Re: [RFC PATCH v15 4/7] sched: Fix runtime accounting w/ split exec & sched contexts

[K Prateek Nayak <kprateek.nayak@amd.com>]

Hello John,

On 3/13/2025 3:41 AM, John Stultz wrote:
> The idea here is we want to charge the scheduler-context task's
> vruntime but charge the execution-context task's sum_exec_runtime.
> 
> This way cputime accounting goes against the task actually running
> but vruntime accounting goes against the rq->donor task so we get
> proper fairness.
> 
> Cc: Joel Fernandes <joelagnelf@nvidia.com>
> Cc: Qais Yousef <qyousef@layalina.io>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Juri Lelli <juri.lelli@redhat.com>
> Cc: Vincent Guittot <vincent.guittot@linaro.org>
> Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> Cc: Valentin Schneider <vschneid@redhat.com>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Ben Segall <bsegall@google.com>
> Cc: Zimuzo Ezeozue <zezeozue@google.com>
> Cc: Mel Gorman <mgorman@suse.de>
> Cc: Will Deacon <will@kernel.org>
> Cc: Waiman Long <longman@redhat.com>
> Cc: Boqun Feng <boqun.feng@gmail.com>
> Cc: "Paul E. McKenney" <paulmck@kernel.org>
> Cc: Metin Kaya <Metin.Kaya@arm.com>
> Cc: Xuewen Yan <xuewen.yan94@gmail.com>
> Cc: K Prateek Nayak <kprateek.nayak@amd.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
> Cc: Suleiman Souhlal <suleiman@google.com>
> Cc: kernel-team@android.com
> Signed-off-by: John Stultz <jstultz@google.com>
> ---
>   kernel/sched/fair.c | 21 ++++++++++++++++-----
>   1 file changed, 16 insertions(+), 5 deletions(-)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index c798d27952431..f8ad3a44b3771 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -1129,22 +1129,33 @@ static void update_tg_load_avg(struct cfs_rq *cfs_rq)
>   }
>   #endif /* CONFIG_SMP */
>   
> -static s64 update_curr_se(struct rq *rq, struct sched_entity *curr)
> +static s64 update_curr_se(struct rq *rq, struct sched_entity *se)
>   {
>   	u64 now = rq_clock_task(rq);
>   	s64 delta_exec;
>   
> -	delta_exec = now - curr->exec_start;
> +	delta_exec = now - se->exec_start;
>   	if (unlikely(delta_exec <= 0))
>   		return delta_exec;
>   
> -	curr->exec_start = now;
> -	curr->sum_exec_runtime += delta_exec;
> +	se->exec_start = now;
> +	if (entity_is_task(se)) {
> +		struct task_struct *running = rq->curr;
> +		/*
> +		 * If se is a task, we account the time against the running
> +		 * task, as w/ proxy-exec they may not be the same.
> +		 */
> +		running->se.exec_start = now;
> +		running->se.sum_exec_runtime += delta_exec;
> +	} else {
> +		/* If not task, account the time against se */
> +		se->sum_exec_runtime += delta_exec;
> +	}
>   
>   	if (schedstat_enabled()) {
>   		struct sched_statistics *stats;
>   
> -		stats = __schedstats_from_se(curr);
> +		stats = __schedstats_from_se(se);
>   		__schedstat_set(stats->exec_max,
>   				max(delta_exec, stats->exec_max));

So I'm slightly confused here - For the case of proxy where
entity_is_task(), we charge the delta_exec to the running task's se but
then we go ahead and update the exec_max against the stats of the
donor's se? That seems odd.

Could we just replace the se with &rq->curr->se if entity_is_task()
returns true and keep the rest as is? The calculations will be same as
what it is above (except for the stats bit) and you'll not require
updating "exec_start" for both current task and proxy's se. Thoughts?


>   	}

-- 
Thanks and Regards,
Prateek