Re: [PATCH 1/4] x86/coco: Add a new CC attribute to unify cache flush during kexec

[Dave Young]

Hi,

On Thu, 22 Feb 2024 at 19:50, Huang, Kai <kai.huang@xxxxxxxxx> wrote:
>
> On Wed, 2024-02-21 at 10:28 +0100, Borislav Petkov wrote:
> > On Tue, Feb 20, 2024 at 04:30:13PM -0600, Tom Lendacky wrote:
> > > I believe the issues were that different Intel systems would hang or reset
> > > and it was bisected to that commit that added the WBINVD. It was a while
> > > ago, but I remember that they were similar to what the 1f5e7eb7868e commit
> > > ended up fixing, which was debugged because sometimes the WBINVD was still
> > > occasionally issued resulting in the following patch
> > >
> > >   9b040453d444 ("x86/smp: Dont access non-existing CPUID leaf")
> > >
> > > It just means that if we go to an unconditional WBINVD, then we need to be
> > > careful.
> >
> > Let's try it.
> >
> > Dave, do you remember what issues
> >
> >   f23d74f6c66c ("x86/mm: Rework wbinvd, hlt operation in stop_this_cpu()")
> >
> > fixed?
> >
> > If so, can you try the below diff ontop of latest tip/master to see if
> > those issues would reappear?
> >
> > Thx.
> >
> > ---
> >
> > diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
> > index ab49ade31b0d..ec4dcc9f70ca 100644
> > --- a/arch/x86/kernel/process.c
> > +++ b/arch/x86/kernel/process.c
> > @@ -824,8 +824,7 @@ void __noreturn stop_this_cpu(void *dummy)
> >        * Test the CPUID bit directly because the machine might've cleared
> >        * X86_FEATURE_SME due to cmdline options.
> >        */
> > -     if (c->extended_cpuid_level >= 0x8000001f && (cpuid_eax(0x8000001f) & BIT(0)))
> > -             native_wbinvd();
> > +     native_wbinvd();
> >
> >       /*
> >        * This brings a cache line back and dirties it, but
> >
>
> I really appreciate if Dave can help here.
>
> I will also reach out to see whether there's anyone in Intel met this before.

I forgot the details now,  let me recall and try the patches if possible.

Thanks
Dave