netdev - Re: [PATCH vhost v6 09/10] virtio_net: set premapped mode by find

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date: Mon, 8 Apr 2024 13:01:56 +0800
From: Jason Wang <jasowang@...hat.com>
To: Xuan Zhuo <xuanzhuo@...ux.alibaba.com>
Cc: virtualization@...ts.linux.dev, "Michael S. Tsirkin" <mst@...hat.com>, 
	"David S. Miller" <davem@...emloft.net>, Eric Dumazet <edumazet@...gle.com>, 
	Jakub Kicinski <kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>, netdev@...r.kernel.org
Subject: Re: [PATCH vhost v6 09/10] virtio_net: set premapped mode by find_vqs()

On Sun, Apr 7, 2024 at 2:03 PM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
>
> On Sun, 7 Apr 2024 12:24:00 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > On Mon, Apr 1, 2024 at 11:10 AM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
> > >
> > > On Mon, 1 Apr 2024 09:40:07 +0800, Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
> > > > On Fri, 29 Mar 2024 11:20:08 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > > > > On Thu, Mar 28, 2024 at 4:27 PM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
> > > > > >
> > > > > > On Thu, 28 Mar 2024 16:05:02 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > > > > > > On Wed, Mar 27, 2024 at 7:14 PM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
> > > > > > > >
> > > > > > > > Now, the virtio core can set the premapped mode by find_vqs().
> > > > > > > > If the premapped can be enabled, the dma array will not be
> > > > > > > > allocated. So virtio-net use the api of find_vqs to enable the
> > > > > > > > premapped.
> > > > > > > >
> > > > > > > > Judge the premapped mode by the vq->premapped instead of saving
> > > > > > > > local variable.
> > > > > > > >
> > > > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@...ux.alibaba.com>
> > > > > > > > ---
> > > > > > >
> > > > > > > I wonder what's the reason to keep a fallback when premapped is not enabled?
> > > > > >
> > > > > > Rethink this.
> > > > > >
> > > > > > I think you are right. We can remove the fallback.
> > > > > >
> > > > > > Because we have the virtio dma apis that wrap all the cases.
> > > > > > So I will remove the fallback from the virtio-net in next version.
> > > > >
> > > > > Ok.
> > > > >
> > > > > >
> > > > > > But we still need to export the premapped to the drivers.
> > > > > > Because we can enable the AF_XDP only when premapped is true.
> > > > >
> > > > > I may miss something but it should work like
> > > > >
> > > > > enable AF_XDP -> enable remapping
> > > > >
> > > > > So can we fail during remapping enablement?
> > > >
> > > >
> > > > YES.
> > > >
> > > > Enabling the premapped mode may fail, then we must stop to enable AF_XDP.
> > > >
> > > > AF-XDP requires that we export the dma dev to the af-xdp.
> > > > We can do that only when the virtio core works with use_dma_api.
> > > > Other other side, if we support the page-pool in future, we may have the
> > > > same requirement.
> > >
> > > Rethink this.
> > >
> > > Enable premapped MUST NOT fail. No care the use_dma_api is true or not, because
> > > we have the DMA APIs for virtio. Then the virtio-net rx will work with
> > > premapped (I will make the big mode work with premapped mode)
> >
> > Just to make sure we're at the same page. Rx will always work in the
> > mode or pre mapping. So we can easily fail the probe if we fail to
> > enable RX premapping?
>
>
> NO, enabling premapped mode can not fail. So the rx will always work
> in the premapping mode.

Ok, kind of my understanding.

>
>
> >
> > >
> > > AF_XDP checks the virtqueue_dma_dev() when enabling.
> > >
> > > But disabling premapped mode may fail, because that virtio ring need to
> > > allocate memory for dma.
> >
> > That's kind of too tricky, what if we just allocate the memory for dma
> > unconditionally?
>
> It's ok, but we waste the memory.

Probably, but the point is to reduce the complexity of the codes for symmetry:

We don't need to have and maintain codes for fallback mode when we
fail to disable premapping.

Thanks

>
> Thanks.
>
>
>
> >
> > Thanks
> >
> > >
> > > Thanks.
> > >
> > >
> > >
> > > >
> > > >
> > > > Thanks.
> > > >
> > > >
> > > > >
> > > > > THanks
> > > > >
> > > > > >
> > > > > > Thanks
> > > > > >
> > > > > >
> > > > > > >
> > > > > > > Thanks
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
>