1 // SPDX-License-Identifier: GPL-2.0
3 * Deadline Scheduling Class (SCHED_DEADLINE)
5 * Earliest Deadline First (EDF) + Constant Bandwidth Server (CBS).
7 * Tasks that periodically executes their instances for less than their
8 * runtime won't miss any of their deadlines.
9 * Tasks that are not periodic or sporadic or that tries to execute more
10 * than their reserved bandwidth will be slowed down (and may potentially
11 * miss some of their deadlines), and won't affect any other task.
13 * Copyright (C) 2012 Dario Faggioli <raistlin@linux.it>,
14 * Juri Lelli <juri.lelli@gmail.com>,
15 * Michael Trimarchi <michael@amarulasolutions.com>,
16 * Fabio Checconi <fchecconi@gmail.com>
19 #include <linux/cpuset.h>
22 * Default limits for DL period; on the top end we guard against small util
23 * tasks still getting ridiculously long effective runtimes, on the bottom end we
24 * guard against timer DoS.
26 static unsigned int sysctl_sched_dl_period_max = 1 << 22; /* ~4 seconds */
27 static unsigned int sysctl_sched_dl_period_min = 100; /* 100 us */
29 static struct ctl_table sched_dl_sysctls[] = {
31 .procname = "sched_deadline_period_max_us",
32 .data = &sysctl_sched_dl_period_max,
33 .maxlen = sizeof(unsigned int),
35 .proc_handler = proc_douintvec_minmax,
36 .extra1 = (void *)&sysctl_sched_dl_period_min,
39 .procname = "sched_deadline_period_min_us",
40 .data = &sysctl_sched_dl_period_min,
41 .maxlen = sizeof(unsigned int),
43 .proc_handler = proc_douintvec_minmax,
44 .extra2 = (void *)&sysctl_sched_dl_period_max,
49 static int __init sched_dl_sysctl_init(void)
51 register_sysctl_init("kernel", sched_dl_sysctls);
54 late_initcall(sched_dl_sysctl_init);
57 static inline struct task_struct *dl_task_of(struct sched_dl_entity *dl_se)
59 return container_of(dl_se, struct task_struct, dl);
62 static inline struct rq *rq_of_dl_rq(struct dl_rq *dl_rq)
64 return container_of(dl_rq, struct rq, dl);
67 static inline struct dl_rq *dl_rq_of_se(struct sched_dl_entity *dl_se)
69 struct task_struct *p = dl_task_of(dl_se);
70 struct rq *rq = task_rq(p);
75 static inline int on_dl_rq(struct sched_dl_entity *dl_se)
77 return !RB_EMPTY_NODE(&dl_se->rb_node);
80 #ifdef CONFIG_RT_MUTEXES
81 static inline struct sched_dl_entity *pi_of(struct sched_dl_entity *dl_se)
86 static inline bool is_dl_boosted(struct sched_dl_entity *dl_se)
88 return pi_of(dl_se) != dl_se;
91 static inline struct sched_dl_entity *pi_of(struct sched_dl_entity *dl_se)
96 static inline bool is_dl_boosted(struct sched_dl_entity *dl_se)
103 static inline struct dl_bw *dl_bw_of(int i)
105 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
106 "sched RCU must be held");
107 return &cpu_rq(i)->rd->dl_bw;
110 static inline int dl_bw_cpus(int i)
112 struct root_domain *rd = cpu_rq(i)->rd;
115 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
116 "sched RCU must be held");
118 if (cpumask_subset(rd->span, cpu_active_mask))
119 return cpumask_weight(rd->span);
123 for_each_cpu_and(i, rd->span, cpu_active_mask)
129 static inline unsigned long __dl_bw_capacity(const struct cpumask *mask)
131 unsigned long cap = 0;
134 for_each_cpu_and(i, mask, cpu_active_mask)
135 cap += arch_scale_cpu_capacity(i);
141 * XXX Fix: If 'rq->rd == def_root_domain' perform AC against capacity
142 * of the CPU the task is running on rather rd's \Sum CPU capacity.
144 static inline unsigned long dl_bw_capacity(int i)
146 if (!sched_asym_cpucap_active() &&
147 arch_scale_cpu_capacity(i) == SCHED_CAPACITY_SCALE) {
148 return dl_bw_cpus(i) << SCHED_CAPACITY_SHIFT;
150 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
151 "sched RCU must be held");
153 return __dl_bw_capacity(cpu_rq(i)->rd->span);
157 static inline bool dl_bw_visited(int cpu, u64 gen)
159 struct root_domain *rd = cpu_rq(cpu)->rd;
161 if (rd->visit_gen == gen)
169 void __dl_update(struct dl_bw *dl_b, s64 bw)
171 struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw);
174 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
175 "sched RCU must be held");
176 for_each_cpu_and(i, rd->span, cpu_active_mask) {
177 struct rq *rq = cpu_rq(i);
179 rq->dl.extra_bw += bw;
183 static inline struct dl_bw *dl_bw_of(int i)
185 return &cpu_rq(i)->dl.dl_bw;
188 static inline int dl_bw_cpus(int i)
193 static inline unsigned long dl_bw_capacity(int i)
195 return SCHED_CAPACITY_SCALE;
198 static inline bool dl_bw_visited(int cpu, u64 gen)
204 void __dl_update(struct dl_bw *dl_b, s64 bw)
206 struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw);
213 void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
215 dl_b->total_bw -= tsk_bw;
216 __dl_update(dl_b, (s32)tsk_bw / cpus);
220 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus)
222 dl_b->total_bw += tsk_bw;
223 __dl_update(dl_b, -((s32)tsk_bw / cpus));
227 __dl_overflow(struct dl_bw *dl_b, unsigned long cap, u64 old_bw, u64 new_bw)
229 return dl_b->bw != -1 &&
230 cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw;
234 void __add_running_bw(u64 dl_bw, struct dl_rq *dl_rq)
236 u64 old = dl_rq->running_bw;
238 lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
239 dl_rq->running_bw += dl_bw;
240 SCHED_WARN_ON(dl_rq->running_bw < old); /* overflow */
241 SCHED_WARN_ON(dl_rq->running_bw > dl_rq->this_bw);
242 /* kick cpufreq (see the comment in kernel/sched/sched.h). */
243 cpufreq_update_util(rq_of_dl_rq(dl_rq), 0);
247 void __sub_running_bw(u64 dl_bw, struct dl_rq *dl_rq)
249 u64 old = dl_rq->running_bw;
251 lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
252 dl_rq->running_bw -= dl_bw;
253 SCHED_WARN_ON(dl_rq->running_bw > old); /* underflow */
254 if (dl_rq->running_bw > old)
255 dl_rq->running_bw = 0;
256 /* kick cpufreq (see the comment in kernel/sched/sched.h). */
257 cpufreq_update_util(rq_of_dl_rq(dl_rq), 0);
261 void __add_rq_bw(u64 dl_bw, struct dl_rq *dl_rq)
263 u64 old = dl_rq->this_bw;
265 lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
266 dl_rq->this_bw += dl_bw;
267 SCHED_WARN_ON(dl_rq->this_bw < old); /* overflow */
271 void __sub_rq_bw(u64 dl_bw, struct dl_rq *dl_rq)
273 u64 old = dl_rq->this_bw;
275 lockdep_assert_rq_held(rq_of_dl_rq(dl_rq));
276 dl_rq->this_bw -= dl_bw;
277 SCHED_WARN_ON(dl_rq->this_bw > old); /* underflow */
278 if (dl_rq->this_bw > old)
280 SCHED_WARN_ON(dl_rq->running_bw > dl_rq->this_bw);
284 void add_rq_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
286 if (!dl_entity_is_special(dl_se))
287 __add_rq_bw(dl_se->dl_bw, dl_rq);
291 void sub_rq_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
293 if (!dl_entity_is_special(dl_se))
294 __sub_rq_bw(dl_se->dl_bw, dl_rq);
298 void add_running_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
300 if (!dl_entity_is_special(dl_se))
301 __add_running_bw(dl_se->dl_bw, dl_rq);
305 void sub_running_bw(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
307 if (!dl_entity_is_special(dl_se))
308 __sub_running_bw(dl_se->dl_bw, dl_rq);
311 static void dl_change_utilization(struct task_struct *p, u64 new_bw)
315 WARN_ON_ONCE(p->dl.flags & SCHED_FLAG_SUGOV);
317 if (task_on_rq_queued(p))
321 if (p->dl.dl_non_contending) {
322 sub_running_bw(&p->dl, &rq->dl);
323 p->dl.dl_non_contending = 0;
325 * If the timer handler is currently running and the
326 * timer cannot be canceled, inactive_task_timer()
327 * will see that dl_not_contending is not set, and
328 * will not touch the rq's active utilization,
329 * so we are still safe.
331 if (hrtimer_try_to_cancel(&p->dl.inactive_timer) == 1)
334 __sub_rq_bw(p->dl.dl_bw, &rq->dl);
335 __add_rq_bw(new_bw, &rq->dl);
339 * The utilization of a task cannot be immediately removed from
340 * the rq active utilization (running_bw) when the task blocks.
341 * Instead, we have to wait for the so called "0-lag time".
343 * If a task blocks before the "0-lag time", a timer (the inactive
344 * timer) is armed, and running_bw is decreased when the timer
347 * If the task wakes up again before the inactive timer fires,
348 * the timer is canceled, whereas if the task wakes up after the
349 * inactive timer fired (and running_bw has been decreased) the
350 * task's utilization has to be added to running_bw again.
351 * A flag in the deadline scheduling entity (dl_non_contending)
352 * is used to avoid race conditions between the inactive timer handler
355 * The following diagram shows how running_bw is updated. A task is
356 * "ACTIVE" when its utilization contributes to running_bw; an
357 * "ACTIVE contending" task is in the TASK_RUNNING state, while an
358 * "ACTIVE non contending" task is a blocked task for which the "0-lag time"
359 * has not passed yet. An "INACTIVE" task is a task for which the "0-lag"
360 * time already passed, which does not contribute to running_bw anymore.
361 * +------------------+
363 * +------------------>+ contending |
364 * | add_running_bw | |
365 * | +----+------+------+
368 * +--------+-------+ | |
369 * | | t >= 0-lag | | wakeup
370 * | INACTIVE |<---------------+ |
371 * | | sub_running_bw | |
372 * +--------+-------+ | |
377 * | +----+------+------+
378 * | sub_running_bw | ACTIVE |
379 * +-------------------+ |
380 * inactive timer | non contending |
381 * fired +------------------+
383 * The task_non_contending() function is invoked when a task
384 * blocks, and checks if the 0-lag time already passed or
385 * not (in the first case, it directly updates running_bw;
386 * in the second case, it arms the inactive timer).
388 * The task_contending() function is invoked when a task wakes
389 * up, and checks if the task is still in the "ACTIVE non contending"
390 * state or not (in the second case, it updates running_bw).
392 static void task_non_contending(struct task_struct *p)
394 struct sched_dl_entity *dl_se = &p->dl;
395 struct hrtimer *timer = &dl_se->inactive_timer;
396 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
397 struct rq *rq = rq_of_dl_rq(dl_rq);
401 * If this is a non-deadline task that has been boosted,
404 if (dl_se->dl_runtime == 0)
407 if (dl_entity_is_special(dl_se))
410 WARN_ON(dl_se->dl_non_contending);
412 zerolag_time = dl_se->deadline -
413 div64_long((dl_se->runtime * dl_se->dl_period),
417 * Using relative times instead of the absolute "0-lag time"
418 * allows to simplify the code
420 zerolag_time -= rq_clock(rq);
423 * If the "0-lag time" already passed, decrease the active
424 * utilization now, instead of starting a timer
426 if ((zerolag_time < 0) || hrtimer_active(&dl_se->inactive_timer)) {
428 sub_running_bw(dl_se, dl_rq);
429 if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) {
430 struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
432 if (READ_ONCE(p->__state) == TASK_DEAD)
433 sub_rq_bw(&p->dl, &rq->dl);
434 raw_spin_lock(&dl_b->lock);
435 __dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
436 raw_spin_unlock(&dl_b->lock);
437 __dl_clear_params(p);
443 dl_se->dl_non_contending = 1;
445 hrtimer_start(timer, ns_to_ktime(zerolag_time), HRTIMER_MODE_REL_HARD);
448 static void task_contending(struct sched_dl_entity *dl_se, int flags)
450 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
453 * If this is a non-deadline task that has been boosted,
456 if (dl_se->dl_runtime == 0)
459 if (flags & ENQUEUE_MIGRATED)
460 add_rq_bw(dl_se, dl_rq);
462 if (dl_se->dl_non_contending) {
463 dl_se->dl_non_contending = 0;
465 * If the timer handler is currently running and the
466 * timer cannot be canceled, inactive_task_timer()
467 * will see that dl_not_contending is not set, and
468 * will not touch the rq's active utilization,
469 * so we are still safe.
471 if (hrtimer_try_to_cancel(&dl_se->inactive_timer) == 1)
472 put_task_struct(dl_task_of(dl_se));
475 * Since "dl_non_contending" is not set, the
476 * task's utilization has already been removed from
477 * active utilization (either when the task blocked,
478 * when the "inactive timer" fired).
481 add_running_bw(dl_se, dl_rq);
485 static inline int is_leftmost(struct task_struct *p, struct dl_rq *dl_rq)
487 struct sched_dl_entity *dl_se = &p->dl;
489 return rb_first_cached(&dl_rq->root) == &dl_se->rb_node;
492 static void init_dl_rq_bw_ratio(struct dl_rq *dl_rq);
494 void init_dl_bw(struct dl_bw *dl_b)
496 raw_spin_lock_init(&dl_b->lock);
497 if (global_rt_runtime() == RUNTIME_INF)
500 dl_b->bw = to_ratio(global_rt_period(), global_rt_runtime());
504 void init_dl_rq(struct dl_rq *dl_rq)
506 dl_rq->root = RB_ROOT_CACHED;
509 /* zero means no -deadline tasks */
510 dl_rq->earliest_dl.curr = dl_rq->earliest_dl.next = 0;
512 dl_rq->overloaded = 0;
513 dl_rq->pushable_dl_tasks_root = RB_ROOT_CACHED;
515 init_dl_bw(&dl_rq->dl_bw);
518 dl_rq->running_bw = 0;
520 init_dl_rq_bw_ratio(dl_rq);
525 static inline int dl_overloaded(struct rq *rq)
527 return atomic_read(&rq->rd->dlo_count);
530 static inline void dl_set_overload(struct rq *rq)
535 cpumask_set_cpu(rq->cpu, rq->rd->dlo_mask);
537 * Must be visible before the overload count is
538 * set (as in sched_rt.c).
540 * Matched by the barrier in pull_dl_task().
543 atomic_inc(&rq->rd->dlo_count);
546 static inline void dl_clear_overload(struct rq *rq)
551 atomic_dec(&rq->rd->dlo_count);
552 cpumask_clear_cpu(rq->cpu, rq->rd->dlo_mask);
555 #define __node_2_pdl(node) \
556 rb_entry((node), struct task_struct, pushable_dl_tasks)
558 static inline bool __pushable_less(struct rb_node *a, const struct rb_node *b)
560 return dl_entity_preempt(&__node_2_pdl(a)->dl, &__node_2_pdl(b)->dl);
563 static inline int has_pushable_dl_tasks(struct rq *rq)
565 return !RB_EMPTY_ROOT(&rq->dl.pushable_dl_tasks_root.rb_root);
569 * The list of pushable -deadline task is not a plist, like in
570 * sched_rt.c, it is an rb-tree with tasks ordered by deadline.
572 static void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
574 struct rb_node *leftmost;
576 WARN_ON_ONCE(!RB_EMPTY_NODE(&p->pushable_dl_tasks));
578 leftmost = rb_add_cached(&p->pushable_dl_tasks,
579 &rq->dl.pushable_dl_tasks_root,
582 rq->dl.earliest_dl.next = p->dl.deadline;
584 if (!rq->dl.overloaded) {
586 rq->dl.overloaded = 1;
590 static void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
592 struct dl_rq *dl_rq = &rq->dl;
593 struct rb_root_cached *root = &dl_rq->pushable_dl_tasks_root;
594 struct rb_node *leftmost;
596 if (RB_EMPTY_NODE(&p->pushable_dl_tasks))
599 leftmost = rb_erase_cached(&p->pushable_dl_tasks, root);
601 dl_rq->earliest_dl.next = __node_2_pdl(leftmost)->dl.deadline;
603 RB_CLEAR_NODE(&p->pushable_dl_tasks);
605 if (!has_pushable_dl_tasks(rq) && rq->dl.overloaded) {
606 dl_clear_overload(rq);
607 rq->dl.overloaded = 0;
611 static int push_dl_task(struct rq *rq);
613 static inline bool need_pull_dl_task(struct rq *rq, struct task_struct *prev)
615 return rq->online && dl_task(prev);
618 static DEFINE_PER_CPU(struct balance_callback, dl_push_head);
619 static DEFINE_PER_CPU(struct balance_callback, dl_pull_head);
621 static void push_dl_tasks(struct rq *);
622 static void pull_dl_task(struct rq *);
624 static inline void deadline_queue_push_tasks(struct rq *rq)
626 if (!has_pushable_dl_tasks(rq))
629 queue_balance_callback(rq, &per_cpu(dl_push_head, rq->cpu), push_dl_tasks);
632 static inline void deadline_queue_pull_task(struct rq *rq)
634 queue_balance_callback(rq, &per_cpu(dl_pull_head, rq->cpu), pull_dl_task);
637 static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq);
639 static struct rq *dl_task_offline_migration(struct rq *rq, struct task_struct *p)
641 struct rq *later_rq = NULL;
644 later_rq = find_lock_later_rq(p, rq);
649 * If we cannot preempt any rq, fall back to pick any
652 cpu = cpumask_any_and(cpu_active_mask, p->cpus_ptr);
653 if (cpu >= nr_cpu_ids) {
655 * Failed to find any suitable CPU.
656 * The task will never come back!
658 WARN_ON_ONCE(dl_bandwidth_enabled());
661 * If admission control is disabled we
662 * try a little harder to let the task
665 cpu = cpumask_any(cpu_active_mask);
667 later_rq = cpu_rq(cpu);
668 double_lock_balance(rq, later_rq);
671 if (p->dl.dl_non_contending || p->dl.dl_throttled) {
673 * Inactive timer is armed (or callback is running, but
674 * waiting for us to release rq locks). In any case, when it
675 * will fire (or continue), it will see running_bw of this
676 * task migrated to later_rq (and correctly handle it).
678 sub_running_bw(&p->dl, &rq->dl);
679 sub_rq_bw(&p->dl, &rq->dl);
681 add_rq_bw(&p->dl, &later_rq->dl);
682 add_running_bw(&p->dl, &later_rq->dl);
684 sub_rq_bw(&p->dl, &rq->dl);
685 add_rq_bw(&p->dl, &later_rq->dl);
689 * And we finally need to fixup root_domain(s) bandwidth accounting,
690 * since p is still hanging out in the old (now moved to default) root
693 dl_b = &rq->rd->dl_bw;
694 raw_spin_lock(&dl_b->lock);
695 __dl_sub(dl_b, p->dl.dl_bw, cpumask_weight(rq->rd->span));
696 raw_spin_unlock(&dl_b->lock);
698 dl_b = &later_rq->rd->dl_bw;
699 raw_spin_lock(&dl_b->lock);
700 __dl_add(dl_b, p->dl.dl_bw, cpumask_weight(later_rq->rd->span));
701 raw_spin_unlock(&dl_b->lock);
703 set_task_cpu(p, later_rq->cpu);
704 double_unlock_balance(later_rq, rq);
712 void enqueue_pushable_dl_task(struct rq *rq, struct task_struct *p)
717 void dequeue_pushable_dl_task(struct rq *rq, struct task_struct *p)
722 void inc_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
727 void dec_dl_migration(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
731 static inline void deadline_queue_push_tasks(struct rq *rq)
735 static inline void deadline_queue_pull_task(struct rq *rq)
738 #endif /* CONFIG_SMP */
740 static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags);
741 static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags);
742 static void wakeup_preempt_dl(struct rq *rq, struct task_struct *p, int flags);
744 static inline void replenish_dl_new_period(struct sched_dl_entity *dl_se,
747 /* for non-boosted task, pi_of(dl_se) == dl_se */
748 dl_se->deadline = rq_clock(rq) + pi_of(dl_se)->dl_deadline;
749 dl_se->runtime = pi_of(dl_se)->dl_runtime;
753 * We are being explicitly informed that a new instance is starting,
754 * and this means that:
755 * - the absolute deadline of the entity has to be placed at
756 * current time + relative deadline;
757 * - the runtime of the entity has to be set to the maximum value.
759 * The capability of specifying such event is useful whenever a -deadline
760 * entity wants to (try to!) synchronize its behaviour with the scheduler's
761 * one, and to (try to!) reconcile itself with its own scheduling
764 static inline void setup_new_dl_entity(struct sched_dl_entity *dl_se)
766 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
767 struct rq *rq = rq_of_dl_rq(dl_rq);
769 WARN_ON(is_dl_boosted(dl_se));
770 WARN_ON(dl_time_before(rq_clock(rq), dl_se->deadline));
773 * We are racing with the deadline timer. So, do nothing because
774 * the deadline timer handler will take care of properly recharging
775 * the runtime and postponing the deadline
777 if (dl_se->dl_throttled)
781 * We use the regular wall clock time to set deadlines in the
782 * future; in fact, we must consider execution overheads (time
783 * spent on hardirq context, etc.).
785 replenish_dl_new_period(dl_se, rq);
789 * Pure Earliest Deadline First (EDF) scheduling does not deal with the
790 * possibility of a entity lasting more than what it declared, and thus
791 * exhausting its runtime.
793 * Here we are interested in making runtime overrun possible, but we do
794 * not want a entity which is misbehaving to affect the scheduling of all
796 * Therefore, a budgeting strategy called Constant Bandwidth Server (CBS)
797 * is used, in order to confine each entity within its own bandwidth.
799 * This function deals exactly with that, and ensures that when the runtime
800 * of a entity is replenished, its deadline is also postponed. That ensures
801 * the overrunning entity can't interfere with other entity in the system and
802 * can't make them miss their deadlines. Reasons why this kind of overruns
803 * could happen are, typically, a entity voluntarily trying to overcome its
804 * runtime, or it just underestimated it during sched_setattr().
806 static void replenish_dl_entity(struct sched_dl_entity *dl_se)
808 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
809 struct rq *rq = rq_of_dl_rq(dl_rq);
811 WARN_ON_ONCE(pi_of(dl_se)->dl_runtime <= 0);
814 * This could be the case for a !-dl task that is boosted.
815 * Just go with full inherited parameters.
817 if (dl_se->dl_deadline == 0)
818 replenish_dl_new_period(dl_se, rq);
820 if (dl_se->dl_yielded && dl_se->runtime > 0)
824 * We keep moving the deadline away until we get some
825 * available runtime for the entity. This ensures correct
826 * handling of situations where the runtime overrun is
829 while (dl_se->runtime <= 0) {
830 dl_se->deadline += pi_of(dl_se)->dl_period;
831 dl_se->runtime += pi_of(dl_se)->dl_runtime;
835 * At this point, the deadline really should be "in
836 * the future" with respect to rq->clock. If it's
837 * not, we are, for some reason, lagging too much!
838 * Anyway, after having warn userspace abut that,
839 * we still try to keep the things running by
840 * resetting the deadline and the budget of the
843 if (dl_time_before(dl_se->deadline, rq_clock(rq))) {
844 printk_deferred_once("sched: DL replenish lagged too much\n");
845 replenish_dl_new_period(dl_se, rq);
848 if (dl_se->dl_yielded)
849 dl_se->dl_yielded = 0;
850 if (dl_se->dl_throttled)
851 dl_se->dl_throttled = 0;
855 * Here we check if --at time t-- an entity (which is probably being
856 * [re]activated or, in general, enqueued) can use its remaining runtime
857 * and its current deadline _without_ exceeding the bandwidth it is
858 * assigned (function returns true if it can't). We are in fact applying
859 * one of the CBS rules: when a task wakes up, if the residual runtime
860 * over residual deadline fits within the allocated bandwidth, then we
861 * can keep the current (absolute) deadline and residual budget without
862 * disrupting the schedulability of the system. Otherwise, we should
863 * refill the runtime and set the deadline a period in the future,
864 * because keeping the current (absolute) deadline of the task would
865 * result in breaking guarantees promised to other tasks (refer to
866 * Documentation/scheduler/sched-deadline.rst for more information).
868 * This function returns true if:
870 * runtime / (deadline - t) > dl_runtime / dl_deadline ,
872 * IOW we can't recycle current parameters.
874 * Notice that the bandwidth check is done against the deadline. For
875 * task with deadline equal to period this is the same of using
876 * dl_period instead of dl_deadline in the equation above.
878 static bool dl_entity_overflow(struct sched_dl_entity *dl_se, u64 t)
883 * left and right are the two sides of the equation above,
884 * after a bit of shuffling to use multiplications instead
887 * Note that none of the time values involved in the two
888 * multiplications are absolute: dl_deadline and dl_runtime
889 * are the relative deadline and the maximum runtime of each
890 * instance, runtime is the runtime left for the last instance
891 * and (deadline - t), since t is rq->clock, is the time left
892 * to the (absolute) deadline. Even if overflowing the u64 type
893 * is very unlikely to occur in both cases, here we scale down
894 * as we want to avoid that risk at all. Scaling down by 10
895 * means that we reduce granularity to 1us. We are fine with it,
896 * since this is only a true/false check and, anyway, thinking
897 * of anything below microseconds resolution is actually fiction
898 * (but still we want to give the user that illusion >;).
900 left = (pi_of(dl_se)->dl_deadline >> DL_SCALE) * (dl_se->runtime >> DL_SCALE);
901 right = ((dl_se->deadline - t) >> DL_SCALE) *
902 (pi_of(dl_se)->dl_runtime >> DL_SCALE);
904 return dl_time_before(right, left);
908 * Revised wakeup rule [1]: For self-suspending tasks, rather then
909 * re-initializing task's runtime and deadline, the revised wakeup
910 * rule adjusts the task's runtime to avoid the task to overrun its
913 * Reasoning: a task may overrun the density if:
914 * runtime / (deadline - t) > dl_runtime / dl_deadline
916 * Therefore, runtime can be adjusted to:
917 * runtime = (dl_runtime / dl_deadline) * (deadline - t)
919 * In such way that runtime will be equal to the maximum density
920 * the task can use without breaking any rule.
922 * [1] Luca Abeni, Giuseppe Lipari, and Juri Lelli. 2015. Constant
923 * bandwidth server revisited. SIGBED Rev. 11, 4 (January 2015), 19-24.
926 update_dl_revised_wakeup(struct sched_dl_entity *dl_se, struct rq *rq)
928 u64 laxity = dl_se->deadline - rq_clock(rq);
931 * If the task has deadline < period, and the deadline is in the past,
932 * it should already be throttled before this check.
934 * See update_dl_entity() comments for further details.
936 WARN_ON(dl_time_before(dl_se->deadline, rq_clock(rq)));
938 dl_se->runtime = (dl_se->dl_density * laxity) >> BW_SHIFT;
942 * Regarding the deadline, a task with implicit deadline has a relative
943 * deadline == relative period. A task with constrained deadline has a
944 * relative deadline <= relative period.
946 * We support constrained deadline tasks. However, there are some restrictions
947 * applied only for tasks which do not have an implicit deadline. See
948 * update_dl_entity() to know more about such restrictions.
950 * The dl_is_implicit() returns true if the task has an implicit deadline.
952 static inline bool dl_is_implicit(struct sched_dl_entity *dl_se)
954 return dl_se->dl_deadline == dl_se->dl_period;
958 * When a deadline entity is placed in the runqueue, its runtime and deadline
959 * might need to be updated. This is done by a CBS wake up rule. There are two
960 * different rules: 1) the original CBS; and 2) the Revisited CBS.
962 * When the task is starting a new period, the Original CBS is used. In this
963 * case, the runtime is replenished and a new absolute deadline is set.
965 * When a task is queued before the begin of the next period, using the
966 * remaining runtime and deadline could make the entity to overflow, see
967 * dl_entity_overflow() to find more about runtime overflow. When such case
968 * is detected, the runtime and deadline need to be updated.
970 * If the task has an implicit deadline, i.e., deadline == period, the Original
971 * CBS is applied. the runtime is replenished and a new absolute deadline is
972 * set, as in the previous cases.
974 * However, the Original CBS does not work properly for tasks with
975 * deadline < period, which are said to have a constrained deadline. By
976 * applying the Original CBS, a constrained deadline task would be able to run
977 * runtime/deadline in a period. With deadline < period, the task would
978 * overrun the runtime/period allowed bandwidth, breaking the admission test.
980 * In order to prevent this misbehave, the Revisited CBS is used for
981 * constrained deadline tasks when a runtime overflow is detected. In the
982 * Revisited CBS, rather than replenishing & setting a new absolute deadline,
983 * the remaining runtime of the task is reduced to avoid runtime overflow.
984 * Please refer to the comments update_dl_revised_wakeup() function to find
985 * more about the Revised CBS rule.
987 static void update_dl_entity(struct sched_dl_entity *dl_se)
989 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
990 struct rq *rq = rq_of_dl_rq(dl_rq);
992 if (dl_time_before(dl_se->deadline, rq_clock(rq)) ||
993 dl_entity_overflow(dl_se, rq_clock(rq))) {
995 if (unlikely(!dl_is_implicit(dl_se) &&
996 !dl_time_before(dl_se->deadline, rq_clock(rq)) &&
997 !is_dl_boosted(dl_se))) {
998 update_dl_revised_wakeup(dl_se, rq);
1002 replenish_dl_new_period(dl_se, rq);
1006 static inline u64 dl_next_period(struct sched_dl_entity *dl_se)
1008 return dl_se->deadline - dl_se->dl_deadline + dl_se->dl_period;
1012 * If the entity depleted all its runtime, and if we want it to sleep
1013 * while waiting for some new execution time to become available, we
1014 * set the bandwidth replenishment timer to the replenishment instant
1015 * and try to activate it.
1017 * Notice that it is important for the caller to know if the timer
1018 * actually started or not (i.e., the replenishment instant is in
1019 * the future or in the past).
1021 static int start_dl_timer(struct task_struct *p)
1023 struct sched_dl_entity *dl_se = &p->dl;
1024 struct hrtimer *timer = &dl_se->dl_timer;
1025 struct rq *rq = task_rq(p);
1029 lockdep_assert_rq_held(rq);
1032 * We want the timer to fire at the deadline, but considering
1033 * that it is actually coming from rq->clock and not from
1034 * hrtimer's time base reading.
1036 act = ns_to_ktime(dl_next_period(dl_se));
1037 now = hrtimer_cb_get_time(timer);
1038 delta = ktime_to_ns(now) - rq_clock(rq);
1039 act = ktime_add_ns(act, delta);
1042 * If the expiry time already passed, e.g., because the value
1043 * chosen as the deadline is too small, don't even try to
1044 * start the timer in the past!
1046 if (ktime_us_delta(act, now) < 0)
1050 * !enqueued will guarantee another callback; even if one is already in
1051 * progress. This ensures a balanced {get,put}_task_struct().
1053 * The race against __run_timer() clearing the enqueued state is
1054 * harmless because we're holding task_rq()->lock, therefore the timer
1055 * expiring after we've done the check will wait on its task_rq_lock()
1056 * and observe our state.
1058 if (!hrtimer_is_queued(timer)) {
1060 hrtimer_start(timer, act, HRTIMER_MODE_ABS_HARD);
1067 * This is the bandwidth enforcement timer callback. If here, we know
1068 * a task is not on its dl_rq, since the fact that the timer was running
1069 * means the task is throttled and needs a runtime replenishment.
1071 * However, what we actually do depends on the fact the task is active,
1072 * (it is on its rq) or has been removed from there by a call to
1073 * dequeue_task_dl(). In the former case we must issue the runtime
1074 * replenishment and add the task back to the dl_rq; in the latter, we just
1075 * do nothing but clearing dl_throttled, so that runtime and deadline
1076 * updating (and the queueing back to dl_rq) will be done by the
1077 * next call to enqueue_task_dl().
1079 static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
1081 struct sched_dl_entity *dl_se = container_of(timer,
1082 struct sched_dl_entity,
1084 struct task_struct *p = dl_task_of(dl_se);
1088 rq = task_rq_lock(p, &rf);
1091 * The task might have changed its scheduling policy to something
1092 * different than SCHED_DEADLINE (through switched_from_dl()).
1098 * The task might have been boosted by someone else and might be in the
1099 * boosting/deboosting path, its not throttled.
1101 if (is_dl_boosted(dl_se))
1105 * Spurious timer due to start_dl_timer() race; or we already received
1106 * a replenishment from rt_mutex_setprio().
1108 if (!dl_se->dl_throttled)
1112 update_rq_clock(rq);
1115 * If the throttle happened during sched-out; like:
1122 * __dequeue_task_dl()
1125 * We can be both throttled and !queued. Replenish the counter
1126 * but do not enqueue -- wait for our wakeup to do that.
1128 if (!task_on_rq_queued(p)) {
1129 replenish_dl_entity(dl_se);
1134 if (unlikely(!rq->online)) {
1136 * If the runqueue is no longer available, migrate the
1137 * task elsewhere. This necessarily changes rq.
1139 lockdep_unpin_lock(__rq_lockp(rq), rf.cookie);
1140 rq = dl_task_offline_migration(rq, p);
1141 rf.cookie = lockdep_pin_lock(__rq_lockp(rq));
1142 update_rq_clock(rq);
1145 * Now that the task has been migrated to the new RQ and we
1146 * have that locked, proceed as normal and enqueue the task
1152 enqueue_task_dl(rq, p, ENQUEUE_REPLENISH);
1153 if (dl_task(rq->curr))
1154 wakeup_preempt_dl(rq, p, 0);
1160 * Queueing this task back might have overloaded rq, check if we need
1161 * to kick someone away.
1163 if (has_pushable_dl_tasks(rq)) {
1165 * Nothing relies on rq->lock after this, so its safe to drop
1168 rq_unpin_lock(rq, &rf);
1170 rq_repin_lock(rq, &rf);
1175 task_rq_unlock(rq, p, &rf);
1178 * This can free the task_struct, including this hrtimer, do not touch
1179 * anything related to that after this.
1183 return HRTIMER_NORESTART;
1186 void init_dl_task_timer(struct sched_dl_entity *dl_se)
1188 struct hrtimer *timer = &dl_se->dl_timer;
1190 hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
1191 timer->function = dl_task_timer;
1195 * During the activation, CBS checks if it can reuse the current task's
1196 * runtime and period. If the deadline of the task is in the past, CBS
1197 * cannot use the runtime, and so it replenishes the task. This rule
1198 * works fine for implicit deadline tasks (deadline == period), and the
1199 * CBS was designed for implicit deadline tasks. However, a task with
1200 * constrained deadline (deadline < period) might be awakened after the
1201 * deadline, but before the next period. In this case, replenishing the
1202 * task would allow it to run for runtime / deadline. As in this case
1203 * deadline < period, CBS enables a task to run for more than the
1204 * runtime / period. In a very loaded system, this can cause a domino
1205 * effect, making other tasks miss their deadlines.
1207 * To avoid this problem, in the activation of a constrained deadline
1208 * task after the deadline but before the next period, throttle the
1209 * task and set the replenishing timer to the begin of the next period,
1210 * unless it is boosted.
1212 static inline void dl_check_constrained_dl(struct sched_dl_entity *dl_se)
1214 struct task_struct *p = dl_task_of(dl_se);
1215 struct rq *rq = rq_of_dl_rq(dl_rq_of_se(dl_se));
1217 if (dl_time_before(dl_se->deadline, rq_clock(rq)) &&
1218 dl_time_before(rq_clock(rq), dl_next_period(dl_se))) {
1219 if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(p)))
1221 dl_se->dl_throttled = 1;
1222 if (dl_se->runtime > 0)
1228 int dl_runtime_exceeded(struct sched_dl_entity *dl_se)
1230 return (dl_se->runtime <= 0);
1234 * This function implements the GRUB accounting rule. According to the
1235 * GRUB reclaiming algorithm, the runtime is not decreased as "dq = -dt",
1236 * but as "dq = -(max{u, (Umax - Uinact - Uextra)} / Umax) dt",
1237 * where u is the utilization of the task, Umax is the maximum reclaimable
1238 * utilization, Uinact is the (per-runqueue) inactive utilization, computed
1239 * as the difference between the "total runqueue utilization" and the
1240 * "runqueue active utilization", and Uextra is the (per runqueue) extra
1241 * reclaimable utilization.
1242 * Since rq->dl.running_bw and rq->dl.this_bw contain utilizations multiplied
1243 * by 2^BW_SHIFT, the result has to be shifted right by BW_SHIFT.
1244 * Since rq->dl.bw_ratio contains 1 / Umax multiplied by 2^RATIO_SHIFT, dl_bw
1245 * is multiped by rq->dl.bw_ratio and shifted right by RATIO_SHIFT.
1246 * Since delta is a 64 bit variable, to have an overflow its value should be
1247 * larger than 2^(64 - 20 - 8), which is more than 64 seconds. So, overflow is
1248 * not an issue here.
1250 static u64 grub_reclaim(u64 delta, struct rq *rq, struct sched_dl_entity *dl_se)
1253 u64 u_inact = rq->dl.this_bw - rq->dl.running_bw; /* Utot - Uact */
1256 * Instead of computing max{u, (u_max - u_inact - u_extra)}, we
1257 * compare u_inact + u_extra with u_max - u, because u_inact + u_extra
1258 * can be larger than u_max. So, u_max - u_inact - u_extra would be
1259 * negative leading to wrong results.
1261 if (u_inact + rq->dl.extra_bw > rq->dl.max_bw - dl_se->dl_bw)
1262 u_act = dl_se->dl_bw;
1264 u_act = rq->dl.max_bw - u_inact - rq->dl.extra_bw;
1266 u_act = (u_act * rq->dl.bw_ratio) >> RATIO_SHIFT;
1267 return (delta * u_act) >> BW_SHIFT;
1271 * Update the current task's runtime statistics (provided it is still
1272 * a -deadline task and has not been removed from the dl_rq).
1274 static void update_curr_dl(struct rq *rq)
1276 struct task_struct *curr = rq->curr;
1277 struct sched_dl_entity *dl_se = &curr->dl;
1278 u64 delta_exec, scaled_delta_exec;
1279 int cpu = cpu_of(rq);
1282 if (!dl_task(curr) || !on_dl_rq(dl_se))
1286 * Consumed budget is computed considering the time as
1287 * observed by schedulable tasks (excluding time spent
1288 * in hardirq context, etc.). Deadlines are instead
1289 * computed using hard walltime. This seems to be the more
1290 * natural solution, but the full ramifications of this
1291 * approach need further study.
1293 now = rq_clock_task(rq);
1294 delta_exec = now - curr->se.exec_start;
1295 if (unlikely((s64)delta_exec <= 0)) {
1296 if (unlikely(dl_se->dl_yielded))
1301 schedstat_set(curr->stats.exec_max,
1302 max(curr->stats.exec_max, delta_exec));
1304 trace_sched_stat_runtime(curr, delta_exec, 0);
1306 update_current_exec_runtime(curr, now, delta_exec);
1308 if (dl_entity_is_special(dl_se))
1312 * For tasks that participate in GRUB, we implement GRUB-PA: the
1313 * spare reclaimed bandwidth is used to clock down frequency.
1315 * For the others, we still need to scale reservation parameters
1316 * according to current frequency and CPU maximum capacity.
1318 if (unlikely(dl_se->flags & SCHED_FLAG_RECLAIM)) {
1319 scaled_delta_exec = grub_reclaim(delta_exec,
1323 unsigned long scale_freq = arch_scale_freq_capacity(cpu);
1324 unsigned long scale_cpu = arch_scale_cpu_capacity(cpu);
1326 scaled_delta_exec = cap_scale(delta_exec, scale_freq);
1327 scaled_delta_exec = cap_scale(scaled_delta_exec, scale_cpu);
1330 dl_se->runtime -= scaled_delta_exec;
1333 if (dl_runtime_exceeded(dl_se) || dl_se->dl_yielded) {
1334 dl_se->dl_throttled = 1;
1336 /* If requested, inform the user about runtime overruns. */
1337 if (dl_runtime_exceeded(dl_se) &&
1338 (dl_se->flags & SCHED_FLAG_DL_OVERRUN))
1339 dl_se->dl_overrun = 1;
1341 __dequeue_task_dl(rq, curr, 0);
1342 if (unlikely(is_dl_boosted(dl_se) || !start_dl_timer(curr)))
1343 enqueue_task_dl(rq, curr, ENQUEUE_REPLENISH);
1345 if (!is_leftmost(curr, &rq->dl))
1350 * Because -- for now -- we share the rt bandwidth, we need to
1351 * account our runtime there too, otherwise actual rt tasks
1352 * would be able to exceed the shared quota.
1354 * Account to the root rt group for now.
1356 * The solution we're working towards is having the RT groups scheduled
1357 * using deadline servers -- however there's a few nasties to figure
1358 * out before that can happen.
1360 if (rt_bandwidth_enabled()) {
1361 struct rt_rq *rt_rq = &rq->rt;
1363 raw_spin_lock(&rt_rq->rt_runtime_lock);
1365 * We'll let actual RT tasks worry about the overflow here, we
1366 * have our own CBS to keep us inline; only account when RT
1367 * bandwidth is relevant.
1369 if (sched_rt_bandwidth_account(rt_rq))
1370 rt_rq->rt_time += delta_exec;
1371 raw_spin_unlock(&rt_rq->rt_runtime_lock);
1375 static enum hrtimer_restart inactive_task_timer(struct hrtimer *timer)
1377 struct sched_dl_entity *dl_se = container_of(timer,
1378 struct sched_dl_entity,
1380 struct task_struct *p = dl_task_of(dl_se);
1384 rq = task_rq_lock(p, &rf);
1387 update_rq_clock(rq);
1389 if (!dl_task(p) || READ_ONCE(p->__state) == TASK_DEAD) {
1390 struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
1392 if (READ_ONCE(p->__state) == TASK_DEAD && dl_se->dl_non_contending) {
1393 sub_running_bw(&p->dl, dl_rq_of_se(&p->dl));
1394 sub_rq_bw(&p->dl, dl_rq_of_se(&p->dl));
1395 dl_se->dl_non_contending = 0;
1398 raw_spin_lock(&dl_b->lock);
1399 __dl_sub(dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
1400 raw_spin_unlock(&dl_b->lock);
1401 __dl_clear_params(p);
1405 if (dl_se->dl_non_contending == 0)
1408 sub_running_bw(dl_se, &rq->dl);
1409 dl_se->dl_non_contending = 0;
1411 task_rq_unlock(rq, p, &rf);
1414 return HRTIMER_NORESTART;
1417 void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se)
1419 struct hrtimer *timer = &dl_se->inactive_timer;
1421 hrtimer_init(timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
1422 timer->function = inactive_task_timer;
1425 #define __node_2_dle(node) \
1426 rb_entry((node), struct sched_dl_entity, rb_node)
1430 static void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
1432 struct rq *rq = rq_of_dl_rq(dl_rq);
1434 if (dl_rq->earliest_dl.curr == 0 ||
1435 dl_time_before(deadline, dl_rq->earliest_dl.curr)) {
1436 if (dl_rq->earliest_dl.curr == 0)
1437 cpupri_set(&rq->rd->cpupri, rq->cpu, CPUPRI_HIGHER);
1438 dl_rq->earliest_dl.curr = deadline;
1439 cpudl_set(&rq->rd->cpudl, rq->cpu, deadline);
1443 static void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline)
1445 struct rq *rq = rq_of_dl_rq(dl_rq);
1448 * Since we may have removed our earliest (and/or next earliest)
1449 * task we must recompute them.
1451 if (!dl_rq->dl_nr_running) {
1452 dl_rq->earliest_dl.curr = 0;
1453 dl_rq->earliest_dl.next = 0;
1454 cpudl_clear(&rq->rd->cpudl, rq->cpu);
1455 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1457 struct rb_node *leftmost = rb_first_cached(&dl_rq->root);
1458 struct sched_dl_entity *entry = __node_2_dle(leftmost);
1460 dl_rq->earliest_dl.curr = entry->deadline;
1461 cpudl_set(&rq->rd->cpudl, rq->cpu, entry->deadline);
1467 static inline void inc_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
1468 static inline void dec_dl_deadline(struct dl_rq *dl_rq, u64 deadline) {}
1470 #endif /* CONFIG_SMP */
1473 void inc_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
1475 int prio = dl_task_of(dl_se)->prio;
1476 u64 deadline = dl_se->deadline;
1478 WARN_ON(!dl_prio(prio));
1479 dl_rq->dl_nr_running++;
1480 add_nr_running(rq_of_dl_rq(dl_rq), 1);
1482 inc_dl_deadline(dl_rq, deadline);
1486 void dec_dl_tasks(struct sched_dl_entity *dl_se, struct dl_rq *dl_rq)
1488 int prio = dl_task_of(dl_se)->prio;
1490 WARN_ON(!dl_prio(prio));
1491 WARN_ON(!dl_rq->dl_nr_running);
1492 dl_rq->dl_nr_running--;
1493 sub_nr_running(rq_of_dl_rq(dl_rq), 1);
1495 dec_dl_deadline(dl_rq, dl_se->deadline);
1498 static inline bool __dl_less(struct rb_node *a, const struct rb_node *b)
1500 return dl_time_before(__node_2_dle(a)->deadline, __node_2_dle(b)->deadline);
1503 static inline struct sched_statistics *
1504 __schedstats_from_dl_se(struct sched_dl_entity *dl_se)
1506 return &dl_task_of(dl_se)->stats;
1510 update_stats_wait_start_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se)
1512 struct sched_statistics *stats;
1514 if (!schedstat_enabled())
1517 stats = __schedstats_from_dl_se(dl_se);
1518 __update_stats_wait_start(rq_of_dl_rq(dl_rq), dl_task_of(dl_se), stats);
1522 update_stats_wait_end_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se)
1524 struct sched_statistics *stats;
1526 if (!schedstat_enabled())
1529 stats = __schedstats_from_dl_se(dl_se);
1530 __update_stats_wait_end(rq_of_dl_rq(dl_rq), dl_task_of(dl_se), stats);
1534 update_stats_enqueue_sleeper_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se)
1536 struct sched_statistics *stats;
1538 if (!schedstat_enabled())
1541 stats = __schedstats_from_dl_se(dl_se);
1542 __update_stats_enqueue_sleeper(rq_of_dl_rq(dl_rq), dl_task_of(dl_se), stats);
1546 update_stats_enqueue_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se,
1549 if (!schedstat_enabled())
1552 if (flags & ENQUEUE_WAKEUP)
1553 update_stats_enqueue_sleeper_dl(dl_rq, dl_se);
1557 update_stats_dequeue_dl(struct dl_rq *dl_rq, struct sched_dl_entity *dl_se,
1560 struct task_struct *p = dl_task_of(dl_se);
1562 if (!schedstat_enabled())
1565 if ((flags & DEQUEUE_SLEEP)) {
1568 state = READ_ONCE(p->__state);
1569 if (state & TASK_INTERRUPTIBLE)
1570 __schedstat_set(p->stats.sleep_start,
1571 rq_clock(rq_of_dl_rq(dl_rq)));
1573 if (state & TASK_UNINTERRUPTIBLE)
1574 __schedstat_set(p->stats.block_start,
1575 rq_clock(rq_of_dl_rq(dl_rq)));
1579 static void __enqueue_dl_entity(struct sched_dl_entity *dl_se)
1581 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
1583 WARN_ON_ONCE(!RB_EMPTY_NODE(&dl_se->rb_node));
1585 rb_add_cached(&dl_se->rb_node, &dl_rq->root, __dl_less);
1587 inc_dl_tasks(dl_se, dl_rq);
1590 static void __dequeue_dl_entity(struct sched_dl_entity *dl_se)
1592 struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
1594 if (RB_EMPTY_NODE(&dl_se->rb_node))
1597 rb_erase_cached(&dl_se->rb_node, &dl_rq->root);
1599 RB_CLEAR_NODE(&dl_se->rb_node);
1601 dec_dl_tasks(dl_se, dl_rq);
1605 enqueue_dl_entity(struct sched_dl_entity *dl_se, int flags)
1607 WARN_ON_ONCE(on_dl_rq(dl_se));
1609 update_stats_enqueue_dl(dl_rq_of_se(dl_se), dl_se, flags);
1612 * If this is a wakeup or a new instance, the scheduling
1613 * parameters of the task might need updating. Otherwise,
1614 * we want a replenishment of its runtime.
1616 if (flags & ENQUEUE_WAKEUP) {
1617 task_contending(dl_se, flags);
1618 update_dl_entity(dl_se);
1619 } else if (flags & ENQUEUE_REPLENISH) {
1620 replenish_dl_entity(dl_se);
1621 } else if ((flags & ENQUEUE_RESTORE) &&
1622 dl_time_before(dl_se->deadline,
1623 rq_clock(rq_of_dl_rq(dl_rq_of_se(dl_se))))) {
1624 setup_new_dl_entity(dl_se);
1627 __enqueue_dl_entity(dl_se);
1630 static void dequeue_dl_entity(struct sched_dl_entity *dl_se)
1632 __dequeue_dl_entity(dl_se);
1635 static void enqueue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1637 if (is_dl_boosted(&p->dl)) {
1639 * Because of delays in the detection of the overrun of a
1640 * thread's runtime, it might be the case that a thread
1641 * goes to sleep in a rt mutex with negative runtime. As
1642 * a consequence, the thread will be throttled.
1644 * While waiting for the mutex, this thread can also be
1645 * boosted via PI, resulting in a thread that is throttled
1646 * and boosted at the same time.
1648 * In this case, the boost overrides the throttle.
1650 if (p->dl.dl_throttled) {
1652 * The replenish timer needs to be canceled. No
1653 * problem if it fires concurrently: boosted threads
1654 * are ignored in dl_task_timer().
1656 hrtimer_try_to_cancel(&p->dl.dl_timer);
1657 p->dl.dl_throttled = 0;
1659 } else if (!dl_prio(p->normal_prio)) {
1661 * Special case in which we have a !SCHED_DEADLINE task that is going
1662 * to be deboosted, but exceeds its runtime while doing so. No point in
1663 * replenishing it, as it's going to return back to its original
1664 * scheduling class after this. If it has been throttled, we need to
1665 * clear the flag, otherwise the task may wake up as throttled after
1666 * being boosted again with no means to replenish the runtime and clear
1669 p->dl.dl_throttled = 0;
1670 if (!(flags & ENQUEUE_REPLENISH))
1671 printk_deferred_once("sched: DL de-boosted task PID %d: REPLENISH flag missing\n",
1678 * Check if a constrained deadline task was activated
1679 * after the deadline but before the next period.
1680 * If that is the case, the task will be throttled and
1681 * the replenishment timer will be set to the next period.
1683 if (!p->dl.dl_throttled && !dl_is_implicit(&p->dl))
1684 dl_check_constrained_dl(&p->dl);
1686 if (p->on_rq == TASK_ON_RQ_MIGRATING || flags & ENQUEUE_RESTORE) {
1687 add_rq_bw(&p->dl, &rq->dl);
1688 add_running_bw(&p->dl, &rq->dl);
1692 * If p is throttled, we do not enqueue it. In fact, if it exhausted
1693 * its budget it needs a replenishment and, since it now is on
1694 * its rq, the bandwidth timer callback (which clearly has not
1695 * run yet) will take care of this.
1696 * However, the active utilization does not depend on the fact
1697 * that the task is on the runqueue or not (but depends on the
1698 * task's state - in GRUB parlance, "inactive" vs "active contending").
1699 * In other words, even if a task is throttled its utilization must
1700 * be counted in the active utilization; hence, we need to call
1703 if (p->dl.dl_throttled && !(flags & ENQUEUE_REPLENISH)) {
1704 if (flags & ENQUEUE_WAKEUP)
1705 task_contending(&p->dl, flags);
1710 check_schedstat_required();
1711 update_stats_wait_start_dl(dl_rq_of_se(&p->dl), &p->dl);
1713 enqueue_dl_entity(&p->dl, flags);
1715 if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
1716 enqueue_pushable_dl_task(rq, p);
1719 static void __dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1721 update_stats_dequeue_dl(&rq->dl, &p->dl, flags);
1722 dequeue_dl_entity(&p->dl);
1723 dequeue_pushable_dl_task(rq, p);
1726 static void dequeue_task_dl(struct rq *rq, struct task_struct *p, int flags)
1729 __dequeue_task_dl(rq, p, flags);
1731 if (p->on_rq == TASK_ON_RQ_MIGRATING || flags & DEQUEUE_SAVE) {
1732 sub_running_bw(&p->dl, &rq->dl);
1733 sub_rq_bw(&p->dl, &rq->dl);
1737 * This check allows to start the inactive timer (or to immediately
1738 * decrease the active utilization, if needed) in two cases:
1739 * when the task blocks and when it is terminating
1740 * (p->state == TASK_DEAD). We can handle the two cases in the same
1741 * way, because from GRUB's point of view the same thing is happening
1742 * (the task moves from "active contending" to "active non contending"
1745 if (flags & DEQUEUE_SLEEP)
1746 task_non_contending(p);
1750 * Yield task semantic for -deadline tasks is:
1752 * get off from the CPU until our next instance, with
1753 * a new runtime. This is of little use now, since we
1754 * don't have a bandwidth reclaiming mechanism. Anyway,
1755 * bandwidth reclaiming is planned for the future, and
1756 * yield_task_dl will indicate that some spare budget
1757 * is available for other task instances to use it.
1759 static void yield_task_dl(struct rq *rq)
1762 * We make the task go to sleep until its current deadline by
1763 * forcing its runtime to zero. This way, update_curr_dl() stops
1764 * it and the bandwidth timer will wake it up and will give it
1765 * new scheduling parameters (thanks to dl_yielded=1).
1767 rq->curr->dl.dl_yielded = 1;
1769 update_rq_clock(rq);
1772 * Tell update_rq_clock() that we've just updated,
1773 * so we don't do microscopic update in schedule()
1774 * and double the fastpath cost.
1776 rq_clock_skip_update(rq);
1781 static inline bool dl_task_is_earliest_deadline(struct task_struct *p,
1784 return (!rq->dl.dl_nr_running ||
1785 dl_time_before(p->dl.deadline,
1786 rq->dl.earliest_dl.curr));
1789 static int find_later_rq(struct task_struct *task);
1792 select_task_rq_dl(struct task_struct *p, int cpu, int flags)
1794 struct task_struct *curr;
1798 if (!(flags & WF_TTWU))
1804 curr = READ_ONCE(rq->curr); /* unlocked access */
1807 * If we are dealing with a -deadline task, we must
1808 * decide where to wake it up.
1809 * If it has a later deadline and the current task
1810 * on this rq can't move (provided the waking task
1811 * can!) we prefer to send it somewhere else. On the
1812 * other hand, if it has a shorter deadline, we
1813 * try to make it stay here, it might be important.
1815 select_rq = unlikely(dl_task(curr)) &&
1816 (curr->nr_cpus_allowed < 2 ||
1817 !dl_entity_preempt(&p->dl, &curr->dl)) &&
1818 p->nr_cpus_allowed > 1;
1821 * Take the capacity of the CPU into account to
1822 * ensure it fits the requirement of the task.
1824 if (sched_asym_cpucap_active())
1825 select_rq |= !dl_task_fits_capacity(p, cpu);
1828 int target = find_later_rq(p);
1831 dl_task_is_earliest_deadline(p, cpu_rq(target)))
1840 static void migrate_task_rq_dl(struct task_struct *p, int new_cpu __maybe_unused)
1845 if (READ_ONCE(p->__state) != TASK_WAKING)
1850 * Since p->state == TASK_WAKING, set_task_cpu() has been called
1851 * from try_to_wake_up(). Hence, p->pi_lock is locked, but
1852 * rq->lock is not... So, lock it
1855 if (p->dl.dl_non_contending) {
1856 update_rq_clock(rq);
1857 sub_running_bw(&p->dl, &rq->dl);
1858 p->dl.dl_non_contending = 0;
1860 * If the timer handler is currently running and the
1861 * timer cannot be canceled, inactive_task_timer()
1862 * will see that dl_not_contending is not set, and
1863 * will not touch the rq's active utilization,
1864 * so we are still safe.
1866 if (hrtimer_try_to_cancel(&p->dl.inactive_timer) == 1)
1869 sub_rq_bw(&p->dl, &rq->dl);
1873 static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
1876 * Current can't be migrated, useless to reschedule,
1877 * let's hope p can move out.
1879 if (rq->curr->nr_cpus_allowed == 1 ||
1880 !cpudl_find(&rq->rd->cpudl, rq->curr, NULL))
1884 * p is migratable, so let's not schedule it and
1885 * see if it is pushed or pulled somewhere else.
1887 if (p->nr_cpus_allowed != 1 &&
1888 cpudl_find(&rq->rd->cpudl, p, NULL))
1894 static int balance_dl(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1896 if (!on_dl_rq(&p->dl) && need_pull_dl_task(rq, p)) {
1898 * This is OK, because current is on_cpu, which avoids it being
1899 * picked for load-balance and preemption/IRQs are still
1900 * disabled avoiding further scheduler activity on it and we've
1901 * not yet started the picking loop.
1903 rq_unpin_lock(rq, rf);
1905 rq_repin_lock(rq, rf);
1908 return sched_stop_runnable(rq) || sched_dl_runnable(rq);
1910 #endif /* CONFIG_SMP */
1913 * Only called when both the current and waking task are -deadline
1916 static void wakeup_preempt_dl(struct rq *rq, struct task_struct *p,
1919 if (dl_entity_preempt(&p->dl, &rq->curr->dl)) {
1926 * In the unlikely case current and p have the same deadline
1927 * let us try to decide what's the best thing to do...
1929 if ((p->dl.deadline == rq->curr->dl.deadline) &&
1930 !test_tsk_need_resched(rq->curr))
1931 check_preempt_equal_dl(rq, p);
1932 #endif /* CONFIG_SMP */
1935 #ifdef CONFIG_SCHED_HRTICK
1936 static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1938 hrtick_start(rq, p->dl.runtime);
1940 #else /* !CONFIG_SCHED_HRTICK */
1941 static void start_hrtick_dl(struct rq *rq, struct task_struct *p)
1946 static void set_next_task_dl(struct rq *rq, struct task_struct *p, bool first)
1948 struct sched_dl_entity *dl_se = &p->dl;
1949 struct dl_rq *dl_rq = &rq->dl;
1951 p->se.exec_start = rq_clock_task(rq);
1952 if (on_dl_rq(&p->dl))
1953 update_stats_wait_end_dl(dl_rq, dl_se);
1955 /* You can't push away the running task */
1956 dequeue_pushable_dl_task(rq, p);
1961 if (hrtick_enabled_dl(rq))
1962 start_hrtick_dl(rq, p);
1964 if (rq->curr->sched_class != &dl_sched_class)
1965 update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 0);
1967 deadline_queue_push_tasks(rq);
1970 static struct sched_dl_entity *pick_next_dl_entity(struct dl_rq *dl_rq)
1972 struct rb_node *left = rb_first_cached(&dl_rq->root);
1977 return __node_2_dle(left);
1980 static struct task_struct *pick_task_dl(struct rq *rq)
1982 struct sched_dl_entity *dl_se;
1983 struct dl_rq *dl_rq = &rq->dl;
1984 struct task_struct *p;
1986 if (!sched_dl_runnable(rq))
1989 dl_se = pick_next_dl_entity(dl_rq);
1990 WARN_ON_ONCE(!dl_se);
1991 p = dl_task_of(dl_se);
1996 static struct task_struct *pick_next_task_dl(struct rq *rq)
1998 struct task_struct *p;
2000 p = pick_task_dl(rq);
2002 set_next_task_dl(rq, p, true);
2007 static void put_prev_task_dl(struct rq *rq, struct task_struct *p)
2009 struct sched_dl_entity *dl_se = &p->dl;
2010 struct dl_rq *dl_rq = &rq->dl;
2012 if (on_dl_rq(&p->dl))
2013 update_stats_wait_start_dl(dl_rq, dl_se);
2017 update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 1);
2018 if (on_dl_rq(&p->dl) && p->nr_cpus_allowed > 1)
2019 enqueue_pushable_dl_task(rq, p);
2023 * scheduler tick hitting a task of our scheduling class.
2025 * NOTE: This function can be called remotely by the tick offload that
2026 * goes along full dynticks. Therefore no local assumption can be made
2027 * and everything must be accessed through the @rq and @curr passed in
2030 static void task_tick_dl(struct rq *rq, struct task_struct *p, int queued)
2034 update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 1);
2036 * Even when we have runtime, update_curr_dl() might have resulted in us
2037 * not being the leftmost task anymore. In that case NEED_RESCHED will
2038 * be set and schedule() will start a new hrtick for the next task.
2040 if (hrtick_enabled_dl(rq) && queued && p->dl.runtime > 0 &&
2041 is_leftmost(p, &rq->dl))
2042 start_hrtick_dl(rq, p);
2045 static void task_fork_dl(struct task_struct *p)
2048 * SCHED_DEADLINE tasks cannot fork and this is achieved through
2055 /* Only try algorithms three times */
2056 #define DL_MAX_TRIES 3
2058 static int pick_dl_task(struct rq *rq, struct task_struct *p, int cpu)
2060 if (!task_on_cpu(rq, p) &&
2061 cpumask_test_cpu(cpu, &p->cpus_mask))
2067 * Return the earliest pushable rq's task, which is suitable to be executed
2068 * on the CPU, NULL otherwise:
2070 static struct task_struct *pick_earliest_pushable_dl_task(struct rq *rq, int cpu)
2072 struct task_struct *p = NULL;
2073 struct rb_node *next_node;
2075 if (!has_pushable_dl_tasks(rq))
2078 next_node = rb_first_cached(&rq->dl.pushable_dl_tasks_root);
2082 p = __node_2_pdl(next_node);
2084 if (pick_dl_task(rq, p, cpu))
2087 next_node = rb_next(next_node);
2094 static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask_dl);
2096 static int find_later_rq(struct task_struct *task)
2098 struct sched_domain *sd;
2099 struct cpumask *later_mask = this_cpu_cpumask_var_ptr(local_cpu_mask_dl);
2100 int this_cpu = smp_processor_id();
2101 int cpu = task_cpu(task);
2103 /* Make sure the mask is initialized first */
2104 if (unlikely(!later_mask))
2107 if (task->nr_cpus_allowed == 1)
2111 * We have to consider system topology and task affinity
2112 * first, then we can look for a suitable CPU.
2114 if (!cpudl_find(&task_rq(task)->rd->cpudl, task, later_mask))
2118 * If we are here, some targets have been found, including
2119 * the most suitable which is, among the runqueues where the
2120 * current tasks have later deadlines than the task's one, the
2121 * rq with the latest possible one.
2123 * Now we check how well this matches with task's
2124 * affinity and system topology.
2126 * The last CPU where the task run is our first
2127 * guess, since it is most likely cache-hot there.
2129 if (cpumask_test_cpu(cpu, later_mask))
2132 * Check if this_cpu is to be skipped (i.e., it is
2133 * not in the mask) or not.
2135 if (!cpumask_test_cpu(this_cpu, later_mask))
2139 for_each_domain(cpu, sd) {
2140 if (sd->flags & SD_WAKE_AFFINE) {
2144 * If possible, preempting this_cpu is
2145 * cheaper than migrating.
2147 if (this_cpu != -1 &&
2148 cpumask_test_cpu(this_cpu, sched_domain_span(sd))) {
2153 best_cpu = cpumask_any_and_distribute(later_mask,
2154 sched_domain_span(sd));
2156 * Last chance: if a CPU being in both later_mask
2157 * and current sd span is valid, that becomes our
2158 * choice. Of course, the latest possible CPU is
2159 * already under consideration through later_mask.
2161 if (best_cpu < nr_cpu_ids) {
2170 * At this point, all our guesses failed, we just return
2171 * 'something', and let the caller sort the things out.
2176 cpu = cpumask_any_distribute(later_mask);
2177 if (cpu < nr_cpu_ids)
2183 /* Locks the rq it finds */
2184 static struct rq *find_lock_later_rq(struct task_struct *task, struct rq *rq)
2186 struct rq *later_rq = NULL;
2190 for (tries = 0; tries < DL_MAX_TRIES; tries++) {
2191 cpu = find_later_rq(task);
2193 if ((cpu == -1) || (cpu == rq->cpu))
2196 later_rq = cpu_rq(cpu);
2198 if (!dl_task_is_earliest_deadline(task, later_rq)) {
2200 * Target rq has tasks of equal or earlier deadline,
2201 * retrying does not release any lock and is unlikely
2202 * to yield a different result.
2208 /* Retry if something changed. */
2209 if (double_lock_balance(rq, later_rq)) {
2210 if (unlikely(task_rq(task) != rq ||
2211 !cpumask_test_cpu(later_rq->cpu, &task->cpus_mask) ||
2212 task_on_cpu(rq, task) ||
2214 is_migration_disabled(task) ||
2215 !task_on_rq_queued(task))) {
2216 double_unlock_balance(rq, later_rq);
2223 * If the rq we found has no -deadline task, or
2224 * its earliest one has a later deadline than our
2225 * task, the rq is a good one.
2227 if (dl_task_is_earliest_deadline(task, later_rq))
2230 /* Otherwise we try again. */
2231 double_unlock_balance(rq, later_rq);
2238 static struct task_struct *pick_next_pushable_dl_task(struct rq *rq)
2240 struct task_struct *p;
2242 if (!has_pushable_dl_tasks(rq))
2245 p = __node_2_pdl(rb_first_cached(&rq->dl.pushable_dl_tasks_root));
2247 WARN_ON_ONCE(rq->cpu != task_cpu(p));
2248 WARN_ON_ONCE(task_current(rq, p));
2249 WARN_ON_ONCE(p->nr_cpus_allowed <= 1);
2251 WARN_ON_ONCE(!task_on_rq_queued(p));
2252 WARN_ON_ONCE(!dl_task(p));
2258 * See if the non running -deadline tasks on this rq
2259 * can be sent to some other CPU where they can preempt
2260 * and start executing.
2262 static int push_dl_task(struct rq *rq)
2264 struct task_struct *next_task;
2265 struct rq *later_rq;
2268 next_task = pick_next_pushable_dl_task(rq);
2274 * If next_task preempts rq->curr, and rq->curr
2275 * can move away, it makes sense to just reschedule
2276 * without going further in pushing next_task.
2278 if (dl_task(rq->curr) &&
2279 dl_time_before(next_task->dl.deadline, rq->curr->dl.deadline) &&
2280 rq->curr->nr_cpus_allowed > 1) {
2285 if (is_migration_disabled(next_task))
2288 if (WARN_ON(next_task == rq->curr))
2291 /* We might release rq lock */
2292 get_task_struct(next_task);
2294 /* Will lock the rq it'll find */
2295 later_rq = find_lock_later_rq(next_task, rq);
2297 struct task_struct *task;
2300 * We must check all this again, since
2301 * find_lock_later_rq releases rq->lock and it is
2302 * then possible that next_task has migrated.
2304 task = pick_next_pushable_dl_task(rq);
2305 if (task == next_task) {
2307 * The task is still there. We don't try
2308 * again, some other CPU will pull it when ready.
2317 put_task_struct(next_task);
2322 deactivate_task(rq, next_task, 0);
2323 set_task_cpu(next_task, later_rq->cpu);
2324 activate_task(later_rq, next_task, 0);
2327 resched_curr(later_rq);
2329 double_unlock_balance(rq, later_rq);
2332 put_task_struct(next_task);
2337 static void push_dl_tasks(struct rq *rq)
2339 /* push_dl_task() will return true if it moved a -deadline task */
2340 while (push_dl_task(rq))
2344 static void pull_dl_task(struct rq *this_rq)
2346 int this_cpu = this_rq->cpu, cpu;
2347 struct task_struct *p, *push_task;
2348 bool resched = false;
2350 u64 dmin = LONG_MAX;
2352 if (likely(!dl_overloaded(this_rq)))
2356 * Match the barrier from dl_set_overloaded; this guarantees that if we
2357 * see overloaded we must also see the dlo_mask bit.
2361 for_each_cpu(cpu, this_rq->rd->dlo_mask) {
2362 if (this_cpu == cpu)
2365 src_rq = cpu_rq(cpu);
2368 * It looks racy, abd it is! However, as in sched_rt.c,
2369 * we are fine with this.
2371 if (this_rq->dl.dl_nr_running &&
2372 dl_time_before(this_rq->dl.earliest_dl.curr,
2373 src_rq->dl.earliest_dl.next))
2376 /* Might drop this_rq->lock */
2378 double_lock_balance(this_rq, src_rq);
2381 * If there are no more pullable tasks on the
2382 * rq, we're done with it.
2384 if (src_rq->dl.dl_nr_running <= 1)
2387 p = pick_earliest_pushable_dl_task(src_rq, this_cpu);
2390 * We found a task to be pulled if:
2391 * - it preempts our current (if there's one),
2392 * - it will preempt the last one we pulled (if any).
2394 if (p && dl_time_before(p->dl.deadline, dmin) &&
2395 dl_task_is_earliest_deadline(p, this_rq)) {
2396 WARN_ON(p == src_rq->curr);
2397 WARN_ON(!task_on_rq_queued(p));
2400 * Then we pull iff p has actually an earlier
2401 * deadline than the current task of its runqueue.
2403 if (dl_time_before(p->dl.deadline,
2404 src_rq->curr->dl.deadline))
2407 if (is_migration_disabled(p)) {
2408 push_task = get_push_task(src_rq);
2410 deactivate_task(src_rq, p, 0);
2411 set_task_cpu(p, this_cpu);
2412 activate_task(this_rq, p, 0);
2413 dmin = p->dl.deadline;
2417 /* Is there any other task even earlier? */
2420 double_unlock_balance(this_rq, src_rq);
2424 raw_spin_rq_unlock(this_rq);
2425 stop_one_cpu_nowait(src_rq->cpu, push_cpu_stop,
2426 push_task, &src_rq->push_work);
2428 raw_spin_rq_lock(this_rq);
2433 resched_curr(this_rq);
2437 * Since the task is not running and a reschedule is not going to happen
2438 * anytime soon on its runqueue, we try pushing it away now.
2440 static void task_woken_dl(struct rq *rq, struct task_struct *p)
2442 if (!task_on_cpu(rq, p) &&
2443 !test_tsk_need_resched(rq->curr) &&
2444 p->nr_cpus_allowed > 1 &&
2445 dl_task(rq->curr) &&
2446 (rq->curr->nr_cpus_allowed < 2 ||
2447 !dl_entity_preempt(&p->dl, &rq->curr->dl))) {
2452 static void set_cpus_allowed_dl(struct task_struct *p,
2453 struct affinity_context *ctx)
2455 struct root_domain *src_rd;
2458 WARN_ON_ONCE(!dl_task(p));
2463 * Migrating a SCHED_DEADLINE task between exclusive
2464 * cpusets (different root_domains) entails a bandwidth
2465 * update. We already made space for us in the destination
2466 * domain (see cpuset_can_attach()).
2468 if (!cpumask_intersects(src_rd->span, ctx->new_mask)) {
2469 struct dl_bw *src_dl_b;
2471 src_dl_b = dl_bw_of(cpu_of(rq));
2473 * We now free resources of the root_domain we are migrating
2474 * off. In the worst case, sched_setattr() may temporary fail
2475 * until we complete the update.
2477 raw_spin_lock(&src_dl_b->lock);
2478 __dl_sub(src_dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
2479 raw_spin_unlock(&src_dl_b->lock);
2482 set_cpus_allowed_common(p, ctx);
2485 /* Assumes rq->lock is held */
2486 static void rq_online_dl(struct rq *rq)
2488 if (rq->dl.overloaded)
2489 dl_set_overload(rq);
2491 cpudl_set_freecpu(&rq->rd->cpudl, rq->cpu);
2492 if (rq->dl.dl_nr_running > 0)
2493 cpudl_set(&rq->rd->cpudl, rq->cpu, rq->dl.earliest_dl.curr);
2496 /* Assumes rq->lock is held */
2497 static void rq_offline_dl(struct rq *rq)
2499 if (rq->dl.overloaded)
2500 dl_clear_overload(rq);
2502 cpudl_clear(&rq->rd->cpudl, rq->cpu);
2503 cpudl_clear_freecpu(&rq->rd->cpudl, rq->cpu);
2506 void __init init_sched_dl_class(void)
2510 for_each_possible_cpu(i)
2511 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask_dl, i),
2512 GFP_KERNEL, cpu_to_node(i));
2515 void dl_add_task_root_domain(struct task_struct *p)
2521 raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2523 raw_spin_unlock_irqrestore(&p->pi_lock, rf.flags);
2527 rq = __task_rq_lock(p, &rf);
2529 dl_b = &rq->rd->dl_bw;
2530 raw_spin_lock(&dl_b->lock);
2532 __dl_add(dl_b, p->dl.dl_bw, cpumask_weight(rq->rd->span));
2534 raw_spin_unlock(&dl_b->lock);
2536 task_rq_unlock(rq, p, &rf);
2539 void dl_clear_root_domain(struct root_domain *rd)
2541 unsigned long flags;
2543 raw_spin_lock_irqsave(&rd->dl_bw.lock, flags);
2544 rd->dl_bw.total_bw = 0;
2545 raw_spin_unlock_irqrestore(&rd->dl_bw.lock, flags);
2548 #endif /* CONFIG_SMP */
2550 static void switched_from_dl(struct rq *rq, struct task_struct *p)
2553 * task_non_contending() can start the "inactive timer" (if the 0-lag
2554 * time is in the future). If the task switches back to dl before
2555 * the "inactive timer" fires, it can continue to consume its current
2556 * runtime using its current deadline. If it stays outside of
2557 * SCHED_DEADLINE until the 0-lag time passes, inactive_task_timer()
2558 * will reset the task parameters.
2560 if (task_on_rq_queued(p) && p->dl.dl_runtime)
2561 task_non_contending(p);
2564 * In case a task is setscheduled out from SCHED_DEADLINE we need to
2565 * keep track of that on its cpuset (for correct bandwidth tracking).
2569 if (!task_on_rq_queued(p)) {
2571 * Inactive timer is armed. However, p is leaving DEADLINE and
2572 * might migrate away from this rq while continuing to run on
2573 * some other class. We need to remove its contribution from
2574 * this rq running_bw now, or sub_rq_bw (below) will complain.
2576 if (p->dl.dl_non_contending)
2577 sub_running_bw(&p->dl, &rq->dl);
2578 sub_rq_bw(&p->dl, &rq->dl);
2582 * We cannot use inactive_task_timer() to invoke sub_running_bw()
2583 * at the 0-lag time, because the task could have been migrated
2584 * while SCHED_OTHER in the meanwhile.
2586 if (p->dl.dl_non_contending)
2587 p->dl.dl_non_contending = 0;
2590 * Since this might be the only -deadline task on the rq,
2591 * this is the right place to try to pull some other one
2592 * from an overloaded CPU, if any.
2594 if (!task_on_rq_queued(p) || rq->dl.dl_nr_running)
2597 deadline_queue_pull_task(rq);
2601 * When switching to -deadline, we may overload the rq, then
2602 * we try to push someone off, if possible.
2604 static void switched_to_dl(struct rq *rq, struct task_struct *p)
2606 if (hrtimer_try_to_cancel(&p->dl.inactive_timer) == 1)
2610 * In case a task is setscheduled to SCHED_DEADLINE we need to keep
2611 * track of that on its cpuset (for correct bandwidth tracking).
2615 /* If p is not queued we will update its parameters at next wakeup. */
2616 if (!task_on_rq_queued(p)) {
2617 add_rq_bw(&p->dl, &rq->dl);
2622 if (rq->curr != p) {
2624 if (p->nr_cpus_allowed > 1 && rq->dl.overloaded)
2625 deadline_queue_push_tasks(rq);
2627 if (dl_task(rq->curr))
2628 wakeup_preempt_dl(rq, p, 0);
2632 update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 0);
2637 * If the scheduling parameters of a -deadline task changed,
2638 * a push or pull operation might be needed.
2640 static void prio_changed_dl(struct rq *rq, struct task_struct *p,
2643 if (!task_on_rq_queued(p))
2648 * This might be too much, but unfortunately
2649 * we don't have the old deadline value, and
2650 * we can't argue if the task is increasing
2651 * or lowering its prio, so...
2653 if (!rq->dl.overloaded)
2654 deadline_queue_pull_task(rq);
2656 if (task_current(rq, p)) {
2658 * If we now have a earlier deadline task than p,
2659 * then reschedule, provided p is still on this
2662 if (dl_time_before(rq->dl.earliest_dl.curr, p->dl.deadline))
2666 * Current may not be deadline in case p was throttled but we
2667 * have just replenished it (e.g. rt_mutex_setprio()).
2669 * Otherwise, if p was given an earlier deadline, reschedule.
2671 if (!dl_task(rq->curr) ||
2672 dl_time_before(p->dl.deadline, rq->curr->dl.deadline))
2677 * We don't know if p has a earlier or later deadline, so let's blindly
2678 * set a (maybe not needed) rescheduling point.
2684 #ifdef CONFIG_SCHED_CORE
2685 static int task_is_throttled_dl(struct task_struct *p, int cpu)
2687 return p->dl.dl_throttled;
2691 DEFINE_SCHED_CLASS(dl) = {
2693 .enqueue_task = enqueue_task_dl,
2694 .dequeue_task = dequeue_task_dl,
2695 .yield_task = yield_task_dl,
2697 .wakeup_preempt = wakeup_preempt_dl,
2699 .pick_next_task = pick_next_task_dl,
2700 .put_prev_task = put_prev_task_dl,
2701 .set_next_task = set_next_task_dl,
2704 .balance = balance_dl,
2705 .pick_task = pick_task_dl,
2706 .select_task_rq = select_task_rq_dl,
2707 .migrate_task_rq = migrate_task_rq_dl,
2708 .set_cpus_allowed = set_cpus_allowed_dl,
2709 .rq_online = rq_online_dl,
2710 .rq_offline = rq_offline_dl,
2711 .task_woken = task_woken_dl,
2712 .find_lock_rq = find_lock_later_rq,
2715 .task_tick = task_tick_dl,
2716 .task_fork = task_fork_dl,
2718 .prio_changed = prio_changed_dl,
2719 .switched_from = switched_from_dl,
2720 .switched_to = switched_to_dl,
2722 .update_curr = update_curr_dl,
2723 #ifdef CONFIG_SCHED_CORE
2724 .task_is_throttled = task_is_throttled_dl,
2728 /* Used for dl_bw check and update, used under sched_rt_handler()::mutex */
2729 static u64 dl_generation;
2731 int sched_dl_global_validate(void)
2733 u64 runtime = global_rt_runtime();
2734 u64 period = global_rt_period();
2735 u64 new_bw = to_ratio(period, runtime);
2736 u64 gen = ++dl_generation;
2738 int cpu, cpus, ret = 0;
2739 unsigned long flags;
2742 * Here we want to check the bandwidth not being set to some
2743 * value smaller than the currently allocated bandwidth in
2744 * any of the root_domains.
2746 for_each_possible_cpu(cpu) {
2747 rcu_read_lock_sched();
2749 if (dl_bw_visited(cpu, gen))
2752 dl_b = dl_bw_of(cpu);
2753 cpus = dl_bw_cpus(cpu);
2755 raw_spin_lock_irqsave(&dl_b->lock, flags);
2756 if (new_bw * cpus < dl_b->total_bw)
2758 raw_spin_unlock_irqrestore(&dl_b->lock, flags);
2761 rcu_read_unlock_sched();
2770 static void init_dl_rq_bw_ratio(struct dl_rq *dl_rq)
2772 if (global_rt_runtime() == RUNTIME_INF) {
2773 dl_rq->bw_ratio = 1 << RATIO_SHIFT;
2774 dl_rq->max_bw = dl_rq->extra_bw = 1 << BW_SHIFT;
2776 dl_rq->bw_ratio = to_ratio(global_rt_runtime(),
2777 global_rt_period()) >> (BW_SHIFT - RATIO_SHIFT);
2778 dl_rq->max_bw = dl_rq->extra_bw =
2779 to_ratio(global_rt_period(), global_rt_runtime());
2783 void sched_dl_do_global(void)
2786 u64 gen = ++dl_generation;
2789 unsigned long flags;
2791 if (global_rt_runtime() != RUNTIME_INF)
2792 new_bw = to_ratio(global_rt_period(), global_rt_runtime());
2794 for_each_possible_cpu(cpu) {
2795 rcu_read_lock_sched();
2797 if (dl_bw_visited(cpu, gen)) {
2798 rcu_read_unlock_sched();
2802 dl_b = dl_bw_of(cpu);
2804 raw_spin_lock_irqsave(&dl_b->lock, flags);
2806 raw_spin_unlock_irqrestore(&dl_b->lock, flags);
2808 rcu_read_unlock_sched();
2809 init_dl_rq_bw_ratio(&cpu_rq(cpu)->dl);
2814 * We must be sure that accepting a new task (or allowing changing the
2815 * parameters of an existing one) is consistent with the bandwidth
2816 * constraints. If yes, this function also accordingly updates the currently
2817 * allocated bandwidth to reflect the new situation.
2819 * This function is called while holding p's rq->lock.
2821 int sched_dl_overflow(struct task_struct *p, int policy,
2822 const struct sched_attr *attr)
2824 u64 period = attr->sched_period ?: attr->sched_deadline;
2825 u64 runtime = attr->sched_runtime;
2826 u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0;
2827 int cpus, err = -1, cpu = task_cpu(p);
2828 struct dl_bw *dl_b = dl_bw_of(cpu);
2831 if (attr->sched_flags & SCHED_FLAG_SUGOV)
2834 /* !deadline task may carry old deadline bandwidth */
2835 if (new_bw == p->dl.dl_bw && task_has_dl_policy(p))
2839 * Either if a task, enters, leave, or stays -deadline but changes
2840 * its parameters, we may need to update accordingly the total
2841 * allocated bandwidth of the container.
2843 raw_spin_lock(&dl_b->lock);
2844 cpus = dl_bw_cpus(cpu);
2845 cap = dl_bw_capacity(cpu);
2847 if (dl_policy(policy) && !task_has_dl_policy(p) &&
2848 !__dl_overflow(dl_b, cap, 0, new_bw)) {
2849 if (hrtimer_active(&p->dl.inactive_timer))
2850 __dl_sub(dl_b, p->dl.dl_bw, cpus);
2851 __dl_add(dl_b, new_bw, cpus);
2853 } else if (dl_policy(policy) && task_has_dl_policy(p) &&
2854 !__dl_overflow(dl_b, cap, p->dl.dl_bw, new_bw)) {
2856 * XXX this is slightly incorrect: when the task
2857 * utilization decreases, we should delay the total
2858 * utilization change until the task's 0-lag point.
2859 * But this would require to set the task's "inactive
2860 * timer" when the task is not inactive.
2862 __dl_sub(dl_b, p->dl.dl_bw, cpus);
2863 __dl_add(dl_b, new_bw, cpus);
2864 dl_change_utilization(p, new_bw);
2866 } else if (!dl_policy(policy) && task_has_dl_policy(p)) {
2868 * Do not decrease the total deadline utilization here,
2869 * switched_from_dl() will take care to do it at the correct
2874 raw_spin_unlock(&dl_b->lock);
2880 * This function initializes the sched_dl_entity of a newly becoming
2881 * SCHED_DEADLINE task.
2883 * Only the static values are considered here, the actual runtime and the
2884 * absolute deadline will be properly calculated when the task is enqueued
2885 * for the first time with its new policy.
2887 void __setparam_dl(struct task_struct *p, const struct sched_attr *attr)
2889 struct sched_dl_entity *dl_se = &p->dl;
2891 dl_se->dl_runtime = attr->sched_runtime;
2892 dl_se->dl_deadline = attr->sched_deadline;
2893 dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
2894 dl_se->flags = attr->sched_flags & SCHED_DL_FLAGS;
2895 dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
2896 dl_se->dl_density = to_ratio(dl_se->dl_deadline, dl_se->dl_runtime);
2899 void __getparam_dl(struct task_struct *p, struct sched_attr *attr)
2901 struct sched_dl_entity *dl_se = &p->dl;
2903 attr->sched_priority = p->rt_priority;
2904 attr->sched_runtime = dl_se->dl_runtime;
2905 attr->sched_deadline = dl_se->dl_deadline;
2906 attr->sched_period = dl_se->dl_period;
2907 attr->sched_flags &= ~SCHED_DL_FLAGS;
2908 attr->sched_flags |= dl_se->flags;
2912 * This function validates the new parameters of a -deadline task.
2913 * We ask for the deadline not being zero, and greater or equal
2914 * than the runtime, as well as the period of being zero or
2915 * greater than deadline. Furthermore, we have to be sure that
2916 * user parameters are above the internal resolution of 1us (we
2917 * check sched_runtime only since it is always the smaller one) and
2918 * below 2^63 ns (we have to check both sched_deadline and
2919 * sched_period, as the latter can be zero).
2921 bool __checkparam_dl(const struct sched_attr *attr)
2923 u64 period, max, min;
2925 /* special dl tasks don't actually use any parameter */
2926 if (attr->sched_flags & SCHED_FLAG_SUGOV)
2930 if (attr->sched_deadline == 0)
2934 * Since we truncate DL_SCALE bits, make sure we're at least
2937 if (attr->sched_runtime < (1ULL << DL_SCALE))
2941 * Since we use the MSB for wrap-around and sign issues, make
2942 * sure it's not set (mind that period can be equal to zero).
2944 if (attr->sched_deadline & (1ULL << 63) ||
2945 attr->sched_period & (1ULL << 63))
2948 period = attr->sched_period;
2950 period = attr->sched_deadline;
2952 /* runtime <= deadline <= period (if period != 0) */
2953 if (period < attr->sched_deadline ||
2954 attr->sched_deadline < attr->sched_runtime)
2957 max = (u64)READ_ONCE(sysctl_sched_dl_period_max) * NSEC_PER_USEC;
2958 min = (u64)READ_ONCE(sysctl_sched_dl_period_min) * NSEC_PER_USEC;
2960 if (period < min || period > max)
2967 * This function clears the sched_dl_entity static params.
2969 void __dl_clear_params(struct task_struct *p)
2971 struct sched_dl_entity *dl_se = &p->dl;
2973 dl_se->dl_runtime = 0;
2974 dl_se->dl_deadline = 0;
2975 dl_se->dl_period = 0;
2978 dl_se->dl_density = 0;
2980 dl_se->dl_throttled = 0;
2981 dl_se->dl_yielded = 0;
2982 dl_se->dl_non_contending = 0;
2983 dl_se->dl_overrun = 0;
2985 #ifdef CONFIG_RT_MUTEXES
2986 dl_se->pi_se = dl_se;
2990 bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr)
2992 struct sched_dl_entity *dl_se = &p->dl;
2994 if (dl_se->dl_runtime != attr->sched_runtime ||
2995 dl_se->dl_deadline != attr->sched_deadline ||
2996 dl_se->dl_period != attr->sched_period ||
2997 dl_se->flags != (attr->sched_flags & SCHED_DL_FLAGS))
3004 int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur,
3005 const struct cpumask *trial)
3007 unsigned long flags, cap;
3008 struct dl_bw *cur_dl_b;
3011 rcu_read_lock_sched();
3012 cur_dl_b = dl_bw_of(cpumask_any(cur));
3013 cap = __dl_bw_capacity(trial);
3014 raw_spin_lock_irqsave(&cur_dl_b->lock, flags);
3015 if (__dl_overflow(cur_dl_b, cap, 0, 0))
3017 raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags);
3018 rcu_read_unlock_sched();
3023 enum dl_bw_request {
3024 dl_bw_req_check_overflow = 0,
3029 static int dl_bw_manage(enum dl_bw_request req, int cpu, u64 dl_bw)
3031 unsigned long flags;
3035 rcu_read_lock_sched();
3036 dl_b = dl_bw_of(cpu);
3037 raw_spin_lock_irqsave(&dl_b->lock, flags);
3039 if (req == dl_bw_req_free) {
3040 __dl_sub(dl_b, dl_bw, dl_bw_cpus(cpu));
3042 unsigned long cap = dl_bw_capacity(cpu);
3044 overflow = __dl_overflow(dl_b, cap, 0, dl_bw);
3046 if (req == dl_bw_req_alloc && !overflow) {
3048 * We reserve space in the destination
3049 * root_domain, as we can't fail after this point.
3050 * We will free resources in the source root_domain
3051 * later on (see set_cpus_allowed_dl()).
3053 __dl_add(dl_b, dl_bw, dl_bw_cpus(cpu));
3057 raw_spin_unlock_irqrestore(&dl_b->lock, flags);
3058 rcu_read_unlock_sched();
3060 return overflow ? -EBUSY : 0;
3063 int dl_bw_check_overflow(int cpu)
3065 return dl_bw_manage(dl_bw_req_check_overflow, cpu, 0);
3068 int dl_bw_alloc(int cpu, u64 dl_bw)
3070 return dl_bw_manage(dl_bw_req_alloc, cpu, dl_bw);
3073 void dl_bw_free(int cpu, u64 dl_bw)
3075 dl_bw_manage(dl_bw_req_free, cpu, dl_bw);
3079 #ifdef CONFIG_SCHED_DEBUG
3080 void print_dl_stats(struct seq_file *m, int cpu)
3082 print_dl_rq(m, cpu, &cpu_rq(cpu)->dl);
3084 #endif /* CONFIG_SCHED_DEBUG */