qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH] target-i386: fix losing XCR0 processor state co


From: Eduardo Habkost
Subject: Re: [Qemu-devel] [PATCH] target-i386: fix losing XCR0 processor state component bits
Date: Wed, 28 Sep 2016 11:57:37 -0300
User-agent: Mutt/1.7.0 (2016-08-17)

On Wed, Sep 28, 2016 at 09:54:27AM +0200, Paolo Bonzini wrote:
> 
> 
> On 28/09/2016 07:31, Wanpeng Li wrote:
> > From: Wanpeng Li <address@hidden>
> > 
> > Commit 96193c22a "target-i386: Move xsave component mask to features array"
> > leverages features array to handle XCR0 processor state component bits, 
> > however, it introduces a regression:
> > 
> > warning: host doesn't support requested feature: CPUID.0DH:EAX [bit 0]
> > warning: host doesn't support requested feature: CPUID.0DH:EAX [bit 1]
> > warning: host doesn't support requested feature: CPUID.0DH:EAX [bit 2]
> > 
> > My desktop doesn't have enough advance features, so just X87,SSE,AVX 
> > warnings are splat when I boot a guest.

Oops. I assume this is reproducible only using "-cpu host"?

> > 
> > The get migratable flags logic in x86_cpu_filter_features() path will 
> > filter out the feature flags which are unsupported and unmigratable. 
> > However, the bits of XCR0 processor state component featureword don't 
> > have feat_names, and some features like SSE/AVX etc have feat_names in 
> > CPUID.01H:EDX, CPUID.01H:ECX, so they are treated as unsupported.
> > 
> > This patch fix it by don't filter out XCR0 processor state components 
> > bits though they don't have feat_names just as before commit 96193c22ab3.
> > 
> > Cc: Paolo Bonzini <address@hidden>
> > Cc: Richard Henderson <address@hidden>
> > Cc: Eduardo Habkost <address@hidden>
> > Cc: Michael S. Tsirkin <address@hidden>
> > Signed-off-by: Wanpeng Li <address@hidden>
> > ---
> >  target-i386/cpu.c | 4 ++++
> >  1 file changed, 4 insertions(+)
> > 
> > diff --git a/target-i386/cpu.c b/target-i386/cpu.c
> > index ad09246..9d24eff 100644
> > --- a/target-i386/cpu.c
> > +++ b/target-i386/cpu.c
> > @@ -2156,6 +2156,10 @@ static uint32_t 
> > x86_cpu_get_supported_feature_word(FeatureWord w,
> >          r = kvm_arch_get_supported_cpuid(kvm_state, wi->cpuid_eax,
> >                                                      wi->cpuid_ecx,
> >                                                      wi->cpuid_reg);
> > +        if ((w == FEAT_XSAVE_COMP_LO) ||
> > +            (w == FEAT_XSAVE_COMP_HI)) {
> > +            return r;
> > +        }
> >      } else if (tcg_enabled()) {
> >          r = wi->tcg_features;
> >      } else {
> > 
> 
> I think the right place to add the test is x86_cpu_get_migratable_flags.

This can be fixed by adding actual property names to the
FEAT_XSAVE_COMP_* bits. This way we will be able to report
meaningful names to management in case GET_SUPPORTED_CPUID says a
given xsave component is not supported yet, and will ensure we
correctly treat still-unknown xsave components as unmigratable.

-- 
Eduardo



reply via email to

[Prev in Thread] Current Thread [Next in Thread]