Re: [PATCH V2 3/3] workqueue: Schedule work on non-idle cpu insteadof current one

[Vincent Guittot]

On 1 October 2012 05:47, Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> On 1 October 2012 06:02, Tejun Heo <tj@xxxxxxxxxx> wrote:
>> It isn't about the CPU being actually idle?
>
> No. Being idle only from scheduler's perspective. :)
>
>> Also, if it's only about timers, shouldn't it be enough to implement
>> it for timer and delayed_work?
>
> What if we need a timer, which must re-arm itself + schedule a work?
> delayed_work will not be sufficient in that case, and we would need
> to use normal work.
>
> If i am not wrong, there can be other future users of this routine too.
> @Vincent: Can you please comment on this?

The goal is to consolidate the place, where the selection of a target
CPU for running something is done. The scheduler should select the CPU
in order to have coherent behaviour of all framework.

A delayed work can be delayed for a long period and the target CPU
that have been selected for the timer could not be the best place to
queue a work any more.

The goal is to be able to migrate a work if the scheduler thinks that'
is necessary.

Vincent

>
>> It would be great if you explain what you're trying to achieve how.  I
>> can't tell what you're aiming for and why that would be beneficial
>> from these submissions.
>
> Following slides are implemented by Vincent and presented during LPC.
> Please have a look at them, they explain the problem statement well:
>
> http://www.linuxplumbersconf.org/2012/wp-content/uploads/2012/08/lpc2012-sched-timer-workqueue.pdf
>
> Specifically slides: 12 & 19.
>
> There aren't too many users with this behavior, but even a single user
> will be sufficient not to let the cpu get idle at all. And that will result in
> less power saving.
>
> --
> viresh
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/