Re: NR_PIRQS vs. NR_IRQS

["Jan Beulich" <jbeulich@novell.com>]

>>> Keir Fraser <keir.fraser@eu.citrix.com> 14.11.08 09:00 >>>
>On 14/11/08 07:54, "Keir Fraser" <keir.fraser@eu.citrix.com> wrote:
>
>>>> I agree with keeping this naming distinction of course, although I think
>>>> allowing NR_IRQS > NR_VECTORS right now is not very useful. But maybe you
>>>> have a box in mind that needs it?
>>> 
>>> I had sent a mail a few days ago on this, where IBM was testing 96 CPU
>>> support (4-node system), and it crashing because of a PIRQ ending up in
>>> DYNIRQ space (kernel perspective), because there being 300+ IO-APIC
>>> pins. While the crash ought to be fixed with the subsequent patch, it's
>>> clear that none of the devices with an accumulated pin number greater
>>> than 255 will actually work on that system.
>> 
>> Oh dear. :-D
>
>Is fixing this actually any harder than just bumping NR_IRQS/NR_PIRQS in Xen
>and NR_PIRQS in Linux? Have IRQS and VECTORS got somehow accidentally tied
>together in Xen?

Unfortunately there was some mix-up. But I think I got this fixed (running
a NR_IRQS=1024 hypervisor at the moment). I'd prefer to submit these
patches only after your NR_PIRQS -> NR_IRQS change comes out of the
staging tree though, to make sure it applies (and works) correctly on top
of it.

>These parameters should probably be build-time configurable.

This too.

Jan