DPDK patches and discussions
 help / color / mirror / Atom feed
From: Ye Xiaolong <xiaolong.ye@intel.com>
To: Joyce Kong <joyce.kong@arm.com>
Cc: maxime.coquelin@redhat.com, stephen@networkplumber.org,
	tiwei.bie@intel.com, zhihong.wang@intel.com, thomas@monjalon.net,
	jerinj@marvell.com, yinan.wang@intel.com,
	honnappa.nagarahalli@arm.com, gavin.hu@arm.com, nd@arm.com,
	dev@dpdk.org
Subject: Re: [dpdk-dev] [PATCH v3 1/2] virtio: one way barrier for split vring used idx
Date: Fri, 17 Apr 2020 14:51:45 +0800	[thread overview]
Message-ID: <20200417065145.GA57965@intel.com> (raw)
In-Reply-To: <20200406152634.606-2-joyce.kong@arm.com>

On 04/06, Joyce Kong wrote:
>In case VIRTIO_F_ORDER_PLATFORM(36) is not negotiated, then the frontend
>and backend are assumed to be implemented in software, that is they can
>run on identical CPUs in an SMP configuration.
>Thus a weak form of memory barriers like rte_smp_r/wmb, other than
>rte_cio_r/wmb, is sufficient for this case(vq->hw->weak_barriers == 1)
>and yields better performance.
>For the above case, this patch helps yielding even better performance
>by replacing the two-way barriers with C11 one-way barriers for used
>index in split ring.
>
>Signed-off-by: Joyce Kong <joyce.kong@arm.com>
>Reviewed-by: Gavin Hu <gavin.hu@arm.com>
>---
> drivers/net/virtio/virtio_ethdev.c            |  9 ++--
> drivers/net/virtio/virtio_ring.h              |  2 +-
> drivers/net/virtio/virtio_rxtx.c              | 46 +++++++++----------
> drivers/net/virtio/virtio_rxtx_simple_neon.c  |  5 +-
> drivers/net/virtio/virtio_rxtx_simple_sse.c   |  5 +-
> .../net/virtio/virtio_user/virtio_user_dev.c  |  8 ++--
> drivers/net/virtio/virtqueue.c                |  2 +-
> drivers/net/virtio/virtqueue.h                | 37 ++++++++++++---
> lib/librte_vhost/virtio_net.c                 |  5 +-
> 9 files changed, 71 insertions(+), 48 deletions(-)
>
>diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
>index f9d0ea70d..a4a865bfa 100644
>--- a/drivers/net/virtio/virtio_ethdev.c
>+++ b/drivers/net/virtio/virtio_ethdev.c
>@@ -285,13 +285,12 @@ virtio_send_command_split(struct virtnet_ctl *cvq,
> 
> 	virtqueue_notify(vq);
> 
>-	rte_rmb();
>-	while (VIRTQUEUE_NUSED(vq) == 0) {
>-		rte_rmb();
>+	/* virtqueue_nused has a load-acquire or rte_cio_rmb inside */
>+	while (virtqueue_nused(vq) == 0)
> 		usleep(100);
>-	}
> 
>-	while (VIRTQUEUE_NUSED(vq)) {
>+	/* virtqueue_nused has a load-acquire or rte_cio_rmb inside */
>+	while (virtqueue_nused(vq)) {
> 		uint32_t idx, desc_idx, used_idx;
> 		struct vring_used_elem *uep;
> 
>diff --git a/drivers/net/virtio/virtio_ring.h b/drivers/net/virtio/virtio_ring.h
>index 7ba34662e..0f6574f68 100644
>--- a/drivers/net/virtio/virtio_ring.h
>+++ b/drivers/net/virtio/virtio_ring.h
>@@ -59,7 +59,7 @@ struct vring_used_elem {
> 
> struct vring_used {
> 	uint16_t flags;
>-	volatile uint16_t idx;
>+	uint16_t idx;
> 	struct vring_used_elem ring[0];
> };
> 
>diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio_rxtx.c
>index 752faa0f6..9ba26fd95 100644
>--- a/drivers/net/virtio/virtio_rxtx.c
>+++ b/drivers/net/virtio/virtio_rxtx.c
>@@ -45,7 +45,7 @@ virtio_dev_rx_queue_done(void *rxq, uint16_t offset)
> 	struct virtnet_rx *rxvq = rxq;
> 	struct virtqueue *vq = rxvq->vq;
> 
>-	return VIRTQUEUE_NUSED(vq) >= offset;
>+	return virtqueue_nused(vq) >= offset;
> }
> 
> void
>@@ -1243,9 +1243,8 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
> 	if (unlikely(hw->started == 0))
> 		return nb_rx;
> 
>-	nb_used = VIRTQUEUE_NUSED(vq);
>-
>-	virtio_rmb(hw->weak_barriers);
>+	/* virtqueue_nused has a load-acquire or rte_cio_rmb inside */

Small nit, I don't think we need to add this comment to every occurrence of 
virtqueue_nused, what about moving it to the definition of this function?

Thanks,
Xiaolong

  reply	other threads:[~2020-04-17  6:56 UTC|newest]

Thread overview: 32+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-02-12  9:24 [dpdk-dev] [PATCH v1 0/2] one way barrier for split vring idx Joyce Kong
2020-02-12  9:24 ` [dpdk-dev] [PATCH v1 1/2] virtio: one way barrier for split vring used idx Joyce Kong
2020-02-12  9:24 ` [dpdk-dev] [PATCH v1 2/2] virtio: one way barrier for split vring avail idx Joyce Kong
2020-04-02  2:57 ` [dpdk-dev] [PATCH v2 0/2] one way barrier for split vring idx Joyce Kong
2020-04-02  2:57 ` [dpdk-dev] [PATCH v2 1/2] virtio: one way barrier for split vring used idx Joyce Kong
2020-04-02 15:47   ` Stephen Hemminger
2020-04-03  8:55     ` Gavin Hu
2020-04-16  4:40       ` Honnappa Nagarahalli
2020-04-16  6:46         ` Joyce Kong
2020-04-02  2:57 ` [dpdk-dev] [PATCH v2 2/2] virtio: one way barrier for split vring avail idx Joyce Kong
2020-04-06 15:26 ` [dpdk-dev] [PATCH v3 0/2] one way barrier for split vring idx Joyce Kong
2020-04-16  9:08   ` Ye Xiaolong
2020-04-06 15:26 ` [dpdk-dev] [PATCH v3 1/2] virtio: one way barrier for split vring used idx Joyce Kong
2020-04-17  6:51   ` Ye Xiaolong [this message]
2020-04-17  8:14     ` Joyce Kong
2020-04-06 15:26 ` [dpdk-dev] [PATCH v3 2/2] virtio: one way barrier for split vring avail idx Joyce Kong
2020-04-24  3:39 ` [dpdk-dev] [PATCH v4 0/2] one way barrier for split vring idx Joyce Kong
2020-04-28 16:06   ` Maxime Coquelin
2020-04-29 17:45   ` Ferruh Yigit
2020-04-30  9:09     ` Maxime Coquelin
2020-04-30  9:16       ` Joyce Kong
2020-04-30  9:24         ` Maxime Coquelin
2020-04-24  3:39 ` [dpdk-dev] [PATCH v4 1/2] virtio: one way barrier for split vring used idx Joyce Kong
2020-04-27  9:03   ` Maxime Coquelin
2020-04-24  3:39 ` [dpdk-dev] [PATCH v4 2/2] virtio: one way barrier for split vring avail idx Joyce Kong
2020-04-27  9:03   ` Maxime Coquelin
2020-04-30  9:14 ` [dpdk-dev] [PATCH v5 0/2] one way barrier for split vring idx Joyce Kong
2020-04-30 20:54   ` Maxime Coquelin
2020-04-30  9:14 ` [dpdk-dev] [PATCH v5 1/2] virtio: one way barrier for split vring used idx Joyce Kong
2020-04-30 22:58   ` Ferruh Yigit
2020-05-04 10:04     ` Maxime Coquelin
2020-04-30  9:14 ` [dpdk-dev] [PATCH v5 2/2] virtio: one way barrier for split vring avail idx Joyce Kong

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200417065145.GA57965@intel.com \
    --to=xiaolong.ye@intel.com \
    --cc=dev@dpdk.org \
    --cc=gavin.hu@arm.com \
    --cc=honnappa.nagarahalli@arm.com \
    --cc=jerinj@marvell.com \
    --cc=joyce.kong@arm.com \
    --cc=maxime.coquelin@redhat.com \
    --cc=nd@arm.com \
    --cc=stephen@networkplumber.org \
    --cc=thomas@monjalon.net \
    --cc=tiwei.bie@intel.com \
    --cc=yinan.wang@intel.com \
    --cc=zhihong.wang@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).