[v4] examples/vhost: fix retry logic on eth rx path

Message ID 20220622092555.1149810-1-yuanx.wang@intel.com (mailing list archive)
State Accepted, archived
Delegated to: Maxime Coquelin
Headers
Series [v4] examples/vhost: fix retry logic on eth rx path |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK
ci/iol-aarch64-compile-testing success Testing PASS
ci/intel-Testing success Testing PASS
ci/iol-x86_64-compile-testing success Testing PASS
ci/iol-aarch64-unit-testing success Testing PASS
ci/iol-x86_64-unit-testing success Testing PASS
ci/iol-intel-Functional success Functional Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/github-robot: build success github build: passed
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-abi-testing warning Testing issues

Commit Message

Wang, YuanX June 22, 2022, 9:25 a.m. UTC
  drain_eth_rx() uses rte_vhost_avail_entries() to calculate
the available entries to determine if a retry is required.
However, this function only works with split rings, and
calculating packed rings will return the wrong value and cause
unnecessary retries resulting in a significant performance penalty.

This patch fix that by using the difference between tx/rx burst
as the retry condition.

Fixes: be800696c26e ("examples/vhost: use burst enqueue and dequeue from lib")
Cc: stable@dpdk.org

Signed-off-by: Yuan Wang <yuanx.wang@intel.com>
Tested-by: Wei Ling <weix.ling@intel.com>
---
V4: Fix fiexs tag.
V3: Fix mbuf index.
V2: Rebase to 22.07 rc1.
---
 examples/vhost/main.c | 28 +++++++++++-----------------
 1 file changed, 11 insertions(+), 17 deletions(-)
  

Comments

Chenbo Xia June 23, 2022, 2:57 a.m. UTC | #1
> -----Original Message-----
> From: Wang, YuanX <yuanx.wang@intel.com>
> Sent: Wednesday, June 22, 2022 5:26 PM
> To: maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>;
> dev@dpdk.org
> Cc: Hu, Jiayu <jiayu.hu@intel.com>; He, Xingguang <xingguang.he@intel.com>;
> Jiang, Cheng1 <cheng1.jiang@intel.com>; Wang, YuanX <yuanx.wang@intel.com>;
> stable@dpdk.org; Ling, WeiX <weix.ling@intel.com>
> Subject: [PATCH v4] examples/vhost: fix retry logic on eth rx path
> 
> drain_eth_rx() uses rte_vhost_avail_entries() to calculate
> the available entries to determine if a retry is required.
> However, this function only works with split rings, and
> calculating packed rings will return the wrong value and cause
> unnecessary retries resulting in a significant performance penalty.
> 
> This patch fix that by using the difference between tx/rx burst
> as the retry condition.
> 
> Fixes: be800696c26e ("examples/vhost: use burst enqueue and dequeue from
> lib")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Yuan Wang <yuanx.wang@intel.com>
> Tested-by: Wei Ling <weix.ling@intel.com>

Yuan & Wei

Please make sure it's re-tested. Patch seems good to me.

Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>

> ---
> V4: Fix fiexs tag.
> V3: Fix mbuf index.
> V2: Rebase to 22.07 rc1.
> ---
>  examples/vhost/main.c | 28 +++++++++++-----------------
>  1 file changed, 11 insertions(+), 17 deletions(-)
> 
> diff --git a/examples/vhost/main.c b/examples/vhost/main.c
> index e7fee5aa1b..0fa6c096c8 100644
> --- a/examples/vhost/main.c
> +++ b/examples/vhost/main.c
> @@ -634,7 +634,7 @@ us_vhost_usage(const char *prgname)
>  {
>  	RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
>  	"		--vm2vm [0|1|2]\n"
> -	"		--rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
> +	"		--rx-retry [0|1] --mergeable [0|1] --stats [0-N]\n"
>  	"		--socket-file <path>\n"
>  	"		--nb-devices ND\n"
>  	"		-p PORTMASK: Set mask for ports to be used by
> application\n"
> @@ -1383,27 +1383,21 @@ drain_eth_rx(struct vhost_dev *vdev)
>  	if (!rx_count)
>  		return;
> 
> -	/*
> -	 * When "enable_retry" is set, here we wait and retry when there
> -	 * is no enough free slots in the queue to hold @rx_count packets,
> -	 * to diminish packet loss.
> -	 */
> -	if (enable_retry &&
> -	    unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
> -			VIRTIO_RXQ))) {
> -		uint32_t retry;
> +	enqueue_count = vdev_queue_ops[vdev->vid].enqueue_pkt_burst(vdev,
> +						VIRTIO_RXQ, pkts, rx_count);
> 
> -		for (retry = 0; retry < burst_rx_retry_num; retry++) {
> +	/* Retry if necessary */
> +	if (enable_retry && unlikely(enqueue_count < rx_count)) {
> +		uint32_t retry = 0;
> +
> +		while (enqueue_count < rx_count && retry++ <
> burst_rx_retry_num) {
>  			rte_delay_us(burst_rx_delay_time);
> -			if (rx_count <= rte_vhost_avail_entries(vdev->vid,
> -					VIRTIO_RXQ))
> -				break;
> +			enqueue_count += vdev_queue_ops[vdev-
> >vid].enqueue_pkt_burst(vdev,
> +							VIRTIO_RXQ, &pkts[enqueue_count],
> +							rx_count - enqueue_count);
>  		}
>  	}
> 
> -	enqueue_count = vdev_queue_ops[vdev->vid].enqueue_pkt_burst(vdev,
> -					VIRTIO_RXQ, pkts, rx_count);
> -
>  	if (enable_stats) {
>  		__atomic_add_fetch(&vdev->stats.rx_total_atomic, rx_count,
>  				__ATOMIC_SEQ_CST);
> --
> 2.25.1
  
Ling, WeiX June 23, 2022, 7:20 a.m. UTC | #2
> -----Original Message-----
> From: Wang, YuanX <yuanx.wang@intel.com>
> Sent: Wednesday, June 22, 2022 5:26 PM
> To: maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>;
> dev@dpdk.org
> Cc: Hu, Jiayu <jiayu.hu@intel.com>; He, Xingguang
> <xingguang.he@intel.com>; Jiang, Cheng1 <cheng1.jiang@intel.com>; Wang,
> YuanX <yuanx.wang@intel.com>; stable@dpdk.org; Ling, WeiX
> <weix.ling@intel.com>
> Subject: [PATCH v4] examples/vhost: fix retry logic on eth rx path
> 
> drain_eth_rx() uses rte_vhost_avail_entries() to calculate the available
> entries to determine if a retry is required.
> However, this function only works with split rings, and calculating packed
> rings will return the wrong value and cause unnecessary retries resulting in a
> significant performance penalty.
> 
> This patch fix that by using the difference between tx/rx burst as the retry
> condition.
> 
> Fixes: be800696c26e ("examples/vhost: use burst enqueue and dequeue
> from lib")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Yuan Wang <yuanx.wang@intel.com>
> Tested-by: Wei Ling <weix.ling@intel.com>
> ---
> V4: Fix fiexs tag.
> V3: Fix mbuf index.
> V2: Rebase to 22.07 rc1.
> ---

Tested-by: Wei Ling <weix.ling@intel.com>
  
Maxime Coquelin July 1, 2022, 1:57 p.m. UTC | #3
On 6/22/22 11:25, Yuan Wang wrote:
> drain_eth_rx() uses rte_vhost_avail_entries() to calculate
> the available entries to determine if a retry is required.
> However, this function only works with split rings, and
> calculating packed rings will return the wrong value and cause
> unnecessary retries resulting in a significant performance penalty.
> 
> This patch fix that by using the difference between tx/rx burst
> as the retry condition.
> 
> Fixes: be800696c26e ("examples/vhost: use burst enqueue and dequeue from lib")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Yuan Wang <yuanx.wang@intel.com>
> Tested-by: Wei Ling <weix.ling@intel.com>
> ---
> V4: Fix fiexs tag.
> V3: Fix mbuf index.
> V2: Rebase to 22.07 rc1.
> ---
>   examples/vhost/main.c | 28 +++++++++++-----------------
>   1 file changed, 11 insertions(+), 17 deletions(-)


Applied to dpdk-next-virtio/main.

Thanks,
Maxime
		__ATOMIC_SEQ_CST);
  

Patch

diff --git a/examples/vhost/main.c b/examples/vhost/main.c
index e7fee5aa1b..0fa6c096c8 100644
--- a/examples/vhost/main.c
+++ b/examples/vhost/main.c
@@ -634,7 +634,7 @@  us_vhost_usage(const char *prgname)
 {
 	RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
 	"		--vm2vm [0|1|2]\n"
-	"		--rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
+	"		--rx-retry [0|1] --mergeable [0|1] --stats [0-N]\n"
 	"		--socket-file <path>\n"
 	"		--nb-devices ND\n"
 	"		-p PORTMASK: Set mask for ports to be used by application\n"
@@ -1383,27 +1383,21 @@  drain_eth_rx(struct vhost_dev *vdev)
 	if (!rx_count)
 		return;
 
-	/*
-	 * When "enable_retry" is set, here we wait and retry when there
-	 * is no enough free slots in the queue to hold @rx_count packets,
-	 * to diminish packet loss.
-	 */
-	if (enable_retry &&
-	    unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
-			VIRTIO_RXQ))) {
-		uint32_t retry;
+	enqueue_count = vdev_queue_ops[vdev->vid].enqueue_pkt_burst(vdev,
+						VIRTIO_RXQ, pkts, rx_count);
 
-		for (retry = 0; retry < burst_rx_retry_num; retry++) {
+	/* Retry if necessary */
+	if (enable_retry && unlikely(enqueue_count < rx_count)) {
+		uint32_t retry = 0;
+
+		while (enqueue_count < rx_count && retry++ < burst_rx_retry_num) {
 			rte_delay_us(burst_rx_delay_time);
-			if (rx_count <= rte_vhost_avail_entries(vdev->vid,
-					VIRTIO_RXQ))
-				break;
+			enqueue_count += vdev_queue_ops[vdev->vid].enqueue_pkt_burst(vdev,
+							VIRTIO_RXQ, &pkts[enqueue_count],
+							rx_count - enqueue_count);
 		}
 	}
 
-	enqueue_count = vdev_queue_ops[vdev->vid].enqueue_pkt_burst(vdev,
-					VIRTIO_RXQ, pkts, rx_count);
-
 	if (enable_stats) {
 		__atomic_add_fetch(&vdev->stats.rx_total_atomic, rx_count,
 				__ATOMIC_SEQ_CST);