[PATCH RFC v2 1/1] x86: default to not accepting Xeon Phi insns

[Jiang, Haochen]

> From: Jan Beulich <jbeulich@suse.com>
> Sent: Monday, August 4, 2025 3:34 PM
> 
> On 04.08.2025 04:49, Jiang, Haochen wrote:
> >> From: Jan Beulich <jbeulich@suse.com>
> >> Sent: Friday, August 1, 2025 9:34 PM
> >>
> >> Gcc15 has Xeon Phi support removed. While we don't want to remove
> >> support, require these deprecated insns to be enabled explicitly.
> >> ---
> >> RFC: Likely this wants holding back until gcc15 was actually released,
> >>      to avoid negatively affecting in particular gcc14. Question is
> >>      whether just the release of gcc15 is enough, or whether we want
> >>      people to be able to build/use gcc14 with newer binutils as well,
> >>      e.g. until that branch is retired.
> >>
> >
> > I agree that the time when GCC15 branch retirement is the safest choice.
> 
> FTAOD, I meant the gcc14 branch though. Gcc15, with Phi support removed,
> doesn't require its default-enabled support in gas.

Yes, it should be GCC14 retirement, which means two years later.

> 
> > While at least for now we did not get complaints about removal in GCC,
> > we could also be more aggressive. Timing like one year after GCC15.2
> > got released could be ok since most people will use GCC15.2 when
> > they start to use GCC15. This will ease rebase effort.
> 
> Hmm, this would feel somewhat random / arbitrary to me.
> 

It is the most aggressive way from my thought. Considering even for that
timing, it is actually only one year before GCC14 branch get retired, maybe
not that needed. Just providing a choice to discuss.

Thx,
Haochen