2 * Real-Time Scheduling Class (mapped to the SCHED_FIFO and SCHED_RR
8 #include <linux/slab.h>
10 int sched_rr_timeslice
= RR_TIMESLICE
;
12 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
);
14 struct rt_bandwidth def_rt_bandwidth
;
16 static enum hrtimer_restart
sched_rt_period_timer(struct hrtimer
*timer
)
18 struct rt_bandwidth
*rt_b
=
19 container_of(timer
, struct rt_bandwidth
, rt_period_timer
);
25 now
= hrtimer_cb_get_time(timer
);
26 overrun
= hrtimer_forward(timer
, now
, rt_b
->rt_period
);
31 idle
= do_sched_rt_period_timer(rt_b
, overrun
);
34 return idle
? HRTIMER_NORESTART
: HRTIMER_RESTART
;
37 void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
)
39 rt_b
->rt_period
= ns_to_ktime(period
);
40 rt_b
->rt_runtime
= runtime
;
42 raw_spin_lock_init(&rt_b
->rt_runtime_lock
);
44 hrtimer_init(&rt_b
->rt_period_timer
,
45 CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
46 rt_b
->rt_period_timer
.function
= sched_rt_period_timer
;
49 static void start_rt_bandwidth(struct rt_bandwidth
*rt_b
)
51 if (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
)
54 if (hrtimer_active(&rt_b
->rt_period_timer
))
57 raw_spin_lock(&rt_b
->rt_runtime_lock
);
58 start_bandwidth_timer(&rt_b
->rt_period_timer
, rt_b
->rt_period
);
59 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
62 void init_rt_rq(struct rt_rq
*rt_rq
, struct rq
*rq
)
64 struct rt_prio_array
*array
;
67 array
= &rt_rq
->active
;
68 for (i
= 0; i
< MAX_RT_PRIO
; i
++) {
69 INIT_LIST_HEAD(array
->queue
+ i
);
70 __clear_bit(i
, array
->bitmap
);
72 /* delimiter for bitsearch: */
73 __set_bit(MAX_RT_PRIO
, array
->bitmap
);
75 #if defined CONFIG_SMP
76 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
77 rt_rq
->highest_prio
.next
= MAX_RT_PRIO
;
78 rt_rq
->rt_nr_migratory
= 0;
79 rt_rq
->overloaded
= 0;
80 plist_head_init(&rt_rq
->pushable_tasks
);
84 rt_rq
->rt_throttled
= 0;
85 rt_rq
->rt_runtime
= 0;
86 raw_spin_lock_init(&rt_rq
->rt_runtime_lock
);
89 #ifdef CONFIG_RT_GROUP_SCHED
90 static void destroy_rt_bandwidth(struct rt_bandwidth
*rt_b
)
92 hrtimer_cancel(&rt_b
->rt_period_timer
);
95 #define rt_entity_is_task(rt_se) (!(rt_se)->my_q)
97 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
99 #ifdef CONFIG_SCHED_DEBUG
100 WARN_ON_ONCE(!rt_entity_is_task(rt_se
));
102 return container_of(rt_se
, struct task_struct
, rt
);
105 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
110 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
115 void free_rt_sched_group(struct task_group
*tg
)
120 destroy_rt_bandwidth(&tg
->rt_bandwidth
);
122 for_each_possible_cpu(i
) {
133 void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
134 struct sched_rt_entity
*rt_se
, int cpu
,
135 struct sched_rt_entity
*parent
)
137 struct rq
*rq
= cpu_rq(cpu
);
139 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
140 rt_rq
->rt_nr_boosted
= 0;
144 tg
->rt_rq
[cpu
] = rt_rq
;
145 tg
->rt_se
[cpu
] = rt_se
;
151 rt_se
->rt_rq
= &rq
->rt
;
153 rt_se
->rt_rq
= parent
->my_q
;
156 rt_se
->parent
= parent
;
157 INIT_LIST_HEAD(&rt_se
->run_list
);
160 int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
)
163 struct sched_rt_entity
*rt_se
;
166 tg
->rt_rq
= kzalloc(sizeof(rt_rq
) * nr_cpu_ids
, GFP_KERNEL
);
169 tg
->rt_se
= kzalloc(sizeof(rt_se
) * nr_cpu_ids
, GFP_KERNEL
);
173 init_rt_bandwidth(&tg
->rt_bandwidth
,
174 ktime_to_ns(def_rt_bandwidth
.rt_period
), 0);
176 for_each_possible_cpu(i
) {
177 rt_rq
= kzalloc_node(sizeof(struct rt_rq
),
178 GFP_KERNEL
, cpu_to_node(i
));
182 rt_se
= kzalloc_node(sizeof(struct sched_rt_entity
),
183 GFP_KERNEL
, cpu_to_node(i
));
187 init_rt_rq(rt_rq
, cpu_rq(i
));
188 rt_rq
->rt_runtime
= tg
->rt_bandwidth
.rt_runtime
;
189 init_tg_rt_entry(tg
, rt_rq
, rt_se
, i
, parent
->rt_se
[i
]);
200 #else /* CONFIG_RT_GROUP_SCHED */
202 #define rt_entity_is_task(rt_se) (1)
204 static inline struct task_struct
*rt_task_of(struct sched_rt_entity
*rt_se
)
206 return container_of(rt_se
, struct task_struct
, rt
);
209 static inline struct rq
*rq_of_rt_rq(struct rt_rq
*rt_rq
)
211 return container_of(rt_rq
, struct rq
, rt
);
214 static inline struct rt_rq
*rt_rq_of_se(struct sched_rt_entity
*rt_se
)
216 struct task_struct
*p
= rt_task_of(rt_se
);
217 struct rq
*rq
= task_rq(p
);
222 void free_rt_sched_group(struct task_group
*tg
) { }
224 int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
)
228 #endif /* CONFIG_RT_GROUP_SCHED */
232 static int pull_rt_task(struct rq
*this_rq
);
234 static inline bool need_pull_rt_task(struct rq
*rq
, struct task_struct
*prev
)
236 /* Try to pull RT tasks here if we lower this rq's prio */
237 return rq
->rt
.highest_prio
.curr
> prev
->prio
;
240 static inline int rt_overloaded(struct rq
*rq
)
242 return atomic_read(&rq
->rd
->rto_count
);
245 static inline void rt_set_overload(struct rq
*rq
)
250 cpumask_set_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
252 * Make sure the mask is visible before we set
253 * the overload count. That is checked to determine
254 * if we should look at the mask. It would be a shame
255 * if we looked at the mask, but the mask was not
258 * Matched by the barrier in pull_rt_task().
261 atomic_inc(&rq
->rd
->rto_count
);
264 static inline void rt_clear_overload(struct rq
*rq
)
269 /* the order here really doesn't matter */
270 atomic_dec(&rq
->rd
->rto_count
);
271 cpumask_clear_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
274 static void update_rt_migration(struct rt_rq
*rt_rq
)
276 if (rt_rq
->rt_nr_migratory
&& rt_rq
->rt_nr_total
> 1) {
277 if (!rt_rq
->overloaded
) {
278 rt_set_overload(rq_of_rt_rq(rt_rq
));
279 rt_rq
->overloaded
= 1;
281 } else if (rt_rq
->overloaded
) {
282 rt_clear_overload(rq_of_rt_rq(rt_rq
));
283 rt_rq
->overloaded
= 0;
287 static void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
289 struct task_struct
*p
;
291 if (!rt_entity_is_task(rt_se
))
294 p
= rt_task_of(rt_se
);
295 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
297 rt_rq
->rt_nr_total
++;
298 if (p
->nr_cpus_allowed
> 1)
299 rt_rq
->rt_nr_migratory
++;
301 update_rt_migration(rt_rq
);
304 static void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
306 struct task_struct
*p
;
308 if (!rt_entity_is_task(rt_se
))
311 p
= rt_task_of(rt_se
);
312 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
314 rt_rq
->rt_nr_total
--;
315 if (p
->nr_cpus_allowed
> 1)
316 rt_rq
->rt_nr_migratory
--;
318 update_rt_migration(rt_rq
);
321 static inline int has_pushable_tasks(struct rq
*rq
)
323 return !plist_head_empty(&rq
->rt
.pushable_tasks
);
326 static inline void set_post_schedule(struct rq
*rq
)
329 * We detect this state here so that we can avoid taking the RQ
330 * lock again later if there is no need to push
332 rq
->post_schedule
= has_pushable_tasks(rq
);
335 static void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
337 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
338 plist_node_init(&p
->pushable_tasks
, p
->prio
);
339 plist_add(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
341 /* Update the highest prio pushable task */
342 if (p
->prio
< rq
->rt
.highest_prio
.next
)
343 rq
->rt
.highest_prio
.next
= p
->prio
;
346 static void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
348 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
350 /* Update the new highest prio pushable task */
351 if (has_pushable_tasks(rq
)) {
352 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
353 struct task_struct
, pushable_tasks
);
354 rq
->rt
.highest_prio
.next
= p
->prio
;
356 rq
->rt
.highest_prio
.next
= MAX_RT_PRIO
;
361 static inline void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
365 static inline void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
370 void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
375 void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
379 static inline bool need_pull_rt_task(struct rq
*rq
, struct task_struct
*prev
)
384 static inline int pull_rt_task(struct rq
*this_rq
)
389 static inline void set_post_schedule(struct rq
*rq
)
392 #endif /* CONFIG_SMP */
394 static inline int on_rt_rq(struct sched_rt_entity
*rt_se
)
396 return !list_empty(&rt_se
->run_list
);
399 #ifdef CONFIG_RT_GROUP_SCHED
401 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
406 return rt_rq
->rt_runtime
;
409 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
411 return ktime_to_ns(rt_rq
->tg
->rt_bandwidth
.rt_period
);
414 typedef struct task_group
*rt_rq_iter_t
;
416 static inline struct task_group
*next_task_group(struct task_group
*tg
)
419 tg
= list_entry_rcu(tg
->list
.next
,
420 typeof(struct task_group
), list
);
421 } while (&tg
->list
!= &task_groups
&& task_group_is_autogroup(tg
));
423 if (&tg
->list
== &task_groups
)
429 #define for_each_rt_rq(rt_rq, iter, rq) \
430 for (iter = container_of(&task_groups, typeof(*iter), list); \
431 (iter = next_task_group(iter)) && \
432 (rt_rq = iter->rt_rq[cpu_of(rq)]);)
434 #define for_each_sched_rt_entity(rt_se) \
435 for (; rt_se; rt_se = rt_se->parent)
437 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
442 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
);
443 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
);
445 static void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
447 struct task_struct
*curr
= rq_of_rt_rq(rt_rq
)->curr
;
448 struct sched_rt_entity
*rt_se
;
450 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
452 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
454 if (rt_rq
->rt_nr_running
) {
455 if (rt_se
&& !on_rt_rq(rt_se
))
456 enqueue_rt_entity(rt_se
, false);
457 if (rt_rq
->highest_prio
.curr
< curr
->prio
)
462 static void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
464 struct sched_rt_entity
*rt_se
;
465 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
467 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
469 if (rt_se
&& on_rt_rq(rt_se
))
470 dequeue_rt_entity(rt_se
);
473 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
475 return rt_rq
->rt_throttled
&& !rt_rq
->rt_nr_boosted
;
478 static int rt_se_boosted(struct sched_rt_entity
*rt_se
)
480 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
481 struct task_struct
*p
;
484 return !!rt_rq
->rt_nr_boosted
;
486 p
= rt_task_of(rt_se
);
487 return p
->prio
!= p
->normal_prio
;
491 static inline const struct cpumask
*sched_rt_period_mask(void)
493 return this_rq()->rd
->span
;
496 static inline const struct cpumask
*sched_rt_period_mask(void)
498 return cpu_online_mask
;
503 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
505 return container_of(rt_b
, struct task_group
, rt_bandwidth
)->rt_rq
[cpu
];
508 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
510 return &rt_rq
->tg
->rt_bandwidth
;
513 #else /* !CONFIG_RT_GROUP_SCHED */
515 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
517 return rt_rq
->rt_runtime
;
520 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
522 return ktime_to_ns(def_rt_bandwidth
.rt_period
);
525 typedef struct rt_rq
*rt_rq_iter_t
;
527 #define for_each_rt_rq(rt_rq, iter, rq) \
528 for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
530 #define for_each_sched_rt_entity(rt_se) \
531 for (; rt_se; rt_se = NULL)
533 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
538 static inline void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
540 if (rt_rq
->rt_nr_running
)
541 resched_task(rq_of_rt_rq(rt_rq
)->curr
);
544 static inline void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
548 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
550 return rt_rq
->rt_throttled
;
553 static inline const struct cpumask
*sched_rt_period_mask(void)
555 return cpu_online_mask
;
559 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
561 return &cpu_rq(cpu
)->rt
;
564 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
566 return &def_rt_bandwidth
;
569 #endif /* CONFIG_RT_GROUP_SCHED */
571 bool sched_rt_bandwidth_account(struct rt_rq
*rt_rq
)
573 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
575 return (hrtimer_active(&rt_b
->rt_period_timer
) ||
576 rt_rq
->rt_time
< rt_b
->rt_runtime
);
581 * We ran out of runtime, see if we can borrow some from our neighbours.
583 static int do_balance_runtime(struct rt_rq
*rt_rq
)
585 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
586 struct root_domain
*rd
= rq_of_rt_rq(rt_rq
)->rd
;
587 int i
, weight
, more
= 0;
590 weight
= cpumask_weight(rd
->span
);
592 raw_spin_lock(&rt_b
->rt_runtime_lock
);
593 rt_period
= ktime_to_ns(rt_b
->rt_period
);
594 for_each_cpu(i
, rd
->span
) {
595 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
601 raw_spin_lock(&iter
->rt_runtime_lock
);
603 * Either all rqs have inf runtime and there's nothing to steal
604 * or __disable_runtime() below sets a specific rq to inf to
605 * indicate its been disabled and disalow stealing.
607 if (iter
->rt_runtime
== RUNTIME_INF
)
611 * From runqueues with spare time, take 1/n part of their
612 * spare time, but no more than our period.
614 diff
= iter
->rt_runtime
- iter
->rt_time
;
616 diff
= div_u64((u64
)diff
, weight
);
617 if (rt_rq
->rt_runtime
+ diff
> rt_period
)
618 diff
= rt_period
- rt_rq
->rt_runtime
;
619 iter
->rt_runtime
-= diff
;
620 rt_rq
->rt_runtime
+= diff
;
622 if (rt_rq
->rt_runtime
== rt_period
) {
623 raw_spin_unlock(&iter
->rt_runtime_lock
);
628 raw_spin_unlock(&iter
->rt_runtime_lock
);
630 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
636 * Ensure this RQ takes back all the runtime it lend to its neighbours.
638 static void __disable_runtime(struct rq
*rq
)
640 struct root_domain
*rd
= rq
->rd
;
644 if (unlikely(!scheduler_running
))
647 for_each_rt_rq(rt_rq
, iter
, rq
) {
648 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
652 raw_spin_lock(&rt_b
->rt_runtime_lock
);
653 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
655 * Either we're all inf and nobody needs to borrow, or we're
656 * already disabled and thus have nothing to do, or we have
657 * exactly the right amount of runtime to take out.
659 if (rt_rq
->rt_runtime
== RUNTIME_INF
||
660 rt_rq
->rt_runtime
== rt_b
->rt_runtime
)
662 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
665 * Calculate the difference between what we started out with
666 * and what we current have, that's the amount of runtime
667 * we lend and now have to reclaim.
669 want
= rt_b
->rt_runtime
- rt_rq
->rt_runtime
;
672 * Greedy reclaim, take back as much as we can.
674 for_each_cpu(i
, rd
->span
) {
675 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
679 * Can't reclaim from ourselves or disabled runqueues.
681 if (iter
== rt_rq
|| iter
->rt_runtime
== RUNTIME_INF
)
684 raw_spin_lock(&iter
->rt_runtime_lock
);
686 diff
= min_t(s64
, iter
->rt_runtime
, want
);
687 iter
->rt_runtime
-= diff
;
690 iter
->rt_runtime
-= want
;
693 raw_spin_unlock(&iter
->rt_runtime_lock
);
699 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
701 * We cannot be left wanting - that would mean some runtime
702 * leaked out of the system.
707 * Disable all the borrow logic by pretending we have inf
708 * runtime - in which case borrowing doesn't make sense.
710 rt_rq
->rt_runtime
= RUNTIME_INF
;
711 rt_rq
->rt_throttled
= 0;
712 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
713 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
717 static void __enable_runtime(struct rq
*rq
)
722 if (unlikely(!scheduler_running
))
726 * Reset each runqueue's bandwidth settings
728 for_each_rt_rq(rt_rq
, iter
, rq
) {
729 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
731 raw_spin_lock(&rt_b
->rt_runtime_lock
);
732 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
733 rt_rq
->rt_runtime
= rt_b
->rt_runtime
;
735 rt_rq
->rt_throttled
= 0;
736 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
737 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
741 static int balance_runtime(struct rt_rq
*rt_rq
)
745 if (!sched_feat(RT_RUNTIME_SHARE
))
748 if (rt_rq
->rt_time
> rt_rq
->rt_runtime
) {
749 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
750 more
= do_balance_runtime(rt_rq
);
751 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
756 #else /* !CONFIG_SMP */
757 static inline int balance_runtime(struct rt_rq
*rt_rq
)
761 #endif /* CONFIG_SMP */
763 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
)
765 int i
, idle
= 1, throttled
= 0;
766 const struct cpumask
*span
;
768 span
= sched_rt_period_mask();
769 #ifdef CONFIG_RT_GROUP_SCHED
771 * FIXME: isolated CPUs should really leave the root task group,
772 * whether they are isolcpus or were isolated via cpusets, lest
773 * the timer run on a CPU which does not service all runqueues,
774 * potentially leaving other CPUs indefinitely throttled. If
775 * isolation is really required, the user will turn the throttle
776 * off to kill the perturbations it causes anyway. Meanwhile,
777 * this maintains functionality for boot and/or troubleshooting.
779 if (rt_b
== &root_task_group
.rt_bandwidth
)
780 span
= cpu_online_mask
;
782 for_each_cpu(i
, span
) {
784 struct rt_rq
*rt_rq
= sched_rt_period_rt_rq(rt_b
, i
);
785 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
787 raw_spin_lock(&rq
->lock
);
788 if (rt_rq
->rt_time
) {
791 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
792 if (rt_rq
->rt_throttled
)
793 balance_runtime(rt_rq
);
794 runtime
= rt_rq
->rt_runtime
;
795 rt_rq
->rt_time
-= min(rt_rq
->rt_time
, overrun
*runtime
);
796 if (rt_rq
->rt_throttled
&& rt_rq
->rt_time
< runtime
) {
797 rt_rq
->rt_throttled
= 0;
801 * Force a clock update if the CPU was idle,
802 * lest wakeup -> unthrottle time accumulate.
804 if (rt_rq
->rt_nr_running
&& rq
->curr
== rq
->idle
)
805 rq
->skip_clock_update
= -1;
807 if (rt_rq
->rt_time
|| rt_rq
->rt_nr_running
)
809 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
810 } else if (rt_rq
->rt_nr_running
) {
812 if (!rt_rq_throttled(rt_rq
))
815 if (rt_rq
->rt_throttled
)
819 sched_rt_rq_enqueue(rt_rq
);
820 raw_spin_unlock(&rq
->lock
);
823 if (!throttled
&& (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
))
829 static inline int rt_se_prio(struct sched_rt_entity
*rt_se
)
831 #ifdef CONFIG_RT_GROUP_SCHED
832 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
835 return rt_rq
->highest_prio
.curr
;
838 return rt_task_of(rt_se
)->prio
;
841 static int sched_rt_runtime_exceeded(struct rt_rq
*rt_rq
)
843 u64 runtime
= sched_rt_runtime(rt_rq
);
845 if (rt_rq
->rt_throttled
)
846 return rt_rq_throttled(rt_rq
);
848 if (runtime
>= sched_rt_period(rt_rq
))
851 balance_runtime(rt_rq
);
852 runtime
= sched_rt_runtime(rt_rq
);
853 if (runtime
== RUNTIME_INF
)
856 if (rt_rq
->rt_time
> runtime
) {
857 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
860 * Don't actually throttle groups that have no runtime assigned
861 * but accrue some time due to boosting.
863 if (likely(rt_b
->rt_runtime
)) {
864 static bool once
= false;
866 rt_rq
->rt_throttled
= 1;
870 printk_sched("sched: RT throttling activated\n");
874 * In case we did anyway, make it go away,
875 * replenishment is a joke, since it will replenish us
881 if (rt_rq_throttled(rt_rq
)) {
882 sched_rt_rq_dequeue(rt_rq
);
891 * Update the current task's runtime statistics. Skip current tasks that
892 * are not in our scheduling class.
894 static void update_curr_rt(struct rq
*rq
)
896 struct task_struct
*curr
= rq
->curr
;
897 struct sched_rt_entity
*rt_se
= &curr
->rt
;
898 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
901 if (curr
->sched_class
!= &rt_sched_class
)
904 delta_exec
= rq_clock_task(rq
) - curr
->se
.exec_start
;
905 if (unlikely((s64
)delta_exec
<= 0))
908 schedstat_set(curr
->se
.statistics
.exec_max
,
909 max(curr
->se
.statistics
.exec_max
, delta_exec
));
911 curr
->se
.sum_exec_runtime
+= delta_exec
;
912 account_group_exec_runtime(curr
, delta_exec
);
914 curr
->se
.exec_start
= rq_clock_task(rq
);
915 cpuacct_charge(curr
, delta_exec
);
917 sched_rt_avg_update(rq
, delta_exec
);
919 if (!rt_bandwidth_enabled())
922 for_each_sched_rt_entity(rt_se
) {
923 rt_rq
= rt_rq_of_se(rt_se
);
925 if (sched_rt_runtime(rt_rq
) != RUNTIME_INF
) {
926 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
927 rt_rq
->rt_time
+= delta_exec
;
928 if (sched_rt_runtime_exceeded(rt_rq
))
930 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
935 #if defined CONFIG_SMP
938 inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
940 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
942 #ifdef CONFIG_RT_GROUP_SCHED
944 * Change rq's cpupri only if rt_rq is the top queue.
946 if (&rq
->rt
!= rt_rq
)
949 if (rq
->online
&& prio
< prev_prio
)
950 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, prio
);
954 dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
956 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
958 #ifdef CONFIG_RT_GROUP_SCHED
960 * Change rq's cpupri only if rt_rq is the top queue.
962 if (&rq
->rt
!= rt_rq
)
965 if (rq
->online
&& rt_rq
->highest_prio
.curr
!= prev_prio
)
966 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rt_rq
->highest_prio
.curr
);
969 #else /* CONFIG_SMP */
972 void inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
974 void dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
976 #endif /* CONFIG_SMP */
978 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
980 inc_rt_prio(struct rt_rq
*rt_rq
, int prio
)
982 int prev_prio
= rt_rq
->highest_prio
.curr
;
984 if (prio
< prev_prio
)
985 rt_rq
->highest_prio
.curr
= prio
;
987 inc_rt_prio_smp(rt_rq
, prio
, prev_prio
);
991 dec_rt_prio(struct rt_rq
*rt_rq
, int prio
)
993 int prev_prio
= rt_rq
->highest_prio
.curr
;
995 if (rt_rq
->rt_nr_running
) {
997 WARN_ON(prio
< prev_prio
);
1000 * This may have been our highest task, and therefore
1001 * we may have some recomputation to do
1003 if (prio
== prev_prio
) {
1004 struct rt_prio_array
*array
= &rt_rq
->active
;
1006 rt_rq
->highest_prio
.curr
=
1007 sched_find_first_bit(array
->bitmap
);
1011 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
1013 dec_rt_prio_smp(rt_rq
, prio
, prev_prio
);
1018 static inline void inc_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
1019 static inline void dec_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
1021 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
1023 #ifdef CONFIG_RT_GROUP_SCHED
1026 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1028 if (rt_se_boosted(rt_se
))
1029 rt_rq
->rt_nr_boosted
++;
1032 start_rt_bandwidth(&rt_rq
->tg
->rt_bandwidth
);
1036 dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1038 if (rt_se_boosted(rt_se
))
1039 rt_rq
->rt_nr_boosted
--;
1041 WARN_ON(!rt_rq
->rt_nr_running
&& rt_rq
->rt_nr_boosted
);
1044 #else /* CONFIG_RT_GROUP_SCHED */
1047 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1049 start_rt_bandwidth(&def_rt_bandwidth
);
1053 void dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
) {}
1055 #endif /* CONFIG_RT_GROUP_SCHED */
1058 void inc_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1060 int prio
= rt_se_prio(rt_se
);
1062 WARN_ON(!rt_prio(prio
));
1063 rt_rq
->rt_nr_running
++;
1065 inc_rt_prio(rt_rq
, prio
);
1066 inc_rt_migration(rt_se
, rt_rq
);
1067 inc_rt_group(rt_se
, rt_rq
);
1071 void dec_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1073 WARN_ON(!rt_prio(rt_se_prio(rt_se
)));
1074 WARN_ON(!rt_rq
->rt_nr_running
);
1075 rt_rq
->rt_nr_running
--;
1077 dec_rt_prio(rt_rq
, rt_se_prio(rt_se
));
1078 dec_rt_migration(rt_se
, rt_rq
);
1079 dec_rt_group(rt_se
, rt_rq
);
1082 static void __enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
1084 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1085 struct rt_prio_array
*array
= &rt_rq
->active
;
1086 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
1087 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1090 * Don't enqueue the group if its throttled, or when empty.
1091 * The latter is a consequence of the former when a child group
1092 * get throttled and the current group doesn't have any other
1095 if (group_rq
&& (rt_rq_throttled(group_rq
) || !group_rq
->rt_nr_running
))
1099 list_add(&rt_se
->run_list
, queue
);
1101 list_add_tail(&rt_se
->run_list
, queue
);
1102 __set_bit(rt_se_prio(rt_se
), array
->bitmap
);
1104 inc_rt_tasks(rt_se
, rt_rq
);
1107 static void __dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
1109 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1110 struct rt_prio_array
*array
= &rt_rq
->active
;
1112 list_del_init(&rt_se
->run_list
);
1113 if (list_empty(array
->queue
+ rt_se_prio(rt_se
)))
1114 __clear_bit(rt_se_prio(rt_se
), array
->bitmap
);
1116 dec_rt_tasks(rt_se
, rt_rq
);
1120 * Because the prio of an upper entry depends on the lower
1121 * entries, we must remove entries top - down.
1123 static void dequeue_rt_stack(struct sched_rt_entity
*rt_se
)
1125 struct sched_rt_entity
*back
= NULL
;
1127 for_each_sched_rt_entity(rt_se
) {
1132 for (rt_se
= back
; rt_se
; rt_se
= rt_se
->back
) {
1133 if (on_rt_rq(rt_se
))
1134 __dequeue_rt_entity(rt_se
);
1138 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
1140 dequeue_rt_stack(rt_se
);
1141 for_each_sched_rt_entity(rt_se
)
1142 __enqueue_rt_entity(rt_se
, head
);
1145 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
1147 dequeue_rt_stack(rt_se
);
1149 for_each_sched_rt_entity(rt_se
) {
1150 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
1152 if (rt_rq
&& rt_rq
->rt_nr_running
)
1153 __enqueue_rt_entity(rt_se
, false);
1158 * Adding/removing a task to/from a priority array:
1161 enqueue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1163 struct sched_rt_entity
*rt_se
= &p
->rt
;
1165 if (flags
& ENQUEUE_WAKEUP
)
1168 enqueue_rt_entity(rt_se
, flags
& ENQUEUE_HEAD
);
1170 if (!task_current(rq
, p
) && p
->nr_cpus_allowed
> 1)
1171 enqueue_pushable_task(rq
, p
);
1176 static void dequeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1178 struct sched_rt_entity
*rt_se
= &p
->rt
;
1181 dequeue_rt_entity(rt_se
);
1183 dequeue_pushable_task(rq
, p
);
1189 * Put task to the head or the end of the run list without the overhead of
1190 * dequeue followed by enqueue.
1193 requeue_rt_entity(struct rt_rq
*rt_rq
, struct sched_rt_entity
*rt_se
, int head
)
1195 if (on_rt_rq(rt_se
)) {
1196 struct rt_prio_array
*array
= &rt_rq
->active
;
1197 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1200 list_move(&rt_se
->run_list
, queue
);
1202 list_move_tail(&rt_se
->run_list
, queue
);
1206 static void requeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int head
)
1208 struct sched_rt_entity
*rt_se
= &p
->rt
;
1209 struct rt_rq
*rt_rq
;
1211 for_each_sched_rt_entity(rt_se
) {
1212 rt_rq
= rt_rq_of_se(rt_se
);
1213 requeue_rt_entity(rt_rq
, rt_se
, head
);
1217 static void yield_task_rt(struct rq
*rq
)
1219 requeue_task_rt(rq
, rq
->curr
, 0);
1223 static int find_lowest_rq(struct task_struct
*task
);
1226 select_task_rq_rt(struct task_struct
*p
, int cpu
, int sd_flag
, int flags
)
1228 struct task_struct
*curr
;
1231 if (p
->nr_cpus_allowed
== 1)
1234 /* For anything but wake ups, just return the task_cpu */
1235 if (sd_flag
!= SD_BALANCE_WAKE
&& sd_flag
!= SD_BALANCE_FORK
)
1241 curr
= ACCESS_ONCE(rq
->curr
); /* unlocked access */
1244 * If the current task on @p's runqueue is an RT task, then
1245 * try to see if we can wake this RT task up on another
1246 * runqueue. Otherwise simply start this RT task
1247 * on its current runqueue.
1249 * We want to avoid overloading runqueues. If the woken
1250 * task is a higher priority, then it will stay on this CPU
1251 * and the lower prio task should be moved to another CPU.
1252 * Even though this will probably make the lower prio task
1253 * lose its cache, we do not want to bounce a higher task
1254 * around just because it gave up its CPU, perhaps for a
1257 * For equal prio tasks, we just let the scheduler sort it out.
1259 * Otherwise, just let it ride on the affined RQ and the
1260 * post-schedule router will push the preempted task away
1262 * This test is optimistic, if we get it wrong the load-balancer
1263 * will have to sort it out.
1265 if (curr
&& unlikely(rt_task(curr
)) &&
1266 (curr
->nr_cpus_allowed
< 2 ||
1267 curr
->prio
<= p
->prio
)) {
1268 int target
= find_lowest_rq(p
);
1279 static void check_preempt_equal_prio(struct rq
*rq
, struct task_struct
*p
)
1281 if (rq
->curr
->nr_cpus_allowed
== 1)
1284 if (p
->nr_cpus_allowed
!= 1
1285 && cpupri_find(&rq
->rd
->cpupri
, p
, NULL
))
1288 if (!cpupri_find(&rq
->rd
->cpupri
, rq
->curr
, NULL
))
1292 * There appears to be other cpus that can accept
1293 * current and none to run 'p', so lets reschedule
1294 * to try and push current away:
1296 requeue_task_rt(rq
, p
, 1);
1297 resched_task(rq
->curr
);
1300 #endif /* CONFIG_SMP */
1303 * Preempt the current task with a newly woken task if needed:
1305 static void check_preempt_curr_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1307 if (p
->prio
< rq
->curr
->prio
) {
1308 resched_task(rq
->curr
);
1316 * - the newly woken task is of equal priority to the current task
1317 * - the newly woken task is non-migratable while current is migratable
1318 * - current will be preempted on the next reschedule
1320 * we should check to see if current can readily move to a different
1321 * cpu. If so, we will reschedule to allow the push logic to try
1322 * to move current somewhere else, making room for our non-migratable
1325 if (p
->prio
== rq
->curr
->prio
&& !test_tsk_need_resched(rq
->curr
))
1326 check_preempt_equal_prio(rq
, p
);
1330 static struct sched_rt_entity
*pick_next_rt_entity(struct rq
*rq
,
1331 struct rt_rq
*rt_rq
)
1333 struct rt_prio_array
*array
= &rt_rq
->active
;
1334 struct sched_rt_entity
*next
= NULL
;
1335 struct list_head
*queue
;
1338 idx
= sched_find_first_bit(array
->bitmap
);
1339 BUG_ON(idx
>= MAX_RT_PRIO
);
1341 queue
= array
->queue
+ idx
;
1342 next
= list_entry(queue
->next
, struct sched_rt_entity
, run_list
);
1347 static struct task_struct
*_pick_next_task_rt(struct rq
*rq
)
1349 struct sched_rt_entity
*rt_se
;
1350 struct task_struct
*p
;
1351 struct rt_rq
*rt_rq
= &rq
->rt
;
1354 rt_se
= pick_next_rt_entity(rq
, rt_rq
);
1356 rt_rq
= group_rt_rq(rt_se
);
1359 p
= rt_task_of(rt_se
);
1360 p
->se
.exec_start
= rq_clock_task(rq
);
1365 static struct task_struct
*
1366 pick_next_task_rt(struct rq
*rq
, struct task_struct
*prev
)
1368 struct task_struct
*p
;
1369 struct rt_rq
*rt_rq
= &rq
->rt
;
1371 if (need_pull_rt_task(rq
, prev
)) {
1374 * pull_rt_task() can drop (and re-acquire) rq->lock; this
1375 * means a dl task can slip in, in which case we need to
1376 * re-start task selection.
1378 if (unlikely(rq
->dl
.dl_nr_running
))
1382 if (!rt_rq
->rt_nr_running
)
1385 if (rt_rq_throttled(rt_rq
))
1388 put_prev_task(rq
, prev
);
1390 p
= _pick_next_task_rt(rq
);
1392 /* The running task is never eligible for pushing */
1394 dequeue_pushable_task(rq
, p
);
1396 set_post_schedule(rq
);
1401 static void put_prev_task_rt(struct rq
*rq
, struct task_struct
*p
)
1406 * The previous task needs to be made eligible for pushing
1407 * if it is still active
1409 if (on_rt_rq(&p
->rt
) && p
->nr_cpus_allowed
> 1)
1410 enqueue_pushable_task(rq
, p
);
1415 /* Only try algorithms three times */
1416 #define RT_MAX_TRIES 3
1418 static int pick_rt_task(struct rq
*rq
, struct task_struct
*p
, int cpu
)
1420 if (!task_running(rq
, p
) &&
1421 cpumask_test_cpu(cpu
, tsk_cpus_allowed(p
)))
1427 * Return the highest pushable rq's task, which is suitable to be executed
1428 * on the cpu, NULL otherwise
1430 static struct task_struct
*pick_highest_pushable_task(struct rq
*rq
, int cpu
)
1432 struct plist_head
*head
= &rq
->rt
.pushable_tasks
;
1433 struct task_struct
*p
;
1435 if (!has_pushable_tasks(rq
))
1438 plist_for_each_entry(p
, head
, pushable_tasks
) {
1439 if (pick_rt_task(rq
, p
, cpu
))
1446 static DEFINE_PER_CPU(cpumask_var_t
, local_cpu_mask
);
1448 static int find_lowest_rq(struct task_struct
*task
)
1450 struct sched_domain
*sd
;
1451 struct cpumask
*lowest_mask
= __get_cpu_var(local_cpu_mask
);
1452 int this_cpu
= smp_processor_id();
1453 int cpu
= task_cpu(task
);
1455 /* Make sure the mask is initialized first */
1456 if (unlikely(!lowest_mask
))
1459 if (task
->nr_cpus_allowed
== 1)
1460 return -1; /* No other targets possible */
1462 if (!cpupri_find(&task_rq(task
)->rd
->cpupri
, task
, lowest_mask
))
1463 return -1; /* No targets found */
1466 * At this point we have built a mask of cpus representing the
1467 * lowest priority tasks in the system. Now we want to elect
1468 * the best one based on our affinity and topology.
1470 * We prioritize the last cpu that the task executed on since
1471 * it is most likely cache-hot in that location.
1473 if (cpumask_test_cpu(cpu
, lowest_mask
))
1477 * Otherwise, we consult the sched_domains span maps to figure
1478 * out which cpu is logically closest to our hot cache data.
1480 if (!cpumask_test_cpu(this_cpu
, lowest_mask
))
1481 this_cpu
= -1; /* Skip this_cpu opt if not among lowest */
1484 for_each_domain(cpu
, sd
) {
1485 if (sd
->flags
& SD_WAKE_AFFINE
) {
1489 * "this_cpu" is cheaper to preempt than a
1492 if (this_cpu
!= -1 &&
1493 cpumask_test_cpu(this_cpu
, sched_domain_span(sd
))) {
1498 best_cpu
= cpumask_first_and(lowest_mask
,
1499 sched_domain_span(sd
));
1500 if (best_cpu
< nr_cpu_ids
) {
1509 * And finally, if there were no matches within the domains
1510 * just give the caller *something* to work with from the compatible
1516 cpu
= cpumask_any(lowest_mask
);
1517 if (cpu
< nr_cpu_ids
)
1522 /* Will lock the rq it finds */
1523 static struct rq
*find_lock_lowest_rq(struct task_struct
*task
, struct rq
*rq
)
1525 struct rq
*lowest_rq
= NULL
;
1529 for (tries
= 0; tries
< RT_MAX_TRIES
; tries
++) {
1530 cpu
= find_lowest_rq(task
);
1532 if ((cpu
== -1) || (cpu
== rq
->cpu
))
1535 lowest_rq
= cpu_rq(cpu
);
1537 /* if the prio of this runqueue changed, try again */
1538 if (double_lock_balance(rq
, lowest_rq
)) {
1540 * We had to unlock the run queue. In
1541 * the mean time, task could have
1542 * migrated already or had its affinity changed.
1543 * Also make sure that it wasn't scheduled on its rq.
1545 if (unlikely(task_rq(task
) != rq
||
1546 !cpumask_test_cpu(lowest_rq
->cpu
,
1547 tsk_cpus_allowed(task
)) ||
1548 task_running(rq
, task
) ||
1551 double_unlock_balance(rq
, lowest_rq
);
1557 /* If this rq is still suitable use it. */
1558 if (lowest_rq
->rt
.highest_prio
.curr
> task
->prio
)
1562 double_unlock_balance(rq
, lowest_rq
);
1569 static struct task_struct
*pick_next_pushable_task(struct rq
*rq
)
1571 struct task_struct
*p
;
1573 if (!has_pushable_tasks(rq
))
1576 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
1577 struct task_struct
, pushable_tasks
);
1579 BUG_ON(rq
->cpu
!= task_cpu(p
));
1580 BUG_ON(task_current(rq
, p
));
1581 BUG_ON(p
->nr_cpus_allowed
<= 1);
1584 BUG_ON(!rt_task(p
));
1590 * If the current CPU has more than one RT task, see if the non
1591 * running task can migrate over to a CPU that is running a task
1592 * of lesser priority.
1594 static int push_rt_task(struct rq
*rq
)
1596 struct task_struct
*next_task
;
1597 struct rq
*lowest_rq
;
1600 if (!rq
->rt
.overloaded
)
1603 next_task
= pick_next_pushable_task(rq
);
1608 if (unlikely(next_task
== rq
->curr
)) {
1614 * It's possible that the next_task slipped in of
1615 * higher priority than current. If that's the case
1616 * just reschedule current.
1618 if (unlikely(next_task
->prio
< rq
->curr
->prio
)) {
1619 resched_task(rq
->curr
);
1623 /* We might release rq lock */
1624 get_task_struct(next_task
);
1626 /* find_lock_lowest_rq locks the rq if found */
1627 lowest_rq
= find_lock_lowest_rq(next_task
, rq
);
1629 struct task_struct
*task
;
1631 * find_lock_lowest_rq releases rq->lock
1632 * so it is possible that next_task has migrated.
1634 * We need to make sure that the task is still on the same
1635 * run-queue and is also still the next task eligible for
1638 task
= pick_next_pushable_task(rq
);
1639 if (task_cpu(next_task
) == rq
->cpu
&& task
== next_task
) {
1641 * The task hasn't migrated, and is still the next
1642 * eligible task, but we failed to find a run-queue
1643 * to push it to. Do not retry in this case, since
1644 * other cpus will pull from us when ready.
1650 /* No more tasks, just exit */
1654 * Something has shifted, try again.
1656 put_task_struct(next_task
);
1661 deactivate_task(rq
, next_task
, 0);
1662 set_task_cpu(next_task
, lowest_rq
->cpu
);
1663 activate_task(lowest_rq
, next_task
, 0);
1666 resched_task(lowest_rq
->curr
);
1668 double_unlock_balance(rq
, lowest_rq
);
1671 put_task_struct(next_task
);
1676 static void push_rt_tasks(struct rq
*rq
)
1678 /* push_rt_task will return true if it moved an RT */
1679 while (push_rt_task(rq
))
1683 static int pull_rt_task(struct rq
*this_rq
)
1685 int this_cpu
= this_rq
->cpu
, ret
= 0, cpu
;
1686 struct task_struct
*p
;
1689 if (likely(!rt_overloaded(this_rq
)))
1693 * Match the barrier from rt_set_overloaded; this guarantees that if we
1694 * see overloaded we must also see the rto_mask bit.
1698 for_each_cpu(cpu
, this_rq
->rd
->rto_mask
) {
1699 if (this_cpu
== cpu
)
1702 src_rq
= cpu_rq(cpu
);
1705 * Don't bother taking the src_rq->lock if the next highest
1706 * task is known to be lower-priority than our current task.
1707 * This may look racy, but if this value is about to go
1708 * logically higher, the src_rq will push this task away.
1709 * And if its going logically lower, we do not care
1711 if (src_rq
->rt
.highest_prio
.next
>=
1712 this_rq
->rt
.highest_prio
.curr
)
1716 * We can potentially drop this_rq's lock in
1717 * double_lock_balance, and another CPU could
1720 double_lock_balance(this_rq
, src_rq
);
1723 * We can pull only a task, which is pushable
1724 * on its rq, and no others.
1726 p
= pick_highest_pushable_task(src_rq
, this_cpu
);
1729 * Do we have an RT task that preempts
1730 * the to-be-scheduled task?
1732 if (p
&& (p
->prio
< this_rq
->rt
.highest_prio
.curr
)) {
1733 WARN_ON(p
== src_rq
->curr
);
1737 * There's a chance that p is higher in priority
1738 * than what's currently running on its cpu.
1739 * This is just that p is wakeing up and hasn't
1740 * had a chance to schedule. We only pull
1741 * p if it is lower in priority than the
1742 * current task on the run queue
1744 if (p
->prio
< src_rq
->curr
->prio
)
1749 deactivate_task(src_rq
, p
, 0);
1750 set_task_cpu(p
, this_cpu
);
1751 activate_task(this_rq
, p
, 0);
1753 * We continue with the search, just in
1754 * case there's an even higher prio task
1755 * in another runqueue. (low likelihood
1760 double_unlock_balance(this_rq
, src_rq
);
1766 static void post_schedule_rt(struct rq
*rq
)
1772 * If we are not running and we are not going to reschedule soon, we should
1773 * try to push tasks away now
1775 static void task_woken_rt(struct rq
*rq
, struct task_struct
*p
)
1777 if (!task_running(rq
, p
) &&
1778 !test_tsk_need_resched(rq
->curr
) &&
1779 has_pushable_tasks(rq
) &&
1780 p
->nr_cpus_allowed
> 1 &&
1781 (dl_task(rq
->curr
) || rt_task(rq
->curr
)) &&
1782 (rq
->curr
->nr_cpus_allowed
< 2 ||
1783 rq
->curr
->prio
<= p
->prio
))
1787 static void set_cpus_allowed_rt(struct task_struct
*p
,
1788 const struct cpumask
*new_mask
)
1793 BUG_ON(!rt_task(p
));
1798 weight
= cpumask_weight(new_mask
);
1801 * Only update if the process changes its state from whether it
1802 * can migrate or not.
1804 if ((p
->nr_cpus_allowed
> 1) == (weight
> 1))
1810 * The process used to be able to migrate OR it can now migrate
1813 if (!task_current(rq
, p
))
1814 dequeue_pushable_task(rq
, p
);
1815 BUG_ON(!rq
->rt
.rt_nr_migratory
);
1816 rq
->rt
.rt_nr_migratory
--;
1818 if (!task_current(rq
, p
))
1819 enqueue_pushable_task(rq
, p
);
1820 rq
->rt
.rt_nr_migratory
++;
1823 update_rt_migration(&rq
->rt
);
1826 /* Assumes rq->lock is held */
1827 static void rq_online_rt(struct rq
*rq
)
1829 if (rq
->rt
.overloaded
)
1830 rt_set_overload(rq
);
1832 __enable_runtime(rq
);
1834 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rq
->rt
.highest_prio
.curr
);
1837 /* Assumes rq->lock is held */
1838 static void rq_offline_rt(struct rq
*rq
)
1840 if (rq
->rt
.overloaded
)
1841 rt_clear_overload(rq
);
1843 __disable_runtime(rq
);
1845 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, CPUPRI_INVALID
);
1849 * When switch from the rt queue, we bring ourselves to a position
1850 * that we might want to pull RT tasks from other runqueues.
1852 static void switched_from_rt(struct rq
*rq
, struct task_struct
*p
)
1855 * If there are other RT tasks then we will reschedule
1856 * and the scheduling of the other RT tasks will handle
1857 * the balancing. But if we are the last RT task
1858 * we may need to handle the pulling of RT tasks
1861 if (!p
->on_rq
|| rq
->rt
.rt_nr_running
)
1864 if (pull_rt_task(rq
))
1865 resched_task(rq
->curr
);
1868 void __init
init_sched_rt_class(void)
1872 for_each_possible_cpu(i
) {
1873 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask
, i
),
1874 GFP_KERNEL
, cpu_to_node(i
));
1877 #endif /* CONFIG_SMP */
1880 * When switching a task to RT, we may overload the runqueue
1881 * with RT tasks. In this case we try to push them off to
1884 static void switched_to_rt(struct rq
*rq
, struct task_struct
*p
)
1886 int check_resched
= 1;
1889 * If we are already running, then there's nothing
1890 * that needs to be done. But if we are not running
1891 * we may need to preempt the current running task.
1892 * If that current running task is also an RT task
1893 * then see if we can move to another run queue.
1895 if (p
->on_rq
&& rq
->curr
!= p
) {
1897 if (rq
->rt
.overloaded
&& push_rt_task(rq
) &&
1898 /* Don't resched if we changed runqueues */
1901 #endif /* CONFIG_SMP */
1902 if (check_resched
&& p
->prio
< rq
->curr
->prio
)
1903 resched_task(rq
->curr
);
1908 * Priority of the task has changed. This may cause
1909 * us to initiate a push or pull.
1912 prio_changed_rt(struct rq
*rq
, struct task_struct
*p
, int oldprio
)
1917 if (rq
->curr
== p
) {
1920 * If our priority decreases while running, we
1921 * may need to pull tasks to this runqueue.
1923 if (oldprio
< p
->prio
)
1926 * If there's a higher priority task waiting to run
1927 * then reschedule. Note, the above pull_rt_task
1928 * can release the rq lock and p could migrate.
1929 * Only reschedule if p is still on the same runqueue.
1931 if (p
->prio
> rq
->rt
.highest_prio
.curr
&& rq
->curr
== p
)
1934 /* For UP simply resched on drop of prio */
1935 if (oldprio
< p
->prio
)
1937 #endif /* CONFIG_SMP */
1940 * This task is not running, but if it is
1941 * greater than the current running task
1944 if (p
->prio
< rq
->curr
->prio
)
1945 resched_task(rq
->curr
);
1949 static void watchdog(struct rq
*rq
, struct task_struct
*p
)
1951 unsigned long soft
, hard
;
1953 /* max may change after cur was read, this will be fixed next tick */
1954 soft
= task_rlimit(p
, RLIMIT_RTTIME
);
1955 hard
= task_rlimit_max(p
, RLIMIT_RTTIME
);
1957 if (soft
!= RLIM_INFINITY
) {
1960 if (p
->rt
.watchdog_stamp
!= jiffies
) {
1962 p
->rt
.watchdog_stamp
= jiffies
;
1965 next
= DIV_ROUND_UP(min(soft
, hard
), USEC_PER_SEC
/HZ
);
1966 if (p
->rt
.timeout
> next
)
1967 p
->cputime_expires
.sched_exp
= p
->se
.sum_exec_runtime
;
1971 static void task_tick_rt(struct rq
*rq
, struct task_struct
*p
, int queued
)
1973 struct sched_rt_entity
*rt_se
= &p
->rt
;
1980 * RR tasks need a special form of timeslice management.
1981 * FIFO tasks have no timeslices.
1983 if (p
->policy
!= SCHED_RR
)
1986 if (--p
->rt
.time_slice
)
1989 p
->rt
.time_slice
= sched_rr_timeslice
;
1992 * Requeue to the end of queue if we (and all of our ancestors) are not
1993 * the only element on the queue
1995 for_each_sched_rt_entity(rt_se
) {
1996 if (rt_se
->run_list
.prev
!= rt_se
->run_list
.next
) {
1997 requeue_task_rt(rq
, p
, 0);
1998 set_tsk_need_resched(p
);
2004 static void set_curr_task_rt(struct rq
*rq
)
2006 struct task_struct
*p
= rq
->curr
;
2008 p
->se
.exec_start
= rq_clock_task(rq
);
2010 /* The running task is never eligible for pushing */
2011 dequeue_pushable_task(rq
, p
);
2014 static unsigned int get_rr_interval_rt(struct rq
*rq
, struct task_struct
*task
)
2017 * Time slice is 0 for SCHED_FIFO tasks
2019 if (task
->policy
== SCHED_RR
)
2020 return sched_rr_timeslice
;
2025 const struct sched_class rt_sched_class
= {
2026 .next
= &fair_sched_class
,
2027 .enqueue_task
= enqueue_task_rt
,
2028 .dequeue_task
= dequeue_task_rt
,
2029 .yield_task
= yield_task_rt
,
2031 .check_preempt_curr
= check_preempt_curr_rt
,
2033 .pick_next_task
= pick_next_task_rt
,
2034 .put_prev_task
= put_prev_task_rt
,
2037 .select_task_rq
= select_task_rq_rt
,
2039 .set_cpus_allowed
= set_cpus_allowed_rt
,
2040 .rq_online
= rq_online_rt
,
2041 .rq_offline
= rq_offline_rt
,
2042 .post_schedule
= post_schedule_rt
,
2043 .task_woken
= task_woken_rt
,
2044 .switched_from
= switched_from_rt
,
2047 .set_curr_task
= set_curr_task_rt
,
2048 .task_tick
= task_tick_rt
,
2050 .get_rr_interval
= get_rr_interval_rt
,
2052 .prio_changed
= prio_changed_rt
,
2053 .switched_to
= switched_to_rt
,
2056 #ifdef CONFIG_SCHED_DEBUG
2057 extern void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
);
2059 void print_rt_stats(struct seq_file
*m
, int cpu
)
2062 struct rt_rq
*rt_rq
;
2065 for_each_rt_rq(rt_rq
, iter
, cpu_rq(cpu
))
2066 print_rt_rq(m
, cpu
, rt_rq
);
2069 #endif /* CONFIG_SCHED_DEBUG */