Re: [PATCH 5.10 462/717] ice, xsk: clear the status bits for the next_to_use descriptor

[Sasha Levin]

On Mon, Dec 28, 2020 at 02:51:05PM -0800, Jakub Kicinski wrote:
> On Mon, 28 Dec 2020 17:29:07 -0500 Sasha Levin wrote:
> > On Mon, Dec 28, 2020 at 10:54:23AM -0800, Jakub Kicinski wrote:
> > >On Mon, 28 Dec 2020 13:47:40 +0100 Greg Kroah-Hartman wrote:
> > >> From: Björn Töpel <bjorn.topel@xxxxxxxxx>
> > >>
> > >> [ Upstream commit 8d14768a7972b92c73259f0c9c45b969d85e3a60 ]
> > >>
> > >> On the Rx side, the next_to_use index points to the next item in the
> > >> HW ring to be refilled/allocated, and next_to_clean points to the next
> > >> item to potentially be processed.
> > >>
> > >> When the HW Rx ring is fully refilled, i.e. no packets has been
> > >> processed, the next_to_use will be next_to_clean - 1. When the ring is
> > >> fully processed next_to_clean will be equal to next_to_use. The latter
> > >> case is where a bug is triggered.
> > >>
> > >> If the next_to_use bits are not cleared, and the "fully processed"
> > >> state is entered, a stale descriptor can be processed.
> > >>
> > >> The skb-path correctly clear the status bit for the next_to_use
> > >> descriptor, but the AF_XDP zero-copy path did not do that.
> > >>
> > >> This change adds the status bits clearing of the next_to_use
> > >> descriptor.
> > >>
> > >> Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
> > >> Signed-off-by: Björn Töpel <bjorn.topel@xxxxxxxxx>
> > >> Signed-off-by: Jakub Kicinski <kuba@xxxxxxxxxx>
> > >> Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
> > >
> > >Oh wow, so much for Sasha waiting longer for code to get tested before
> > >auto-pulling things into stable :/
> >
> > The timeline is usually for a commit to appear in a release, and it did.
> > Was it too early?
>
> Hm, I'm not sure of exact semantics but I meant a final release,
> not an -rc.
>
> Plus I thought the point of things being part of a release is that
> people actually get a chance to test that release. -rc1 was cut 24
> hours ago. I guess a "release" is used as a yardstick here, to
> measure time, not for practical reasons?

Note that it wasn't actually released yet, at this point folks are
supposed to be testing 5.10.4-rc1 to make sure that those patches are
okay.

I still think that there are no significant users of Linus's tree, so
the idea of having a patch "in a release" doesn't mean as much as folks
think it does. Sure, we have a lot of folks who test -rc releases, but
are you aware of anyone who runs -rc on real world workloads to test it?

> > >I have this change and other changes here queued, but haven't sent the
> > >submission yet.
> >
> > What do you mean with "queued"? Its in Linus's tree for about two weeks
> > now.
>
> Networking maintainers have their own queue for patches that will go to
> stable:
>
> https://patchwork.kernel.org/bundle/netdev/stable/?state=*

This part has always been tricky to me: some parts of net/ and
drivers/net/ don't go through netdev, and some do. I have a filter to
ignore net/ completely, but I found that quite a lot of drivers/net/
wasn't covered by this process.

How could I blacklist/ignore the parts of the tree you're looking at?

Also, is drivers/net/ stuff covered as well as net/? I found in the past
that it's not the case when I was looking at missing patches for the
hyper-v driver.

--
Thanks,
Sasha