RE: [PATCH v2 5/8] x86/mtrr: revert commit 90b926e68f50

[Michael Kelley (LINUX)]

From: Juergen Gross <jgross@xxxxxxxx>
> 
> On 13.02.23 12:46, Christian Kujau wrote:
> > On Mon, 13 Feb 2023, Juergen Gross wrote:
> >> On 10.02.23 19:59, Linux regression tracking (Thorsten Leemhuis) wrote:
> >>> Hi, this is your Linux kernel regression tracker.
> >>>
> >>> On 09.02.23 08:22, Juergen Gross wrote:
> >>>> Commit 90b926e68f50 ("x86/pat: Fix pat_x_mtrr_type() for MTRR disabled
> >>>> case") has introduced a regression with Xen.
> >>>>
> >>>> Revert the patch.
> >>>
> >>> That regression you refer to is afaics one I'm tracking[1] that was
> >>> introduced this cycle. That makes me wonder: could this patch be applied
> >>> directly to fix the issue quickly? Or are patches 1 to 4 needed as well
> >>> (or the whole series?) to avoid other problems?
> >>
> >> Patches 1-4 are needed, too, as otherwise the issue claimed to be fixed
> >> with patch 5 would show up again.
> >
> > The (last?) -rc8 version was released yesterday. Would it be possible to
> > include at least (only) the revert in mainline so that 6.2 will be
> > released with a working storage configuration under Xen?
> 
> Hmm, this would make Hyper-V SEV-SNP guests slow again.
> 
> I'm not completely against it, but OTOH I'm a little bit biased as the
> maintainer of the Xen code. :-)
> 
> Michael, would you see major problems with doing the revert before having
> the final patches for fixing your issue, too?
> 

I'm OK with doing the revert.  It's probably the right tradeoff for the
broader community because the Hyper-V use case is more narrow and
requires more curation for other reasons.  The use case is the Azure
public cloud, and we can pretty much make sure that one of the solutions
is applied to kernels used with SEV-SNP in that environment.

Michael

> > Otherwise one would have to carry around that single revert manually until
> > this patch series has landed in mainline, or convince all the
> > distributions to do so :-\
> >
> > Anyway, thanks for fixing this problem, I did not expect this to be such a
> > complicated issue when I reported that thing :-)
> 
> Yes, I have opened a can of worms with my MTRR/PAT disentangling.
> 
> 
> Juergen