Re: [PATCH 2/4] sched: Correctly handle nohz ticks cpu load accounting

[Peter Zijlstra]

On Fri, Apr 01, 2016 at 03:23:05PM +0200, Frederic Weisbecker wrote:
> Ticks can happen in the middle of a nohz frame and

I'm still miffed with that.. And this changelog doesn't even explain why
and how.

> cpu_load_update_active() doesn't handle these correctly. It forgets the
> whole previous tickless load and just records the current tick, ignoring
> potentially long idle periods.
> 
> In order to solve this, record the load on nohz frame entry so we know
> what to record in case of nohz interruptions, then use this recorded load
> to account the tickless load on nohz ticks and nohz frame end.


> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index f33764d..394f008 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -4527,9 +4527,9 @@ decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
>   * term. See the @active paramter.

^^^^^^^^^^^^^^

What active parameter... you need to update that comment.

>   */
>  static void __cpu_load_update(struct rq *this_rq, unsigned long this_load,
> -			      unsigned long pending_updates, int active)
> +			      unsigned long pending_updates)
>  {
> -	unsigned long tickless_load = active ? this_rq->cpu_load[0] : 0;
> +	unsigned long tickless_load = this_rq->cpu_load[0];
>  	int i, scale;
>  
>  	this_rq->nr_load_updates++;