Re: [PATCH 1/2] sched/fair: Fix value reported by hot tasks pulled in /proc/schedstat

[Swapnil Sapkal]

Hello Peter,

On 6/19/2023 2:52 PM, Peter Zijlstra wrote:
> On Wed, Jun 14, 2023 at 10:22:23AM +0000, Swapnil Sapkal wrote:
> > In /proc/schedstat, lb_hot_gained reports the number hot tasks pulled
> > during load balance. This value is incremented in can_migrate_task()
> > if the task is migratable and hot. After incrementing the value,
> > load balancer can still decide not to migrate this task leading to wrong
> > accounting. Fix this by incrementing stats when hot tasks are detached.
> > This issue only exits in detach_tasks() where we can decide to not
> > migrate hot task even if it is migratable. However, in detach_one_task(),
> > we migrate it unconditionally.
> >
> > Fixes: d31980846f96 ("sched: Move up affinity check to mitigate useless redoing overhead")
> > Reported-by: Gautham R. Shenoy <gautham.shenoy@xxxxxxx>
> > Signed-off-by: Swapnil Sapkal <swapnil.sapkal@xxxxxxx>
> > ---
> >   kernel/sched/fair.c | 47 +++++++++++++++++++++++++++++----------------
> >   1 file changed, 30 insertions(+), 17 deletions(-)
>
> All this for just a number hardly anybody looks at :-(
>
> Does this also work?

Thank you for this patch which looks much simpler.
This will work. In addition we need to handle the following case: when the task is
hot and migratable, and detach_tasks() decides not to migrate the task, then we
should be incrementing nr_failed_migrations_hot. I am appending this hunk at the
end of this mail.

> Please double check the order of the task_struct::sched_bitfield thing,
> I've not had much wake-up juice
>
> ---
>   include/linux/sched.h |  1 +
>   kernel/sched/fair.c   | 14 ++++++++++----
>   2 files changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 1292d38d66cc..eba0a78ac2a9 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -887,6 +887,7 @@ struct task_struct {
>   	unsigned			sched_reset_on_fork:1;
>   	unsigned			sched_contributes_to_load:1;
>   	unsigned			sched_migrated:1;
> +	unsigned			sched_task_hot:1;
>   
>   	/* Force alignment to the next boundary: */
>   	unsigned			:0;
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 6189d1a45635..a88577132b20 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -8569,6 +8569,8 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
>   	int tsk_cache_hot;
>   
>   	lockdep_assert_rq_held(env->src_rq);
> +	if (p->sched_task_hot)
> +		p->sched_task_hot = 0;
>   
>   	/*
>   	 * We do not migrate tasks that are:
> @@ -8641,10 +8643,8 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
>   
>   	if (tsk_cache_hot <= 0 ||
>   	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
> -		if (tsk_cache_hot == 1) {
> -			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
> -			schedstat_inc(p->stats.nr_forced_migrations);
> -		}
> +		if (tsk_cache_hot == 1)
> +			p->sched_task_hot = 1;
>   		return 1;
>   	}
>   
> @@ -8659,6 +8659,12 @@ static void detach_task(struct task_struct *p, struct lb_env *env)
>   {
>   	lockdep_assert_rq_held(env->src_rq);
>   
> +	if (p->sched_task_hot) {
> +		p->sched_task_hot = 0;
> +		schedstat_inc(env->sd->lb_hot_gained[env->idle]);
> +		schedstat_inc(p->stats.nr_forced_migrations);
> +	}
> +
>   	deactivate_task(env->src_rq, p, DEQUEUE_NOCLOCK);
>   	set_task_cpu(p, env->dst_cpu);
>   }
---
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index b6a738514047..eb836629560f 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -8772,6 +8772,9 @@ static int detach_tasks(struct lb_env *env)

                continue;
 next:
+               if (p->sched_task_hot)
+                       schedstat_inc(p->stats.nr_failed_migrations_hot);
+
                list_move(&p->se.group_node, tasks);
        }