net/virtio: separate AVX Rx/Tx
[dpdk.git] / drivers / net / virtio / virtio_rxtx.c
index 02f8d94..622d4bf 100644 (file)
@@ -45,7 +45,7 @@ virtio_dev_rx_queue_done(void *rxq, uint16_t offset)
        struct virtnet_rx *rxvq = rxq;
        struct virtqueue *vq = rxvq->vq;
 
-       return VIRTQUEUE_NUSED(vq) >= offset;
+       return virtqueue_nused(vq) >= offset;
 }
 
 void
@@ -89,21 +89,46 @@ vq_ring_free_chain(struct virtqueue *vq, uint16_t desc_idx)
        dp->next = VQ_RING_DESC_CHAIN_END;
 }
 
-static void
-vq_ring_free_id_packed(struct virtqueue *vq, uint16_t id)
+void
+virtio_update_packet_stats(struct virtnet_stats *stats, struct rte_mbuf *mbuf)
 {
-       struct vq_desc_extra *dxp;
+       uint32_t s = mbuf->pkt_len;
+       struct rte_ether_addr *ea;
 
-       dxp = &vq->vq_descx[id];
-       vq->vq_free_cnt += dxp->ndescs;
+       stats->bytes += s;
 
-       if (vq->vq_desc_tail_idx == VQ_RING_DESC_CHAIN_END)
-               vq->vq_desc_head_idx = id;
-       else
-               vq->vq_descx[vq->vq_desc_tail_idx].next = id;
+       if (s == 64) {
+               stats->size_bins[1]++;
+       } else if (s > 64 && s < 1024) {
+               uint32_t bin;
 
-       vq->vq_desc_tail_idx = id;
-       dxp->next = VQ_RING_DESC_CHAIN_END;
+               /* count zeros, and offset into correct bin */
+               bin = (sizeof(s) * 8) - __builtin_clz(s) - 5;
+               stats->size_bins[bin]++;
+       } else {
+               if (s < 64)
+                       stats->size_bins[0]++;
+               else if (s < 1519)
+                       stats->size_bins[6]++;
+               else
+                       stats->size_bins[7]++;
+       }
+
+       ea = rte_pktmbuf_mtod(mbuf, struct rte_ether_addr *);
+       if (rte_is_multicast_ether_addr(ea)) {
+               if (rte_is_broadcast_ether_addr(ea))
+                       stats->broadcast++;
+               else
+                       stats->multicast++;
+       }
+}
+
+static inline void
+virtio_rx_stats_updated(struct virtnet_rx *rxvq, struct rte_mbuf *m)
+{
+       VIRTIO_DUMP_PACKET(m, m->data_len);
+
+       virtio_update_packet_stats(&rxvq->stats, m);
 }
 
 static uint16_t
@@ -118,13 +143,15 @@ virtqueue_dequeue_burst_rx_packed(struct virtqueue *vq,
        struct vring_packed_desc *desc;
        uint16_t i;
 
-       desc = vq->vq_packed.ring.desc_packed;
+       desc = vq->vq_packed.ring.desc;
 
        for (i = 0; i < num; i++) {
                used_idx = vq->vq_used_cons_idx;
+               /* desc_is_used has a load-acquire or rte_io_rmb inside
+                * and wait for used desc in virtqueue.
+                */
                if (!desc_is_used(&desc[used_idx], vq))
                        return i;
-               virtio_rmb(vq->hw->weak_barriers);
                len[i] = desc[used_idx].len;
                id = desc[used_idx].id;
                cookie = (struct rte_mbuf *)vq->vq_descx[id].cookie;
@@ -220,126 +247,6 @@ virtqueue_dequeue_rx_inorder(struct virtqueue *vq,
        return i;
 }
 
-#ifndef DEFAULT_TX_FREE_THRESH
-#define DEFAULT_TX_FREE_THRESH 32
-#endif
-
-static void
-virtio_xmit_cleanup_inorder_packed(struct virtqueue *vq, int num)
-{
-       uint16_t used_idx, id, curr_id, free_cnt = 0;
-       uint16_t size = vq->vq_nentries;
-       struct vring_packed_desc *desc = vq->vq_packed.ring.desc_packed;
-       struct vq_desc_extra *dxp;
-
-       used_idx = vq->vq_used_cons_idx;
-       while (num > 0 && desc_is_used(&desc[used_idx], vq)) {
-               virtio_rmb(vq->hw->weak_barriers);
-               id = desc[used_idx].id;
-               do {
-                       curr_id = used_idx;
-                       dxp = &vq->vq_descx[used_idx];
-                       used_idx += dxp->ndescs;
-                       free_cnt += dxp->ndescs;
-                       num -= dxp->ndescs;
-                       if (used_idx >= size) {
-                               used_idx -= size;
-                               vq->vq_packed.used_wrap_counter ^= 1;
-                       }
-                       if (dxp->cookie != NULL) {
-                               rte_pktmbuf_free(dxp->cookie);
-                               dxp->cookie = NULL;
-                       }
-               } while (curr_id != id);
-       }
-       vq->vq_used_cons_idx = used_idx;
-       vq->vq_free_cnt += free_cnt;
-}
-
-static void
-virtio_xmit_cleanup_normal_packed(struct virtqueue *vq, int num)
-{
-       uint16_t used_idx, id;
-       uint16_t size = vq->vq_nentries;
-       struct vring_packed_desc *desc = vq->vq_packed.ring.desc_packed;
-       struct vq_desc_extra *dxp;
-
-       used_idx = vq->vq_used_cons_idx;
-       while (num-- && desc_is_used(&desc[used_idx], vq)) {
-               virtio_rmb(vq->hw->weak_barriers);
-               id = desc[used_idx].id;
-               dxp = &vq->vq_descx[id];
-               vq->vq_used_cons_idx += dxp->ndescs;
-               if (vq->vq_used_cons_idx >= size) {
-                       vq->vq_used_cons_idx -= size;
-                       vq->vq_packed.used_wrap_counter ^= 1;
-               }
-               vq_ring_free_id_packed(vq, id);
-               if (dxp->cookie != NULL) {
-                       rte_pktmbuf_free(dxp->cookie);
-                       dxp->cookie = NULL;
-               }
-               used_idx = vq->vq_used_cons_idx;
-       }
-}
-
-/* Cleanup from completed transmits. */
-static inline void
-virtio_xmit_cleanup_packed(struct virtqueue *vq, int num, int in_order)
-{
-       if (in_order)
-               virtio_xmit_cleanup_inorder_packed(vq, num);
-       else
-               virtio_xmit_cleanup_normal_packed(vq, num);
-}
-
-static void
-virtio_xmit_cleanup(struct virtqueue *vq, uint16_t num)
-{
-       uint16_t i, used_idx, desc_idx;
-       for (i = 0; i < num; i++) {
-               struct vring_used_elem *uep;
-               struct vq_desc_extra *dxp;
-
-               used_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
-               uep = &vq->vq_split.ring.used->ring[used_idx];
-
-               desc_idx = (uint16_t) uep->id;
-               dxp = &vq->vq_descx[desc_idx];
-               vq->vq_used_cons_idx++;
-               vq_ring_free_chain(vq, desc_idx);
-
-               if (dxp->cookie != NULL) {
-                       rte_pktmbuf_free(dxp->cookie);
-                       dxp->cookie = NULL;
-               }
-       }
-}
-
-/* Cleanup from completed inorder transmits. */
-static void
-virtio_xmit_cleanup_inorder(struct virtqueue *vq, uint16_t num)
-{
-       uint16_t i, idx = vq->vq_used_cons_idx;
-       int16_t free_cnt = 0;
-       struct vq_desc_extra *dxp = NULL;
-
-       if (unlikely(num == 0))
-               return;
-
-       for (i = 0; i < num; i++) {
-               dxp = &vq->vq_descx[idx++ & (vq->vq_nentries - 1)];
-               free_cnt += dxp->ndescs;
-               if (dxp->cookie != NULL) {
-                       rte_pktmbuf_free(dxp->cookie);
-                       dxp->cookie = NULL;
-               }
-       }
-
-       vq->vq_free_cnt += free_cnt;
-       vq->vq_used_cons_idx = idx;
-}
-
 static inline int
 virtqueue_enqueue_refill_inorder(struct virtqueue *vq,
                        struct rte_mbuf **cookies,
@@ -430,7 +337,7 @@ static inline int
 virtqueue_enqueue_recv_refill_packed(struct virtqueue *vq,
                                     struct rte_mbuf **cookie, uint16_t num)
 {
-       struct vring_packed_desc *start_dp = vq->vq_packed.ring.desc_packed;
+       struct vring_packed_desc *start_dp = vq->vq_packed.ring.desc;
        uint16_t flags = vq->vq_packed.cached_flags;
        struct virtio_hw *hw = vq->hw;
        struct vq_desc_extra *dxp;
@@ -456,12 +363,14 @@ virtqueue_enqueue_recv_refill_packed(struct virtqueue *vq,
                vq->vq_desc_head_idx = dxp->next;
                if (vq->vq_desc_head_idx == VQ_RING_DESC_CHAIN_END)
                        vq->vq_desc_tail_idx = vq->vq_desc_head_idx;
-               virtio_wmb(hw->weak_barriers);
-               start_dp[idx].flags = flags;
+
+               virtqueue_store_flags_packed(&start_dp[idx], flags,
+                                            hw->weak_barriers);
+
                if (++vq->vq_avail_idx >= vq->vq_nentries) {
                        vq->vq_avail_idx -= vq->vq_nentries;
                        vq->vq_packed.cached_flags ^=
-                               VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
+                               VRING_PACKED_DESC_F_AVAIL_USED;
                        flags = vq->vq_packed.cached_flags;
                }
        }
@@ -479,13 +388,14 @@ virtio_tso_fix_cksum(struct rte_mbuf *m)
        /* common case: header is not fragmented */
        if (likely(rte_pktmbuf_data_len(m) >= m->l2_len + m->l3_len +
                        m->l4_len)) {
-               struct ipv4_hdr *iph;
-               struct ipv6_hdr *ip6h;
-               struct tcp_hdr *th;
+               struct rte_ipv4_hdr *iph;
+               struct rte_ipv6_hdr *ip6h;
+               struct rte_tcp_hdr *th;
                uint16_t prev_cksum, new_cksum, ip_len, ip_paylen;
                uint32_t tmp;
 
-               iph = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, m->l2_len);
+               iph = rte_pktmbuf_mtod_offset(m,
+                                       struct rte_ipv4_hdr *, m->l2_len);
                th = RTE_PTR_ADD(iph, m->l3_len);
                if ((iph->version_ihl >> 4) == 4) {
                        iph->hdr_checksum = 0;
@@ -494,7 +404,7 @@ virtio_tso_fix_cksum(struct rte_mbuf *m)
                        ip_paylen = rte_cpu_to_be_16(rte_be_to_cpu_16(ip_len) -
                                m->l3_len);
                } else {
-                       ip6h = (struct ipv6_hdr *)iph;
+                       ip6h = (struct rte_ipv6_hdr *)iph;
                        ip_paylen = ip6h->payload_len;
                }
 
@@ -511,69 +421,7 @@ virtio_tso_fix_cksum(struct rte_mbuf *m)
 }
 
 
-/* avoid write operation when necessary, to lessen cache issues */
-#define ASSIGN_UNLESS_EQUAL(var, val) do {     \
-       if ((var) != (val))                     \
-               (var) = (val);                  \
-} while (0)
-
-#define virtqueue_clear_net_hdr(_hdr) do {             \
-       ASSIGN_UNLESS_EQUAL((_hdr)->csum_start, 0);     \
-       ASSIGN_UNLESS_EQUAL((_hdr)->csum_offset, 0);    \
-       ASSIGN_UNLESS_EQUAL((_hdr)->flags, 0);          \
-       ASSIGN_UNLESS_EQUAL((_hdr)->gso_type, 0);       \
-       ASSIGN_UNLESS_EQUAL((_hdr)->gso_size, 0);       \
-       ASSIGN_UNLESS_EQUAL((_hdr)->hdr_len, 0);        \
-} while (0)
-
-static inline void
-virtqueue_xmit_offload(struct virtio_net_hdr *hdr,
-                       struct rte_mbuf *cookie,
-                       bool offload)
-{
-       if (offload) {
-               if (cookie->ol_flags & PKT_TX_TCP_SEG)
-                       cookie->ol_flags |= PKT_TX_TCP_CKSUM;
-
-               switch (cookie->ol_flags & PKT_TX_L4_MASK) {
-               case PKT_TX_UDP_CKSUM:
-                       hdr->csum_start = cookie->l2_len + cookie->l3_len;
-                       hdr->csum_offset = offsetof(struct udp_hdr,
-                               dgram_cksum);
-                       hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
-                       break;
-
-               case PKT_TX_TCP_CKSUM:
-                       hdr->csum_start = cookie->l2_len + cookie->l3_len;
-                       hdr->csum_offset = offsetof(struct tcp_hdr, cksum);
-                       hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
-                       break;
 
-               default:
-                       ASSIGN_UNLESS_EQUAL(hdr->csum_start, 0);
-                       ASSIGN_UNLESS_EQUAL(hdr->csum_offset, 0);
-                       ASSIGN_UNLESS_EQUAL(hdr->flags, 0);
-                       break;
-               }
-
-               /* TCP Segmentation Offload */
-               if (cookie->ol_flags & PKT_TX_TCP_SEG) {
-                       virtio_tso_fix_cksum(cookie);
-                       hdr->gso_type = (cookie->ol_flags & PKT_TX_IPV6) ?
-                               VIRTIO_NET_HDR_GSO_TCPV6 :
-                               VIRTIO_NET_HDR_GSO_TCPV4;
-                       hdr->gso_size = cookie->tso_segsz;
-                       hdr->hdr_len =
-                               cookie->l2_len +
-                               cookie->l3_len +
-                               cookie->l4_len;
-               } else {
-                       ASSIGN_UNLESS_EQUAL(hdr->gso_type, 0);
-                       ASSIGN_UNLESS_EQUAL(hdr->gso_size, 0);
-                       ASSIGN_UNLESS_EQUAL(hdr->hdr_len, 0);
-               }
-       }
-}
 
 static inline void
 virtqueue_enqueue_xmit_inorder(struct virtnet_tx *txvq,
@@ -585,7 +433,7 @@ virtqueue_enqueue_xmit_inorder(struct virtnet_tx *txvq,
        struct vring_desc *start_dp;
        struct virtio_net_hdr *hdr;
        uint16_t idx;
-       uint16_t head_size = vq->hw->vtnet_hdr_size;
+       int16_t head_size = vq->hw->vtnet_hdr_size;
        uint16_t i = 0;
 
        idx = vq->vq_desc_head_idx;
@@ -596,10 +444,10 @@ virtqueue_enqueue_xmit_inorder(struct virtnet_tx *txvq,
                dxp = &vq->vq_descx[vq->vq_avail_idx & (vq->vq_nentries - 1)];
                dxp->cookie = (void *)cookies[i];
                dxp->ndescs = 1;
+               virtio_update_packet_stats(&txvq->stats, cookies[i]);
 
-               hdr = (struct virtio_net_hdr *)
-                       rte_pktmbuf_prepend(cookies[i], head_size);
-               cookies[i]->pkt_len -= head_size;
+               hdr = rte_pktmbuf_mtod_offset(cookies[i],
+                               struct virtio_net_hdr *, -head_size);
 
                /* if offload disabled, hdr is not zeroed yet, do it now */
                if (!vq->hw->has_tx_offload)
@@ -607,10 +455,12 @@ virtqueue_enqueue_xmit_inorder(struct virtnet_tx *txvq,
                else
                        virtqueue_xmit_offload(hdr, cookies[i], true);
 
-               start_dp[idx].addr  = VIRTIO_MBUF_DATA_DMA_ADDR(cookies[i], vq);
-               start_dp[idx].len   = cookies[i]->data_len;
+               start_dp[idx].addr  =
+                       VIRTIO_MBUF_DATA_DMA_ADDR(cookies[i], vq) - head_size;
+               start_dp[idx].len   = cookies[i]->data_len + head_size;
                start_dp[idx].flags = 0;
 
+
                vq_update_avail_ring(vq, idx);
 
                idx++;
@@ -630,12 +480,12 @@ virtqueue_enqueue_xmit_packed_fast(struct virtnet_tx *txvq,
        struct vring_packed_desc *dp;
        struct vq_desc_extra *dxp;
        uint16_t idx, id, flags;
-       uint16_t head_size = vq->hw->vtnet_hdr_size;
+       int16_t head_size = vq->hw->vtnet_hdr_size;
        struct virtio_net_hdr *hdr;
 
        id = in_order ? vq->vq_avail_idx : vq->vq_desc_head_idx;
        idx = vq->vq_avail_idx;
-       dp = &vq->vq_packed.ring.desc_packed[idx];
+       dp = &vq->vq_packed.ring.desc[idx];
 
        dxp = &vq->vq_descx[id];
        dxp->ndescs = 1;
@@ -644,9 +494,8 @@ virtqueue_enqueue_xmit_packed_fast(struct virtnet_tx *txvq,
        flags = vq->vq_packed.cached_flags;
 
        /* prepend cannot fail, checked by caller */
-       hdr = (struct virtio_net_hdr *)
-               rte_pktmbuf_prepend(cookie, head_size);
-       cookie->pkt_len -= head_size;
+       hdr = rte_pktmbuf_mtod_offset(cookie, struct virtio_net_hdr *,
+                                     -head_size);
 
        /* if offload disabled, hdr is not zeroed yet, do it now */
        if (!vq->hw->has_tx_offload)
@@ -654,14 +503,13 @@ virtqueue_enqueue_xmit_packed_fast(struct virtnet_tx *txvq,
        else
                virtqueue_xmit_offload(hdr, cookie, true);
 
-       dp->addr = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq);
-       dp->len  = cookie->data_len;
+       dp->addr = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq) - head_size;
+       dp->len  = cookie->data_len + head_size;
        dp->id   = id;
 
        if (++vq->vq_avail_idx >= vq->vq_nentries) {
                vq->vq_avail_idx -= vq->vq_nentries;
-               vq->vq_packed.cached_flags ^=
-                       VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
+               vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
        }
 
        vq->vq_free_cnt--;
@@ -672,100 +520,7 @@ virtqueue_enqueue_xmit_packed_fast(struct virtnet_tx *txvq,
                        vq->vq_desc_tail_idx = VQ_RING_DESC_CHAIN_END;
        }
 
-       virtio_wmb(vq->hw->weak_barriers);
-       dp->flags = flags;
-}
-
-static inline void
-virtqueue_enqueue_xmit_packed(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
-                             uint16_t needed, int can_push, int in_order)
-{
-       struct virtio_tx_region *txr = txvq->virtio_net_hdr_mz->addr;
-       struct vq_desc_extra *dxp;
-       struct virtqueue *vq = txvq->vq;
-       struct vring_packed_desc *start_dp, *head_dp;
-       uint16_t idx, id, head_idx, head_flags;
-       uint16_t head_size = vq->hw->vtnet_hdr_size;
-       struct virtio_net_hdr *hdr;
-       uint16_t prev;
-
-       id = in_order ? vq->vq_avail_idx : vq->vq_desc_head_idx;
-
-       dxp = &vq->vq_descx[id];
-       dxp->ndescs = needed;
-       dxp->cookie = cookie;
-
-       head_idx = vq->vq_avail_idx;
-       idx = head_idx;
-       prev = head_idx;
-       start_dp = vq->vq_packed.ring.desc_packed;
-
-       head_dp = &vq->vq_packed.ring.desc_packed[idx];
-       head_flags = cookie->next ? VRING_DESC_F_NEXT : 0;
-       head_flags |= vq->vq_packed.cached_flags;
-
-       if (can_push) {
-               /* prepend cannot fail, checked by caller */
-               hdr = (struct virtio_net_hdr *)
-                       rte_pktmbuf_prepend(cookie, head_size);
-               /* rte_pktmbuf_prepend() counts the hdr size to the pkt length,
-                * which is wrong. Below subtract restores correct pkt size.
-                */
-               cookie->pkt_len -= head_size;
-
-               /* if offload disabled, it is not zeroed below, do it now */
-               if (!vq->hw->has_tx_offload)
-                       virtqueue_clear_net_hdr(hdr);
-       } else {
-               /* setup first tx ring slot to point to header
-                * stored in reserved region.
-                */
-               start_dp[idx].addr  = txvq->virtio_net_hdr_mem +
-                       RTE_PTR_DIFF(&txr[idx].tx_hdr, txr);
-               start_dp[idx].len   = vq->hw->vtnet_hdr_size;
-               hdr = (struct virtio_net_hdr *)&txr[idx].tx_hdr;
-               idx++;
-               if (idx >= vq->vq_nentries) {
-                       idx -= vq->vq_nentries;
-                       vq->vq_packed.cached_flags ^=
-                               VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
-               }
-       }
-
-       virtqueue_xmit_offload(hdr, cookie, vq->hw->has_tx_offload);
-
-       do {
-               uint16_t flags;
-
-               start_dp[idx].addr = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq);
-               start_dp[idx].len  = cookie->data_len;
-               if (likely(idx != head_idx)) {
-                       flags = cookie->next ? VRING_DESC_F_NEXT : 0;
-                       flags |= vq->vq_packed.cached_flags;
-                       start_dp[idx].flags = flags;
-               }
-               prev = idx;
-               idx++;
-               if (idx >= vq->vq_nentries) {
-                       idx -= vq->vq_nentries;
-                       vq->vq_packed.cached_flags ^=
-                               VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
-               }
-       } while ((cookie = cookie->next) != NULL);
-
-       start_dp[prev].id = id;
-
-       vq->vq_free_cnt = (uint16_t)(vq->vq_free_cnt - needed);
-       vq->vq_avail_idx = idx;
-
-       if (!in_order) {
-               vq->vq_desc_head_idx = dxp->next;
-               if (vq->vq_desc_head_idx == VQ_RING_DESC_CHAIN_END)
-                       vq->vq_desc_tail_idx = VQ_RING_DESC_CHAIN_END;
-       }
-
-       virtio_wmb(vq->hw->weak_barriers);
-       head_dp->flags = head_flags;
+       virtqueue_store_flags_packed(dp, flags, vq->hw->weak_barriers);
 }
 
 static inline void
@@ -779,7 +534,8 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
        struct vring_desc *start_dp;
        uint16_t seg_num = cookie->nb_segs;
        uint16_t head_idx, idx;
-       uint16_t head_size = vq->hw->vtnet_hdr_size;
+       int16_t head_size = vq->hw->vtnet_hdr_size;
+       bool prepend_header = false;
        struct virtio_net_hdr *hdr;
 
        head_idx = vq->vq_desc_head_idx;
@@ -795,12 +551,9 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
 
        if (can_push) {
                /* prepend cannot fail, checked by caller */
-               hdr = (struct virtio_net_hdr *)
-                       rte_pktmbuf_prepend(cookie, head_size);
-               /* rte_pktmbuf_prepend() counts the hdr size to the pkt length,
-                * which is wrong. Below subtract restores correct pkt size.
-                */
-               cookie->pkt_len -= head_size;
+               hdr = rte_pktmbuf_mtod_offset(cookie, struct virtio_net_hdr *,
+                                             -head_size);
+               prepend_header = true;
 
                /* if offload disabled, it is not zeroed below, do it now */
                if (!vq->hw->has_tx_offload)
@@ -839,6 +592,11 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
        do {
                start_dp[idx].addr  = VIRTIO_MBUF_DATA_DMA_ADDR(cookie, vq);
                start_dp[idx].len   = cookie->data_len;
+               if (prepend_header) {
+                       start_dp[idx].addr -= head_size;
+                       start_dp[idx].len += head_size;
+                       prepend_header = false;
+               }
                start_dp[idx].flags = cookie->next ? VRING_DESC_F_NEXT : 0;
                idx = start_dp[idx].next;
        } while ((cookie = cookie->next) != NULL);
@@ -873,16 +631,44 @@ virtio_dev_rx_queue_setup(struct rte_eth_dev *dev,
                        uint16_t queue_idx,
                        uint16_t nb_desc,
                        unsigned int socket_id __rte_unused,
-                       const struct rte_eth_rxconf *rx_conf __rte_unused,
+                       const struct rte_eth_rxconf *rx_conf,
                        struct rte_mempool *mp)
 {
        uint16_t vtpci_queue_idx = 2 * queue_idx + VTNET_SQ_RQ_QUEUE_IDX;
        struct virtio_hw *hw = dev->data->dev_private;
        struct virtqueue *vq = hw->vqs[vtpci_queue_idx];
        struct virtnet_rx *rxvq;
+       uint16_t rx_free_thresh;
 
        PMD_INIT_FUNC_TRACE();
 
+       if (rx_conf->rx_deferred_start) {
+               PMD_INIT_LOG(ERR, "Rx deferred start is not supported");
+               return -EINVAL;
+       }
+
+       rx_free_thresh = rx_conf->rx_free_thresh;
+       if (rx_free_thresh == 0)
+               rx_free_thresh =
+                       RTE_MIN(vq->vq_nentries / 4, DEFAULT_RX_FREE_THRESH);
+
+       if (rx_free_thresh & 0x3) {
+               RTE_LOG(ERR, PMD, "rx_free_thresh must be multiples of four."
+                       " (rx_free_thresh=%u port=%u queue=%u)\n",
+                       rx_free_thresh, dev->data->port_id, queue_idx);
+               return -EINVAL;
+       }
+
+       if (rx_free_thresh >= vq->vq_nentries) {
+               RTE_LOG(ERR, PMD, "rx_free_thresh must be less than the "
+                       "number of RX entries (%u)."
+                       " (rx_free_thresh=%u port=%u queue=%u)\n",
+                       vq->vq_nentries,
+                       rx_free_thresh, dev->data->port_id, queue_idx);
+               return -EINVAL;
+       }
+       vq->vq_free_thresh = rx_free_thresh;
+
        if (nb_desc == 0 || nb_desc > vq->vq_nentries)
                nb_desc = vq->vq_nentries;
        vq->vq_free_cnt = RTE_MIN(vq->vq_free_cnt, nb_desc);
@@ -890,11 +676,6 @@ virtio_dev_rx_queue_setup(struct rte_eth_dev *dev,
        rxvq = &vq->rxq;
        rxvq->queue_id = queue_idx;
        rxvq->mpool = mp;
-       if (rxvq->mpool == NULL) {
-               rte_exit(EXIT_FAILURE,
-                       "Cannot allocate mbufs for rx virtqueue");
-       }
-
        dev->data->rx_queues[queue_idx] = rxvq;
 
        return 0;
@@ -910,13 +691,14 @@ virtio_dev_rx_queue_setup_finish(struct rte_eth_dev *dev, uint16_t queue_idx)
        struct rte_mbuf *m;
        uint16_t desc_idx;
        int error, nbufs, i;
+       bool in_order = vtpci_with_feature(hw, VIRTIO_F_IN_ORDER);
 
        PMD_INIT_FUNC_TRACE();
 
        /* Allocate blank mbufs for the each rx descriptor */
        nbufs = 0;
 
-       if (hw->use_simple_rx) {
+       if (hw->use_vec_rx && !vtpci_packed_queue(hw)) {
                for (desc_idx = 0; desc_idx < vq->vq_nentries;
                     desc_idx++) {
                        vq->vq_split.ring.avail->ring[desc_idx] = desc_idx;
@@ -934,12 +716,12 @@ virtio_dev_rx_queue_setup_finish(struct rte_eth_dev *dev, uint16_t queue_idx)
                        &rxvq->fake_mbuf;
        }
 
-       if (hw->use_simple_rx) {
+       if (hw->use_vec_rx && !vtpci_packed_queue(hw)) {
                while (vq->vq_free_cnt >= RTE_VIRTIO_VPMD_RX_REARM_THRESH) {
                        virtio_rxq_rearm_vec(rxvq);
                        nbufs += RTE_VIRTIO_VPMD_RX_REARM_THRESH;
                }
-       } else if (hw->use_inorder_rx) {
+       } else if (!vtpci_packed_queue(vq->hw) && in_order) {
                if ((!virtqueue_full(vq))) {
                        uint16_t free_cnt = vq->vq_free_cnt;
                        struct rte_mbuf *pkts[free_cnt];
@@ -1011,6 +793,11 @@ virtio_dev_tx_queue_setup(struct rte_eth_dev *dev,
 
        PMD_INIT_FUNC_TRACE();
 
+       if (tx_conf->tx_deferred_start) {
+               PMD_INIT_LOG(ERR, "Tx deferred start is not supported");
+               return -EINVAL;
+       }
+
        if (nb_desc == 0 || nb_desc > vq->vq_nentries)
                nb_desc = vq->vq_nentries;
        vq->vq_free_cnt = RTE_MIN(vq->vq_free_cnt, nb_desc);
@@ -1024,7 +811,7 @@ virtio_dev_tx_queue_setup(struct rte_eth_dev *dev,
                        RTE_MIN(vq->vq_nentries / 4, DEFAULT_TX_FREE_THRESH);
 
        if (tx_free_thresh >= (vq->vq_nentries - 3)) {
-               RTE_LOG(ERR, PMD, "tx_free_thresh must be less than the "
+               PMD_DRV_LOG(ERR, "tx_free_thresh must be less than the "
                        "number of TX entries minus 3 (%u)."
                        " (tx_free_thresh=%u port=%u queue=%u)\n",
                        vq->vq_nentries - 3,
@@ -1049,7 +836,7 @@ virtio_dev_tx_queue_setup_finish(struct rte_eth_dev *dev,
        PMD_INIT_FUNC_TRACE();
 
        if (!vtpci_packed_queue(hw)) {
-               if (hw->use_inorder_tx)
+               if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER))
                        vq->vq_split.ring.desc[vq->vq_nentries - 1].next = 0;
        }
 
@@ -1072,7 +859,7 @@ virtio_discard_rxbuf(struct virtqueue *vq, struct rte_mbuf *m)
                error = virtqueue_enqueue_recv_refill(vq, &m, 1);
 
        if (unlikely(error)) {
-               RTE_LOG(ERR, PMD, "cannot requeue discarded mbuf");
+               PMD_DRV_LOG(ERR, "cannot requeue discarded mbuf");
                rte_pktmbuf_free(m);
        }
 }
@@ -1084,53 +871,11 @@ virtio_discard_rxbuf_inorder(struct virtqueue *vq, struct rte_mbuf *m)
 
        error = virtqueue_enqueue_refill_inorder(vq, &m, 1);
        if (unlikely(error)) {
-               RTE_LOG(ERR, PMD, "cannot requeue discarded mbuf");
+               PMD_DRV_LOG(ERR, "cannot requeue discarded mbuf");
                rte_pktmbuf_free(m);
        }
 }
 
-static inline void
-virtio_update_packet_stats(struct virtnet_stats *stats, struct rte_mbuf *mbuf)
-{
-       uint32_t s = mbuf->pkt_len;
-       struct ether_addr *ea;
-
-       stats->bytes += s;
-
-       if (s == 64) {
-               stats->size_bins[1]++;
-       } else if (s > 64 && s < 1024) {
-               uint32_t bin;
-
-               /* count zeros, and offset into correct bin */
-               bin = (sizeof(s) * 8) - __builtin_clz(s) - 5;
-               stats->size_bins[bin]++;
-       } else {
-               if (s < 64)
-                       stats->size_bins[0]++;
-               else if (s < 1519)
-                       stats->size_bins[6]++;
-               else if (s >= 1519)
-                       stats->size_bins[7]++;
-       }
-
-       ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
-       if (is_multicast_ether_addr(ea)) {
-               if (is_broadcast_ether_addr(ea))
-                       stats->broadcast++;
-               else
-                       stats->multicast++;
-       }
-}
-
-static inline void
-virtio_rx_stats_updated(struct virtnet_rx *rxvq, struct rte_mbuf *m)
-{
-       VIRTIO_DUMP_PACKET(m, m->data_len);
-
-       virtio_update_packet_stats(&rxvq->stats, m);
-}
-
 /* Optionally fill offload information in structure */
 static inline int
 virtio_rx_offload(struct rte_mbuf *m, struct virtio_net_hdr *hdr)
@@ -1165,9 +910,10 @@ virtio_rx_offload(struct rte_mbuf *m, struct virtio_net_hdr *hdr)
                         */
                        uint16_t csum = 0, off;
 
-                       rte_raw_cksum_mbuf(m, hdr->csum_start,
+                       if (rte_raw_cksum_mbuf(m, hdr->csum_start,
                                rte_pktmbuf_pkt_len(m) - hdr->csum_start,
-                               &csum);
+                               &csum) < 0)
+                               return -EINVAL;
                        if (likely(csum != 0xffff))
                                csum = ~csum;
                        off = hdr->csum_offset + hdr->csum_start;
@@ -1203,7 +949,6 @@ virtio_rx_offload(struct rte_mbuf *m, struct virtio_net_hdr *hdr)
        return 0;
 }
 
-#define VIRTIO_MBUF_BURST_SZ 64
 #define DESC_PER_CACHELINE (RTE_CACHE_LINE_SIZE / sizeof(struct vring_desc))
 uint16_t
 virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
@@ -1211,7 +956,7 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        struct virtnet_rx *rxvq = rx_queue;
        struct virtqueue *vq = rxvq->vq;
        struct virtio_hw *hw = vq->hw;
-       struct rte_mbuf *rxm, *new_mbuf;
+       struct rte_mbuf *rxm;
        uint16_t nb_used, num, nb_rx;
        uint32_t len[VIRTIO_MBUF_BURST_SZ];
        struct rte_mbuf *rcv_pkts[VIRTIO_MBUF_BURST_SZ];
@@ -1224,9 +969,7 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        if (unlikely(hw->started == 0))
                return nb_rx;
 
-       nb_used = VIRTQUEUE_NUSED(vq);
-
-       virtio_rmb(hw->weak_barriers);
+       nb_used = virtqueue_nused(vq);
 
        num = likely(nb_used <= nb_pkts) ? nb_used : nb_pkts;
        if (unlikely(num > VIRTIO_MBUF_BURST_SZ))
@@ -1245,7 +988,7 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
                PMD_RX_LOG(DEBUG, "packet len:%d", len[i]);
 
-               if (unlikely(len[i] < hdr_size + ETHER_HDR_LEN)) {
+               if (unlikely(len[i] < hdr_size + RTE_ETHER_HDR_LEN)) {
                        PMD_RX_LOG(ERR, "Packet drop");
                        nb_enqueued++;
                        virtio_discard_rxbuf(vq, rxm);
@@ -1281,20 +1024,24 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        rxvq->stats.packets += nb_rx;
 
        /* Allocate new mbuf for the used descriptor */
-       while (likely(!virtqueue_full(vq))) {
-               new_mbuf = rte_mbuf_raw_alloc(rxvq->mpool);
-               if (unlikely(new_mbuf == NULL)) {
-                       struct rte_eth_dev *dev
-                               = &rte_eth_devices[rxvq->port_id];
-                       dev->data->rx_mbuf_alloc_failed++;
-                       break;
-               }
-               error = virtqueue_enqueue_recv_refill(vq, &new_mbuf, 1);
-               if (unlikely(error)) {
-                       rte_pktmbuf_free(new_mbuf);
-                       break;
+       if (likely(!virtqueue_full(vq))) {
+               uint16_t free_cnt = vq->vq_free_cnt;
+               struct rte_mbuf *new_pkts[free_cnt];
+
+               if (likely(rte_pktmbuf_alloc_bulk(rxvq->mpool, new_pkts,
+                                               free_cnt) == 0)) {
+                       error = virtqueue_enqueue_recv_refill(vq, new_pkts,
+                                       free_cnt);
+                       if (unlikely(error)) {
+                               for (i = 0; i < free_cnt; i++)
+                                       rte_pktmbuf_free(new_pkts[i]);
+                       }
+                       nb_enqueued += free_cnt;
+               } else {
+                       struct rte_eth_dev *dev =
+                               &rte_eth_devices[rxvq->port_id];
+                       dev->data->rx_mbuf_alloc_failed += free_cnt;
                }
-               nb_enqueued++;
        }
 
        if (likely(nb_enqueued)) {
@@ -1316,7 +1063,7 @@ virtio_recv_pkts_packed(void *rx_queue, struct rte_mbuf **rx_pkts,
        struct virtnet_rx *rxvq = rx_queue;
        struct virtqueue *vq = rxvq->vq;
        struct virtio_hw *hw = vq->hw;
-       struct rte_mbuf *rxm, *new_mbuf;
+       struct rte_mbuf *rxm;
        uint16_t num, nb_rx;
        uint32_t len[VIRTIO_MBUF_BURST_SZ];
        struct rte_mbuf *rcv_pkts[VIRTIO_MBUF_BURST_SZ];
@@ -1344,7 +1091,7 @@ virtio_recv_pkts_packed(void *rx_queue, struct rte_mbuf **rx_pkts,
 
                PMD_RX_LOG(DEBUG, "packet len:%d", len[i]);
 
-               if (unlikely(len[i] < hdr_size + ETHER_HDR_LEN)) {
+               if (unlikely(len[i] < hdr_size + RTE_ETHER_HDR_LEN)) {
                        PMD_RX_LOG(ERR, "Packet drop");
                        nb_enqueued++;
                        virtio_discard_rxbuf(vq, rxm);
@@ -1380,20 +1127,24 @@ virtio_recv_pkts_packed(void *rx_queue, struct rte_mbuf **rx_pkts,
        rxvq->stats.packets += nb_rx;
 
        /* Allocate new mbuf for the used descriptor */
-       while (likely(!virtqueue_full(vq))) {
-               new_mbuf = rte_mbuf_raw_alloc(rxvq->mpool);
-               if (unlikely(new_mbuf == NULL)) {
+       if (likely(!virtqueue_full(vq))) {
+               uint16_t free_cnt = vq->vq_free_cnt;
+               struct rte_mbuf *new_pkts[free_cnt];
+
+               if (likely(rte_pktmbuf_alloc_bulk(rxvq->mpool, new_pkts,
+                                               free_cnt) == 0)) {
+                       error = virtqueue_enqueue_recv_refill_packed(vq,
+                                       new_pkts, free_cnt);
+                       if (unlikely(error)) {
+                               for (i = 0; i < free_cnt; i++)
+                                       rte_pktmbuf_free(new_pkts[i]);
+                       }
+                       nb_enqueued += free_cnt;
+               } else {
                        struct rte_eth_dev *dev =
                                &rte_eth_devices[rxvq->port_id];
-                       dev->data->rx_mbuf_alloc_failed++;
-                       break;
-               }
-               error = virtqueue_enqueue_recv_refill_packed(vq, &new_mbuf, 1);
-               if (unlikely(error)) {
-                       rte_pktmbuf_free(new_mbuf);
-                       break;
+                       dev->data->rx_mbuf_alloc_failed += free_cnt;
                }
-               nb_enqueued++;
        }
 
        if (likely(nb_enqueued)) {
@@ -1416,7 +1167,7 @@ virtio_recv_pkts_inorder(void *rx_queue,
        struct virtqueue *vq = rxvq->vq;
        struct virtio_hw *hw = vq->hw;
        struct rte_mbuf *rxm;
-       struct rte_mbuf *prev;
+       struct rte_mbuf *prev = NULL;
        uint16_t nb_used, num, nb_rx;
        uint32_t len[VIRTIO_MBUF_BURST_SZ];
        struct rte_mbuf *rcv_pkts[VIRTIO_MBUF_BURST_SZ];
@@ -1431,12 +1182,10 @@ virtio_recv_pkts_inorder(void *rx_queue,
        if (unlikely(hw->started == 0))
                return nb_rx;
 
-       nb_used = VIRTQUEUE_NUSED(vq);
+       nb_used = virtqueue_nused(vq);
        nb_used = RTE_MIN(nb_used, nb_pkts);
        nb_used = RTE_MIN(nb_used, VIRTIO_MBUF_BURST_SZ);
 
-       virtio_rmb(hw->weak_barriers);
-
        PMD_RX_LOG(DEBUG, "used:%d", nb_used);
 
        nb_enqueued = 0;
@@ -1454,7 +1203,7 @@ virtio_recv_pkts_inorder(void *rx_queue,
 
                rxm = rcv_pkts[i];
 
-               if (unlikely(len[i] < hdr_size + ETHER_HDR_LEN)) {
+               if (unlikely(len[i] < hdr_size + RTE_ETHER_HDR_LEN)) {
                        PMD_RX_LOG(ERR, "Packet drop");
                        nb_enqueued++;
                        virtio_discard_rxbuf_inorder(vq, rxm);
@@ -1508,11 +1257,8 @@ virtio_recv_pkts_inorder(void *rx_queue,
                        rxm->data_len = (uint16_t)(len[i]);
 
                        rx_pkts[nb_rx]->pkt_len += (uint32_t)(len[i]);
-                       rx_pkts[nb_rx]->data_len += (uint16_t)(len[i]);
-
-                       if (prev)
-                               prev->next = rxm;
 
+                       prev->next = rxm;
                        prev = rxm;
                        seg_res -= 1;
                }
@@ -1528,9 +1274,7 @@ virtio_recv_pkts_inorder(void *rx_queue,
                uint16_t rcv_cnt = RTE_MIN((uint16_t)seg_res,
                                        VIRTIO_MBUF_BURST_SZ);
 
-               prev = rcv_pkts[nb_rx];
-               if (likely(VIRTQUEUE_NUSED(vq) >= rcv_cnt)) {
-                       virtio_rmb(hw->weak_barriers);
+               if (likely(virtqueue_nused(vq) >= rcv_cnt)) {
                        num = virtqueue_dequeue_rx_inorder(vq, rcv_pkts, len,
                                                           rcv_cnt);
                        uint16_t extra_idx = 0;
@@ -1545,7 +1289,6 @@ virtio_recv_pkts_inorder(void *rx_queue,
                                prev->next = rxm;
                                prev = rxm;
                                rx_pkts[nb_rx]->pkt_len += len[extra_idx];
-                               rx_pkts[nb_rx]->data_len += len[extra_idx];
                                extra_idx += 1;
                        };
                        seg_res -= rcv_cnt;
@@ -1557,7 +1300,7 @@ virtio_recv_pkts_inorder(void *rx_queue,
                } else {
                        PMD_RX_LOG(ERR,
                                        "No enough segments for packet.");
-                       virtio_discard_rxbuf_inorder(vq, prev);
+                       rte_pktmbuf_free(rx_pkts[nb_rx]);
                        rxvq->stats.errors++;
                        break;
                }
@@ -1608,7 +1351,7 @@ virtio_recv_mergeable_pkts(void *rx_queue,
        struct virtqueue *vq = rxvq->vq;
        struct virtio_hw *hw = vq->hw;
        struct rte_mbuf *rxm;
-       struct rte_mbuf *prev;
+       struct rte_mbuf *prev = NULL;
        uint16_t nb_used, num, nb_rx = 0;
        uint32_t len[VIRTIO_MBUF_BURST_SZ];
        struct rte_mbuf *rcv_pkts[VIRTIO_MBUF_BURST_SZ];
@@ -1622,9 +1365,7 @@ virtio_recv_mergeable_pkts(void *rx_queue,
        if (unlikely(hw->started == 0))
                return nb_rx;
 
-       nb_used = VIRTQUEUE_NUSED(vq);
-
-       virtio_rmb(hw->weak_barriers);
+       nb_used = virtqueue_nused(vq);
 
        PMD_RX_LOG(DEBUG, "used:%d", nb_used);
 
@@ -1646,7 +1387,7 @@ virtio_recv_mergeable_pkts(void *rx_queue,
 
                rxm = rcv_pkts[i];
 
-               if (unlikely(len[i] < hdr_size + ETHER_HDR_LEN)) {
+               if (unlikely(len[i] < hdr_size + RTE_ETHER_HDR_LEN)) {
                        PMD_RX_LOG(ERR, "Packet drop");
                        nb_enqueued++;
                        virtio_discard_rxbuf(vq, rxm);
@@ -1695,11 +1436,8 @@ virtio_recv_mergeable_pkts(void *rx_queue,
                        rxm->data_len = (uint16_t)(len[i]);
 
                        rx_pkts[nb_rx]->pkt_len += (uint32_t)(len[i]);
-                       rx_pkts[nb_rx]->data_len += (uint16_t)(len[i]);
-
-                       if (prev)
-                               prev->next = rxm;
 
+                       prev->next = rxm;
                        prev = rxm;
                        seg_res -= 1;
                }
@@ -1715,9 +1453,7 @@ virtio_recv_mergeable_pkts(void *rx_queue,
                uint16_t rcv_cnt = RTE_MIN((uint16_t)seg_res,
                                        VIRTIO_MBUF_BURST_SZ);
 
-               prev = rcv_pkts[nb_rx];
-               if (likely(VIRTQUEUE_NUSED(vq) >= rcv_cnt)) {
-                       virtio_rmb(hw->weak_barriers);
+               if (likely(virtqueue_nused(vq) >= rcv_cnt)) {
                        num = virtqueue_dequeue_burst_rx(vq, rcv_pkts, len,
                                                           rcv_cnt);
                        uint16_t extra_idx = 0;
@@ -1732,7 +1468,6 @@ virtio_recv_mergeable_pkts(void *rx_queue,
                                prev->next = rxm;
                                prev = rxm;
                                rx_pkts[nb_rx]->pkt_len += len[extra_idx];
-                               rx_pkts[nb_rx]->data_len += len[extra_idx];
                                extra_idx += 1;
                        };
                        seg_res -= rcv_cnt;
@@ -1744,7 +1479,7 @@ virtio_recv_mergeable_pkts(void *rx_queue,
                } else {
                        PMD_RX_LOG(ERR,
                                        "No enough segments for packet.");
-                       virtio_discard_rxbuf(vq, prev);
+                       rte_pktmbuf_free(rx_pkts[nb_rx]);
                        rxvq->stats.errors++;
                        break;
                }
@@ -1825,7 +1560,7 @@ virtio_recv_mergeable_pkts_packed(void *rx_queue,
 
                rxm = rcv_pkts[i];
 
-               if (unlikely(len[i] < hdr_size + ETHER_HDR_LEN)) {
+               if (unlikely(len[i] < hdr_size + RTE_ETHER_HDR_LEN)) {
                        PMD_RX_LOG(ERR, "Packet drop");
                        nb_enqueued++;
                        virtio_discard_rxbuf(vq, rxm);
@@ -1873,11 +1608,8 @@ virtio_recv_mergeable_pkts_packed(void *rx_queue,
                        rxm->data_len = (uint16_t)(len[i]);
 
                        rx_pkts[nb_rx]->pkt_len += (uint32_t)(len[i]);
-                       rx_pkts[nb_rx]->data_len += (uint16_t)(len[i]);
-
-                       if (prev)
-                               prev->next = rxm;
 
+                       prev->next = rxm;
                        prev = rxm;
                        seg_res -= 1;
                }
@@ -1892,39 +1624,33 @@ virtio_recv_mergeable_pkts_packed(void *rx_queue,
        while (seg_res != 0) {
                uint16_t rcv_cnt = RTE_MIN((uint16_t)seg_res,
                                        VIRTIO_MBUF_BURST_SZ);
-               if (likely(vq->vq_free_cnt >= rcv_cnt)) {
-                       num = virtqueue_dequeue_burst_rx_packed(vq, rcv_pkts,
-                                       len, rcv_cnt);
-                       uint16_t extra_idx = 0;
+               uint16_t extra_idx = 0;
 
-                       rcv_cnt = num;
+               rcv_cnt = virtqueue_dequeue_burst_rx_packed(vq, rcv_pkts,
+                               len, rcv_cnt);
+               if (unlikely(rcv_cnt == 0)) {
+                       PMD_RX_LOG(ERR, "No enough segments for packet.");
+                       rte_pktmbuf_free(rx_pkts[nb_rx]);
+                       rxvq->stats.errors++;
+                       break;
+               }
 
-                       while (extra_idx < rcv_cnt) {
-                               rxm = rcv_pkts[extra_idx];
+               while (extra_idx < rcv_cnt) {
+                       rxm = rcv_pkts[extra_idx];
 
-                               rxm->data_off =
-                                       RTE_PKTMBUF_HEADROOM - hdr_size;
-                               rxm->pkt_len = (uint32_t)(len[extra_idx]);
-                               rxm->data_len = (uint16_t)(len[extra_idx]);
+                       rxm->data_off = RTE_PKTMBUF_HEADROOM - hdr_size;
+                       rxm->pkt_len = (uint32_t)(len[extra_idx]);
+                       rxm->data_len = (uint16_t)(len[extra_idx]);
 
-                               prev->next = rxm;
-                               prev = rxm;
-                               rx_pkts[nb_rx]->pkt_len += len[extra_idx];
-                               rx_pkts[nb_rx]->data_len += len[extra_idx];
-                               extra_idx += 1;
-                       }
-                       seg_res -= rcv_cnt;
-                       if (!seg_res) {
-                               virtio_rx_stats_updated(rxvq, rx_pkts[nb_rx]);
-                               nb_rx++;
-                       }
-               } else {
-                       PMD_RX_LOG(ERR,
-                                       "No enough segments for packet.");
-                       if (prev)
-                               virtio_discard_rxbuf(vq, prev);
-                       rxvq->stats.errors++;
-                       break;
+                       prev->next = rxm;
+                       prev = rxm;
+                       rx_pkts[nb_rx]->pkt_len += len[extra_idx];
+                       extra_idx += 1;
+               }
+               seg_res -= rcv_cnt;
+               if (!seg_res) {
+                       virtio_rx_stats_updated(rxvq, rx_pkts[nb_rx]);
+                       nb_rx++;
                }
        }
 
@@ -1961,6 +1687,51 @@ virtio_recv_mergeable_pkts_packed(void *rx_queue,
        return nb_rx;
 }
 
+uint16_t
+virtio_xmit_pkts_prepare(void *tx_queue __rte_unused, struct rte_mbuf **tx_pkts,
+                       uint16_t nb_pkts)
+{
+       uint16_t nb_tx;
+       int error;
+
+       for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
+               struct rte_mbuf *m = tx_pkts[nb_tx];
+
+#ifdef RTE_LIBRTE_ETHDEV_DEBUG
+               error = rte_validate_tx_offload(m);
+               if (unlikely(error)) {
+                       rte_errno = -error;
+                       break;
+               }
+#endif
+
+               /* Do VLAN tag insertion */
+               if (unlikely(m->ol_flags & PKT_TX_VLAN_PKT)) {
+                       error = rte_vlan_insert(&m);
+                       /* rte_vlan_insert() may change pointer
+                        * even in the case of failure
+                        */
+                       tx_pkts[nb_tx] = m;
+
+                       if (unlikely(error)) {
+                               rte_errno = -error;
+                               break;
+                       }
+               }
+
+               error = rte_net_intel_cksum_prepare(m);
+               if (unlikely(error)) {
+                       rte_errno = -error;
+                       break;
+               }
+
+               if (m->ol_flags & PKT_TX_TCP_SEG)
+                       virtio_tso_fix_cksum(m);
+       }
+
+       return nb_tx;
+}
+
 uint16_t
 virtio_xmit_pkts_packed(void *tx_queue, struct rte_mbuf **tx_pkts,
                        uint16_t nb_pkts)
@@ -1970,8 +1741,7 @@ virtio_xmit_pkts_packed(void *tx_queue, struct rte_mbuf **tx_pkts,
        struct virtio_hw *hw = vq->hw;
        uint16_t hdr_size = hw->vtnet_hdr_size;
        uint16_t nb_tx = 0;
-       bool in_order = hw->use_inorder_tx;
-       int error;
+       bool in_order = vtpci_with_feature(hw, VIRTIO_F_IN_ORDER);
 
        if (unlikely(hw->started == 0 && tx_pkts != hw->inject_pkts))
                return nb_tx;
@@ -1987,16 +1757,7 @@ virtio_xmit_pkts_packed(void *tx_queue, struct rte_mbuf **tx_pkts,
 
        for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
                struct rte_mbuf *txm = tx_pkts[nb_tx];
-               int can_push = 0, slots, need;
-
-               /* Do VLAN tag insertion */
-               if (unlikely(txm->ol_flags & PKT_TX_VLAN_PKT)) {
-                       error = rte_vlan_insert(&txm);
-                       if (unlikely(error)) {
-                               rte_pktmbuf_free(txm);
-                               continue;
-                       }
-               }
+               int can_push = 0, use_indirect = 0, slots, need;
 
                /* optimize ring usage */
                if ((vtpci_with_feature(hw, VIRTIO_F_ANY_LAYOUT) ||
@@ -2008,12 +1769,15 @@ virtio_xmit_pkts_packed(void *tx_queue, struct rte_mbuf **tx_pkts,
                    rte_is_aligned(rte_pktmbuf_mtod(txm, char *),
                           __alignof__(struct virtio_net_hdr_mrg_rxbuf)))
                        can_push = 1;
-
+               else if (vtpci_with_feature(hw, VIRTIO_RING_F_INDIRECT_DESC) &&
+                        txm->nb_segs < VIRTIO_MAX_TX_INDIRECT)
+                       use_indirect = 1;
                /* How many main ring entries are needed to this Tx?
+                * indirect   => 1
                 * any_layout => number of segments
                 * default    => number of segments + 1
                 */
-               slots = txm->nb_segs + !can_push;
+               slots = use_indirect ? 1 : (txm->nb_segs + !can_push);
                need = slots - vq->vq_free_cnt;
 
                /* Positive value indicates it need free vring descriptors */
@@ -2031,7 +1795,8 @@ virtio_xmit_pkts_packed(void *tx_queue, struct rte_mbuf **tx_pkts,
                if (can_push)
                        virtqueue_enqueue_xmit_packed_fast(txvq, txm, in_order);
                else
-                       virtqueue_enqueue_xmit_packed(txvq, txm, slots, 0,
+                       virtqueue_enqueue_xmit_packed(txvq, txm, slots,
+                                                     use_indirect, 0,
                                                      in_order);
 
                virtio_update_packet_stats(&txvq->stats, txm);
@@ -2057,7 +1822,6 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
        struct virtio_hw *hw = vq->hw;
        uint16_t hdr_size = hw->vtnet_hdr_size;
        uint16_t nb_used, nb_tx = 0;
-       int error;
 
        if (unlikely(hw->started == 0 && tx_pkts != hw->inject_pkts))
                return nb_tx;
@@ -2066,9 +1830,9 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                return nb_pkts;
 
        PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
-       nb_used = VIRTQUEUE_NUSED(vq);
 
-       virtio_rmb(hw->weak_barriers);
+       nb_used = virtqueue_nused(vq);
+
        if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
                virtio_xmit_cleanup(vq, nb_used);
 
@@ -2076,15 +1840,6 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                struct rte_mbuf *txm = tx_pkts[nb_tx];
                int can_push = 0, use_indirect = 0, slots, need;
 
-               /* Do VLAN tag insertion */
-               if (unlikely(txm->ol_flags & PKT_TX_VLAN_PKT)) {
-                       error = rte_vlan_insert(&txm);
-                       if (unlikely(error)) {
-                               rte_pktmbuf_free(txm);
-                               continue;
-                       }
-               }
-
                /* optimize ring usage */
                if ((vtpci_with_feature(hw, VIRTIO_F_ANY_LAYOUT) ||
                      vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) &&
@@ -2109,8 +1864,8 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 
                /* Positive value indicates it need free vring descriptors */
                if (unlikely(need > 0)) {
-                       nb_used = VIRTQUEUE_NUSED(vq);
-                       virtio_rmb(hw->weak_barriers);
+                       nb_used = virtqueue_nused(vq);
+
                        need = RTE_MIN(need, (int)nb_used);
 
                        virtio_xmit_cleanup(vq, need);
@@ -2143,6 +1898,20 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
        return nb_tx;
 }
 
+static __rte_always_inline int
+virtio_xmit_try_cleanup_inorder(struct virtqueue *vq, uint16_t need)
+{
+       uint16_t nb_used, nb_clean, nb_descs;
+
+       nb_descs = vq->vq_free_cnt + need;
+       nb_used = virtqueue_nused(vq);
+       nb_clean = RTE_MIN(need, (int)nb_used);
+
+       virtio_xmit_cleanup_inorder(vq, nb_clean);
+
+       return nb_descs - vq->vq_free_cnt;
+}
+
 uint16_t
 virtio_xmit_pkts_inorder(void *tx_queue,
                        struct rte_mbuf **tx_pkts,
@@ -2152,9 +1921,9 @@ virtio_xmit_pkts_inorder(void *tx_queue,
        struct virtqueue *vq = txvq->vq;
        struct virtio_hw *hw = vq->hw;
        uint16_t hdr_size = hw->vtnet_hdr_size;
-       uint16_t nb_used, nb_avail, nb_tx = 0, nb_inorder_pkts = 0;
+       uint16_t nb_used, nb_tx = 0, nb_inorder_pkts = 0;
        struct rte_mbuf *inorder_pkts[nb_pkts];
-       int error;
+       int need;
 
        if (unlikely(hw->started == 0 && tx_pkts != hw->inject_pkts))
                return nb_tx;
@@ -2164,29 +1933,14 @@ virtio_xmit_pkts_inorder(void *tx_queue,
 
        VIRTQUEUE_DUMP(vq);
        PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
-       nb_used = VIRTQUEUE_NUSED(vq);
+       nb_used = virtqueue_nused(vq);
 
-       virtio_rmb(hw->weak_barriers);
        if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
                virtio_xmit_cleanup_inorder(vq, nb_used);
 
-       if (unlikely(!vq->vq_free_cnt))
-               virtio_xmit_cleanup_inorder(vq, nb_used);
-
-       nb_avail = RTE_MIN(vq->vq_free_cnt, nb_pkts);
-
-       for (nb_tx = 0; nb_tx < nb_avail; nb_tx++) {
+       for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
                struct rte_mbuf *txm = tx_pkts[nb_tx];
-               int slots, need;
-
-               /* Do VLAN tag insertion */
-               if (unlikely(txm->ol_flags & PKT_TX_VLAN_PKT)) {
-                       error = rte_vlan_insert(&txm);
-                       if (unlikely(error)) {
-                               rte_pktmbuf_free(txm);
-                               continue;
-                       }
-               }
+               int slots;
 
                /* optimize ring usage */
                if ((vtpci_with_feature(hw, VIRTIO_F_ANY_LAYOUT) ||
@@ -2200,11 +1954,21 @@ virtio_xmit_pkts_inorder(void *tx_queue,
                        inorder_pkts[nb_inorder_pkts] = txm;
                        nb_inorder_pkts++;
 
-                       virtio_update_packet_stats(&txvq->stats, txm);
                        continue;
                }
 
                if (nb_inorder_pkts) {
+                       need = nb_inorder_pkts - vq->vq_free_cnt;
+                       if (unlikely(need > 0)) {
+                               need = virtio_xmit_try_cleanup_inorder(vq,
+                                                                      need);
+                               if (unlikely(need > 0)) {
+                                       PMD_TX_LOG(ERR,
+                                               "No free tx descriptors to "
+                                               "transmit");
+                                       break;
+                               }
+                       }
                        virtqueue_enqueue_xmit_inorder(txvq, inorder_pkts,
                                                        nb_inorder_pkts);
                        nb_inorder_pkts = 0;
@@ -2213,13 +1977,7 @@ virtio_xmit_pkts_inorder(void *tx_queue,
                slots = txm->nb_segs + 1;
                need = slots - vq->vq_free_cnt;
                if (unlikely(need > 0)) {
-                       nb_used = VIRTQUEUE_NUSED(vq);
-                       virtio_rmb(hw->weak_barriers);
-                       need = RTE_MIN(need, (int)nb_used);
-
-                       virtio_xmit_cleanup_inorder(vq, need);
-
-                       need = slots - vq->vq_free_cnt;
+                       need = virtio_xmit_try_cleanup_inorder(vq, slots);
 
                        if (unlikely(need > 0)) {
                                PMD_TX_LOG(ERR,
@@ -2234,9 +1992,22 @@ virtio_xmit_pkts_inorder(void *tx_queue,
        }
 
        /* Transmit all inorder packets */
-       if (nb_inorder_pkts)
+       if (nb_inorder_pkts) {
+               need = nb_inorder_pkts - vq->vq_free_cnt;
+               if (unlikely(need > 0)) {
+                       need = virtio_xmit_try_cleanup_inorder(vq,
+                                                                 need);
+                       if (unlikely(need > 0)) {
+                               PMD_TX_LOG(ERR,
+                                       "No free tx descriptors to transmit");
+                               nb_inorder_pkts = vq->vq_free_cnt;
+                               nb_tx -= need;
+                       }
+               }
+
                virtqueue_enqueue_xmit_inorder(txvq, inorder_pkts,
                                                nb_inorder_pkts);
+       }
 
        txvq->stats.packets += nb_tx;
 
@@ -2253,3 +2024,19 @@ virtio_xmit_pkts_inorder(void *tx_queue,
 
        return nb_tx;
 }
+
+__rte_weak uint16_t
+virtio_recv_pkts_packed_vec(void *rx_queue __rte_unused,
+                           struct rte_mbuf **rx_pkts __rte_unused,
+                           uint16_t nb_pkts __rte_unused)
+{
+       return 0;
+}
+
+__rte_weak uint16_t
+virtio_xmit_pkts_packed_vec(void *tx_queue __rte_unused,
+                           struct rte_mbuf **tx_pkts __rte_unused,
+                           uint16_t nb_pkts __rte_unused)
+{
+       return 0;
+}