Re: [RFC] vmstat: Avoid waking up idle-cpu to service shepherd work

[Viresh Kumar <viresh.kumar@linaro.org>]

On 30 March 2015 at 18:17, Peter Zijlstra <peterz@infradead.org> wrote:
> No, I means something else with that. We can remove the
> tvec_base::running_timer field. Everything that uses that can use
> tbase_running() AFAICT.

Okay, there is one instance which still needs it.

migrate_timers():

        BUG_ON(old_base->running_timer);

What I wasn't sure about it is if we get can drop this statement or not.
If we decide not to drop it, then we can convert running_timer into a bool.

> Drop yes, racy not so much I think.
>
>
> diff --git a/kernel/time/timer.c b/kernel/time/timer.c
> index 2d3f5c504939..1394f9540348 100644
> --- a/kernel/time/timer.c
> +++ b/kernel/time/timer.c
> @@ -1189,12 +1189,39 @@ static inline void __run_timers(struct tvec_base *base)
>                         cascade(base, &base->tv5, INDEX(3));
>                 ++base->timer_jiffies;
>                 list_replace_init(base->tv1.vec + index, head);
> +
> +again:
>                 while (!list_empty(head)) {
>                         void (*fn)(unsigned long);
>                         unsigned long data;
>                         bool irqsafe;
>
> -                       timer = list_first_entry(head, struct timer_list,entry);
> +                       timer = list_first_entry(head, struct timer_list, entry);
> +                       if (unlikely(tbase_running(timer))) {
> +                               /* Only one timer on the list, force wait. */
> +                               if (unlikely(head->next == head->prev)) {
> +                                       spin_unlock(&base->lock);
> +
> +                                       /*
> +                                        * The only way to get here is if the
> +                                        * handler requeued itself on another
> +                                        * base, this guarantees the timer will
> +                                        * not go away.
> +                                        */
> +                                       while (tbase_running(timer))
> +                                               cpu_relax();
> +
> +                                       spin_lock(&base->lock);
> +                               } else  {
> +                                       /*
> +                                        * Otherwise, rotate the list and try
> +                                        * someone else.
> +                                        */
> +                                       list_move_tail(&timer->entry, head);
> +                               }
> +                               goto again;
> +                       }
> +
>                         fn = timer->function;
>                         data = timer->data;
>                         irqsafe = tbase_get_irqsafe(timer->base);

Yeah, so I have written something similar only. Wasn't sure about what you wrote
earlier. Thanks for the clarification.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>