Re: [PATCH V2] idle/intel_powerclamp: Redesign idle injection to use bandwidth control mechanism

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Feb 09, 2015 at 10:19:43AM +0530, Preeti U Murthy wrote:
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 8db31ef..6a7ccb2 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -3002,6 +3002,12 @@ extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
>  
>  #ifdef CONFIG_CGROUP_SCHED
>  extern struct task_group root_task_group;
> +extern int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us);
> +extern int tg_set_cfs_period(struct task_group *tg, long cfs_period_us);
> +#else
> +
> +static inline int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us);
> +static inline int tg_set_cfs_period(struct task_group *tg, long cfs_period_us);
>  #endif /* CONFIG_CGROUP_SCHED */

Instead you might want to make the whole powerclamp thing depend on
CONFIG_CFS_BANDWIDTH.

Also, exposing these and root_task_group is of course vile. Not to
mention you change the user (cgroup) interface without mention.

In any case, I cannot see how this could ever work. Bandwidth is shared
across CPUs; nothing will even attempt to get CPUs to idle at the same
time.