diff --git a/lib/dp-packet.h b/lib/dp-packet.h
index 596cfe69133..ba91e589187 100644
--- a/lib/dp-packet.h
+++ b/lib/dp-packet.h
@@ -691,6 +691,19 @@ reset_dp_packet_checksum_ol_flags(struct dp_packet *p)
 #define reset_dp_packet_checksum_ol_flags(arg)
 #endif
 
+static inline bool
+dp_packet_has_flow_mark(struct dp_packet *p OVS_UNUSED,
+                        uint32_t *mark OVS_UNUSED)
+{
+#ifdef DPDK_NETDEV
+    if (p->mbuf.ol_flags & PKT_RX_FDIR_ID) {
+        *mark = p->mbuf.hash.fdir.hi;
+        return true;
+    }
+#endif
+    return false;
+}
+
 enum { NETDEV_MAX_BURST = 32 }; /* Maximum number packets in a batch. */
 
 struct dp_packet_batch {
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index e11106e7565..cb5934f8df5 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -2119,6 +2119,23 @@ flow_mark_flush(struct dp_netdev_pmd_thread *pmd)
     }
 }
 
+static struct dp_netdev_flow *
+mark_to_flow_find(const struct dp_netdev_pmd_thread *pmd,
+                  const uint32_t mark)
+{
+    struct dp_netdev_flow *flow;
+
+    CMAP_FOR_EACH_WITH_HASH (flow, mark_node, hash_int(mark, 0),
+                             &flow_mark.mark_to_flow) {
+        if (flow->mark == mark && flow->pmd_id == pmd->core_id &&
+            flow->dead == false) {
+            return flow;
+        }
+    }
+
+    return NULL;
+}
+
 static void
 dp_netdev_pmd_remove_flow(struct dp_netdev_pmd_thread *pmd,
                           struct dp_netdev_flow *flow)
@@ -5366,10 +5383,10 @@ struct packet_batch_per_flow {
 static inline void
 packet_batch_per_flow_update(struct packet_batch_per_flow *batch,
                              struct dp_packet *packet,
-                             const struct miniflow *mf)
+                             uint16_t tcp_flags)
 {
     batch->byte_count += dp_packet_size(packet);
-    batch->tcp_flags |= miniflow_get_tcp_flags(mf);
+    batch->tcp_flags |= tcp_flags;
     batch->array.packets[batch->array.count++] = packet;
 }
 
@@ -5403,7 +5420,7 @@ packet_batch_per_flow_execute(struct packet_batch_per_flow *batch,
 
 static inline void
 dp_netdev_queue_batches(struct dp_packet *pkt,
-                        struct dp_netdev_flow *flow, const struct miniflow *mf,
+                        struct dp_netdev_flow *flow, uint16_t tcp_flags,
                         struct packet_batch_per_flow *batches,
                         size_t *n_batches)
 {
@@ -5414,7 +5431,7 @@ dp_netdev_queue_batches(struct dp_packet *pkt,
         packet_batch_per_flow_init(batch, flow);
     }
 
-    packet_batch_per_flow_update(batch, pkt, mf);
+    packet_batch_per_flow_update(batch, pkt, tcp_flags);
 }
 
 /* Try to process all ('cnt') the 'packets' using only the exact match cache
@@ -5445,6 +5462,7 @@ emc_processing(struct dp_netdev_pmd_thread *pmd,
     const size_t cnt = dp_packet_batch_size(packets_);
     uint32_t cur_min;
     int i;
+    uint16_t tcp_flags;
 
     atomic_read_relaxed(&pmd->dp->emc_insert_min, &cur_min);
     pmd_perf_update_counter(&pmd->perf_stats,
@@ -5453,6 +5471,7 @@ emc_processing(struct dp_netdev_pmd_thread *pmd,
 
     DP_PACKET_BATCH_REFILL_FOR_EACH (i, cnt, packet, packets_) {
         struct dp_netdev_flow *flow;
+        uint32_t mark;
 
         if (OVS_UNLIKELY(dp_packet_size(packet) < ETH_HEADER_LEN)) {
             dp_packet_delete(packet);
@@ -5470,6 +5489,18 @@ emc_processing(struct dp_netdev_pmd_thread *pmd,
         if (!md_is_valid) {
             pkt_metadata_init(&packet->md, port_no);
         }
+
+        if ((*recirc_depth_get() == 0) &&
+            dp_packet_has_flow_mark(packet, &mark)) {
+            flow = mark_to_flow_find(pmd, mark);
+            if (flow) {
+                tcp_flags = parse_tcp_flags(packet);
+                dp_netdev_queue_batches(packet, flow, tcp_flags, batches,
+                                        n_batches);
+                continue;
+            }
+        }
+
         miniflow_extract(packet, &key->mf);
         key->len = 0; /* Not computed yet. */
         /* If EMC is disabled skip hash computation and emc_lookup */
@@ -5485,7 +5516,8 @@ emc_processing(struct dp_netdev_pmd_thread *pmd,
             flow = NULL;
         }
         if (OVS_LIKELY(flow)) {
-            dp_netdev_queue_batches(packet, flow, &key->mf, batches,
+            tcp_flags = miniflow_get_tcp_flags(&key->mf);
+            dp_netdev_queue_batches(packet, flow, tcp_flags, batches,
                                     n_batches);
         } else {
             /* Exact match cache missed. Group missed packets together at
@@ -5672,7 +5704,9 @@ fast_path_processing(struct dp_netdev_pmd_thread *pmd,
         flow = dp_netdev_flow_cast(rules[i]);
 
         emc_probabilistic_insert(pmd, &keys[i], flow);
-        dp_netdev_queue_batches(packet, flow, &keys[i].mf, batches, n_batches);
+        dp_netdev_queue_batches(packet, flow,
+                                miniflow_get_tcp_flags(&keys[i].mf),
+                                batches, n_batches);
     }
 
     pmd_perf_update_counter(&pmd->perf_stats, PMD_STAT_MASKED_HIT,
diff --git a/lib/flow.c b/lib/flow.c
index cb8b2df1922..a785e63a82f 100644
--- a/lib/flow.c
+++ b/lib/flow.c
@@ -1019,6 +1019,73 @@ parse_dl_type(const struct eth_header *data_, size_t size)
     return parse_ethertype(&data, &size);
 }
 
+uint16_t
+parse_tcp_flags(struct dp_packet *packet)
+{
+    const void *data = dp_packet_data(packet);
+    const char *frame = (const char *)data;
+    size_t size = dp_packet_size(packet);
+    ovs_be16 dl_type;
+    uint8_t nw_frag = 0, nw_proto = 0;
+
+    if (packet->packet_type != htonl(PT_ETH)) {
+        return 0;
+    }
+
+    dp_packet_reset_offsets(packet);
+
+    data_pull(&data, &size, ETH_ADDR_LEN * 2);
+    dl_type = parse_ethertype(&data, &size);
+    if (OVS_UNLIKELY(eth_type_mpls(dl_type))) {
+        packet->l2_5_ofs = (char *)data - frame;
+    }
+    if (OVS_LIKELY(dl_type == htons(ETH_TYPE_IP))) {
+        const struct ip_header *nh = data;
+        int ip_len;
+        uint16_t tot_len;
+
+        if (OVS_UNLIKELY(!ipv4_sanity_check(nh, size, &ip_len, &tot_len))) {
+            return 0;
+        }
+        dp_packet_set_l2_pad_size(packet, size - tot_len);
+        packet->l3_ofs = (uint16_t)((char *)nh - frame);
+        nw_proto = nh->ip_proto;
+        nw_frag = ipv4_get_nw_frag(nh);
+
+        size = tot_len;   /* Never pull padding. */
+        data_pull(&data, &size, ip_len);
+    } else if (dl_type == htons(ETH_TYPE_IPV6)) {
+        const struct ovs_16aligned_ip6_hdr *nh = data;
+        uint16_t plen;
+
+        if (OVS_UNLIKELY(!ipv6_sanity_check(nh, size))) {
+            return 0;
+        }
+        packet->l3_ofs = (uint16_t)((char *)nh - frame);
+        data_pull(&data, &size, sizeof *nh);
+
+        plen = ntohs(nh->ip6_plen); /* Never pull padding. */
+        dp_packet_set_l2_pad_size(packet, size - plen);
+        size = plen;
+        if (!parse_ipv6_ext_hdrs__(&data, &size, &nw_proto, &nw_frag)) {
+            return 0;
+        }
+        nw_proto = nh->ip6_nxt;
+    } else {
+        return 0;
+    }
+
+    packet->l4_ofs = (uint16_t)((char *)data - frame);
+    if (!(nw_frag & FLOW_NW_FRAG_LATER) && nw_proto == IPPROTO_TCP &&
+        size >= TCP_HEADER_LEN) {
+        const struct tcp_header *tcp = data;
+
+        return TCP_FLAGS(tcp->tcp_ctl);
+    }
+
+    return 0;
+}
+
 /* For every bit of a field that is wildcarded in 'wildcards', sets the
  * corresponding bit in 'flow' to zero. */
 void
diff --git a/lib/flow.h b/lib/flow.h
index 5b6585f11d9..af7b5e921b3 100644
--- a/lib/flow.h
+++ b/lib/flow.h
@@ -133,6 +133,7 @@ bool parse_ipv6_ext_hdrs(const void **datap, size_t *sizep, uint8_t *nw_proto,
                          uint8_t *nw_frag);
 ovs_be16 parse_dl_type(const struct eth_header *data_, size_t size);
 bool parse_nsh(const void **datap, size_t *sizep, struct ovs_key_nsh *key);
+uint16_t parse_tcp_flags(struct dp_packet *packet);
 
 static inline uint64_t
 flow_get_xreg(const struct flow *flow, int idx)