On Fri, 27 Nov 2015, Jiang Liu wrote:
Please trim your replies.
> On 2015/11/26 5:12, Thomas Gleixner wrote:
> > Looks a bit overkill with the extra cpumask. I tried a simple counter
> > but that does not work versus cpu unplug as we do not know whether the
> > outgoing cpu is involved in the cleanup or not. And if the cpu is
> > involved we starve assign_irq_vector() ....
> >
> > The upside of this is that we get rid of that atomic allocation in
> > __send_cleanup_vector().
>
> Maybe more headache for you now:)
> It seems there are still rooms for improvements. First it
> seems we could just reuse old_domain instead of adding cleanup_mask.
I really like to get rid of that atomic allocation in
__send_cleanup_vector()
> Second I found another race window among x86_vector_free_irqs(),
> __send_cleanup_vector() and smp_irq_move_cleanup_interrupt().
What's the race there?
> I'm trying to refine your patch based following rules:
> 1) move_in_progress controls whether we need to send IPIs
> 2) old_domain controls which CPUs we should do clean up
> 3) assign_irq_vector checks both move_in_progress and old_domain.
> Will send out the patch soon for comments:)
Sure.
Thanks,
tglx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
