Re: [regression] commit d298b03506d3 ("x86/fpu: Restore the masking out of reserved MXCSR bits")

[Ville Syrjälä]

On Thu, Oct 14, 2021 at 05:34:14PM +0300, Ville Syrjälä wrote:
> On Thu, Oct 14, 2021 at 04:27:07PM +0200, Borislav Petkov wrote:
> > On Thu, Oct 14, 2021 at 02:44:33PM +0300, Ville Syrjälä wrote:
> > > I have a 32bit installation here that stopped working. Bisected it
> > > to commit d298b03506d3 ("x86/fpu: Restore the masking out of reserved
> > > MXCSR bits").
> > 
> > Lemme make sure I understand this correctly: this patch is bad and with
> > it reverted it works?
> 
> Yes.
> 
> >
> > Because before this patch, the restoring would be a more restrictive
> > than before and this patch reverts the code to the old behavior for
> > invalid MXCSR bits.

Yeah, it's a bit weird.

Hmm. Actually I just stared at the code a bit more it looks
a bit funny. Was it supposed to do this instead?

- fpu->state.fxsave.mxcsr &= ~mxcsr_feature_mask;
+ fpu->state.fxsave.mxcsr &= mxcsr_feature_mask;

-- 
Ville Syrjälä
Intel