Re: Bug in SCHED_IDLE interaction with group scheduling?

[Ingo Molnar <mingo@elte.hu>]

* Paul Turner <pjt@google.com> wrote:

> Hi Peter,
> 
> It seems like there's an assumption in the sched_idle policy check in 
> place_entity that the sched_entity we are looking at belongs to a task?
> 
> - Paul
> 
> --
> 
> sched: bug in SCHED_IDLE interaction with group scheduling
> 
> One of the isolation modifications for SCHED_IDLE is the 
> unitization of sleeper credit.  However the check for this assumes 
> that the sched_entity we're placing always belongs to a task.
> 
> This is potentially not true with group scheduling and leaves us 
> rummaging randomly when we try to pull the policy.
> 
> Signed-off-by: Paul Turner <pjt@google.com>
> ---
>  kernel/sched_fair.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
> 
> diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
> index ba7fd6e..7c248dc 100644
> --- a/kernel/sched_fair.c
> +++ b/kernel/sched_fair.c
> @@ -687,7 +687,8 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
>  			 * all of which have the same weight.
>  			 */
>  			if (sched_feat(NORMALIZED_SLEEPER) &&
> -					task_of(se)->policy != SCHED_IDLE)
> +					(!entity_is_task(se) ||
> +					 task_of(se)->policy != SCHED_IDLE))
>  				thresh = calc_delta_fair(thresh, se);

Ah, nice catch!

( Sidenote: we keep having these subtle bugs where we call 
  task_of(se) on a non-task entity - it's been the fifth such 
  incident or so. Perhaps we should add an active debug check to 
  task_of(), in the case of CONFIG_SCHED_DEBUG=y. )

	Ingo