Re: Xen DOM0 pirq panic with lots of cores

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Mon, Apr 13, 2026 at 03:03:01PM -0400, Greg Troxel wrote:
> Staffan Thomen <staffan%shangtai.net@localhost> writes:
> 
> > The Xen howto wiki page states that dom0_max_vcpus and dom0_vcpus_pin
> > is no longer necessary;
> >
> > "With NetBSD 10 and up, there does not seem to be an argument that
> > pinning or limiting CPUs is a good idea."
> 
> I probably wrote that, and if so it was meant in terms of the intended
> purposes not being useful, and it did not mean that I was sure there
> were no bugs that would be avoided.  Don't take it too seriously.
> 
> > [   1.0000030] panic: pirq 16639 out of bound, increase  NR_PIRQS
> >
> > Limiting the CPUs to 1 like I used to avoid this issue an everything
> > works like I'd expect.
> >
> > It seems like some discrepancy between GENERIC vs. XEN3_DOM0 here?
> >
> > The system is not yet in production so I can do tests for a brief time
> > if necessary.
> 
> I (who is unlikely to fix this) would be interested in
> 
>   what is the highest number of cpus that is stable?  even roughly from
>   binary search, might be interesting.
> 
>   what is the value of NR_PIRQS in GENERIC and XEN3_DOM0?   Is this as
>   simple as a default being different?

NR_PIRQS doens't exists in GENERIC.
It's an array mapping physical IRQs to Xen events. An array doesn't scale
well on hardware with lots of core; something else is needed here.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--