Re: [PATCHv2] sched: Account I/O waits independently of delayed dequeue

[Kayra Cizmeci <kayracizmeci@gmail.com>]

On Thu, Oct 15, 2026 at 12:37:54PM +0100, Christian Loehle wrote:
Normally I wouldn't add these. But I changed my setup. It adds
these itself. Not a big fan of changes but this is way more easier...

> ---
>  include/linux/sched.h |  2 ++
>  kernel/sched/core.c   | 33 ++++++++++++++++++++-------------
>  kernel/sched/sched.h  |  5 -----
>  3 files changed, 22 insertions(+), 18 deletions(-)
> 
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index d35ae49a991f..d558816cc05a 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -880,6 +880,8 @@ struct task_struct {
>  	 */
>  	int				recent_used_cpu;
>  	int				wake_cpu;
> +	/* CPU charged while is_blocked && in_iowait. */
> +	int				iowait_cpu;
>  
>  	int				prio;
>  	int				static_prio;

I wondered if the charge could follow the task while it changes CPUs,
but I think that it would break down more easily than this.

Still, there could be a better approach but right now, I can't find
any. Maybe there's a section that we could reuse? But IDK.

> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 1fe40de6ebe3..85eba26f00d6 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -636,7 +636,7 @@ int task_llc(const struct task_struct *p)
>   *
>   * p->is_blocked <- { 0, 1 }:
>   *
> - *   is set by try_to_block_task() and cleared by ttwu_do_wakeup() and tracks
> + *   is set by try_to_block_task() and cleared by unblock_task() and tracks
>   *   if the task is blocked. Traditionally this would mirror p->on_rq, however
>   *   due things like DELAY_DEQUEUE and PROXY_EXEC, this can diverge.
>   *
> @@ -3737,12 +3737,22 @@ ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
>  		__schedstat_inc(p->stats.nr_wakeups_sync);
>  }
>  
> +static inline void unblock_task(struct task_struct *p)
> +{
> +	if (p->is_blocked && p->in_iowait) {
> +		delayacct_blkio_end(p);
> +		atomic_dec(&cpu_rq(p->iowait_cpu)->nr_iowait);
> +	}
> +
> +	p->is_blocked = 0;
> +}
> +
>  /*
>   * Mark the task runnable.
>   */
>  static inline void ttwu_do_wakeup(struct task_struct *p)
>  {
> -	p->is_blocked = 0;
> +	unblock_task(p);
>  	WRITE_ONCE(p->__state, TASK_RUNNING);
>  	trace_sched_wakeup(p);
>  }
> @@ -3839,10 +3849,6 @@ ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
>  		en_flags |= ENQUEUE_RQ_SELECTED;
>  	if (wake_flags & WF_MIGRATED)
>  		en_flags |= ENQUEUE_MIGRATED;
> -	else if (p->in_iowait) {
> -		delayacct_blkio_end(p);
> -		atomic_dec(&task_rq(p)->nr_iowait);
> -	}
>  
>  	activate_task(rq, p, en_flags);
>  	wakeup_preempt(rq, p, wake_flags);
> @@ -4415,11 +4421,6 @@ int try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
>  
>  		cpu = select_task_rq(p, p->wake_cpu, &wake_flags);
>  		if (task_cpu(p) != cpu) {
> -			if (p->in_iowait) {
> -				delayacct_blkio_end(p);
> -				atomic_dec(&task_rq(p)->nr_iowait);
> -			}
> -
>  			wake_flags |= WF_MIGRATED;
>  			psi_ttwu_dequeue(p);
>  			set_task_cpu(p, cpu);
> @@ -6762,6 +6763,12 @@ static bool try_to_block_task(struct rq *rq, struct task_struct *p,
>  
>  	p->is_blocked = 1;
>  
> +	if (p->in_iowait) {
> +		p->iowait_cpu = cpu_of(rq);
> +		atomic_inc(&rq->nr_iowait);
> +		delayacct_blkio_start();
> +	}
> +

So now we increase this nr_iowait on try_to_block_task(), and decrease it on unblock_task().

Also, there's a wrong comment about this in here:

 /*
  * This task is about to go to sleep on IO. Increment rq->nr_iowait so 
  * that process accounting knows that this is a task in IO wait state.
  */
 long __sched io_schedule_timeout(long timeout)
 {
 	int token;
 	long ret;
 
 	token = io_schedule_prepare();
 	ret = schedule_timeout(timeout);
 	io_schedule_finish(token);
 
 	return ret;
 }

This doesn't even touch nr_iowait, so this is stale.
Maybe remove this comment or just correct this while in this?

Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>