Re: [PATCH v1] sched/uclamp: Skip uclamp_rq_dec() for non-final dequeue of delayed tasks

[Zihuan Zhang]

Hi Xuewen,

在 2025/7/1 18:49, Xuewen Yan 写道:
> Hi zihuan,
>
> On Tue, Jul 1, 2025 at 5:34 PM Zihuan Zhang <zhangzihuan@xxxxxxxxxx> wrote:
> > Currently, uclamp_rq_inc() skips updating the clamp aggregation for
> > delayed tasks unless ENQUEUE_DELAYED is set, to ensure we only track the
> > real enqueue of a task that was previously marked as sched_delayed.
> >
> > However, the corresponding uclamp_rq_dec() path only checks
> > sched_delayed, and misses the DEQUEUE_DELAYED flag. As a result, we may
> > skip dec for a delayed task that is now being truly dequeued, leading to
> > uclamp aggregation mismatch.
> >
> > This patch makes uclamp_rq_dec() consistent with uclamp_rq_inc() by
> > checking both sched_delayed and DEQUEUE_DELAYED, ensuring correct
> > accounting symmetry.
> >
> > Fixes: 90ca9410dab2 ("sched/uclamp: Align uclamp and util_est and call before freq update")
> > Signed-off-by: Zihuan Zhang <zhangzihuan@xxxxxxxxxx>
> > ---
> >   kernel/sched/core.c | 9 +++++----
> >   1 file changed, 5 insertions(+), 4 deletions(-)
> >
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 8988d38d46a3..99f1542cff7d 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -1781,7 +1781,7 @@ static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p, int flags
> >                  rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE;
> >   }
> >
> > -static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p)
> > +static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p, int flags)
> >   {
> >          enum uclamp_id clamp_id;
> >
> > @@ -1797,7 +1797,8 @@ static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p)
> >          if (unlikely(!p->sched_class->uclamp_enabled))
> >                  return;
> >
> > -       if (p->se.sched_delayed)
> > +       /* Skip dec if this is a delayed task not being truly dequeued */
> > +       if (p->se.sched_delayed && !(flags & DEQUEUE_DELAYED))
> >                  return;
> Consider the __sched_setscheduler()，when changing the delayed-task's
> sched-class，your patch would dec the uclamp twice.

Thanks for your clarification. I’ve understood that the kernel 
unconditionally decrements the uclamp counter when dequeuing a task,

regardless of whether it’s currently enqueued. That makes sense.

Appreciate your time and explanation!
> BR
> ---
> xuewen
>
> >          for_each_clamp_id(clamp_id)
> > @@ -2039,7 +2040,7 @@ static void __init init_uclamp(void)
> >
> >   #else /* !CONFIG_UCLAMP_TASK */
> >   static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p, int flags) { }
> > -static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { }
> > +static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p, int flags) { }
> >   static inline void uclamp_fork(struct task_struct *p) { }
> >   static inline void uclamp_post_fork(struct task_struct *p) { }
> >   static inline void init_uclamp(void) { }
> > @@ -2112,7 +2113,7 @@ inline bool dequeue_task(struct rq *rq, struct task_struct *p, int flags)
> >           * Must be before ->dequeue_task() because ->dequeue_task() can 'fail'
> >           * and mark the task ->sched_delayed.
> >           */
> > -       uclamp_rq_dec(rq, p);
> > +       uclamp_rq_dec(rq, p, flags);
> >          return p->sched_class->dequeue_task(rq, p, flags);
> >   }
> >
> > --
> > 2.25.1
Best regards,
Zihuan