Re: [PATCH net-next] virtio_net: drop netdev_tx_reset_queue() from virtnet_enable_queue_pair()

From: Koichiro Den
Date: Mon Dec 02 2024 - 22:39:06 EST


On Tue, Dec 03, 2024 at 10:25:14AM +0800, Jason Wang wrote:
> On Tue, Dec 3, 2024 at 10:14 AM Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
> >
> > On Mon, 2 Dec 2024 12:22:53 +0800 Jason Wang wrote:
> > > > Fixes: c8bd1f7f3e61 ("virtio_net: add support for Byte Queue Limits")
> > > > Cc: <stable@xxxxxxxxxxxxxxx> # v6.11+
> > > > Signed-off-by: Koichiro Den <koichiro.den@xxxxxxxxxxxxx>
> > >
> > > Acked-by: Jason Wang <jasowang@xxxxxxxxxx>
> >
> > I see Tx skb flush in:
> >
> > virtnet_freeze() -> remove_vq_common() -> free_unused_bufs() -> virtnet_sq_free_unused_buf()
> >
> > do we need to reset the BQL state in that case?
>
> Yes, I think so. And I spot another path which is:
>
> virtnet_tx_resize() -> virtqueue_resize() -> virtnet_sq_free_unused_buf().
>
> > Rule of thumb is netdev_tx_reset_queue() should follow any flush
> > (IOW skb freeing not followed by netdev_tx_completed_queue()).
> >
>
> Right.
>
> Koichiro, I think this fixes the problem of open/stop but may break
> freeze/restore(). Let's fix that.
>
> For resizing, it's a path that has been buggy since the introduction of BQL.
>
> Thanks
>

It makes sense, I'll send v2 soon. Thanks for the review.
-Koichiro Den