2 * Deadline Scheduling Class (SCHED_DEADLINE)
4 * Earliest Deadline First (EDF) + Constant Bandwidth Server (CBS).
6 * Tasks that periodically executes their instances for less than their
7 * runtime won't miss any of their deadlines.
8 * Tasks that are not periodic or sporadic or that tries to execute more
9 * than their reserved bandwidth will be slowed down (and may potentially
10 * miss some of their deadlines), and won't affect any other task.
12 * Copyright (C) 2012 Dario Faggioli <raistlin@linux.it>,
13 * Juri Lelli <juri.lelli@gmail.com>,
14 * Michael Trimarchi <michael@amarulasolutions.com>,
15 * Fabio Checconi <fchecconi@gmail.com>
19 #include <linux/slab.h>
21 struct dl_bandwidth def_dl_bandwidth;
23 static inline struct task_struct *dl_task_of(struct sched_dl_entity *dl_se)
25 return container_of(dl_se, struct task_struct, dl);
28 static inline struct rq *rq_of_dl_rq(struct dl_rq *dl_rq)
30 return container_of(dl_rq, struct rq, dl);
33 static inline struct dl_rq *dl_rq_of_se(struct sched_dl_entity *dl_se)
35 struct task_struct *p = dl_task_of(dl_se);
36 struct rq *rq = task_rq(p);
41 static inline int on_dl_rq(struct sched_dl_entity *dl_se)
43 return !RB_EMPTY_NODE(&dl_se->rb_node);
46 static inline int is_leftmost(struct task_struct *p, struct dl_rq *dl_rq)
48 struct sched_dl_entity *dl_se = &p->dl;
50 return dl_rq->rb_leftmost == &dl_se->rb_node;
53 void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime)
55 raw_spin_lock_init(&dl_b->dl_runtime_lock);
56 dl_b->dl_period = period;
57 dl_b->dl_runtime = runtime;
60 void init_dl_bw(struct dl_bw *dl_b)
62 raw_spin_lock_init(&dl_b->lock);
63 raw_spin_lock(&def_dl_bandwidth.dl_runtime_lock);
64 if (global_rt_runtime() == RUNTIME_INF)
67 dl_b->bw = to_ratio(global_rt_period(), global_rt_runtime());
68 raw_spin_unlock(&def_dl_bandwidth.dl_runtime_lock);
72 void init_dl_rq(struct dl_rq *dl_rq)
74 dl_rq->rb_root = RB_ROOT;
77 /* zero means no -deadline tasks */
78 dl_rq->earliest_dl.curr = dl_rq->earliest_dl.next = 0;
80 dl_rq->dl_nr_migratory = 0;
81 dl_rq->overloaded = 0;
82 dl_rq->pushable_dl_tasks_root = RB_ROOT;
84 init_dl_bw(&dl_rq->dl_bw);
90 static inline int dl_overloaded(struct rq *rq)
92 return atomic_read(&rq->rd->dlo_count);
95 static inline void dl_set_overload(struct rq *rq)
100 cpumask_set_cpu(rq->cpu, rq->rd->dlo_mask);
102 * Must be visible before the overload count is
103 * set (as in sched_rt.c).
105 * Matched by the barrier in pull_dl_task().
108 atomic_inc(&rq->rd->dlo_count);
111 static inline void dl_clear_overload(struct rq *rq)
116 atomic_dec(&rq->rd->dlo_count);
117 cpumask_clear_cpu(rq->cpu, rq->rd->dlo_mask);
120 static void update_dl_migration(struct dl_rq *dl_rq)
122 if (dl_rq->dl_nr_migratory && dl_rq->dl_nr_running > 1) {
123 if (!dl_rq->overloaded) {
124 dl_set_overload(rq_of_dl_rq(dl_rq));
125 dl_rq->overloaded = 1;
127 } else if (dl_rq->overloaded) {
128 dl_clear_overload(rq_of_dl_rq(dl_rq));
129 dl_rq->overloaded = 0;
133 static void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
135 struct task_struct *p = dl_task_of(dl_se);
137 if (p->nr_cpus_allowed > 1)
138 dl_rq->dl_nr_migratory++;
140 update_dl_migration(dl_rq);
143 static void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
145 struct task_struct *p = dl_task_of(dl_se);
147 if (p->nr_cpus_allowed > 1)
148 dl_rq->dl_nr_migratory--;
150 update_dl_migration(dl_rq);
154 * The list of pushable -deadline task is not a plist, like in
155 * sched_rt.c, it is an rb-tree with tasks ordered by deadline.
157 static void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
159 struct dl_rq *dl_rq = &rq->dl;
160 struct rb_node **link = &dl_rq->pushable_dl_tasks_root.rb_node;
161 struct rb_node *parent = NULL;
162 struct task_struct *entry;
165 BUG_ON(!RB_EMPTY_NODE(&p->pushable_dl_tasks));
169 entry = rb_entry(parent, struct task_struct,
171 if (dl_entity_preempt(&p->dl, &entry->dl))
172 link = &parent->rb_left;
174 link = &parent->rb_right;
180 dl_rq->pushable_dl_tasks_leftmost = &p->pushable_dl_tasks;
182 rb_link_node(&p->pushable_dl_tasks, parent, link);
183 rb_insert_color(&p->pushable_dl_tasks, &dl_rq->pushable_dl_tasks_root);
186 static void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
188 struct dl_rq *dl_rq = &rq->dl;
190 if (RB_EMPTY_NODE(&p->pushable_dl_tasks))
193 if (dl_rq->pushable_dl_tasks_leftmost == &p->pushable_dl_tasks) {
194 struct rb_node *next_node;
196 next_node = rb_next(&p->pushable_dl_tasks);
197 dl_rq->pushable_dl_tasks_leftmost = next_node;
200 rb_erase(&p->pushable_dl_tasks, &dl_rq->pushable_dl_tasks_root);
201 RB_CLEAR_NODE(&p->pushable_dl_tasks);
204 static inline int has_pushable_dl_tasks(struct rq *rq)
206 return !RB_EMPTY_ROOT(&rq->dl.pushable_dl_tasks_root);
209 static int push_dl_task(struct rq *rq);
211 static inline bool need_pull_dl_task(struct rq *rq, struct task_struct *prev)
213 return dl_task(prev);
216 static inline void set_post_schedule(struct rq *rq)
218 rq->post_schedule = has_pushable_dl_tasks(rq);
221 static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq);
223 static void dl_task_offline_migration(struct rq *rq, struct task_struct *p)
225 struct rq *later_rq = NULL;
226 bool fallback = false;
228 later_rq = find_lock_later_rq(p, rq);
234 * If we cannot preempt any rq, fall back to pick any
238 cpu = cpumask_any_and(cpu_active_mask, tsk_cpus_allowed(p));
239 if (cpu >= nr_cpu_ids) {
241 * Fail to find any suitable cpu.
242 * The task will never come back!
244 BUG_ON(dl_bandwidth_enabled());
247 * If admission control is disabled we
248 * try a little harder to let the task
251 cpu = cpumask_any(cpu_active_mask);
253 later_rq = cpu_rq(cpu);
254 double_lock_balance(rq, later_rq);
257 deactivate_task(rq, p, 0);
258 set_task_cpu(p, later_rq->cpu);
259 activate_task(later_rq, p, ENQUEUE_REPLENISH);
262 resched_curr(later_rq);
264 double_unlock_balance(rq, later_rq);
270 void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
275 void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
280 void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
285 void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
289 static inline bool need_pull_dl_task(struct rq *rq, struct task_struct *prev)
294 static inline int pull_dl_task(struct rq *rq)
299 static inline void set_post_schedule(struct rq *rq)
302 #endif /* CONFIG_SMP */
304 static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags);
305 static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags);
306 static void check_preempt_curr_dl(struct rq *rq, struct task_struct *p,
310 * We are being explicitly informed that a new instance is starting,
311 * and this means that:
312 * - the absolute deadline of the entity has to be placed at
313 * current time + relative deadline;
314 * - the runtime of the entity has to be set to the maximum value.
316 * The capability of specifying such event is useful whenever a -deadline
317 * entity wants to (try to!) synchronize its behaviour with the scheduler's
318 * one, and to (try to!) reconcile itself with its own scheduling
321 static inline void setup_new_dl_entity(struct sched_dl_entity *dl_se,
322 struct sched_dl_entity *pi_se)
324 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
325 struct rq *rq = rq_of_dl_rq(dl_rq);
327 WARN_ON(!dl_se->dl_new || dl_se->dl_throttled);
330 * We use the regular wall clock time to set deadlines in the
331 * future; in fact, we must consider execution overheads (time
332 * spent on hardirq context, etc.).
334 dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
335 dl_se->runtime = pi_se->dl_runtime;
340 * Pure Earliest Deadline First (EDF) scheduling does not deal with the
341 * possibility of a entity lasting more than what it declared, and thus
342 * exhausting its runtime.
344 * Here we are interested in making runtime overrun possible, but we do
345 * not want a entity which is misbehaving to affect the scheduling of all
347 * Therefore, a budgeting strategy called Constant Bandwidth Server (CBS)
348 * is used, in order to confine each entity within its own bandwidth.
350 * This function deals exactly with that, and ensures that when the runtime
351 * of a entity is replenished, its deadline is also postponed. That ensures
352 * the overrunning entity can't interfere with other entity in the system and
353 * can't make them miss their deadlines. Reasons why this kind of overruns
354 * could happen are, typically, a entity voluntarily trying to overcome its
355 * runtime, or it just underestimated it during sched_setattr().
357 static void replenish_dl_entity(struct sched_dl_entity *dl_se,
358 struct sched_dl_entity *pi_se)
360 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
361 struct rq *rq = rq_of_dl_rq(dl_rq);
363 BUG_ON(pi_se->dl_runtime <= 0);
366 * This could be the case for a !-dl task that is boosted.
367 * Just go with full inherited parameters.
369 if (dl_se->dl_deadline == 0) {
370 dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
371 dl_se->runtime = pi_se->dl_runtime;
375 * We keep moving the deadline away until we get some
376 * available runtime for the entity. This ensures correct
377 * handling of situations where the runtime overrun is
380 while (dl_se->runtime <= 0) {
381 dl_se->deadline += pi_se->dl_period;
382 dl_se->runtime += pi_se->dl_runtime;
386 * At this point, the deadline really should be "in
387 * the future" with respect to rq->clock. If it's
388 * not, we are, for some reason, lagging too much!
389 * Anyway, after having warn userspace abut that,
390 * we still try to keep the things running by
391 * resetting the deadline and the budget of the
394 if (dl_time_before(dl_se->deadline, rq_clock(rq))) {
395 printk_deferred_once("sched: DL replenish lagged to much\n");
396 dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
397 dl_se->runtime = pi_se->dl_runtime;
400 if (dl_se->dl_yielded)
401 dl_se->dl_yielded = 0;
402 if (dl_se->dl_throttled)
403 dl_se->dl_throttled = 0;
407 * Here we check if --at time t-- an entity (which is probably being
408 * [re]activated or, in general, enqueued) can use its remaining runtime
409 * and its current deadline _without_ exceeding the bandwidth it is
410 * assigned (function returns true if it can't). We are in fact applying
411 * one of the CBS rules: when a task wakes up, if the residual runtime
412 * over residual deadline fits within the allocated bandwidth, then we
413 * can keep the current (absolute) deadline and residual budget without
414 * disrupting the schedulability of the system. Otherwise, we should
415 * refill the runtime and set the deadline a period in the future,
416 * because keeping the current (absolute) deadline of the task would
417 * result in breaking guarantees promised to other tasks (refer to
418 * Documentation/scheduler/sched-deadline.txt for more informations).
420 * This function returns true if:
422 * runtime / (deadline - t) > dl_runtime / dl_period ,
424 * IOW we can't recycle current parameters.
426 * Notice that the bandwidth check is done against the period. For
427 * task with deadline equal to period this is the same of using
428 * dl_deadline instead of dl_period in the equation above.
430 static bool dl_entity_overflow(struct sched_dl_entity *dl_se,
431 struct sched_dl_entity *pi_se, u64 t)
436 * left and right are the two sides of the equation above,
437 * after a bit of shuffling to use multiplications instead
440 * Note that none of the time values involved in the two
441 * multiplications are absolute: dl_deadline and dl_runtime
442 * are the relative deadline and the maximum runtime of each
443 * instance, runtime is the runtime left for the last instance
444 * and (deadline - t), since t is rq->clock, is the time left
445 * to the (absolute) deadline. Even if overflowing the u64 type
446 * is very unlikely to occur in both cases, here we scale down
447 * as we want to avoid that risk at all. Scaling down by 10
448 * means that we reduce granularity to 1us. We are fine with it,
449 * since this is only a true/false check and, anyway, thinking
450 * of anything below microseconds resolution is actually fiction
451 * (but still we want to give the user that illusion >;).
453 left = (pi_se->dl_period >> DL_SCALE) * (dl_se->runtime >> DL_SCALE);
454 right = ((dl_se->deadline - t) >> DL_SCALE) *
455 (pi_se->dl_runtime >> DL_SCALE);
457 return dl_time_before(right, left);
461 * When a -deadline entity is queued back on the runqueue, its runtime and
462 * deadline might need updating.
464 * The policy here is that we update the deadline of the entity only if:
465 * - the current deadline is in the past,
466 * - using the remaining runtime with the current deadline would make
467 * the entity exceed its bandwidth.
469 static void update_dl_entity(struct sched_dl_entity *dl_se,
470 struct sched_dl_entity *pi_se)
472 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
473 struct rq *rq = rq_of_dl_rq(dl_rq);
476 * The arrival of a new instance needs special treatment, i.e.,
477 * the actual scheduling parameters have to be "renewed".
480 setup_new_dl_entity(dl_se, pi_se);
484 if (dl_time_before(dl_se->deadline, rq_clock(rq)) ||
485 dl_entity_overflow(dl_se, pi_se, rq_clock(rq))) {
486 dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
487 dl_se->runtime = pi_se->dl_runtime;
492 * If the entity depleted all its runtime, and if we want it to sleep
493 * while waiting for some new execution time to become available, we
494 * set the bandwidth enforcement timer to the replenishment instant
495 * and try to activate it.
497 * Notice that it is important for the caller to know if the timer
498 * actually started or not (i.e., the replenishment instant is in
499 * the future or in the past).
501 static int start_dl_timer(struct sched_dl_entity *dl_se, bool boosted)
503 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
504 struct rq *rq = rq_of_dl_rq(dl_rq);
511 * We want the timer to fire at the deadline, but considering
512 * that it is actually coming from rq->clock and not from
513 * hrtimer's time base reading.
515 act = ns_to_ktime(dl_se->deadline);
516 now = hrtimer_cb_get_time(&dl_se->dl_timer);
517 delta = ktime_to_ns(now) - rq_clock(rq);
518 act = ktime_add_ns(act, delta);
521 * If the expiry time already passed, e.g., because the value
522 * chosen as the deadline is too small, don't even try to
523 * start the timer in the past!
525 if (ktime_us_delta(act, now) < 0)
528 hrtimer_start(&dl_se->dl_timer, act, HRTIMER_MODE_ABS);
534 * This is the bandwidth enforcement timer callback. If here, we know
535 * a task is not on its dl_rq, since the fact that the timer was running
536 * means the task is throttled and needs a runtime replenishment.
538 * However, what we actually do depends on the fact the task is active,
539 * (it is on its rq) or has been removed from there by a call to
540 * dequeue_task_dl(). In the former case we must issue the runtime
541 * replenishment and add the task back to the dl_rq; in the latter, we just
542 * do nothing but clearing dl_throttled, so that runtime and deadline
543 * updating (and the queueing back to dl_rq) will be done by the
544 * next call to enqueue_task_dl().
546 static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
548 struct sched_dl_entity *dl_se = container_of(timer,
549 struct sched_dl_entity,
551 struct task_struct *p = dl_task_of(dl_se);
555 rq = task_rq_lock(p, &flags);
558 * We need to take care of several possible races here:
560 * - the task might have changed its scheduling policy
561 * to something different than SCHED_DEADLINE
562 * - the task might have changed its reservation parameters
563 * (through sched_setattr())
564 * - the task might have been boosted by someone else and
565 * might be in the boosting/deboosting path
567 * In all this cases we bail out, as the task is already
568 * in the runqueue or is going to be enqueued back anyway.
570 if (!dl_task(p) || dl_se->dl_new ||
571 dl_se->dl_boosted || !dl_se->dl_throttled)
579 * If we find that the rq the task was on is no longer
580 * available, we need to select a new rq.
582 if (unlikely(!rq->online)) {
583 dl_task_offline_migration(rq, p);
589 * If the throttle happened during sched-out; like:
596 * __dequeue_task_dl()
599 * We can be both throttled and !queued. Replenish the counter
600 * but do not enqueue -- wait for our wakeup to do that.
602 if (!task_on_rq_queued(p)) {
603 replenish_dl_entity(dl_se, dl_se);
607 enqueue_task_dl(rq, p, ENQUEUE_REPLENISH);
608 if (dl_task(rq->curr))
609 check_preempt_curr_dl(rq, p, 0);
614 * Queueing this task back might have overloaded rq,
615 * check if we need to kick someone away.
617 if (has_pushable_dl_tasks(rq))
621 task_rq_unlock(rq, p, &flags);
623 return HRTIMER_NORESTART;
626 void init_dl_task_timer(struct sched_dl_entity *dl_se)
628 struct hrtimer *timer = &dl_se->dl_timer;
630 hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
631 timer->function = dl_task_timer;
635 int dl_runtime_exceeded(struct sched_dl_entity *dl_se)
637 return (dl_se->runtime <= 0);
640 extern bool sched_rt_bandwidth_account(struct rt_rq *rt_rq);
643 * Update the current task's runtime statistics (provided it is still
644 * a -deadline task and has not been removed from the dl_rq).
646 static void update_curr_dl(struct rq *rq)
648 struct task_struct *curr = rq->curr;
649 struct sched_dl_entity *dl_se = &curr->dl;
652 if (!dl_task(curr) || !on_dl_rq(dl_se))
656 * Consumed budget is computed considering the time as
657 * observed by schedulable tasks (excluding time spent
658 * in hardirq context, etc.). Deadlines are instead
659 * computed using hard walltime. This seems to be the more
660 * natural solution, but the full ramifications of this
661 * approach need further study.
663 delta_exec = rq_clock_task(rq) - curr->se.exec_start;
664 if (unlikely((s64)delta_exec <= 0))
667 schedstat_set(curr->se.statistics.exec_max,
668 max(curr->se.statistics.exec_max, delta_exec));
670 curr->se.sum_exec_runtime += delta_exec;
671 account_group_exec_runtime(curr, delta_exec);
673 curr->se.exec_start = rq_clock_task(rq);
674 cpuacct_charge(curr, delta_exec);
676 sched_rt_avg_update(rq, delta_exec);
678 dl_se->runtime -= dl_se->dl_yielded ? 0 : delta_exec;
679 if (dl_runtime_exceeded(dl_se)) {
680 dl_se->dl_throttled = 1;
681 __dequeue_task_dl(rq, curr, 0);
682 if (unlikely(!start_dl_timer(dl_se, curr->dl.dl_boosted)))
683 enqueue_task_dl(rq, curr, ENQUEUE_REPLENISH);
685 if (!is_leftmost(curr, &rq->dl))
690 * Because -- for now -- we share the rt bandwidth, we need to
691 * account our runtime there too, otherwise actual rt tasks
692 * would be able to exceed the shared quota.
694 * Account to the root rt group for now.
696 * The solution we're working towards is having the RT groups scheduled
697 * using deadline servers -- however there's a few nasties to figure
698 * out before that can happen.
700 if (rt_bandwidth_enabled()) {
701 struct rt_rq *rt_rq = &rq->rt;
703 raw_spin_lock(&rt_rq->rt_runtime_lock);
705 * We'll let actual RT tasks worry about the overflow here, we
706 * have our own CBS to keep us inline; only account when RT
707 * bandwidth is relevant.
709 if (sched_rt_bandwidth_account(rt_rq))
710 rt_rq->rt_time += delta_exec;
711 raw_spin_unlock(&rt_rq->rt_runtime_lock);
717 static struct task_struct *pick_next_earliest_dl_task(struct rq *rq, int cpu);
719 static inline u64 next_deadline(struct rq *rq)
721 struct task_struct *next = pick_next_earliest_dl_task(rq, rq->cpu);
723 if (next && dl_prio(next->prio))
724 return next->dl.deadline;
729 static void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
731 struct rq *rq = rq_of_dl_rq(dl_rq);
733 if (dl_rq->earliest_dl.curr == 0 ||
734 dl_time_before(deadline, dl_rq->earliest_dl.curr)) {
736 * If the dl_rq had no -deadline tasks, or if the new task
737 * has shorter deadline than the current one on dl_rq, we
738 * know that the previous earliest becomes our next earliest,
739 * as the new task becomes the earliest itself.
741 dl_rq->earliest_dl.next = dl_rq->earliest_dl.curr;
742 dl_rq->earliest_dl.curr = deadline;
743 cpudl_set(&rq->rd->cpudl, rq->cpu, deadline, 1);
744 } else if (dl_rq->earliest_dl.next == 0 ||
745 dl_time_before(deadline, dl_rq->earliest_dl.next)) {
747 * On the other hand, if the new -deadline task has a
748 * a later deadline than the earliest one on dl_rq, but
749 * it is earlier than the next (if any), we must
750 * recompute the next-earliest.
752 dl_rq->earliest_dl.next = next_deadline(rq);
756 static void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
758 struct rq *rq = rq_of_dl_rq(dl_rq);
761 * Since we may have removed our earliest (and/or next earliest)
762 * task we must recompute them.
764 if (!dl_rq->dl_nr_running) {
765 dl_rq->earliest_dl.curr = 0;
766 dl_rq->earliest_dl.next = 0;
767 cpudl_set(&rq->rd->cpudl, rq->cpu, 0, 0);
769 struct rb_node *leftmost = dl_rq->rb_leftmost;
770 struct sched_dl_entity *entry;
772 entry = rb_entry(leftmost, struct sched_dl_entity, rb_node);
773 dl_rq->earliest_dl.curr = entry->deadline;
774 dl_rq->earliest_dl.next = next_deadline(rq);
775 cpudl_set(&rq->rd->cpudl, rq->cpu, entry->deadline, 1);
781 static inline void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
782 static inline void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
784 #endif /* CONFIG_SMP */
787 void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
789 int prio = dl_task_of(dl_se)->prio;
790 u64 deadline = dl_se->deadline;
792 WARN_ON(!dl_prio(prio));
793 dl_rq->dl_nr_running++;
794 add_nr_running(rq_of_dl_rq(dl_rq), 1);
796 inc_dl_deadline(dl_rq, deadline);
797 inc_dl_migration(dl_se, dl_rq);
801 void dec_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
803 int prio = dl_task_of(dl_se)->prio;
805 WARN_ON(!dl_prio(prio));
806 WARN_ON(!dl_rq->dl_nr_running);
807 dl_rq->dl_nr_running--;
808 sub_nr_running(rq_of_dl_rq(dl_rq), 1);
810 dec_dl_deadline(dl_rq, dl_se->deadline);
811 dec_dl_migration(dl_se, dl_rq);
814 static void __enqueue_dl_entity(struct sched_dl_entity *dl_se)
816 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
817 struct rb_node **link = &dl_rq->rb_root.rb_node;
818 struct rb_node *parent = NULL;
819 struct sched_dl_entity *entry;
822 BUG_ON(!RB_EMPTY_NODE(&dl_se->rb_node));
826 entry = rb_entry(parent, struct sched_dl_entity, rb_node);
827 if (dl_time_before(dl_se->deadline, entry->deadline))
828 link = &parent->rb_left;
830 link = &parent->rb_right;
836 dl_rq->rb_leftmost = &dl_se->rb_node;
838 rb_link_node(&dl_se->rb_node, parent, link);
839 rb_insert_color(&dl_se->rb_node, &dl_rq->rb_root);
841 inc_dl_tasks(dl_se, dl_rq);
844 static void __dequeue_dl_entity(struct sched_dl_entity *dl_se)
846 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
848 if (RB_EMPTY_NODE(&dl_se->rb_node))
851 if (dl_rq->rb_leftmost == &dl_se->rb_node) {
852 struct rb_node *next_node;
854 next_node = rb_next(&dl_se->rb_node);
855 dl_rq->rb_leftmost = next_node;
858 rb_erase(&dl_se->rb_node, &dl_rq->rb_root);
859 RB_CLEAR_NODE(&dl_se->rb_node);
861 dec_dl_tasks(dl_se, dl_rq);
865 enqueue_dl_entity(struct sched_dl_entity *dl_se,
866 struct sched_dl_entity *pi_se, int flags)
868 BUG_ON(on_dl_rq(dl_se));
871 * If this is a wakeup or a new instance, the scheduling
872 * parameters of the task might need updating. Otherwise,
873 * we want a replenishment of its runtime.
875 if (dl_se->dl_new || flags & ENQUEUE_WAKEUP)
876 update_dl_entity(dl_se, pi_se);
877 else if (flags & ENQUEUE_REPLENISH)
878 replenish_dl_entity(dl_se, pi_se);
880 __enqueue_dl_entity(dl_se);
883 static void dequeue_dl_entity(struct sched_dl_entity *dl_se)
885 __dequeue_dl_entity(dl_se);
888 static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags)
890 struct task_struct *pi_task = rt_mutex_get_top_task(p);
891 struct sched_dl_entity *pi_se = &p->dl;
894 * Use the scheduling parameters of the top pi-waiter
895 * task if we have one and its (relative) deadline is
896 * smaller than our one... OTW we keep our runtime and
899 if (pi_task && p->dl.dl_boosted && dl_prio(pi_task->normal_prio)) {
900 pi_se = &pi_task->dl;
901 } else if (!dl_prio(p->normal_prio)) {
903 * Special case in which we have a !SCHED_DEADLINE task
904 * that is going to be deboosted, but exceedes its
905 * runtime while doing so. No point in replenishing
906 * it, as it's going to return back to its original
907 * scheduling class after this.
909 BUG_ON(!p->dl.dl_boosted || flags != ENQUEUE_REPLENISH);
914 * If p is throttled, we do nothing. In fact, if it exhausted
915 * its budget it needs a replenishment and, since it now is on
916 * its rq, the bandwidth timer callback (which clearly has not
917 * run yet) will take care of this.
919 if (p->dl.dl_throttled && !(flags & ENQUEUE_REPLENISH))
922 enqueue_dl_entity(&p->dl, pi_se, flags);
924 if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
925 enqueue_pushable_dl_task(rq, p);
928 static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
930 dequeue_dl_entity(&p->dl);
931 dequeue_pushable_dl_task(rq, p);
934 static void dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
937 __dequeue_task_dl(rq, p, flags);
941 * Yield task semantic for -deadline tasks is:
943 * get off from the CPU until our next instance, with
944 * a new runtime. This is of little use now, since we
945 * don't have a bandwidth reclaiming mechanism. Anyway,
946 * bandwidth reclaiming is planned for the future, and
947 * yield_task_dl will indicate that some spare budget
948 * is available for other task instances to use it.
950 static void yield_task_dl(struct rq *rq)
952 struct task_struct *p = rq->curr;
955 * We make the task go to sleep until its current deadline by
956 * forcing its runtime to zero. This way, update_curr_dl() stops
957 * it and the bandwidth timer will wake it up and will give it
958 * new scheduling parameters (thanks to dl_yielded=1).
960 if (p->dl.runtime > 0) {
961 rq->curr->dl.dl_yielded = 1;
967 * Tell update_rq_clock() that we've just updated,
968 * so we don't do microscopic update in schedule()
969 * and double the fastpath cost.
971 rq_clock_skip_update(rq, true);
976 static int find_later_rq(struct task_struct *task);
979 select_task_rq_dl(struct task_struct *p, int cpu, int sd_flag, int flags)
981 struct task_struct *curr;
984 if (sd_flag != SD_BALANCE_WAKE)
990 curr = READ_ONCE(rq->curr); /* unlocked access */
993 * If we are dealing with a -deadline task, we must
994 * decide where to wake it up.
995 * If it has a later deadline and the current task
996 * on this rq can't move (provided the waking task
997 * can!) we prefer to send it somewhere else. On the
998 * other hand, if it has a shorter deadline, we
999 * try to make it stay here, it might be important.
1001 if (unlikely(dl_task(curr)) &&
1002 (curr->nr_cpus_allowed < 2 ||
1003 !dl_entity_preempt(&p->dl, &curr->dl)) &&
1004 (p->nr_cpus_allowed > 1)) {
1005 int target = find_later_rq(p);
1008 dl_time_before(p->dl.deadline,
1009 cpu_rq(target)->dl.earliest_dl.curr))
1018 static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
1021 * Current can't be migrated, useless to reschedule,
1022 * let's hope p can move out.
1024 if (rq->curr->nr_cpus_allowed == 1 ||
1025 cpudl_find(&rq->rd->cpudl, rq->curr, NULL) == -1)
1029 * p is migratable, so let's not schedule it and
1030 * see if it is pushed or pulled somewhere else.
1032 if (p->nr_cpus_allowed != 1 &&
1033 cpudl_find(&rq->rd->cpudl, p, NULL) != -1)
1039 static int pull_dl_task(struct rq *this_rq);
1041 #endif /* CONFIG_SMP */
1044 * Only called when both the current and waking task are -deadline
1047 static void check_preempt_curr_dl(struct rq *rq, struct task_struct *p,
1050 if (dl_entity_preempt(&p->dl, &rq->curr->dl)) {
1057 * In the unlikely case current and p have the same deadline
1058 * let us try to decide what's the best thing to do...
1060 if ((p->dl.deadline == rq->curr->dl.deadline) &&
1061 !test_tsk_need_resched(rq->curr))
1062 check_preempt_equal_dl(rq, p);
1063 #endif /* CONFIG_SMP */
1066 #ifdef CONFIG_SCHED_HRTICK
1067 static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1069 hrtick_start(rq, p->dl.runtime);
1071 #else /* !CONFIG_SCHED_HRTICK */
1072 static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1077 static struct sched_dl_entity *pick_next_dl_entity(struct rq *rq,
1078 struct dl_rq *dl_rq)
1080 struct rb_node *left = dl_rq->rb_leftmost;
1085 return rb_entry(left, struct sched_dl_entity, rb_node);
1088 struct task_struct *pick_next_task_dl(struct rq *rq, struct task_struct *prev)
1090 struct sched_dl_entity *dl_se;
1091 struct task_struct *p;
1092 struct dl_rq *dl_rq;
1096 if (need_pull_dl_task(rq, prev)) {
1099 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1100 * means a stop task can slip in, in which case we need to
1101 * re-start task selection.
1103 if (rq->stop && task_on_rq_queued(rq->stop))
1108 * When prev is DL, we may throttle it in put_prev_task().
1109 * So, we update time before we check for dl_nr_running.
1111 if (prev->sched_class == &dl_sched_class)
1114 if (unlikely(!dl_rq->dl_nr_running))
1117 put_prev_task(rq, prev);
1119 dl_se = pick_next_dl_entity(rq, dl_rq);
1122 p = dl_task_of(dl_se);
1123 p->se.exec_start = rq_clock_task(rq);
1125 /* Running task will never be pushed. */
1126 dequeue_pushable_dl_task(rq, p);
1128 if (hrtick_enabled(rq))
1129 start_hrtick_dl(rq, p);
1131 set_post_schedule(rq);
1136 static void put_prev_task_dl(struct rq *rq, struct task_struct *p)
1140 if (on_dl_rq(&p->dl) && p->nr_cpus_allowed > 1)
1141 enqueue_pushable_dl_task(rq, p);
1144 static void task_tick_dl(struct rq *rq, struct task_struct *p, int queued)
1149 * Even when we have runtime, update_curr_dl() might have resulted in us
1150 * not being the leftmost task anymore. In that case NEED_RESCHED will
1151 * be set and schedule() will start a new hrtick for the next task.
1153 if (hrtick_enabled(rq) && queued && p->dl.runtime > 0 &&
1154 is_leftmost(p, &rq->dl))
1155 start_hrtick_dl(rq, p);
1158 static void task_fork_dl(struct task_struct *p)
1161 * SCHED_DEADLINE tasks cannot fork and this is achieved through
1166 static void task_dead_dl(struct task_struct *p)
1168 struct hrtimer *timer = &p->dl.dl_timer;
1169 struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
1172 * Since we are TASK_DEAD we won't slip out of the domain!
1174 raw_spin_lock_irq(&dl_b->lock);
1175 /* XXX we should retain the bw until 0-lag */
1176 dl_b->total_bw -= p->dl.dl_bw;
1177 raw_spin_unlock_irq(&dl_b->lock);
1179 hrtimer_cancel(timer);
1182 static void set_curr_task_dl(struct rq *rq)
1184 struct task_struct *p = rq->curr;
1186 p->se.exec_start = rq_clock_task(rq);
1188 /* You can't push away the running task */
1189 dequeue_pushable_dl_task(rq, p);
1194 /* Only try algorithms three times */
1195 #define DL_MAX_TRIES 3
1197 static int pick_dl_task(struct rq *rq, struct task_struct *p, int cpu)
1199 if (!task_running(rq, p) &&
1200 cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
1205 /* Returns the second earliest -deadline task, NULL otherwise */
1206 static struct task_struct *pick_next_earliest_dl_task(struct rq *rq, int cpu)
1208 struct rb_node *next_node = rq->dl.rb_leftmost;
1209 struct sched_dl_entity *dl_se;
1210 struct task_struct *p = NULL;
1213 next_node = rb_next(next_node);
1215 dl_se = rb_entry(next_node, struct sched_dl_entity, rb_node);
1216 p = dl_task_of(dl_se);
1218 if (pick_dl_task(rq, p, cpu))
1228 * Return the earliest pushable rq's task, which is suitable to be executed
1229 * on the CPU, NULL otherwise:
1231 static struct task_struct *pick_earliest_pushable_dl_task(struct rq *rq, int cpu)
1233 struct rb_node *next_node = rq->dl.pushable_dl_tasks_leftmost;
1234 struct task_struct *p = NULL;
1236 if (!has_pushable_dl_tasks(rq))
1241 p = rb_entry(next_node, struct task_struct, pushable_dl_tasks);
1243 if (pick_dl_task(rq, p, cpu))
1246 next_node = rb_next(next_node);
1253 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask_dl);
1255 static int find_later_rq(struct task_struct *task)
1257 struct sched_domain *sd;
1258 struct cpumask *later_mask = this_cpu_cpumask_var_ptr(local_cpu_mask_dl);
1259 int this_cpu = smp_processor_id();
1260 int best_cpu, cpu = task_cpu(task);
1262 /* Make sure the mask is initialized first */
1263 if (unlikely(!later_mask))
1266 if (task->nr_cpus_allowed == 1)
1270 * We have to consider system topology and task affinity
1271 * first, then we can look for a suitable cpu.
1273 best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
1279 * If we are here, some target has been found,
1280 * the most suitable of which is cached in best_cpu.
1281 * This is, among the runqueues where the current tasks
1282 * have later deadlines than the task's one, the rq
1283 * with the latest possible one.
1285 * Now we check how well this matches with task's
1286 * affinity and system topology.
1288 * The last cpu where the task run is our first
1289 * guess, since it is most likely cache-hot there.
1291 if (cpumask_test_cpu(cpu, later_mask))
1294 * Check if this_cpu is to be skipped (i.e., it is
1295 * not in the mask) or not.
1297 if (!cpumask_test_cpu(this_cpu, later_mask))
1301 for_each_domain(cpu, sd) {
1302 if (sd->flags & SD_WAKE_AFFINE) {
1305 * If possible, preempting this_cpu is
1306 * cheaper than migrating.
1308 if (this_cpu != -1 &&
1309 cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
1315 * Last chance: if best_cpu is valid and is
1316 * in the mask, that becomes our choice.
1318 if (best_cpu < nr_cpu_ids &&
1319 cpumask_test_cpu(best_cpu, sched_domain_span(sd))) {
1328 * At this point, all our guesses failed, we just return
1329 * 'something', and let the caller sort the things out.
1334 cpu = cpumask_any(later_mask);
1335 if (cpu < nr_cpu_ids)
1341 /* Locks the rq it finds */
1342 static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq)
1344 struct rq *later_rq = NULL;
1348 for (tries = 0; tries < DL_MAX_TRIES; tries++) {
1349 cpu = find_later_rq(task);
1351 if ((cpu == -1) || (cpu == rq->cpu))
1354 later_rq = cpu_rq(cpu);
1356 if (!dl_time_before(task->dl.deadline,
1357 later_rq->dl.earliest_dl.curr)) {
1359 * Target rq has tasks of equal or earlier deadline,
1360 * retrying does not release any lock and is unlikely
1361 * to yield a different result.
1367 /* Retry if something changed. */
1368 if (double_lock_balance(rq, later_rq)) {
1369 if (unlikely(task_rq(task) != rq ||
1370 !cpumask_test_cpu(later_rq->cpu,
1371 &task->cpus_allowed) ||
1372 task_running(rq, task) ||
1373 !task_on_rq_queued(task))) {
1374 double_unlock_balance(rq, later_rq);
1381 * If the rq we found has no -deadline task, or
1382 * its earliest one has a later deadline than our
1383 * task, the rq is a good one.
1385 if (!later_rq->dl.dl_nr_running ||
1386 dl_time_before(task->dl.deadline,
1387 later_rq->dl.earliest_dl.curr))
1390 /* Otherwise we try again. */
1391 double_unlock_balance(rq, later_rq);
1398 static struct task_struct *pick_next_pushable_dl_task(struct rq *rq)
1400 struct task_struct *p;
1402 if (!has_pushable_dl_tasks(rq))
1405 p = rb_entry(rq->dl.pushable_dl_tasks_leftmost,
1406 struct task_struct, pushable_dl_tasks);
1408 BUG_ON(rq->cpu != task_cpu(p));
1409 BUG_ON(task_current(rq, p));
1410 BUG_ON(p->nr_cpus_allowed <= 1);
1412 BUG_ON(!task_on_rq_queued(p));
1413 BUG_ON(!dl_task(p));
1419 * See if the non running -deadline tasks on this rq
1420 * can be sent to some other CPU where they can preempt
1421 * and start executing.
1423 static int push_dl_task(struct rq *rq)
1425 struct task_struct *next_task;
1426 struct rq *later_rq;
1429 if (!rq->dl.overloaded)
1432 next_task = pick_next_pushable_dl_task(rq);
1437 if (unlikely(next_task == rq->curr)) {
1443 * If next_task preempts rq->curr, and rq->curr
1444 * can move away, it makes sense to just reschedule
1445 * without going further in pushing next_task.
1447 if (dl_task(rq->curr) &&
1448 dl_time_before(next_task->dl.deadline, rq->curr->dl.deadline) &&
1449 rq->curr->nr_cpus_allowed > 1) {
1454 /* We might release rq lock */
1455 get_task_struct(next_task);
1457 /* Will lock the rq it'll find */
1458 later_rq = find_lock_later_rq(next_task, rq);
1460 struct task_struct *task;
1463 * We must check all this again, since
1464 * find_lock_later_rq releases rq->lock and it is
1465 * then possible that next_task has migrated.
1467 task = pick_next_pushable_dl_task(rq);
1468 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1470 * The task is still there. We don't try
1471 * again, some other cpu will pull it when ready.
1480 put_task_struct(next_task);
1485 deactivate_task(rq, next_task, 0);
1486 set_task_cpu(next_task, later_rq->cpu);
1487 activate_task(later_rq, next_task, 0);
1490 resched_curr(later_rq);
1492 double_unlock_balance(rq, later_rq);
1495 put_task_struct(next_task);
1500 static void push_dl_tasks(struct rq *rq)
1502 /* Terminates as it moves a -deadline task */
1503 while (push_dl_task(rq))
1507 static int pull_dl_task(struct rq *this_rq)
1509 int this_cpu = this_rq->cpu, ret = 0, cpu;
1510 struct task_struct *p;
1512 u64 dmin = LONG_MAX;
1514 if (likely(!dl_overloaded(this_rq)))
1518 * Match the barrier from dl_set_overloaded; this guarantees that if we
1519 * see overloaded we must also see the dlo_mask bit.
1523 for_each_cpu(cpu, this_rq->rd->dlo_mask) {
1524 if (this_cpu == cpu)
1527 src_rq = cpu_rq(cpu);
1530 * It looks racy, abd it is! However, as in sched_rt.c,
1531 * we are fine with this.
1533 if (this_rq->dl.dl_nr_running &&
1534 dl_time_before(this_rq->dl.earliest_dl.curr,
1535 src_rq->dl.earliest_dl.next))
1538 /* Might drop this_rq->lock */
1539 double_lock_balance(this_rq, src_rq);
1542 * If there are no more pullable tasks on the
1543 * rq, we're done with it.
1545 if (src_rq->dl.dl_nr_running <= 1)
1548 p = pick_earliest_pushable_dl_task(src_rq, this_cpu);
1551 * We found a task to be pulled if:
1552 * - it preempts our current (if there's one),
1553 * - it will preempt the last one we pulled (if any).
1555 if (p && dl_time_before(p->dl.deadline, dmin) &&
1556 (!this_rq->dl.dl_nr_running ||
1557 dl_time_before(p->dl.deadline,
1558 this_rq->dl.earliest_dl.curr))) {
1559 WARN_ON(p == src_rq->curr);
1560 WARN_ON(!task_on_rq_queued(p));
1563 * Then we pull iff p has actually an earlier
1564 * deadline than the current task of its runqueue.
1566 if (dl_time_before(p->dl.deadline,
1567 src_rq->curr->dl.deadline))
1572 deactivate_task(src_rq, p, 0);
1573 set_task_cpu(p, this_cpu);
1574 activate_task(this_rq, p, 0);
1575 dmin = p->dl.deadline;
1577 /* Is there any other task even earlier? */
1580 double_unlock_balance(this_rq, src_rq);
1586 static void post_schedule_dl(struct rq *rq)
1592 * Since the task is not running and a reschedule is not going to happen
1593 * anytime soon on its runqueue, we try pushing it away now.
1595 static void task_woken_dl(struct rq *rq, struct task_struct *p)
1597 if (!task_running(rq, p) &&
1598 !test_tsk_need_resched(rq->curr) &&
1599 has_pushable_dl_tasks(rq) &&
1600 p->nr_cpus_allowed > 1 &&
1601 dl_task(rq->curr) &&
1602 (rq->curr->nr_cpus_allowed < 2 ||
1603 !dl_entity_preempt(&p->dl, &rq->curr->dl))) {
1608 static void set_cpus_allowed_dl(struct task_struct *p,
1609 const struct cpumask *new_mask)
1612 struct root_domain *src_rd;
1615 BUG_ON(!dl_task(p));
1620 * Migrating a SCHED_DEADLINE task between exclusive
1621 * cpusets (different root_domains) entails a bandwidth
1622 * update. We already made space for us in the destination
1623 * domain (see cpuset_can_attach()).
1625 if (!cpumask_intersects(src_rd->span, new_mask)) {
1626 struct dl_bw *src_dl_b;
1628 src_dl_b = dl_bw_of(cpu_of(rq));
1630 * We now free resources of the root_domain we are migrating
1631 * off. In the worst case, sched_setattr() may temporary fail
1632 * until we complete the update.
1634 raw_spin_lock(&src_dl_b->lock);
1635 __dl_clear(src_dl_b, p->dl.dl_bw);
1636 raw_spin_unlock(&src_dl_b->lock);
1640 * Update only if the task is actually running (i.e.,
1641 * it is on the rq AND it is not throttled).
1643 if (!on_dl_rq(&p->dl))
1646 weight = cpumask_weight(new_mask);
1649 * Only update if the process changes its state from whether it
1650 * can migrate or not.
1652 if ((p->nr_cpus_allowed > 1) == (weight > 1))
1656 * The process used to be able to migrate OR it can now migrate
1659 if (!task_current(rq, p))
1660 dequeue_pushable_dl_task(rq, p);
1661 BUG_ON(!rq->dl.dl_nr_migratory);
1662 rq->dl.dl_nr_migratory--;
1664 if (!task_current(rq, p))
1665 enqueue_pushable_dl_task(rq, p);
1666 rq->dl.dl_nr_migratory++;
1669 update_dl_migration(&rq->dl);
1672 /* Assumes rq->lock is held */
1673 static void rq_online_dl(struct rq *rq)
1675 if (rq->dl.overloaded)
1676 dl_set_overload(rq);
1678 cpudl_set_freecpu(&rq->rd->cpudl, rq->cpu);
1679 if (rq->dl.dl_nr_running > 0)
1680 cpudl_set(&rq->rd->cpudl, rq->cpu, rq->dl.earliest_dl.curr, 1);
1683 /* Assumes rq->lock is held */
1684 static void rq_offline_dl(struct rq *rq)
1686 if (rq->dl.overloaded)
1687 dl_clear_overload(rq);
1689 cpudl_set(&rq->rd->cpudl, rq->cpu, 0, 0);
1690 cpudl_clear_freecpu(&rq->rd->cpudl, rq->cpu);
1693 void __init init_sched_dl_class(void)
1697 for_each_possible_cpu(i)
1698 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask_dl, i),
1699 GFP_KERNEL, cpu_to_node(i));
1702 #endif /* CONFIG_SMP */
1705 * Ensure p's dl_timer is cancelled. May drop rq->lock for a while.
1707 static void cancel_dl_timer(struct rq *rq, struct task_struct *p)
1709 struct hrtimer *dl_timer = &p->dl.dl_timer;
1711 /* Nobody will change task's class if pi_lock is held */
1712 lockdep_assert_held(&p->pi_lock);
1714 if (hrtimer_active(dl_timer)) {
1715 int ret = hrtimer_try_to_cancel(dl_timer);
1717 if (unlikely(ret == -1)) {
1719 * Note, p may migrate OR new deadline tasks
1720 * may appear in rq when we are unlocking it.
1721 * A caller of us must be fine with that.
1723 raw_spin_unlock(&rq->lock);
1724 hrtimer_cancel(dl_timer);
1725 raw_spin_lock(&rq->lock);
1730 static void switched_from_dl(struct rq *rq, struct task_struct *p)
1732 /* XXX we should retain the bw until 0-lag */
1733 cancel_dl_timer(rq, p);
1734 __dl_clear_params(p);
1737 * Since this might be the only -deadline task on the rq,
1738 * this is the right place to try to pull some other one
1739 * from an overloaded cpu, if any.
1741 if (!task_on_rq_queued(p) || rq->dl.dl_nr_running)
1744 if (pull_dl_task(rq))
1749 * When switching to -deadline, we may overload the rq, then
1750 * we try to push someone off, if possible.
1752 static void switched_to_dl(struct rq *rq, struct task_struct *p)
1754 int check_resched = 1;
1756 if (task_on_rq_queued(p) && rq->curr != p) {
1758 if (p->nr_cpus_allowed > 1 && rq->dl.overloaded &&
1759 push_dl_task(rq) && rq != task_rq(p))
1760 /* Only reschedule if pushing failed */
1762 #endif /* CONFIG_SMP */
1763 if (check_resched) {
1764 if (dl_task(rq->curr))
1765 check_preempt_curr_dl(rq, p, 0);
1773 * If the scheduling parameters of a -deadline task changed,
1774 * a push or pull operation might be needed.
1776 static void prio_changed_dl(struct rq *rq, struct task_struct *p,
1779 if (task_on_rq_queued(p) || rq->curr == p) {
1782 * This might be too much, but unfortunately
1783 * we don't have the old deadline value, and
1784 * we can't argue if the task is increasing
1785 * or lowering its prio, so...
1787 if (!rq->dl.overloaded)
1791 * If we now have a earlier deadline task than p,
1792 * then reschedule, provided p is still on this
1795 if (dl_time_before(rq->dl.earliest_dl.curr, p->dl.deadline) &&
1800 * Again, we don't know if p has a earlier
1801 * or later deadline, so let's blindly set a
1802 * (maybe not needed) rescheduling point.
1805 #endif /* CONFIG_SMP */
1807 switched_to_dl(rq, p);
1810 const struct sched_class dl_sched_class = {
1811 .next = &rt_sched_class,
1812 .enqueue_task = enqueue_task_dl,
1813 .dequeue_task = dequeue_task_dl,
1814 .yield_task = yield_task_dl,
1816 .check_preempt_curr = check_preempt_curr_dl,
1818 .pick_next_task = pick_next_task_dl,
1819 .put_prev_task = put_prev_task_dl,
1822 .select_task_rq = select_task_rq_dl,
1823 .set_cpus_allowed = set_cpus_allowed_dl,
1824 .rq_online = rq_online_dl,
1825 .rq_offline = rq_offline_dl,
1826 .post_schedule = post_schedule_dl,
1827 .task_woken = task_woken_dl,
1830 .set_curr_task = set_curr_task_dl,
1831 .task_tick = task_tick_dl,
1832 .task_fork = task_fork_dl,
1833 .task_dead = task_dead_dl,
1835 .prio_changed = prio_changed_dl,
1836 .switched_from = switched_from_dl,
1837 .switched_to = switched_to_dl,
1839 .update_curr = update_curr_dl,
1842 #ifdef CONFIG_SCHED_DEBUG
1843 extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq);
1845 void print_dl_stats(struct seq_file *m, int cpu)
1847 print_dl_rq(m, cpu, &cpu_rq(cpu)->dl);
1849 #endif /* CONFIG_SCHED_DEBUG */