[Qemu-devel] Re: [PATCH] qemu: Enable XSAVE related CPUID

[Sheng Yang <sheng@linux.intel.com>]

On Thursday 27 May 2010 20:56:17 Avi Kivity wrote:
> On 05/27/2010 12:50 PM, Sheng Yang wrote:
> > We can support it in KVM now. The initial values are the minimal
> > requirement of XSAVE capable processor.
> > 
> > Signed-off-by: Sheng Yang<sheng@linux.intel.com>
> > ---
> > 
> >   target-i386/cpuid.c |   32 ++++++++++++++++++++++++++++++++
> >   1 files changed, 32 insertions(+), 0 deletions(-)
> > 
> > diff --git a/target-i386/cpuid.c b/target-i386/cpuid.c
> > index eebf038..cbf5595 100644
> > --- a/target-i386/cpuid.c
> > +++ b/target-i386/cpuid.c
> > @@ -1067,6 +1067,38 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t
> > index, uint32_t count,
> > 
> >           *ecx = 0;
> >           *edx = 0;
> >           break;
> > 
> > +    case 0xD:
> > +        /* Processor Extended State */
> > +        if (!(env->cpuid_ext_features&  CPUID_EXT_XSAVE)) {
> > +            *eax = 0;
> > +            *ebx = 0;
> > +            *ecx = 0;
> > +            *edx = 0;
> > +            break;
> > +        }
> > +        if (count == 0) {
> > +            *eax = 0x7; /* FP | SSE | YMM */
> > +            *ebx = 0x340; /* FP + SSE + YMM size */
> > +            *ecx = 0x340; /* FP + SSE + YMM size */
> 
> For -cpu host, we should pick these from KVM_GET_SUPPORTED_CPUID.  For
> canned cpu types (e.g. qemu64), we need to return what we always did.

Yes, I also prefer this way. didn't do this because it's somehow out of current 
QEmu CPUID setting mechanism.
> 
> We can also add a new cpu type that has them built in (there's no cpu on
> the market with avx, right?)

Right... Let's use -cpu host now.

> 
> > +            *edx = 0;
> > +        } else if (count == 1) {
> > +            /* eax = 1, so we can continue with others */
> > +            *eax = 1;
> > +            *ebx = 0;
> > +            *ecx = 0;
> > +            *edx = 0;
> > +        } else if (count == 2) {
> > +            *eax = 0x100; /* YMM size */
> > +            *ebx = 0x240; /* YMM offset */
> > +            *ecx = 0;
> > +            *edx = 0;
> 
> These, too.

Sure.

--
regards
Yang, Sheng

> 
> > +        } else {
> > +            *eax = 0;
> > +            *ebx = 0;
> > +            *ecx = 0;
> > +            *edx = 0;
> > +        }
> > +        break;
> > 
> >       case 0x80000000:
> >           *eax = env->cpuid_xlevel;
> >           *ebx = env->cpuid_vendor1;