Re: [PATCH] KVM: VMX: Add a quirk to (not) honor guest PAT on CPUs that support self-snoop

[Sean Christopherson]

On Wed, Apr 02, 2025, Yan Zhao wrote:
> On Tue, Apr 01, 2025 at 03:11:07PM -0700, Sean Christopherson wrote:
> > Add back support for honoring guest PAT on Intel CPUs that support self-
> > snoop (and don't have errata), but guarded by a quirk so as not to break
> > existing setups that subtly relied on KVM forcing WB for synthetic
> > devices.
> > 
> > This effectively reverts commit 9d70f3fec14421e793ffbc0ec2f739b24e534900
> > and reapplies 377b2f359d1f71c75f8cc352b5c81f2210312d83, but with a quirk.
> > 
> > Cc: Yan Zhao <yan.y.zhao@xxxxxxxxx>
> > Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx>
> > ---
> >
> Hi Sean,
> 
> > AFAIK, we don't have an answer as to whether the slow UC behavior on CLX+
> > is working as intended or a CPU flaw, which Paolo was hoping we would get
> We did answer the slow UC behavior is working as intended at [1].
> 
> "After consulting with CPU architects,
>  it's told that this behavior is expected on ICX/SPR Xeon platforms due to
>  the snooping implementation."
> 
> Paolo then help update the series to v2 [2] /v3 [3].
> 
> Did you overlook those series, or is there something I missed?

Nope, you didn't miss anything.  I have that series in my TODO folder, but only
glanced at it when it flew by and completely missed that it quirks ignoring
guest PAT.  Not sure how I missed the cover letter subject though...

Anyways, ignore this, my bad.  Thanks for the update, and sorry for the noise!