Re: [PATCH v3 1/2] x86: KVM: Advertise FSRS and FSRC on AMD to userspace
From: Sean Christopherson
Date: Fri Apr 11 2025 - 09:55:20 EST
On Fri, Apr 11, 2025, Borislav Petkov wrote:
> On Wed, Dec 04, 2024 at 04:43:44PM +0300, Maksim Davydov wrote:
> > diff --git a/arch/x86/include/asm/cpufeatures.h b/arch/x86/include/asm/cpufeatures.h
> > index 17b6590748c0..45f87a026bba 100644
> > --- a/arch/x86/include/asm/cpufeatures.h
> > +++ b/arch/x86/include/asm/cpufeatures.h
> > @@ -460,6 +460,8 @@
> > #define X86_FEATURE_NULL_SEL_CLR_BASE (20*32+ 6) /* Null Selector Clears Base */
> > #define X86_FEATURE_AUTOIBRS (20*32+ 8) /* Automatic IBRS */
> > #define X86_FEATURE_NO_SMM_CTL_MSR (20*32+ 9) /* SMM_CTL MSR is not present */
> > +#define X86_FEATURE_AMD_FSRS (20*32+10) /* AMD Fast short REP STOSB supported */
> > +#define X86_FEATURE_AMD_FSRC (20*32+11) /* AMD Fast short REP CMPSB supported */
>
> Since Intel has the same flags, you should do
>
> if (cpu_has(c, X86_FEATURE_AMD_FSRS))
> set_cpu_cap(c, X86_FEATURE_FSRS);
>
> and the other one too. Probably in init_amd() so that guest userspace doesn't
> need to differentiate between the two and you don't have to do...
>
> > #define X86_FEATURE_SBPB (20*32+27) /* Selective Branch Prediction Barrier */
> > #define X86_FEATURE_IBPB_BRTYPE (20*32+28) /* MSR_PRED_CMD[IBPB] flushes all branch type predictions */
> > diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
> > index 097bdc022d0f..7bc095add8ee 100644
> > --- a/arch/x86/kvm/cpuid.c
> > +++ b/arch/x86/kvm/cpuid.c
> > @@ -799,8 +799,8 @@ void kvm_set_cpu_caps(void)
> >
> > kvm_cpu_cap_mask(CPUID_8000_0021_EAX,
> > F(NO_NESTED_DATA_BP) | F(LFENCE_RDTSC) | 0 /* SmmPgCfgLock */ |
> > - F(NULL_SEL_CLR_BASE) | F(AUTOIBRS) | 0 /* PrefetchCtlMsr */ |
> > - F(WRMSR_XX_BASE_NS)
> > + F(NULL_SEL_CLR_BASE) | F(AUTOIBRS) | F(AMD_FSRS) |
> > + F(AMD_FSRC) | 0 /* PrefetchCtlMsr */ | F(WRMSR_XX_BASE_NS)
>
> ... this.
KVM should still explicitly advertise support for AMD's flavor. There are KVM
use cases where KVM's advertised CPUID support is used almost verbatim, in which
case not advertising AMD_FSRC would result it the features not be enumerated to
the guest. Linux might cross-pollinate, but KVM can't rely on all software to do
so.