Re: [patch] Add allowed_affinity to the irq_desc to make it possible to have restricted irqs

[ebiederm@xmission.com (Eric W. Biederman)]

Ingo Molnar <mingo@elte.hu> writes:

> * Eric W. Biederman <ebiederm@xmission.com> wrote:
>
>> In addition the cases I can think of allowed_affinity is the wrong 
>> name.  suggested_affinity sounds like what you are trying to implement 
>> and when it is merely a suggestion and not a hard limit it doesn't 
>> make sense to export like this.
>
> well, there are interrupts that must be tied to a single CPU and must 
> never be moved away. For example per-CPU clock-events-source interrupts 
> are such. So allowed_affinity very much exists.

Although in that case since it is a single cpu there is a much
more elegant implementation.  We don't need a full cpumask_t to
describe it.

> also there might be hardware that can only route a given IRQ to a subset 
> of CPUs. While setting set_affinity allows the irqbalance-daemon to 
> 'probe' this mask, it's a far from optimal API.

I agree, I am just arguing that adding another awkward interface to
the current situation does not really make the situation better, and
it increases our support burden.

For a bunch of this it is arguable that the way to go is simply to
parse the irq type in /proc/interrupts.  All of the really weird cases
will have a distinct type there.  This certainly captures the MSI-X
case.  There is still a question of how to handle the NUMA case but...

Eric