return 0;
 }
 
+static void
+nfp_tun_mutual_link(struct nfp_predt_entry *predt,
+                   struct nfp_neigh_entry *neigh)
+{
+       struct nfp_fl_payload *flow_pay = predt->flow_pay;
+       struct nfp_tun_neigh_ext *ext;
+       struct nfp_tun_neigh *common;
+
+       if (flow_pay->pre_tun_rule.is_ipv6 != neigh->is_ipv6)
+               return;
+
+       /* In the case of bonding it is possible that there might already
+        * be a flow linked (as the MAC address gets shared). If a flow
+        * is already linked just return.
+        */
+       if (neigh->flow)
+               return;
+
+       common = neigh->is_ipv6 ?
+                &((struct nfp_tun_neigh_v6 *)neigh->payload)->common :
+                &((struct nfp_tun_neigh_v4 *)neigh->payload)->common;
+       ext = neigh->is_ipv6 ?
+                &((struct nfp_tun_neigh_v6 *)neigh->payload)->ext :
+                &((struct nfp_tun_neigh_v4 *)neigh->payload)->ext;
+
+       if (memcmp(flow_pay->pre_tun_rule.loc_mac,
+                  common->src_addr, ETH_ALEN) ||
+           memcmp(flow_pay->pre_tun_rule.rem_mac,
+                  common->dst_addr, ETH_ALEN))
+               return;
+
+       list_add(&neigh->list_head, &predt->nn_list);
+       neigh->flow = predt;
+       ext->host_ctx = flow_pay->meta.host_ctx_id;
+       ext->vlan_tci = flow_pay->pre_tun_rule.vlan_tci;
+       ext->vlan_tpid = flow_pay->pre_tun_rule.vlan_tpid;
+}
+
+static void
+nfp_tun_link_predt_entries(struct nfp_app *app,
+                          struct nfp_neigh_entry *nn_entry)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_predt_entry *predt, *tmp;
+
+       list_for_each_entry_safe(predt, tmp, &priv->predt_list, list_head) {
+               nfp_tun_mutual_link(predt, nn_entry);
+       }
+}
+
+void nfp_tun_link_and_update_nn_entries(struct nfp_app *app,
+                                       struct nfp_predt_entry *predt)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_neigh_entry *nn_entry;
+       struct rhashtable_iter iter;
+       size_t neigh_size;
+       u8 type;
+
+       rhashtable_walk_enter(&priv->neigh_table, &iter);
+       rhashtable_walk_start(&iter);
+       while ((nn_entry = rhashtable_walk_next(&iter)) != NULL) {
+               if (IS_ERR(nn_entry))
+                       continue;
+               nfp_tun_mutual_link(predt, nn_entry);
+               neigh_size = nn_entry->is_ipv6 ?
+                            sizeof(struct nfp_tun_neigh_v6) :
+                            sizeof(struct nfp_tun_neigh_v4);
+               type = nn_entry->is_ipv6 ? NFP_FLOWER_CMSG_TYPE_TUN_NEIGH_V6 :
+                                          NFP_FLOWER_CMSG_TYPE_TUN_NEIGH;
+               nfp_flower_xmit_tun_conf(app, type, neigh_size,
+                                        nn_entry->payload,
+                                        GFP_ATOMIC);
+       }
+       rhashtable_walk_stop(&iter);
+       rhashtable_walk_exit(&iter);
+}
+
+static void nfp_tun_cleanup_nn_entries(struct nfp_app *app)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_neigh_entry *neigh;
+       struct nfp_tun_neigh_ext *ext;
+       struct rhashtable_iter iter;
+       size_t neigh_size;
+       u8 type;
+
+       rhashtable_walk_enter(&priv->neigh_table, &iter);
+       rhashtable_walk_start(&iter);
+       while ((neigh = rhashtable_walk_next(&iter)) != NULL) {
+               if (IS_ERR(neigh))
+                       continue;
+               ext = neigh->is_ipv6 ?
+                        &((struct nfp_tun_neigh_v6 *)neigh->payload)->ext :
+                        &((struct nfp_tun_neigh_v4 *)neigh->payload)->ext;
+               ext->host_ctx = cpu_to_be32(U32_MAX);
+               ext->vlan_tpid = cpu_to_be16(U16_MAX);
+               ext->vlan_tci = cpu_to_be16(U16_MAX);
+
+               neigh_size = neigh->is_ipv6 ?
+                            sizeof(struct nfp_tun_neigh_v6) :
+                            sizeof(struct nfp_tun_neigh_v4);
+               type = neigh->is_ipv6 ? NFP_FLOWER_CMSG_TYPE_TUN_NEIGH_V6 :
+                                          NFP_FLOWER_CMSG_TYPE_TUN_NEIGH;
+               nfp_flower_xmit_tun_conf(app, type, neigh_size, neigh->payload,
+                                        GFP_ATOMIC);
+
+               rhashtable_remove_fast(&priv->neigh_table, &neigh->ht_node,
+                                      neigh_table_params);
+               if (neigh->flow)
+                       list_del(&neigh->list_head);
+               kfree(neigh);
+       }
+       rhashtable_walk_stop(&iter);
+       rhashtable_walk_exit(&iter);
+}
+
+void nfp_tun_unlink_and_update_nn_entries(struct nfp_app *app,
+                                         struct nfp_predt_entry *predt)
+{
+       struct nfp_neigh_entry *neigh, *tmp;
+       struct nfp_tun_neigh_ext *ext;
+       size_t neigh_size;
+       u8 type;
+
+       list_for_each_entry_safe(neigh, tmp, &predt->nn_list, list_head) {
+               ext = neigh->is_ipv6 ?
+                        &((struct nfp_tun_neigh_v6 *)neigh->payload)->ext :
+                        &((struct nfp_tun_neigh_v4 *)neigh->payload)->ext;
+               neigh->flow = NULL;
+               ext->host_ctx = cpu_to_be32(U32_MAX);
+               ext->vlan_tpid = cpu_to_be16(U16_MAX);
+               ext->vlan_tci = cpu_to_be16(U16_MAX);
+               list_del(&neigh->list_head);
+               neigh_size = neigh->is_ipv6 ?
+                            sizeof(struct nfp_tun_neigh_v6) :
+                            sizeof(struct nfp_tun_neigh_v4);
+               type = neigh->is_ipv6 ? NFP_FLOWER_CMSG_TYPE_TUN_NEIGH_V6 :
+                                          NFP_FLOWER_CMSG_TYPE_TUN_NEIGH;
+               nfp_flower_xmit_tun_conf(app, type, neigh_size, neigh->payload,
+                                        GFP_ATOMIC);
+       }
+}
+
 static bool
 __nfp_tun_has_route(struct list_head *route_list, spinlock_t *list_lock,
                    void *add, int add_len)
                        nfp_tun_add_route_to_cache_v4(app, &payload->dst_ipv4);
                }
 
+               nfp_tun_link_predt_entries(app, nn_entry);
                nfp_flower_xmit_tun_conf(app, mtype, neigh_size,
                                         nn_entry->payload,
                                         GFP_ATOMIC);
        /* Destroy rhash. Entries should be cleaned on netdev notifier unreg. */
        rhashtable_free_and_destroy(&priv->tun.offloaded_macs,
                                    nfp_check_rhashtable_empty, NULL);
+
+       nfp_tun_cleanup_nn_entries(app);
 }