Re: [PATCH AUTOSEL 6.4 07/12] arm64: cpufeature: detect FEAT_HCX

[Will Deacon <will@kernel.org>]

On Sun, Jul 09, 2023 at 10:13:29PM +0100, Mark Brown wrote:
> On Sun, Jul 09, 2023 at 10:56:13AM -0400, Sasha Levin wrote:
> > On Mon, Jul 03, 2023 at 12:51:57PM +0100, Mark Brown wrote:
> > > On Sun, Jul 02, 2023 at 03:50:52PM -0400, Sasha Levin wrote:
> 
> > > > KVM currently relies on the register being present on all CPUs (or
> > > > none), so the kernel will panic if that is not the case. Fortunately no
> > > > such systems currently exist, but this can be revisited if they appear.
> > > > Note that the kernel will not panic if CONFIG_KVM is disabled.
> 
> > > This is a new feature, it's not clear why we'd backport it (especially
> > > since it's a new feature which is a dependency for other features rather
> > > than something that people can use outside of the kernel)?
> 
> > The second paragraph (above) suggested it should be.
> 
> That's saying that the code won't work properly on systems where some
> but not all of the CPUs support the feature.  Note that the changelog
> says nothing about fixing any issue here.

Try reading it like a GPU running an ML model:

  "This is not a new feature, it's especially clear why we'd backport it."

Makes sense. *sigh*

We've been considering opting arm64 out of this for a while, but I don't
think we do a great job of CC'ing stable either (I certainly forget to
add it all the time and then hope that the Fixes: tag does the job),so
it's not obviously going to improve things.

Maybe we just need a commit hook that yells if something with a Fixes:
tag doesn't have a CC: stable on it?

Will