[PATCH] vhost: support sync copy when data buffer is small

[PATCH] vhost: support sync copy when data buffer is small

[Wenwu Ma]

In async datapath, if the length of data buffer
is less than 256, the data will be copied by CPU
instead of DMA.

Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
---
 lib/vhost/vhost.h      |  6 ++-
 lib/vhost/virtio_net.c | 96 ++++++++++++++++++++++++++++++------------
 2 files changed, 73 insertions(+), 29 deletions(-)

diff --git a/lib/vhost/vhost.h b/lib/vhost/vhost.h
index 9209558465..d0da53aa46 100644
--- a/lib/vhost/vhost.h
+++ b/lib/vhost/vhost.h
@@ -125,8 +125,10 @@ struct vring_used_elem_packed {
  * iovec
  */
 struct vhost_iovec {
-	void *src_addr;
-	void *dst_addr;
+	void *src_io_addr;
+	void *dst_io_addr;
+	void *src_virt_addr;
+	void *dst_virt_addr;
 	size_t len;
 };
 
diff --git a/lib/vhost/virtio_net.c b/lib/vhost/virtio_net.c
index 3085905d17..46f35ac05f 100644
--- a/lib/vhost/virtio_net.c
+++ b/lib/vhost/virtio_net.c
@@ -26,6 +26,8 @@
 
 #define MAX_BATCH_LEN 256
 
+#define CPU_COPY_THRESHOLD_LEN 256
+
 /* DMA device copy operation tracking array. */
 struct async_dma_info dma_copy_track[RTE_DMADEV_DEFAULT_MAX];
 
@@ -61,28 +63,37 @@ vhost_async_dma_transfer_one(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	int copy_idx = 0;
 	uint32_t nr_segs = pkt->nr_segs;
 	uint16_t i;
+	bool is_cpu_copy = true;
 
 	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",
+		if (iov[i].len > CPU_COPY_THRESHOLD_LEN) {
+			copy_idx = rte_dma_copy(dma_id, vchan_id,
+					(rte_iova_t)iov[i].src_io_addr,
+					(rte_iova_t)iov[i].dst_io_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;
+					vhost_async_dma_copy_log = true;
+				}
+				return -1;
 			}
-			return -1;
+			is_cpu_copy = false;
+		} else {
+			rte_memcpy(iov[i].dst_virt_addr, iov[i].src_virt_addr, iov[i].len);
 		}
 	}
 
@@ -90,7 +101,13 @@ vhost_async_dma_transfer_one(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	 * 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];
+	if (is_cpu_copy == false) {
+		dma_info->pkts_cmpl_flag_addr[copy_idx & ring_mask] =
+			&vq->async->pkts_cmpl_flag[flag_idx];
+	} else {
+		vq->async->pkts_cmpl_flag[flag_idx] = true;
+		nr_segs = 0;
+	}
 
 	return nr_segs;
 }
@@ -123,6 +140,19 @@ vhost_async_dma_transfer(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
 	rte_spinlock_unlock(&dma_info->dma_lock);
 
+	if (unlikely(ret < 0 && pkt_idx > 0)) {
+		do {
+			head_idx = (head_idx == 0) ? vq->size : head_idx - 1;
+			if (vq->async->pkts_cmpl_flag[head_idx] == false)
+				break;
+
+			pkt_idx--;
+			vq->async->pkts_cmpl_flag[head_idx] = false;
+			if (pkt_idx == 0)
+				break;
+		} while (1);
+	}
+
 	return pkt_idx;
 }
 
@@ -943,7 +973,7 @@ async_iter_initialize(struct virtio_net *dev, struct vhost_async *async)
 
 static __rte_always_inline int
 async_iter_add_iovec(struct virtio_net *dev, struct vhost_async *async,
-		void *src, void *dst, size_t len)
+		void *io_src, void *io_dst, void *virt_src, void *virt_dst, size_t len)
 {
 	struct vhost_iov_iter *iter;
 	struct vhost_iovec *iovec;
@@ -962,8 +992,10 @@ async_iter_add_iovec(struct virtio_net *dev, struct vhost_async *async,
 	iter = async->iov_iter + async->iter_idx;
 	iovec = async->iovec + async->iovec_idx;
 
-	iovec->src_addr = src;
-	iovec->dst_addr = dst;
+	iovec->src_io_addr = io_src;
+	iovec->dst_io_addr = io_dst;
+	iovec->src_virt_addr = virt_src;
+	iovec->dst_virt_addr = virt_dst;
 	iovec->len = len;
 
 	iter->nr_segs++;
@@ -999,12 +1031,13 @@ async_iter_reset(struct vhost_async *async)
 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)
+		uint64_t buf_addr, 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 *io_src, *io_dst;
+	void *virt_src, *virt_dst;
 	void *host_iova;
 
 	while (cpy_len) {
@@ -1016,16 +1049,23 @@ async_fill_seg(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			return -1;
 		}
 
+
 		if (to_desc) {
-			src = (void *)(uintptr_t)rte_pktmbuf_iova_offset(m, mbuf_offset);
-			dst = host_iova;
+			io_src = (void *)(uintptr_t)rte_pktmbuf_iova_offset(m, mbuf_offset);
+			io_dst = host_iova;
+
+			virt_src = rte_pktmbuf_mtod_offset(m, void *, mbuf_offset);
+			virt_dst = (void *)(buf_addr + buf_offset);
 		} else {
-			src = host_iova;
-			dst = (void *)(uintptr_t)rte_pktmbuf_iova_offset(m, mbuf_offset);
+			io_src = host_iova;
+			io_dst = (void *)(uintptr_t)rte_pktmbuf_iova_offset(m, mbuf_offset);
+
+			virt_src = (void *)(buf_addr + buf_offset);
+			virt_dst = rte_pktmbuf_mtod_offset(m, void *, mbuf_offset);
 		}
 
-		if (unlikely(async_iter_add_iovec(dev, async, src, dst,
-						 (size_t)mapped_len)))
+		if (unlikely(async_iter_add_iovec(dev, async, io_src, io_dst,
+						virt_src, virt_dst, (size_t)mapped_len)))
 			return -1;
 
 		cpy_len -= (uint32_t)mapped_len;
@@ -1169,6 +1209,7 @@ mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
 		if (is_async) {
 			if (async_fill_seg(dev, vq, m, mbuf_offset,
+						buf_addr + buf_offset,
 						buf_iova + buf_offset, cpy_len, true) < 0)
 				goto error;
 		} else {
@@ -2562,6 +2603,7 @@ desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
 		if (is_async) {
 			if (async_fill_seg(dev, vq, m, mbuf_offset,
+						buf_addr + buf_offset,
 						buf_iova + buf_offset, cpy_len, false) < 0)
 				goto error;
 		} else {
-- 
2.25.1

Re: [PATCH] vhost: support sync copy when data buffer is small

[Maxime Coquelin]

Hi Wenwu,

On 4/11/22 15:38, Wenwu Ma wrote:
> In async datapath, if the length of data buffer
> is less than 256, the data will be copied by CPU
> instead of DMA.
> 
> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
> ---
>   lib/vhost/vhost.h      |  6 ++-
>   lib/vhost/virtio_net.c | 96 ++++++++++++++++++++++++++++++------------
>   2 files changed, 73 insertions(+), 29 deletions(-)
> 

As I mentioned in the past, let's have a DMA-only solution working
before trying to support CPU copy for small packets.

Thanks,
Maxime

RE: [PATCH] vhost: support sync copy when data buffer is small

[Hu, Jiayu]

Hi Maxime,

> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin@redhat.com>
> Sent: Tuesday, April 19, 2022 5:10 PM
> To: Ma, WenwuX <wenwux.ma@intel.com>; Xia, Chenbo
> <chenbo.xia@intel.com>; dev@dpdk.org
> Cc: Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan <yinan.wang@intel.com>;
> He, Xingguang <xingguang.he@intel.com>; Ling, WeiX <weix.ling@intel.com>
> Subject: Re: [PATCH] vhost: support sync copy when data buffer is small
> 
> Hi Wenwu,
> 
> On 4/11/22 15:38, Wenwu Ma wrote:
> > In async datapath, if the length of data buffer is less than 256, the
> > data will be copied by CPU instead of DMA.
> >
> > Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
> > ---
> >   lib/vhost/vhost.h      |  6 ++-
> >   lib/vhost/virtio_net.c | 96 ++++++++++++++++++++++++++++++------------
> >   2 files changed, 73 insertions(+), 29 deletions(-)
> >
> 
> As I mentioned in the past, let's have a DMA-only solution working before
> trying to support CPU copy for small packets.

Thanks for your suggestion.

For "DMA-only solution working", do you mean community reaches an
agreement on OVS DSA design? The reason of asking this question is
that we will have some following patches around async vhost in 22.11,
like vhostpmd supporting async vhost data-path. Regarding your suggestion,
do you think they also need to be deferred until "DMA-only solution working"?

Thanks,
Jiayu

> 
> Thanks,
> Maxime

Re: [PATCH] vhost: support sync copy when data buffer is small

[Maxime Coquelin]

Hi Jiayu,

On 5/23/22 02:38, Hu, Jiayu wrote:
> Hi Maxime,
> 
>> -----Original Message-----
>> From: Maxime Coquelin <maxime.coquelin@redhat.com>
>> Sent: Tuesday, April 19, 2022 5:10 PM
>> To: Ma, WenwuX <wenwux.ma@intel.com>; Xia, Chenbo
>> <chenbo.xia@intel.com>; dev@dpdk.org
>> Cc: Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan <yinan.wang@intel.com>;
>> He, Xingguang <xingguang.he@intel.com>; Ling, WeiX <weix.ling@intel.com>
>> Subject: Re: [PATCH] vhost: support sync copy when data buffer is small
>>
>> Hi Wenwu,
>>
>> On 4/11/22 15:38, Wenwu Ma wrote:
>>> In async datapath, if the length of data buffer is less than 256, the
>>> data will be copied by CPU instead of DMA.
>>>
>>> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
>>> ---
>>>    lib/vhost/vhost.h      |  6 ++-
>>>    lib/vhost/virtio_net.c | 96 ++++++++++++++++++++++++++++++------------
>>>    2 files changed, 73 insertions(+), 29 deletions(-)
>>>
>>
>> As I mentioned in the past, let's have a DMA-only solution working before
>> trying to support CPU copy for small packets.
> 
> Thanks for your suggestion.
> 
> For "DMA-only solution working", do you mean community reaches an
> agreement on OVS DSA design? The reason of asking this question is
> that we will have some following patches around async vhost in 22.11,
> like vhostpmd supporting async vhost data-path. Regarding your suggestion,
> do you think they also need to be deferred until "DMA-only solution working"?

No, I was just referring to CPU copy for small buffers features.

Thanks,
Maxime

> Thanks,
> Jiayu
> 
>>
>> Thanks,
>> Maxime
>