examples: skip build when missing dependencies
[dpdk.git] / lib / vhost / virtio_net.c
index 29254a1..cef4bcf 100644 (file)
@@ -792,192 +792,156 @@ copy_vnet_hdr_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 }
 
 static __rte_always_inline int
-copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
-                           struct rte_mbuf *m, struct buf_vector *buf_vec,
-                           uint16_t nr_vec, uint16_t num_buffers)
+async_iter_initialize(struct vhost_async *async)
 {
-       uint32_t vec_idx = 0;
-       uint32_t mbuf_offset, mbuf_avail;
-       uint32_t buf_offset, buf_avail;
-       uint64_t buf_addr, buf_iova, buf_len;
-       uint32_t cpy_len;
-       uint64_t hdr_addr;
-       struct rte_mbuf *hdr_mbuf;
-       struct batch_copy_elem *batch_copy = vq->batch_copy_elems;
-       struct virtio_net_hdr_mrg_rxbuf tmp_hdr, *hdr = NULL;
-       int error = 0;
-
-       if (unlikely(m == NULL)) {
-               error = -1;
-               goto out;
-       }
-
-       buf_addr = buf_vec[vec_idx].buf_addr;
-       buf_iova = buf_vec[vec_idx].buf_iova;
-       buf_len = buf_vec[vec_idx].buf_len;
+       struct rte_vhost_iov_iter *iter;
 
-       if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
-               error = -1;
-               goto out;
+       if (unlikely(async->iovec_idx >= VHOST_MAX_ASYNC_VEC)) {
+               VHOST_LOG_DATA(ERR, "no more async iovec available\n");
+               return -1;
        }
 
-       hdr_mbuf = m;
-       hdr_addr = buf_addr;
-       if (unlikely(buf_len < dev->vhost_hlen)) {
-               memset(&tmp_hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
-               hdr = &tmp_hdr;
-       } else
-               hdr = (struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)hdr_addr;
-
-       VHOST_LOG_DATA(DEBUG, "(%d) RX: num merge buffers %d\n",
-               dev->vid, num_buffers);
+       iter = async->iov_iter + async->iter_idx;
+       iter->iov = async->iovec + async->iovec_idx;
+       iter->nr_segs = 0;
 
-       if (unlikely(buf_len < dev->vhost_hlen)) {
-               buf_offset = dev->vhost_hlen - buf_len;
-               vec_idx++;
-               buf_addr = buf_vec[vec_idx].buf_addr;
-               buf_iova = buf_vec[vec_idx].buf_iova;
-               buf_len = buf_vec[vec_idx].buf_len;
-               buf_avail = buf_len - buf_offset;
-       } else {
-               buf_offset = dev->vhost_hlen;
-               buf_avail = buf_len - dev->vhost_hlen;
-       }
+       return 0;
+}
 
-       mbuf_avail  = rte_pktmbuf_data_len(m);
-       mbuf_offset = 0;
-       while (mbuf_avail != 0 || m->next != NULL) {
-               /* done with current buf, get the next one */
-               if (buf_avail == 0) {
-                       vec_idx++;
-                       if (unlikely(vec_idx >= nr_vec)) {
-                               error = -1;
-                               goto out;
-                       }
+static __rte_always_inline int
+async_iter_add_iovec(struct vhost_async *async, void *src, void *dst, size_t len)
+{
+       struct rte_vhost_iov_iter *iter;
+       struct rte_vhost_iovec *iovec;
 
-                       buf_addr = buf_vec[vec_idx].buf_addr;
-                       buf_iova = buf_vec[vec_idx].buf_iova;
-                       buf_len = buf_vec[vec_idx].buf_len;
+       if (unlikely(async->iovec_idx >= VHOST_MAX_ASYNC_VEC)) {
+               static bool vhost_max_async_vec_log;
 
-                       buf_offset = 0;
-                       buf_avail  = buf_len;
+               if (!vhost_max_async_vec_log) {
+                       VHOST_LOG_DATA(ERR, "no more async iovec available\n");
+                       vhost_max_async_vec_log = true;
                }
 
-               /* done with current mbuf, get the next one */
-               if (mbuf_avail == 0) {
-                       m = m->next;
-
-                       mbuf_offset = 0;
-                       mbuf_avail  = rte_pktmbuf_data_len(m);
-               }
-
-               if (hdr_addr) {
-                       virtio_enqueue_offload(hdr_mbuf, &hdr->hdr);
-                       if (rxvq_is_mergeable(dev))
-                               ASSIGN_UNLESS_EQUAL(hdr->num_buffers,
-                                               num_buffers);
+               return -1;
+       }
 
-                       if (unlikely(hdr == &tmp_hdr)) {
-                               copy_vnet_hdr_to_desc(dev, vq, buf_vec, hdr);
-                       } else {
-                               PRINT_PACKET(dev, (uintptr_t)hdr_addr,
-                                               dev->vhost_hlen, 0);
-                               vhost_log_cache_write_iova(dev, vq,
-                                               buf_vec[0].buf_iova,
-                                               dev->vhost_hlen);
-                       }
+       iter = async->iov_iter + async->iter_idx;
+       iovec = async->iovec + async->iovec_idx;
 
-                       hdr_addr = 0;
-               }
+       iovec->src_addr = src;
+       iovec->dst_addr = dst;
+       iovec->len = len;
 
-               cpy_len = RTE_MIN(buf_avail, mbuf_avail);
+       iter->nr_segs++;
+       async->iovec_idx++;
 
-               if (likely(cpy_len > MAX_BATCH_LEN ||
-                                       vq->batch_copy_nb_elems >= vq->size)) {
-                       rte_memcpy((void *)((uintptr_t)(buf_addr + buf_offset)),
-                               rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
-                               cpy_len);
-                       vhost_log_cache_write_iova(dev, vq,
-                                                  buf_iova + buf_offset,
-                                                  cpy_len);
-                       PRINT_PACKET(dev, (uintptr_t)(buf_addr + buf_offset),
-                               cpy_len, 0);
-               } else {
-                       batch_copy[vq->batch_copy_nb_elems].dst =
-                               (void *)((uintptr_t)(buf_addr + buf_offset));
-                       batch_copy[vq->batch_copy_nb_elems].src =
-                               rte_pktmbuf_mtod_offset(m, void *, mbuf_offset);
-                       batch_copy[vq->batch_copy_nb_elems].log_addr =
-                               buf_iova + buf_offset;
-                       batch_copy[vq->batch_copy_nb_elems].len = cpy_len;
-                       vq->batch_copy_nb_elems++;
-               }
+       return 0;
+}
 
-               mbuf_avail  -= cpy_len;
-               mbuf_offset += cpy_len;
-               buf_avail  -= cpy_len;
-               buf_offset += cpy_len;
-       }
+static __rte_always_inline void
+async_iter_finalize(struct vhost_async *async)
+{
+       async->iter_idx++;
+}
 
-out:
+static __rte_always_inline void
+async_iter_cancel(struct vhost_async *async)
+{
+       struct rte_vhost_iov_iter *iter;
 
-       return error;
+       iter = async->iov_iter + async->iter_idx;
+       async->iovec_idx -= iter->nr_segs;
+       iter->nr_segs = 0;
+       iter->iov = NULL;
 }
 
 static __rte_always_inline void
-async_fill_vec(struct rte_vhost_iovec *v, void *src, void *dst, size_t len)
+async_iter_reset(struct vhost_async *async)
 {
-       v->src_addr = src;
-       v->dst_addr = dst;
-       v->len = len;
+       async->iter_idx = 0;
+       async->iovec_idx = 0;
 }
 
-static __rte_always_inline void
-async_fill_iter(struct rte_vhost_iov_iter *it, struct rte_vhost_iovec *vec, unsigned long nr_seg)
+static __rte_always_inline int
+async_mbuf_to_desc_seg(struct virtio_net *dev, struct vhost_virtqueue *vq,
+               struct rte_mbuf *m, uint32_t mbuf_offset,
+               uint64_t buf_iova, uint32_t cpy_len)
 {
-       it->iov = vec;
-       it->nr_segs = nr_seg;
+       struct vhost_async *async = vq->async;
+       uint64_t mapped_len;
+       uint32_t buf_offset = 0;
+       void *hpa;
+
+       while (cpy_len) {
+               hpa = (void *)(uintptr_t)gpa_to_first_hpa(dev,
+                               buf_iova + buf_offset, cpy_len, &mapped_len);
+               if (unlikely(!hpa)) {
+                       VHOST_LOG_DATA(ERR, "(%d) %s: failed to get hpa.\n", dev->vid, __func__);
+                       return -1;
+               }
+
+               if (unlikely(async_iter_add_iovec(async,
+                                               (void *)(uintptr_t)rte_pktmbuf_iova_offset(m,
+                                                       mbuf_offset),
+                                               hpa, (size_t)mapped_len)))
+                       return -1;
+
+               cpy_len -= (uint32_t)mapped_len;
+               mbuf_offset += (uint32_t)mapped_len;
+               buf_offset += (uint32_t)mapped_len;
+       }
+
+       return 0;
 }
 
 static __rte_always_inline void
-async_fill_desc(struct rte_vhost_async_desc *desc, struct rte_vhost_iov_iter *iter)
+sync_mbuf_to_desc_seg(struct virtio_net *dev, struct vhost_virtqueue *vq,
+               struct rte_mbuf *m, uint32_t mbuf_offset,
+               uint64_t buf_addr, uint64_t buf_iova, uint32_t cpy_len)
 {
-       desc->iter = iter;
+       struct batch_copy_elem *batch_copy = vq->batch_copy_elems;
+
+       if (likely(cpy_len > MAX_BATCH_LEN || vq->batch_copy_nb_elems >= vq->size)) {
+               rte_memcpy((void *)((uintptr_t)(buf_addr)),
+                               rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
+                               cpy_len);
+               vhost_log_cache_write_iova(dev, vq, buf_iova, cpy_len);
+               PRINT_PACKET(dev, (uintptr_t)(buf_addr), cpy_len, 0);
+       } else {
+               batch_copy[vq->batch_copy_nb_elems].dst =
+                       (void *)((uintptr_t)(buf_addr));
+               batch_copy[vq->batch_copy_nb_elems].src =
+                       rte_pktmbuf_mtod_offset(m, void *, mbuf_offset);
+               batch_copy[vq->batch_copy_nb_elems].log_addr = buf_iova;
+               batch_copy[vq->batch_copy_nb_elems].len = cpy_len;
+               vq->batch_copy_nb_elems++;
+       }
 }
 
 static __rte_always_inline int
-async_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
-                       struct rte_mbuf *m, struct buf_vector *buf_vec,
-                       uint16_t nr_vec, uint16_t num_buffers,
-                       struct rte_vhost_iovec *iovec, struct rte_vhost_iov_iter *iter)
+mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
+               struct rte_mbuf *m, struct buf_vector *buf_vec,
+               uint16_t nr_vec, uint16_t num_buffers, bool is_async)
 {
-       struct rte_mbuf *hdr_mbuf;
-       struct virtio_net_hdr_mrg_rxbuf tmp_hdr, *hdr = NULL;
-       uint64_t buf_addr, buf_iova;
-       uint64_t hdr_addr;
-       uint64_t mapped_len;
        uint32_t vec_idx = 0;
        uint32_t mbuf_offset, mbuf_avail;
        uint32_t buf_offset, buf_avail;
-       uint32_t cpy_len, buf_len;
-       int error = 0;
-
-       int tvec_idx = 0;
-       void *hpa;
+       uint64_t buf_addr, buf_iova, buf_len;
+       uint32_t cpy_len;
+       uint64_t hdr_addr;
+       struct rte_mbuf *hdr_mbuf;
+       struct virtio_net_hdr_mrg_rxbuf tmp_hdr, *hdr = NULL;
+       struct vhost_async *async = vq->async;
 
-       if (unlikely(m == NULL)) {
-               error = -1;
-               goto out;
-       }
+       if (unlikely(m == NULL))
+               return -1;
 
        buf_addr = buf_vec[vec_idx].buf_addr;
        buf_iova = buf_vec[vec_idx].buf_iova;
        buf_len = buf_vec[vec_idx].buf_len;
 
-       if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
-               error = -1;
-               goto out;
-       }
+       if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1))
+               return -1;
 
        hdr_mbuf = m;
        hdr_addr = buf_addr;
@@ -1005,21 +969,24 @@ async_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
        mbuf_avail  = rte_pktmbuf_data_len(m);
        mbuf_offset = 0;
 
+       if (is_async) {
+               if (async_iter_initialize(async))
+                       return -1;
+       }
+
        while (mbuf_avail != 0 || m->next != NULL) {
                /* done with current buf, get the next one */
                if (buf_avail == 0) {
                        vec_idx++;
-                       if (unlikely(vec_idx >= nr_vec)) {
-                               error = -1;
-                               goto out;
-                       }
+                       if (unlikely(vec_idx >= nr_vec))
+                               goto error;
 
                        buf_addr = buf_vec[vec_idx].buf_addr;
                        buf_iova = buf_vec[vec_idx].buf_iova;
                        buf_len = buf_vec[vec_idx].buf_len;
 
                        buf_offset = 0;
-                       buf_avail = buf_len;
+                       buf_avail  = buf_len;
                }
 
                /* done with current mbuf, get the next one */
@@ -1027,7 +994,7 @@ async_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
                        m = m->next;
 
                        mbuf_offset = 0;
-                       mbuf_avail = rte_pktmbuf_data_len(m);
+                       mbuf_avail  = rte_pktmbuf_data_len(m);
                }
 
                if (hdr_addr) {
@@ -1051,33 +1018,31 @@ async_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
                cpy_len = RTE_MIN(buf_avail, mbuf_avail);
 
-               while (unlikely(cpy_len)) {
-                       hpa = (void *)(uintptr_t)gpa_to_first_hpa(dev,
-                                       buf_iova + buf_offset,
-                                       cpy_len, &mapped_len);
-                       if (unlikely(!hpa)) {
-                               VHOST_LOG_DATA(ERR, "(%d) %s: failed to get hpa.\n",
-                               dev->vid, __func__);
-                               error = -1;
-                               goto out;
-                       }
-
-                       async_fill_vec(iovec + tvec_idx,
-                               (void *)(uintptr_t)rte_pktmbuf_iova_offset(m,
-                               mbuf_offset), hpa, (size_t)mapped_len);
-
-                       cpy_len -= (uint32_t)mapped_len;
-                       mbuf_avail  -= (uint32_t)mapped_len;
-                       mbuf_offset += (uint32_t)mapped_len;
-                       buf_avail  -= (uint32_t)mapped_len;
-                       buf_offset += (uint32_t)mapped_len;
-                       tvec_idx++;
+               if (is_async) {
+                       if (async_mbuf_to_desc_seg(dev, vq, m, mbuf_offset,
+                                               buf_iova + buf_offset, cpy_len) < 0)
+                               goto error;
+               } else {
+                       sync_mbuf_to_desc_seg(dev, vq, m, mbuf_offset,
+                                       buf_addr + buf_offset,
+                                       buf_iova + buf_offset, cpy_len);
                }
+
+               mbuf_avail  -= cpy_len;
+               mbuf_offset += cpy_len;
+               buf_avail  -= cpy_len;
+               buf_offset += cpy_len;
        }
 
-       async_fill_iter(iter, iovec, tvec_idx);
-out:
-       return error;
+       if (is_async)
+               async_iter_finalize(async);
+
+       return 0;
+error:
+       if (is_async)
+               async_iter_cancel(async);
+
+       return -1;
 }
 
 static __rte_always_inline int
@@ -1134,7 +1099,7 @@ vhost_enqueue_single_packed(struct virtio_net *dev,
                        avail_idx -= vq->size;
        }
 
-       if (copy_mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec, num_buffers) < 0)
+       if (mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec, num_buffers, false) < 0)
                return -1;
 
        vhost_shadow_enqueue_single_packed(dev, vq, buffer_len, buffer_buf_id,
@@ -1178,9 +1143,8 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
                        dev->vid, vq->last_avail_idx,
                        vq->last_avail_idx + num_buffers);
 
-               if (copy_mbuf_to_desc(dev, vq, pkts[pkt_idx],
-                                               buf_vec, nr_vec,
-                                               num_buffers) < 0) {
+               if (mbuf_to_desc(dev, vq, pkts[pkt_idx], buf_vec, nr_vec,
+                                       num_buffers, false) < 0) {
                        vq->shadow_used_idx -= num_buffers;
                        break;
                }
@@ -1441,11 +1405,14 @@ rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
 }
 
 static __rte_always_inline uint16_t
-virtio_dev_rx_async_get_info_idx(uint16_t pkts_idx,
-       uint16_t vq_size, uint16_t n_inflight)
+async_get_first_inflight_pkt_idx(struct vhost_virtqueue *vq)
 {
-       return pkts_idx > n_inflight ? (pkts_idx - n_inflight) :
-               (vq_size - n_inflight + pkts_idx) % vq_size;
+       struct vhost_async *async = vq->async;
+
+       if (async->pkts_idx >= async->pkts_inflight_n)
+               return async->pkts_idx - async->pkts_inflight_n;
+       else
+               return vq->size - async->pkts_inflight_n + async->pkts_idx;
 }
 
 static __rte_always_inline void
@@ -1487,18 +1454,15 @@ virtio_dev_rx_async_submit_split(struct virtio_net *dev,
        struct rte_mbuf **pkts, uint32_t count)
 {
        struct buf_vector buf_vec[BUF_VECTOR_MAX];
-       uint32_t pkt_idx = 0, pkt_burst_idx = 0;
+       uint32_t pkt_idx = 0;
        uint16_t num_buffers;
        uint16_t avail_head;
 
        struct vhost_async *async = vq->async;
-       struct rte_vhost_iov_iter *iter = async->iov_iter;
-       struct rte_vhost_async_desc tdes[MAX_PKT_BURST];
-       struct rte_vhost_iovec *iovec = async->iovec;
        struct async_inflight_info *pkts_info = async->pkts_info;
-       uint32_t n_pkts = 0, pkt_err = 0;
+       uint32_t pkt_err = 0;
        int32_t n_xfer;
-       uint16_t iovec_idx = 0, it_idx = 0, slot_idx = 0;
+       uint16_t slot_idx = 0;
 
        /*
         * The ordering between avail index and desc reads need to be enforced.
@@ -1507,95 +1471,51 @@ virtio_dev_rx_async_submit_split(struct virtio_net *dev,
 
        rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 
+       async_iter_reset(async);
+
        for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
                uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
                uint16_t nr_vec = 0;
 
-               if (unlikely(reserve_avail_buf_split(dev, vq,
-                                               pkt_len, buf_vec, &num_buffers,
-                                               avail_head, &nr_vec) < 0)) {
-                       VHOST_LOG_DATA(DEBUG,
-                               "(%d) failed to get enough desc from vring\n",
-                               dev->vid);
+               if (unlikely(reserve_avail_buf_split(dev, vq, pkt_len, buf_vec,
+                                               &num_buffers, avail_head, &nr_vec) < 0)) {
+                       VHOST_LOG_DATA(DEBUG, "(%d) failed to get enough desc from vring\n",
+                                       dev->vid);
                        vq->shadow_used_idx -= num_buffers;
                        break;
                }
 
                VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
-                       dev->vid, vq->last_avail_idx,
-                       vq->last_avail_idx + num_buffers);
+                       dev->vid, vq->last_avail_idx, vq->last_avail_idx + num_buffers);
 
-               if (async_mbuf_to_desc(dev, vq, pkts[pkt_idx], buf_vec, nr_vec, num_buffers,
-                               &iovec[iovec_idx], &iter[it_idx]) < 0) {
+               if (mbuf_to_desc(dev, vq, pkts[pkt_idx], buf_vec, nr_vec, num_buffers, true) < 0) {
                        vq->shadow_used_idx -= num_buffers;
                        break;
                }
 
-               async_fill_desc(&tdes[pkt_burst_idx++], &iter[it_idx]);
-
                slot_idx = (async->pkts_idx + pkt_idx) & (vq->size - 1);
                pkts_info[slot_idx].descs = num_buffers;
                pkts_info[slot_idx].mbuf = pkts[pkt_idx];
 
-               iovec_idx += iter[it_idx].nr_segs;
-               it_idx++;
-
                vq->last_avail_idx += num_buffers;
-
-               /*
-                * condition to trigger async device transfer:
-                * - unused async iov number is less than max vhost vector
-                */
-               if (unlikely(VHOST_MAX_ASYNC_VEC - iovec_idx < BUF_VECTOR_MAX)) {
-                       n_xfer = async->ops.transfer_data(dev->vid,
-                                       queue_id, tdes, 0, pkt_burst_idx);
-                       if (likely(n_xfer >= 0)) {
-                               n_pkts = n_xfer;
-                       } else {
-                               VHOST_LOG_DATA(ERR,
-                                       "(%d) %s: failed to transfer data for queue id %d.\n",
-                                       dev->vid, __func__, queue_id);
-                               n_pkts = 0;
-                       }
-
-                       iovec_idx = 0;
-                       it_idx = 0;
-
-                       if (unlikely(n_pkts < pkt_burst_idx)) {
-                               /*
-                                * log error packets number here and do actual
-                                * error processing when applications poll
-                                * completion
-                                */
-                               pkt_err = pkt_burst_idx - n_pkts;
-                               pkt_idx++;
-                               pkt_burst_idx = 0;
-                               break;
-                       }
-
-                       pkt_burst_idx = 0;
-               }
        }
 
-       if (pkt_burst_idx) {
-               n_xfer = async->ops.transfer_data(dev->vid, queue_id, tdes, 0, pkt_burst_idx);
-               if (likely(n_xfer >= 0)) {
-                       n_pkts = n_xfer;
-               } else {
-                       VHOST_LOG_DATA(ERR, "(%d) %s: failed to transfer data for queue id %d.\n",
-                               dev->vid, __func__, queue_id);
-                       n_pkts = 0;
-               }
+       if (unlikely(pkt_idx == 0))
+               return 0;
 
-               if (unlikely(n_pkts < pkt_burst_idx))
-                       pkt_err = pkt_burst_idx - n_pkts;
+       n_xfer = async->ops.transfer_data(dev->vid, queue_id, async->iov_iter, 0, pkt_idx);
+       if (unlikely(n_xfer < 0)) {
+               VHOST_LOG_DATA(ERR, "(%d) %s: failed to transfer data for queue id %d.\n",
+                               dev->vid, __func__, queue_id);
+               n_xfer = 0;
        }
 
+       pkt_err = pkt_idx - n_xfer;
        if (unlikely(pkt_err)) {
                uint16_t num_descs = 0;
 
                /* update number of completed packets */
-               pkt_idx -= pkt_err;
+               pkt_idx = n_xfer;
 
                /* calculate the sum of descriptors to revert */
                while (pkt_err-- > 0) {
@@ -1617,7 +1537,11 @@ virtio_dev_rx_async_submit_split(struct virtio_net *dev,
                                vq->shadow_used_idx);
 
                async->desc_idx_split += vq->shadow_used_idx;
+
                async->pkts_idx += pkt_idx;
+               if (async->pkts_idx >= vq->size)
+                       async->pkts_idx -= vq->size;
+
                async->pkts_inflight_n += pkt_idx;
                vq->shadow_used_idx = 0;
        }
@@ -1686,9 +1610,7 @@ vhost_enqueue_async_packed(struct virtio_net *dev,
                            struct rte_mbuf *pkt,
                            struct buf_vector *buf_vec,
                            uint16_t *nr_descs,
-                           uint16_t *nr_buffers,
-                           struct rte_vhost_iovec *iovec,
-                           struct rte_vhost_iov_iter *iter)
+                           uint16_t *nr_buffers)
 {
        uint16_t nr_vec = 0;
        uint16_t avail_idx = vq->last_avail_idx;
@@ -1735,8 +1657,7 @@ vhost_enqueue_async_packed(struct virtio_net *dev,
                        avail_idx -= vq->size;
        }
 
-       if (unlikely(async_mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec,
-                                       *nr_buffers, iovec, iter) < 0))
+       if (unlikely(mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec, *nr_buffers, true) < 0))
                return -1;
 
        vhost_shadow_enqueue_packed(vq, buffer_len, buffer_buf_id, buffer_desc_count, *nr_buffers);
@@ -1746,13 +1667,12 @@ vhost_enqueue_async_packed(struct virtio_net *dev,
 
 static __rte_always_inline int16_t
 virtio_dev_rx_async_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
-                           struct rte_mbuf *pkt, uint16_t *nr_descs, uint16_t *nr_buffers,
-                           struct rte_vhost_iovec *iovec, struct rte_vhost_iov_iter *iter)
+                           struct rte_mbuf *pkt, uint16_t *nr_descs, uint16_t *nr_buffers)
 {
        struct buf_vector buf_vec[BUF_VECTOR_MAX];
 
-       if (unlikely(vhost_enqueue_async_packed(dev, vq, pkt, buf_vec, nr_descs, nr_buffers,
-                                                iovec, iter) < 0)) {
+       if (unlikely(vhost_enqueue_async_packed(dev, vq, pkt, buf_vec,
+                                       nr_descs, nr_buffers) < 0)) {
                VHOST_LOG_DATA(DEBUG, "(%d) failed to get enough desc from vring\n", dev->vid);
                return -1;
        }
@@ -1794,20 +1714,16 @@ virtio_dev_rx_async_submit_packed(struct virtio_net *dev,
        struct vhost_virtqueue *vq, uint16_t queue_id,
        struct rte_mbuf **pkts, uint32_t count)
 {
-       uint32_t pkt_idx = 0, pkt_burst_idx = 0;
+       uint32_t pkt_idx = 0;
        uint32_t remained = count;
        int32_t n_xfer;
        uint16_t num_buffers;
        uint16_t num_descs;
 
        struct vhost_async *async = vq->async;
-       struct rte_vhost_iov_iter *iter = async->iov_iter;
-       struct rte_vhost_async_desc tdes[MAX_PKT_BURST];
-       struct rte_vhost_iovec *iovec = async->iovec;
        struct async_inflight_info *pkts_info = async->pkts_info;
-       uint32_t n_pkts = 0, pkt_err = 0;
+       uint32_t pkt_err = 0;
        uint16_t slot_idx = 0;
-       uint16_t iovec_idx = 0, it_idx = 0;
 
        do {
                rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
@@ -1815,71 +1731,34 @@ virtio_dev_rx_async_submit_packed(struct virtio_net *dev,
                num_buffers = 0;
                num_descs = 0;
                if (unlikely(virtio_dev_rx_async_packed(dev, vq, pkts[pkt_idx],
-                                               &num_descs, &num_buffers,
-                                               &iovec[iovec_idx], &iter[it_idx]) < 0))
+                                               &num_descs, &num_buffers) < 0))
                        break;
 
                slot_idx = (async->pkts_idx + pkt_idx) % vq->size;
 
-               async_fill_desc(&tdes[pkt_burst_idx++], &iter[it_idx]);
                pkts_info[slot_idx].descs = num_descs;
                pkts_info[slot_idx].nr_buffers = num_buffers;
                pkts_info[slot_idx].mbuf = pkts[pkt_idx];
-               iovec_idx += iter[it_idx].nr_segs;
-               it_idx++;
 
                pkt_idx++;
                remained--;
                vq_inc_last_avail_packed(vq, num_descs);
-
-               /*
-                * condition to trigger async device transfer:
-                * - unused async iov number is less than max vhost vector
-                */
-               if (unlikely(VHOST_MAX_ASYNC_VEC - iovec_idx < BUF_VECTOR_MAX)) {
-                       n_xfer = async->ops.transfer_data(dev->vid,
-                                       queue_id, tdes, 0, pkt_burst_idx);
-                       if (likely(n_xfer >= 0)) {
-                               n_pkts = n_xfer;
-                       } else {
-                               VHOST_LOG_DATA(ERR,
-                                       "(%d) %s: failed to transfer data for queue id %d.\n",
-                                       dev->vid, __func__, queue_id);
-                               n_pkts = 0;
-                       }
-
-                       iovec_idx = 0;
-                       it_idx = 0;
-
-                       if (unlikely(n_pkts < pkt_burst_idx)) {
-                               /*
-                                * log error packets number here and do actual
-                                * error processing when applications poll
-                                * completion
-                                */
-                               pkt_err = pkt_burst_idx - n_pkts;
-                               pkt_burst_idx = 0;
-                               break;
-                       }
-
-                       pkt_burst_idx = 0;
-               }
        } while (pkt_idx < count);
 
-       if (pkt_burst_idx) {
-               n_xfer = async->ops.transfer_data(dev->vid, queue_id, tdes, 0, pkt_burst_idx);
-               if (likely(n_xfer >= 0)) {
-                       n_pkts = n_xfer;
-               } else {
-                       VHOST_LOG_DATA(ERR, "(%d) %s: failed to transfer data for queue id %d.\n",
-                               dev->vid, __func__, queue_id);
-                       n_pkts = 0;
-               }
+       if (unlikely(pkt_idx == 0))
+               return 0;
 
-               if (unlikely(n_pkts < pkt_burst_idx))
-                       pkt_err = pkt_burst_idx - n_pkts;
+       n_xfer = async->ops.transfer_data(dev->vid, queue_id, async->iov_iter, 0, pkt_idx);
+       if (unlikely(n_xfer < 0)) {
+               VHOST_LOG_DATA(ERR, "(%d) %s: failed to transfer data for queue id %d.\n",
+                               dev->vid, __func__, queue_id);
+               n_xfer = 0;
        }
 
+       pkt_err = pkt_idx - n_xfer;
+
+       async_iter_reset(async);
+
        if (unlikely(pkt_err))
                dma_error_handler_packed(vq, slot_idx, pkt_err, &pkt_idx);
 
@@ -1963,68 +1842,43 @@ static __rte_always_inline uint16_t
 vhost_poll_enqueue_completed(struct virtio_net *dev, uint16_t queue_id,
                struct rte_mbuf **pkts, uint16_t count)
 {
-       struct vhost_virtqueue *vq;
-       struct vhost_async *async;
-       struct async_inflight_info *pkts_info;
+       struct vhost_virtqueue *vq = dev->virtqueue[queue_id];
+       struct vhost_async *async = vq->async;
+       struct async_inflight_info *pkts_info = async->pkts_info;
        int32_t n_cpl;
-       uint16_t n_pkts_cpl = 0, n_pkts_put = 0, n_descs = 0, n_buffers = 0;
-       uint16_t start_idx, pkts_idx, vq_size;
-       uint16_t from, i;
+       uint16_t n_descs = 0, n_buffers = 0;
+       uint16_t start_idx, from, i;
 
-       vq = dev->virtqueue[queue_id];
-       async = vq->async;
-       pkts_idx = async->pkts_idx % vq->size;
-       pkts_info = async->pkts_info;
-       vq_size = vq->size;
-       start_idx = virtio_dev_rx_async_get_info_idx(pkts_idx,
-               vq_size, async->pkts_inflight_n);
-
-       if (count > async->last_pkts_n) {
-               n_cpl = async->ops.check_completed_copies(dev->vid,
-                       queue_id, 0, count - async->last_pkts_n);
-               if (likely(n_cpl >= 0)) {
-                       n_pkts_cpl = n_cpl;
-               } else {
-                       VHOST_LOG_DATA(ERR,
-                               "(%d) %s: failed to check completed copies for queue id %d.\n",
+       n_cpl = async->ops.check_completed_copies(dev->vid, queue_id, 0, count);
+       if (unlikely(n_cpl < 0)) {
+               VHOST_LOG_DATA(ERR, "(%d) %s: failed to check completed copies for queue id %d.\n",
                                dev->vid, __func__, queue_id);
-                       n_pkts_cpl = 0;
-               }
+               return 0;
        }
 
-       n_pkts_cpl += async->last_pkts_n;
-       n_pkts_put = RTE_MIN(n_pkts_cpl, count);
-       if (unlikely(n_pkts_put == 0)) {
-               async->last_pkts_n = n_pkts_cpl;
+       if (n_cpl == 0)
                return 0;
-       }
 
-       if (vq_is_packed(dev)) {
-               for (i = 0; i < n_pkts_put; i++) {
-                       from = (start_idx + i) % vq_size;
-                       n_buffers += pkts_info[from].nr_buffers;
-                       pkts[i] = pkts_info[from].mbuf;
-               }
-       } else {
-               for (i = 0; i < n_pkts_put; i++) {
-                       from = (start_idx + i) & (vq_size - 1);
-                       n_descs += pkts_info[from].descs;
-                       pkts[i] = pkts_info[from].mbuf;
-               }
+       start_idx = async_get_first_inflight_pkt_idx(vq);
+
+       for (i = 0; i < n_cpl; i++) {
+               from = (start_idx + i) % vq->size;
+               /* Only used with packed ring */
+               n_buffers += pkts_info[from].nr_buffers;
+               /* Only used with split ring */
+               n_descs += pkts_info[from].descs;
+               pkts[i] = pkts_info[from].mbuf;
        }
-       async->last_pkts_n = n_pkts_cpl - n_pkts_put;
-       async->pkts_inflight_n -= n_pkts_put;
+
+       async->pkts_inflight_n -= n_cpl;
 
        if (likely(vq->enabled && vq->access_ok)) {
                if (vq_is_packed(dev)) {
                        write_back_completed_descs_packed(vq, n_buffers);
-
                        vhost_vring_call_packed(dev, vq);
                } else {
                        write_back_completed_descs_split(vq, n_descs);
-
-                       __atomic_add_fetch(&vq->used->idx, n_descs,
-                                       __ATOMIC_RELEASE);
+                       __atomic_add_fetch(&vq->used->idx, n_descs, __ATOMIC_RELEASE);
                        vhost_vring_call_split(dev, vq);
                }
        } else {
@@ -2037,7 +1891,7 @@ vhost_poll_enqueue_completed(struct virtio_net *dev, uint16_t queue_id,
                }
        }
 
-       return n_pkts_put;
+       return n_cpl;
 }
 
 uint16_t