On Mon, Feb 15, 2016 at 8:23 PM, Russell King - ARM Linux
[off-list ref] wrote:

On Mon, Feb 15, 2016 at 07:03:33PM +0000, Marc Zyngier wrote:

quoted

On 15/02/16 18:54, Rafael J. Wysocki wrote:

quoted

That would explain it, thanks.

So it looks like we should always use irq_work_queue() on UP even if
CONFIG_SMP is set, shouldn't we?

Something like that, yes. CONFIG_SMP is not an indication of an SMP
system anymore (we've even dropped the config option on arm64).

Hopefully num_possible_cpus() is reliable enough to let you do the right
thing...

CONFIG_SMP just says whether to include support for SMP.  It doesn't
mandate running on a SMP system. :)

I've been looking around the usages of irq_work_queue_on in kernel/
in -rc4, and some places seem to check for "this CPU":

        /*
         * It is possible that a restart caused this CPU to be
         * chosen again. Don't bother with an IPI, just see if we
         * have more to push.
         */
        if (unlikely(cpu == rq->cpu))
                goto again;

        /* Try the next RT overloaded CPU */
        irq_work_queue_on(&rt_rq->push_work, cpu);

I'm not sure about tell_cpu_to_push().

It's also called via tick_nohz_full_kick_cpu(), and the core scheduler
avoids calling this for the current CPU:

        if (tick_nohz_full_cpu(cpu)) {
                if (cpu != smp_processor_id() ||
                    tick_nohz_tick_stopped())
                        tick_nohz_full_kick_cpu(cpu);

I'm not sure about add_nr_running() in kernel/sched/sched.h - I think
that _could_ be a problem even without Rafael's cpufreq change.

So... the question is what do we do with irq_work_queue_on() in general
when called on non-SMP systems.

I guess it might fall back to arch_irq_work_raise() when asked to
queue on the same CPU, so long as that will always do the right thing
(ie. actually queue on the same one).

Thanks,
Rafael