All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Michael S. Tsirkin" <mst@redhat.com>
To: Vishwanath Seshagiri <vishs@meta.com>
Cc: Omar Elghoul <oelghoul@linux.ibm.com>,
	andrew+netdev@lunn.ch, davem@davemloft.net, dw@davidwei.uk,
	edumazet@google.com, eperezma@redhat.com,
	ilias.apalodimas@linaro.org, jasowang@redhat.com,
	kernel-team@meta.com, kuba@kernel.org,
	linux-kernel@vger.kernel.org, netdev@vger.kernel.org,
	pabeni@redhat.com, technoboy85@gmail.com,
	virtualization@lists.linux.dev, xuanzhuo@linux.alibaba.com
Subject: Re: [PATCH net-next v11] virtio_net: add page_pool support for buffer allocation
Date: Mon, 23 Mar 2026 19:37:55 -0400	[thread overview]
Message-ID: <20260323193722-mutt-send-email-mst@kernel.org> (raw)
In-Reply-To: <f91a42f7-1e90-4db0-923a-371dac8245e0@meta.com>

On Mon, Mar 23, 2026 at 01:50:29PM -0400, Vishwanath Seshagiri wrote:
> 
> 
> On 3/23/26 10:39 PM, Omar Elghoul wrote:
> > On 3/23/26 12:58 PM, Michael S. Tsirkin wrote:
> > 
> > > On Mon, Mar 23, 2026 at 11:52:34AM -0400, Michael S. Tsirkin wrote:
> > > > On Mon, Mar 23, 2026 at 11:01:31AM -0400, Omar Elghoul wrote:
> > > > > [...]
> > > > Well... I am not sure how I missed it. Obvious in hindsight:
> > > > 
> > > > static void receive_buf(struct virtnet_info *vi, struct
> > > > receive_queue *rq,
> > > >                          void *buf, unsigned int len, void **ctx,
> > > >                          unsigned int *xdp_xmit,
> > > >                          struct virtnet_rq_stats *stats)
> > > > {
> > > >          struct net_device *dev = vi->dev;
> > > >          struct sk_buff *skb;
> > > >          u8 flags;
> > > >          if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
> > > >                  pr_debug("%s: short packet %i\n", dev->name, len);
> > > >                  DEV_STATS_INC(dev, rx_length_errors);
> > > >                  virtnet_rq_free_buf(vi, rq, buf);
> > > >                  return;
> > > >          }
> > > >          /* About the flags below:
> > > >           * 1. Save the flags early, as the XDP program might
> > > > overwrite them.
> > > >           * These flags ensure packets marked as
> > > > VIRTIO_NET_HDR_F_DATA_VALID
> > > >           * stay valid after XDP processing.
> > > >           * 2. XDP doesn't work with partially checksummed
> > > > packets (refer to
> > > >           * virtnet_xdp_set()), so packets marked as
> > > >           * VIRTIO_NET_HDR_F_NEEDS_CSUM get dropped during XDP
> > > > processing.
> > > >           */
> > > >          if (vi->mergeable_rx_bufs) {
> > > >                  flags = ((struct virtio_net_common_hdr *)buf)-
> > > > >hdr.flags;
> > > >                  skb = receive_mergeable(dev, vi, rq, buf, ctx,
> > > > len, xdp_xmit,
> > > >                                          stats);
> > > >          } else if (vi->big_packets) {
> > > >                  void *p = page_address((struct page *)buf);
> > > >                  flags = ((struct virtio_net_common_hdr *)p)->hdr.flags;
> > > >                  skb = receive_big(dev, vi, rq, buf, len, stats);
> > > >          } else {
> > > >                  flags = ((struct virtio_net_common_hdr *)buf)-
> > > > >hdr.flags;
> > > >                  skb = receive_small(dev, vi, rq, buf, ctx, len,
> > > > xdp_xmit, stats);
> > > >          }
> > > > 
> > > > 
> > > > So we are reading the header, before dma sync, which is within
> > > > receive_mergeable and friends:
> > > > 
> > > > static struct sk_buff *receive_mergeable(struct net_device *dev,
> > > >                                           struct virtnet_info *vi,
> > > >                                           struct receive_queue *rq,
> > > >                                           void *buf,
> > > >                                           void *ctx,
> > > >                                           unsigned int len,
> > > >                                           unsigned int *xdp_xmit,
> > > >                                           struct
> > > > virtnet_rq_stats *stats)
> > > > {
> > > >          struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
> > > >          int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
> > > >          struct page *page = virt_to_head_page(buf);
> > > >          int offset = buf - page_address(page);
> > > >          struct sk_buff *head_skb, *curr_skb;
> > > >          unsigned int truesize = mergeable_ctx_to_truesize(ctx);
> > > >          unsigned int headroom = mergeable_ctx_to_headroom(ctx);
> > > >          head_skb = NULL;
> > > >          if (rq->use_page_pool_dma)
> > > >                  page_pool_dma_sync_for_cpu(rq->page_pool, page,
> > > > offset, len);
> > > > 
> > > > 
> > > > Just as a test, the below should fix it (compiled only), but the real
> > > > fix is more complex since we need to be careful to avoid
> > > > expensive syncing
> > > > twice.
> > > > 
> > > > 
> > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > index 97035b49bae7..57b4f5954bed 100644
> > > > --- a/drivers/net/virtio_net.c
> > > > +++ b/drivers/net/virtio_net.c
> > > > @@ -931,9 +931,19 @@ static struct sk_buff *page_to_skb(struct
> > > > virtnet_info *vi,
> > > >   static void *virtnet_rq_get_buf(struct receive_queue *rq, u32
> > > > *len, void **ctx)
> > > >   {
> > > > +    void *buf;
> > > > +
> > > >       BUG_ON(!rq->page_pool);
> > > > -    return virtqueue_get_buf_ctx(rq->vq, len, ctx);
> > > > +    buf = virtqueue_get_buf_ctx(rq->vq, len, ctx);
> > > > +    if (buf && rq->use_page_pool_dma && *len) {
> > > > +        struct page *page = virt_to_head_page(buf);
> > > > +        int offset = buf - page_address(page);
> > > > +
> > > > +        page_pool_dma_sync_for_cpu(rq->page_pool, page, offset, *len);
> > > > +    }
> > > > +
> > > > +    return buf;
> > > >   }
> > > >   static void virtnet_rq_unmap_free_buf(struct virtqueue *vq, void *buf)
> > > > 
> > > > 
> > > > 
> > > > 
> > > > -- 
> > > > MST
> > > or maybe like this:
> > Both of these patches resolve the issue on my end.
> > > 
> > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > index 97035b49bae7..835f52651006 100644
> > > --- a/drivers/net/virtio_net.c
> > > +++ b/drivers/net/virtio_net.c
> > > @@ -1956,13 +1956,6 @@ static struct sk_buff *receive_small(struct
> > > net_device *dev,
> > >        */
> > >       buf -= VIRTNET_RX_PAD + xdp_headroom;
> > > -    if (rq->use_page_pool_dma) {
> > > -        int offset = buf - page_address(page) +
> > > -                 VIRTNET_RX_PAD + xdp_headroom;
> > > -
> > > -        page_pool_dma_sync_for_cpu(rq->page_pool, page, offset, len);
> > > -    }
> > > -
> > >       len -= vi->hdr_len;
> > >       u64_stats_add(&stats->bytes, len);
> > > @@ -2398,9 +2391,6 @@ static struct sk_buff
> > > *receive_mergeable(struct net_device *dev,
> > >       head_skb = NULL;
> > > -    if (rq->use_page_pool_dma)
> > > -        page_pool_dma_sync_for_cpu(rq->page_pool, page, offset, len);
> > > -
> > >       u64_stats_add(&stats->bytes, len - vi->hdr_len);
> > >       if (check_mergeable_len(dev, ctx, len))
> > > @@ -2563,6 +2553,13 @@ static void receive_buf(struct virtnet_info
> > > *vi, struct receive_queue *rq,
> > >           return;
> > >       }
> > > +    if (rq->use_page_pool_dma) {
> > > +        struct page *page = virt_to_head_page(buf);
> > > +        int offset = buf - page_address(page);
> > > +
> > > +        page_pool_dma_sync_for_cpu(rq->page_pool, page, offset, len);
> > > +    }
> > > +
> > >       /* About the flags below:
> > >        * 1. Save the flags early, as the XDP program might overwrite
> > > them.
> > >        * These flags ensure packets marked as VIRTIO_NET_HDR_F_DATA_VALID
> > > 
> 
> Hello Omar. Thank you for the bisect, apologies for breaking the
> functionality. Thank you Michael for the patch, I will be careful next
> time about DMA sync issues. Looks like the bug is fixed, is there
> anything else needed from my end?
> 


Would be great to have you test the patch on your end and report
the performace. Thanks!

-- 
MST


  reply	other threads:[~2026-03-23 23:38 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-03-10 18:31 [PATCH net-next v11] virtio_net: add page_pool support for buffer allocation Vishwanath Seshagiri
2026-03-13  7:51 ` Jason Wang
2026-03-13  9:26   ` Vishwanath Seshagiri
2026-03-16  7:41     ` Jason Wang
2026-03-13 16:50   ` Vishwanath Seshagiri
2026-03-16  7:35     ` Jason Wang
2026-03-16  9:56 ` Michael S. Tsirkin
2026-03-16 10:43   ` Michael S. Tsirkin
2026-03-16 11:57     ` Vishwanath Seshagiri
2026-03-16 12:04       ` Michael S. Tsirkin
2026-03-17  2:30 ` patchwork-bot+netdevbpf
2026-03-23 15:01 ` Omar Elghoul
2026-03-23 15:01   ` Omar Elghoul
2026-03-23 15:52   ` Michael S. Tsirkin
2026-03-23 16:54     ` Omar Elghoul
2026-03-23 17:10       ` Michael S. Tsirkin
2026-03-23 16:58     ` Michael S. Tsirkin
2026-03-23 17:09       ` Omar Elghoul
2026-03-23 17:50         ` Vishwanath Seshagiri
2026-03-23 23:37           ` Michael S. Tsirkin [this message]
2026-03-24  0:34       ` Jason Wang
2026-03-24  8:20       ` Aithal, Srikanth

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=20260323193722-mutt-send-email-mst@kernel.org \
    --to=mst@redhat.com \
    --cc=andrew+netdev@lunn.ch \
    --cc=davem@davemloft.net \
    --cc=dw@davidwei.uk \
    --cc=edumazet@google.com \
    --cc=eperezma@redhat.com \
    --cc=ilias.apalodimas@linaro.org \
    --cc=jasowang@redhat.com \
    --cc=kernel-team@meta.com \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=oelghoul@linux.ibm.com \
    --cc=pabeni@redhat.com \
    --cc=technoboy85@gmail.com \
    --cc=virtualization@lists.linux.dev \
    --cc=vishs@meta.com \
    --cc=xuanzhuo@linux.alibaba.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.