Re: [PATCH 3/3] x86/HVM: make hvm_efer_valid() honor guest features

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 08/01/15 16:04, Jan Beulich wrote:
>>>> On 08.01.15 at 16:56, <tim@xen.org> wrote:
>> At 15:23 +0000 on 08 Jan (1420727005), Jan Beulich wrote:
>>> +    if ( (value & EFER_LMSLE) && !cpu_has_lmsl )
>>> +        return 0;
>> I see this bit has no CPUID flag, and the docs don't seem to suggest
>> that it would ever not be valid.  Are there real CPUs where it can't
>> be set?
> Yes, I'm afraid AMD introduced this subsequent to SVM, iirc upon
> VMware request.
>
>> My reviewed-by: stands anyway as this is an improvement over the
>> current code.  Just wondering whether this needs to go on the list
>> of interesting edge cazses for Andrew's feature-levelling rewrite. 
> Yes, dealing with this ought to be interesting.

:(

I had pessimistically assumed I wasn't going to be able to get away with
purely architectural state in the new shiny world.  I will take care to
accommodate this in the next iteration.

~Andrew