*
+        * This item and later ones comprise a
+        * match specification of type ACTION.
+        */
+       ctx_mae->match_spec = ctx_mae->match_spec_action;
+
+       /* This item and later ones use non-encap. EFX MAE field IDs. */
+       ctx_mae->field_ids_remap = field_ids_no_remap;
+
+       if (spec == NULL)
+               return 0;
+
+       /*
+        * Field EFX_MAE_FIELD_ENC_VNET_ID_BE is a 32-bit one.
+        * Copy 24-bit VNI, which is BE, at offset 1 in it.
+        * The extra byte is 0 both in the mask and in the value.
+        */
+       vxp = (const struct rte_flow_item_vxlan *)spec;
+       memcpy(vnet_id_v + 1, &vxp->vni, sizeof(vxp->vni));
+
+       vxp = (const struct rte_flow_item_vxlan *)mask;
+       memcpy(vnet_id_m + 1, &vxp->vni, sizeof(vxp->vni));
+
+       rc = efx_mae_match_spec_field_set(ctx_mae->match_spec,
+                                         EFX_MAE_FIELD_ENC_VNET_ID_BE,
+                                         sizeof(vnet_id_v), vnet_id_v,
+                                         sizeof(vnet_id_m), vnet_id_m);
+       if (rc != 0) {
+               rc = rte_flow_error_set(error, rc, RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item, "Failed to set VXLAN VNI");
+       }
+
+       return rc;
+}
+
+static const struct sfc_flow_item sfc_flow_items[] = {
+       {
+               .type = RTE_FLOW_ITEM_TYPE_PORT_ID,
+               /*
+                * In terms of RTE flow, this item is a META one,
+                * and its position in the pattern is don't care.
+                */
+               .prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_port_id,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_PHY_PORT,
+               /*
+                * In terms of RTE flow, this item is a META one,
+                * and its position in the pattern is don't care.
+                */
+               .prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_phy_port,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_PF,
+               /*
+                * In terms of RTE flow, this item is a META one,
+                * and its position in the pattern is don't care.
+                */
+               .prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_pf,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_VF,
+               /*
+                * In terms of RTE flow, this item is a META one,
+                * and its position in the pattern is don't care.
+                */
+               .prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .layer = SFC_FLOW_ITEM_ANY_LAYER,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_vf,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_ETH,
+               .prev_layer = SFC_FLOW_ITEM_START_LAYER,
+               .layer = SFC_FLOW_ITEM_L2,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_eth,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_VLAN,
+               .prev_layer = SFC_FLOW_ITEM_L2,
+               .layer = SFC_FLOW_ITEM_L2,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_vlan,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_IPV4,
+               .prev_layer = SFC_FLOW_ITEM_L2,
+               .layer = SFC_FLOW_ITEM_L3,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_ipv4,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_IPV6,
+               .prev_layer = SFC_FLOW_ITEM_L2,
+               .layer = SFC_FLOW_ITEM_L3,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_ipv6,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_TCP,
+               .prev_layer = SFC_FLOW_ITEM_L3,
+               .layer = SFC_FLOW_ITEM_L4,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_tcp,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_UDP,
+               .prev_layer = SFC_FLOW_ITEM_L3,
+               .layer = SFC_FLOW_ITEM_L4,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_udp,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_VXLAN,
+               .prev_layer = SFC_FLOW_ITEM_L4,
+               .layer = SFC_FLOW_ITEM_START_LAYER,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_tunnel,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_GENEVE,
+               .prev_layer = SFC_FLOW_ITEM_L4,
+               .layer = SFC_FLOW_ITEM_START_LAYER,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_tunnel,
+       },
+       {
+               .type = RTE_FLOW_ITEM_TYPE_NVGRE,
+               .prev_layer = SFC_FLOW_ITEM_L3,
+               .layer = SFC_FLOW_ITEM_START_LAYER,
+               .ctx_type = SFC_FLOW_PARSE_CTX_MAE,
+               .parse = sfc_mae_rule_parse_item_tunnel,
+       },
+};
+
+static int
+sfc_mae_rule_process_outer(struct sfc_adapter *sa,
+                          struct sfc_mae_parse_ctx *ctx,
+                          struct sfc_mae_outer_rule **rulep,
+                          struct rte_flow_error *error)
+{
+       struct sfc_mae_outer_rule *rule;
+       int rc;
+
+       if (ctx->encap_type == EFX_TUNNEL_PROTOCOL_NONE) {
+               *rulep = NULL;
+               return 0;
+       }
+
+       SFC_ASSERT(ctx->match_spec_outer != NULL);
+
+       if (!efx_mae_match_spec_is_valid(sa->nic, ctx->match_spec_outer)) {
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+                                         "Inconsistent pattern (outer)");
+       }
+
+       *rulep = sfc_mae_outer_rule_attach(sa, ctx->match_spec_outer,
+                                          ctx->encap_type);
+       if (*rulep != NULL) {
+               efx_mae_match_spec_fini(sa->nic, ctx->match_spec_outer);
+       } else {
+               rc = sfc_mae_outer_rule_add(sa, ctx->match_spec_outer,
+                                           ctx->encap_type, rulep);
+               if (rc != 0) {
+                       return rte_flow_error_set(error, rc,
+                                       RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+                                       "Failed to process the pattern");
+               }
+       }
+
+       /* The spec has now been tracked by the outer rule entry. */
+       ctx->match_spec_outer = NULL;
+
+       /*
+        * Depending on whether we reuse an existing outer rule or create a
+        * new one (see above), outer rule ID is either a valid value or
+        * EFX_MAE_RSRC_ID_INVALID. Set it in the action rule match
+        * specification (and the full mask, too) in order to have correct
+        * class comparisons of the new rule with existing ones.
+        * Also, action rule match specification will be validated shortly,
+        * and having the full mask set for outer rule ID indicates that we
+        * will use this field, and support for this field has to be checked.
+        */
+       rule = *rulep;
+       rc = efx_mae_match_spec_outer_rule_id_set(ctx->match_spec_action,
+                                                 &rule->fw_rsrc.rule_id);
+       if (rc != 0) {
+               sfc_mae_outer_rule_del(sa, *rulep);
+               *rulep = NULL;
+
+               return rte_flow_error_set(error, rc,
+                                         RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+                                         "Failed to process the pattern");
+       }
+
+       return 0;
+}
+
+static int
+sfc_mae_rule_encap_parse_init(struct sfc_adapter *sa,
+                             const struct rte_flow_item pattern[],
+                             struct sfc_mae_parse_ctx *ctx,
+                             struct rte_flow_error *error)
+{
+       struct sfc_mae *mae = &sa->mae;
+       int rc;
+
+       if (pattern == NULL) {
+               rte_flow_error_set(error, EINVAL,
+                                  RTE_FLOW_ERROR_TYPE_ITEM_NUM, NULL,
+                                  "NULL pattern");
+               return -rte_errno;
+       }
+
+       for (;;) {
+               switch (pattern->type) {
+               case RTE_FLOW_ITEM_TYPE_VXLAN:
+                       ctx->encap_type = EFX_TUNNEL_PROTOCOL_VXLAN;
+                       ctx->tunnel_def_mask = &rte_flow_item_vxlan_mask;
+                       ctx->tunnel_def_mask_size =
+                               sizeof(rte_flow_item_vxlan_mask);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_GENEVE:
+                       ctx->encap_type = EFX_TUNNEL_PROTOCOL_GENEVE;
+                       ctx->tunnel_def_mask = &rte_flow_item_geneve_mask;
+                       ctx->tunnel_def_mask_size =
+                               sizeof(rte_flow_item_geneve_mask);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_NVGRE:
+                       ctx->encap_type = EFX_TUNNEL_PROTOCOL_NVGRE;
+                       ctx->tunnel_def_mask = &rte_flow_item_nvgre_mask;
+                       ctx->tunnel_def_mask_size =
+                               sizeof(rte_flow_item_nvgre_mask);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_END:
+                       break;
+               default:
+                       ++pattern;
+                       continue;
+               };
+
+               break;
+       }
+
+       if (pattern->type == RTE_FLOW_ITEM_TYPE_END)
+               return 0;
+
+       if ((mae->encap_types_supported & (1U << ctx->encap_type)) == 0) {
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ITEM,
+                                         pattern, "Unsupported tunnel item");
+       }
+
+       if (ctx->priority >= mae->nb_outer_rule_prios_max) {
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
+                                         NULL, "Unsupported priority level");
+       }
+
+       rc = efx_mae_match_spec_init(sa->nic, EFX_MAE_RULE_OUTER, ctx->priority,
+                                    &ctx->match_spec_outer);
+       if (rc != 0) {
+               return rte_flow_error_set(error, rc,
+                       RTE_FLOW_ERROR_TYPE_ITEM, pattern,
+                       "Failed to initialise outer rule match specification");
+       }
+
+       /* Outermost items comprise a match specification of type OUTER. */
+       ctx->match_spec = ctx->match_spec_outer;
+
+       /* Outermost items use "ENC" EFX MAE field IDs. */
+       ctx->field_ids_remap = field_ids_remap_to_encap;
+
+       return 0;
+}
+
+static void
+sfc_mae_rule_encap_parse_fini(struct sfc_adapter *sa,
+                             struct sfc_mae_parse_ctx *ctx)
+{
+       if (ctx->encap_type == EFX_TUNNEL_PROTOCOL_NONE)
+               return;
+
+       if (ctx->match_spec_outer != NULL)
+               efx_mae_match_spec_fini(sa->nic, ctx->match_spec_outer);
+}
+
+int
+sfc_mae_rule_parse_pattern(struct sfc_adapter *sa,
+                          const struct rte_flow_item pattern[],
+                          struct sfc_flow_spec_mae *spec,
+                          struct rte_flow_error *error)
+{
+       struct sfc_mae_parse_ctx ctx_mae;
+       struct sfc_flow_parse_ctx ctx;
+       int rc;
+
+       memset(&ctx_mae, 0, sizeof(ctx_mae));
+       ctx_mae.priority = spec->priority;
+       ctx_mae.sa = sa;
+
+       rc = efx_mae_match_spec_init(sa->nic, EFX_MAE_RULE_ACTION,
+                                    spec->priority,
+                                    &ctx_mae.match_spec_action);
+       if (rc != 0) {
+               rc = rte_flow_error_set(error, rc,
+                       RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+                       "Failed to initialise action rule match specification");
+               goto fail_init_match_spec_action;
+       }
+
+       /*
+        * As a preliminary setting, assume that there is no encapsulation
+        * in the pattern. That is, pattern items are about to comprise a
+        * match specification of type ACTION and use non-encap. field IDs.
+        *
+        * sfc_mae_rule_encap_parse_init() below may override this.
+        */
+       ctx_mae.encap_type = EFX_TUNNEL_PROTOCOL_NONE;
+       ctx_mae.match_spec = ctx_mae.match_spec_action;
+       ctx_mae.field_ids_remap = field_ids_no_remap;
+
+       ctx.type = SFC_FLOW_PARSE_CTX_MAE;
+       ctx.mae = &ctx_mae;
+
+       rc = sfc_mae_rule_encap_parse_init(sa, pattern, &ctx_mae, error);
+       if (rc != 0)
+               goto fail_encap_parse_init;
+
+       rc = sfc_flow_parse_pattern(sfc_flow_items, RTE_DIM(sfc_flow_items),
+                                   pattern, &ctx, error);
+       if (rc != 0)
+               goto fail_parse_pattern;
+
+       rc = sfc_mae_rule_process_pattern_data(&ctx_mae, error);
+       if (rc != 0)
+               goto fail_process_pattern_data;
+
+       rc = sfc_mae_rule_process_outer(sa, &ctx_mae, &spec->outer_rule, error);
+       if (rc != 0)
+               goto fail_process_outer;
+
+       if (!efx_mae_match_spec_is_valid(sa->nic, ctx_mae.match_spec_action)) {
+               rc = rte_flow_error_set(error, ENOTSUP,
+                                       RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+                                       "Inconsistent pattern");
+               goto fail_validate_match_spec_action;
+       }
+
+       spec->match_spec = ctx_mae.match_spec_action;
+
+       return 0;
+
+fail_validate_match_spec_action:
+fail_process_outer:
+fail_process_pattern_data:
+fail_parse_pattern:
+       sfc_mae_rule_encap_parse_fini(sa, &ctx_mae);
+
+fail_encap_parse_init:
+       efx_mae_match_spec_fini(sa->nic, ctx_mae.match_spec_action);
+
+fail_init_match_spec_action:
+       return rc;
+}
+
+/*
+ * An action supported by MAE may correspond to a bundle of RTE flow actions,
+ * in example, VLAN_PUSH = OF_PUSH_VLAN + OF_VLAN_SET_VID + OF_VLAN_SET_PCP.
+ * That is, related RTE flow actions need to be tracked as parts of a whole
+ * so that they can be combined into a single action and submitted to MAE
+ * representation of a given rule's action set.
+ *
+ * Each RTE flow action provided by an application gets classified as
+ * one belonging to some bundle type. If an action is not supposed to
+ * belong to any bundle, or if this action is END, it is described as
+ * one belonging to a dummy bundle of type EMPTY.
+ *
+ * A currently tracked bundle will be submitted if a repeating
+ * action or an action of different bundle type follows.
+ */
+
+enum sfc_mae_actions_bundle_type {
+       SFC_MAE_ACTIONS_BUNDLE_EMPTY = 0,
+       SFC_MAE_ACTIONS_BUNDLE_VLAN_PUSH,
+};
+
+struct sfc_mae_actions_bundle {
+       enum sfc_mae_actions_bundle_type        type;
+
+       /* Indicates actions already tracked by the current bundle */
+       uint64_t                                actions_mask;
+
+       /* Parameters used by SFC_MAE_ACTIONS_BUNDLE_VLAN_PUSH */
+       rte_be16_t                              vlan_push_tpid;
+       rte_be16_t                              vlan_push_tci;
+};
+
+/*
+ * Combine configuration of RTE flow actions tracked by the bundle into a
+ * single action and submit the result to MAE action set specification.
+ * Do nothing in the case of dummy action bundle.
+ */
+static int
+sfc_mae_actions_bundle_submit(const struct sfc_mae_actions_bundle *bundle,
+                             efx_mae_actions_t *spec)
+{
+       int rc = 0;
+
+       switch (bundle->type) {
+       case SFC_MAE_ACTIONS_BUNDLE_EMPTY:
+               break;
+       case SFC_MAE_ACTIONS_BUNDLE_VLAN_PUSH:
+               rc = efx_mae_action_set_populate_vlan_push(
+                       spec, bundle->vlan_push_tpid, bundle->vlan_push_tci);
+               break;
+       default:
+               SFC_ASSERT(B_FALSE);
+               break;
+       }
+
+       return rc;
+}
+
+/*
+ * Given the type of the next RTE flow action in the line, decide
+ * whether a new bundle is about to start, and, if this is the case,
+ * submit and reset the current bundle.
+ */
+static int
+sfc_mae_actions_bundle_sync(const struct rte_flow_action *action,
+                           struct sfc_mae_actions_bundle *bundle,
+                           efx_mae_actions_t *spec,
+                           struct rte_flow_error *error)
+{
+       enum sfc_mae_actions_bundle_type bundle_type_new;
+       int rc;
+
+       switch (action->type) {
+       case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
+       case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
+       case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
+               bundle_type_new = SFC_MAE_ACTIONS_BUNDLE_VLAN_PUSH;
+               break;
+       default:
+               /*
+                * Self-sufficient actions, including END, are handled in this
+                * case. No checks for unsupported actions are needed here
+                * because parsing doesn't occur at this point.
+                */
+               bundle_type_new = SFC_MAE_ACTIONS_BUNDLE_EMPTY;
+               break;
+       }
+
+       if (bundle_type_new != bundle->type ||
+           (bundle->actions_mask & (1ULL << action->type)) != 0) {
+               rc = sfc_mae_actions_bundle_submit(bundle, spec);
+               if (rc != 0)
+                       goto fail_submit;
+
+               memset(bundle, 0, sizeof(*bundle));
+       }
+
+       bundle->type = bundle_type_new;
+
+       return 0;
+
+fail_submit:
+       return rte_flow_error_set(error, rc,
+                       RTE_FLOW_ERROR_TYPE_ACTION, NULL,
+                       "Failed to request the (group of) action(s)");
+}
+
+static void
+sfc_mae_rule_parse_action_of_push_vlan(
+                           const struct rte_flow_action_of_push_vlan *conf,
+                           struct sfc_mae_actions_bundle *bundle)
+{
+       bundle->vlan_push_tpid = conf->ethertype;
+}
+
+static void
+sfc_mae_rule_parse_action_of_set_vlan_vid(
+                           const struct rte_flow_action_of_set_vlan_vid *conf,
+                           struct sfc_mae_actions_bundle *bundle)
+{
+       bundle->vlan_push_tci |= (conf->vlan_vid &
+                                 rte_cpu_to_be_16(RTE_LEN2MASK(12, uint16_t)));
+}
+
+static void
+sfc_mae_rule_parse_action_of_set_vlan_pcp(
+                           const struct rte_flow_action_of_set_vlan_pcp *conf,
+                           struct sfc_mae_actions_bundle *bundle)
+{
+       uint16_t vlan_tci_pcp = (uint16_t)(conf->vlan_pcp &
+                                          RTE_LEN2MASK(3, uint8_t)) << 13;
+
+       bundle->vlan_push_tci |= rte_cpu_to_be_16(vlan_tci_pcp);
+}
+
+struct sfc_mae_parsed_item {
+       const struct rte_flow_item      *item;
+       size_t                          proto_header_ofst;
+       size_t                          proto_header_size;
+};
+
+/*
+ * For each 16-bit word of the given header, override
+ * bits enforced by the corresponding 16-bit mask.
+ */
+static void
+sfc_mae_header_force_item_masks(uint8_t *header_buf,
+                               const struct sfc_mae_parsed_item *parsed_items,
+                               unsigned int nb_parsed_items)
+{
+       unsigned int item_idx;
+
+       for (item_idx = 0; item_idx < nb_parsed_items; ++item_idx) {
+               const struct sfc_mae_parsed_item *parsed_item;
+               const struct rte_flow_item *item;
+               size_t proto_header_size;
+               size_t ofst;
+
+               parsed_item = &parsed_items[item_idx];
+               proto_header_size = parsed_item->proto_header_size;
+               item = parsed_item->item;
+
+               for (ofst = 0; ofst < proto_header_size;
+                    ofst += sizeof(rte_be16_t)) {
+                       rte_be16_t *wp = RTE_PTR_ADD(header_buf, ofst);
+                       const rte_be16_t *w_maskp;
+                       const rte_be16_t *w_specp;
+
+                       w_maskp = RTE_PTR_ADD(item->mask, ofst);
+                       w_specp = RTE_PTR_ADD(item->spec, ofst);
+
+                       *wp &= ~(*w_maskp);
+                       *wp |= (*w_specp & *w_maskp);
+               }
+
+               header_buf += proto_header_size;
+       }
+}
+
+#define SFC_IPV4_TTL_DEF       0x40
+#define SFC_IPV6_VTC_FLOW_DEF  0x60000000
+#define SFC_IPV6_HOP_LIMITS_DEF        0xff
+#define SFC_VXLAN_FLAGS_DEF    0x08000000
+
+static int
+sfc_mae_rule_parse_action_vxlan_encap(
+                           struct sfc_mae *mae,
+                           const struct rte_flow_action_vxlan_encap *conf,
+                           efx_mae_actions_t *spec,
+                           struct rte_flow_error *error)
+{
+       struct sfc_mae_bounce_eh *bounce_eh = &mae->bounce_eh;
+       struct rte_flow_item *pattern = conf->definition;
+       uint8_t *buf = bounce_eh->buf;
 
-       sfc_log_init(sa, "entry");
+       /* This array will keep track of non-VOID pattern items. */
+       struct sfc_mae_parsed_item parsed_items[1 /* Ethernet */ +
+                                               2 /* VLAN tags */ +
+                                               1 /* IPv4 or IPv6 */ +
+                                               1 /* UDP */ +
+                                               1 /* VXLAN */];
+       unsigned int nb_parsed_items = 0;
 
-       if (!encp->enc_mae_supported) {
-               mae->status = SFC_MAE_STATUS_UNSUPPORTED;
-               return 0;
+       size_t eth_ethertype_ofst = offsetof(struct rte_ether_hdr, ether_type);
+       uint8_t dummy_buf[RTE_MAX(sizeof(struct rte_ipv4_hdr),
+                                 sizeof(struct rte_ipv6_hdr))];
+       struct rte_ipv4_hdr *ipv4 = (void *)dummy_buf;
+       struct rte_ipv6_hdr *ipv6 = (void *)dummy_buf;
+       struct rte_vxlan_hdr *vxlan = NULL;
+       struct rte_udp_hdr *udp = NULL;
+       unsigned int nb_vlan_tags = 0;
+       size_t next_proto_ofst = 0;
+       size_t ethertype_ofst = 0;
+       uint64_t exp_items;
+
+       if (pattern == NULL) {
+               return rte_flow_error_set(error, EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                               "The encap. header definition is NULL");
        }
 
-       sfc_log_init(sa, "init MAE");
-       rc = efx_mae_init(sa->nic);
-       if (rc != 0)
-               goto fail_mae_init;
+       bounce_eh->type = EFX_TUNNEL_PROTOCOL_VXLAN;
+       bounce_eh->size = 0;
 
-       sfc_log_init(sa, "get MAE limits");
-       rc = efx_mae_get_limits(sa->nic, &limits);
-       if (rc != 0)
-               goto fail_mae_get_limits;
+       /*
+        * Process pattern items and remember non-VOID ones.
+        * Defer applying masks until after the complete header
+        * has been built from the pattern items.
+        */
+       exp_items = RTE_BIT64(RTE_FLOW_ITEM_TYPE_ETH);
 
-       mae->status = SFC_MAE_STATUS_SUPPORTED;
-       mae->nb_action_rule_prios_max = limits.eml_max_n_action_prios;
+       for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; ++pattern) {
+               struct sfc_mae_parsed_item *parsed_item;
+               const uint64_t exp_items_extra_vlan[] = {
+                       RTE_BIT64(RTE_FLOW_ITEM_TYPE_VLAN), 0
+               };
+               size_t proto_header_size;
+               rte_be16_t *ethertypep;
+               uint8_t *next_protop;
+               uint8_t *buf_cur;
 
-       sfc_log_init(sa, "done");
+               if (pattern->spec == NULL) {
+                       return rte_flow_error_set(error, EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "NULL item spec in the encap. header");
+               }
 
-       return 0;
+               if (pattern->mask == NULL) {
+                       return rte_flow_error_set(error, EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "NULL item mask in the encap. header");
+               }
 
-fail_mae_get_limits:
-       efx_mae_fini(sa->nic);
+               if (pattern->last != NULL) {
+                       /* This is not a match pattern, so disallow range. */
+                       return rte_flow_error_set(error, EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "Range item in the encap. header");
+               }
 
-fail_mae_init:
-       sfc_log_init(sa, "failed %d", rc);
+               if (pattern->type == RTE_FLOW_ITEM_TYPE_VOID) {
+                       /* Handle VOID separately, for clarity. */
+                       continue;
+               }
 
-       return rc;
+               if ((exp_items & RTE_BIT64(pattern->type)) == 0) {
+                       return rte_flow_error_set(error, ENOTSUP,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "Unexpected item in the encap. header");
+               }
+
+               parsed_item = &parsed_items[nb_parsed_items];
+               buf_cur = buf + bounce_eh->size;
+
+               switch (pattern->type) {
+               case RTE_FLOW_ITEM_TYPE_ETH:
+                       SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ITEM_TYPE_ETH,
+                                              exp_items);
+                       RTE_BUILD_BUG_ON(offsetof(struct rte_flow_item_eth,
+                                                 hdr) != 0);
+
+                       proto_header_size = sizeof(struct rte_ether_hdr);
+
+                       ethertype_ofst = eth_ethertype_ofst;
+
+                       exp_items = RTE_BIT64(RTE_FLOW_ITEM_TYPE_VLAN) |
+                                   RTE_BIT64(RTE_FLOW_ITEM_TYPE_IPV4) |
+                                   RTE_BIT64(RTE_FLOW_ITEM_TYPE_IPV6);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_VLAN:
+                       SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ITEM_TYPE_VLAN,
+                                              exp_items);
+                       RTE_BUILD_BUG_ON(offsetof(struct rte_flow_item_vlan,
+                                                 hdr) != 0);
+
+                       proto_header_size = sizeof(struct rte_vlan_hdr);
+
+                       ethertypep = RTE_PTR_ADD(buf, eth_ethertype_ofst);
+                       *ethertypep = RTE_BE16(RTE_ETHER_TYPE_QINQ);
+
+                       ethertypep = RTE_PTR_ADD(buf, ethertype_ofst);
+                       *ethertypep = RTE_BE16(RTE_ETHER_TYPE_VLAN);
+
+                       ethertype_ofst =
+                           bounce_eh->size +
+                           offsetof(struct rte_vlan_hdr, eth_proto);
+
+                       exp_items = RTE_BIT64(RTE_FLOW_ITEM_TYPE_IPV4) |
+                                   RTE_BIT64(RTE_FLOW_ITEM_TYPE_IPV6);
+                       exp_items |= exp_items_extra_vlan[nb_vlan_tags];
+
+                       ++nb_vlan_tags;
+                       break;
+               case RTE_FLOW_ITEM_TYPE_IPV4:
+                       SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ITEM_TYPE_IPV4,
+                                              exp_items);
+                       RTE_BUILD_BUG_ON(offsetof(struct rte_flow_item_ipv4,
+                                                 hdr) != 0);
+
+                       proto_header_size = sizeof(struct rte_ipv4_hdr);
+
+                       ethertypep = RTE_PTR_ADD(buf, ethertype_ofst);
+                       *ethertypep = RTE_BE16(RTE_ETHER_TYPE_IPV4);
+
+                       next_proto_ofst =
+                           bounce_eh->size +
+                           offsetof(struct rte_ipv4_hdr, next_proto_id);
+
+                       ipv4 = (struct rte_ipv4_hdr *)buf_cur;
+
+                       exp_items = RTE_BIT64(RTE_FLOW_ITEM_TYPE_UDP);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_IPV6:
+                       SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ITEM_TYPE_IPV6,
+                                              exp_items);
+                       RTE_BUILD_BUG_ON(offsetof(struct rte_flow_item_ipv6,
+                                                 hdr) != 0);
+
+                       proto_header_size = sizeof(struct rte_ipv6_hdr);
+
+                       ethertypep = RTE_PTR_ADD(buf, ethertype_ofst);
+                       *ethertypep = RTE_BE16(RTE_ETHER_TYPE_IPV6);
+
+                       next_proto_ofst = bounce_eh->size +
+                                         offsetof(struct rte_ipv6_hdr, proto);
+
+                       ipv6 = (struct rte_ipv6_hdr *)buf_cur;
+
+                       exp_items = RTE_BIT64(RTE_FLOW_ITEM_TYPE_UDP);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_UDP:
+                       SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ITEM_TYPE_UDP,
+                                              exp_items);
+                       RTE_BUILD_BUG_ON(offsetof(struct rte_flow_item_udp,
+                                                 hdr) != 0);
+
+                       proto_header_size = sizeof(struct rte_udp_hdr);
+
+                       next_protop = RTE_PTR_ADD(buf, next_proto_ofst);
+                       *next_protop = IPPROTO_UDP;
+
+                       udp = (struct rte_udp_hdr *)buf_cur;
+
+                       exp_items = RTE_BIT64(RTE_FLOW_ITEM_TYPE_VXLAN);
+                       break;
+               case RTE_FLOW_ITEM_TYPE_VXLAN:
+                       SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ITEM_TYPE_VXLAN,
+                                              exp_items);
+                       RTE_BUILD_BUG_ON(offsetof(struct rte_flow_item_vxlan,
+                                                 hdr) != 0);
+
+                       proto_header_size = sizeof(struct rte_vxlan_hdr);
+
+                       vxlan = (struct rte_vxlan_hdr *)buf_cur;
+
+                       udp->dst_port = RTE_BE16(RTE_VXLAN_DEFAULT_PORT);
+                       udp->dgram_len = RTE_BE16(sizeof(*udp) +
+                                                 sizeof(*vxlan));
+                       udp->dgram_cksum = 0;
+
+                       exp_items = 0;
+                       break;
+               default:
+                       return rte_flow_error_set(error, ENOTSUP,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "Unknown item in the encap. header");
+               }
+
+               if (bounce_eh->size + proto_header_size > bounce_eh->buf_size) {
+                       return rte_flow_error_set(error, E2BIG,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "The encap. header is too big");
+               }
+
+               if ((proto_header_size & 1) != 0) {
+                       return rte_flow_error_set(error, EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "Odd layer size in the encap. header");
+               }
+
+               rte_memcpy(buf_cur, pattern->spec, proto_header_size);
+               bounce_eh->size += proto_header_size;
+
+               parsed_item->item = pattern;
+               parsed_item->proto_header_size = proto_header_size;
+               ++nb_parsed_items;
+       }
+
+       if (exp_items != 0) {
+               /* Parsing item VXLAN would have reset exp_items to 0. */
+               return rte_flow_error_set(error, ENOTSUP,
+                                       RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
+                                       "No item VXLAN in the encap. header");
+       }
+
+       /* One of the pointers (ipv4, ipv6) refers to a dummy area. */
+       ipv4->version_ihl = RTE_IPV4_VHL_DEF;
+       ipv4->time_to_live = SFC_IPV4_TTL_DEF;
+       ipv4->total_length = RTE_BE16(sizeof(*ipv4) + sizeof(*udp) +
+                                     sizeof(*vxlan));
+       /* The HW cannot compute this checksum. */
+       ipv4->hdr_checksum = 0;
+       ipv4->hdr_checksum = rte_ipv4_cksum(ipv4);
+
+       ipv6->vtc_flow = RTE_BE32(SFC_IPV6_VTC_FLOW_DEF);
+       ipv6->hop_limits = SFC_IPV6_HOP_LIMITS_DEF;
+       ipv6->payload_len = udp->dgram_len;
+
+       vxlan->vx_flags = RTE_BE32(SFC_VXLAN_FLAGS_DEF);
+
+       /* Take care of the masks. */
+       sfc_mae_header_force_item_masks(buf, parsed_items, nb_parsed_items);
+
+       return (spec != NULL) ? efx_mae_action_set_populate_encap(spec) : 0;
 }
 
-void
-sfc_mae_detach(struct sfc_adapter *sa)
+static int
+sfc_mae_rule_parse_action_mark(const struct rte_flow_action_mark *conf,
+                              efx_mae_actions_t *spec)
 {
-       struct sfc_mae *mae = &sa->mae;
-       enum sfc_mae_status status_prev = mae->status;
+       return efx_mae_action_set_populate_mark(spec, conf->id);
+}
 
-       sfc_log_init(sa, "entry");
+static int
+sfc_mae_rule_parse_action_phy_port(struct sfc_adapter *sa,
+                                  const struct rte_flow_action_phy_port *conf,
+                                  efx_mae_actions_t *spec)
+{
+       efx_mport_sel_t mport;
+       uint32_t phy_port;
+       int rc;
 
-       mae->nb_action_rule_prios_max = 0;
-       mae->status = SFC_MAE_STATUS_UNKNOWN;
+       if (conf->original != 0)
+               phy_port = efx_nic_cfg_get(sa->nic)->enc_assigned_port;
+       else
+               phy_port = conf->index;
 
-       if (status_prev != SFC_MAE_STATUS_SUPPORTED)
-               return;
+       rc = efx_mae_mport_by_phy_port(phy_port, &mport);
+       if (rc != 0)
+               return rc;
 
-       efx_mae_fini(sa->nic);
+       return efx_mae_action_set_populate_deliver(spec, &mport);
+}
 
-       sfc_log_init(sa, "done");
+static int
+sfc_mae_rule_parse_action_pf_vf(struct sfc_adapter *sa,
+                               const struct rte_flow_action_vf *vf_conf,
+                               efx_mae_actions_t *spec)
+{
+       const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
+       efx_mport_sel_t mport;
+       uint32_t vf;
+       int rc;
+
+       if (vf_conf == NULL)
+               vf = EFX_PCI_VF_INVALID;
+       else if (vf_conf->original != 0)
+               vf = encp->enc_vf;
+       else
+               vf = vf_conf->id;
+
+       rc = efx_mae_mport_by_pcie_function(encp->enc_pf, vf, &mport);
+       if (rc != 0)
+               return rc;
+
+       return efx_mae_action_set_populate_deliver(spec, &mport);
 }
 
-void
-sfc_mae_flow_cleanup(struct sfc_adapter *sa,
-                    struct rte_flow *flow)
+static int
+sfc_mae_rule_parse_action_port_id(struct sfc_adapter *sa,
+                                 const struct rte_flow_action_port_id *conf,
+                                 efx_mae_actions_t *spec)
 {
-       struct sfc_flow_spec *spec;
-       struct sfc_flow_spec_mae *spec_mae;
+       struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
+       struct sfc_mae *mae = &sa->mae;
+       efx_mport_sel_t mport;
+       uint16_t port_id;
+       int rc;
 
-       if (flow == NULL)
-               return;
+       port_id = (conf->original != 0) ? sas->port_id : conf->id;
 
-       spec = &flow->spec;
+       rc = sfc_mae_switch_port_by_ethdev(mae->switch_domain_id,
+                                          port_id, &mport);
+       if (rc != 0)
+               return rc;
 
-       if (spec == NULL)
-               return;
+       return efx_mae_action_set_populate_deliver(spec, &mport);
+}
 
-       spec_mae = &spec->mae;
+static int
+sfc_mae_rule_parse_action(struct sfc_adapter *sa,
+                         const struct rte_flow_action *action,
+                         const struct sfc_mae_outer_rule *outer_rule,
+                         struct sfc_mae_actions_bundle *bundle,
+                         efx_mae_actions_t *spec,
+                         struct rte_flow_error *error)
+{
+       bool custom_error = B_FALSE;
+       int rc = 0;
 
-       if (spec_mae->match_spec != NULL)
-               efx_mae_match_spec_fini(sa->nic, spec_mae->match_spec);
+       switch (action->type) {
+       case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_VXLAN_DECAP,
+                                      bundle->actions_mask);
+               if (outer_rule == NULL ||
+                   outer_rule->encap_type != EFX_TUNNEL_PROTOCOL_VXLAN)
+                       rc = EINVAL;
+               else
+                       rc = efx_mae_action_set_populate_decap(spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_OF_POP_VLAN,
+                                      bundle->actions_mask);
+               rc = efx_mae_action_set_populate_vlan_pop(spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN,
+                                      bundle->actions_mask);
+               sfc_mae_rule_parse_action_of_push_vlan(action->conf, bundle);
+               break;
+       case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID,
+                                      bundle->actions_mask);
+               sfc_mae_rule_parse_action_of_set_vlan_vid(action->conf, bundle);
+               break;
+       case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP,
+                                      bundle->actions_mask);
+               sfc_mae_rule_parse_action_of_set_vlan_pcp(action->conf, bundle);
+               break;
+       case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP,
+                                      bundle->actions_mask);
+               rc = sfc_mae_rule_parse_action_vxlan_encap(&sa->mae,
+                                                          action->conf,
+                                                          spec, error);
+               custom_error = B_TRUE;
+               break;
+       case RTE_FLOW_ACTION_TYPE_FLAG:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_FLAG,
+                                      bundle->actions_mask);
+               rc = efx_mae_action_set_populate_flag(spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_MARK:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_MARK,
+                                      bundle->actions_mask);
+               rc = sfc_mae_rule_parse_action_mark(action->conf, spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_PHY_PORT:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_PHY_PORT,
+                                      bundle->actions_mask);
+               rc = sfc_mae_rule_parse_action_phy_port(sa, action->conf, spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_PF:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_PF,
+                                      bundle->actions_mask);
+               rc = sfc_mae_rule_parse_action_pf_vf(sa, NULL, spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_VF:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_VF,
+                                      bundle->actions_mask);
+               rc = sfc_mae_rule_parse_action_pf_vf(sa, action->conf, spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_PORT_ID:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_PORT_ID,
+                                      bundle->actions_mask);
+               rc = sfc_mae_rule_parse_action_port_id(sa, action->conf, spec);
+               break;
+       case RTE_FLOW_ACTION_TYPE_DROP:
+               SFC_BUILD_SET_OVERFLOW(RTE_FLOW_ACTION_TYPE_DROP,
+                                      bundle->actions_mask);
+               rc = efx_mae_action_set_populate_drop(spec);
+               break;
+       default:
+               return rte_flow_error_set(error, ENOTSUP,
+                               RTE_FLOW_ERROR_TYPE_ACTION, NULL,
+                               "Unsupported action");
+       }
+
+       if (rc == 0) {
+               bundle->actions_mask |= (1ULL << action->type);
+       } else if (!custom_error) {
+               rc = rte_flow_error_set(error, rc, RTE_FLOW_ERROR_TYPE_ACTION,
+                               NULL, "Failed to request the action");
+       }
+
+       return rc;
 }
 
-static const struct sfc_flow_item sfc_flow_items[] = {
-};
+static void
+sfc_mae_bounce_eh_invalidate(struct sfc_mae_bounce_eh *bounce_eh)
+{
+       bounce_eh->type = EFX_TUNNEL_PROTOCOL_NONE;
+}
+
+static int
+sfc_mae_process_encap_header(struct sfc_adapter *sa,
+                            const struct sfc_mae_bounce_eh *bounce_eh,
+                            struct sfc_mae_encap_header **encap_headerp)
+{
+       if (bounce_eh->type == EFX_TUNNEL_PROTOCOL_NONE) {
+               encap_headerp = NULL;
+               return 0;
+       }
+
+       *encap_headerp = sfc_mae_encap_header_attach(sa, bounce_eh);
+       if (*encap_headerp != NULL)
+               return 0;
+
+       return sfc_mae_encap_header_add(sa, bounce_eh, encap_headerp);
+}
 
 int
-sfc_mae_rule_parse_pattern(struct sfc_adapter *sa,
-                          const struct rte_flow_item pattern[],
-                          struct sfc_flow_spec_mae *spec,
+sfc_mae_rule_parse_actions(struct sfc_adapter *sa,
+                          const struct rte_flow_action actions[],
+                          struct sfc_flow_spec_mae *spec_mae,
                           struct rte_flow_error *error)
 {
-       struct sfc_mae_parse_ctx ctx_mae;
-       struct sfc_flow_parse_ctx ctx;
+       struct sfc_mae_encap_header *encap_header = NULL;
+       struct sfc_mae_actions_bundle bundle = {0};
+       const struct rte_flow_action *action;
+       struct sfc_mae *mae = &sa->mae;
+       efx_mae_actions_t *spec;
        int rc;
 
-       memset(&ctx_mae, 0, sizeof(ctx_mae));
+       rte_errno = 0;
 
-       rc = efx_mae_match_spec_init(sa->nic, EFX_MAE_RULE_ACTION,
-                                    spec->priority,
-                                    &ctx_mae.match_spec_action);
-       if (rc != 0) {
-               rc = rte_flow_error_set(error, rc,
-                       RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
-                       "Failed to initialise action rule match specification");
-               goto fail_init_match_spec_action;
+       if (actions == NULL) {
+               return rte_flow_error_set(error, EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ACTION_NUM, NULL,
+                               "NULL actions");
        }
 
-       ctx.type = SFC_FLOW_PARSE_CTX_MAE;
-       ctx.mae = &ctx_mae;
+       rc = efx_mae_action_set_spec_init(sa->nic, &spec);
+       if (rc != 0)
+               goto fail_action_set_spec_init;
 
-       rc = sfc_flow_parse_pattern(sfc_flow_items, RTE_DIM(sfc_flow_items),
-                                   pattern, &ctx, error);
+       /* Cleanup after previous encap. header bounce buffer usage. */
+       sfc_mae_bounce_eh_invalidate(&mae->bounce_eh);
+
+       for (action = actions;
+            action->type != RTE_FLOW_ACTION_TYPE_END; ++action) {
+               rc = sfc_mae_actions_bundle_sync(action, &bundle, spec, error);
+               if (rc != 0)
+                       goto fail_rule_parse_action;
+
+               rc = sfc_mae_rule_parse_action(sa, action, spec_mae->outer_rule,
+                                              &bundle, spec, error);
+               if (rc != 0)
+                       goto fail_rule_parse_action;
+       }
+
+       rc = sfc_mae_actions_bundle_sync(action, &bundle, spec, error);
        if (rc != 0)
-               goto fail_parse_pattern;
+               goto fail_rule_parse_action;
 
-       if (!efx_mae_match_spec_is_valid(sa->nic, ctx_mae.match_spec_action)) {
-               rc = rte_flow_error_set(error, ENOTSUP,
-                                       RTE_FLOW_ERROR_TYPE_ITEM, NULL,
-                                       "Inconsistent pattern");
-               goto fail_validate_match_spec_action;
+       rc = sfc_mae_process_encap_header(sa, &mae->bounce_eh, &encap_header);
+       if (rc != 0)
+               goto fail_process_encap_header;
+
+       spec_mae->action_set = sfc_mae_action_set_attach(sa, encap_header,
+                                                        spec);
+       if (spec_mae->action_set != NULL) {
+               sfc_mae_encap_header_del(sa, encap_header);
+               efx_mae_action_set_spec_fini(sa->nic, spec);
+               return 0;
        }
 
-       spec->match_spec = ctx_mae.match_spec_action;
+       rc = sfc_mae_action_set_add(sa, spec, encap_header,
+                                   &spec_mae->action_set);
+       if (rc != 0)
+               goto fail_action_set_add;
 
        return 0;
 
-fail_validate_match_spec_action:
-fail_parse_pattern:
-       efx_mae_match_spec_fini(sa->nic, ctx_mae.match_spec_action);
+fail_action_set_add:
+       sfc_mae_encap_header_del(sa, encap_header);
 
-fail_init_match_spec_action:
+fail_process_encap_header:
+fail_rule_parse_action:
+       efx_mae_action_set_spec_fini(sa->nic, spec);
+
+fail_action_set_spec_init:
+       if (rc > 0 && rte_errno == 0) {
+               rc = rte_flow_error_set(error, rc,
+                       RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                       NULL, "Failed to process the action");
+       }
        return rc;
 }
 
@@ -163,6 +2716,37 @@ sfc_mae_rules_class_cmp(struct sfc_adapter *sa,
        return (rc == 0) ? have_same_class : false;
 }
 
+static int
+sfc_mae_outer_rule_class_verify(struct sfc_adapter *sa,
+                               struct sfc_mae_outer_rule *rule)
+{
+       struct sfc_mae_fw_rsrc *fw_rsrc = &rule->fw_rsrc;
+       struct sfc_mae_outer_rule *entry;
+       struct sfc_mae *mae = &sa->mae;
+
+       if (fw_rsrc->rule_id.id != EFX_MAE_RSRC_ID_INVALID) {
+               /* An active rule is reused. It's class is wittingly valid. */
+               return 0;
+       }
+
+       TAILQ_FOREACH_REVERSE(entry, &mae->outer_rules,
+                             sfc_mae_outer_rules, entries) {
+               const efx_mae_match_spec_t *left = entry->match_spec;
+               const efx_mae_match_spec_t *right = rule->match_spec;
+
+               if (entry == rule)
+                       continue;
+
+               if (sfc_mae_rules_class_cmp(sa, left, right))
+                       return 0;
+       }
+
+       sfc_info(sa, "for now, the HW doesn't support rule validation, and HW "
+                "support for outer frame pattern items is not guaranteed; "
+                "other than that, the items are valid from SW standpoint");
+       return 0;
+}
+
 static int
 sfc_mae_action_rule_class_verify(struct sfc_adapter *sa,
                                 struct sfc_flow_spec_mae *spec)
@@ -215,11 +2799,94 @@ sfc_mae_flow_verify(struct sfc_adapter *sa,
 {
        struct sfc_flow_spec *spec = &flow->spec;
        struct sfc_flow_spec_mae *spec_mae = &spec->mae;
+       struct sfc_mae_outer_rule *outer_rule = spec_mae->outer_rule;
+       int rc;
 
        SFC_ASSERT(sfc_adapter_is_locked(sa));
 
        if (sa->state != SFC_ADAPTER_STARTED)
                return EAGAIN;
 
+       if (outer_rule != NULL) {
+               rc = sfc_mae_outer_rule_class_verify(sa, outer_rule);
+               if (rc != 0)
+                       return rc;
+       }
+
        return sfc_mae_action_rule_class_verify(sa, spec_mae);
 }
+
+int
+sfc_mae_flow_insert(struct sfc_adapter *sa,
+                   struct rte_flow *flow)
+{
+       struct sfc_flow_spec *spec = &flow->spec;
+       struct sfc_flow_spec_mae *spec_mae = &spec->mae;
+       struct sfc_mae_outer_rule *outer_rule = spec_mae->outer_rule;
+       struct sfc_mae_action_set *action_set = spec_mae->action_set;
+       struct sfc_mae_fw_rsrc *fw_rsrc = &action_set->fw_rsrc;
+       int rc;
+
+       SFC_ASSERT(spec_mae->rule_id.id == EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(action_set != NULL);
+
+       if (outer_rule != NULL) {
+               rc = sfc_mae_outer_rule_enable(sa, outer_rule,
+                                              spec_mae->match_spec);
+               if (rc != 0)
+                       goto fail_outer_rule_enable;
+       }
+
+       rc = sfc_mae_action_set_enable(sa, action_set);
+       if (rc != 0)
+               goto fail_action_set_enable;
+
+       rc = efx_mae_action_rule_insert(sa->nic, spec_mae->match_spec,
+                                       NULL, &fw_rsrc->aset_id,
+                                       &spec_mae->rule_id);
+       if (rc != 0)
+               goto fail_action_rule_insert;
+
+       return 0;
+
+fail_action_rule_insert:
+       (void)sfc_mae_action_set_disable(sa, action_set);
+
+fail_action_set_enable:
+       if (outer_rule != NULL)
+               (void)sfc_mae_outer_rule_disable(sa, outer_rule);
+
+fail_outer_rule_enable:
+       return rc;
+}
+
+int
+sfc_mae_flow_remove(struct sfc_adapter *sa,
+                   struct rte_flow *flow)
+{
+       struct sfc_flow_spec *spec = &flow->spec;
+       struct sfc_flow_spec_mae *spec_mae = &spec->mae;
+       struct sfc_mae_action_set *action_set = spec_mae->action_set;
+       struct sfc_mae_outer_rule *outer_rule = spec_mae->outer_rule;
+       int rc;
+
+       SFC_ASSERT(spec_mae->rule_id.id != EFX_MAE_RSRC_ID_INVALID);
+       SFC_ASSERT(action_set != NULL);
+
+       rc = efx_mae_action_rule_remove(sa->nic, &spec_mae->rule_id);
+       if (rc != 0)
+               return rc;
+
+       spec_mae->rule_id.id = EFX_MAE_RSRC_ID_INVALID;
+
+       rc = sfc_mae_action_set_disable(sa, action_set);
+       if (rc != 0) {
+               sfc_err(sa, "failed to disable the action set (rc = %d)", rc);
+               /* Despite the error, proceed with outer rule removal. */
+       }
+
+       if (outer_rule != NULL)
+               return sfc_mae_outer_rule_disable(sa, outer_rule);
+
+       return 0;
+}