Re: [RFC,v3 1/2] vhost: support async dequeue for split ring

[Maxime Coquelin <maxime.coquelin@redhat.com>]

On 3/10/22 07:54, xuan.ding@intel.com wrote:
> From: Xuan Ding <xuan.ding@intel.com>
> 
> This patch implements asynchronous dequeue data path for vhost split
> ring, with dmadev library integrated.
> 
> Signed-off-by: Xuan Ding <xuan.ding@intel.com>
> Signed-off-by: Yuan Wang <yuanx.wang@intel.com>
> ---
>   lib/vhost/rte_vhost_async.h |  37 ++-
>   lib/vhost/version.map       |   1 +
>   lib/vhost/vhost.h           |   1 +
>   lib/vhost/virtio_net.c      | 504 ++++++++++++++++++++++++++++++++++++
>   4 files changed, 541 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/vhost/rte_vhost_async.h b/lib/vhost/rte_vhost_async.h
> index f1293c6a9d..b6ab0b06a2 100644
> --- a/lib/vhost/rte_vhost_async.h
> +++ b/lib/vhost/rte_vhost_async.h
> @@ -155,9 +155,9 @@ int rte_vhost_async_get_inflight(int vid, uint16_t queue_id);
>    * @param count
>    *  Size of the packet array
>    * @param dma_id
> - *  the identifier of DMA device
> + *  The identifier of DMA device
>    * @param vchan_id
> - *  the identifier of virtual DMA channel
> + *  The identifier of virtual DMA channel

This is unrelated to the purpose of this patch, it can be moved in a
dedicated trivial patch.

>    * @return
>    *  Number of packets returned
>    */
> @@ -187,6 +187,39 @@ uint16_t rte_vhost_clear_queue_thread_unsafe(int vid, uint16_t queue_id,
>   __rte_experimental
>   int rte_vhost_async_dma_configure(int16_t dma_id, uint16_t vchan_id);
>   
> +/**
> + * This function tries to receive packets from the guest with offloading
> + * copies to the async channel. The packets that are transfer completed
> + * are returned in "pkts". The other packets that their copies are submitted to
> + * the async channel but not completed are called "in-flight packets".
> + * This function will not return in-flight packets until their copies are
> + * completed by the async channel.
> + *
> + * @param vid
> + *  ID of vhost device to dequeue data
> + * @param queue_id
> + *  ID of virtqueue to dequeue data
> + * @param mbuf_pool
> + *  Mbuf_pool where host mbuf is allocated
> + * @param pkts
> + *  Blank array to keep successfully dequeued packets
> + * @param count
> + *  Size of the packet array
> + * @param nr_inflight
> + *  The amount of in-flight packets. If error occurred, its value is set to -1.
> + * @param dma_id
> + *  The identifier of DMA device
> + * @param vchan_id
> + *  The identifier of virtual DMA channel
> + * @return
> + *  Number of successfully dequeued packets
> + */
> +__rte_experimental
> +uint16_t
> +rte_vhost_async_try_dequeue_burst(int vid, uint16_t queue_id,
> +	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count,
> +	int *nr_inflight, uint16_t dma_id, uint16_t vchan_id);
> +
>   #ifdef __cplusplus
>   }
>   #endif
> diff --git a/lib/vhost/version.map b/lib/vhost/version.map
> index 0a66c5840c..968d6d4290 100644
> --- a/lib/vhost/version.map
> +++ b/lib/vhost/version.map
> @@ -87,6 +87,7 @@ EXPERIMENTAL {
>   
>   	# added in 22.03
>   	rte_vhost_async_dma_configure;

# added in 22.07

> +	rte_vhost_async_try_dequeue_burst;
>   };
>   
>   INTERNAL {
> diff --git a/lib/vhost/vhost.h b/lib/vhost/vhost.h
> index a9edc271aa..3799d41089 100644
> --- a/lib/vhost/vhost.h
> +++ b/lib/vhost/vhost.h
> @@ -178,6 +178,7 @@ extern struct async_dma_info dma_copy_track[RTE_DMADEV_DEFAULT_MAX];
>    */
>   struct async_inflight_info {
>   	struct rte_mbuf *mbuf;
> +	struct virtio_net_hdr nethdr;
>   	uint16_t descs; /* num of descs inflight */
>   	uint16_t nr_buffers; /* num of buffers inflight for packed ring */
>   };
> diff --git a/lib/vhost/virtio_net.c b/lib/vhost/virtio_net.c
> index 5f432b0d77..3816caca79 100644
> --- a/lib/vhost/virtio_net.c
> +++ b/lib/vhost/virtio_net.c
> @@ -3141,3 +3141,507 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
>   
>   	return count;
>   }
> +
> +static __rte_always_inline int
> +async_desc_to_mbuf_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)
> +{
> +	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 (unlikely(async_iter_add_iovec(dev, async, host_iova,
> +				(void *)(uintptr_t)rte_pktmbuf_iova_offset(m, mbuf_offset),
> +				(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;
> +}

It looks really similar to async_mbuf_to_desc_seg(), just the direction
of the DMA copoy is changed.

Maybe we could refactor the two functions in a single one with adding a 
parameter for the direction. Something like this:

static __rte_always_inline int
async_fill_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, bool to_desc)
{
	struct vhost_async *async = vq->async;
	uint64_t mapped_len;
	uint32_t buf_offset = 0;
	void *src, *dst
	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 (to_desc) {
			src = (void *)(uintptr_t)rte_pktmbuf_iova_offset(m, mbuff_offset);
			dst = host_iova
		} else {
			dst = host_iova
			src = (void *)(uintptr_t)rte_pktmbuf_iova_offset(m, mbuff_offset);
		}

		if (unlikely(async_iter_add_iovec(dev, async, src, dst, 
(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;
}


Then if you don't pass a variable but a static true/false value for
to_desc, the compiler should generate the same code, so no performance
degradation.

> +
> +static __rte_always_inline int
> +async_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
> +		  struct buf_vector *buf_vec, uint16_t nr_vec,
> +		  struct rte_mbuf *m, struct rte_mempool *mbuf_pool,
> +		  struct virtio_net_hdr *nethdr)
> +{
> +	uint64_t buf_addr, buf_iova;
> +	uint32_t buf_avail, buf_offset, buf_len;
> +	uint32_t mbuf_avail, mbuf_offset;
> +	uint32_t cpy_len;
> +	/* A counter to avoid desc dead loop chain */
> +	uint16_t vec_idx = 0;
> +	struct rte_mbuf *cur = m, *prev = m;
> +	struct virtio_net_hdr tmp_hdr;
> +	struct virtio_net_hdr *hdr = NULL;
> +	struct vhost_async *async = vq->async;
> +
> +	buf_addr = buf_vec[vec_idx].buf_addr;
> +	buf_len = buf_vec[vec_idx].buf_len;
> +	buf_iova = buf_vec[vec_idx].buf_iova;
> +
> +	if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1))
> +		return -1;
> +
> +	if (virtio_net_with_host_offload(dev)) {
> +		if (unlikely(buf_len < sizeof(struct virtio_net_hdr))) {
> +			/*
> +			 * No luck, the virtio-net header doesn't fit
> +			 * in a contiguous virtual area.
> +			 */
> +			copy_vnet_hdr_from_desc(&tmp_hdr, buf_vec);
> +			hdr = &tmp_hdr;
> +		} else {
> +			hdr = (struct virtio_net_hdr *)((uintptr_t)buf_addr);
> +		}
> +	}
> +
> +	/*
> +	 * A virtio driver normally uses at least 2 desc buffers
> +	 * for Tx: the first for storing the header, and others
> +	 * for storing the data.
> +	 */
> +	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 if (buf_len == dev->vhost_hlen) {
> +		if (unlikely(++vec_idx >= nr_vec))
> +			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;
> +
> +		buf_offset = 0;
> +		buf_avail = buf_len;
> +	} else {
> +		buf_offset = dev->vhost_hlen;
> +		buf_avail = buf_vec[vec_idx].buf_len - dev->vhost_hlen;
> +	}
> +
> +	PRINT_PACKET(dev, (uintptr_t)(buf_addr + buf_offset), (uint32_t)buf_avail, 0);
> +
> +	mbuf_offset = 0;
> +	mbuf_avail  = m->buf_len - RTE_PKTMBUF_HEADROOM;
> +
> +	if (async_iter_initialize(dev, async))
> +		return -1;
> +
> +	while (1) {
> +		cpy_len = RTE_MIN(buf_avail, mbuf_avail);
> +
> +		if (async_desc_to_mbuf_seg(dev, vq, cur, mbuf_offset, buf_iova + buf_offset,
> +					   cpy_len) < 0)
> +			goto error;
> +
> +		mbuf_avail -= cpy_len;
> +		buf_avail -= cpy_len;
> +		mbuf_offset += cpy_len;
> +		buf_offset += cpy_len;
> +
> +		/* This buf reaches to its end, get the next one */
> +		if (buf_avail == 0) {
> +			if (++vec_idx >= nr_vec)
> +				break;
> +
> +			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;
> +
> +			PRINT_PACKET(dev, (uintptr_t)buf_addr, (uint32_t)buf_avail, 0);
> +		}
> +
> +		/*
> +		 * This mbuf reaches to its end, get a new one
> +		 * to hold more data.
> +		 */
> +		if (mbuf_avail == 0) {
> +			cur = rte_pktmbuf_alloc(mbuf_pool);
> +			if (unlikely(cur == NULL)) {
> +				VHOST_LOG_DATA(ERR,
> +					"(%s) %s: failed to allocate memory for mbuf.\n",
> +					dev->ifname, __func__);
> +				goto error;
> +			}
> +
> +			prev->next = cur;
> +			prev->data_len = mbuf_offset;
> +			m->nb_segs += 1;
> +			m->pkt_len += mbuf_offset;
> +			prev = cur;
> +
> +			mbuf_offset = 0;
> +			mbuf_avail = cur->buf_len - RTE_PKTMBUF_HEADROOM;
> +		}
> +	}
> +
> +	prev->data_len = mbuf_offset;
> +	m->pkt_len += mbuf_offset;
> +
> +	async_iter_finalize(async);
> +	if (hdr)
> +		*nethdr = *hdr;
> +
> +	return 0;
> +
> +error:
> +	async_iter_cancel(async);
> +	return -1;
> +}


You can do here the same refactoring I did for the enqueue path, i.e.
merging copy_desc_to_mbuf and async_desc_to_mbuf in a single function.

> +static __rte_always_inline uint16_t
> +async_poll_dequeue_completed_split(struct virtio_net *dev, uint16_t queue_id,
> +		struct rte_mbuf **pkts, uint16_t count, uint16_t dma_id,
> +		uint16_t vchan_id, bool legacy_ol_flags)
> +{
> +	uint16_t start_idx, from, i;
> +	uint16_t nr_cpl_pkts = 0;
> +	struct async_inflight_info *pkts_info;
> +	struct vhost_virtqueue *vq = dev->virtqueue[queue_id];
> +
> +	pkts_info = vq->async->pkts_info;
> +
> +	vhost_async_dma_check_completed(dev, dma_id, vchan_id, VHOST_DMA_MAX_COPY_COMPLETE);
> +
> +	start_idx = async_get_first_inflight_pkt_idx(vq);
> +
> +	from = start_idx;
> +	while (vq->async->pkts_cmpl_flag[from] && count--) {
> +		vq->async->pkts_cmpl_flag[from] = false;
> +		from = (from + 1) & (vq->size - 1);
> +		nr_cpl_pkts++;
> +	}
> +
> +	if (nr_cpl_pkts == 0)
> +		return 0;
> +
> +	for (i = 0; i < nr_cpl_pkts; i++) {
> +		from = (start_idx + i) & (vq->size - 1);
> +		pkts[i] = pkts_info[from].mbuf;
> +
> +		if (virtio_net_with_host_offload(dev))
> +			vhost_dequeue_offload(dev, &pkts_info[from].nethdr, pkts[i],
> +					      legacy_ol_flags);
> +	}
> +
> +	/* write back completed descs to used ring and update used idx */
> +	write_back_completed_descs_split(vq, nr_cpl_pkts);
> +	__atomic_add_fetch(&vq->used->idx, nr_cpl_pkts, __ATOMIC_RELEASE);
> +	vhost_vring_call_split(dev, vq);
> +
> +	vq->async->pkts_inflight_n -= nr_cpl_pkts;
> +
> +	return nr_cpl_pkts;
> +}
> +
> +static __rte_always_inline uint16_t
> +virtio_dev_tx_async_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
> +		uint16_t queue_id, struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts,
> +		uint16_t count, uint16_t dma_id, uint16_t vchan_id, bool legacy_ol_flags)
> +{
> +	static bool allocerr_warned;
> +	bool dropped = false;
> +	uint16_t free_entries;
> +	uint16_t pkt_idx, slot_idx = 0;
> +	uint16_t nr_done_pkts = 0;
> +	uint16_t pkt_err = 0;
> +	uint16_t n_xfer;
> +	struct vhost_async *async = vq->async;
> +	struct async_inflight_info *pkts_info = async->pkts_info;
> +	struct rte_mbuf *pkts_prealloc[MAX_PKT_BURST];
> +	uint16_t pkts_size = count;
> +
> +	/**
> +	 * The ordering between avail index and
> +	 * desc reads needs to be enforced.
> +	 */
> +	free_entries = __atomic_load_n(&vq->avail->idx, __ATOMIC_ACQUIRE) - vq->last_avail_idx;
> +	if (free_entries == 0)
> +		goto out;
> +
> +	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
> +
> +	async_iter_reset(async);
> +
> +	count = RTE_MIN(count, MAX_PKT_BURST);
> +	count = RTE_MIN(count, free_entries);
> +	VHOST_LOG_DATA(DEBUG, "(%s) about to dequeue %u buffers\n", dev->ifname, count);
> +
> +	if (rte_pktmbuf_alloc_bulk(mbuf_pool, pkts_prealloc, count))
> +		goto out;
> +
> +	for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
> +		uint16_t head_idx = 0;
> +		uint16_t nr_vec = 0;
> +		uint16_t to;
> +		uint32_t buf_len;
> +		int err;
> +		struct buf_vector buf_vec[BUF_VECTOR_MAX];
> +		struct rte_mbuf *pkt = pkts_prealloc[pkt_idx];
> +
> +		if (unlikely(fill_vec_buf_split(dev, vq, vq->last_avail_idx,
> +						&nr_vec, buf_vec,
> +						&head_idx, &buf_len,
> +						VHOST_ACCESS_RO) < 0)) {
> +			dropped = true;
> +			break;
> +		}
> +
> +		err = virtio_dev_pktmbuf_prep(dev, pkt, buf_len);
> +		if (unlikely(err)) {
> +			/**
> +			 * mbuf allocation fails for jumbo packets when external
> +			 * buffer allocation is not allowed and linear buffer
> +			 * is required. Drop this packet.
> +			 */
> +			if (!allocerr_warned) {
> +				VHOST_LOG_DATA(ERR,
> +					"(%s) %s: Failed mbuf alloc of size %d from %s\n",
> +					dev->ifname, __func__, buf_len, mbuf_pool->name);
> +				allocerr_warned = true;
> +			}
> +			dropped = true;
> +			break;
> +		}
> +
> +		slot_idx = (async->pkts_idx + pkt_idx) & (vq->size - 1);
> +		err = async_desc_to_mbuf(dev, vq, buf_vec, nr_vec, pkt, mbuf_pool,
> +					&pkts_info[slot_idx].nethdr);
> +		if (unlikely(err)) {
> +			if (!allocerr_warned) {
> +				VHOST_LOG_DATA(ERR,
> +					"(%s) %s: Failed to offload copies to async channel.\n",
> +					dev->ifname, __func__);
> +				allocerr_warned = true;
> +			}
> +			dropped = true;
> +			break;
> +		}
> +
> +		pkts_info[slot_idx].mbuf = pkt;
> +
> +		/* store used descs */
> +		to = async->desc_idx_split & (vq->size - 1);
> +		async->descs_split[to].id = head_idx;
> +		async->descs_split[to].len = 0;
> +		async->desc_idx_split++;
> +
> +		vq->last_avail_idx++;
> +	}
> +
> +	if (unlikely(dropped))
> +		rte_pktmbuf_free_bulk(&pkts_prealloc[pkt_idx], count - pkt_idx);
> +
> +	n_xfer = vhost_async_dma_transfer(dev, vq, dma_id, vchan_id, async->pkts_idx,
> +					  async->iov_iter, pkt_idx);
> +
> +	async->pkts_inflight_n += n_xfer;
> +
> +	pkt_err = pkt_idx - n_xfer;
> +	if (unlikely(pkt_err)) {
> +		VHOST_LOG_DATA(DEBUG,
> +			"(%s) %s: failed to transfer data for queue id %d.\n",
> +			dev->ifname, __func__, queue_id);
> +
> +		pkt_idx = n_xfer;
> +		/* recover available ring */
> +		vq->last_avail_idx -= pkt_err;
> +
> +		/**
> +		 * recover async channel copy related structures and free pktmbufs
> +		 * for error pkts.
> +		 */
> +		async->desc_idx_split -= pkt_err;
> +		while (pkt_err-- > 0) {
> +			rte_pktmbuf_free(pkts_info[slot_idx & (vq->size - 1)].mbuf);
> +			slot_idx--;
> +		}
> +	}
> +
> +	async->pkts_idx += pkt_idx;
> +	if (async->pkts_idx >= vq->size)
> +		async->pkts_idx -= vq->size;
> +
> +out:
> +	/* DMA device may serve other queues, unconditionally check completed. */
> +	nr_done_pkts = async_poll_dequeue_completed_split(dev, queue_id, pkts, pkts_size,
> +							  dma_id, vchan_id, legacy_ol_flags);
> +
> +	return nr_done_pkts;
> +}
> +
> +__rte_noinline
> +static uint16_t
> +virtio_dev_tx_async_split_legacy(struct virtio_net *dev,
> +		struct vhost_virtqueue *vq, uint16_t queue_id,
> +		struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts,
> +		uint16_t count, uint16_t dma_id, uint16_t vchan_id)
> +{
> +	return virtio_dev_tx_async_split(dev, vq, queue_id, mbuf_pool,
> +				pkts, count, dma_id, vchan_id, true);
> +}
> +
> +__rte_noinline
> +static uint16_t
> +virtio_dev_tx_async_split_compliant(struct virtio_net *dev,
> +		struct vhost_virtqueue *vq, uint16_t queue_id,
> +		struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts,
> +		uint16_t count, uint16_t dma_id, uint16_t vchan_id)
> +{
> +	return virtio_dev_tx_async_split(dev, vq, queue_id, mbuf_pool,
> +				pkts, count, dma_id, vchan_id, false);
> +}
> +
> +uint16_t
> +rte_vhost_async_try_dequeue_burst(int vid, uint16_t queue_id,
> +	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count,
> +	int *nr_inflight, uint16_t dma_id, uint16_t vchan_id)
> +{
> +	struct virtio_net *dev;
> +	struct rte_mbuf *rarp_mbuf = NULL;
> +	struct vhost_virtqueue *vq;
> +	int16_t success = 1;
> +
> +	*nr_inflight = -1;
> +
> +	dev = get_device(vid);
> +	if (!dev)
> +		return 0;
> +
> +	if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
> +		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,
> +			"(%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(rte_spinlock_trylock(&vq->access_lock) == 0))
> +		return 0;
> +
> +	if (unlikely(vq->enabled == 0)) {
> +		count = 0;
> +		goto out_access_unlock;
> +	}
> +
> +	if (unlikely(!vq->async)) {
> +		VHOST_LOG_DATA(ERR, "(%s) %s: async not registered for queue id %d.\n",
> +			dev->ifname, __func__, queue_id);
> +		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
> +	 * __atomic_compare_exchange_n causes a write if performed compare
> +	 * and exchange. This could result in false sharing between enqueue
> +	 * and dequeue.
> +	 *
> +	 * Prevent unnecessary false sharing by reading broadcast_rarp first
> +	 * and only performing compare and exchange if the read indicates it
> +	 * is likely to be set.
> +	 */
> +	if (unlikely(__atomic_load_n(&dev->broadcast_rarp, __ATOMIC_ACQUIRE) &&
> +			__atomic_compare_exchange_n(&dev->broadcast_rarp,
> +			&success, 0, 0, __ATOMIC_RELEASE, __ATOMIC_RELAXED))) {
> +
> +		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");
> +			count = 0;
> +			goto out;
> +		}
> +		/*
> +		 * Inject it to the head of "pkts" array, so that switch's mac
> +		 * learning table will get updated first.
> +		 */
> +		pkts[0] = rarp_mbuf;
> +		pkts++;
> +		count -= 1;
> +	}
> +
> +	if (unlikely(vq_is_packed(dev))) {
> +		static bool not_support_pack_log;
> +		if (!not_support_pack_log) {
> +			VHOST_LOG_DATA(ERR,
> +				"(%s) %s: async dequeue does not support packed ring.\n",
> +				dev->ifname, __func__);
> +			not_support_pack_log = true;
> +		}
> +		count = 0;
> +		goto out;
> +	}
> +
> +	if (dev->flags & VIRTIO_DEV_LEGACY_OL_FLAGS)
> +		count = virtio_dev_tx_async_split_legacy(dev, vq, queue_id,
> +				mbuf_pool, pkts, count, dma_id, vchan_id);
> +	else
> +		count = virtio_dev_tx_async_split_compliant(dev, vq, queue_id,
> +				mbuf_pool, pkts, count, dma_id, vchan_id);
> +
> +	*nr_inflight = vq->async->pkts_inflight_n;
> +
> +out:
> +	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
> +		vhost_user_iotlb_rd_unlock(vq);
> +
> +out_access_unlock:
> +	rte_spinlock_unlock(&vq->access_lock);
> +
> +	if (unlikely(rarp_mbuf != NULL))
> +		count += 1;
> +
> +	return count;
> +}