Re: [PATCH 3/5] irq_work: Implement remote queueing

[Frederic Weisbecker <fweisbec@gmail.com>]

On Tue, Jun 03, 2014 at 01:02:39PM -0700, Eric Dumazet wrote:
> On Tue, 2014-06-03 at 18:37 +0200, Peter Zijlstra wrote:
> > On Tue, Jun 03, 2014 at 09:29:07AM -0700, Eric Dumazet wrote:
> > > > +bool irq_work_queue_on(struct irq_work *work, int cpu)
> > > > +{
> > > > +	/* All work should have been flushed before going offline */
> > > > +	WARN_ON_ONCE(cpu_is_offline(cpu));
> > > > +
> > > > +	/* Arch remote IPI send/receive backend aren't NMI safe */
> > > > +	WARN_ON_ONCE(in_nmi());
> > > > +
> > > > +	/* Only queue if not already pending */
> > > > +	if (!irq_work_claim(work))
> > > > +		return false;
> > > > +
> > > > +	if (llist_add(&work->llnode, &per_cpu(raised_list, cpu)))
> > > > +		native_send_call_func_single_ipi(cpu);
> > > > +
> > > > +	return true;
> > > > +}
> > > > +EXPORT_SYMBOL_GPL(irq_work_queue_on);
> > > > +
> > > 
> > > I am curious, this should only compile on x86, right ?
> > 
> > Oh, you tease, you forgot to say why you think this.
> > 
> > Are you referring to the in_nmi() usage? that's from
> > include/linux/hardirq.h, hardly x86 specific.
> 
> No, my eyes were attracted by native_send_call_func_single_ipi()

Right, should be arch_send_call_function_single_ipi(). I'm like unable
to get a correct patchset before at least 9 takes...