Re: [Xen-devel] [PATCH RFC] x86/resume: take care of fully eager FPU around system suspend

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 18.06.18 at 10:45, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 17/06/18 16:56, Jan Beulich wrote:
>> --- a/xen/arch/x86/acpi/suspend.c
>> +++ b/xen/arch/x86/acpi/suspend.c
>> @@ -92,8 +92,11 @@ void restore_rest_processor_state(void)
>>          write_debugreg(7, curr->arch.debugreg[7]);
>>      }
>>  
>> -    /* Reload FPU state on next FPU use. */
>> -    stts();
>> +    /* Reload FPU state immediately or on next FPU use. */
>> +    if ( curr->arch.fully_eager_fpu )
>> +        vcpu_restore_fpu_eager(curr);
>> +    else
>> +        stts();
> 
> Hmm - thinking about it, this should either be nothing, or a straight
> call to vcpu_restore_fpu_eager(curr).
> 
> If it is necessary here, then I cant see how the non-lazy xsave states
> currently work.

Hmm, indeed, and same then in the EFI case.

> OTOH, we could probably drop a large quantity of this custom
> save/restore logic if we force a transition to full idle before taking
> the S3 path, at which point the normal ctxt_switch_to() path should DTRT.

Right, but that's not something I'd consider for 4.11. In particular I'd
want such a change to be throughly (over an extended period of time)
tested by someone actively using S3. It also is not really a good option
for the EFI side of things.

Jan



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel