Re: [Xen-devel] [PATCH v2 11/30] xen/x86: Calculate maximum host and guest featuresets

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 15/02/16 15:07, Jan Beulich wrote:
>
>>>> +uint32_t __read_mostly raw_featureset[FSCAPINTS];
>>>> +uint32_t __read_mostly host_featureset[FSCAPINTS];
>>>> +uint32_t __read_mostly pv_featureset[FSCAPINTS];
>>>> +uint32_t __read_mostly hvm_featureset[FSCAPINTS];
>>>> +
>>>> +static void sanitise_featureset(uint32_t *fs)
>>>> +{
>>>> +    unsigned int i;
>>>> +
>>>> +    for ( i = 0; i < FSCAPINTS; ++i )
>>>> +    {
>>>> +        /* Clamp to known mask. */
>>>> +        fs[i] &= known_features[i];
>>>> +    }
>>>> +
>>>> +    switch ( boot_cpu_data.x86_vendor )
>>>> +    {
>>>> +    case X86_VENDOR_INTEL:
>>>> +        /* Intel clears the common bits in e1d. */
>>>> +        fs[FEATURESET_e1d] &= ~COMMON_1D;
>>>> +        break;
>>>> +
>>>> +    case X86_VENDOR_AMD:
>>>> +        /* AMD duplicates the common bits between 1d and e1d. */
>>>> +        fs[FEATURESET_e1d] = ((fs[FEATURESET_1d]  &  COMMON_1D) |
>>>> +                              (fs[FEATURESET_e1d] & ~COMMON_1D));
>>>> +        break;
>>>> +    }
>>> How is this meant to work with cross vendor migration?
>> I don't see how cross-vendor is relevant here.  This is about reporting
>> the hosts modified featureset accurately to the toolstack.
> I.e. you're not later going to use what you generate here to also
> massage (or at least validate) what guests are going to see?

Oh right - as currently implemented, this will clobber features on an
Intel host attempting to emulate AMD through cross-vendor.

I will reconsider the logic in v3.  This isn't trivial to fix,
especially given that we don't yet have per-domain policies.

~Andrew

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