Re: [Xen-devel] [PATCH 4/6] x86/hvm: Move hvm_funcs.cpuid_intercept() handling into hvm_cpuid()

["Tian, Kevin" <kevin.tian@xxxxxxxxx>]

> From: Boris Ostrovsky [mailto:boris.ostrovsky@xxxxxxxxxx]
> Sent: Thursday, November 17, 2016 1:45 AM
> 
> On 11/16/2016 12:34 PM, Andrew Cooper wrote:
> > On 16/11/16 17:34, Boris Ostrovsky wrote:
> >> On 11/16/2016 12:10 PM, Andrew Cooper wrote:
> >>> On 16/11/16 16:40, Boris Ostrovsky wrote:
> >>>> On 11/16/2016 07:31 AM, Andrew Cooper wrote:
> >>>>> @@ -3700,6 +3701,14 @@ void hvm_cpuid(unsigned int input, unsigned int 
> >>>>> *eax,
> unsigned int *ebx,
> >>>>>
> >>>>>          *ebx &= hvm_featureset[FEATURESET_e8b];
> >>>>>          break;
> >>>>> +
> >>>>> +    case 0x8000001c:
> >>>>> +        if ( !(v->arch.xcr0 & XSTATE_LWP) )
> >>>>> +            *eax = 0;
> >>>>> +        else if ( cpu_has_svm && cpu_has_lwp )
> >>>>> +            /* Turn on available bit and other features specified in 
> >>>>> lwp_cfg. */
> >>>>> +            *eax = (*edx & v->arch.hvm_svm.guest_lwp_cfg) | 1;
> >>>>> +        break;
> >>>>>      }
> >>>> You don't think this whole case should be under cpu_has_svm (or
> >>>> X86_VENDOR_AMD)?
> >>> LWP, being independently identifiable state should be gated on that
> >>> alone, even if in reality, it only exists on AMD hardware.
> >>>
> >>> The use of cpu_has_svm is only because guest_lwp_cfg is in an svm
> >>> union.  Were guest_lwp_cfg to move, the condition should be relaxed.
> >> I was thinking about the first 'if' clause. I believe 0x8000001c doesn't
> >> exist on Intel yet but if they add it we will clear eax for no good reason.
> >>
> >> OTOH we wouldn't be handling the leaf correctly anyway so maybe it's OK.
> > What do you think about Jan's suggestion, which is slightly better
> > overall anyway?
> 
> I don't think it changes anything wrt clearing eax on Intel.
> 

I don't know whether Intel/AMD will coordinate on those leaf code 
definitions (a question I need to ask internally). If there is no guarantee
that this leaf will mean same feature cross vendors, might be safer to
have whole trunk under svm condition?

Thanks
Kevin
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel