1 // SPDX-License-Identifier: GPL-2.0
3 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
9 #include <linux/slab.h>
10 #include <linux/irq_work.h>
12 int sched_rr_timeslice
= RR_TIMESLICE
;
13 int sysctl_sched_rr_timeslice
= (MSEC_PER_SEC
/ HZ
) * RR_TIMESLICE
;
15 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
);
17 struct rt_bandwidth def_rt_bandwidth
;
19 static enum hrtimer_restart
sched_rt_period_timer(struct hrtimer
*timer
)
21 struct rt_bandwidth
*rt_b
=
22 container_of(timer
, struct rt_bandwidth
, rt_period_timer
);
26 raw_spin_lock(&rt_b
->rt_runtime_lock
);
28 overrun
= hrtimer_forward_now(timer
, rt_b
->rt_period
);
32 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
33 idle
= do_sched_rt_period_timer(rt_b
, overrun
);
34 raw_spin_lock(&rt_b
->rt_runtime_lock
);
37 rt_b
->rt_period_active
= 0;
38 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
40 return idle
? HRTIMER_NORESTART
: HRTIMER_RESTART
;
43 void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
)
45 rt_b
->rt_period
= ns_to_ktime(period
);
46 rt_b
->rt_runtime
= runtime
;
48 raw_spin_lock_init(&rt_b
->rt_runtime_lock
);
50 hrtimer_init(&rt_b
->rt_period_timer
,
51 CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
52 rt_b
->rt_period_timer
.function
= sched_rt_period_timer
;
55 static void start_rt_bandwidth(struct rt_bandwidth
*rt_b
)
57 if (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
)
60 raw_spin_lock(&rt_b
->rt_runtime_lock
);
61 if (!rt_b
->rt_period_active
) {
62 rt_b
->rt_period_active
= 1;
64 * SCHED_DEADLINE updates the bandwidth, as a run away
65 * RT task with a DL task could hog a CPU. But DL does
66 * not reset the period. If a deadline task was running
67 * without an RT task running, it can cause RT tasks to
68 * throttle when they start up. Kick the timer right away
69 * to update the period.
71 hrtimer_forward_now(&rt_b
->rt_period_timer
, ns_to_ktime(0));
72 hrtimer_start_expires(&rt_b
->rt_period_timer
, HRTIMER_MODE_ABS_PINNED
);
74 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
77 #if defined(CONFIG_SMP) && defined(HAVE_RT_PUSH_IPI)
78 static void push_irq_work_func(struct irq_work
*work
);
81 void init_rt_rq(struct rt_rq
*rt_rq
)
83 struct rt_prio_array
*array
;
86 array
= &rt_rq
->active
;
87 for (i
= 0; i
< MAX_RT_PRIO
; i
++) {
88 INIT_LIST_HEAD(array
->queue
+ i
);
89 __clear_bit(i
, array
->bitmap
);
91 /* delimiter for bitsearch: */
92 __set_bit(MAX_RT_PRIO
, array
->bitmap
);
94 #if defined CONFIG_SMP
95 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
96 rt_rq
->highest_prio
.next
= MAX_RT_PRIO
;
97 rt_rq
->rt_nr_migratory
= 0;
98 rt_rq
->overloaded
= 0;
99 plist_head_init(&rt_rq
->pushable_tasks
);
101 #ifdef HAVE_RT_PUSH_IPI
102 rt_rq
->push_flags
= 0;
103 rt_rq
->push_cpu
= nr_cpu_ids
;
104 raw_spin_lock_init(&rt_rq
->push_lock
);
105 init_irq_work(&rt_rq
->push_work
, push_irq_work_func
);
107 #endif /* CONFIG_SMP */
108 /* We start is dequeued state, because no RT tasks are queued */
109 rt_rq
->rt_queued
= 0;
112 rt_rq
->rt_throttled
= 0;
113 rt_rq
->rt_runtime
= 0;
114 raw_spin_lock_init(&rt_rq
->rt_runtime_lock
);
117 #ifdef CONFIG_RT_GROUP_SCHED
118 static void destroy_rt_bandwidth(struct rt_bandwidth
*rt_b
)
120 hrtimer_cancel(&rt_b
->rt_period_timer
);
123 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
125 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
127 #ifdef CONFIG_SCHED_DEBUG
128 WARN_ON_ONCE(!rt_entity_is_task(rt_se
));
130 return container_of(rt_se
, struct task_struct
, rt
);
133 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
138 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
143 static inline struct rq
*rq_of_rt_se(struct sched_rt_entity
*rt_se
)
145 struct rt_rq
*rt_rq
= rt_se
->rt_rq
;
150 void free_rt_sched_group(struct task_group
*tg
)
155 destroy_rt_bandwidth(&tg
->rt_bandwidth
);
157 for_each_possible_cpu(i
) {
168 void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
169 struct sched_rt_entity
*rt_se
, int cpu
,
170 struct sched_rt_entity
*parent
)
172 struct rq
*rq
= cpu_rq(cpu
);
174 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
175 rt_rq
->rt_nr_boosted
= 0;
179 tg
->rt_rq
[cpu
] = rt_rq
;
180 tg
->rt_se
[cpu
] = rt_se
;
186 rt_se
->rt_rq
= &rq
->rt
;
188 rt_se
->rt_rq
= parent
->my_q
;
191 rt_se
->parent
= parent
;
192 INIT_LIST_HEAD(&rt_se
->run_list
);
195 int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
)
198 struct sched_rt_entity
*rt_se
;
201 tg
->rt_rq
= kzalloc(sizeof(rt_rq
) * nr_cpu_ids
, GFP_KERNEL
);
204 tg
->rt_se
= kzalloc(sizeof(rt_se
) * nr_cpu_ids
, GFP_KERNEL
);
208 init_rt_bandwidth(&tg
->rt_bandwidth
,
209 ktime_to_ns(def_rt_bandwidth
.rt_period
), 0);
211 for_each_possible_cpu(i
) {
212 rt_rq
= kzalloc_node(sizeof(struct rt_rq
),
213 GFP_KERNEL
, cpu_to_node(i
));
217 rt_se
= kzalloc_node(sizeof(struct sched_rt_entity
),
218 GFP_KERNEL
, cpu_to_node(i
));
223 rt_rq
->rt_runtime
= tg
->rt_bandwidth
.rt_runtime
;
224 init_tg_rt_entry(tg
, rt_rq
, rt_se
, i
, parent
->rt_se
[i
]);
235 #else /* CONFIG_RT_GROUP_SCHED */
237 #define rt_entity_is_task(rt_se) (1)
239 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
241 return container_of(rt_se
, struct task_struct
, rt
);
244 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
246 return container_of(rt_rq
, struct rq
, rt
);
249 static inline struct rq
*rq_of_rt_se(struct sched_rt_entity
*rt_se
)
251 struct task_struct
*p
= rt_task_of(rt_se
);
256 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
258 struct rq
*rq
= rq_of_rt_se(rt_se
);
263 void free_rt_sched_group(struct task_group
*tg
) { }
265 int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
)
269 #endif /* CONFIG_RT_GROUP_SCHED */
273 static void pull_rt_task(struct rq
*this_rq
);
275 static inline bool need_pull_rt_task(struct rq
*rq
, struct task_struct
*prev
)
277 /* Try to pull RT tasks here if we lower this rq's prio */
278 return rq
->rt
.highest_prio
.curr
> prev
->prio
;
281 static inline int rt_overloaded(struct rq
*rq
)
283 return atomic_read(&rq
->rd
->rto_count
);
286 static inline void rt_set_overload(struct rq
*rq
)
291 cpumask_set_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
293 * Make sure the mask is visible before we set
294 * the overload count. That is checked to determine
295 * if we should look at the mask. It would be a shame
296 * if we looked at the mask, but the mask was not
299 * Matched by the barrier in pull_rt_task().
302 atomic_inc(&rq
->rd
->rto_count
);
305 static inline void rt_clear_overload(struct rq
*rq
)
310 /* the order here really doesn't matter */
311 atomic_dec(&rq
->rd
->rto_count
);
312 cpumask_clear_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
315 static void update_rt_migration(struct rt_rq
*rt_rq
)
317 if (rt_rq
->rt_nr_migratory
&& rt_rq
->rt_nr_total
> 1) {
318 if (!rt_rq
->overloaded
) {
319 rt_set_overload(rq_of_rt_rq(rt_rq
));
320 rt_rq
->overloaded
= 1;
322 } else if (rt_rq
->overloaded
) {
323 rt_clear_overload(rq_of_rt_rq(rt_rq
));
324 rt_rq
->overloaded
= 0;
328 static void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
330 struct task_struct
*p
;
332 if (!rt_entity_is_task(rt_se
))
335 p
= rt_task_of(rt_se
);
336 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
338 rt_rq
->rt_nr_total
++;
339 if (p
->nr_cpus_allowed
> 1)
340 rt_rq
->rt_nr_migratory
++;
342 update_rt_migration(rt_rq
);
345 static void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
347 struct task_struct
*p
;
349 if (!rt_entity_is_task(rt_se
))
352 p
= rt_task_of(rt_se
);
353 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
355 rt_rq
->rt_nr_total
--;
356 if (p
->nr_cpus_allowed
> 1)
357 rt_rq
->rt_nr_migratory
--;
359 update_rt_migration(rt_rq
);
362 static inline int has_pushable_tasks(struct rq
*rq
)
364 return !plist_head_empty(&rq
->rt
.pushable_tasks
);
367 static DEFINE_PER_CPU(struct callback_head
, rt_push_head
);
368 static DEFINE_PER_CPU(struct callback_head
, rt_pull_head
);
370 static void push_rt_tasks(struct rq
*);
371 static void pull_rt_task(struct rq
*);
373 static inline void queue_push_tasks(struct rq
*rq
)
375 if (!has_pushable_tasks(rq
))
378 queue_balance_callback(rq
, &per_cpu(rt_push_head
, rq
->cpu
), push_rt_tasks
);
381 static inline void queue_pull_task(struct rq
*rq
)
383 queue_balance_callback(rq
, &per_cpu(rt_pull_head
, rq
->cpu
), pull_rt_task
);
386 static void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
388 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
389 plist_node_init(&p
->pushable_tasks
, p
->prio
);
390 plist_add(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
392 /* Update the highest prio pushable task */
393 if (p
->prio
< rq
->rt
.highest_prio
.next
)
394 rq
->rt
.highest_prio
.next
= p
->prio
;
397 static void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
399 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
401 /* Update the new highest prio pushable task */
402 if (has_pushable_tasks(rq
)) {
403 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
404 struct task_struct
, pushable_tasks
);
405 rq
->rt
.highest_prio
.next
= p
->prio
;
407 rq
->rt
.highest_prio
.next
= MAX_RT_PRIO
;
412 static inline void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
416 static inline void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
421 void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
426 void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
430 static inline bool need_pull_rt_task(struct rq
*rq
, struct task_struct
*prev
)
435 static inline void pull_rt_task(struct rq
*this_rq
)
439 static inline void queue_push_tasks(struct rq
*rq
)
442 #endif /* CONFIG_SMP */
444 static void enqueue_top_rt_rq(struct rt_rq
*rt_rq
);
445 static void dequeue_top_rt_rq(struct rt_rq
*rt_rq
);
447 static inline int on_rt_rq(struct sched_rt_entity
*rt_se
)
452 #ifdef CONFIG_RT_GROUP_SCHED
454 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
459 return rt_rq
->rt_runtime
;
462 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
464 return ktime_to_ns(rt_rq
->tg
->rt_bandwidth
.rt_period
);
467 typedef struct task_group
*rt_rq_iter_t
;
469 static inline struct task_group
*next_task_group(struct task_group
*tg
)
472 tg
= list_entry_rcu(tg
->list
.next
,
473 typeof(struct task_group
), list
);
474 } while (&tg
->list
!= &task_groups
&& task_group_is_autogroup(tg
));
476 if (&tg
->list
== &task_groups
)
482 #define for_each_rt_rq(rt_rq, iter, rq) \
483 for (iter = container_of(&task_groups, typeof(*iter), list); \
484 (iter = next_task_group(iter)) && \
485 (rt_rq = iter->rt_rq[cpu_of(rq)]);)
487 #define for_each_sched_rt_entity(rt_se) \
488 for (; rt_se; rt_se = rt_se->parent)
490 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
495 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, unsigned int flags
);
496 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
, unsigned int flags
);
498 static void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
500 struct task_struct
*curr
= rq_of_rt_rq(rt_rq
)->curr
;
501 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
502 struct sched_rt_entity
*rt_se
;
504 int cpu
= cpu_of(rq
);
506 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
508 if (rt_rq
->rt_nr_running
) {
510 enqueue_top_rt_rq(rt_rq
);
511 else if (!on_rt_rq(rt_se
))
512 enqueue_rt_entity(rt_se
, 0);
514 if (rt_rq
->highest_prio
.curr
< curr
->prio
)
519 static void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
521 struct sched_rt_entity
*rt_se
;
522 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
524 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
527 dequeue_top_rt_rq(rt_rq
);
528 else if (on_rt_rq(rt_se
))
529 dequeue_rt_entity(rt_se
, 0);
532 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
534 return rt_rq
->rt_throttled
&& !rt_rq
->rt_nr_boosted
;
537 static int rt_se_boosted(struct sched_rt_entity
*rt_se
)
539 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
540 struct task_struct
*p
;
543 return !!rt_rq
->rt_nr_boosted
;
545 p
= rt_task_of(rt_se
);
546 return p
->prio
!= p
->normal_prio
;
550 static inline const struct cpumask
*sched_rt_period_mask(void)
552 return this_rq()->rd
->span
;
555 static inline const struct cpumask
*sched_rt_period_mask(void)
557 return cpu_online_mask
;
562 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
564 return container_of(rt_b
, struct task_group
, rt_bandwidth
)->rt_rq
[cpu
];
567 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
569 return &rt_rq
->tg
->rt_bandwidth
;
572 #else /* !CONFIG_RT_GROUP_SCHED */
574 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
576 return rt_rq
->rt_runtime
;
579 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
581 return ktime_to_ns(def_rt_bandwidth
.rt_period
);
584 typedef struct rt_rq
*rt_rq_iter_t
;
586 #define for_each_rt_rq(rt_rq, iter, rq) \
587 for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
589 #define for_each_sched_rt_entity(rt_se) \
590 for (; rt_se; rt_se = NULL)
592 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
597 static inline void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
599 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
601 if (!rt_rq
->rt_nr_running
)
604 enqueue_top_rt_rq(rt_rq
);
608 static inline void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
610 dequeue_top_rt_rq(rt_rq
);
613 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
615 return rt_rq
->rt_throttled
;
618 static inline const struct cpumask
*sched_rt_period_mask(void)
620 return cpu_online_mask
;
624 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
626 return &cpu_rq(cpu
)->rt
;
629 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
631 return &def_rt_bandwidth
;
634 #endif /* CONFIG_RT_GROUP_SCHED */
636 bool sched_rt_bandwidth_account(struct rt_rq
*rt_rq
)
638 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
640 return (hrtimer_active(&rt_b
->rt_period_timer
) ||
641 rt_rq
->rt_time
< rt_b
->rt_runtime
);
646 * We ran out of runtime, see if we can borrow some from our neighbours.
648 static void do_balance_runtime(struct rt_rq
*rt_rq
)
650 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
651 struct root_domain
*rd
= rq_of_rt_rq(rt_rq
)->rd
;
655 weight
= cpumask_weight(rd
->span
);
657 raw_spin_lock(&rt_b
->rt_runtime_lock
);
658 rt_period
= ktime_to_ns(rt_b
->rt_period
);
659 for_each_cpu(i
, rd
->span
) {
660 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
666 raw_spin_lock(&iter
->rt_runtime_lock
);
668 * Either all rqs have inf runtime and there's nothing to steal
669 * or __disable_runtime() below sets a specific rq to inf to
670 * indicate its been disabled and disalow stealing.
672 if (iter
->rt_runtime
== RUNTIME_INF
)
676 * From runqueues with spare time, take 1/n part of their
677 * spare time, but no more than our period.
679 diff
= iter
->rt_runtime
- iter
->rt_time
;
681 diff
= div_u64((u64
)diff
, weight
);
682 if (rt_rq
->rt_runtime
+ diff
> rt_period
)
683 diff
= rt_period
- rt_rq
->rt_runtime
;
684 iter
->rt_runtime
-= diff
;
685 rt_rq
->rt_runtime
+= diff
;
686 if (rt_rq
->rt_runtime
== rt_period
) {
687 raw_spin_unlock(&iter
->rt_runtime_lock
);
692 raw_spin_unlock(&iter
->rt_runtime_lock
);
694 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
698 * Ensure this RQ takes back all the runtime it lend to its neighbours.
700 static void __disable_runtime(struct rq
*rq
)
702 struct root_domain
*rd
= rq
->rd
;
706 if (unlikely(!scheduler_running
))
709 for_each_rt_rq(rt_rq
, iter
, rq
) {
710 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
714 raw_spin_lock(&rt_b
->rt_runtime_lock
);
715 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
717 * Either we're all inf and nobody needs to borrow, or we're
718 * already disabled and thus have nothing to do, or we have
719 * exactly the right amount of runtime to take out.
721 if (rt_rq
->rt_runtime
== RUNTIME_INF
||
722 rt_rq
->rt_runtime
== rt_b
->rt_runtime
)
724 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
727 * Calculate the difference between what we started out with
728 * and what we current have, that's the amount of runtime
729 * we lend and now have to reclaim.
731 want
= rt_b
->rt_runtime
- rt_rq
->rt_runtime
;
734 * Greedy reclaim, take back as much as we can.
736 for_each_cpu(i
, rd
->span
) {
737 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
741 * Can't reclaim from ourselves or disabled runqueues.
743 if (iter
== rt_rq
|| iter
->rt_runtime
== RUNTIME_INF
)
746 raw_spin_lock(&iter
->rt_runtime_lock
);
748 diff
= min_t(s64
, iter
->rt_runtime
, want
);
749 iter
->rt_runtime
-= diff
;
752 iter
->rt_runtime
-= want
;
755 raw_spin_unlock(&iter
->rt_runtime_lock
);
761 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
763 * We cannot be left wanting - that would mean some runtime
764 * leaked out of the system.
769 * Disable all the borrow logic by pretending we have inf
770 * runtime - in which case borrowing doesn't make sense.
772 rt_rq
->rt_runtime
= RUNTIME_INF
;
773 rt_rq
->rt_throttled
= 0;
774 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
775 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
777 /* Make rt_rq available for pick_next_task() */
778 sched_rt_rq_enqueue(rt_rq
);
782 static void __enable_runtime(struct rq
*rq
)
787 if (unlikely(!scheduler_running
))
791 * Reset each runqueue's bandwidth settings
793 for_each_rt_rq(rt_rq
, iter
, rq
) {
794 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
796 raw_spin_lock(&rt_b
->rt_runtime_lock
);
797 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
798 rt_rq
->rt_runtime
= rt_b
->rt_runtime
;
800 rt_rq
->rt_throttled
= 0;
801 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
802 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
806 static void balance_runtime(struct rt_rq
*rt_rq
)
808 if (!sched_feat(RT_RUNTIME_SHARE
))
811 if (rt_rq
->rt_time
> rt_rq
->rt_runtime
) {
812 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
813 do_balance_runtime(rt_rq
);
814 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
817 #else /* !CONFIG_SMP */
818 static inline void balance_runtime(struct rt_rq
*rt_rq
) {}
819 #endif /* CONFIG_SMP */
821 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
)
823 int i
, idle
= 1, throttled
= 0;
824 const struct cpumask
*span
;
826 span
= sched_rt_period_mask();
827 #ifdef CONFIG_RT_GROUP_SCHED
829 * FIXME: isolated CPUs should really leave the root task group,
830 * whether they are isolcpus or were isolated via cpusets, lest
831 * the timer run on a CPU which does not service all runqueues,
832 * potentially leaving other CPUs indefinitely throttled. If
833 * isolation is really required, the user will turn the throttle
834 * off to kill the perturbations it causes anyway. Meanwhile,
835 * this maintains functionality for boot and/or troubleshooting.
837 if (rt_b
== &root_task_group
.rt_bandwidth
)
838 span
= cpu_online_mask
;
840 for_each_cpu(i
, span
) {
842 struct rt_rq
*rt_rq
= sched_rt_period_rt_rq(rt_b
, i
);
843 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
847 * When span == cpu_online_mask, taking each rq->lock
848 * can be time-consuming. Try to avoid it when possible.
850 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
851 skip
= !rt_rq
->rt_time
&& !rt_rq
->rt_nr_running
;
852 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
856 raw_spin_lock(&rq
->lock
);
857 if (rt_rq
->rt_time
) {
860 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
861 if (rt_rq
->rt_throttled
)
862 balance_runtime(rt_rq
);
863 runtime
= rt_rq
->rt_runtime
;
864 rt_rq
->rt_time
-= min(rt_rq
->rt_time
, overrun
*runtime
);
865 if (rt_rq
->rt_throttled
&& rt_rq
->rt_time
< runtime
) {
866 rt_rq
->rt_throttled
= 0;
870 * When we're idle and a woken (rt) task is
871 * throttled check_preempt_curr() will set
872 * skip_update and the time between the wakeup
873 * and this unthrottle will get accounted as
876 if (rt_rq
->rt_nr_running
&& rq
->curr
== rq
->idle
)
877 rq_clock_skip_update(rq
, false);
879 if (rt_rq
->rt_time
|| rt_rq
->rt_nr_running
)
881 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
882 } else if (rt_rq
->rt_nr_running
) {
884 if (!rt_rq_throttled(rt_rq
))
887 if (rt_rq
->rt_throttled
)
891 sched_rt_rq_enqueue(rt_rq
);
892 raw_spin_unlock(&rq
->lock
);
895 if (!throttled
&& (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
))
901 static inline int rt_se_prio(struct sched_rt_entity
*rt_se
)
903 #ifdef CONFIG_RT_GROUP_SCHED
904 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
907 return rt_rq
->highest_prio
.curr
;
910 return rt_task_of(rt_se
)->prio
;
913 static int sched_rt_runtime_exceeded(struct rt_rq
*rt_rq
)
915 u64 runtime
= sched_rt_runtime(rt_rq
);
917 if (rt_rq
->rt_throttled
)
918 return rt_rq_throttled(rt_rq
);
920 if (runtime
>= sched_rt_period(rt_rq
))
923 balance_runtime(rt_rq
);
924 runtime
= sched_rt_runtime(rt_rq
);
925 if (runtime
== RUNTIME_INF
)
928 if (rt_rq
->rt_time
> runtime
) {
929 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
932 * Don't actually throttle groups that have no runtime assigned
933 * but accrue some time due to boosting.
935 if (likely(rt_b
->rt_runtime
)) {
936 rt_rq
->rt_throttled
= 1;
937 printk_deferred_once("sched: RT throttling activated\n");
940 * In case we did anyway, make it go away,
941 * replenishment is a joke, since it will replenish us
947 if (rt_rq_throttled(rt_rq
)) {
948 sched_rt_rq_dequeue(rt_rq
);
957 * Update the current task's runtime statistics. Skip current tasks that
958 * are not in our scheduling class.
960 static void update_curr_rt(struct rq
*rq
)
962 struct task_struct
*curr
= rq
->curr
;
963 struct sched_rt_entity
*rt_se
= &curr
->rt
;
966 if (curr
->sched_class
!= &rt_sched_class
)
969 delta_exec
= rq_clock_task(rq
) - curr
->se
.exec_start
;
970 if (unlikely((s64
)delta_exec
<= 0))
973 /* Kick cpufreq (see the comment in kernel/sched/sched.h). */
974 cpufreq_update_util(rq
, SCHED_CPUFREQ_RT
);
976 schedstat_set(curr
->se
.statistics
.exec_max
,
977 max(curr
->se
.statistics
.exec_max
, delta_exec
));
979 curr
->se
.sum_exec_runtime
+= delta_exec
;
980 account_group_exec_runtime(curr
, delta_exec
);
982 curr
->se
.exec_start
= rq_clock_task(rq
);
983 cpuacct_charge(curr
, delta_exec
);
985 sched_rt_avg_update(rq
, delta_exec
);
987 if (!rt_bandwidth_enabled())
990 for_each_sched_rt_entity(rt_se
) {
991 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
993 if (sched_rt_runtime(rt_rq
) != RUNTIME_INF
) {
994 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
995 rt_rq
->rt_time
+= delta_exec
;
996 if (sched_rt_runtime_exceeded(rt_rq
))
998 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
1004 dequeue_top_rt_rq(struct rt_rq
*rt_rq
)
1006 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
1008 BUG_ON(&rq
->rt
!= rt_rq
);
1010 if (!rt_rq
->rt_queued
)
1013 BUG_ON(!rq
->nr_running
);
1015 sub_nr_running(rq
, rt_rq
->rt_nr_running
);
1016 rt_rq
->rt_queued
= 0;
1020 enqueue_top_rt_rq(struct rt_rq
*rt_rq
)
1022 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
1024 BUG_ON(&rq
->rt
!= rt_rq
);
1026 if (rt_rq
->rt_queued
)
1028 if (rt_rq_throttled(rt_rq
) || !rt_rq
->rt_nr_running
)
1031 add_nr_running(rq
, rt_rq
->rt_nr_running
);
1032 rt_rq
->rt_queued
= 1;
1035 #if defined CONFIG_SMP
1038 inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
1040 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
1042 #ifdef CONFIG_RT_GROUP_SCHED
1044 * Change rq's cpupri only if rt_rq is the top queue.
1046 if (&rq
->rt
!= rt_rq
)
1049 if (rq
->online
&& prio
< prev_prio
)
1050 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, prio
);
1054 dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
1056 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
1058 #ifdef CONFIG_RT_GROUP_SCHED
1060 * Change rq's cpupri only if rt_rq is the top queue.
1062 if (&rq
->rt
!= rt_rq
)
1065 if (rq
->online
&& rt_rq
->highest_prio
.curr
!= prev_prio
)
1066 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rt_rq
->highest_prio
.curr
);
1069 #else /* CONFIG_SMP */
1072 void inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
1074 void dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
1076 #endif /* CONFIG_SMP */
1078 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
1080 inc_rt_prio(struct rt_rq
*rt_rq
, int prio
)
1082 int prev_prio
= rt_rq
->highest_prio
.curr
;
1084 if (prio
< prev_prio
)
1085 rt_rq
->highest_prio
.curr
= prio
;
1087 inc_rt_prio_smp(rt_rq
, prio
, prev_prio
);
1091 dec_rt_prio(struct rt_rq
*rt_rq
, int prio
)
1093 int prev_prio
= rt_rq
->highest_prio
.curr
;
1095 if (rt_rq
->rt_nr_running
) {
1097 WARN_ON(prio
< prev_prio
);
1100 * This may have been our highest task, and therefore
1101 * we may have some recomputation to do
1103 if (prio
== prev_prio
) {
1104 struct rt_prio_array
*array
= &rt_rq
->active
;
1106 rt_rq
->highest_prio
.curr
=
1107 sched_find_first_bit(array
->bitmap
);
1111 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
1113 dec_rt_prio_smp(rt_rq
, prio
, prev_prio
);
1118 static inline void inc_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
1119 static inline void dec_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
1121 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1123 #ifdef CONFIG_RT_GROUP_SCHED
1126 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1128 if (rt_se_boosted(rt_se
))
1129 rt_rq
->rt_nr_boosted
++;
1132 start_rt_bandwidth(&rt_rq
->tg
->rt_bandwidth
);
1136 dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1138 if (rt_se_boosted(rt_se
))
1139 rt_rq
->rt_nr_boosted
--;
1141 WARN_ON(!rt_rq
->rt_nr_running
&& rt_rq
->rt_nr_boosted
);
1144 #else /* CONFIG_RT_GROUP_SCHED */
1147 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1149 start_rt_bandwidth(&def_rt_bandwidth
);
1153 void dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
) {}
1155 #endif /* CONFIG_RT_GROUP_SCHED */
1158 unsigned int rt_se_nr_running(struct sched_rt_entity
*rt_se
)
1160 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
1163 return group_rq
->rt_nr_running
;
1169 unsigned int rt_se_rr_nr_running(struct sched_rt_entity
*rt_se
)
1171 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
1172 struct task_struct
*tsk
;
1175 return group_rq
->rr_nr_running
;
1177 tsk
= rt_task_of(rt_se
);
1179 return (tsk
->policy
== SCHED_RR
) ? 1 : 0;
1183 void inc_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1185 int prio
= rt_se_prio(rt_se
);
1187 WARN_ON(!rt_prio(prio
));
1188 rt_rq
->rt_nr_running
+= rt_se_nr_running(rt_se
);
1189 rt_rq
->rr_nr_running
+= rt_se_rr_nr_running(rt_se
);
1191 inc_rt_prio(rt_rq
, prio
);
1192 inc_rt_migration(rt_se
, rt_rq
);
1193 inc_rt_group(rt_se
, rt_rq
);
1197 void dec_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1199 WARN_ON(!rt_prio(rt_se_prio(rt_se
)));
1200 WARN_ON(!rt_rq
->rt_nr_running
);
1201 rt_rq
->rt_nr_running
-= rt_se_nr_running(rt_se
);
1202 rt_rq
->rr_nr_running
-= rt_se_rr_nr_running(rt_se
);
1204 dec_rt_prio(rt_rq
, rt_se_prio(rt_se
));
1205 dec_rt_migration(rt_se
, rt_rq
);
1206 dec_rt_group(rt_se
, rt_rq
);
1210 * Change rt_se->run_list location unless SAVE && !MOVE
1212 * assumes ENQUEUE/DEQUEUE flags match
1214 static inline bool move_entity(unsigned int flags
)
1216 if ((flags
& (DEQUEUE_SAVE
| DEQUEUE_MOVE
)) == DEQUEUE_SAVE
)
1222 static void __delist_rt_entity(struct sched_rt_entity
*rt_se
, struct rt_prio_array
*array
)
1224 list_del_init(&rt_se
->run_list
);
1226 if (list_empty(array
->queue
+ rt_se_prio(rt_se
)))
1227 __clear_bit(rt_se_prio(rt_se
), array
->bitmap
);
1232 static void __enqueue_rt_entity(struct sched_rt_entity
*rt_se
, unsigned int flags
)
1234 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1235 struct rt_prio_array
*array
= &rt_rq
->active
;
1236 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
1237 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1240 * Don't enqueue the group if its throttled, or when empty.
1241 * The latter is a consequence of the former when a child group
1242 * get throttled and the current group doesn't have any other
1245 if (group_rq
&& (rt_rq_throttled(group_rq
) || !group_rq
->rt_nr_running
)) {
1247 __delist_rt_entity(rt_se
, array
);
1251 if (move_entity(flags
)) {
1252 WARN_ON_ONCE(rt_se
->on_list
);
1253 if (flags
& ENQUEUE_HEAD
)
1254 list_add(&rt_se
->run_list
, queue
);
1256 list_add_tail(&rt_se
->run_list
, queue
);
1258 __set_bit(rt_se_prio(rt_se
), array
->bitmap
);
1263 inc_rt_tasks(rt_se
, rt_rq
);
1266 static void __dequeue_rt_entity(struct sched_rt_entity
*rt_se
, unsigned int flags
)
1268 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1269 struct rt_prio_array
*array
= &rt_rq
->active
;
1271 if (move_entity(flags
)) {
1272 WARN_ON_ONCE(!rt_se
->on_list
);
1273 __delist_rt_entity(rt_se
, array
);
1277 dec_rt_tasks(rt_se
, rt_rq
);
1281 * Because the prio of an upper entry depends on the lower
1282 * entries, we must remove entries top - down.
1284 static void dequeue_rt_stack(struct sched_rt_entity
*rt_se
, unsigned int flags
)
1286 struct sched_rt_entity
*back
= NULL
;
1288 for_each_sched_rt_entity(rt_se
) {
1293 dequeue_top_rt_rq(rt_rq_of_se(back
));
1295 for (rt_se
= back
; rt_se
; rt_se
= rt_se
->back
) {
1296 if (on_rt_rq(rt_se
))
1297 __dequeue_rt_entity(rt_se
, flags
);
1301 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, unsigned int flags
)
1303 struct rq
*rq
= rq_of_rt_se(rt_se
);
1305 dequeue_rt_stack(rt_se
, flags
);
1306 for_each_sched_rt_entity(rt_se
)
1307 __enqueue_rt_entity(rt_se
, flags
);
1308 enqueue_top_rt_rq(&rq
->rt
);
1311 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
, unsigned int flags
)
1313 struct rq
*rq
= rq_of_rt_se(rt_se
);
1315 dequeue_rt_stack(rt_se
, flags
);
1317 for_each_sched_rt_entity(rt_se
) {
1318 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
1320 if (rt_rq
&& rt_rq
->rt_nr_running
)
1321 __enqueue_rt_entity(rt_se
, flags
);
1323 enqueue_top_rt_rq(&rq
->rt
);
1327 * Adding/removing a task to/from a priority array:
1330 enqueue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1332 struct sched_rt_entity
*rt_se
= &p
->rt
;
1334 if (flags
& ENQUEUE_WAKEUP
)
1337 enqueue_rt_entity(rt_se
, flags
);
1339 if (!task_current(rq
, p
) && p
->nr_cpus_allowed
> 1)
1340 enqueue_pushable_task(rq
, p
);
1343 static void dequeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1345 struct sched_rt_entity
*rt_se
= &p
->rt
;
1348 dequeue_rt_entity(rt_se
, flags
);
1350 dequeue_pushable_task(rq
, p
);
1354 * Put task to the head or the end of the run list without the overhead of
1355 * dequeue followed by enqueue.
1358 requeue_rt_entity(struct rt_rq
*rt_rq
, struct sched_rt_entity
*rt_se
, int head
)
1360 if (on_rt_rq(rt_se
)) {
1361 struct rt_prio_array
*array
= &rt_rq
->active
;
1362 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1365 list_move(&rt_se
->run_list
, queue
);
1367 list_move_tail(&rt_se
->run_list
, queue
);
1371 static void requeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int head
)
1373 struct sched_rt_entity
*rt_se
= &p
->rt
;
1374 struct rt_rq
*rt_rq
;
1376 for_each_sched_rt_entity(rt_se
) {
1377 rt_rq
= rt_rq_of_se(rt_se
);
1378 requeue_rt_entity(rt_rq
, rt_se
, head
);
1382 static void yield_task_rt(struct rq
*rq
)
1384 requeue_task_rt(rq
, rq
->curr
, 0);
1388 static int find_lowest_rq(struct task_struct
*task
);
1391 select_task_rq_rt(struct task_struct
*p
, int cpu
, int sd_flag
, int flags
)
1393 struct task_struct
*curr
;
1396 /* For anything but wake ups, just return the task_cpu */
1397 if (sd_flag
!= SD_BALANCE_WAKE
&& sd_flag
!= SD_BALANCE_FORK
)
1403 curr
= READ_ONCE(rq
->curr
); /* unlocked access */
1406 * If the current task on @p's runqueue is an RT task, then
1407 * try to see if we can wake this RT task up on another
1408 * runqueue. Otherwise simply start this RT task
1409 * on its current runqueue.
1411 * We want to avoid overloading runqueues. If the woken
1412 * task is a higher priority, then it will stay on this CPU
1413 * and the lower prio task should be moved to another CPU.
1414 * Even though this will probably make the lower prio task
1415 * lose its cache, we do not want to bounce a higher task
1416 * around just because it gave up its CPU, perhaps for a
1419 * For equal prio tasks, we just let the scheduler sort it out.
1421 * Otherwise, just let it ride on the affined RQ and the
1422 * post-schedule router will push the preempted task away
1424 * This test is optimistic, if we get it wrong the load-balancer
1425 * will have to sort it out.
1427 if (curr
&& unlikely(rt_task(curr
)) &&
1428 (curr
->nr_cpus_allowed
< 2 ||
1429 curr
->prio
<= p
->prio
)) {
1430 int target
= find_lowest_rq(p
);
1433 * Don't bother moving it if the destination CPU is
1434 * not running a lower priority task.
1437 p
->prio
< cpu_rq(target
)->rt
.highest_prio
.curr
)
1446 static void check_preempt_equal_prio(struct rq
*rq
, struct task_struct
*p
)
1449 * Current can't be migrated, useless to reschedule,
1450 * let's hope p can move out.
1452 if (rq
->curr
->nr_cpus_allowed
== 1 ||
1453 !cpupri_find(&rq
->rd
->cpupri
, rq
->curr
, NULL
))
1457 * p is migratable, so let's not schedule it and
1458 * see if it is pushed or pulled somewhere else.
1460 if (p
->nr_cpus_allowed
!= 1
1461 && cpupri_find(&rq
->rd
->cpupri
, p
, NULL
))
1465 * There appears to be other cpus that can accept
1466 * current and none to run 'p', so lets reschedule
1467 * to try and push current away:
1469 requeue_task_rt(rq
, p
, 1);
1473 #endif /* CONFIG_SMP */
1476 * Preempt the current task with a newly woken task if needed:
1478 static void check_preempt_curr_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1480 if (p
->prio
< rq
->curr
->prio
) {
1489 * - the newly woken task is of equal priority to the current task
1490 * - the newly woken task is non-migratable while current is migratable
1491 * - current will be preempted on the next reschedule
1493 * we should check to see if current can readily move to a different
1494 * cpu. If so, we will reschedule to allow the push logic to try
1495 * to move current somewhere else, making room for our non-migratable
1498 if (p
->prio
== rq
->curr
->prio
&& !test_tsk_need_resched(rq
->curr
))
1499 check_preempt_equal_prio(rq
, p
);
1503 static struct sched_rt_entity
*pick_next_rt_entity(struct rq
*rq
,
1504 struct rt_rq
*rt_rq
)
1506 struct rt_prio_array
*array
= &rt_rq
->active
;
1507 struct sched_rt_entity
*next
= NULL
;
1508 struct list_head
*queue
;
1511 idx
= sched_find_first_bit(array
->bitmap
);
1512 BUG_ON(idx
>= MAX_RT_PRIO
);
1514 queue
= array
->queue
+ idx
;
1515 next
= list_entry(queue
->next
, struct sched_rt_entity
, run_list
);
1520 static struct task_struct
*_pick_next_task_rt(struct rq
*rq
)
1522 struct sched_rt_entity
*rt_se
;
1523 struct task_struct
*p
;
1524 struct rt_rq
*rt_rq
= &rq
->rt
;
1527 rt_se
= pick_next_rt_entity(rq
, rt_rq
);
1529 rt_rq
= group_rt_rq(rt_se
);
1532 p
= rt_task_of(rt_se
);
1533 p
->se
.exec_start
= rq_clock_task(rq
);
1538 static struct task_struct
*
1539 pick_next_task_rt(struct rq
*rq
, struct task_struct
*prev
, struct rq_flags
*rf
)
1541 struct task_struct
*p
;
1542 struct rt_rq
*rt_rq
= &rq
->rt
;
1544 if (need_pull_rt_task(rq
, prev
)) {
1546 * This is OK, because current is on_cpu, which avoids it being
1547 * picked for load-balance and preemption/IRQs are still
1548 * disabled avoiding further scheduler activity on it and we're
1549 * being very careful to re-start the picking loop.
1551 rq_unpin_lock(rq
, rf
);
1553 rq_repin_lock(rq
, rf
);
1555 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1556 * means a dl or stop task can slip in, in which case we need
1557 * to re-start task selection.
1559 if (unlikely((rq
->stop
&& task_on_rq_queued(rq
->stop
)) ||
1560 rq
->dl
.dl_nr_running
))
1565 * We may dequeue prev's rt_rq in put_prev_task().
1566 * So, we update time before rt_nr_running check.
1568 if (prev
->sched_class
== &rt_sched_class
)
1571 if (!rt_rq
->rt_queued
)
1574 put_prev_task(rq
, prev
);
1576 p
= _pick_next_task_rt(rq
);
1578 /* The running task is never eligible for pushing */
1579 dequeue_pushable_task(rq
, p
);
1581 queue_push_tasks(rq
);
1586 static void put_prev_task_rt(struct rq
*rq
, struct task_struct
*p
)
1591 * The previous task needs to be made eligible for pushing
1592 * if it is still active
1594 if (on_rt_rq(&p
->rt
) && p
->nr_cpus_allowed
> 1)
1595 enqueue_pushable_task(rq
, p
);
1600 /* Only try algorithms three times */
1601 #define RT_MAX_TRIES 3
1603 static int pick_rt_task(struct rq
*rq
, struct task_struct
*p
, int cpu
)
1605 if (!task_running(rq
, p
) &&
1606 cpumask_test_cpu(cpu
, &p
->cpus_allowed
))
1612 * Return the highest pushable rq's task, which is suitable to be executed
1613 * on the cpu, NULL otherwise
1615 static struct task_struct
*pick_highest_pushable_task(struct rq
*rq
, int cpu
)
1617 struct plist_head
*head
= &rq
->rt
.pushable_tasks
;
1618 struct task_struct
*p
;
1620 if (!has_pushable_tasks(rq
))
1623 plist_for_each_entry(p
, head
, pushable_tasks
) {
1624 if (pick_rt_task(rq
, p
, cpu
))
1631 static DEFINE_PER_CPU(cpumask_var_t
, local_cpu_mask
);
1633 static int find_lowest_rq(struct task_struct
*task
)
1635 struct sched_domain
*sd
;
1636 struct cpumask
*lowest_mask
= this_cpu_cpumask_var_ptr(local_cpu_mask
);
1637 int this_cpu
= smp_processor_id();
1638 int cpu
= task_cpu(task
);
1640 /* Make sure the mask is initialized first */
1641 if (unlikely(!lowest_mask
))
1644 if (task
->nr_cpus_allowed
== 1)
1645 return -1; /* No other targets possible */
1647 if (!cpupri_find(&task_rq(task
)->rd
->cpupri
, task
, lowest_mask
))
1648 return -1; /* No targets found */
1651 * At this point we have built a mask of cpus representing the
1652 * lowest priority tasks in the system. Now we want to elect
1653 * the best one based on our affinity and topology.
1655 * We prioritize the last cpu that the task executed on since
1656 * it is most likely cache-hot in that location.
1658 if (cpumask_test_cpu(cpu
, lowest_mask
))
1662 * Otherwise, we consult the sched_domains span maps to figure
1663 * out which cpu is logically closest to our hot cache data.
1665 if (!cpumask_test_cpu(this_cpu
, lowest_mask
))
1666 this_cpu
= -1; /* Skip this_cpu opt if not among lowest */
1669 for_each_domain(cpu
, sd
) {
1670 if (sd
->flags
& SD_WAKE_AFFINE
) {
1674 * "this_cpu" is cheaper to preempt than a
1677 if (this_cpu
!= -1 &&
1678 cpumask_test_cpu(this_cpu
, sched_domain_span(sd
))) {
1683 best_cpu
= cpumask_first_and(lowest_mask
,
1684 sched_domain_span(sd
));
1685 if (best_cpu
< nr_cpu_ids
) {
1694 * And finally, if there were no matches within the domains
1695 * just give the caller *something* to work with from the compatible
1701 cpu
= cpumask_any(lowest_mask
);
1702 if (cpu
< nr_cpu_ids
)
1707 /* Will lock the rq it finds */
1708 static struct rq
*find_lock_lowest_rq(struct task_struct
*task
, struct rq
*rq
)
1710 struct rq
*lowest_rq
= NULL
;
1714 for (tries
= 0; tries
< RT_MAX_TRIES
; tries
++) {
1715 cpu
= find_lowest_rq(task
);
1717 if ((cpu
== -1) || (cpu
== rq
->cpu
))
1720 lowest_rq
= cpu_rq(cpu
);
1722 if (lowest_rq
->rt
.highest_prio
.curr
<= task
->prio
) {
1724 * Target rq has tasks of equal or higher priority,
1725 * retrying does not release any lock and is unlikely
1726 * to yield a different result.
1732 /* if the prio of this runqueue changed, try again */
1733 if (double_lock_balance(rq
, lowest_rq
)) {
1735 * We had to unlock the run queue. In
1736 * the mean time, task could have
1737 * migrated already or had its affinity changed.
1738 * Also make sure that it wasn't scheduled on its rq.
1740 if (unlikely(task_rq(task
) != rq
||
1741 !cpumask_test_cpu(lowest_rq
->cpu
, &task
->cpus_allowed
) ||
1742 task_running(rq
, task
) ||
1744 !task_on_rq_queued(task
))) {
1746 double_unlock_balance(rq
, lowest_rq
);
1752 /* If this rq is still suitable use it. */
1753 if (lowest_rq
->rt
.highest_prio
.curr
> task
->prio
)
1757 double_unlock_balance(rq
, lowest_rq
);
1764 static struct task_struct
*pick_next_pushable_task(struct rq
*rq
)
1766 struct task_struct
*p
;
1768 if (!has_pushable_tasks(rq
))
1771 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
1772 struct task_struct
, pushable_tasks
);
1774 BUG_ON(rq
->cpu
!= task_cpu(p
));
1775 BUG_ON(task_current(rq
, p
));
1776 BUG_ON(p
->nr_cpus_allowed
<= 1);
1778 BUG_ON(!task_on_rq_queued(p
));
1779 BUG_ON(!rt_task(p
));
1785 * If the current CPU has more than one RT task, see if the non
1786 * running task can migrate over to a CPU that is running a task
1787 * of lesser priority.
1789 static int push_rt_task(struct rq
*rq
)
1791 struct task_struct
*next_task
;
1792 struct rq
*lowest_rq
;
1795 if (!rq
->rt
.overloaded
)
1798 next_task
= pick_next_pushable_task(rq
);
1803 if (unlikely(next_task
== rq
->curr
)) {
1809 * It's possible that the next_task slipped in of
1810 * higher priority than current. If that's the case
1811 * just reschedule current.
1813 if (unlikely(next_task
->prio
< rq
->curr
->prio
)) {
1818 /* We might release rq lock */
1819 get_task_struct(next_task
);
1821 /* find_lock_lowest_rq locks the rq if found */
1822 lowest_rq
= find_lock_lowest_rq(next_task
, rq
);
1824 struct task_struct
*task
;
1826 * find_lock_lowest_rq releases rq->lock
1827 * so it is possible that next_task has migrated.
1829 * We need to make sure that the task is still on the same
1830 * run-queue and is also still the next task eligible for
1833 task
= pick_next_pushable_task(rq
);
1834 if (task
== next_task
) {
1836 * The task hasn't migrated, and is still the next
1837 * eligible task, but we failed to find a run-queue
1838 * to push it to. Do not retry in this case, since
1839 * other cpus will pull from us when ready.
1845 /* No more tasks, just exit */
1849 * Something has shifted, try again.
1851 put_task_struct(next_task
);
1856 deactivate_task(rq
, next_task
, 0);
1857 set_task_cpu(next_task
, lowest_rq
->cpu
);
1858 activate_task(lowest_rq
, next_task
, 0);
1861 resched_curr(lowest_rq
);
1863 double_unlock_balance(rq
, lowest_rq
);
1866 put_task_struct(next_task
);
1871 static void push_rt_tasks(struct rq
*rq
)
1873 /* push_rt_task will return true if it moved an RT */
1874 while (push_rt_task(rq
))
1878 #ifdef HAVE_RT_PUSH_IPI
1880 * The search for the next cpu always starts at rq->cpu and ends
1881 * when we reach rq->cpu again. It will never return rq->cpu.
1882 * This returns the next cpu to check, or nr_cpu_ids if the loop
1885 * rq->rt.push_cpu holds the last cpu returned by this function,
1886 * or if this is the first instance, it must hold rq->cpu.
1888 static int rto_next_cpu(struct rq
*rq
)
1890 int prev_cpu
= rq
->rt
.push_cpu
;
1893 cpu
= cpumask_next(prev_cpu
, rq
->rd
->rto_mask
);
1896 * If the previous cpu is less than the rq's CPU, then it already
1897 * passed the end of the mask, and has started from the beginning.
1898 * We end if the next CPU is greater or equal to rq's CPU.
1900 if (prev_cpu
< rq
->cpu
) {
1904 } else if (cpu
>= nr_cpu_ids
) {
1906 * We passed the end of the mask, start at the beginning.
1907 * If the result is greater or equal to the rq's CPU, then
1908 * the loop is finished.
1910 cpu
= cpumask_first(rq
->rd
->rto_mask
);
1914 rq
->rt
.push_cpu
= cpu
;
1916 /* Return cpu to let the caller know if the loop is finished or not */
1920 static int find_next_push_cpu(struct rq
*rq
)
1926 cpu
= rto_next_cpu(rq
);
1927 if (cpu
>= nr_cpu_ids
)
1929 next_rq
= cpu_rq(cpu
);
1931 /* Make sure the next rq can push to this rq */
1932 if (next_rq
->rt
.highest_prio
.next
< rq
->rt
.highest_prio
.curr
)
1939 #define RT_PUSH_IPI_EXECUTING 1
1940 #define RT_PUSH_IPI_RESTART 2
1943 * When a high priority task schedules out from a CPU and a lower priority
1944 * task is scheduled in, a check is made to see if there's any RT tasks
1945 * on other CPUs that are waiting to run because a higher priority RT task
1946 * is currently running on its CPU. In this case, the CPU with multiple RT
1947 * tasks queued on it (overloaded) needs to be notified that a CPU has opened
1948 * up that may be able to run one of its non-running queued RT tasks.
1950 * On large CPU boxes, there's the case that several CPUs could schedule
1951 * a lower priority task at the same time, in which case it will look for
1952 * any overloaded CPUs that it could pull a task from. To do this, the runqueue
1953 * lock must be taken from that overloaded CPU. Having 10s of CPUs all fighting
1954 * for a single overloaded CPU's runqueue lock can produce a large latency.
1955 * (This has actually been observed on large boxes running cyclictest).
1956 * Instead of taking the runqueue lock of the overloaded CPU, each of the
1957 * CPUs that scheduled a lower priority task simply sends an IPI to the
1958 * overloaded CPU. An IPI is much cheaper than taking an runqueue lock with
1959 * lots of contention. The overloaded CPU will look to push its non-running
1960 * RT task off, and if it does, it can then ignore the other IPIs coming
1961 * in, and just pass those IPIs off to any other overloaded CPU.
1963 * When a CPU schedules a lower priority task, it only sends an IPI to
1964 * the "next" CPU that has overloaded RT tasks. This prevents IPI storms,
1965 * as having 10 CPUs scheduling lower priority tasks and 10 CPUs with
1966 * RT overloaded tasks, would cause 100 IPIs to go out at once.
1968 * The overloaded RT CPU, when receiving an IPI, will try to push off its
1969 * overloaded RT tasks and then send an IPI to the next CPU that has
1970 * overloaded RT tasks. This stops when all CPUs with overloaded RT tasks
1971 * have completed. Just because a CPU may have pushed off its own overloaded
1972 * RT task does not mean it should stop sending the IPI around to other
1973 * overloaded CPUs. There may be another RT task waiting to run on one of
1974 * those CPUs that are of higher priority than the one that was just
1977 * An optimization that could possibly be made is to make a CPU array similar
1978 * to the cpupri array mask of all running RT tasks, but for the overloaded
1979 * case, then the IPI could be sent to only the CPU with the highest priority
1980 * RT task waiting, and that CPU could send off further IPIs to the CPU with
1981 * the next highest waiting task. Since the overloaded case is much less likely
1982 * to happen, the complexity of this implementation may not be worth it.
1983 * Instead, just send an IPI around to all overloaded CPUs.
1985 * The rq->rt.push_flags holds the status of the IPI that is going around.
1986 * A run queue can only send out a single IPI at a time. The possible flags
1987 * for rq->rt.push_flags are:
1989 * (None or zero): No IPI is going around for the current rq
1990 * RT_PUSH_IPI_EXECUTING: An IPI for the rq is being passed around
1991 * RT_PUSH_IPI_RESTART: The priority of the running task for the rq
1992 * has changed, and the IPI should restart
1993 * circulating the overloaded CPUs again.
1995 * rq->rt.push_cpu contains the CPU that is being sent the IPI. It is updated
1996 * before sending to the next CPU.
1998 * Instead of having all CPUs that schedule a lower priority task send
1999 * an IPI to the same "first" CPU in the RT overload mask, they send it
2000 * to the next overloaded CPU after their own CPU. This helps distribute
2001 * the work when there's more than one overloaded CPU and multiple CPUs
2002 * scheduling in lower priority tasks.
2004 * When a rq schedules a lower priority task than what was currently
2005 * running, the next CPU with overloaded RT tasks is examined first.
2006 * That is, if CPU 1 and 5 are overloaded, and CPU 3 schedules a lower
2007 * priority task, it will send an IPI first to CPU 5, then CPU 5 will
2008 * send to CPU 1 if it is still overloaded. CPU 1 will clear the
2009 * rq->rt.push_flags if RT_PUSH_IPI_RESTART is not set.
2011 * The first CPU to notice IPI_RESTART is set, will clear that flag and then
2012 * send an IPI to the next overloaded CPU after the rq->cpu and not the next
2013 * CPU after push_cpu. That is, if CPU 1, 4 and 5 are overloaded when CPU 3
2014 * schedules a lower priority task, and the IPI_RESTART gets set while the
2015 * handling is being done on CPU 5, it will clear the flag and send it back to
2016 * CPU 4 instead of CPU 1.
2018 * Note, the above logic can be disabled by turning off the sched_feature
2019 * RT_PUSH_IPI. Then the rq lock of the overloaded CPU will simply be
2020 * taken by the CPU requesting a pull and the waiting RT task will be pulled
2021 * by that CPU. This may be fine for machines with few CPUs.
2023 static void tell_cpu_to_push(struct rq
*rq
)
2027 if (rq
->rt
.push_flags
& RT_PUSH_IPI_EXECUTING
) {
2028 raw_spin_lock(&rq
->rt
.push_lock
);
2029 /* Make sure it's still executing */
2030 if (rq
->rt
.push_flags
& RT_PUSH_IPI_EXECUTING
) {
2032 * Tell the IPI to restart the loop as things have
2033 * changed since it started.
2035 rq
->rt
.push_flags
|= RT_PUSH_IPI_RESTART
;
2036 raw_spin_unlock(&rq
->rt
.push_lock
);
2039 raw_spin_unlock(&rq
->rt
.push_lock
);
2042 /* When here, there's no IPI going around */
2044 rq
->rt
.push_cpu
= rq
->cpu
;
2045 cpu
= find_next_push_cpu(rq
);
2046 if (cpu
>= nr_cpu_ids
)
2049 rq
->rt
.push_flags
= RT_PUSH_IPI_EXECUTING
;
2051 irq_work_queue_on(&rq
->rt
.push_work
, cpu
);
2054 /* Called from hardirq context */
2055 static void try_to_push_tasks(void *arg
)
2057 struct rt_rq
*rt_rq
= arg
;
2058 struct rq
*rq
, *src_rq
;
2062 this_cpu
= rt_rq
->push_cpu
;
2064 /* Paranoid check */
2065 BUG_ON(this_cpu
!= smp_processor_id());
2067 rq
= cpu_rq(this_cpu
);
2068 src_rq
= rq_of_rt_rq(rt_rq
);
2071 if (has_pushable_tasks(rq
)) {
2072 raw_spin_lock(&rq
->lock
);
2074 raw_spin_unlock(&rq
->lock
);
2077 /* Pass the IPI to the next rt overloaded queue */
2078 raw_spin_lock(&rt_rq
->push_lock
);
2080 * If the source queue changed since the IPI went out,
2081 * we need to restart the search from that CPU again.
2083 if (rt_rq
->push_flags
& RT_PUSH_IPI_RESTART
) {
2084 rt_rq
->push_flags
&= ~RT_PUSH_IPI_RESTART
;
2085 rt_rq
->push_cpu
= src_rq
->cpu
;
2088 cpu
= find_next_push_cpu(src_rq
);
2090 if (cpu
>= nr_cpu_ids
)
2091 rt_rq
->push_flags
&= ~RT_PUSH_IPI_EXECUTING
;
2092 raw_spin_unlock(&rt_rq
->push_lock
);
2094 if (cpu
>= nr_cpu_ids
)
2098 * It is possible that a restart caused this CPU to be
2099 * chosen again. Don't bother with an IPI, just see if we
2100 * have more to push.
2102 if (unlikely(cpu
== rq
->cpu
))
2105 /* Try the next RT overloaded CPU */
2106 irq_work_queue_on(&rt_rq
->push_work
, cpu
);
2109 static void push_irq_work_func(struct irq_work
*work
)
2111 struct rt_rq
*rt_rq
= container_of(work
, struct rt_rq
, push_work
);
2113 try_to_push_tasks(rt_rq
);
2115 #endif /* HAVE_RT_PUSH_IPI */
2117 static void pull_rt_task(struct rq
*this_rq
)
2119 int this_cpu
= this_rq
->cpu
, cpu
;
2120 bool resched
= false;
2121 struct task_struct
*p
;
2124 if (likely(!rt_overloaded(this_rq
)))
2128 * Match the barrier from rt_set_overloaded; this guarantees that if we
2129 * see overloaded we must also see the rto_mask bit.
2133 #ifdef HAVE_RT_PUSH_IPI
2134 if (sched_feat(RT_PUSH_IPI
)) {
2135 tell_cpu_to_push(this_rq
);
2140 for_each_cpu(cpu
, this_rq
->rd
->rto_mask
) {
2141 if (this_cpu
== cpu
)
2144 src_rq
= cpu_rq(cpu
);
2147 * Don't bother taking the src_rq->lock if the next highest
2148 * task is known to be lower-priority than our current task.
2149 * This may look racy, but if this value is about to go
2150 * logically higher, the src_rq will push this task away.
2151 * And if its going logically lower, we do not care
2153 if (src_rq
->rt
.highest_prio
.next
>=
2154 this_rq
->rt
.highest_prio
.curr
)
2158 * We can potentially drop this_rq's lock in
2159 * double_lock_balance, and another CPU could
2162 double_lock_balance(this_rq
, src_rq
);
2165 * We can pull only a task, which is pushable
2166 * on its rq, and no others.
2168 p
= pick_highest_pushable_task(src_rq
, this_cpu
);
2171 * Do we have an RT task that preempts
2172 * the to-be-scheduled task?
2174 if (p
&& (p
->prio
< this_rq
->rt
.highest_prio
.curr
)) {
2175 WARN_ON(p
== src_rq
->curr
);
2176 WARN_ON(!task_on_rq_queued(p
));
2179 * There's a chance that p is higher in priority
2180 * than what's currently running on its cpu.
2181 * This is just that p is wakeing up and hasn't
2182 * had a chance to schedule. We only pull
2183 * p if it is lower in priority than the
2184 * current task on the run queue
2186 if (p
->prio
< src_rq
->curr
->prio
)
2191 deactivate_task(src_rq
, p
, 0);
2192 set_task_cpu(p
, this_cpu
);
2193 activate_task(this_rq
, p
, 0);
2195 * We continue with the search, just in
2196 * case there's an even higher prio task
2197 * in another runqueue. (low likelihood
2202 double_unlock_balance(this_rq
, src_rq
);
2206 resched_curr(this_rq
);
2210 * If we are not running and we are not going to reschedule soon, we should
2211 * try to push tasks away now
2213 static void task_woken_rt(struct rq
*rq
, struct task_struct
*p
)
2215 if (!task_running(rq
, p
) &&
2216 !test_tsk_need_resched(rq
->curr
) &&
2217 p
->nr_cpus_allowed
> 1 &&
2218 (dl_task(rq
->curr
) || rt_task(rq
->curr
)) &&
2219 (rq
->curr
->nr_cpus_allowed
< 2 ||
2220 rq
->curr
->prio
<= p
->prio
))
2224 /* Assumes rq->lock is held */
2225 static void rq_online_rt(struct rq
*rq
)
2227 if (rq
->rt
.overloaded
)
2228 rt_set_overload(rq
);
2230 __enable_runtime(rq
);
2232 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rq
->rt
.highest_prio
.curr
);
2235 /* Assumes rq->lock is held */
2236 static void rq_offline_rt(struct rq
*rq
)
2238 if (rq
->rt
.overloaded
)
2239 rt_clear_overload(rq
);
2241 __disable_runtime(rq
);
2243 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, CPUPRI_INVALID
);
2247 * When switch from the rt queue, we bring ourselves to a position
2248 * that we might want to pull RT tasks from other runqueues.
2250 static void switched_from_rt(struct rq
*rq
, struct task_struct
*p
)
2253 * If there are other RT tasks then we will reschedule
2254 * and the scheduling of the other RT tasks will handle
2255 * the balancing. But if we are the last RT task
2256 * we may need to handle the pulling of RT tasks
2259 if (!task_on_rq_queued(p
) || rq
->rt
.rt_nr_running
)
2262 queue_pull_task(rq
);
2265 void __init
init_sched_rt_class(void)
2269 for_each_possible_cpu(i
) {
2270 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask
, i
),
2271 GFP_KERNEL
, cpu_to_node(i
));
2274 #endif /* CONFIG_SMP */
2277 * When switching a task to RT, we may overload the runqueue
2278 * with RT tasks. In this case we try to push them off to
2281 static void switched_to_rt(struct rq
*rq
, struct task_struct
*p
)
2284 * If we are already running, then there's nothing
2285 * that needs to be done. But if we are not running
2286 * we may need to preempt the current running task.
2287 * If that current running task is also an RT task
2288 * then see if we can move to another run queue.
2290 if (task_on_rq_queued(p
) && rq
->curr
!= p
) {
2292 if (p
->nr_cpus_allowed
> 1 && rq
->rt
.overloaded
)
2293 queue_push_tasks(rq
);
2294 #endif /* CONFIG_SMP */
2295 if (p
->prio
< rq
->curr
->prio
)
2301 * Priority of the task has changed. This may cause
2302 * us to initiate a push or pull.
2305 prio_changed_rt(struct rq
*rq
, struct task_struct
*p
, int oldprio
)
2307 if (!task_on_rq_queued(p
))
2310 if (rq
->curr
== p
) {
2313 * If our priority decreases while running, we
2314 * may need to pull tasks to this runqueue.
2316 if (oldprio
< p
->prio
)
2317 queue_pull_task(rq
);
2320 * If there's a higher priority task waiting to run
2323 if (p
->prio
> rq
->rt
.highest_prio
.curr
)
2326 /* For UP simply resched on drop of prio */
2327 if (oldprio
< p
->prio
)
2329 #endif /* CONFIG_SMP */
2332 * This task is not running, but if it is
2333 * greater than the current running task
2336 if (p
->prio
< rq
->curr
->prio
)
2341 #ifdef CONFIG_POSIX_TIMERS
2342 static void watchdog(struct rq
*rq
, struct task_struct
*p
)
2344 unsigned long soft
, hard
;
2346 /* max may change after cur was read, this will be fixed next tick */
2347 soft
= task_rlimit(p
, RLIMIT_RTTIME
);
2348 hard
= task_rlimit_max(p
, RLIMIT_RTTIME
);
2350 if (soft
!= RLIM_INFINITY
) {
2353 if (p
->rt
.watchdog_stamp
!= jiffies
) {
2355 p
->rt
.watchdog_stamp
= jiffies
;
2358 next
= DIV_ROUND_UP(min(soft
, hard
), USEC_PER_SEC
/HZ
);
2359 if (p
->rt
.timeout
> next
)
2360 p
->cputime_expires
.sched_exp
= p
->se
.sum_exec_runtime
;
2364 static inline void watchdog(struct rq
*rq
, struct task_struct
*p
) { }
2367 static void task_tick_rt(struct rq
*rq
, struct task_struct
*p
, int queued
)
2369 struct sched_rt_entity
*rt_se
= &p
->rt
;
2376 * RR tasks need a special form of timeslice management.
2377 * FIFO tasks have no timeslices.
2379 if (p
->policy
!= SCHED_RR
)
2382 if (--p
->rt
.time_slice
)
2385 p
->rt
.time_slice
= sched_rr_timeslice
;
2388 * Requeue to the end of queue if we (and all of our ancestors) are not
2389 * the only element on the queue
2391 for_each_sched_rt_entity(rt_se
) {
2392 if (rt_se
->run_list
.prev
!= rt_se
->run_list
.next
) {
2393 requeue_task_rt(rq
, p
, 0);
2400 static void set_curr_task_rt(struct rq
*rq
)
2402 struct task_struct
*p
= rq
->curr
;
2404 p
->se
.exec_start
= rq_clock_task(rq
);
2406 /* The running task is never eligible for pushing */
2407 dequeue_pushable_task(rq
, p
);
2410 static unsigned int get_rr_interval_rt(struct rq
*rq
, struct task_struct
*task
)
2413 * Time slice is 0 for SCHED_FIFO tasks
2415 if (task
->policy
== SCHED_RR
)
2416 return sched_rr_timeslice
;
2421 const struct sched_class rt_sched_class
= {
2422 .next
= &fair_sched_class
,
2423 .enqueue_task
= enqueue_task_rt
,
2424 .dequeue_task
= dequeue_task_rt
,
2425 .yield_task
= yield_task_rt
,
2427 .check_preempt_curr
= check_preempt_curr_rt
,
2429 .pick_next_task
= pick_next_task_rt
,
2430 .put_prev_task
= put_prev_task_rt
,
2433 .select_task_rq
= select_task_rq_rt
,
2435 .set_cpus_allowed
= set_cpus_allowed_common
,
2436 .rq_online
= rq_online_rt
,
2437 .rq_offline
= rq_offline_rt
,
2438 .task_woken
= task_woken_rt
,
2439 .switched_from
= switched_from_rt
,
2442 .set_curr_task
= set_curr_task_rt
,
2443 .task_tick
= task_tick_rt
,
2445 .get_rr_interval
= get_rr_interval_rt
,
2447 .prio_changed
= prio_changed_rt
,
2448 .switched_to
= switched_to_rt
,
2450 .update_curr
= update_curr_rt
,
2453 #ifdef CONFIG_RT_GROUP_SCHED
2455 * Ensure that the real time constraints are schedulable.
2457 static DEFINE_MUTEX(rt_constraints_mutex
);
2459 /* Must be called with tasklist_lock held */
2460 static inline int tg_has_rt_tasks(struct task_group
*tg
)
2462 struct task_struct
*g
, *p
;
2465 * Autogroups do not have RT tasks; see autogroup_create().
2467 if (task_group_is_autogroup(tg
))
2470 for_each_process_thread(g
, p
) {
2471 if (rt_task(p
) && task_group(p
) == tg
)
2478 struct rt_schedulable_data
{
2479 struct task_group
*tg
;
2484 static int tg_rt_schedulable(struct task_group
*tg
, void *data
)
2486 struct rt_schedulable_data
*d
= data
;
2487 struct task_group
*child
;
2488 unsigned long total
, sum
= 0;
2489 u64 period
, runtime
;
2491 period
= ktime_to_ns(tg
->rt_bandwidth
.rt_period
);
2492 runtime
= tg
->rt_bandwidth
.rt_runtime
;
2495 period
= d
->rt_period
;
2496 runtime
= d
->rt_runtime
;
2500 * Cannot have more runtime than the period.
2502 if (runtime
> period
&& runtime
!= RUNTIME_INF
)
2506 * Ensure we don't starve existing RT tasks.
2508 if (rt_bandwidth_enabled() && !runtime
&& tg_has_rt_tasks(tg
))
2511 total
= to_ratio(period
, runtime
);
2514 * Nobody can have more than the global setting allows.
2516 if (total
> to_ratio(global_rt_period(), global_rt_runtime()))
2520 * The sum of our children's runtime should not exceed our own.
2522 list_for_each_entry_rcu(child
, &tg
->children
, siblings
) {
2523 period
= ktime_to_ns(child
->rt_bandwidth
.rt_period
);
2524 runtime
= child
->rt_bandwidth
.rt_runtime
;
2526 if (child
== d
->tg
) {
2527 period
= d
->rt_period
;
2528 runtime
= d
->rt_runtime
;
2531 sum
+= to_ratio(period
, runtime
);
2540 static int __rt_schedulable(struct task_group
*tg
, u64 period
, u64 runtime
)
2544 struct rt_schedulable_data data
= {
2546 .rt_period
= period
,
2547 .rt_runtime
= runtime
,
2551 ret
= walk_tg_tree(tg_rt_schedulable
, tg_nop
, &data
);
2557 static int tg_set_rt_bandwidth(struct task_group
*tg
,
2558 u64 rt_period
, u64 rt_runtime
)
2563 * Disallowing the root group RT runtime is BAD, it would disallow the
2564 * kernel creating (and or operating) RT threads.
2566 if (tg
== &root_task_group
&& rt_runtime
== 0)
2569 /* No period doesn't make any sense. */
2573 mutex_lock(&rt_constraints_mutex
);
2574 read_lock(&tasklist_lock
);
2575 err
= __rt_schedulable(tg
, rt_period
, rt_runtime
);
2579 raw_spin_lock_irq(&tg
->rt_bandwidth
.rt_runtime_lock
);
2580 tg
->rt_bandwidth
.rt_period
= ns_to_ktime(rt_period
);
2581 tg
->rt_bandwidth
.rt_runtime
= rt_runtime
;
2583 for_each_possible_cpu(i
) {
2584 struct rt_rq
*rt_rq
= tg
->rt_rq
[i
];
2586 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
2587 rt_rq
->rt_runtime
= rt_runtime
;
2588 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
2590 raw_spin_unlock_irq(&tg
->rt_bandwidth
.rt_runtime_lock
);
2592 read_unlock(&tasklist_lock
);
2593 mutex_unlock(&rt_constraints_mutex
);
2598 int sched_group_set_rt_runtime(struct task_group
*tg
, long rt_runtime_us
)
2600 u64 rt_runtime
, rt_period
;
2602 rt_period
= ktime_to_ns(tg
->rt_bandwidth
.rt_period
);
2603 rt_runtime
= (u64
)rt_runtime_us
* NSEC_PER_USEC
;
2604 if (rt_runtime_us
< 0)
2605 rt_runtime
= RUNTIME_INF
;
2607 return tg_set_rt_bandwidth(tg
, rt_period
, rt_runtime
);
2610 long sched_group_rt_runtime(struct task_group
*tg
)
2614 if (tg
->rt_bandwidth
.rt_runtime
== RUNTIME_INF
)
2617 rt_runtime_us
= tg
->rt_bandwidth
.rt_runtime
;
2618 do_div(rt_runtime_us
, NSEC_PER_USEC
);
2619 return rt_runtime_us
;
2622 int sched_group_set_rt_period(struct task_group
*tg
, u64 rt_period_us
)
2624 u64 rt_runtime
, rt_period
;
2626 rt_period
= rt_period_us
* NSEC_PER_USEC
;
2627 rt_runtime
= tg
->rt_bandwidth
.rt_runtime
;
2629 return tg_set_rt_bandwidth(tg
, rt_period
, rt_runtime
);
2632 long sched_group_rt_period(struct task_group
*tg
)
2636 rt_period_us
= ktime_to_ns(tg
->rt_bandwidth
.rt_period
);
2637 do_div(rt_period_us
, NSEC_PER_USEC
);
2638 return rt_period_us
;
2641 static int sched_rt_global_constraints(void)
2645 mutex_lock(&rt_constraints_mutex
);
2646 read_lock(&tasklist_lock
);
2647 ret
= __rt_schedulable(NULL
, 0, 0);
2648 read_unlock(&tasklist_lock
);
2649 mutex_unlock(&rt_constraints_mutex
);
2654 int sched_rt_can_attach(struct task_group
*tg
, struct task_struct
*tsk
)
2656 /* Don't accept realtime tasks when there is no way for them to run */
2657 if (rt_task(tsk
) && tg
->rt_bandwidth
.rt_runtime
== 0)
2663 #else /* !CONFIG_RT_GROUP_SCHED */
2664 static int sched_rt_global_constraints(void)
2666 unsigned long flags
;
2669 raw_spin_lock_irqsave(&def_rt_bandwidth
.rt_runtime_lock
, flags
);
2670 for_each_possible_cpu(i
) {
2671 struct rt_rq
*rt_rq
= &cpu_rq(i
)->rt
;
2673 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
2674 rt_rq
->rt_runtime
= global_rt_runtime();
2675 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
2677 raw_spin_unlock_irqrestore(&def_rt_bandwidth
.rt_runtime_lock
, flags
);
2681 #endif /* CONFIG_RT_GROUP_SCHED */
2683 static int sched_rt_global_validate(void)
2685 if (sysctl_sched_rt_period
<= 0)
2688 if ((sysctl_sched_rt_runtime
!= RUNTIME_INF
) &&
2689 (sysctl_sched_rt_runtime
> sysctl_sched_rt_period
))
2695 static void sched_rt_do_global(void)
2697 def_rt_bandwidth
.rt_runtime
= global_rt_runtime();
2698 def_rt_bandwidth
.rt_period
= ns_to_ktime(global_rt_period());
2701 int sched_rt_handler(struct ctl_table
*table
, int write
,
2702 void __user
*buffer
, size_t *lenp
,
2705 int old_period
, old_runtime
;
2706 static DEFINE_MUTEX(mutex
);
2710 old_period
= sysctl_sched_rt_period
;
2711 old_runtime
= sysctl_sched_rt_runtime
;
2713 ret
= proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
2715 if (!ret
&& write
) {
2716 ret
= sched_rt_global_validate();
2720 ret
= sched_dl_global_validate();
2724 ret
= sched_rt_global_constraints();
2728 sched_rt_do_global();
2729 sched_dl_do_global();
2733 sysctl_sched_rt_period
= old_period
;
2734 sysctl_sched_rt_runtime
= old_runtime
;
2736 mutex_unlock(&mutex
);
2741 int sched_rr_handler(struct ctl_table
*table
, int write
,
2742 void __user
*buffer
, size_t *lenp
,
2746 static DEFINE_MUTEX(mutex
);
2749 ret
= proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
2751 * Make sure that internally we keep jiffies.
2752 * Also, writing zero resets the timeslice to default:
2754 if (!ret
&& write
) {
2755 sched_rr_timeslice
=
2756 sysctl_sched_rr_timeslice
<= 0 ? RR_TIMESLICE
:
2757 msecs_to_jiffies(sysctl_sched_rr_timeslice
);
2759 mutex_unlock(&mutex
);
2763 #ifdef CONFIG_SCHED_DEBUG
2764 extern void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
);
2766 void print_rt_stats(struct seq_file
*m
, int cpu
)
2769 struct rt_rq
*rt_rq
;
2772 for_each_rt_rq(rt_rq
, iter
, cpu_rq(cpu
))
2773 print_rt_rq(m
, cpu
, rt_rq
);
2776 #endif /* CONFIG_SCHED_DEBUG */