Re: [Xen-devel] [PATCH] x86: correct CPUID output for out of bounds input

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 02.09.16 at 17:14, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 01/09/16 16:27, Jan Beulich wrote:
>>
>>>> +        {
>>>> +            if ( d->arch.x86_vendor == X86_VENDOR_AMD )
>>>> +            {
>>>> +                *eax = 0;
>>>> +                *ebx = 0;
>>>> +                *ecx = 0;
>>>> +                *edx = 0;
>>>> +                return;
>>>> +            }
>>>> +            if ( input >> 16 )
>>>> +                hvm_cpuid(0, &lvl, NULL, NULL, NULL);
>>> Is this really the right way round?  The AMD method of "reserved always
>>> as zero" is the more sane default to take.
>> If anything I'd then say let's _always_ follow the AMD model.
> 
> It would certainly be better to default to AMD, and special case others
> on an as-needed basis.
> 
> Strictly speaking, following the AMD model is compatible with the
> "Reserved" nature specified for Intel.
> 
> Lets just go with this.

Done. But before sending v2, just to be clear: The group check
which you also didn't like won't go away. That's because if we didn't
do it, we'd hide all CPUID info outside the basic and extended group,
in particular (in case we run virtualized ourselves) and leaves coming
from the lower level hypervisor (most notably our own ones, if it's
another Xen underneath).

Jan


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