Re: [PATCH 03/20 v3] sched/eevdf: Reset lag at wakeup when migrating and prev has been idle

[Kayra Cizmeci <kayracizmeci@gmail.com>]

On Fri, Nov 06, 2026 at 04:19:30PM +0100, Vincent Guittot wrote:
> When prev CPU entered idle while a task was sleeping, we assume that the
> latter received enough running time and the lag can be cleared similarly to
> what happens for a negative lag.
> While migrating the waking task to a new cpu, Check if prev CPU entered
> idle in the meantime.
> 
> Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
> ---
>  kernel/sched/fair.c | 27 +++++++++++++++++++++++++++
>  1 file changed, 27 insertions(+)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 79f53811274e..945de0215abb 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -932,6 +932,27 @@ bool update_entity_lag(struct cfs_rq *cfs_rq, struct sched_entity *se)
>  static inline unsigned long cfs_rq_load_avg(struct cfs_rq *cfs_rq);
>  static unsigned long task_h_load(struct task_struct *p);
>  
> +static __always_inline
> +void entity_clear_lag(struct task_struct *p)
> +{
> +	struct sched_entity *se;
> +	struct cfs_rq *cfs_rq;
> +
> +	if (READ_ONCE(p->__state) != TASK_WAKING)
> +		return;
> +
> +	se = &p->se;
> +	/* Negative lag implies delayed dequeue */
> +	if (se->vlag <= 0)
> +		return;
> +
> +	cfs_rq = &task_rq(p)->cfs;
> +
> +	/* CPU has been idle in between so the lag has been removed */
> +	if (se->vlag_seq != READ_ONCE(cfs_rq->idle_seq))
> +		se->vlag = 0;
> +}
> +
>  static __always_inline
>  void decay_entity_lag(struct cfs_rq *cfs_rq, struct task_struct *p, int flags)
>  {
> @@ -10049,6 +10070,12 @@ static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
>  	struct sched_entity *se = &p->se;
>  
>  	if (!task_on_rq_migrating(p)) {
> +		/*
> +		 * Clear the positive lag of a waking task before migrating to
> +		 * the new cpu.
> +		 */
> +		entity_clear_lag(p);
> +
>  		remove_entity_load_avg(se);
>  
>  		/*
> -- 
> 2.53.0
> 

Yeah well looks good. I need to test this on my head tho:

CPU1:    CPU2:
TA       TD
TB
TC

Let's say TA sleeps and has 12 lags. And while its sleeping TB and
TC leaves:

CPU1:
TA (Sleeping)

Then some new tasks come:

CPU1:
TA (Sleeping)
TW
TR

Now TA wakes up, we enter try_to_wake_up(), from there we pick CPU2 and
enter set_task_cpu(), from there migrate_task_rq_fair(), and to
entity_clear_lag().

And from there this just works correctly.


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



Cheers,

Kayra