Re: [PATCH net-next 2/5] virtio_net: Add page_pool support to improve performance

From: Xuan Zhuo
Date: Thu Nov 30 2023 - 20:39:32 EST


On Thu, 30 Nov 2023 13:30:40 +0800, Zhu Yanjun <yanjun.zhu@xxxxxxxxx> wrote:
>
> 在 2023/11/30 10:34, Xuan Zhuo 写道:
> > On Wed, 29 Nov 2023 23:29:10 +0800, Zhu Yanjun <yanjun.zhu@xxxxxxxxx> wrote:
> >> 在 2023/11/29 23:22, Zhu Yanjun 写道:
> >>> 在 2023/11/29 22:59, Michael S. Tsirkin 写道:
> >>>> On Wed, Nov 29, 2023 at 10:50:57PM +0800, Zhu Yanjun wrote:
> >>>>> 在 2023/5/26 13:46, Liang Chen 写道:
> >>>> what made you respond to a patch from May, now?
> >>> I want to apply page_pool to our virtio_net. This virtio_net works on
> >>> our device.
> >>>
> >>> I want to verify whether page_pool on virtio_net with our device can
> >>> improve the performance or not.
> >>>
> >>> And I found that ethtool is wrong.
> >>>
> >>> I use virtio_net on our device. I found that page member variable in
> >>> rq is not used in recv path.
> >>>
> >>> When virtio_net is modprobe, I checked page member variable in rq with
> >>> kprobe or crash tool.  page member variable in rq is always NULL.
> >>>
> >>> But sg in recv path is used.
> >>>
> >>> So how to use page member variable in rq? If page member variable in
> >>> rq is always NULL, can we remove it?
> >>>
> >>> BTW, I use ping and iperf tool to make tests with virtio_net. In the
> >>> tests, page member variable in rq is always NULL.
> >>
> >> And I replaced page member variable in rq with page_pool, but the
> >> statistics of page_pool are always 0.
> >>
> >> It is interesting that page_pool member variable in rq is not used in
> >> ping and iperf tests.
> >>
> >> I am not sure what tests can make page member variable not NULL. ^_^
> > Do you mean rq->pages?
> >
> > That is for big mode.
>
> Hi, Xuan
>
> Got it. What is big mode? Do you mean big packet size? I run iperf with
> the packet size 2^23.
>
> The rq->pages is still NULL.
>
> It is interesting.

You may need to check the code of virtnet_probe().

Thanks.


>
> Zhu Yanjun
>
>
> >
> > Thanks.
> >
> >
> >> Best Regards,
> >>
> >> Zhu Yanjun
> >>
> >>
> >>> It is interesting.
> >>>
> >>> Zhu Yanjun
> >>>