Re: [RESEND PATCH v4 09/17] sched/deadline: Reset the deadline and the runtime with the parameters

From: sashiko-bot

Date: Thu Nov 19 2026 - 11:50:13 EST


Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [Medium] sched/deadline: donor runtime and deadline retained after PI
deboost
- [Medium] sched/deadline: uncanceled dl_timer in __dl_clear_params()
--

commit 2c25e238b8ef61f400892fff248fbabebd5863e5
Author: Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx>

sched/deadline: Reset the deadline and the runtime with the parameters

This commit resets dl_se->deadline and dl_se->runtime in __dl_clear_params()
and in switched_from_dl() when a boosted task with no deadline parameters of
its own leaves SCHED_DEADLINE, so that the next entry starts a new period.

> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index 71b1c155c2424..d6d2657dcddd6 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -3569,6 +3569,16 @@ static void switched_from_dl(struct rq *rq, struct task_struct *p)
> if (task_on_rq_queued(p) && p->dl.dl_runtime)
> task_non_contending(&p->dl, false);
>
> + /*
> + * A task with no parameters of its own was boosted into the class.
> + * Its deadline and runtime are the donor's. Reset them, so that its
> + * next entry into the class starts a new period.
> + */
> + if (!p->dl.dl_runtime) {
> + p->dl.deadline = 0;
> + p->dl.runtime = 0;
> + }

[Severity: Medium]
Can a task still retain the donor's deadline and runtime after PI deboost when
p->dl.dl_runtime is non-zero?

If a non-deadline task is PI-boosted by a SCHED_DEADLINE waiter and calls
sched_setattr() to enter SCHED_DEADLINE while still boosted, prev_class and
next_class in __sched_setscheduler() are both &dl_sched_class, so neither
switched_from_dl() nor switched_to_dl() runs. __setparam_dl() sets
p->dl.dl_runtime without clearing the donor's deadline or runtime (and if p is
sleeping, skipping switched_to_dl() also skips add_rq_bw(), later triggering
WARN_ON_ONCE(dl_rq->running_bw > dl_rq->this_bw) and underflow in
__sub_rq_bw()). When rt_mutex_setprio() later deboosts p, it stays in
dl_sched_class and restores p->dl.pi_se = &p->dl without ENQUEUE_REPLENISH, so
enqueue_dl_entity() sees the donor's future deadline and skips
setup_new_dl_entity(), letting p run unboosted on the donor's remaining
runtime.

Similarly, if a task leaves SCHED_DEADLINE before its 0-lag time (or changes
policy while boosted via __setscheduler_dl_pi()) and is replenished from a
donor while p->dl.dl_runtime is still non-zero, switched_from_dl() on deboost
skips this !p->dl.dl_runtime reset and passes the donor's deadline and runtime
to task_non_contending().

> @@ -3984,7 +3994,8 @@ bool __checkparam_dl(const struct sched_attr *attr)
> }
>
> /*
> - * This function clears the sched_dl_entity static params.
> + * Clear the static parameters of @dl_se. Also reset its deadline and its
> + * runtime, so that its next enqueue starts a new period.
> */
> static void __dl_clear_params(struct sched_dl_entity *dl_se)
> {
> @@ -3995,6 +4006,9 @@ static void __dl_clear_params(struct sched_dl_entity *dl_se)
> dl_se->dl_bw = 0;
> dl_se->dl_density = 0;
>
> + dl_se->deadline = 0;
> + dl_se->runtime = 0;
> +
> dl_se->dl_throttled = 0;
> dl_se->dl_yielded = 0;
> dl_se->dl_non_contending = 0;

[Severity: Medium]
Does resetting deadline, runtime, and dl_throttled in __dl_clear_params()
without canceling dl_se->dl_timer leave a stale replenishment timer queued?

When a task yields with remaining runtime (or uses a constrained deadline with
dl_deadline < dl_period) and is throttled, its 0-lag time arrives before
dl_se->dl_timer fires at dl_next_period(dl_se). If the task switches out of
SCHED_DEADLINE while throttled, __dl_clear_params() runs at the 0-lag time and
clears deadline and dl_throttled while dl_se->dl_timer remains queued. If the
task re-enters SCHED_DEADLINE before that old timer expires,
enqueue_dl_entity() starts a new period, and when the task throttles in the
new period, start_dl_timer() bails out because hrtimer_is_queued(timer) is
already true:

kernel/sched/deadline.c:start_dl_timer() {
...
if (!hrtimer_is_queued(timer)) {
if (!dl_server(dl_se))
get_task_struct(dl_task_of(dl_se));
hrtimer_start(timer, act, HRTIMER_MODE_ABS_HARD);
}

return 1;
}

When the stale dl_timer fires at the old period boundary, dl_task_timer()
replenishes and unthrottles the task in the middle of its new period, allowing
it to exceed its reservation.

--
Sashiko AI review · https://sashiko.dev/#/patchset/cover.1795105408.git.mathieu.desnoyers@xxxxxxxxxxxx?part=9