Re: [PATCH v2] Subject: x86/PAT: Report PAT on CPUs that support PAT without MTRR

[Chuck Zmudzinski]

On 7/18/2022 7:39 AM, Jan Beulich wrote:
> On 18.07.2022 13:31, Chuck Zmudzinski wrote:
> > On 7/18/2022 2:07 AM, Jan Beulich wrote:
> >> On 15.07.2022 21:53, Chuck Zmudzinski wrote:
> >>> Two things I see here in my efforts to get a patch to fix this regression:
> >>>
> >>> 1. Does Xen have plans to give Linux running in Dom0 write-access to the
> >>> PAT MSR?
> >>
> >> No, as this is not technically feasible (all physical CPUs should run
> >> with the same value in the MSR, or else other issues arise).
> >>
> >>> 2. Does Xen have plans to expose MTRRs to Linux running in Dom0?
> >>
> >> Yen does expose MTRRs to PV Dom0, but via a hypercall mechanism. I
> >> don't think there are plans on the Xen side to support the MSR
> >> interface (and hence to expose the CPUID bit), and iirc there are
> >> no plans on the Linux side to use the MTRR interface. This also
> >> wouldn't really make sense anymore now that it has become quite
> >> clear that Linux wants to have PAT working without depending on
> >> MTRR.
> > 
> > I am not so sure about that, given what Borislav Petkov
> > said when commenting on your patch here:
> > 
> > https://lore.kernel.org/lkml/YsRjX%2FU1XN8rq+8u@xxxxxxx/
> > 
> > Specifically, Borislav Petkov wrote on Tue, 5 Jul 2022 18:14:23 +0200:
> > 
> > Actually, the current goal is to adjust Xen dom0 because:
> > 
> > 1. it uses the PAT code
> > 
> > 2. but then it does something special and hides the MTRRs
> > 
> > which is not something real hardware does.
> > 
> > So this one-off thing should be prominent, visible and not get in the
> > way.
> > 
> > --------------end of Borislav Petkov quote-----------
>
> And then, a day later, he said
>
> "So I'm being told that it would be generally beneficial for all kinds of
>  virtualization solutions to be able to support PAT only, without MTRRs
>  so it would be interesting to see how ugly it would become to decouple
>  PAT from MTRRs in Linux..."

What if it proves to be too ugly to decouple PAT from MTRRs? Then
I doubt that "Linux wants to have PAT working without depending
on MTRR." We can hope that Juergen's work to decouple PAT from
MTRRs is not so ugly that it cannot be done, but that is by no means
certain at this point. At the very least, this means the fix to the
regression
will be at least delayed considerably, and possibly this means this
regression will never be fixed in the mainline kernel release.

Chuck