X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=lib%2Flibrte_vhost%2Fvirtio_net.c;h=f97ec76f9b9498957bc65857dda2a0f1953a8f94;hb=6d13ea8e8e49ab957deae2bba5ecf4a4bfe747d1;hp=6339296c706899fe9c17b910737f2810dcc98db8;hpb=fd68b4739d2c60d10023762603d2b65127d65fa1;p=dpdk.git

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 6339296c70..f97ec76f9b 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -37,52 +37,13 @@ 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 struct vring_desc *
-alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
-					 struct vring_desc *desc)
-{
-	struct vring_desc *idesc;
-	uint64_t src, dst;
-	uint64_t len, remain = desc->len;
-	uint64_t desc_addr = desc->addr;
-
-	idesc = rte_malloc(__func__, desc->len, 0);
-	if (unlikely(!idesc))
-		return 0;
-
-	dst = (uint64_t)(uintptr_t)idesc;
-
-	while (remain) {
-		len = remain;
-		src = vhost_iova_to_vva(dev, vq, desc_addr, &len,
-				VHOST_ACCESS_RO);
-		if (unlikely(!src || !len)) {
-			rte_free(idesc);
-			return 0;
-		}
-
-		rte_memcpy((void *)(uintptr_t)dst, (void *)(uintptr_t)src, len);
-
-		remain -= len;
-		dst += len;
-		desc_addr += len;
-	}
-
-	return idesc;
-}
-
 static __rte_always_inline void
-free_ind_table(struct vring_desc *idesc)
-{
-	rte_free(idesc);
-}
-
-static __rte_always_inline void
-do_flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq,
-			  uint16_t to, uint16_t from, uint16_t size)
+do_flush_shadow_used_ring_split(struct virtio_net *dev,
+			struct vhost_virtqueue *vq,
+			uint16_t to, uint16_t from, uint16_t size)
 {
 	rte_memcpy(&vq->used->ring[to],
-			&vq->shadow_used_ring[from],
+			&vq->shadow_used_split[from],
 			size * sizeof(struct vring_used_elem));
 	vhost_log_cache_used_vring(dev, vq,
 			offsetof(struct vring_used, ring[to]),
@@ -90,22 +51,22 @@ do_flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 }
 
 static __rte_always_inline void
-flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq)
+flush_shadow_used_ring_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
 	uint16_t used_idx = vq->last_used_idx & (vq->size - 1);
 
 	if (used_idx + vq->shadow_used_idx <= vq->size) {
-		do_flush_shadow_used_ring(dev, vq, used_idx, 0,
+		do_flush_shadow_used_ring_split(dev, vq, used_idx, 0,
 					  vq->shadow_used_idx);
 	} else {
 		uint16_t size;
 
 		/* update used ring interval [used_idx, vq->size] */
 		size = vq->size - used_idx;
-		do_flush_shadow_used_ring(dev, vq, used_idx, 0, size);
+		do_flush_shadow_used_ring_split(dev, vq, used_idx, 0, size);
 
 		/* update the left half used ring interval [0, left_size] */
-		do_flush_shadow_used_ring(dev, vq, 0, size,
+		do_flush_shadow_used_ring_split(dev, vq, 0, size,
 					  vq->shadow_used_idx - size);
 	}
 	vq->last_used_idx += vq->shadow_used_idx;
@@ -115,18 +76,97 @@ flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq)
 	vhost_log_cache_sync(dev, vq);
 
 	*(volatile uint16_t *)&vq->used->idx += vq->shadow_used_idx;
+	vq->shadow_used_idx = 0;
 	vhost_log_used_vring(dev, vq, offsetof(struct vring_used, idx),
 		sizeof(vq->used->idx));
 }
 
 static __rte_always_inline void
-update_shadow_used_ring(struct vhost_virtqueue *vq,
-			 uint16_t desc_idx, uint16_t len)
+update_shadow_used_ring_split(struct vhost_virtqueue *vq,
+			 uint16_t desc_idx, uint32_t len)
+{
+	uint16_t i = vq->shadow_used_idx++;
+
+	vq->shadow_used_split[i].id  = desc_idx;
+	vq->shadow_used_split[i].len = len;
+}
+
+static __rte_always_inline void
+flush_shadow_used_ring_packed(struct virtio_net *dev,
+			struct vhost_virtqueue *vq)
+{
+	int i;
+	uint16_t used_idx = vq->last_used_idx;
+	uint16_t head_idx = vq->last_used_idx;
+	uint16_t head_flags = 0;
+
+	/* Split loop in two to save memory barriers */
+	for (i = 0; i < vq->shadow_used_idx; i++) {
+		vq->desc_packed[used_idx].id = vq->shadow_used_packed[i].id;
+		vq->desc_packed[used_idx].len = vq->shadow_used_packed[i].len;
+
+		used_idx += vq->shadow_used_packed[i].count;
+		if (used_idx >= vq->size)
+			used_idx -= vq->size;
+	}
+
+	rte_smp_wmb();
+
+	for (i = 0; i < vq->shadow_used_idx; 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;
+
+			vhost_log_cache_used_vring(dev, vq,
+					vq->last_used_idx *
+					sizeof(struct vring_packed_desc),
+					sizeof(struct vring_packed_desc));
+		} else {
+			head_idx = vq->last_used_idx;
+			head_flags = flags;
+		}
+
+		vq->last_used_idx += vq->shadow_used_packed[i].count;
+		if (vq->last_used_idx >= vq->size) {
+			vq->used_wrap_counter ^= 1;
+			vq->last_used_idx -= vq->size;
+		}
+	}
+
+	vq->desc_packed[head_idx].flags = head_flags;
+
+	vhost_log_cache_used_vring(dev, vq,
+				head_idx *
+				sizeof(struct vring_packed_desc),
+				sizeof(struct vring_packed_desc));
+
+	vq->shadow_used_idx = 0;
+	vhost_log_cache_sync(dev, vq);
+}
+
+static __rte_always_inline void
+update_shadow_used_ring_packed(struct vhost_virtqueue *vq,
+			 uint16_t desc_idx, uint32_t len, uint16_t count)
 {
 	uint16_t i = vq->shadow_used_idx++;
 
-	vq->shadow_used_ring[i].id  = desc_idx;
-	vq->shadow_used_ring[i].len = len;
+	vq->shadow_used_packed[i].id  = desc_idx;
+	vq->shadow_used_packed[i].len = len;
+	vq->shadow_used_packed[i].count = count;
 }
 
 static inline void
@@ -141,6 +181,8 @@ do_data_copy_enqueue(struct virtio_net *dev, struct vhost_virtqueue *vq)
 		vhost_log_cache_write(dev, vq, elem[i].log_addr, elem[i].len);
 		PRINT_PACKET(dev, (uintptr_t)elem[i].dst, elem[i].len, 0);
 	}
+
+	vq->batch_copy_nb_elems = 0;
 }
 
 static inline void
@@ -152,6 +194,8 @@ do_data_copy_dequeue(struct vhost_virtqueue *vq)
 
 	for (i = 0; i < count; i++)
 		rte_memcpy(elem[i].dst, elem[i].src, elem[i].len);
+
+	vq->batch_copy_nb_elems = 0;
 }
 
 /* avoid write operation when necessary, to lessen cache issues */
@@ -222,22 +266,65 @@ virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
 }
 
 static __rte_always_inline int
-fill_vec_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
-			 uint32_t avail_idx, uint32_t *vec_idx,
+map_one_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
+		struct buf_vector *buf_vec, uint16_t *vec_idx,
+		uint64_t desc_iova, uint64_t desc_len, uint8_t perm)
+{
+	uint16_t vec_id = *vec_idx;
+
+	while (desc_len) {
+		uint64_t desc_addr;
+		uint64_t desc_chunck_len = desc_len;
+
+		if (unlikely(vec_id >= BUF_VECTOR_MAX))
+			return -1;
+
+		desc_addr = vhost_iova_to_vva(dev, vq,
+				desc_iova,
+				&desc_chunck_len,
+				perm);
+		if (unlikely(!desc_addr))
+			return -1;
+
+		buf_vec[vec_id].buf_iova = desc_iova;
+		buf_vec[vec_id].buf_addr = desc_addr;
+		buf_vec[vec_id].buf_len  = desc_chunck_len;
+
+		desc_len -= desc_chunck_len;
+		desc_iova += desc_chunck_len;
+		vec_id++;
+	}
+	*vec_idx = vec_id;
+
+	return 0;
+}
+
+static __rte_always_inline int
+fill_vec_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
+			 uint32_t avail_idx, uint16_t *vec_idx,
 			 struct buf_vector *buf_vec, uint16_t *desc_chain_head,
-			 uint16_t *desc_chain_len, uint8_t perm)
+			 uint32_t *desc_chain_len, uint8_t perm)
 {
 	uint16_t idx = vq->avail->ring[avail_idx & (vq->size - 1)];
-	uint32_t vec_id = *vec_idx;
+	uint16_t vec_id = *vec_idx;
 	uint32_t len    = 0;
-	uint64_t dlen, desc_avail, desc_iova;
+	uint64_t dlen;
+	uint32_t nr_descs = vq->size;
+	uint32_t cnt    = 0;
 	struct vring_desc *descs = vq->desc;
 	struct vring_desc *idesc = NULL;
 
+	if (unlikely(idx >= vq->size))
+		return -1;
+
 	*desc_chain_head = idx;
 
 	if (vq->desc[idx].flags & VRING_DESC_F_INDIRECT) {
 		dlen = vq->desc[idx].len;
+		nr_descs = dlen / sizeof(struct vring_desc);
+		if (unlikely(nr_descs > vq->size))
+			return -1;
+
 		descs = (struct vring_desc *)(uintptr_t)
 			vhost_iova_to_vva(dev, vq, vq->desc[idx].addr,
 						&dlen,
@@ -250,7 +337,8 @@ fill_vec_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			 * The indirect desc table is not contiguous
 			 * in process VA space, we have to copy it.
 			 */
-			idesc = alloc_copy_ind_table(dev, vq, &vq->desc[idx]);
+			idesc = alloc_copy_ind_table(dev, vq,
+					vq->desc[idx].addr, vq->desc[idx].len);
 			if (unlikely(!idesc))
 				return -1;
 
@@ -261,42 +349,18 @@ fill_vec_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	}
 
 	while (1) {
-		if (unlikely(idx >= vq->size)) {
+		if (unlikely(idx >= nr_descs || cnt++ >= nr_descs)) {
 			free_ind_table(idesc);
 			return -1;
 		}
 
-
 		len += descs[idx].len;
-		desc_avail = descs[idx].len;
-		desc_iova = descs[idx].addr;
-
-		while (desc_avail) {
-			uint64_t desc_addr;
-			uint64_t desc_chunck_len = desc_avail;
-
-			if (unlikely(vec_id >= BUF_VECTOR_MAX)) {
-				free_ind_table(idesc);
-				return -1;
-			}
-
-			desc_addr = vhost_iova_to_vva(dev, vq,
-					desc_iova,
-					&desc_chunck_len,
-					perm);
-			if (unlikely(!desc_addr)) {
-				free_ind_table(idesc);
-				return -1;
-			}
-
-			buf_vec[vec_id].buf_iova = desc_iova;
-			buf_vec[vec_id].buf_addr = desc_addr;
-			buf_vec[vec_id].buf_len  = desc_chunck_len;
-			buf_vec[vec_id].desc_idx = idx;
 
-			desc_avail -= desc_chunck_len;
-			desc_iova += desc_chunck_len;
-			vec_id++;
+		if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
+						descs[idx].addr, descs[idx].len,
+						perm))) {
+			free_ind_table(idesc);
+			return -1;
 		}
 
 		if ((descs[idx].flags & VRING_DESC_F_NEXT) == 0)
@@ -318,49 +382,227 @@ fill_vec_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
  * Returns -1 on fail, 0 on success
  */
 static inline int
-reserve_avail_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
+reserve_avail_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 				uint32_t size, struct buf_vector *buf_vec,
 				uint16_t *num_buffers, uint16_t avail_head,
 				uint16_t *nr_vec)
 {
 	uint16_t cur_idx;
-	uint32_t vec_idx = 0;
+	uint16_t vec_idx = 0;
 	uint16_t max_tries, tries = 0;
 
 	uint16_t head_idx = 0;
-	uint16_t len = 0;
+	uint32_t len = 0;
 
 	*num_buffers = 0;
 	cur_idx  = vq->last_avail_idx;
 
 	if (rxvq_is_mergeable(dev))
-		max_tries = vq->size;
+		max_tries = vq->size - 1;
 	else
 		max_tries = 1;
 
 	while (size > 0) {
 		if (unlikely(cur_idx == avail_head))
 			return -1;
+		/*
+		 * if we tried all available ring items, and still
+		 * can't get enough buf, it means something abnormal
+		 * happened.
+		 */
+		if (unlikely(++tries > max_tries))
+			return -1;
 
-		if (unlikely(fill_vec_buf(dev, vq, cur_idx, &vec_idx, buf_vec,
+		if (unlikely(fill_vec_buf_split(dev, vq, cur_idx,
+						&vec_idx, buf_vec,
 						&head_idx, &len,
 						VHOST_ACCESS_RW) < 0))
 			return -1;
 		len = RTE_MIN(len, size);
-		update_shadow_used_ring(vq, head_idx, len);
+		update_shadow_used_ring_split(vq, head_idx, len);
 		size -= len;
 
 		cur_idx++;
-		tries++;
 		*num_buffers += 1;
+	}
+
+	*nr_vec = vec_idx;
+
+	return 0;
+}
+
+static __rte_always_inline int
+fill_vec_buf_packed_indirect(struct virtio_net *dev,
+			struct vhost_virtqueue *vq,
+			struct vring_packed_desc *desc, uint16_t *vec_idx,
+			struct buf_vector *buf_vec, uint32_t *len, uint8_t perm)
+{
+	uint16_t i;
+	uint32_t nr_descs;
+	uint16_t vec_id = *vec_idx;
+	uint64_t dlen;
+	struct vring_packed_desc *descs, *idescs = NULL;
+
+	dlen = desc->len;
+	descs = (struct vring_packed_desc *)(uintptr_t)
+		vhost_iova_to_vva(dev, vq, desc->addr, &dlen, VHOST_ACCESS_RO);
+	if (unlikely(!descs))
+		return -1;
+
+	if (unlikely(dlen < desc->len)) {
+		/*
+		 * The indirect desc table is not contiguous
+		 * in process VA space, we have to copy it.
+		 */
+		idescs = alloc_copy_ind_table(dev, vq, desc->addr, desc->len);
+		if (unlikely(!idescs))
+			return -1;
+
+		descs = idescs;
+	}
+
+	nr_descs =  desc->len / sizeof(struct vring_packed_desc);
+	if (unlikely(nr_descs >= vq->size)) {
+		free_ind_table(idescs);
+		return -1;
+	}
+
+	for (i = 0; i < nr_descs; i++) {
+		if (unlikely(vec_id >= BUF_VECTOR_MAX)) {
+			free_ind_table(idescs);
+			return -1;
+		}
+
+		*len += descs[i].len;
+		if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
+						descs[i].addr, descs[i].len,
+						perm)))
+			return -1;
+	}
+	*vec_idx = vec_id;
+
+	if (unlikely(!!idescs))
+		free_ind_table(idescs);
+
+	return 0;
+}
+
+static __rte_always_inline int
+fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+				uint16_t avail_idx, uint16_t *desc_count,
+				struct buf_vector *buf_vec, uint16_t *vec_idx,
+				uint16_t *buf_id, uint32_t *len, uint8_t perm)
+{
+	bool wrap_counter = vq->avail_wrap_counter;
+	struct vring_packed_desc *descs = vq->desc_packed;
+	uint16_t vec_id = *vec_idx;
+
+	if (avail_idx < vq->last_avail_idx)
+		wrap_counter ^= 1;
+
+	if (unlikely(!desc_is_avail(&descs[avail_idx], wrap_counter)))
+		return -1;
+
+	/*
+	 * The ordering between desc flags and desc
+	 * content reads need to be enforced.
+	 */
+	rte_smp_rmb();
+
+	*desc_count = 0;
+	*len = 0;
+
+	while (1) {
+		if (unlikely(vec_id >= BUF_VECTOR_MAX))
+			return -1;
+
+		if (unlikely(*desc_count >= vq->size))
+			return -1;
+
+		*desc_count += 1;
+		*buf_id = descs[avail_idx].id;
+
+		if (descs[avail_idx].flags & VRING_DESC_F_INDIRECT) {
+			if (unlikely(fill_vec_buf_packed_indirect(dev, vq,
+							&descs[avail_idx],
+							&vec_id, buf_vec,
+							len, perm) < 0))
+				return -1;
+		} else {
+			*len += descs[avail_idx].len;
+
+			if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
+							descs[avail_idx].addr,
+							descs[avail_idx].len,
+							perm)))
+				return -1;
+		}
+
+		if ((descs[avail_idx].flags & VRING_DESC_F_NEXT) == 0)
+			break;
+
+		if (++avail_idx >= vq->size) {
+			avail_idx -= vq->size;
+			wrap_counter ^= 1;
+		}
+	}
+
+	*vec_idx = vec_id;
+
+	return 0;
+}
+
+/*
+ * Returns -1 on fail, 0 on success
+ */
+static inline int
+reserve_avail_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+				uint32_t size, struct buf_vector *buf_vec,
+				uint16_t *nr_vec, uint16_t *num_buffers,
+				uint16_t *nr_descs)
+{
+	uint16_t avail_idx;
+	uint16_t vec_idx = 0;
+	uint16_t max_tries, tries = 0;
+
+	uint16_t buf_id = 0;
+	uint32_t len = 0;
+	uint16_t desc_count;
+
+	*num_buffers = 0;
+	avail_idx = vq->last_avail_idx;
+
+	if (rxvq_is_mergeable(dev))
+		max_tries = vq->size - 1;
+	else
+		max_tries = 1;
 
+	while (size > 0) {
 		/*
 		 * if we tried all available ring items, and still
 		 * can't get enough buf, it means something abnormal
 		 * happened.
 		 */
-		if (unlikely(tries > max_tries))
+		if (unlikely(++tries > max_tries))
+			return -1;
+
+		if (unlikely(fill_vec_buf_packed(dev, vq,
+						avail_idx, &desc_count,
+						buf_vec, &vec_idx,
+						&buf_id, &len,
+						VHOST_ACCESS_RW) < 0))
 			return -1;
+
+		len = RTE_MIN(len, size);
+		update_shadow_used_ring_packed(vq, buf_id, len, desc_count);
+		size -= len;
+
+		avail_idx += desc_count;
+		if (avail_idx >= vq->size)
+			avail_idx -= vq->size;
+
+		*nr_descs += desc_count;
+		*num_buffers += 1;
 	}
 
 	*nr_vec = vec_idx;
@@ -393,6 +635,9 @@ copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	buf_iova = buf_vec[vec_idx].buf_iova;
 	buf_len = buf_vec[vec_idx].buf_len;
 
+	if (nr_vec > 1)
+		rte_prefetch0((void *)(uintptr_t)buf_vec[1].buf_addr);
+
 	if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
 		error = -1;
 		goto out;
@@ -404,7 +649,6 @@ copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		hdr = &tmp_hdr;
 	else
 		hdr = (struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)hdr_addr;
-	rte_prefetch0((void *)(uintptr_t)hdr_addr);
 
 	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) RX: num merge buffers %d\n",
 		dev->vid, num_buffers);
@@ -436,8 +680,10 @@ copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			buf_iova = buf_vec[vec_idx].buf_iova;
 			buf_len = buf_vec[vec_idx].buf_len;
 
-			/* Prefetch buffer address. */
-			rte_prefetch0((void *)(uintptr_t)buf_addr);
+			/* Prefetch next buffer address. */
+			if (vec_idx + 1 < nr_vec)
+				rte_prefetch0((void *)(uintptr_t)
+						buf_vec[vec_idx + 1].buf_addr);
 			buf_offset = 0;
 			buf_avail  = buf_len;
 		}
@@ -464,7 +710,8 @@ copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 				uint16_t hdr_vec_idx = 0;
 
 				while (remain) {
-					len = remain;
+					len = RTE_MIN(remain,
+						buf_vec[hdr_vec_idx].buf_len);
 					dst = buf_vec[hdr_vec_idx].buf_addr;
 					rte_memcpy((void *)(uintptr_t)dst,
 							(void *)(uintptr_t)src,
@@ -491,7 +738,7 @@ copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			hdr_addr = 0;
 		}
 
-		cpy_len = RTE_MIN(buf_len, mbuf_avail);
+		cpy_len = RTE_MIN(buf_avail, mbuf_avail);
 
 		if (likely(cpy_len > MAX_BATCH_LEN ||
 					vq->batch_copy_nb_elems >= vq->size)) {
@@ -525,53 +772,80 @@ out:
 }
 
 static __rte_always_inline uint32_t
-virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
+virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	struct rte_mbuf **pkts, uint32_t count)
 {
-	struct vhost_virtqueue *vq;
 	uint32_t pkt_idx = 0;
 	uint16_t num_buffers;
 	struct buf_vector buf_vec[BUF_VECTOR_MAX];
 	uint16_t avail_head;
 
-	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
-	if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
-		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
-			dev->vid, __func__, queue_id);
-		return 0;
-	}
+	avail_head = *((volatile uint16_t *)&vq->avail->idx);
 
-	vq = dev->virtqueue[queue_id];
+	/*
+	 * The ordering between avail index and
+	 * desc reads needs to be enforced.
+	 */
+	rte_smp_rmb();
 
-	rte_spinlock_lock(&vq->access_lock);
+	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 
-	if (unlikely(vq->enabled == 0))
-		goto out_access_unlock;
+	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 (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
-		vhost_user_iotlb_rd_lock(vq);
+		if (unlikely(reserve_avail_buf_split(dev, vq,
+						pkt_len, buf_vec, &num_buffers,
+						avail_head, &nr_vec) < 0)) {
+			VHOST_LOG_DEBUG(VHOST_DATA,
+				"(%d) failed to get enough desc from vring\n",
+				dev->vid);
+			vq->shadow_used_idx -= num_buffers;
+			break;
+		}
 
-	if (unlikely(vq->access_ok == 0))
-		if (unlikely(vring_translate(dev, vq) < 0))
-			goto out;
+		rte_prefetch0((void *)(uintptr_t)buf_vec[0].buf_addr);
 
-	count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
-	if (count == 0)
-		goto out;
+		VHOST_LOG_DEBUG(VHOST_DATA, "(%d) current index %d | end index %d\n",
+			dev->vid, vq->last_avail_idx,
+			vq->last_avail_idx + num_buffers);
 
-	vq->batch_copy_nb_elems = 0;
+		if (copy_mbuf_to_desc(dev, vq, pkts[pkt_idx],
+						buf_vec, nr_vec,
+						num_buffers) < 0) {
+			vq->shadow_used_idx -= num_buffers;
+			break;
+		}
 
-	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
+		vq->last_avail_idx += num_buffers;
+	}
+
+	do_data_copy_enqueue(dev, vq);
+
+	if (likely(vq->shadow_used_idx)) {
+		flush_shadow_used_ring_split(dev, vq);
+		vhost_vring_call_split(dev, vq);
+	}
+
+	return pkt_idx;
+}
+
+static __rte_always_inline uint32_t
+virtio_dev_rx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+	struct rte_mbuf **pkts, uint32_t count)
+{
+	uint32_t pkt_idx = 0;
+	uint16_t num_buffers;
+	struct buf_vector buf_vec[BUF_VECTOR_MAX];
 
-	vq->shadow_used_idx = 0;
-	avail_head = *((volatile uint16_t *)&vq->avail->idx);
 	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;
+		uint16_t nr_descs = 0;
 
-		if (unlikely(reserve_avail_buf(dev, vq,
-						pkt_len, buf_vec, &num_buffers,
-						avail_head, &nr_vec) < 0)) {
+		if (unlikely(reserve_avail_buf_packed(dev, vq,
+						pkt_len, buf_vec, &nr_vec,
+						&num_buffers, &nr_descs) < 0)) {
 			VHOST_LOG_DEBUG(VHOST_DATA,
 				"(%d) failed to get enough desc from vring\n",
 				dev->vid);
@@ -579,6 +853,8 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 			break;
 		}
 
+		rte_prefetch0((void *)(uintptr_t)buf_vec[0].buf_addr);
+
 		VHOST_LOG_DEBUG(VHOST_DATA, "(%d) current index %d | end index %d\n",
 			dev->vid, vq->last_avail_idx,
 			vq->last_avail_idx + num_buffers);
@@ -590,15 +866,59 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 			break;
 		}
 
-		vq->last_avail_idx += num_buffers;
+		vq->last_avail_idx += nr_descs;
+		if (vq->last_avail_idx >= vq->size) {
+			vq->last_avail_idx -= vq->size;
+			vq->avail_wrap_counter ^= 1;
+		}
 	}
 
 	do_data_copy_enqueue(dev, vq);
 
-	if (likely(vq->shadow_used_idx)) {
-		flush_shadow_used_ring(dev, vq);
-		vhost_vring_call(dev, vq);
-	}
+	if (likely(vq->shadow_used_idx)) {
+		flush_shadow_used_ring_packed(dev, vq);
+		vhost_vring_call_packed(dev, vq);
+	}
+
+	return pkt_idx;
+}
+
+static __rte_always_inline uint32_t
+virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
+	struct rte_mbuf **pkts, uint32_t count)
+{
+	struct vhost_virtqueue *vq;
+	uint32_t nb_tx = 0;
+
+	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
+	if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
+		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
+			dev->vid, __func__, queue_id);
+		return 0;
+	}
+
+	vq = dev->virtqueue[queue_id];
+
+	rte_spinlock_lock(&vq->access_lock);
+
+	if (unlikely(vq->enabled == 0))
+		goto out_access_unlock;
+
+	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
+		vhost_user_iotlb_rd_lock(vq);
+
+	if (unlikely(vq->access_ok == 0))
+		if (unlikely(vring_translate(dev, vq) < 0))
+			goto out;
+
+	count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
+	if (count == 0)
+		goto out;
+
+	if (vq_is_packed(dev))
+		nb_tx = virtio_dev_rx_packed(dev, vq, pkts, count);
+	else
+		nb_tx = virtio_dev_rx_split(dev, vq, pkts, count);
 
 out:
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
@@ -607,7 +927,7 @@ out:
 out_access_unlock:
 	rte_spinlock_unlock(&vq->access_lock);
 
-	return pkt_idx;
+	return nb_tx;
 }
 
 uint16_t
@@ -649,18 +969,19 @@ parse_ethernet(struct rte_mbuf *m, uint16_t *l4_proto, void **l4_hdr)
 	struct ipv4_hdr *ipv4_hdr;
 	struct ipv6_hdr *ipv6_hdr;
 	void *l3_hdr = NULL;
-	struct ether_hdr *eth_hdr;
+	struct rte_ether_hdr *eth_hdr;
 	uint16_t ethertype;
 
-	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
+	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 
-	m->l2_len = sizeof(struct ether_hdr);
+	m->l2_len = sizeof(struct rte_ether_hdr);
 	ethertype = rte_be_to_cpu_16(eth_hdr->ether_type);
 
 	if (ethertype == ETHER_TYPE_VLAN) {
-		struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
+		struct rte_vlan_hdr *vlan_hdr =
+			(struct rte_vlan_hdr *)(eth_hdr + 1);
 
-		m->l2_len += sizeof(struct vlan_hdr);
+		m->l2_len += sizeof(struct rte_vlan_hdr);
 		ethertype = rte_be_to_cpu_16(vlan_hdr->eth_proto);
 	}
 
@@ -743,12 +1064,6 @@ vhost_dequeue_offload(struct virtio_net_hdr *hdr, struct rte_mbuf *m)
 	}
 }
 
-static __rte_always_inline void
-put_zmbuf(struct zcopy_mbuf *zmbuf)
-{
-	zmbuf->in_use = 0;
-}
-
 static __rte_always_inline int
 copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		  struct buf_vector *buf_vec, uint16_t nr_vec,
@@ -791,7 +1106,8 @@ copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			 * in a contiguous virtual area.
 			 */
 			while (remain) {
-				len = remain;
+				len = RTE_MIN(remain,
+					buf_vec[hdr_vec_idx].buf_len);
 				src = buf_vec[hdr_vec_idx].buf_addr;
 				rte_memcpy((void *)(uintptr_t)dst,
 						   (void *)(uintptr_t)src, len);
@@ -902,7 +1218,13 @@ copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			buf_iova = buf_vec[vec_idx].buf_iova;
 			buf_len = buf_vec[vec_idx].buf_len;
 
-			rte_prefetch0((void *)(uintptr_t)buf_addr);
+			/*
+			 * Prefecth desc n + 1 buffer while
+			 * desc n buffer is processed.
+			 */
+			if (vec_idx + 1 < nr_vec)
+				rte_prefetch0((void *)(uintptr_t)
+						buf_vec[vec_idx + 1].buf_addr);
 
 			buf_offset = 0;
 			buf_avail  = buf_len;
@@ -979,90 +1301,146 @@ again:
 	return NULL;
 }
 
-static __rte_always_inline bool
-mbuf_is_consumed(struct rte_mbuf *m)
+static __rte_always_inline uint16_t
+virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
+	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
 {
-	while (m) {
-		if (rte_mbuf_refcnt_read(m) > 1)
-			return false;
-		m = m->next;
-	}
+	uint16_t i;
+	uint16_t free_entries;
 
-	return true;
-}
+	if (unlikely(dev->dequeue_zero_copy)) {
+		struct zcopy_mbuf *zmbuf, *next;
 
-static __rte_always_inline void
-restore_mbuf(struct rte_mbuf *m)
-{
-	uint32_t mbuf_size, priv_size;
+		for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
+		     zmbuf != NULL; zmbuf = next) {
+			next = TAILQ_NEXT(zmbuf, next);
 
-	while (m) {
-		priv_size = rte_pktmbuf_priv_size(m->pool);
-		mbuf_size = sizeof(struct rte_mbuf) + priv_size;
-		/* start of buffer is after mbuf structure and priv data */
+			if (mbuf_is_consumed(zmbuf->mbuf)) {
+				update_shadow_used_ring_split(vq,
+						zmbuf->desc_idx, 0);
+				TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
+				restore_mbuf(zmbuf->mbuf);
+				rte_pktmbuf_free(zmbuf->mbuf);
+				put_zmbuf(zmbuf);
+				vq->nr_zmbuf -= 1;
+			}
+		}
 
-		m->buf_addr = (char *)m + mbuf_size;
-		m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
-		m = m->next;
+		if (likely(vq->shadow_used_idx)) {
+			flush_shadow_used_ring_split(dev, vq);
+			vhost_vring_call_split(dev, vq);
+		}
 	}
-}
 
-uint16_t
-rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
-	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
-{
-	struct virtio_net *dev;
-	struct rte_mbuf *rarp_mbuf = NULL;
-	struct vhost_virtqueue *vq;
-	uint32_t i = 0;
-	uint16_t free_entries;
-
-	dev = get_device(vid);
-	if (!dev)
+	free_entries = *((volatile uint16_t *)&vq->avail->idx) -
+			vq->last_avail_idx;
+	if (free_entries == 0)
 		return 0;
 
-	if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
-		RTE_LOG(ERR, VHOST_DATA,
-			"(%d) %s: built-in vhost net backend is disabled.\n",
-			dev->vid, __func__);
-		return 0;
-	}
+	/*
+	 * The ordering between avail index and
+	 * desc reads needs to be enforced.
+	 */
+	rte_smp_rmb();
 
-	if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->nr_vring))) {
-		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
-			dev->vid, __func__, queue_id);
-		return 0;
-	}
+	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 
-	vq = dev->virtqueue[queue_id];
+	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
 
-	if (unlikely(rte_spinlock_trylock(&vq->access_lock) == 0))
-		return 0;
+	count = RTE_MIN(count, MAX_PKT_BURST);
+	count = RTE_MIN(count, free_entries);
+	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) about to dequeue %u buffers\n",
+			dev->vid, count);
 
-	if (unlikely(vq->enabled == 0))
-		goto out_access_unlock;
+	for (i = 0; i < count; i++) {
+		struct buf_vector buf_vec[BUF_VECTOR_MAX];
+		uint16_t head_idx;
+		uint32_t dummy_len;
+		uint16_t nr_vec = 0;
+		int err;
 
-	vq->batch_copy_nb_elems = 0;
-	vq->shadow_used_idx = 0;
+		if (unlikely(fill_vec_buf_split(dev, vq,
+						vq->last_avail_idx + i,
+						&nr_vec, buf_vec,
+						&head_idx, &dummy_len,
+						VHOST_ACCESS_RO) < 0))
+			break;
 
-	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
-		vhost_user_iotlb_rd_lock(vq);
+		if (likely(dev->dequeue_zero_copy == 0))
+			update_shadow_used_ring_split(vq, head_idx, 0);
 
-	if (unlikely(vq->access_ok == 0))
-		if (unlikely(vring_translate(dev, vq) < 0))
-			goto out;
+		rte_prefetch0((void *)(uintptr_t)buf_vec[0].buf_addr);
+
+		pkts[i] = rte_pktmbuf_alloc(mbuf_pool);
+		if (unlikely(pkts[i] == NULL)) {
+			RTE_LOG(ERR, VHOST_DATA,
+				"Failed to allocate memory for mbuf.\n");
+			break;
+		}
+
+		err = copy_desc_to_mbuf(dev, vq, buf_vec, nr_vec, pkts[i],
+				mbuf_pool);
+		if (unlikely(err)) {
+			rte_pktmbuf_free(pkts[i]);
+			break;
+		}
+
+		if (unlikely(dev->dequeue_zero_copy)) {
+			struct zcopy_mbuf *zmbuf;
+
+			zmbuf = get_zmbuf(vq);
+			if (!zmbuf) {
+				rte_pktmbuf_free(pkts[i]);
+				break;
+			}
+			zmbuf->mbuf = pkts[i];
+			zmbuf->desc_idx = head_idx;
+
+			/*
+			 * Pin lock the mbuf; we will check later to see
+			 * whether the mbuf is freed (when we are the last
+			 * user) or not. If that's the case, we then could
+			 * update the used ring safely.
+			 */
+			rte_mbuf_refcnt_update(pkts[i], 1);
+
+			vq->nr_zmbuf += 1;
+			TAILQ_INSERT_TAIL(&vq->zmbuf_list, zmbuf, next);
+		}
+	}
+	vq->last_avail_idx += i;
+
+	if (likely(dev->dequeue_zero_copy == 0)) {
+		do_data_copy_dequeue(vq);
+		if (unlikely(i < count))
+			vq->shadow_used_idx = i;
+		if (likely(vq->shadow_used_idx)) {
+			flush_shadow_used_ring_split(dev, vq);
+			vhost_vring_call_split(dev, vq);
+		}
+	}
+
+	return i;
+}
+
+static __rte_always_inline uint16_t
+virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
+{
+	uint16_t i;
 
 	if (unlikely(dev->dequeue_zero_copy)) {
 		struct zcopy_mbuf *zmbuf, *next;
-		int nr_updated = 0;
 
 		for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 		     zmbuf != NULL; zmbuf = next) {
 			next = TAILQ_NEXT(zmbuf, next);
 
 			if (mbuf_is_consumed(zmbuf->mbuf)) {
-				update_shadow_used_ring(vq, zmbuf->desc_idx, 0);
-				nr_updated += 1;
+				update_shadow_used_ring_packed(vq,
+						zmbuf->desc_idx,
+						0,
+						zmbuf->desc_count);
 
 				TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 				restore_mbuf(zmbuf->mbuf);
@@ -1072,67 +1450,37 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 			}
 		}
 
-		flush_shadow_used_ring(dev, vq);
-		vhost_vring_call(dev, vq);
-		vq->shadow_used_idx = 0;
-	}
-
-	/*
-	 * Construct a RARP broadcast packet, and inject it to the "pkts"
-	 * array, to looks like that guest actually send such packet.
-	 *
-	 * Check user_send_rarp() for more information.
-	 *
-	 * broadcast_rarp shares a cacheline in the virtio_net structure
-	 * with some fields that are accessed during enqueue and
-	 * rte_atomic16_cmpset() causes a write if using cmpxchg. This could
-	 * result in false sharing between enqueue and dequeue.
-	 *
-	 * Prevent unnecessary false sharing by reading broadcast_rarp first
-	 * and only performing cmpset if the read indicates it is likely to
-	 * be set.
-	 */
-
-	if (unlikely(rte_atomic16_read(&dev->broadcast_rarp) &&
-			rte_atomic16_cmpset((volatile uint16_t *)
-				&dev->broadcast_rarp.cnt, 1, 0))) {
-
-		rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
-		if (rarp_mbuf == NULL) {
-			RTE_LOG(ERR, VHOST_DATA,
-				"Failed to make RARP packet.\n");
-			return 0;
+		if (likely(vq->shadow_used_idx)) {
+			flush_shadow_used_ring_packed(dev, vq);
+			vhost_vring_call_packed(dev, vq);
 		}
-		count -= 1;
 	}
 
-	free_entries = *((volatile uint16_t *)&vq->avail->idx) -
-			vq->last_avail_idx;
-	if (free_entries == 0)
-		goto out;
-
 	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
 
 	count = RTE_MIN(count, MAX_PKT_BURST);
-	count = RTE_MIN(count, free_entries);
 	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) about to dequeue %u buffers\n",
 			dev->vid, count);
 
 	for (i = 0; i < count; i++) {
 		struct buf_vector buf_vec[BUF_VECTOR_MAX];
-		uint16_t head_idx, dummy_len;
-		uint32_t nr_vec = 0;
+		uint16_t buf_id;
+		uint32_t dummy_len;
+		uint16_t desc_count, nr_vec = 0;
 		int err;
 
-		if (unlikely(fill_vec_buf(dev, vq,
-						vq->last_avail_idx + i,
-						&nr_vec, buf_vec,
-						&head_idx, &dummy_len,
+		if (unlikely(fill_vec_buf_packed(dev, vq,
+						vq->last_avail_idx, &desc_count,
+						buf_vec, &nr_vec,
+						&buf_id, &dummy_len,
 						VHOST_ACCESS_RO) < 0))
 			break;
 
 		if (likely(dev->dequeue_zero_copy == 0))
-			update_shadow_used_ring(vq, head_idx, 0);
+			update_shadow_used_ring_packed(vq, buf_id, 0,
+					desc_count);
+
+		rte_prefetch0((void *)(uintptr_t)buf_vec[0].buf_addr);
 
 		pkts[i] = rte_pktmbuf_alloc(mbuf_pool);
 		if (unlikely(pkts[i] == NULL)) {
@@ -1157,7 +1505,8 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 				break;
 			}
 			zmbuf->mbuf = pkts[i];
-			zmbuf->desc_idx = head_idx;
+			zmbuf->desc_idx = buf_id;
+			zmbuf->desc_count = desc_count;
 
 			/*
 			 * Pin lock the mbuf; we will check later to see
@@ -1170,17 +1519,105 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 			vq->nr_zmbuf += 1;
 			TAILQ_INSERT_TAIL(&vq->zmbuf_list, zmbuf, next);
 		}
+
+		vq->last_avail_idx += desc_count;
+		if (vq->last_avail_idx >= vq->size) {
+			vq->last_avail_idx -= vq->size;
+			vq->avail_wrap_counter ^= 1;
+		}
 	}
-	vq->last_avail_idx += i;
 
 	if (likely(dev->dequeue_zero_copy == 0)) {
 		do_data_copy_dequeue(vq);
 		if (unlikely(i < count))
 			vq->shadow_used_idx = i;
-		flush_shadow_used_ring(dev, vq);
-		vhost_vring_call(dev, vq);
+		if (likely(vq->shadow_used_idx)) {
+			flush_shadow_used_ring_packed(dev, vq);
+			vhost_vring_call_packed(dev, vq);
+		}
+	}
+
+	return i;
+}
+
+uint16_t
+rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
+	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
+{
+	struct virtio_net *dev;
+	struct rte_mbuf *rarp_mbuf = NULL;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (!dev)
+		return 0;
+
+	if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
+		RTE_LOG(ERR, VHOST_DATA,
+			"(%d) %s: built-in vhost net backend is disabled.\n",
+			dev->vid, __func__);
+		return 0;
+	}
+
+	if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->nr_vring))) {
+		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
+			dev->vid, __func__, queue_id);
+		return 0;
+	}
+
+	vq = dev->virtqueue[queue_id];
+
+	if (unlikely(rte_spinlock_trylock(&vq->access_lock) == 0))
+		return 0;
+
+	if (unlikely(vq->enabled == 0)) {
+		count = 0;
+		goto out_access_unlock;
+	}
+
+	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
+		vhost_user_iotlb_rd_lock(vq);
+
+	if (unlikely(vq->access_ok == 0))
+		if (unlikely(vring_translate(dev, vq) < 0)) {
+			count = 0;
+			goto out;
+		}
+
+	/*
+	 * Construct a RARP broadcast packet, and inject it to the "pkts"
+	 * array, to looks like that guest actually send such packet.
+	 *
+	 * Check user_send_rarp() for more information.
+	 *
+	 * broadcast_rarp shares a cacheline in the virtio_net structure
+	 * with some fields that are accessed during enqueue and
+	 * rte_atomic16_cmpset() causes a write if using cmpxchg. This could
+	 * result in false sharing between enqueue and dequeue.
+	 *
+	 * Prevent unnecessary false sharing by reading broadcast_rarp first
+	 * and only performing cmpset if the read indicates it is likely to
+	 * be set.
+	 */
+	if (unlikely(rte_atomic16_read(&dev->broadcast_rarp) &&
+			rte_atomic16_cmpset((volatile uint16_t *)
+				&dev->broadcast_rarp.cnt, 1, 0))) {
+
+		rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
+		if (rarp_mbuf == NULL) {
+			RTE_LOG(ERR, VHOST_DATA,
+				"Failed to make RARP packet.\n");
+			count = 0;
+			goto out;
+		}
+		count -= 1;
 	}
 
+	if (vq_is_packed(dev))
+		count = virtio_dev_tx_packed(dev, vq, mbuf_pool, pkts, count);
+	else
+		count = virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count);
+
 out:
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 		vhost_user_iotlb_rd_unlock(vq);
@@ -1193,10 +1630,10 @@ out_access_unlock:
 		 * Inject it to the head of "pkts" array, so that switch's mac
 		 * learning table will get updated first.
 		 */
-		memmove(&pkts[1], pkts, i * sizeof(struct rte_mbuf *));
+		memmove(&pkts[1], pkts, count * sizeof(struct rte_mbuf *));
 		pkts[0] = rarp_mbuf;
-		i += 1;
+		count += 1;
 	}
 
-	return i;
+	return count;
 }