Re: fsincos emulation on AMD CPUs

[Keir Fraser <keir@xen.org>]

On 15/12/2011 13:19, "Jan Beulich" <JBeulich@suse.com> wrote:

>>>> On 15.12.11 at 14:13, Keir Fraser <keir@xen.org> wrote:
>> On 15/12/2011 13:08, "Jan Beulich" <JBeulich@suse.com> wrote:
>> 
>>>>>> On 15.12.11 at 13:33, Keir Fraser <keir@xen.org> wrote:
>>>> More detail: the full patch is ugly and hard to test all cases. And there's
>>>> no practical scenario where we want to emulate FSINCOS on AMD -- we don't
>>>> emulate realmode on AMD, FSINCOS on a shadowed page certainly indicates
>>>> that
>>>> we should unshadow the page, FSINCOS on MMIO is mad or malicious.
>>> 
>>> Those latter two cases can't really happen, as fsincos has no memory
>>> operand.
>> 
>> Possibly if the instruction itself was in a recycled page-table page? Or in
>> an MMIO page, or the malicious race that Paolo described --- definitely
>> malicious either way.
>> 
>> Anyhow the short answer is we never want to emulate it on AMD. :-)
> 
> I just sent out the patch as quoted in the reply to Paolo, but you're
> suggesting to be even more drastic and ignore the CPU family. If
> you really want it done that way, I wonder whether we should bail on
> AMD for *all* x87 operations not having memory operands.

Your patch is fine.

 -- Keir

> Jan
>