2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
6 #ifdef CONFIG_RT_GROUP_SCHED
8 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
10 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
12 #ifdef CONFIG_SCHED_DEBUG
13 WARN_ON_ONCE(!rt_entity_is_task(rt_se
));
15 return container_of(rt_se
, struct task_struct
, rt
);
18 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
23 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
28 #else /* CONFIG_RT_GROUP_SCHED */
30 #define rt_entity_is_task(rt_se) (1)
32 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
34 return container_of(rt_se
, struct task_struct
, rt
);
37 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
39 return container_of(rt_rq
, struct rq
, rt
);
42 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
44 struct task_struct
*p
= rt_task_of(rt_se
);
45 struct rq
*rq
= task_rq(p
);
50 #endif /* CONFIG_RT_GROUP_SCHED */
54 static inline int rt_overloaded(struct rq
*rq
)
56 return atomic_read(&rq
->rd
->rto_count
);
59 static inline void rt_set_overload(struct rq
*rq
)
64 cpumask_set_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
66 * Make sure the mask is visible before we set
67 * the overload count. That is checked to determine
68 * if we should look at the mask. It would be a shame
69 * if we looked at the mask, but the mask was not
73 atomic_inc(&rq
->rd
->rto_count
);
76 static inline void rt_clear_overload(struct rq
*rq
)
81 /* the order here really doesn't matter */
82 atomic_dec(&rq
->rd
->rto_count
);
83 cpumask_clear_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
86 static void update_rt_migration(struct rt_rq
*rt_rq
)
88 if (rt_rq
->rt_nr_migratory
&& rt_rq
->rt_nr_total
> 1) {
89 if (!rt_rq
->overloaded
) {
90 rt_set_overload(rq_of_rt_rq(rt_rq
));
91 rt_rq
->overloaded
= 1;
93 } else if (rt_rq
->overloaded
) {
94 rt_clear_overload(rq_of_rt_rq(rt_rq
));
95 rt_rq
->overloaded
= 0;
99 static void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
101 if (!rt_entity_is_task(rt_se
))
104 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
106 rt_rq
->rt_nr_total
++;
107 if (rt_se
->nr_cpus_allowed
> 1)
108 rt_rq
->rt_nr_migratory
++;
110 update_rt_migration(rt_rq
);
113 static void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
115 if (!rt_entity_is_task(rt_se
))
118 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
120 rt_rq
->rt_nr_total
--;
121 if (rt_se
->nr_cpus_allowed
> 1)
122 rt_rq
->rt_nr_migratory
--;
124 update_rt_migration(rt_rq
);
127 static void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
129 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
130 plist_node_init(&p
->pushable_tasks
, p
->prio
);
131 plist_add(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
134 static void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
136 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
139 static inline int has_pushable_tasks(struct rq
*rq
)
141 return !plist_head_empty(&rq
->rt
.pushable_tasks
);
146 static inline void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
150 static inline void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
155 void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
160 void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
164 #endif /* CONFIG_SMP */
166 static inline int on_rt_rq(struct sched_rt_entity
*rt_se
)
168 return !list_empty(&rt_se
->run_list
);
171 #ifdef CONFIG_RT_GROUP_SCHED
173 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
178 return rt_rq
->rt_runtime
;
181 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
183 return ktime_to_ns(rt_rq
->tg
->rt_bandwidth
.rt_period
);
186 #define for_each_leaf_rt_rq(rt_rq, rq) \
187 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
189 #define for_each_sched_rt_entity(rt_se) \
190 for (; rt_se; rt_se = rt_se->parent)
192 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
197 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
);
198 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
);
200 static void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
202 struct task_struct
*curr
= rq_of_rt_rq(rt_rq
)->curr
;
203 struct sched_rt_entity
*rt_se
= rt_rq
->rt_se
;
205 if (rt_rq
->rt_nr_running
) {
206 if (rt_se
&& !on_rt_rq(rt_se
))
207 enqueue_rt_entity(rt_se
, false);
208 if (rt_rq
->highest_prio
.curr
< curr
->prio
)
213 static void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
215 struct sched_rt_entity
*rt_se
= rt_rq
->rt_se
;
217 if (rt_se
&& on_rt_rq(rt_se
))
218 dequeue_rt_entity(rt_se
);
221 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
223 return rt_rq
->rt_throttled
&& !rt_rq
->rt_nr_boosted
;
226 static int rt_se_boosted(struct sched_rt_entity
*rt_se
)
228 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
229 struct task_struct
*p
;
232 return !!rt_rq
->rt_nr_boosted
;
234 p
= rt_task_of(rt_se
);
235 return p
->prio
!= p
->normal_prio
;
239 static inline const struct cpumask
*sched_rt_period_mask(void)
241 return cpu_rq(smp_processor_id())->rd
->span
;
244 static inline const struct cpumask
*sched_rt_period_mask(void)
246 return cpu_online_mask
;
251 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
253 return container_of(rt_b
, struct task_group
, rt_bandwidth
)->rt_rq
[cpu
];
256 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
258 return &rt_rq
->tg
->rt_bandwidth
;
261 #else /* !CONFIG_RT_GROUP_SCHED */
263 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
265 return rt_rq
->rt_runtime
;
268 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
270 return ktime_to_ns(def_rt_bandwidth
.rt_period
);
273 #define for_each_leaf_rt_rq(rt_rq, rq) \
274 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
276 #define for_each_sched_rt_entity(rt_se) \
277 for (; rt_se; rt_se = NULL)
279 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
284 static inline void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
286 if (rt_rq
->rt_nr_running
)
287 resched_task(rq_of_rt_rq(rt_rq
)->curr
);
290 static inline void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
294 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
296 return rt_rq
->rt_throttled
;
299 static inline const struct cpumask
*sched_rt_period_mask(void)
301 return cpu_online_mask
;
305 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
307 return &cpu_rq(cpu
)->rt
;
310 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
312 return &def_rt_bandwidth
;
315 #endif /* CONFIG_RT_GROUP_SCHED */
319 * We ran out of runtime, see if we can borrow some from our neighbours.
321 static int do_balance_runtime(struct rt_rq
*rt_rq
)
323 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
324 struct root_domain
*rd
= cpu_rq(smp_processor_id())->rd
;
325 int i
, weight
, more
= 0;
328 weight
= cpumask_weight(rd
->span
);
330 spin_lock(&rt_b
->rt_runtime_lock
);
331 rt_period
= ktime_to_ns(rt_b
->rt_period
);
332 for_each_cpu(i
, rd
->span
) {
333 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
339 spin_lock(&iter
->rt_runtime_lock
);
341 * Either all rqs have inf runtime and there's nothing to steal
342 * or __disable_runtime() below sets a specific rq to inf to
343 * indicate its been disabled and disalow stealing.
345 if (iter
->rt_runtime
== RUNTIME_INF
)
349 * From runqueues with spare time, take 1/n part of their
350 * spare time, but no more than our period.
352 diff
= iter
->rt_runtime
- iter
->rt_time
;
354 diff
= div_u64((u64
)diff
, weight
);
355 if (rt_rq
->rt_runtime
+ diff
> rt_period
)
356 diff
= rt_period
- rt_rq
->rt_runtime
;
357 iter
->rt_runtime
-= diff
;
358 rt_rq
->rt_runtime
+= diff
;
360 if (rt_rq
->rt_runtime
== rt_period
) {
361 spin_unlock(&iter
->rt_runtime_lock
);
366 spin_unlock(&iter
->rt_runtime_lock
);
368 spin_unlock(&rt_b
->rt_runtime_lock
);
374 * Ensure this RQ takes back all the runtime it lend to its neighbours.
376 static void __disable_runtime(struct rq
*rq
)
378 struct root_domain
*rd
= rq
->rd
;
381 if (unlikely(!scheduler_running
))
384 for_each_leaf_rt_rq(rt_rq
, rq
) {
385 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
389 spin_lock(&rt_b
->rt_runtime_lock
);
390 spin_lock(&rt_rq
->rt_runtime_lock
);
392 * Either we're all inf and nobody needs to borrow, or we're
393 * already disabled and thus have nothing to do, or we have
394 * exactly the right amount of runtime to take out.
396 if (rt_rq
->rt_runtime
== RUNTIME_INF
||
397 rt_rq
->rt_runtime
== rt_b
->rt_runtime
)
399 spin_unlock(&rt_rq
->rt_runtime_lock
);
402 * Calculate the difference between what we started out with
403 * and what we current have, that's the amount of runtime
404 * we lend and now have to reclaim.
406 want
= rt_b
->rt_runtime
- rt_rq
->rt_runtime
;
409 * Greedy reclaim, take back as much as we can.
411 for_each_cpu(i
, rd
->span
) {
412 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
416 * Can't reclaim from ourselves or disabled runqueues.
418 if (iter
== rt_rq
|| iter
->rt_runtime
== RUNTIME_INF
)
421 spin_lock(&iter
->rt_runtime_lock
);
423 diff
= min_t(s64
, iter
->rt_runtime
, want
);
424 iter
->rt_runtime
-= diff
;
427 iter
->rt_runtime
-= want
;
430 spin_unlock(&iter
->rt_runtime_lock
);
436 spin_lock(&rt_rq
->rt_runtime_lock
);
438 * We cannot be left wanting - that would mean some runtime
439 * leaked out of the system.
444 * Disable all the borrow logic by pretending we have inf
445 * runtime - in which case borrowing doesn't make sense.
447 rt_rq
->rt_runtime
= RUNTIME_INF
;
448 spin_unlock(&rt_rq
->rt_runtime_lock
);
449 spin_unlock(&rt_b
->rt_runtime_lock
);
453 static void disable_runtime(struct rq
*rq
)
457 spin_lock_irqsave(&rq
->lock
, flags
);
458 __disable_runtime(rq
);
459 spin_unlock_irqrestore(&rq
->lock
, flags
);
462 static void __enable_runtime(struct rq
*rq
)
466 if (unlikely(!scheduler_running
))
470 * Reset each runqueue's bandwidth settings
472 for_each_leaf_rt_rq(rt_rq
, rq
) {
473 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
475 spin_lock(&rt_b
->rt_runtime_lock
);
476 spin_lock(&rt_rq
->rt_runtime_lock
);
477 rt_rq
->rt_runtime
= rt_b
->rt_runtime
;
479 rt_rq
->rt_throttled
= 0;
480 spin_unlock(&rt_rq
->rt_runtime_lock
);
481 spin_unlock(&rt_b
->rt_runtime_lock
);
485 static void enable_runtime(struct rq
*rq
)
489 spin_lock_irqsave(&rq
->lock
, flags
);
490 __enable_runtime(rq
);
491 spin_unlock_irqrestore(&rq
->lock
, flags
);
494 static int balance_runtime(struct rt_rq
*rt_rq
)
498 if (rt_rq
->rt_time
> rt_rq
->rt_runtime
) {
499 spin_unlock(&rt_rq
->rt_runtime_lock
);
500 more
= do_balance_runtime(rt_rq
);
501 spin_lock(&rt_rq
->rt_runtime_lock
);
506 #else /* !CONFIG_SMP */
507 static inline int balance_runtime(struct rt_rq
*rt_rq
)
511 #endif /* CONFIG_SMP */
513 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
)
516 const struct cpumask
*span
;
518 if (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
)
521 span
= sched_rt_period_mask();
522 for_each_cpu(i
, span
) {
524 struct rt_rq
*rt_rq
= sched_rt_period_rt_rq(rt_b
, i
);
525 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
527 spin_lock(&rq
->lock
);
528 if (rt_rq
->rt_time
) {
531 spin_lock(&rt_rq
->rt_runtime_lock
);
532 if (rt_rq
->rt_throttled
)
533 balance_runtime(rt_rq
);
534 runtime
= rt_rq
->rt_runtime
;
535 rt_rq
->rt_time
-= min(rt_rq
->rt_time
, overrun
*runtime
);
536 if (rt_rq
->rt_throttled
&& rt_rq
->rt_time
< runtime
) {
537 rt_rq
->rt_throttled
= 0;
540 if (rt_rq
->rt_time
|| rt_rq
->rt_nr_running
)
542 spin_unlock(&rt_rq
->rt_runtime_lock
);
543 } else if (rt_rq
->rt_nr_running
)
547 sched_rt_rq_enqueue(rt_rq
);
548 spin_unlock(&rq
->lock
);
554 static inline int rt_se_prio(struct sched_rt_entity
*rt_se
)
556 #ifdef CONFIG_RT_GROUP_SCHED
557 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
560 return rt_rq
->highest_prio
.curr
;
563 return rt_task_of(rt_se
)->prio
;
566 static int sched_rt_runtime_exceeded(struct rt_rq
*rt_rq
)
568 u64 runtime
= sched_rt_runtime(rt_rq
);
570 if (rt_rq
->rt_throttled
)
571 return rt_rq_throttled(rt_rq
);
573 if (sched_rt_runtime(rt_rq
) >= sched_rt_period(rt_rq
))
576 balance_runtime(rt_rq
);
577 runtime
= sched_rt_runtime(rt_rq
);
578 if (runtime
== RUNTIME_INF
)
581 if (rt_rq
->rt_time
> runtime
) {
582 rt_rq
->rt_throttled
= 1;
583 if (rt_rq_throttled(rt_rq
)) {
584 sched_rt_rq_dequeue(rt_rq
);
593 * Update the current task's runtime statistics. Skip current tasks that
594 * are not in our scheduling class.
596 static void update_curr_rt(struct rq
*rq
)
598 struct task_struct
*curr
= rq
->curr
;
599 struct sched_rt_entity
*rt_se
= &curr
->rt
;
600 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
603 if (!task_has_rt_policy(curr
))
606 delta_exec
= rq
->clock_task
- curr
->se
.exec_start
;
607 if (unlikely((s64
)delta_exec
< 0))
610 schedstat_set(curr
->se
.exec_max
, max(curr
->se
.exec_max
, delta_exec
));
612 curr
->se
.sum_exec_runtime
+= delta_exec
;
613 account_group_exec_runtime(curr
, delta_exec
);
615 curr
->se
.exec_start
= rq
->clock_task
;
616 cpuacct_charge(curr
, delta_exec
);
618 sched_rt_avg_update(rq
, delta_exec
);
620 if (!rt_bandwidth_enabled())
623 for_each_sched_rt_entity(rt_se
) {
624 rt_rq
= rt_rq_of_se(rt_se
);
626 if (sched_rt_runtime(rt_rq
) != RUNTIME_INF
) {
627 spin_lock(&rt_rq
->rt_runtime_lock
);
628 rt_rq
->rt_time
+= delta_exec
;
629 if (sched_rt_runtime_exceeded(rt_rq
))
631 spin_unlock(&rt_rq
->rt_runtime_lock
);
636 #if defined CONFIG_SMP
638 static struct task_struct
*pick_next_highest_task_rt(struct rq
*rq
, int cpu
);
640 static inline int next_prio(struct rq
*rq
)
642 struct task_struct
*next
= pick_next_highest_task_rt(rq
, rq
->cpu
);
644 if (next
&& rt_prio(next
->prio
))
651 inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
653 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
655 if (prio
< prev_prio
) {
658 * If the new task is higher in priority than anything on the
659 * run-queue, we know that the previous high becomes our
662 rt_rq
->highest_prio
.next
= prev_prio
;
665 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, prio
);
667 } else if (prio
== rt_rq
->highest_prio
.curr
)
669 * If the next task is equal in priority to the highest on
670 * the run-queue, then we implicitly know that the next highest
671 * task cannot be any lower than current
673 rt_rq
->highest_prio
.next
= prio
;
674 else if (prio
< rt_rq
->highest_prio
.next
)
676 * Otherwise, we need to recompute next-highest
678 rt_rq
->highest_prio
.next
= next_prio(rq
);
682 dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
684 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
686 if (rt_rq
->rt_nr_running
&& (prio
<= rt_rq
->highest_prio
.next
))
687 rt_rq
->highest_prio
.next
= next_prio(rq
);
689 if (rq
->online
&& rt_rq
->highest_prio
.curr
!= prev_prio
)
690 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rt_rq
->highest_prio
.curr
);
693 #else /* CONFIG_SMP */
696 void inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
698 void dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
700 #endif /* CONFIG_SMP */
702 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
704 inc_rt_prio(struct rt_rq
*rt_rq
, int prio
)
706 int prev_prio
= rt_rq
->highest_prio
.curr
;
708 if (prio
< prev_prio
)
709 rt_rq
->highest_prio
.curr
= prio
;
711 inc_rt_prio_smp(rt_rq
, prio
, prev_prio
);
715 dec_rt_prio(struct rt_rq
*rt_rq
, int prio
)
717 int prev_prio
= rt_rq
->highest_prio
.curr
;
719 if (rt_rq
->rt_nr_running
) {
721 WARN_ON(prio
< prev_prio
);
724 * This may have been our highest task, and therefore
725 * we may have some recomputation to do
727 if (prio
== prev_prio
) {
728 struct rt_prio_array
*array
= &rt_rq
->active
;
730 rt_rq
->highest_prio
.curr
=
731 sched_find_first_bit(array
->bitmap
);
735 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
737 dec_rt_prio_smp(rt_rq
, prio
, prev_prio
);
742 static inline void inc_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
743 static inline void dec_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
745 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
747 #ifdef CONFIG_RT_GROUP_SCHED
750 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
752 if (rt_se_boosted(rt_se
))
753 rt_rq
->rt_nr_boosted
++;
756 start_rt_bandwidth(&rt_rq
->tg
->rt_bandwidth
);
760 dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
762 if (rt_se_boosted(rt_se
))
763 rt_rq
->rt_nr_boosted
--;
765 WARN_ON(!rt_rq
->rt_nr_running
&& rt_rq
->rt_nr_boosted
);
768 #else /* CONFIG_RT_GROUP_SCHED */
771 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
773 start_rt_bandwidth(&def_rt_bandwidth
);
777 void dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
) {}
779 #endif /* CONFIG_RT_GROUP_SCHED */
782 void inc_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
784 int prio
= rt_se_prio(rt_se
);
786 WARN_ON(!rt_prio(prio
));
787 rt_rq
->rt_nr_running
++;
789 inc_rt_prio(rt_rq
, prio
);
790 inc_rt_migration(rt_se
, rt_rq
);
791 inc_rt_group(rt_se
, rt_rq
);
795 void dec_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
797 WARN_ON(!rt_prio(rt_se_prio(rt_se
)));
798 WARN_ON(!rt_rq
->rt_nr_running
);
799 rt_rq
->rt_nr_running
--;
801 dec_rt_prio(rt_rq
, rt_se_prio(rt_se
));
802 dec_rt_migration(rt_se
, rt_rq
);
803 dec_rt_group(rt_se
, rt_rq
);
806 static void __enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
808 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
809 struct rt_prio_array
*array
= &rt_rq
->active
;
810 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
811 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
814 * Don't enqueue the group if its throttled, or when empty.
815 * The latter is a consequence of the former when a child group
816 * get throttled and the current group doesn't have any other
819 if (group_rq
&& (rt_rq_throttled(group_rq
) || !group_rq
->rt_nr_running
))
823 list_add(&rt_se
->run_list
, queue
);
825 list_add_tail(&rt_se
->run_list
, queue
);
826 __set_bit(rt_se_prio(rt_se
), array
->bitmap
);
828 inc_rt_tasks(rt_se
, rt_rq
);
831 static void __dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
833 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
834 struct rt_prio_array
*array
= &rt_rq
->active
;
836 list_del_init(&rt_se
->run_list
);
837 if (list_empty(array
->queue
+ rt_se_prio(rt_se
)))
838 __clear_bit(rt_se_prio(rt_se
), array
->bitmap
);
840 dec_rt_tasks(rt_se
, rt_rq
);
844 * Because the prio of an upper entry depends on the lower
845 * entries, we must remove entries top - down.
847 static void dequeue_rt_stack(struct sched_rt_entity
*rt_se
)
849 struct sched_rt_entity
*back
= NULL
;
851 for_each_sched_rt_entity(rt_se
) {
856 for (rt_se
= back
; rt_se
; rt_se
= rt_se
->back
) {
858 __dequeue_rt_entity(rt_se
);
862 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
864 dequeue_rt_stack(rt_se
);
865 for_each_sched_rt_entity(rt_se
)
866 __enqueue_rt_entity(rt_se
, head
);
869 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
871 dequeue_rt_stack(rt_se
);
873 for_each_sched_rt_entity(rt_se
) {
874 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
876 if (rt_rq
&& rt_rq
->rt_nr_running
)
877 __enqueue_rt_entity(rt_se
, false);
882 * Adding/removing a task to/from a priority array:
885 enqueue_task_rt(struct rq
*rq
, struct task_struct
*p
, int wakeup
, bool head
)
887 struct sched_rt_entity
*rt_se
= &p
->rt
;
892 enqueue_rt_entity(rt_se
, head
);
894 if (!task_current(rq
, p
) && p
->rt
.nr_cpus_allowed
> 1)
895 enqueue_pushable_task(rq
, p
);
898 static void dequeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int sleep
)
900 struct sched_rt_entity
*rt_se
= &p
->rt
;
903 dequeue_rt_entity(rt_se
);
905 dequeue_pushable_task(rq
, p
);
909 * Put task to the end of the run list without the overhead of dequeue
910 * followed by enqueue.
913 requeue_rt_entity(struct rt_rq
*rt_rq
, struct sched_rt_entity
*rt_se
, int head
)
915 if (on_rt_rq(rt_se
)) {
916 struct rt_prio_array
*array
= &rt_rq
->active
;
917 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
920 list_move(&rt_se
->run_list
, queue
);
922 list_move_tail(&rt_se
->run_list
, queue
);
926 static void requeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int head
)
928 struct sched_rt_entity
*rt_se
= &p
->rt
;
931 for_each_sched_rt_entity(rt_se
) {
932 rt_rq
= rt_rq_of_se(rt_se
);
933 requeue_rt_entity(rt_rq
, rt_se
, head
);
937 static void yield_task_rt(struct rq
*rq
)
939 requeue_task_rt(rq
, rq
->curr
, 0);
943 static int find_lowest_rq(struct task_struct
*task
);
946 select_task_rq_rt(struct rq
*rq
, struct task_struct
*p
, int sd_flag
, int flags
)
948 if (sd_flag
!= SD_BALANCE_WAKE
)
949 return smp_processor_id();
952 * If the current task is an RT task, then
953 * try to see if we can wake this RT task up on another
954 * runqueue. Otherwise simply start this RT task
955 * on its current runqueue.
957 * We want to avoid overloading runqueues. If the woken
958 * task is a higher priority, then it will stay on this CPU
959 * and the lower prio task should be moved to another CPU.
960 * Even though this will probably make the lower prio task
961 * lose its cache, we do not want to bounce a higher task
962 * around just because it gave up its CPU, perhaps for a
965 * For equal prio tasks, we just let the scheduler sort it out.
967 if (unlikely(rt_task(rq
->curr
)) &&
968 (rq
->curr
->rt
.nr_cpus_allowed
< 2 ||
969 rq
->curr
->prio
< p
->prio
) &&
970 (p
->rt
.nr_cpus_allowed
> 1)) {
971 int cpu
= find_lowest_rq(p
);
973 return (cpu
== -1) ? task_cpu(p
) : cpu
;
977 * Otherwise, just let it ride on the affined RQ and the
978 * post-schedule router will push the preempted task away
983 static void check_preempt_equal_prio(struct rq
*rq
, struct task_struct
*p
)
985 if (rq
->curr
->rt
.nr_cpus_allowed
== 1)
988 if (p
->rt
.nr_cpus_allowed
!= 1
989 && cpupri_find(&rq
->rd
->cpupri
, p
, NULL
))
992 if (!cpupri_find(&rq
->rd
->cpupri
, rq
->curr
, NULL
))
996 * There appears to be other cpus that can accept
997 * current and none to run 'p', so lets reschedule
998 * to try and push current away:
1000 requeue_task_rt(rq
, p
, 1);
1001 resched_task(rq
->curr
);
1004 #endif /* CONFIG_SMP */
1007 * Preempt the current task with a newly woken task if needed:
1009 static void check_preempt_curr_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1011 if (p
->prio
< rq
->curr
->prio
) {
1012 resched_task(rq
->curr
);
1020 * - the newly woken task is of equal priority to the current task
1021 * - the newly woken task is non-migratable while current is migratable
1022 * - current will be preempted on the next reschedule
1024 * we should check to see if current can readily move to a different
1025 * cpu. If so, we will reschedule to allow the push logic to try
1026 * to move current somewhere else, making room for our non-migratable
1029 if (p
->prio
== rq
->curr
->prio
&& !need_resched())
1030 check_preempt_equal_prio(rq
, p
);
1034 static struct sched_rt_entity
*pick_next_rt_entity(struct rq
*rq
,
1035 struct rt_rq
*rt_rq
)
1037 struct rt_prio_array
*array
= &rt_rq
->active
;
1038 struct sched_rt_entity
*next
= NULL
;
1039 struct list_head
*queue
;
1042 idx
= sched_find_first_bit(array
->bitmap
);
1043 BUG_ON(idx
>= MAX_RT_PRIO
);
1045 queue
= array
->queue
+ idx
;
1046 next
= list_entry(queue
->next
, struct sched_rt_entity
, run_list
);
1051 static struct task_struct
*_pick_next_task_rt(struct rq
*rq
)
1053 struct sched_rt_entity
*rt_se
;
1054 struct task_struct
*p
;
1055 struct rt_rq
*rt_rq
;
1059 if (unlikely(!rt_rq
->rt_nr_running
))
1062 if (rt_rq_throttled(rt_rq
))
1066 rt_se
= pick_next_rt_entity(rq
, rt_rq
);
1068 rt_rq
= group_rt_rq(rt_se
);
1071 p
= rt_task_of(rt_se
);
1072 p
->se
.exec_start
= rq
->clock_task
;
1077 static struct task_struct
*pick_next_task_rt(struct rq
*rq
)
1079 struct task_struct
*p
= _pick_next_task_rt(rq
);
1081 /* The running task is never eligible for pushing */
1083 dequeue_pushable_task(rq
, p
);
1087 * We detect this state here so that we can avoid taking the RQ
1088 * lock again later if there is no need to push
1090 rq
->post_schedule
= has_pushable_tasks(rq
);
1096 static void put_prev_task_rt(struct rq
*rq
, struct task_struct
*p
)
1099 p
->se
.exec_start
= 0;
1102 * The previous task needs to be made eligible for pushing
1103 * if it is still active
1105 if (p
->se
.on_rq
&& p
->rt
.nr_cpus_allowed
> 1)
1106 enqueue_pushable_task(rq
, p
);
1111 /* Only try algorithms three times */
1112 #define RT_MAX_TRIES 3
1114 static void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int sleep
);
1116 static int pick_rt_task(struct rq
*rq
, struct task_struct
*p
, int cpu
)
1118 if (!task_running(rq
, p
) &&
1119 (cpu
< 0 || cpumask_test_cpu(cpu
, &p
->cpus_allowed
)) &&
1120 (p
->rt
.nr_cpus_allowed
> 1))
1125 /* Return the second highest RT task, NULL otherwise */
1126 static struct task_struct
*pick_next_highest_task_rt(struct rq
*rq
, int cpu
)
1128 struct task_struct
*next
= NULL
;
1129 struct sched_rt_entity
*rt_se
;
1130 struct rt_prio_array
*array
;
1131 struct rt_rq
*rt_rq
;
1134 for_each_leaf_rt_rq(rt_rq
, rq
) {
1135 array
= &rt_rq
->active
;
1136 idx
= sched_find_first_bit(array
->bitmap
);
1138 if (idx
>= MAX_RT_PRIO
)
1140 if (next
&& next
->prio
< idx
)
1142 list_for_each_entry(rt_se
, array
->queue
+ idx
, run_list
) {
1143 struct task_struct
*p
= rt_task_of(rt_se
);
1144 if (pick_rt_task(rq
, p
, cpu
)) {
1150 idx
= find_next_bit(array
->bitmap
, MAX_RT_PRIO
, idx
+1);
1158 static DEFINE_PER_CPU(cpumask_var_t
, local_cpu_mask
);
1160 static inline int pick_optimal_cpu(int this_cpu
,
1161 const struct cpumask
*mask
)
1165 /* "this_cpu" is cheaper to preempt than a remote processor */
1166 if ((this_cpu
!= -1) && cpumask_test_cpu(this_cpu
, mask
))
1169 first
= cpumask_first(mask
);
1170 if (first
< nr_cpu_ids
)
1176 static int find_lowest_rq(struct task_struct
*task
)
1178 struct sched_domain
*sd
;
1179 struct cpumask
*lowest_mask
= __get_cpu_var(local_cpu_mask
);
1180 int this_cpu
= smp_processor_id();
1181 int cpu
= task_cpu(task
);
1182 cpumask_var_t domain_mask
;
1184 if (task
->rt
.nr_cpus_allowed
== 1)
1185 return -1; /* No other targets possible */
1187 if (!cpupri_find(&task_rq(task
)->rd
->cpupri
, task
, lowest_mask
))
1188 return -1; /* No targets found */
1191 * At this point we have built a mask of cpus representing the
1192 * lowest priority tasks in the system. Now we want to elect
1193 * the best one based on our affinity and topology.
1195 * We prioritize the last cpu that the task executed on since
1196 * it is most likely cache-hot in that location.
1198 if (cpumask_test_cpu(cpu
, lowest_mask
))
1202 * Otherwise, we consult the sched_domains span maps to figure
1203 * out which cpu is logically closest to our hot cache data.
1205 if (this_cpu
== cpu
)
1206 this_cpu
= -1; /* Skip this_cpu opt if the same */
1208 if (alloc_cpumask_var(&domain_mask
, GFP_ATOMIC
)) {
1209 for_each_domain(cpu
, sd
) {
1210 if (sd
->flags
& SD_WAKE_AFFINE
) {
1213 cpumask_and(domain_mask
,
1214 sched_domain_span(sd
),
1217 best_cpu
= pick_optimal_cpu(this_cpu
,
1220 if (best_cpu
!= -1) {
1221 free_cpumask_var(domain_mask
);
1226 free_cpumask_var(domain_mask
);
1230 * And finally, if there were no matches within the domains
1231 * just give the caller *something* to work with from the compatible
1234 return pick_optimal_cpu(this_cpu
, lowest_mask
);
1237 /* Will lock the rq it finds */
1238 static struct rq
*find_lock_lowest_rq(struct task_struct
*task
, struct rq
*rq
)
1240 struct rq
*lowest_rq
= NULL
;
1244 for (tries
= 0; tries
< RT_MAX_TRIES
; tries
++) {
1245 cpu
= find_lowest_rq(task
);
1247 if ((cpu
== -1) || (cpu
== rq
->cpu
))
1250 lowest_rq
= cpu_rq(cpu
);
1252 /* if the prio of this runqueue changed, try again */
1253 if (double_lock_balance(rq
, lowest_rq
)) {
1255 * We had to unlock the run queue. In
1256 * the mean time, task could have
1257 * migrated already or had its affinity changed.
1258 * Also make sure that it wasn't scheduled on its rq.
1260 if (unlikely(task_rq(task
) != rq
||
1261 !cpumask_test_cpu(lowest_rq
->cpu
,
1262 &task
->cpus_allowed
) ||
1263 task_running(rq
, task
) ||
1266 spin_unlock(&lowest_rq
->lock
);
1272 /* If this rq is still suitable use it. */
1273 if (lowest_rq
->rt
.highest_prio
.curr
> task
->prio
)
1277 double_unlock_balance(rq
, lowest_rq
);
1284 static struct task_struct
*pick_next_pushable_task(struct rq
*rq
)
1286 struct task_struct
*p
;
1288 if (!has_pushable_tasks(rq
))
1291 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
1292 struct task_struct
, pushable_tasks
);
1294 BUG_ON(rq
->cpu
!= task_cpu(p
));
1295 BUG_ON(task_current(rq
, p
));
1296 BUG_ON(p
->rt
.nr_cpus_allowed
<= 1);
1298 BUG_ON(!p
->se
.on_rq
);
1299 BUG_ON(!rt_task(p
));
1305 * If the current CPU has more than one RT task, see if the non
1306 * running task can migrate over to a CPU that is running a task
1307 * of lesser priority.
1309 static int push_rt_task(struct rq
*rq
)
1311 struct task_struct
*next_task
;
1312 struct rq
*lowest_rq
;
1314 if (!rq
->rt
.overloaded
)
1317 next_task
= pick_next_pushable_task(rq
);
1322 if (unlikely(next_task
== rq
->curr
)) {
1328 * It's possible that the next_task slipped in of
1329 * higher priority than current. If that's the case
1330 * just reschedule current.
1332 if (unlikely(next_task
->prio
< rq
->curr
->prio
)) {
1333 resched_task(rq
->curr
);
1337 /* We might release rq lock */
1338 get_task_struct(next_task
);
1340 /* find_lock_lowest_rq locks the rq if found */
1341 lowest_rq
= find_lock_lowest_rq(next_task
, rq
);
1343 struct task_struct
*task
;
1345 * find lock_lowest_rq releases rq->lock
1346 * so it is possible that next_task has migrated.
1348 * We need to make sure that the task is still on the same
1349 * run-queue and is also still the next task eligible for
1352 task
= pick_next_pushable_task(rq
);
1353 if (task_cpu(next_task
) == rq
->cpu
&& task
== next_task
) {
1355 * If we get here, the task hasnt moved at all, but
1356 * it has failed to push. We will not try again,
1357 * since the other cpus will pull from us when they
1360 dequeue_pushable_task(rq
, next_task
);
1365 /* No more tasks, just exit */
1369 * Something has shifted, try again.
1371 put_task_struct(next_task
);
1376 deactivate_task(rq
, next_task
, 0);
1377 set_task_cpu(next_task
, lowest_rq
->cpu
);
1378 activate_task(lowest_rq
, next_task
, 0);
1380 resched_task(lowest_rq
->curr
);
1382 double_unlock_balance(rq
, lowest_rq
);
1385 put_task_struct(next_task
);
1390 static void push_rt_tasks(struct rq
*rq
)
1392 /* push_rt_task will return true if it moved an RT */
1393 while (push_rt_task(rq
))
1397 static int pull_rt_task(struct rq
*this_rq
)
1399 int this_cpu
= this_rq
->cpu
, ret
= 0, cpu
;
1400 struct task_struct
*p
;
1403 if (likely(!rt_overloaded(this_rq
)))
1406 for_each_cpu(cpu
, this_rq
->rd
->rto_mask
) {
1407 if (this_cpu
== cpu
)
1410 src_rq
= cpu_rq(cpu
);
1413 * Don't bother taking the src_rq->lock if the next highest
1414 * task is known to be lower-priority than our current task.
1415 * This may look racy, but if this value is about to go
1416 * logically higher, the src_rq will push this task away.
1417 * And if its going logically lower, we do not care
1419 if (src_rq
->rt
.highest_prio
.next
>=
1420 this_rq
->rt
.highest_prio
.curr
)
1424 * We can potentially drop this_rq's lock in
1425 * double_lock_balance, and another CPU could
1428 double_lock_balance(this_rq
, src_rq
);
1431 * Are there still pullable RT tasks?
1433 if (src_rq
->rt
.rt_nr_running
<= 1)
1436 p
= pick_next_highest_task_rt(src_rq
, this_cpu
);
1439 * Do we have an RT task that preempts
1440 * the to-be-scheduled task?
1442 if (p
&& (p
->prio
< this_rq
->rt
.highest_prio
.curr
)) {
1443 WARN_ON(p
== src_rq
->curr
);
1444 WARN_ON(!p
->se
.on_rq
);
1447 * There's a chance that p is higher in priority
1448 * than what's currently running on its cpu.
1449 * This is just that p is wakeing up and hasn't
1450 * had a chance to schedule. We only pull
1451 * p if it is lower in priority than the
1452 * current task on the run queue
1454 if (p
->prio
< src_rq
->curr
->prio
)
1459 deactivate_task(src_rq
, p
, 0);
1460 set_task_cpu(p
, this_cpu
);
1461 activate_task(this_rq
, p
, 0);
1463 * We continue with the search, just in
1464 * case there's an even higher prio task
1465 * in another runqueue. (low likelyhood
1470 double_unlock_balance(this_rq
, src_rq
);
1476 static void pre_schedule_rt(struct rq
*rq
, struct task_struct
*prev
)
1478 /* Try to pull RT tasks here if we lower this rq's prio */
1479 if (unlikely(rt_task(prev
)) && rq
->rt
.highest_prio
.curr
> prev
->prio
)
1483 static void post_schedule_rt(struct rq
*rq
)
1489 * If we are not running and we are not going to reschedule soon, we should
1490 * try to push tasks away now
1492 static void task_woken_rt(struct rq
*rq
, struct task_struct
*p
)
1494 if (!task_running(rq
, p
) &&
1495 !test_tsk_need_resched(rq
->curr
) &&
1496 has_pushable_tasks(rq
) &&
1497 p
->rt
.nr_cpus_allowed
> 1 &&
1498 rt_task(rq
->curr
) &&
1499 (rq
->curr
->rt
.nr_cpus_allowed
< 2 ||
1500 rq
->curr
->prio
< p
->prio
))
1504 static unsigned long
1505 load_balance_rt(struct rq
*this_rq
, int this_cpu
, struct rq
*busiest
,
1506 unsigned long max_load_move
,
1507 struct sched_domain
*sd
, enum cpu_idle_type idle
,
1508 int *all_pinned
, int *this_best_prio
)
1510 /* don't touch RT tasks */
1515 move_one_task_rt(struct rq
*this_rq
, int this_cpu
, struct rq
*busiest
,
1516 struct sched_domain
*sd
, enum cpu_idle_type idle
)
1518 /* don't touch RT tasks */
1522 static void set_cpus_allowed_rt(struct task_struct
*p
,
1523 const struct cpumask
*new_mask
)
1525 int weight
= cpumask_weight(new_mask
);
1527 BUG_ON(!rt_task(p
));
1530 * Update the migration status of the RQ if we have an RT task
1531 * which is running AND changing its weight value.
1533 if (p
->se
.on_rq
&& (weight
!= p
->rt
.nr_cpus_allowed
)) {
1534 struct rq
*rq
= task_rq(p
);
1536 if (!task_current(rq
, p
)) {
1538 * Make sure we dequeue this task from the pushable list
1539 * before going further. It will either remain off of
1540 * the list because we are no longer pushable, or it
1543 if (p
->rt
.nr_cpus_allowed
> 1)
1544 dequeue_pushable_task(rq
, p
);
1547 * Requeue if our weight is changing and still > 1
1550 enqueue_pushable_task(rq
, p
);
1554 if ((p
->rt
.nr_cpus_allowed
<= 1) && (weight
> 1)) {
1555 rq
->rt
.rt_nr_migratory
++;
1556 } else if ((p
->rt
.nr_cpus_allowed
> 1) && (weight
<= 1)) {
1557 BUG_ON(!rq
->rt
.rt_nr_migratory
);
1558 rq
->rt
.rt_nr_migratory
--;
1561 update_rt_migration(&rq
->rt
);
1564 cpumask_copy(&p
->cpus_allowed
, new_mask
);
1565 p
->rt
.nr_cpus_allowed
= weight
;
1568 /* Assumes rq->lock is held */
1569 static void rq_online_rt(struct rq
*rq
)
1571 if (rq
->rt
.overloaded
)
1572 rt_set_overload(rq
);
1574 __enable_runtime(rq
);
1576 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rq
->rt
.highest_prio
.curr
);
1579 /* Assumes rq->lock is held */
1580 static void rq_offline_rt(struct rq
*rq
)
1582 if (rq
->rt
.overloaded
)
1583 rt_clear_overload(rq
);
1585 __disable_runtime(rq
);
1587 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, CPUPRI_INVALID
);
1591 * When switch from the rt queue, we bring ourselves to a position
1592 * that we might want to pull RT tasks from other runqueues.
1594 static void switched_from_rt(struct rq
*rq
, struct task_struct
*p
,
1598 * If there are other RT tasks then we will reschedule
1599 * and the scheduling of the other RT tasks will handle
1600 * the balancing. But if we are the last RT task
1601 * we may need to handle the pulling of RT tasks
1604 if (!rq
->rt
.rt_nr_running
)
1608 static inline void init_sched_rt_class(void)
1612 for_each_possible_cpu(i
)
1613 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask
, i
),
1614 GFP_KERNEL
, cpu_to_node(i
));
1616 #endif /* CONFIG_SMP */
1619 * When switching a task to RT, we may overload the runqueue
1620 * with RT tasks. In this case we try to push them off to
1623 static void switched_to_rt(struct rq
*rq
, struct task_struct
*p
,
1626 int check_resched
= 1;
1629 * If we are already running, then there's nothing
1630 * that needs to be done. But if we are not running
1631 * we may need to preempt the current running task.
1632 * If that current running task is also an RT task
1633 * then see if we can move to another run queue.
1637 if (rq
->rt
.overloaded
&& push_rt_task(rq
) &&
1638 /* Don't resched if we changed runqueues */
1641 #endif /* CONFIG_SMP */
1642 if (check_resched
&& p
->prio
< rq
->curr
->prio
)
1643 resched_task(rq
->curr
);
1648 * Priority of the task has changed. This may cause
1649 * us to initiate a push or pull.
1651 static void prio_changed_rt(struct rq
*rq
, struct task_struct
*p
,
1652 int oldprio
, int running
)
1657 * If our priority decreases while running, we
1658 * may need to pull tasks to this runqueue.
1660 if (oldprio
< p
->prio
)
1663 * If there's a higher priority task waiting to run
1664 * then reschedule. Note, the above pull_rt_task
1665 * can release the rq lock and p could migrate.
1666 * Only reschedule if p is still on the same runqueue.
1668 if (p
->prio
> rq
->rt
.highest_prio
.curr
&& rq
->curr
== p
)
1671 /* For UP simply resched on drop of prio */
1672 if (oldprio
< p
->prio
)
1674 #endif /* CONFIG_SMP */
1677 * This task is not running, but if it is
1678 * greater than the current running task
1681 if (p
->prio
< rq
->curr
->prio
)
1682 resched_task(rq
->curr
);
1686 static void watchdog(struct rq
*rq
, struct task_struct
*p
)
1688 unsigned long soft
, hard
;
1693 soft
= p
->signal
->rlim
[RLIMIT_RTTIME
].rlim_cur
;
1694 hard
= p
->signal
->rlim
[RLIMIT_RTTIME
].rlim_max
;
1696 if (soft
!= RLIM_INFINITY
) {
1700 next
= DIV_ROUND_UP(min(soft
, hard
), USEC_PER_SEC
/HZ
);
1701 if (p
->rt
.timeout
> next
)
1702 p
->cputime_expires
.sched_exp
= p
->se
.sum_exec_runtime
;
1706 static void task_tick_rt(struct rq
*rq
, struct task_struct
*p
, int queued
)
1713 * RR tasks need a special form of timeslice management.
1714 * FIFO tasks have no timeslices.
1716 if (p
->policy
!= SCHED_RR
)
1719 if (--p
->rt
.time_slice
)
1722 p
->rt
.time_slice
= DEF_TIMESLICE
;
1725 * Requeue to the end of queue if we are not the only element
1728 if (p
->rt
.run_list
.prev
!= p
->rt
.run_list
.next
) {
1729 requeue_task_rt(rq
, p
, 0);
1730 set_tsk_need_resched(p
);
1734 static void set_curr_task_rt(struct rq
*rq
)
1736 struct task_struct
*p
= rq
->curr
;
1738 p
->se
.exec_start
= rq
->clock_task
;
1740 /* The running task is never eligible for pushing */
1741 dequeue_pushable_task(rq
, p
);
1744 unsigned int get_rr_interval_rt(struct rq
*rq
, struct task_struct
*task
)
1747 * Time slice is 0 for SCHED_FIFO tasks
1749 if (task
->policy
== SCHED_RR
)
1750 return DEF_TIMESLICE
;
1755 static const struct sched_class rt_sched_class
= {
1756 .next
= &fair_sched_class
,
1757 .enqueue_task
= enqueue_task_rt
,
1758 .dequeue_task
= dequeue_task_rt
,
1759 .yield_task
= yield_task_rt
,
1761 .check_preempt_curr
= check_preempt_curr_rt
,
1763 .pick_next_task
= pick_next_task_rt
,
1764 .put_prev_task
= put_prev_task_rt
,
1767 .select_task_rq
= select_task_rq_rt
,
1769 .load_balance
= load_balance_rt
,
1770 .move_one_task
= move_one_task_rt
,
1771 .set_cpus_allowed
= set_cpus_allowed_rt
,
1772 .rq_online
= rq_online_rt
,
1773 .rq_offline
= rq_offline_rt
,
1774 .pre_schedule
= pre_schedule_rt
,
1775 .post_schedule
= post_schedule_rt
,
1776 .task_woken
= task_woken_rt
,
1777 .switched_from
= switched_from_rt
,
1780 .set_curr_task
= set_curr_task_rt
,
1781 .task_tick
= task_tick_rt
,
1783 .get_rr_interval
= get_rr_interval_rt
,
1785 .prio_changed
= prio_changed_rt
,
1786 .switched_to
= switched_to_rt
,
1789 #ifdef CONFIG_SCHED_DEBUG
1790 extern void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
);
1792 static void print_rt_stats(struct seq_file
*m
, int cpu
)
1794 struct rt_rq
*rt_rq
;
1797 for_each_leaf_rt_rq(rt_rq
, cpu_rq(cpu
))
1798 print_rt_rq(m
, cpu
, rt_rq
);
1801 #endif /* CONFIG_SCHED_DEBUG */