net/bnxt: fix null dereference in session cleanup
[dpdk.git] / lib / vhost / virtio_net.c
index 3c8be48..5f432b0 100644 (file)
@@ -11,6 +11,7 @@
 #include <rte_net.h>
 #include <rte_ether.h>
 #include <rte_ip.h>
+#include <rte_dmadev.h>
 #include <rte_vhost.h>
 #include <rte_tcp.h>
 #include <rte_udp.h>
@@ -25,6 +26,9 @@
 
 #define MAX_BATCH_LEN 256
 
+/* DMA device copy operation tracking array. */
+struct async_dma_info dma_copy_track[RTE_DMADEV_DEFAULT_MAX];
+
 static  __rte_always_inline bool
 rxvq_is_mergeable(struct virtio_net *dev)
 {
@@ -43,6 +47,135 @@ is_valid_virt_queue_idx(uint32_t idx, int is_tx, uint32_t nr_vring)
        return (is_tx ^ (idx & 1)) == 0 && idx < nr_vring;
 }
 
+static __rte_always_inline int64_t
+vhost_async_dma_transfer_one(struct virtio_net *dev, struct vhost_virtqueue *vq,
+               int16_t dma_id, uint16_t vchan_id, uint16_t flag_idx,
+               struct vhost_iov_iter *pkt)
+{
+       struct async_dma_vchan_info *dma_info = &dma_copy_track[dma_id].vchans[vchan_id];
+       uint16_t ring_mask = dma_info->ring_mask;
+       static bool vhost_async_dma_copy_log;
+
+
+       struct vhost_iovec *iov = pkt->iov;
+       int copy_idx = 0;
+       uint32_t nr_segs = pkt->nr_segs;
+       uint16_t i;
+
+       if (rte_dma_burst_capacity(dma_id, vchan_id) < nr_segs)
+               return -1;
+
+       for (i = 0; i < nr_segs; i++) {
+               copy_idx = rte_dma_copy(dma_id, vchan_id, (rte_iova_t)iov[i].src_addr,
+                               (rte_iova_t)iov[i].dst_addr, iov[i].len, RTE_DMA_OP_FLAG_LLC);
+               /**
+                * Since all memory is pinned and DMA vChannel
+                * ring has enough space, failure should be a
+                * rare case. If failure happens, it means DMA
+                * device encounters serious errors; in this
+                * case, please stop async data-path and check
+                * what has happened to DMA device.
+                */
+               if (unlikely(copy_idx < 0)) {
+                       if (!vhost_async_dma_copy_log) {
+                               VHOST_LOG_DATA(ERR, "(%s) DMA copy failed for channel %d:%u\n",
+                                               dev->ifname, dma_id, vchan_id);
+                               vhost_async_dma_copy_log = true;
+                       }
+                       return -1;
+               }
+       }
+
+       /**
+        * Only store packet completion flag address in the last copy's
+        * slot, and other slots are set to NULL.
+        */
+       dma_info->pkts_cmpl_flag_addr[copy_idx & ring_mask] = &vq->async->pkts_cmpl_flag[flag_idx];
+
+       return nr_segs;
+}
+
+static __rte_always_inline uint16_t
+vhost_async_dma_transfer(struct virtio_net *dev, struct vhost_virtqueue *vq,
+               int16_t dma_id, uint16_t vchan_id, uint16_t head_idx,
+               struct vhost_iov_iter *pkts, uint16_t nr_pkts)
+{
+       struct async_dma_vchan_info *dma_info = &dma_copy_track[dma_id].vchans[vchan_id];
+       int64_t ret, nr_copies = 0;
+       uint16_t pkt_idx;
+
+       rte_spinlock_lock(&dma_info->dma_lock);
+
+       for (pkt_idx = 0; pkt_idx < nr_pkts; pkt_idx++) {
+               ret = vhost_async_dma_transfer_one(dev, vq, dma_id, vchan_id, head_idx,
+                               &pkts[pkt_idx]);
+               if (unlikely(ret < 0))
+                       break;
+
+               nr_copies += ret;
+               head_idx++;
+               if (head_idx >= vq->size)
+                       head_idx -= vq->size;
+       }
+
+       if (likely(nr_copies > 0))
+               rte_dma_submit(dma_id, vchan_id);
+
+       rte_spinlock_unlock(&dma_info->dma_lock);
+
+       return pkt_idx;
+}
+
+static __rte_always_inline uint16_t
+vhost_async_dma_check_completed(struct virtio_net *dev, int16_t dma_id, uint16_t vchan_id,
+               uint16_t max_pkts)
+{
+       struct async_dma_vchan_info *dma_info = &dma_copy_track[dma_id].vchans[vchan_id];
+       uint16_t ring_mask = dma_info->ring_mask;
+       uint16_t last_idx = 0;
+       uint16_t nr_copies;
+       uint16_t copy_idx;
+       uint16_t i;
+       bool has_error = false;
+       static bool vhost_async_dma_complete_log;
+
+       rte_spinlock_lock(&dma_info->dma_lock);
+
+       /**
+        * Print error log for debugging, if DMA reports error during
+        * DMA transfer. We do not handle error in vhost level.
+        */
+       nr_copies = rte_dma_completed(dma_id, vchan_id, max_pkts, &last_idx, &has_error);
+       if (unlikely(!vhost_async_dma_complete_log && has_error)) {
+               VHOST_LOG_DATA(ERR, "(%s) DMA completion failure on channel %d:%u\n", dev->ifname,
+                               dma_id, vchan_id);
+               vhost_async_dma_complete_log = true;
+       } else if (nr_copies == 0) {
+               goto out;
+       }
+
+       copy_idx = last_idx - nr_copies + 1;
+       for (i = 0; i < nr_copies; i++) {
+               bool *flag;
+
+               flag = dma_info->pkts_cmpl_flag_addr[copy_idx & ring_mask];
+               if (flag) {
+                       /**
+                        * Mark the packet flag as received. The flag
+                        * could belong to another virtqueue but write
+                        * is atomic.
+                        */
+                       *flag = true;
+                       dma_info->pkts_cmpl_flag_addr[copy_idx & ring_mask] = NULL;
+               }
+               copy_idx++;
+       }
+
+out:
+       rte_spinlock_unlock(&dma_info->dma_lock);
+       return nr_copies;
+}
+
 static inline void
 do_data_copy_enqueue(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
@@ -792,202 +925,158 @@ 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 virtio_net *dev, 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;
-
-       if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
-               error = -1;
-               goto out;
-       }
-
-       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);
+       struct vhost_iov_iter *iter;
 
-       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;
+       if (unlikely(async->iovec_idx >= VHOST_MAX_ASYNC_VEC)) {
+               VHOST_LOG_DATA(ERR, "(%s) no more async iovec available\n", dev->ifname);
+               return -1;
        }
 
-       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;
-                       }
+       iter = async->iov_iter + async->iter_idx;
+       iter->iov = async->iovec + async->iovec_idx;
+       iter->nr_segs = 0;
 
-                       buf_addr = buf_vec[vec_idx].buf_addr;
-                       buf_iova = buf_vec[vec_idx].buf_iova;
-                       buf_len = buf_vec[vec_idx].buf_len;
+       return 0;
+}
 
-                       buf_offset = 0;
-                       buf_avail  = buf_len;
-               }
+static __rte_always_inline int
+async_iter_add_iovec(struct virtio_net *dev, struct vhost_async *async,
+               void *src, void *dst, size_t len)
+{
+       struct vhost_iov_iter *iter;
+       struct vhost_iovec *iovec;
 
-               /* done with current mbuf, get the next one */
-               if (mbuf_avail == 0) {
-                       m = m->next;
+       if (unlikely(async->iovec_idx >= VHOST_MAX_ASYNC_VEC)) {
+               static bool vhost_max_async_vec_log;
 
-                       mbuf_offset = 0;
-                       mbuf_avail  = rte_pktmbuf_data_len(m);
+               if (!vhost_max_async_vec_log) {
+                       VHOST_LOG_DATA(ERR, "(%s) no more async iovec available\n", dev->ifname);
+                       vhost_max_async_vec_log = true;
                }
 
-               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 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, size_t count,
-       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->offset = 0;
-       it->count = count;
+       struct vhost_async *async = vq->async;
+       uint64_t mapped_len;
+       uint32_t buf_offset = 0;
+       void *host_iova;
+
+       while (cpy_len) {
+               host_iova = (void *)(uintptr_t)gpa_to_first_hpa(dev,
+                               buf_iova + buf_offset, cpy_len, &mapped_len);
+               if (unlikely(!host_iova)) {
+                       VHOST_LOG_DATA(ERR, "(%s) %s: failed to get host iova.\n",
+                                      dev->ifname, __func__);
+                       return -1;
+               }
 
-       if (count) {
-               it->iov = vec;
-               it->nr_segs = nr_seg;
-       } else {
-               it->iov = 0;
-               it->nr_segs = 0;
+               if (unlikely(async_iter_add_iovec(dev, async,
+                                               (void *)(uintptr_t)rte_pktmbuf_iova_offset(m,
+                                                       mbuf_offset),
+                                               host_iova, (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;
-
-       uint32_t tlen = 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;
@@ -997,8 +1086,8 @@ async_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
        } 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);
+       VHOST_LOG_DATA(DEBUG, "(%s) RX: num merge buffers %d\n",
+               dev->ifname, num_buffers);
 
        if (unlikely(buf_len < dev->vhost_hlen)) {
                buf_offset = dev->vhost_hlen - buf_len;
@@ -1015,21 +1104,24 @@ 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(dev, 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 */
@@ -1037,7 +1129,7 @@ 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) {
@@ -1061,34 +1153,31 @@ 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);
-
-                       tlen += (uint32_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, tlen, 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
@@ -1145,7 +1234,7 @@ 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,
@@ -1179,19 +1268,18 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *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);
+                               "(%s) failed to get enough desc from vring\n",
+                               dev->ifname);
                        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,
+               VHOST_LOG_DATA(DEBUG, "(%s) current index %d | end index %d\n",
+                       dev->ifname, 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;
                }
@@ -1334,14 +1422,13 @@ virtio_dev_rx_single_packed(struct virtio_net *dev,
 
        if (unlikely(vhost_enqueue_single_packed(dev, vq, pkt, buf_vec,
                                                 &nr_descs) < 0)) {
-               VHOST_LOG_DATA(DEBUG,
-                               "(%d) failed to get enough desc from vring\n",
-                               dev->vid);
+               VHOST_LOG_DATA(DEBUG, "(%s) failed to get enough desc from vring\n",
+                               dev->ifname);
                return -1;
        }
 
-       VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
-                       dev->vid, vq->last_avail_idx,
+       VHOST_LOG_DATA(DEBUG, "(%s) current index %d | end index %d\n",
+                       dev->ifname, vq->last_avail_idx,
                        vq->last_avail_idx + nr_descs);
 
        vq_inc_last_avail_packed(vq, nr_descs);
@@ -1392,10 +1479,10 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
        struct vhost_virtqueue *vq;
        uint32_t nb_tx = 0;
 
-       VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
+       VHOST_LOG_DATA(DEBUG, "(%s) %s\n", dev->ifname, __func__);
        if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
-               VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
-                       dev->vid, __func__, queue_id);
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid virtqueue idx %d.\n",
+                       dev->ifname, __func__, queue_id);
                return 0;
        }
 
@@ -1442,9 +1529,8 @@ rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
                return 0;
 
        if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
-               VHOST_LOG_DATA(ERR,
-                       "(%d) %s: built-in vhost net backend is disabled.\n",
-                       dev->vid, __func__);
+               VHOST_LOG_DATA(ERR, "(%s) %s: built-in vhost net backend is disabled.\n",
+                       dev->ifname, __func__);
                return 0;
        }
 
@@ -1452,11 +1538,14 @@ 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
@@ -1493,23 +1582,20 @@ store_dma_desc_info_packed(struct vring_used_elem_packed *s_ring,
 }
 
 static __rte_noinline uint32_t
-virtio_dev_rx_async_submit_split(struct virtio_net *dev,
-       struct vhost_virtqueue *vq, uint16_t queue_id,
-       struct rte_mbuf **pkts, uint32_t count)
+virtio_dev_rx_async_submit_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
+               uint16_t queue_id, struct rte_mbuf **pkts, uint32_t count,
+               int16_t dma_id, uint16_t vchan_id)
 {
        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;
-       int32_t n_xfer;
-       uint16_t iovec_idx = 0, it_idx = 0, slot_idx = 0;
+       uint32_t pkt_err = 0;
+       uint16_t n_xfer;
+       uint16_t slot_idx = 0;
 
        /*
         * The ordering between avail index and desc reads need to be enforced.
@@ -1518,95 +1604,50 @@ 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, "(%s) failed to get enough desc from vring\n",
+                                       dev->ifname);
                        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);
+               VHOST_LOG_DATA(DEBUG, "(%s) current index %d | end index %d\n",
+                       dev->ifname, 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 = vhost_async_dma_transfer(dev, vq, dma_id, vchan_id, async->pkts_idx,
+                       async->iov_iter, pkt_idx);
 
+       pkt_err = pkt_idx - n_xfer;
        if (unlikely(pkt_err)) {
                uint16_t num_descs = 0;
 
+               VHOST_LOG_DATA(DEBUG, "(%s) %s: failed to transfer %u packets for queue %u.\n",
+                               dev->ifname, __func__, pkt_err, queue_id);
+
                /* update number of completed packets */
-               pkt_idx -= pkt_err;
+               pkt_idx = n_xfer;
 
                /* calculate the sum of descriptors to revert */
                while (pkt_err-- > 0) {
@@ -1628,7 +1669,11 @@ 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;
        }
@@ -1636,60 +1681,6 @@ virtio_dev_rx_async_submit_split(struct virtio_net *dev,
        return pkt_idx;
 }
 
-static __rte_always_inline void
-vhost_update_used_packed(struct vhost_virtqueue *vq,
-                       struct vring_used_elem_packed *shadow_ring,
-                       uint16_t count)
-{
-       int i;
-       uint16_t used_idx = vq->last_used_idx;
-       uint16_t head_idx = vq->last_used_idx;
-       uint16_t head_flags = 0;
-
-       if (count == 0)
-               return;
-
-       /* Split loop in two to save memory barriers */
-       for (i = 0; i < count; i++) {
-               vq->desc_packed[used_idx].id = shadow_ring[i].id;
-               vq->desc_packed[used_idx].len = shadow_ring[i].len;
-
-               used_idx += shadow_ring[i].count;
-               if (used_idx >= vq->size)
-                       used_idx -= vq->size;
-       }
-
-       /* The ordering for storing desc flags needs to be enforced. */
-       rte_atomic_thread_fence(__ATOMIC_RELEASE);
-
-       for (i = 0; i < count; i++) {
-               uint16_t flags;
-
-               if (vq->shadow_used_packed[i].len)
-                       flags = VRING_DESC_F_WRITE;
-               else
-                       flags = 0;
-
-               if (vq->used_wrap_counter) {
-                       flags |= VRING_DESC_F_USED;
-                       flags |= VRING_DESC_F_AVAIL;
-               } else {
-                       flags &= ~VRING_DESC_F_USED;
-                       flags &= ~VRING_DESC_F_AVAIL;
-               }
-
-               if (i > 0) {
-                       vq->desc_packed[vq->last_used_idx].flags = flags;
-               } else {
-                       head_idx = vq->last_used_idx;
-                       head_flags = flags;
-               }
-
-               vq_inc_last_used_packed(vq, shadow_ring[i].count);
-       }
-
-       vq->desc_packed[head_idx].flags = head_flags;
-}
 
 static __rte_always_inline int
 vhost_enqueue_async_packed(struct virtio_net *dev,
@@ -1697,9 +1688,7 @@ 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;
@@ -1746,8 +1735,7 @@ 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);
@@ -1757,19 +1745,18 @@ 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)) {
-               VHOST_LOG_DATA(DEBUG, "(%d) failed to get enough desc from vring\n", dev->vid);
+       if (unlikely(vhost_enqueue_async_packed(dev, vq, pkt, buf_vec,
+                                       nr_descs, nr_buffers) < 0)) {
+               VHOST_LOG_DATA(DEBUG, "(%s) failed to get enough desc from vring\n", dev->ifname);
                return -1;
        }
 
-       VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
-                       dev->vid, vq->last_avail_idx, vq->last_avail_idx + *nr_descs);
+       VHOST_LOG_DATA(DEBUG, "(%s) current index %d | end index %d\n",
+                       dev->ifname, vq->last_avail_idx, vq->last_avail_idx + *nr_descs);
 
        return 0;
 }
@@ -1801,24 +1788,20 @@ dma_error_handler_packed(struct vhost_virtqueue *vq, uint16_t slot_idx,
 }
 
 static __rte_noinline uint32_t
-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)
+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,
+               int16_t dma_id, uint16_t vchan_id)
 {
-       uint32_t pkt_idx = 0, pkt_burst_idx = 0;
+       uint32_t pkt_idx = 0;
        uint32_t remained = count;
-       int32_t n_xfer;
+       uint16_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]);
@@ -1826,73 +1809,34 @@ 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 = vhost_async_dma_transfer(dev, vq, dma_id, vchan_id, async->pkts_idx,
+                       async->iov_iter, pkt_idx);
+
+       async_iter_reset(async);
 
-       if (unlikely(pkt_err))
+       pkt_err = pkt_idx - n_xfer;
+       if (unlikely(pkt_err)) {
+               VHOST_LOG_DATA(DEBUG, "(%s) %s: failed to transfer %u packets for queue %u.\n",
+                               dev->ifname, __func__, pkt_err, queue_id);
                dma_error_handler_packed(vq, slot_idx, pkt_err, &pkt_idx);
+       }
 
        if (likely(vq->shadow_used_idx)) {
                /* keep used descriptors. */
@@ -1951,91 +1895,118 @@ write_back_completed_descs_packed(struct vhost_virtqueue *vq,
                                uint16_t n_buffers)
 {
        struct vhost_async *async = vq->async;
-       uint16_t nr_left = n_buffers;
-       uint16_t from, to;
+       uint16_t from = async->last_buffer_idx_packed;
+       uint16_t used_idx = vq->last_used_idx;
+       uint16_t head_idx = vq->last_used_idx;
+       uint16_t head_flags = 0;
+       uint16_t i;
 
-       do {
-               from = async->last_buffer_idx_packed;
-               to = (from + nr_left) % vq->size;
-               if (to > from) {
-                       vhost_update_used_packed(vq, async->buffers_packed + from, to - from);
-                       async->last_buffer_idx_packed += nr_left;
-                       nr_left = 0;
+       /* Split loop in two to save memory barriers */
+       for (i = 0; i < n_buffers; i++) {
+               vq->desc_packed[used_idx].id = async->buffers_packed[from].id;
+               vq->desc_packed[used_idx].len = async->buffers_packed[from].len;
+
+               used_idx += async->buffers_packed[from].count;
+               if (used_idx >= vq->size)
+                       used_idx -= vq->size;
+
+               from++;
+               if (from >= vq->size)
+                       from = 0;
+       }
+
+       /* The ordering for storing desc flags needs to be enforced. */
+       rte_atomic_thread_fence(__ATOMIC_RELEASE);
+
+       from = async->last_buffer_idx_packed;
+
+       for (i = 0; i < n_buffers; i++) {
+               uint16_t flags;
+
+               if (async->buffers_packed[from].len)
+                       flags = VRING_DESC_F_WRITE;
+               else
+                       flags = 0;
+
+               if (vq->used_wrap_counter) {
+                       flags |= VRING_DESC_F_USED;
+                       flags |= VRING_DESC_F_AVAIL;
                } else {
-                       vhost_update_used_packed(vq, async->buffers_packed + from,
-                               vq->size - from);
-                       async->last_buffer_idx_packed = 0;
-                       nr_left -= vq->size - from;
+                       flags &= ~VRING_DESC_F_USED;
+                       flags &= ~VRING_DESC_F_AVAIL;
                }
-       } while (nr_left > 0);
+
+               if (i > 0) {
+                       vq->desc_packed[vq->last_used_idx].flags = flags;
+               } else {
+                       head_idx = vq->last_used_idx;
+                       head_flags = flags;
+               }
+
+               vq_inc_last_used_packed(vq, async->buffers_packed[from].count);
+
+               from++;
+               if (from == vq->size)
+                       from = 0;
+       }
+
+       vq->desc_packed[head_idx].flags = head_flags;
+       async->last_buffer_idx_packed = from;
 }
 
 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 rte_mbuf **pkts, uint16_t count, int16_t dma_id,
+               uint16_t vchan_id)
 {
-       struct vhost_virtqueue *vq;
-       struct vhost_async *async;
-       struct async_inflight_info *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;
-
-       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",
-                               dev->vid, __func__, queue_id);
-                       n_pkts_cpl = 0;
-               }
+       struct vhost_virtqueue *vq = dev->virtqueue[queue_id];
+       struct vhost_async *async = vq->async;
+       struct async_inflight_info *pkts_info = async->pkts_info;
+       uint16_t nr_cpl_pkts = 0;
+       uint16_t n_descs = 0, n_buffers = 0;
+       uint16_t start_idx, from, i;
+
+       /* Check completed copies for the given DMA vChannel */
+       vhost_async_dma_check_completed(dev, dma_id, vchan_id, VHOST_DMA_MAX_COPY_COMPLETE);
+
+       start_idx = async_get_first_inflight_pkt_idx(vq);
+       /**
+        * Calculate the number of copy completed packets.
+        * Note that there may be completed packets even if
+        * no copies are reported done by the given DMA vChannel,
+        * as it's possible that a virtqueue uses multiple DMA
+        * vChannels.
+        */
+       from = start_idx;
+       while (vq->async->pkts_cmpl_flag[from] && count--) {
+               vq->async->pkts_cmpl_flag[from] = false;
+               from++;
+               if (from >= vq->size)
+                       from -= vq->size;
+               nr_cpl_pkts++;
        }
 
-       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 (nr_cpl_pkts == 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;
-               }
+       for (i = 0; i < nr_cpl_pkts; 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 -= nr_cpl_pkts;
 
        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 {
@@ -2048,12 +2019,13 @@ vhost_poll_enqueue_completed(struct virtio_net *dev, uint16_t queue_id,
                }
        }
 
-       return n_pkts_put;
+       return nr_cpl_pkts;
 }
 
 uint16_t
 rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id,
-               struct rte_mbuf **pkts, uint16_t count)
+               struct rte_mbuf **pkts, uint16_t count, int16_t dma_id,
+               uint16_t vchan_id)
 {
        struct virtio_net *dev = get_device(vid);
        struct vhost_virtqueue *vq;
@@ -2062,25 +2034,37 @@ rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id,
        if (unlikely(!dev))
                return 0;
 
-       VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
+       VHOST_LOG_DATA(DEBUG, "(%s) %s\n", dev->ifname, __func__);
        if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
-               VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
-                       dev->vid, __func__, queue_id);
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid virtqueue idx %d.\n",
+                       dev->ifname, __func__, queue_id);
+               return 0;
+       }
+
+       if (unlikely(!dma_copy_track[dma_id].vchans ||
+                               !dma_copy_track[dma_id].vchans[vchan_id].pkts_cmpl_flag_addr)) {
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid channel %d:%u.\n", dev->ifname, __func__,
+                              dma_id, vchan_id);
                return 0;
        }
 
        vq = dev->virtqueue[queue_id];
 
-       if (unlikely(!vq->async)) {
-               VHOST_LOG_DATA(ERR, "(%d) %s: async not registered for queue id %d.\n",
-                       dev->vid, __func__, queue_id);
+       if (!rte_spinlock_trylock(&vq->access_lock)) {
+               VHOST_LOG_DATA(DEBUG, "(%s) %s: virtqueue %u is busy.\n", dev->ifname, __func__,
+                               queue_id);
                return 0;
        }
 
-       rte_spinlock_lock(&vq->access_lock);
+       if (unlikely(!vq->async)) {
+               VHOST_LOG_DATA(ERR, "(%s) %s: async not registered for virtqueue %d.\n",
+                               dev->ifname, __func__, queue_id);
+               goto out;
+       }
 
-       n_pkts_cpl = vhost_poll_enqueue_completed(dev, queue_id, pkts, count);
+       n_pkts_cpl = vhost_poll_enqueue_completed(dev, queue_id, pkts, count, dma_id, vchan_id);
 
+out:
        rte_spinlock_unlock(&vq->access_lock);
 
        return n_pkts_cpl;
@@ -2088,7 +2072,8 @@ rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id,
 
 uint16_t
 rte_vhost_clear_queue_thread_unsafe(int vid, uint16_t queue_id,
-               struct rte_mbuf **pkts, uint16_t count)
+               struct rte_mbuf **pkts, uint16_t count, int16_t dma_id,
+               uint16_t vchan_id)
 {
        struct virtio_net *dev = get_device(vid);
        struct vhost_virtqueue *vq;
@@ -2097,37 +2082,51 @@ rte_vhost_clear_queue_thread_unsafe(int vid, uint16_t queue_id,
        if (!dev)
                return 0;
 
-       VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
+       VHOST_LOG_DATA(DEBUG, "(%s) %s\n", dev->ifname, __func__);
        if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
-               VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
-                       dev->vid, __func__, queue_id);
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid virtqueue idx %d.\n",
+                       dev->ifname, __func__, queue_id);
                return 0;
        }
 
        vq = dev->virtqueue[queue_id];
 
        if (unlikely(!vq->async)) {
-               VHOST_LOG_DATA(ERR, "(%d) %s: async not registered for queue id %d.\n",
-                       dev->vid, __func__, queue_id);
+               VHOST_LOG_DATA(ERR, "(%s) %s: async not registered for queue id %d.\n",
+                       dev->ifname, __func__, queue_id);
+               return 0;
+       }
+
+       if (unlikely(!dma_copy_track[dma_id].vchans ||
+                               !dma_copy_track[dma_id].vchans[vchan_id].pkts_cmpl_flag_addr)) {
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid channel %d:%u.\n", dev->ifname, __func__,
+                               dma_id, vchan_id);
                return 0;
        }
 
-       n_pkts_cpl = vhost_poll_enqueue_completed(dev, queue_id, pkts, count);
+       n_pkts_cpl = vhost_poll_enqueue_completed(dev, queue_id, pkts, count, dma_id, vchan_id);
 
        return n_pkts_cpl;
 }
 
 static __rte_always_inline uint32_t
 virtio_dev_rx_async_submit(struct virtio_net *dev, uint16_t queue_id,
-       struct rte_mbuf **pkts, uint32_t count)
+       struct rte_mbuf **pkts, uint32_t count, int16_t dma_id, uint16_t vchan_id)
 {
        struct vhost_virtqueue *vq;
        uint32_t nb_tx = 0;
 
-       VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
+       VHOST_LOG_DATA(DEBUG, "(%s) %s\n", dev->ifname, __func__);
        if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
-               VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
-                       dev->vid, __func__, queue_id);
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid virtqueue idx %d.\n",
+                       dev->ifname, __func__, queue_id);
+               return 0;
+       }
+
+       if (unlikely(!dma_copy_track[dma_id].vchans ||
+                               !dma_copy_track[dma_id].vchans[vchan_id].pkts_cmpl_flag_addr)) {
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid channel %d:%u.\n", dev->ifname, __func__,
+                              dma_id, vchan_id);
                return 0;
        }
 
@@ -2151,10 +2150,10 @@ virtio_dev_rx_async_submit(struct virtio_net *dev, uint16_t queue_id,
 
        if (vq_is_packed(dev))
                nb_tx = virtio_dev_rx_async_submit_packed(dev, vq, queue_id,
-                               pkts, count);
+                               pkts, count, dma_id, vchan_id);
        else
                nb_tx = virtio_dev_rx_async_submit_split(dev, vq, queue_id,
-                               pkts, count);
+                               pkts, count, dma_id, vchan_id);
 
 out:
        if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
@@ -2168,7 +2167,8 @@ out_access_unlock:
 
 uint16_t
 rte_vhost_submit_enqueue_burst(int vid, uint16_t queue_id,
-               struct rte_mbuf **pkts, uint16_t count)
+               struct rte_mbuf **pkts, uint16_t count, int16_t dma_id,
+               uint16_t vchan_id)
 {
        struct virtio_net *dev = get_device(vid);
 
@@ -2176,13 +2176,12 @@ rte_vhost_submit_enqueue_burst(int vid, uint16_t queue_id,
                return 0;
 
        if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
-               VHOST_LOG_DATA(ERR,
-                       "(%d) %s: built-in vhost net backend is disabled.\n",
-                       dev->vid, __func__);
+               VHOST_LOG_DATA(ERR, "(%s) %s: built-in vhost net backend is disabled.\n",
+                       dev->ifname, __func__);
                return 0;
        }
 
-       return virtio_dev_rx_async_submit(dev, queue_id, pkts, count);
+       return virtio_dev_rx_async_submit(dev, queue_id, pkts, count, dma_id, vchan_id);
 }
 
 static inline bool
@@ -2285,7 +2284,8 @@ error:
 }
 
 static __rte_always_inline void
-vhost_dequeue_offload_legacy(struct virtio_net_hdr *hdr, struct rte_mbuf *m)
+vhost_dequeue_offload_legacy(struct virtio_net *dev, struct virtio_net_hdr *hdr,
+               struct rte_mbuf *m)
 {
        uint8_t l4_proto = 0;
        struct rte_tcp_hdr *tcp_hdr = NULL;
@@ -2345,8 +2345,8 @@ vhost_dequeue_offload_legacy(struct virtio_net_hdr *hdr, struct rte_mbuf *m)
                        m->l4_len = sizeof(struct rte_udp_hdr);
                        break;
                default:
-                       VHOST_LOG_DATA(WARNING,
-                               "unsupported gso type %u.\n", hdr->gso_type);
+                       VHOST_LOG_DATA(WARNING, "(%s) unsupported gso type %u.\n",
+                                       dev->ifname, hdr->gso_type);
                        goto error;
                }
        }
@@ -2359,8 +2359,8 @@ error:
 }
 
 static __rte_always_inline void
-vhost_dequeue_offload(struct virtio_net_hdr *hdr, struct rte_mbuf *m,
-       bool legacy_ol_flags)
+vhost_dequeue_offload(struct virtio_net *dev, struct virtio_net_hdr *hdr,
+               struct rte_mbuf *m, bool legacy_ol_flags)
 {
        struct rte_net_hdr_lens hdr_lens;
        int l4_supported = 0;
@@ -2370,7 +2370,7 @@ vhost_dequeue_offload(struct virtio_net_hdr *hdr, struct rte_mbuf *m,
                return;
 
        if (legacy_ol_flags) {
-               vhost_dequeue_offload_legacy(hdr, m);
+               vhost_dequeue_offload_legacy(dev, hdr, m);
                return;
        }
 
@@ -2583,8 +2583,8 @@ copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
                if (mbuf_avail == 0) {
                        cur = rte_pktmbuf_alloc(mbuf_pool);
                        if (unlikely(cur == NULL)) {
-                               VHOST_LOG_DATA(ERR, "Failed to "
-                                       "allocate memory for mbuf.\n");
+                               VHOST_LOG_DATA(ERR, "(%s) failed to allocate memory for mbuf.\n",
+                                               dev->ifname);
                                error = -1;
                                goto out;
                        }
@@ -2604,7 +2604,7 @@ copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
        m->pkt_len    += mbuf_offset;
 
        if (hdr)
-               vhost_dequeue_offload(hdr, m, legacy_ol_flags);
+               vhost_dequeue_offload(dev, hdr, m, legacy_ol_flags);
 
 out:
 
@@ -2618,7 +2618,7 @@ virtio_dev_extbuf_free(void *addr __rte_unused, void *opaque)
 }
 
 static int
-virtio_dev_extbuf_alloc(struct rte_mbuf *pkt, uint32_t size)
+virtio_dev_extbuf_alloc(struct virtio_net *dev, struct rte_mbuf *pkt, uint32_t size)
 {
        struct rte_mbuf_ext_shared_info *shinfo = NULL;
        uint32_t total_len = RTE_PKTMBUF_HEADROOM + size;
@@ -2642,7 +2642,7 @@ virtio_dev_extbuf_alloc(struct rte_mbuf *pkt, uint32_t size)
                                                virtio_dev_extbuf_free, buf);
        if (unlikely(shinfo == NULL)) {
                rte_free(buf);
-               VHOST_LOG_DATA(ERR, "Failed to init shinfo\n");
+               VHOST_LOG_DATA(ERR, "(%s) failed to init shinfo\n", dev->ifname);
                return -1;
        }
 
@@ -2664,7 +2664,7 @@ virtio_dev_pktmbuf_prep(struct virtio_net *dev, struct rte_mbuf *pkt,
                return 0;
 
        /* attach an external buffer if supported */
-       if (dev->extbuf && !virtio_dev_extbuf_alloc(pkt, data_len))
+       if (dev->extbuf && !virtio_dev_extbuf_alloc(dev, pkt, data_len))
                return 0;
 
        /* check if chained buffers are allowed */
@@ -2696,12 +2696,12 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
        rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 
-       VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
+       VHOST_LOG_DATA(DEBUG, "(%s) %s\n", dev->ifname, __func__);
 
        count = RTE_MIN(count, MAX_PKT_BURST);
        count = RTE_MIN(count, free_entries);
-       VHOST_LOG_DATA(DEBUG, "(%d) about to dequeue %u buffers\n",
-                       dev->vid, count);
+       VHOST_LOG_DATA(DEBUG, "(%s) about to dequeue %u buffers\n",
+                       dev->ifname, count);
 
        if (rte_pktmbuf_alloc_bulk(mbuf_pool, pkts, count))
                return 0;
@@ -2730,9 +2730,8 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
                         * is required. Drop this packet.
                         */
                        if (!allocerr_warned) {
-                               VHOST_LOG_DATA(ERR,
-                                       "Failed mbuf alloc of size %d from %s on %s.\n",
-                                       buf_len, mbuf_pool->name, dev->ifname);
+                               VHOST_LOG_DATA(ERR, "(%s) failed mbuf alloc of size %d from %s.\n",
+                                       dev->ifname, buf_len, mbuf_pool->name);
                                allocerr_warned = true;
                        }
                        dropped += 1;
@@ -2744,8 +2743,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
                                mbuf_pool, legacy_ol_flags);
                if (unlikely(err)) {
                        if (!allocerr_warned) {
-                               VHOST_LOG_DATA(ERR,
-                                       "Failed to copy desc to mbuf on %s.\n",
+                               VHOST_LOG_DATA(ERR, "(%s) failed to copy desc to mbuf.\n",
                                        dev->ifname);
                                allocerr_warned = true;
                        }
@@ -2888,7 +2886,7 @@ virtio_dev_tx_batch_packed(struct virtio_net *dev,
        if (virtio_net_with_host_offload(dev)) {
                vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
                        hdr = (struct virtio_net_hdr *)(desc_addrs[i]);
-                       vhost_dequeue_offload(hdr, pkts[i], legacy_ol_flags);
+                       vhost_dequeue_offload(dev, hdr, pkts[i], legacy_ol_flags);
                }
        }
 
@@ -2927,9 +2925,8 @@ vhost_dequeue_single_packed(struct virtio_net *dev,
 
        if (unlikely(virtio_dev_pktmbuf_prep(dev, pkts, buf_len))) {
                if (!allocerr_warned) {
-                       VHOST_LOG_DATA(ERR,
-                               "Failed mbuf alloc of size %d from %s on %s.\n",
-                               buf_len, mbuf_pool->name, dev->ifname);
+                       VHOST_LOG_DATA(ERR, "(%s) failed mbuf alloc of size %d from %s.\n",
+                               dev->ifname, buf_len, mbuf_pool->name);
                        allocerr_warned = true;
                }
                return -1;
@@ -2939,8 +2936,7 @@ vhost_dequeue_single_packed(struct virtio_net *dev,
                                mbuf_pool, legacy_ol_flags);
        if (unlikely(err)) {
                if (!allocerr_warned) {
-                       VHOST_LOG_DATA(ERR,
-                               "Failed to copy desc to mbuf on %s.\n",
+                       VHOST_LOG_DATA(ERR, "(%s) failed to copy desc to mbuf.\n",
                                dev->ifname);
                        allocerr_warned = true;
                }
@@ -3056,16 +3052,14 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
                return 0;
 
        if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
-               VHOST_LOG_DATA(ERR,
-                       "(%d) %s: built-in vhost net backend is disabled.\n",
-                       dev->vid, __func__);
+               VHOST_LOG_DATA(ERR, "(%s) %s: built-in vhost net backend is disabled.\n",
+                               dev->ifname, __func__);
                return 0;
        }
 
        if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->nr_vring))) {
-               VHOST_LOG_DATA(ERR,
-                       "(%d) %s: invalid virtqueue idx %d.\n",
-                       dev->vid, __func__, queue_id);
+               VHOST_LOG_DATA(ERR, "(%s) %s: invalid virtqueue idx %d.\n",
+                               dev->ifname, __func__, queue_id);
                return 0;
        }
 
@@ -3110,7 +3104,7 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 
                rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
                if (rarp_mbuf == NULL) {
-                       VHOST_LOG_DATA(ERR, "Failed to make RARP packet.\n");
+                       VHOST_LOG_DATA(ERR, "(%s) failed to make RARP packet.\n", dev->ifname);
                        count = 0;
                        goto out;
                }