Re: [PATCH v8 26/33] sched/deadline: Start a period for a task that enters the class blocked
From: sashiko-bot
Date: Thu Nov 19 2026 - 15:57:19 EST
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Medium] sched/deadline: dl_task_pi_blocked() misses tasks sleeping
off rt_mutex
- [Medium] sched/deadline: missing period start in
__sched_setscheduler() when boosted
- [Medium] sched/deadline: dl_boost_blocked() skips departed tasks with
dl_deadline
--
commit 38735e4f5a4290a90781132f54e4120fda6dbb23
Author: Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx>
sched/deadline: Start a period for a task that enters the class blocked
This commit starts a new period in switched_to_dl() and dl_boost_blocked()
when a task blocked on an rt_mutex enters the deadline class or receives a
new priority-inheritance donor.
> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index 8ac78958a17af..77a1d1f463f6d 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -125,6 +125,12 @@ void dl_deboost(struct sched_dl_entity *dl_se)
> }
> dl_drop_boost(dl_se);
> }
> +
> +/* Return true if @p waits on an rt_mutex. The caller holds its pi_lock. */
> +static inline bool dl_task_pi_blocked(struct task_struct *p)
> +{
> + return p->pi_blocked_on;
> +}
[Severity: Medium]
Does gating replenish_dl_new_period() on dl_task_pi_blocked() leave
p->dl.deadline at 0 for tasks that sleep with p->pi_blocked_on == NULL?
In futex_wait_requeue_pi(), a task sleeps on the non-PI source futex with
p->pi_blocked_on == NULL before being requeued. If it enters the deadline
class while sleeping there (via sched_setattr() or a boost on another
rt_mutex it owns), switched_to_dl() and dl_boost_blocked() skip starting a
period. When FUTEX_CMP_REQUEUE_PI then proxy-locks the task onto the PI
futex without waking it, task_blocks_on_rt_mutex() copies p->dl.deadline
(still 0) into waiter->tree.deadline before setting p->pi_blocked_on,
placing the waiter ahead of all existing deadline waiters and boosting the
owner with deadline 0:
kernel/locking/rtmutex.c:task_blocks_on_rt_mutex() {
...
waiter_update_prio(waiter, task);
waiter_clone_prio(waiter, task);
...
task->pi_blocked_on = waiter;
...
}
Similarly, if a task holding an rt_mutex sleeps on a non-rt_mutex primitive
(p->pi_blocked_on == NULL), calling sched_setattr(SCHED_DEADLINE) leaves
p->dl.deadline at 0 so an earlier-deadline waiter blocking on that rt_mutex
sees dl_entity_preempt() return false and fails to boost it; or if the
sleeping owner has no reservation of its own and gets boosted by a second
waiter with an earlier deadline than the first, dl_boost_blocked() skips
the period start and ENQUEUE_REPLENISH is discarded because !scope->queued.
[ ... ]
> @@ -1116,6 +1127,30 @@ static inline void replenish_dl_new_period(struct sched_dl_entity *dl_se,
> trace_sched_dl_replenish_tp(dl_se, cpu_of(rq), dl_get_type(dl_se, rq));
> }
>
> +#ifdef CONFIG_RT_MUTEXES
> +/*
> + * Start a period from the new donor of @p, if @p has no reservation of its
> + * own and waits on an rt_mutex. The enqueue does this for a queued task, in
> + * replenish_dl_entity(). A blocked task is not enqueued, but the chain walk
> + * copies its deadline into its waiter.
> + *
> + * The caller calls it only when the donor changes. A new period changes the
> + * deadline, so the chain walk finds the waiter out of date and goes on. In a
> + * ww_mutex cycle, a new period at each step would never let the walk stop.
> + *
> + * The caller holds the rq lock and the pi_lock of @p, and has updated the
> + * rq clock.
> + */
> +void dl_boost_blocked(struct task_struct *p)
> +{
> + struct sched_dl_entity *dl_se = &p->dl;
> +
> + if (!dl_se->dl_deadline && !dl_se->dl_throttled &&
> + dl_task_pi_blocked(p))
> + replenish_dl_new_period(dl_se, task_rq(p));
> +}
> +#endif
[Severity: Medium]
Can checking !dl_se->dl_deadline in dl_boost_blocked() miss a non-deadline
task that recently left SCHED_DEADLINE before its 0-lag time?
When a task switches from SCHED_DEADLINE to a non-deadline policy before
its 0-lag time, switched_from_dl() arms inactive_timer and leaves
dl_se->dl_deadline and dl_se->deadline non-zero until the timer fires. If
the task holds one rt_mutex, blocks on a second rt_mutex, and gets boosted
into the deadline class before that 0-lag time, dl_boost_blocked() sees
dl_se->dl_deadline != 0 and skips starting a period from the donor.
switched_to_dl() then cancels inactive_timer (so __dl_clear_params() will
not run while the task stays boosted) and also skips
replenish_dl_new_period() while the old dl_se->deadline has not yet passed.
Once that old deadline passes while the task remains blocked and a new
donor boosts it, switched_to_dl() does not run because the task is already
in dl_sched_class, and dl_boost_blocked() still sees dl_se->dl_deadline != 0,
requeuing the waiter on the second rt_mutex with the expired deadline ahead
of waiters with current deadlines. Should this check !dl_prio(p->normal_prio)
instead of !dl_se->dl_deadline?
[ ... ]
> @@ -4565,6 +4600,17 @@ static void switched_to_dl(struct rq *rq, struct task_struct *p)
> if (!task_on_rq_queued(p)) {
> add_rq_bw(&p->dl, &rq->dl);
>
> + /*
> + * A task that waits on an rt_mutex cannot wait for its wakeup.
> + * The chain walk sorts its waiter by its deadline first. A task
> + * new to the class has a deadline of 0, which goes before every
> + * real one. Start a new period now, as setup_new_dl_entity()
> + * does at an enqueue.
> + */
> + if (dl_task_pi_blocked(p) && !p->dl.dl_throttled &&
> + dl_time_before(p->dl.deadline, rq_clock(rq)))
> + replenish_dl_new_period(&p->dl, rq);
> +
> return;
> }
[Severity: Medium]
What happens when sched_setattr() gives a SCHED_DEADLINE reservation to a
blocked task that is already in dl_sched_class (for instance, because it
was boosted into the class by priority inheritance)?
In __sched_setscheduler(), prev_class and next_class are both
&dl_sched_class, so DEQUEUE_CLASS is not set and sched_change_end() calls
prio_changed_dl() (which returns early when !task_on_rq_queued(p)) instead
of switched_to_dl(). Meanwhile, __setscheduler_dl_pi() only acts when
!dl_policy(policy):
kernel/sched/syscalls.c:__setscheduler_dl_pi() {
if (!dl_prio(p->prio))
return;
if (dl_prio(newprio) && !dl_policy(policy)) {
struct task_struct *pi_task = rt_mutex_get_top_task(p);
...
}
If a non-deadline task holding rt_mutex L1 and blocked on rt_mutex L2 is
boosted by a deadline waiter on L1, and sched_setattr(SCHED_DEADLINE) then
gives the task its own reservation with an earlier deadline than its donor
(or if the donor's deadline on the task has already expired), the task is
neither deboosted nor given a period from its new parameters, leaving its
waiter on L2 ordered by the donor's later or expired deadline.
--
Sashiko AI review · https://sashiko.dev/#/patchset/cover.1795116790.git.mathieu.desnoyers@xxxxxxxxxxxx?part=26