Re: [PATCH] timers/nohz: Update nohz load even if tick already stopped

[Scott Wood]

On Wed, 2019-10-30 at 14:31 +0100, Peter Zijlstra wrote:
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index eb42b71faab9..d02d1b8f40af 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -3660,21 +3660,17 @@ static void sched_tick_remote(struct work_struct
> *work)
>  	u64 delta;
>  	int os;
>  
> -	/*
> -	 * Handle the tick only if it appears the remote CPU is running in
> full
> -	 * dynticks mode. The check is racy by nature, but missing a tick or
> -	 * having one too much is no big deal because the scheduler tick
> updates
> -	 * statistics and checks timeslices in a time-independent way,
> regardless
> -	 * of when exactly it is running.
> -	 */
> -	if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu))
> +	if (!tick_nohz_tick_stopped_cpu(cpu))
>  		goto out_requeue;
>  
>  	rq_lock_irq(rq, &rf);
> -	curr = rq->curr;
> -	if (is_idle_task(curr) || cpu_is_offline(cpu))
> +	/*
> +	 * We must not call calc_load_nohz_remote() when not in NOHZ mode.
> +	 */
> +	if (cpu_is_offline(cpu) || !tick_nohz_tick_stopped(cpu))
>  		goto out_unlock;

Is it really a problem if calc_load_nohz_remote() gets called in
non-NOHZ?  It won't race due to rq lock -- and we're already mixing
remote and non-remote updates because the normal tick timer can still be
run while "stopped".

-Scott