Re: [PATCH v2 4/7] x86/irq: restrict CPU movement in set_desc_affinity()

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Tue, Jun 11, 2024 at 12:20:33PM +0200, Jan Beulich wrote:
> On 10.06.2024 16:20, Roger Pau Monne wrote:
> > If external interrupts are using logical mode it's possible to have an 
> > overlap
> > between the current ->arch.cpu_mask and the provided mask (or TARGET_CPUS). 
> >  If
> > that's the case avoid assigning a new vector and just move the interrupt to 
> > a
> > member of ->arch.cpu_mask that overlaps with the provided mask and is 
> > online.
> 
> What I'm kind of missing here is an explanation of why what 
> _assign_irq_vector()
> does to avoid unnecessary migration (very first conditional there) isn't
> sufficient.

Somehow I looked at that and think it wasn't enough, but now I cannot
figure out why, so it might be just fine, and this patch is not
needed.  Let me test again and get back to you, for the time being
ignore this patch.

Thanks, Roger.