Re: [RFC] Processing of raised_list can stall if an IPI/interrupt is missed

[Herton Krzesinski]

On Tue, Mar 3, 2026 at 7:39 PM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>
> On Tue, Mar 03 2026 at 20:29, Peter Zijlstra wrote:
>
> > On Tue, Mar 03, 2026 at 04:07:15PM -0300, Herton R. Krzesinski wrote:
> >
> >> Or may be it's not worth changing this since this is rare and missed self IPI should
> >> not be expected?
> >
> > IPIs going missing is certainly unexpected; although kernel/smp.c has
> > much debugging crud for similar scenarios (AFAIK all of them related to
> > virt).
> >
> > If IPIs go missing your system *will* go funny in one way or another.
> > I'm not totally against building in some fallback, but we should
> > definitely consider it dodgy/buggy if we do detect one has gone
> > walk-about.
>
> Indeed and there is enough code by now which relies on the pristine IPI
> context when the architecture supports it, so basically reverting this
> change is going to create a boatload of other hard to debug problems.

Ok thank you Peter and Thomas. So far it's very rare for it to occur from the
reports I have, not worth changing/doing the revert then. I assume it was
broken virt in those cases.

>
> Thanks
>
>         tglx
>