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 inline int rt_overloaded(struct rq
*rq
)
234 return atomic_read(&rq
->rd
->rto_count
);
237 static inline void rt_set_overload(struct rq
*rq
)
242 cpumask_set_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
244 * Make sure the mask is visible before we set
245 * the overload count. That is checked to determine
246 * if we should look at the mask. It would be a shame
247 * if we looked at the mask, but the mask was not
251 atomic_inc(&rq
->rd
->rto_count
);
254 static inline void rt_clear_overload(struct rq
*rq
)
259 /* the order here really doesn't matter */
260 atomic_dec(&rq
->rd
->rto_count
);
261 cpumask_clear_cpu(rq
->cpu
, rq
->rd
->rto_mask
);
264 static void update_rt_migration(struct rt_rq
*rt_rq
)
266 if (rt_rq
->rt_nr_migratory
&& rt_rq
->rt_nr_total
> 1) {
267 if (!rt_rq
->overloaded
) {
268 rt_set_overload(rq_of_rt_rq(rt_rq
));
269 rt_rq
->overloaded
= 1;
271 } else if (rt_rq
->overloaded
) {
272 rt_clear_overload(rq_of_rt_rq(rt_rq
));
273 rt_rq
->overloaded
= 0;
277 static void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
279 struct task_struct
*p
;
281 if (!rt_entity_is_task(rt_se
))
284 p
= rt_task_of(rt_se
);
285 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
287 rt_rq
->rt_nr_total
++;
288 if (p
->nr_cpus_allowed
> 1)
289 rt_rq
->rt_nr_migratory
++;
291 update_rt_migration(rt_rq
);
294 static void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
296 struct task_struct
*p
;
298 if (!rt_entity_is_task(rt_se
))
301 p
= rt_task_of(rt_se
);
302 rt_rq
= &rq_of_rt_rq(rt_rq
)->rt
;
304 rt_rq
->rt_nr_total
--;
305 if (p
->nr_cpus_allowed
> 1)
306 rt_rq
->rt_nr_migratory
--;
308 update_rt_migration(rt_rq
);
311 static inline int has_pushable_tasks(struct rq
*rq
)
313 return !plist_head_empty(&rq
->rt
.pushable_tasks
);
316 static void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
318 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
319 plist_node_init(&p
->pushable_tasks
, p
->prio
);
320 plist_add(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
322 /* Update the highest prio pushable task */
323 if (p
->prio
< rq
->rt
.highest_prio
.next
)
324 rq
->rt
.highest_prio
.next
= p
->prio
;
327 static void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
329 plist_del(&p
->pushable_tasks
, &rq
->rt
.pushable_tasks
);
331 /* Update the new highest prio pushable task */
332 if (has_pushable_tasks(rq
)) {
333 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
334 struct task_struct
, pushable_tasks
);
335 rq
->rt
.highest_prio
.next
= p
->prio
;
337 rq
->rt
.highest_prio
.next
= MAX_RT_PRIO
;
342 static inline void enqueue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
346 static inline void dequeue_pushable_task(struct rq
*rq
, struct task_struct
*p
)
351 void inc_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
356 void dec_rt_migration(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
360 #endif /* CONFIG_SMP */
362 static inline int on_rt_rq(struct sched_rt_entity
*rt_se
)
364 return !list_empty(&rt_se
->run_list
);
367 #ifdef CONFIG_RT_GROUP_SCHED
369 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
374 return rt_rq
->rt_runtime
;
377 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
379 return ktime_to_ns(rt_rq
->tg
->rt_bandwidth
.rt_period
);
382 typedef struct task_group
*rt_rq_iter_t
;
384 static inline struct task_group
*next_task_group(struct task_group
*tg
)
387 tg
= list_entry_rcu(tg
->list
.next
,
388 typeof(struct task_group
), list
);
389 } while (&tg
->list
!= &task_groups
&& task_group_is_autogroup(tg
));
391 if (&tg
->list
== &task_groups
)
397 #define for_each_rt_rq(rt_rq, iter, rq) \
398 for (iter = container_of(&task_groups, typeof(*iter), list); \
399 (iter = next_task_group(iter)) && \
400 (rt_rq = iter->rt_rq[cpu_of(rq)]);)
402 static inline void list_add_leaf_rt_rq(struct rt_rq
*rt_rq
)
404 list_add_rcu(&rt_rq
->leaf_rt_rq_list
,
405 &rq_of_rt_rq(rt_rq
)->leaf_rt_rq_list
);
408 static inline void list_del_leaf_rt_rq(struct rt_rq
*rt_rq
)
410 list_del_rcu(&rt_rq
->leaf_rt_rq_list
);
413 #define for_each_leaf_rt_rq(rt_rq, rq) \
414 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
416 #define for_each_sched_rt_entity(rt_se) \
417 for (; rt_se; rt_se = rt_se->parent)
419 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
424 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
);
425 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
);
427 static void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
429 struct task_struct
*curr
= rq_of_rt_rq(rt_rq
)->curr
;
430 struct sched_rt_entity
*rt_se
;
432 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
434 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
436 if (rt_rq
->rt_nr_running
) {
437 if (rt_se
&& !on_rt_rq(rt_se
))
438 enqueue_rt_entity(rt_se
, false);
439 if (rt_rq
->highest_prio
.curr
< curr
->prio
)
444 static void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
446 struct sched_rt_entity
*rt_se
;
447 int cpu
= cpu_of(rq_of_rt_rq(rt_rq
));
449 rt_se
= rt_rq
->tg
->rt_se
[cpu
];
451 if (rt_se
&& on_rt_rq(rt_se
))
452 dequeue_rt_entity(rt_se
);
455 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
457 return rt_rq
->rt_throttled
&& !rt_rq
->rt_nr_boosted
;
460 static int rt_se_boosted(struct sched_rt_entity
*rt_se
)
462 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
463 struct task_struct
*p
;
466 return !!rt_rq
->rt_nr_boosted
;
468 p
= rt_task_of(rt_se
);
469 return p
->prio
!= p
->normal_prio
;
473 static inline const struct cpumask
*sched_rt_period_mask(void)
475 return cpu_rq(smp_processor_id())->rd
->span
;
478 static inline const struct cpumask
*sched_rt_period_mask(void)
480 return cpu_online_mask
;
485 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
487 return container_of(rt_b
, struct task_group
, rt_bandwidth
)->rt_rq
[cpu
];
490 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
492 return &rt_rq
->tg
->rt_bandwidth
;
495 #else /* !CONFIG_RT_GROUP_SCHED */
497 static inline u64
sched_rt_runtime(struct rt_rq
*rt_rq
)
499 return rt_rq
->rt_runtime
;
502 static inline u64
sched_rt_period(struct rt_rq
*rt_rq
)
504 return ktime_to_ns(def_rt_bandwidth
.rt_period
);
507 typedef struct rt_rq
*rt_rq_iter_t
;
509 #define for_each_rt_rq(rt_rq, iter, rq) \
510 for ((void) iter, rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
512 static inline void list_add_leaf_rt_rq(struct rt_rq
*rt_rq
)
516 static inline void list_del_leaf_rt_rq(struct rt_rq
*rt_rq
)
520 #define for_each_leaf_rt_rq(rt_rq, rq) \
521 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
523 #define for_each_sched_rt_entity(rt_se) \
524 for (; rt_se; rt_se = NULL)
526 static inline struct rt_rq
*group_rt_rq(struct sched_rt_entity
*rt_se
)
531 static inline void sched_rt_rq_enqueue(struct rt_rq
*rt_rq
)
533 if (rt_rq
->rt_nr_running
)
534 resched_task(rq_of_rt_rq(rt_rq
)->curr
);
537 static inline void sched_rt_rq_dequeue(struct rt_rq
*rt_rq
)
541 static inline int rt_rq_throttled(struct rt_rq
*rt_rq
)
543 return rt_rq
->rt_throttled
;
546 static inline const struct cpumask
*sched_rt_period_mask(void)
548 return cpu_online_mask
;
552 struct rt_rq
*sched_rt_period_rt_rq(struct rt_bandwidth
*rt_b
, int cpu
)
554 return &cpu_rq(cpu
)->rt
;
557 static inline struct rt_bandwidth
*sched_rt_bandwidth(struct rt_rq
*rt_rq
)
559 return &def_rt_bandwidth
;
562 #endif /* CONFIG_RT_GROUP_SCHED */
566 * We ran out of runtime, see if we can borrow some from our neighbours.
568 static int do_balance_runtime(struct rt_rq
*rt_rq
)
570 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
571 struct root_domain
*rd
= rq_of_rt_rq(rt_rq
)->rd
;
572 int i
, weight
, more
= 0;
575 weight
= cpumask_weight(rd
->span
);
577 raw_spin_lock(&rt_b
->rt_runtime_lock
);
578 rt_period
= ktime_to_ns(rt_b
->rt_period
);
579 for_each_cpu(i
, rd
->span
) {
580 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
586 raw_spin_lock(&iter
->rt_runtime_lock
);
588 * Either all rqs have inf runtime and there's nothing to steal
589 * or __disable_runtime() below sets a specific rq to inf to
590 * indicate its been disabled and disalow stealing.
592 if (iter
->rt_runtime
== RUNTIME_INF
)
596 * From runqueues with spare time, take 1/n part of their
597 * spare time, but no more than our period.
599 diff
= iter
->rt_runtime
- iter
->rt_time
;
601 diff
= div_u64((u64
)diff
, weight
);
602 if (rt_rq
->rt_runtime
+ diff
> rt_period
)
603 diff
= rt_period
- rt_rq
->rt_runtime
;
604 iter
->rt_runtime
-= diff
;
605 rt_rq
->rt_runtime
+= diff
;
607 if (rt_rq
->rt_runtime
== rt_period
) {
608 raw_spin_unlock(&iter
->rt_runtime_lock
);
613 raw_spin_unlock(&iter
->rt_runtime_lock
);
615 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
621 * Ensure this RQ takes back all the runtime it lend to its neighbours.
623 static void __disable_runtime(struct rq
*rq
)
625 struct root_domain
*rd
= rq
->rd
;
629 if (unlikely(!scheduler_running
))
632 for_each_rt_rq(rt_rq
, iter
, rq
) {
633 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
637 raw_spin_lock(&rt_b
->rt_runtime_lock
);
638 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
640 * Either we're all inf and nobody needs to borrow, or we're
641 * already disabled and thus have nothing to do, or we have
642 * exactly the right amount of runtime to take out.
644 if (rt_rq
->rt_runtime
== RUNTIME_INF
||
645 rt_rq
->rt_runtime
== rt_b
->rt_runtime
)
647 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
650 * Calculate the difference between what we started out with
651 * and what we current have, that's the amount of runtime
652 * we lend and now have to reclaim.
654 want
= rt_b
->rt_runtime
- rt_rq
->rt_runtime
;
657 * Greedy reclaim, take back as much as we can.
659 for_each_cpu(i
, rd
->span
) {
660 struct rt_rq
*iter
= sched_rt_period_rt_rq(rt_b
, i
);
664 * Can't reclaim from ourselves or disabled runqueues.
666 if (iter
== rt_rq
|| iter
->rt_runtime
== RUNTIME_INF
)
669 raw_spin_lock(&iter
->rt_runtime_lock
);
671 diff
= min_t(s64
, iter
->rt_runtime
, want
);
672 iter
->rt_runtime
-= diff
;
675 iter
->rt_runtime
-= want
;
678 raw_spin_unlock(&iter
->rt_runtime_lock
);
684 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
686 * We cannot be left wanting - that would mean some runtime
687 * leaked out of the system.
692 * Disable all the borrow logic by pretending we have inf
693 * runtime - in which case borrowing doesn't make sense.
695 rt_rq
->rt_runtime
= RUNTIME_INF
;
696 rt_rq
->rt_throttled
= 0;
697 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
698 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
702 static void __enable_runtime(struct rq
*rq
)
707 if (unlikely(!scheduler_running
))
711 * Reset each runqueue's bandwidth settings
713 for_each_rt_rq(rt_rq
, iter
, rq
) {
714 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
716 raw_spin_lock(&rt_b
->rt_runtime_lock
);
717 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
718 rt_rq
->rt_runtime
= rt_b
->rt_runtime
;
720 rt_rq
->rt_throttled
= 0;
721 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
722 raw_spin_unlock(&rt_b
->rt_runtime_lock
);
726 static int balance_runtime(struct rt_rq
*rt_rq
)
730 if (!sched_feat(RT_RUNTIME_SHARE
))
733 if (rt_rq
->rt_time
> rt_rq
->rt_runtime
) {
734 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
735 more
= do_balance_runtime(rt_rq
);
736 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
741 #else /* !CONFIG_SMP */
742 static inline int balance_runtime(struct rt_rq
*rt_rq
)
746 #endif /* CONFIG_SMP */
748 static int do_sched_rt_period_timer(struct rt_bandwidth
*rt_b
, int overrun
)
750 int i
, idle
= 1, throttled
= 0;
751 const struct cpumask
*span
;
753 span
= sched_rt_period_mask();
754 #ifdef CONFIG_RT_GROUP_SCHED
756 * FIXME: isolated CPUs should really leave the root task group,
757 * whether they are isolcpus or were isolated via cpusets, lest
758 * the timer run on a CPU which does not service all runqueues,
759 * potentially leaving other CPUs indefinitely throttled. If
760 * isolation is really required, the user will turn the throttle
761 * off to kill the perturbations it causes anyway. Meanwhile,
762 * this maintains functionality for boot and/or troubleshooting.
764 if (rt_b
== &root_task_group
.rt_bandwidth
)
765 span
= cpu_online_mask
;
767 for_each_cpu(i
, span
) {
769 struct rt_rq
*rt_rq
= sched_rt_period_rt_rq(rt_b
, i
);
770 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
772 raw_spin_lock(&rq
->lock
);
773 if (rt_rq
->rt_time
) {
776 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
777 if (rt_rq
->rt_throttled
)
778 balance_runtime(rt_rq
);
779 runtime
= rt_rq
->rt_runtime
;
780 rt_rq
->rt_time
-= min(rt_rq
->rt_time
, overrun
*runtime
);
781 if (rt_rq
->rt_throttled
&& rt_rq
->rt_time
< runtime
) {
782 rt_rq
->rt_throttled
= 0;
786 * Force a clock update if the CPU was idle,
787 * lest wakeup -> unthrottle time accumulate.
789 if (rt_rq
->rt_nr_running
&& rq
->curr
== rq
->idle
)
790 rq
->skip_clock_update
= -1;
792 if (rt_rq
->rt_time
|| rt_rq
->rt_nr_running
)
794 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
795 } else if (rt_rq
->rt_nr_running
) {
797 if (!rt_rq_throttled(rt_rq
))
800 if (rt_rq
->rt_throttled
)
804 sched_rt_rq_enqueue(rt_rq
);
805 raw_spin_unlock(&rq
->lock
);
808 if (!throttled
&& (!rt_bandwidth_enabled() || rt_b
->rt_runtime
== RUNTIME_INF
))
814 static inline int rt_se_prio(struct sched_rt_entity
*rt_se
)
816 #ifdef CONFIG_RT_GROUP_SCHED
817 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
820 return rt_rq
->highest_prio
.curr
;
823 return rt_task_of(rt_se
)->prio
;
826 static int sched_rt_runtime_exceeded(struct rt_rq
*rt_rq
)
828 u64 runtime
= sched_rt_runtime(rt_rq
);
830 if (rt_rq
->rt_throttled
)
831 return rt_rq_throttled(rt_rq
);
833 if (runtime
>= sched_rt_period(rt_rq
))
836 balance_runtime(rt_rq
);
837 runtime
= sched_rt_runtime(rt_rq
);
838 if (runtime
== RUNTIME_INF
)
841 if (rt_rq
->rt_time
> runtime
) {
842 struct rt_bandwidth
*rt_b
= sched_rt_bandwidth(rt_rq
);
845 * Don't actually throttle groups that have no runtime assigned
846 * but accrue some time due to boosting.
848 if (likely(rt_b
->rt_runtime
)) {
849 static bool once
= false;
851 rt_rq
->rt_throttled
= 1;
855 printk_deferred("sched: RT throttling activated on %s(%d)\n",
856 current
->comm
, current
->pid
);
860 * In case we did anyway, make it go away,
861 * replenishment is a joke, since it will replenish us
867 if (rt_rq_throttled(rt_rq
)) {
868 sched_rt_rq_dequeue(rt_rq
);
877 * Update the current task's runtime statistics. Skip current tasks that
878 * are not in our scheduling class.
880 static void update_curr_rt(struct rq
*rq
)
882 struct task_struct
*curr
= rq
->curr
;
883 struct sched_rt_entity
*rt_se
= &curr
->rt
;
884 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
887 if (curr
->sched_class
!= &rt_sched_class
)
890 delta_exec
= rq
->clock_task
- curr
->se
.exec_start
;
891 if (unlikely((s64
)delta_exec
<= 0))
894 schedstat_set(curr
->se
.statistics
.exec_max
,
895 max(curr
->se
.statistics
.exec_max
, delta_exec
));
897 curr
->se
.sum_exec_runtime
+= delta_exec
;
898 account_group_exec_runtime(curr
, delta_exec
);
900 curr
->se
.exec_start
= rq
->clock_task
;
901 cpuacct_charge(curr
, delta_exec
);
903 sched_rt_avg_update(rq
, delta_exec
);
905 if (!rt_bandwidth_enabled())
908 for_each_sched_rt_entity(rt_se
) {
909 rt_rq
= rt_rq_of_se(rt_se
);
911 if (sched_rt_runtime(rt_rq
) != RUNTIME_INF
) {
912 raw_spin_lock(&rt_rq
->rt_runtime_lock
);
913 rt_rq
->rt_time
+= delta_exec
;
914 if (sched_rt_runtime_exceeded(rt_rq
))
916 raw_spin_unlock(&rt_rq
->rt_runtime_lock
);
921 #if defined CONFIG_SMP
924 inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
926 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
928 if (rq
->online
&& prio
< prev_prio
)
929 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, prio
);
933 dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
)
935 struct rq
*rq
= rq_of_rt_rq(rt_rq
);
937 if (rq
->online
&& rt_rq
->highest_prio
.curr
!= prev_prio
)
938 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rt_rq
->highest_prio
.curr
);
941 #else /* CONFIG_SMP */
944 void inc_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
946 void dec_rt_prio_smp(struct rt_rq
*rt_rq
, int prio
, int prev_prio
) {}
948 #endif /* CONFIG_SMP */
950 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
952 inc_rt_prio(struct rt_rq
*rt_rq
, int prio
)
954 int prev_prio
= rt_rq
->highest_prio
.curr
;
956 if (prio
< prev_prio
)
957 rt_rq
->highest_prio
.curr
= prio
;
959 inc_rt_prio_smp(rt_rq
, prio
, prev_prio
);
963 dec_rt_prio(struct rt_rq
*rt_rq
, int prio
)
965 int prev_prio
= rt_rq
->highest_prio
.curr
;
967 if (rt_rq
->rt_nr_running
) {
969 WARN_ON(prio
< prev_prio
);
972 * This may have been our highest task, and therefore
973 * we may have some recomputation to do
975 if (prio
== prev_prio
) {
976 struct rt_prio_array
*array
= &rt_rq
->active
;
978 rt_rq
->highest_prio
.curr
=
979 sched_find_first_bit(array
->bitmap
);
983 rt_rq
->highest_prio
.curr
= MAX_RT_PRIO
;
985 dec_rt_prio_smp(rt_rq
, prio
, prev_prio
);
990 static inline void inc_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
991 static inline void dec_rt_prio(struct rt_rq
*rt_rq
, int prio
) {}
993 #endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
995 #ifdef CONFIG_RT_GROUP_SCHED
998 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1000 if (rt_se_boosted(rt_se
))
1001 rt_rq
->rt_nr_boosted
++;
1004 start_rt_bandwidth(&rt_rq
->tg
->rt_bandwidth
);
1008 dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1010 if (rt_se_boosted(rt_se
))
1011 rt_rq
->rt_nr_boosted
--;
1013 WARN_ON(!rt_rq
->rt_nr_running
&& rt_rq
->rt_nr_boosted
);
1016 #else /* CONFIG_RT_GROUP_SCHED */
1019 inc_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1021 start_rt_bandwidth(&def_rt_bandwidth
);
1025 void dec_rt_group(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
) {}
1027 #endif /* CONFIG_RT_GROUP_SCHED */
1030 void inc_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1032 int prio
= rt_se_prio(rt_se
);
1034 WARN_ON(!rt_prio(prio
));
1035 rt_rq
->rt_nr_running
++;
1037 inc_rt_prio(rt_rq
, prio
);
1038 inc_rt_migration(rt_se
, rt_rq
);
1039 inc_rt_group(rt_se
, rt_rq
);
1043 void dec_rt_tasks(struct sched_rt_entity
*rt_se
, struct rt_rq
*rt_rq
)
1045 WARN_ON(!rt_prio(rt_se_prio(rt_se
)));
1046 WARN_ON(!rt_rq
->rt_nr_running
);
1047 rt_rq
->rt_nr_running
--;
1049 dec_rt_prio(rt_rq
, rt_se_prio(rt_se
));
1050 dec_rt_migration(rt_se
, rt_rq
);
1051 dec_rt_group(rt_se
, rt_rq
);
1054 static void __enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
1056 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1057 struct rt_prio_array
*array
= &rt_rq
->active
;
1058 struct rt_rq
*group_rq
= group_rt_rq(rt_se
);
1059 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1062 * Don't enqueue the group if its throttled, or when empty.
1063 * The latter is a consequence of the former when a child group
1064 * get throttled and the current group doesn't have any other
1067 if (group_rq
&& (rt_rq_throttled(group_rq
) || !group_rq
->rt_nr_running
))
1070 if (!rt_rq
->rt_nr_running
)
1071 list_add_leaf_rt_rq(rt_rq
);
1074 list_add(&rt_se
->run_list
, queue
);
1076 list_add_tail(&rt_se
->run_list
, queue
);
1077 __set_bit(rt_se_prio(rt_se
), array
->bitmap
);
1079 inc_rt_tasks(rt_se
, rt_rq
);
1082 static void __dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
1084 struct rt_rq
*rt_rq
= rt_rq_of_se(rt_se
);
1085 struct rt_prio_array
*array
= &rt_rq
->active
;
1087 list_del_init(&rt_se
->run_list
);
1088 if (list_empty(array
->queue
+ rt_se_prio(rt_se
)))
1089 __clear_bit(rt_se_prio(rt_se
), array
->bitmap
);
1091 dec_rt_tasks(rt_se
, rt_rq
);
1092 if (!rt_rq
->rt_nr_running
)
1093 list_del_leaf_rt_rq(rt_rq
);
1097 * Because the prio of an upper entry depends on the lower
1098 * entries, we must remove entries top - down.
1100 static void dequeue_rt_stack(struct sched_rt_entity
*rt_se
)
1102 struct sched_rt_entity
*back
= NULL
;
1104 for_each_sched_rt_entity(rt_se
) {
1109 for (rt_se
= back
; rt_se
; rt_se
= rt_se
->back
) {
1110 if (on_rt_rq(rt_se
))
1111 __dequeue_rt_entity(rt_se
);
1115 static void enqueue_rt_entity(struct sched_rt_entity
*rt_se
, bool head
)
1117 dequeue_rt_stack(rt_se
);
1118 for_each_sched_rt_entity(rt_se
)
1119 __enqueue_rt_entity(rt_se
, head
);
1122 static void dequeue_rt_entity(struct sched_rt_entity
*rt_se
)
1124 dequeue_rt_stack(rt_se
);
1126 for_each_sched_rt_entity(rt_se
) {
1127 struct rt_rq
*rt_rq
= group_rt_rq(rt_se
);
1129 if (rt_rq
&& rt_rq
->rt_nr_running
)
1130 __enqueue_rt_entity(rt_se
, false);
1135 * Adding/removing a task to/from a priority array:
1138 enqueue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1140 struct sched_rt_entity
*rt_se
= &p
->rt
;
1142 if (flags
& ENQUEUE_WAKEUP
)
1145 enqueue_rt_entity(rt_se
, flags
& ENQUEUE_HEAD
);
1147 if (!task_current(rq
, p
) && p
->nr_cpus_allowed
> 1)
1148 enqueue_pushable_task(rq
, p
);
1153 static void dequeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1155 struct sched_rt_entity
*rt_se
= &p
->rt
;
1158 dequeue_rt_entity(rt_se
);
1160 dequeue_pushable_task(rq
, p
);
1166 * Put task to the head or the end of the run list without the overhead of
1167 * dequeue followed by enqueue.
1170 requeue_rt_entity(struct rt_rq
*rt_rq
, struct sched_rt_entity
*rt_se
, int head
)
1172 if (on_rt_rq(rt_se
)) {
1173 struct rt_prio_array
*array
= &rt_rq
->active
;
1174 struct list_head
*queue
= array
->queue
+ rt_se_prio(rt_se
);
1177 list_move(&rt_se
->run_list
, queue
);
1179 list_move_tail(&rt_se
->run_list
, queue
);
1183 static void requeue_task_rt(struct rq
*rq
, struct task_struct
*p
, int head
)
1185 struct sched_rt_entity
*rt_se
= &p
->rt
;
1186 struct rt_rq
*rt_rq
;
1188 for_each_sched_rt_entity(rt_se
) {
1189 rt_rq
= rt_rq_of_se(rt_se
);
1190 requeue_rt_entity(rt_rq
, rt_se
, head
);
1194 static void yield_task_rt(struct rq
*rq
)
1196 requeue_task_rt(rq
, rq
->curr
, 0);
1200 static int find_lowest_rq(struct task_struct
*task
);
1203 select_task_rq_rt(struct task_struct
*p
, int sd_flag
, int flags
)
1205 struct task_struct
*curr
;
1211 if (p
->nr_cpus_allowed
== 1)
1214 /* For anything but wake ups, just return the task_cpu */
1215 if (sd_flag
!= SD_BALANCE_WAKE
&& sd_flag
!= SD_BALANCE_FORK
)
1221 curr
= ACCESS_ONCE(rq
->curr
); /* unlocked access */
1224 * If the current task on @p's runqueue is an RT task, then
1225 * try to see if we can wake this RT task up on another
1226 * runqueue. Otherwise simply start this RT task
1227 * on its current runqueue.
1229 * We want to avoid overloading runqueues. If the woken
1230 * task is a higher priority, then it will stay on this CPU
1231 * and the lower prio task should be moved to another CPU.
1232 * Even though this will probably make the lower prio task
1233 * lose its cache, we do not want to bounce a higher task
1234 * around just because it gave up its CPU, perhaps for a
1237 * For equal prio tasks, we just let the scheduler sort it out.
1239 * Otherwise, just let it ride on the affined RQ and the
1240 * post-schedule router will push the preempted task away
1242 * This test is optimistic, if we get it wrong the load-balancer
1243 * will have to sort it out.
1245 if (curr
&& unlikely(rt_task(curr
)) &&
1246 (curr
->nr_cpus_allowed
< 2 ||
1247 curr
->prio
<= p
->prio
) &&
1248 (p
->nr_cpus_allowed
> 1)) {
1249 int target
= find_lowest_rq(p
);
1260 static void check_preempt_equal_prio(struct rq
*rq
, struct task_struct
*p
)
1262 if (rq
->curr
->nr_cpus_allowed
== 1)
1265 if (p
->nr_cpus_allowed
!= 1
1266 && cpupri_find(&rq
->rd
->cpupri
, p
, NULL
))
1269 if (!cpupri_find(&rq
->rd
->cpupri
, rq
->curr
, NULL
))
1273 * There appears to be other cpus that can accept
1274 * current and none to run 'p', so lets reschedule
1275 * to try and push current away:
1277 requeue_task_rt(rq
, p
, 1);
1278 resched_task(rq
->curr
);
1281 #endif /* CONFIG_SMP */
1284 * Preempt the current task with a newly woken task if needed:
1286 static void check_preempt_curr_rt(struct rq
*rq
, struct task_struct
*p
, int flags
)
1288 if (p
->prio
< rq
->curr
->prio
) {
1289 resched_task(rq
->curr
);
1297 * - the newly woken task is of equal priority to the current task
1298 * - the newly woken task is non-migratable while current is migratable
1299 * - current will be preempted on the next reschedule
1301 * we should check to see if current can readily move to a different
1302 * cpu. If so, we will reschedule to allow the push logic to try
1303 * to move current somewhere else, making room for our non-migratable
1306 if (p
->prio
== rq
->curr
->prio
&& !test_tsk_need_resched(rq
->curr
))
1307 check_preempt_equal_prio(rq
, p
);
1311 static struct sched_rt_entity
*pick_next_rt_entity(struct rq
*rq
,
1312 struct rt_rq
*rt_rq
)
1314 struct rt_prio_array
*array
= &rt_rq
->active
;
1315 struct sched_rt_entity
*next
= NULL
;
1316 struct list_head
*queue
;
1319 idx
= sched_find_first_bit(array
->bitmap
);
1320 BUG_ON(idx
>= MAX_RT_PRIO
);
1322 queue
= array
->queue
+ idx
;
1323 next
= list_entry(queue
->next
, struct sched_rt_entity
, run_list
);
1328 static struct task_struct
*_pick_next_task_rt(struct rq
*rq
)
1330 struct sched_rt_entity
*rt_se
;
1331 struct task_struct
*p
;
1332 struct rt_rq
*rt_rq
;
1336 if (!rt_rq
->rt_nr_running
)
1339 if (rt_rq_throttled(rt_rq
))
1343 rt_se
= pick_next_rt_entity(rq
, rt_rq
);
1345 rt_rq
= group_rt_rq(rt_se
);
1348 p
= rt_task_of(rt_se
);
1349 p
->se
.exec_start
= rq
->clock_task
;
1354 static struct task_struct
*pick_next_task_rt(struct rq
*rq
)
1356 struct task_struct
*p
= _pick_next_task_rt(rq
);
1358 /* The running task is never eligible for pushing */
1360 dequeue_pushable_task(rq
, p
);
1364 * We detect this state here so that we can avoid taking the RQ
1365 * lock again later if there is no need to push
1367 rq
->post_schedule
= has_pushable_tasks(rq
);
1373 static void put_prev_task_rt(struct rq
*rq
, struct task_struct
*p
)
1378 * The previous task needs to be made eligible for pushing
1379 * if it is still active
1381 if (on_rt_rq(&p
->rt
) && p
->nr_cpus_allowed
> 1)
1382 enqueue_pushable_task(rq
, p
);
1387 /* Only try algorithms three times */
1388 #define RT_MAX_TRIES 3
1390 static int pick_rt_task(struct rq
*rq
, struct task_struct
*p
, int cpu
)
1392 if (!task_running(rq
, p
) &&
1393 cpumask_test_cpu(cpu
, tsk_cpus_allowed(p
)))
1398 /* Return the second highest RT task, NULL otherwise */
1399 static struct task_struct
*pick_next_highest_task_rt(struct rq
*rq
, int cpu
)
1401 struct task_struct
*next
= NULL
;
1402 struct sched_rt_entity
*rt_se
;
1403 struct rt_prio_array
*array
;
1404 struct rt_rq
*rt_rq
;
1407 for_each_leaf_rt_rq(rt_rq
, rq
) {
1408 array
= &rt_rq
->active
;
1409 idx
= sched_find_first_bit(array
->bitmap
);
1411 if (idx
>= MAX_RT_PRIO
)
1413 if (next
&& next
->prio
<= idx
)
1415 list_for_each_entry(rt_se
, array
->queue
+ idx
, run_list
) {
1416 struct task_struct
*p
;
1418 if (!rt_entity_is_task(rt_se
))
1421 p
= rt_task_of(rt_se
);
1422 if (pick_rt_task(rq
, p
, cpu
)) {
1428 idx
= find_next_bit(array
->bitmap
, MAX_RT_PRIO
, idx
+1);
1436 static DEFINE_PER_CPU(cpumask_var_t
, local_cpu_mask
);
1438 static int find_lowest_rq(struct task_struct
*task
)
1440 struct sched_domain
*sd
;
1441 struct cpumask
*lowest_mask
= __get_cpu_var(local_cpu_mask
);
1442 int this_cpu
= smp_processor_id();
1443 int cpu
= task_cpu(task
);
1445 /* Make sure the mask is initialized first */
1446 if (unlikely(!lowest_mask
))
1449 if (task
->nr_cpus_allowed
== 1)
1450 return -1; /* No other targets possible */
1452 if (!cpupri_find(&task_rq(task
)->rd
->cpupri
, task
, lowest_mask
))
1453 return -1; /* No targets found */
1456 * At this point we have built a mask of cpus representing the
1457 * lowest priority tasks in the system. Now we want to elect
1458 * the best one based on our affinity and topology.
1460 * We prioritize the last cpu that the task executed on since
1461 * it is most likely cache-hot in that location.
1463 if (cpumask_test_cpu(cpu
, lowest_mask
))
1467 * Otherwise, we consult the sched_domains span maps to figure
1468 * out which cpu is logically closest to our hot cache data.
1470 if (!cpumask_test_cpu(this_cpu
, lowest_mask
))
1471 this_cpu
= -1; /* Skip this_cpu opt if not among lowest */
1474 for_each_domain(cpu
, sd
) {
1475 if (sd
->flags
& SD_WAKE_AFFINE
) {
1479 * "this_cpu" is cheaper to preempt than a
1482 if (this_cpu
!= -1 &&
1483 cpumask_test_cpu(this_cpu
, sched_domain_span(sd
))) {
1488 best_cpu
= cpumask_first_and(lowest_mask
,
1489 sched_domain_span(sd
));
1490 if (best_cpu
< nr_cpu_ids
) {
1499 * And finally, if there were no matches within the domains
1500 * just give the caller *something* to work with from the compatible
1506 cpu
= cpumask_any(lowest_mask
);
1507 if (cpu
< nr_cpu_ids
)
1512 /* Will lock the rq it finds */
1513 static struct rq
*find_lock_lowest_rq(struct task_struct
*task
, struct rq
*rq
)
1515 struct rq
*lowest_rq
= NULL
;
1519 for (tries
= 0; tries
< RT_MAX_TRIES
; tries
++) {
1520 cpu
= find_lowest_rq(task
);
1522 if ((cpu
== -1) || (cpu
== rq
->cpu
))
1525 lowest_rq
= cpu_rq(cpu
);
1527 /* if the prio of this runqueue changed, try again */
1528 if (double_lock_balance(rq
, lowest_rq
)) {
1530 * We had to unlock the run queue. In
1531 * the mean time, task could have
1532 * migrated already or had its affinity changed.
1533 * Also make sure that it wasn't scheduled on its rq.
1535 if (unlikely(task_rq(task
) != rq
||
1536 !cpumask_test_cpu(lowest_rq
->cpu
,
1537 tsk_cpus_allowed(task
)) ||
1538 task_running(rq
, task
) ||
1541 double_unlock_balance(rq
, lowest_rq
);
1547 /* If this rq is still suitable use it. */
1548 if (lowest_rq
->rt
.highest_prio
.curr
> task
->prio
)
1552 double_unlock_balance(rq
, lowest_rq
);
1559 static struct task_struct
*pick_next_pushable_task(struct rq
*rq
)
1561 struct task_struct
*p
;
1563 if (!has_pushable_tasks(rq
))
1566 p
= plist_first_entry(&rq
->rt
.pushable_tasks
,
1567 struct task_struct
, pushable_tasks
);
1569 BUG_ON(rq
->cpu
!= task_cpu(p
));
1570 BUG_ON(task_current(rq
, p
));
1571 BUG_ON(p
->nr_cpus_allowed
<= 1);
1574 BUG_ON(!rt_task(p
));
1580 * If the current CPU has more than one RT task, see if the non
1581 * running task can migrate over to a CPU that is running a task
1582 * of lesser priority.
1584 static int push_rt_task(struct rq
*rq
)
1586 struct task_struct
*next_task
;
1587 struct rq
*lowest_rq
;
1590 if (!rq
->rt
.overloaded
)
1593 next_task
= pick_next_pushable_task(rq
);
1598 if (unlikely(next_task
== rq
->curr
)) {
1604 * It's possible that the next_task slipped in of
1605 * higher priority than current. If that's the case
1606 * just reschedule current.
1608 if (unlikely(next_task
->prio
< rq
->curr
->prio
)) {
1609 resched_task(rq
->curr
);
1613 /* We might release rq lock */
1614 get_task_struct(next_task
);
1616 /* find_lock_lowest_rq locks the rq if found */
1617 lowest_rq
= find_lock_lowest_rq(next_task
, rq
);
1619 struct task_struct
*task
;
1621 * find_lock_lowest_rq releases rq->lock
1622 * so it is possible that next_task has migrated.
1624 * We need to make sure that the task is still on the same
1625 * run-queue and is also still the next task eligible for
1628 task
= pick_next_pushable_task(rq
);
1629 if (task_cpu(next_task
) == rq
->cpu
&& task
== next_task
) {
1631 * The task hasn't migrated, and is still the next
1632 * eligible task, but we failed to find a run-queue
1633 * to push it to. Do not retry in this case, since
1634 * other cpus will pull from us when ready.
1640 /* No more tasks, just exit */
1644 * Something has shifted, try again.
1646 put_task_struct(next_task
);
1651 deactivate_task(rq
, next_task
, 0);
1652 set_task_cpu(next_task
, lowest_rq
->cpu
);
1653 activate_task(lowest_rq
, next_task
, 0);
1656 resched_task(lowest_rq
->curr
);
1658 double_unlock_balance(rq
, lowest_rq
);
1661 put_task_struct(next_task
);
1666 static void push_rt_tasks(struct rq
*rq
)
1668 /* push_rt_task will return true if it moved an RT */
1669 while (push_rt_task(rq
))
1673 static int pull_rt_task(struct rq
*this_rq
)
1675 int this_cpu
= this_rq
->cpu
, ret
= 0, cpu
;
1676 struct task_struct
*p
;
1679 if (likely(!rt_overloaded(this_rq
)))
1682 for_each_cpu(cpu
, this_rq
->rd
->rto_mask
) {
1683 if (this_cpu
== cpu
)
1686 src_rq
= cpu_rq(cpu
);
1689 * Don't bother taking the src_rq->lock if the next highest
1690 * task is known to be lower-priority than our current task.
1691 * This may look racy, but if this value is about to go
1692 * logically higher, the src_rq will push this task away.
1693 * And if its going logically lower, we do not care
1695 if (src_rq
->rt
.highest_prio
.next
>=
1696 this_rq
->rt
.highest_prio
.curr
)
1700 * We can potentially drop this_rq's lock in
1701 * double_lock_balance, and another CPU could
1704 double_lock_balance(this_rq
, src_rq
);
1707 * Are there still pullable RT tasks?
1709 if (src_rq
->rt
.rt_nr_running
<= 1)
1712 p
= pick_next_highest_task_rt(src_rq
, this_cpu
);
1715 * Do we have an RT task that preempts
1716 * the to-be-scheduled task?
1718 if (p
&& (p
->prio
< this_rq
->rt
.highest_prio
.curr
)) {
1719 WARN_ON(p
== src_rq
->curr
);
1723 * There's a chance that p is higher in priority
1724 * than what's currently running on its cpu.
1725 * This is just that p is wakeing up and hasn't
1726 * had a chance to schedule. We only pull
1727 * p if it is lower in priority than the
1728 * current task on the run queue
1730 if (p
->prio
< src_rq
->curr
->prio
)
1735 deactivate_task(src_rq
, p
, 0);
1736 set_task_cpu(p
, this_cpu
);
1737 activate_task(this_rq
, p
, 0);
1739 * We continue with the search, just in
1740 * case there's an even higher prio task
1741 * in another runqueue. (low likelihood
1746 double_unlock_balance(this_rq
, src_rq
);
1752 static void pre_schedule_rt(struct rq
*rq
, struct task_struct
*prev
)
1754 /* Try to pull RT tasks here if we lower this rq's prio */
1755 if (rq
->rt
.highest_prio
.curr
> prev
->prio
)
1759 static void post_schedule_rt(struct rq
*rq
)
1765 * If we are not running and we are not going to reschedule soon, we should
1766 * try to push tasks away now
1768 static void task_woken_rt(struct rq
*rq
, struct task_struct
*p
)
1770 if (!task_running(rq
, p
) &&
1771 !test_tsk_need_resched(rq
->curr
) &&
1772 has_pushable_tasks(rq
) &&
1773 p
->nr_cpus_allowed
> 1 &&
1774 rt_task(rq
->curr
) &&
1775 (rq
->curr
->nr_cpus_allowed
< 2 ||
1776 rq
->curr
->prio
<= p
->prio
))
1780 static void set_cpus_allowed_rt(struct task_struct
*p
,
1781 const struct cpumask
*new_mask
)
1786 BUG_ON(!rt_task(p
));
1791 weight
= cpumask_weight(new_mask
);
1794 * Only update if the process changes its state from whether it
1795 * can migrate or not.
1797 if ((p
->nr_cpus_allowed
> 1) == (weight
> 1))
1803 * The process used to be able to migrate OR it can now migrate
1806 if (!task_current(rq
, p
))
1807 dequeue_pushable_task(rq
, p
);
1808 BUG_ON(!rq
->rt
.rt_nr_migratory
);
1809 rq
->rt
.rt_nr_migratory
--;
1811 if (!task_current(rq
, p
))
1812 enqueue_pushable_task(rq
, p
);
1813 rq
->rt
.rt_nr_migratory
++;
1816 update_rt_migration(&rq
->rt
);
1819 /* Assumes rq->lock is held */
1820 static void rq_online_rt(struct rq
*rq
)
1822 if (rq
->rt
.overloaded
)
1823 rt_set_overload(rq
);
1825 __enable_runtime(rq
);
1827 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, rq
->rt
.highest_prio
.curr
);
1830 /* Assumes rq->lock is held */
1831 static void rq_offline_rt(struct rq
*rq
)
1833 if (rq
->rt
.overloaded
)
1834 rt_clear_overload(rq
);
1836 __disable_runtime(rq
);
1838 cpupri_set(&rq
->rd
->cpupri
, rq
->cpu
, CPUPRI_INVALID
);
1842 * When switch from the rt queue, we bring ourselves to a position
1843 * that we might want to pull RT tasks from other runqueues.
1845 static void switched_from_rt(struct rq
*rq
, struct task_struct
*p
)
1848 * If there are other RT tasks then we will reschedule
1849 * and the scheduling of the other RT tasks will handle
1850 * the balancing. But if we are the last RT task
1851 * we may need to handle the pulling of RT tasks
1854 if (!p
->on_rq
|| rq
->rt
.rt_nr_running
)
1857 if (pull_rt_task(rq
))
1858 resched_task(rq
->curr
);
1861 void init_sched_rt_class(void)
1865 for_each_possible_cpu(i
) {
1866 zalloc_cpumask_var_node(&per_cpu(local_cpu_mask
, i
),
1867 GFP_KERNEL
, cpu_to_node(i
));
1870 #endif /* CONFIG_SMP */
1873 * When switching a task to RT, we may overload the runqueue
1874 * with RT tasks. In this case we try to push them off to
1877 static void switched_to_rt(struct rq
*rq
, struct task_struct
*p
)
1879 int check_resched
= 1;
1882 * If we are already running, then there's nothing
1883 * that needs to be done. But if we are not running
1884 * we may need to preempt the current running task.
1885 * If that current running task is also an RT task
1886 * then see if we can move to another run queue.
1888 if (p
->on_rq
&& rq
->curr
!= p
) {
1890 if (rq
->rt
.overloaded
&& push_rt_task(rq
) &&
1891 /* Don't resched if we changed runqueues */
1894 #endif /* CONFIG_SMP */
1895 if (check_resched
&& p
->prio
< rq
->curr
->prio
)
1896 resched_task(rq
->curr
);
1901 * Priority of the task has changed. This may cause
1902 * us to initiate a push or pull.
1905 prio_changed_rt(struct rq
*rq
, struct task_struct
*p
, int oldprio
)
1910 if (rq
->curr
== p
) {
1913 * If our priority decreases while running, we
1914 * may need to pull tasks to this runqueue.
1916 if (oldprio
< p
->prio
)
1919 * If there's a higher priority task waiting to run
1920 * then reschedule. Note, the above pull_rt_task
1921 * can release the rq lock and p could migrate.
1922 * Only reschedule if p is still on the same runqueue.
1924 if (p
->prio
> rq
->rt
.highest_prio
.curr
&& rq
->curr
== p
)
1927 /* For UP simply resched on drop of prio */
1928 if (oldprio
< p
->prio
)
1930 #endif /* CONFIG_SMP */
1933 * This task is not running, but if it is
1934 * greater than the current running task
1937 if (p
->prio
< rq
->curr
->prio
)
1938 resched_task(rq
->curr
);
1942 static void watchdog(struct rq
*rq
, struct task_struct
*p
)
1944 unsigned long soft
, hard
;
1946 /* max may change after cur was read, this will be fixed next tick */
1947 soft
= task_rlimit(p
, RLIMIT_RTTIME
);
1948 hard
= task_rlimit_max(p
, RLIMIT_RTTIME
);
1950 if (soft
!= RLIM_INFINITY
) {
1953 if (p
->rt
.watchdog_stamp
!= jiffies
) {
1955 p
->rt
.watchdog_stamp
= jiffies
;
1958 next
= DIV_ROUND_UP(min(soft
, hard
), USEC_PER_SEC
/HZ
);
1959 if (p
->rt
.timeout
> next
)
1960 p
->cputime_expires
.sched_exp
= p
->se
.sum_exec_runtime
;
1964 static void task_tick_rt(struct rq
*rq
, struct task_struct
*p
, int queued
)
1966 struct sched_rt_entity
*rt_se
= &p
->rt
;
1973 * RR tasks need a special form of timeslice management.
1974 * FIFO tasks have no timeslices.
1976 if (p
->policy
!= SCHED_RR
)
1979 if (--p
->rt
.time_slice
)
1982 p
->rt
.time_slice
= sched_rr_timeslice
;
1985 * Requeue to the end of queue if we (and all of our ancestors) are the
1986 * only element on the queue
1988 for_each_sched_rt_entity(rt_se
) {
1989 if (rt_se
->run_list
.prev
!= rt_se
->run_list
.next
) {
1990 requeue_task_rt(rq
, p
, 0);
1991 set_tsk_need_resched(p
);
1997 static void set_curr_task_rt(struct rq
*rq
)
1999 struct task_struct
*p
= rq
->curr
;
2001 p
->se
.exec_start
= rq
->clock_task
;
2003 /* The running task is never eligible for pushing */
2004 dequeue_pushable_task(rq
, p
);
2007 static unsigned int get_rr_interval_rt(struct rq
*rq
, struct task_struct
*task
)
2010 * Time slice is 0 for SCHED_FIFO tasks
2012 if (task
->policy
== SCHED_RR
)
2013 return sched_rr_timeslice
;
2018 const struct sched_class rt_sched_class
= {
2019 .next
= &fair_sched_class
,
2020 .enqueue_task
= enqueue_task_rt
,
2021 .dequeue_task
= dequeue_task_rt
,
2022 .yield_task
= yield_task_rt
,
2024 .check_preempt_curr
= check_preempt_curr_rt
,
2026 .pick_next_task
= pick_next_task_rt
,
2027 .put_prev_task
= put_prev_task_rt
,
2030 .select_task_rq
= select_task_rq_rt
,
2032 .set_cpus_allowed
= set_cpus_allowed_rt
,
2033 .rq_online
= rq_online_rt
,
2034 .rq_offline
= rq_offline_rt
,
2035 .pre_schedule
= pre_schedule_rt
,
2036 .post_schedule
= post_schedule_rt
,
2037 .task_woken
= task_woken_rt
,
2038 .switched_from
= switched_from_rt
,
2041 .set_curr_task
= set_curr_task_rt
,
2042 .task_tick
= task_tick_rt
,
2044 .get_rr_interval
= get_rr_interval_rt
,
2046 .prio_changed
= prio_changed_rt
,
2047 .switched_to
= switched_to_rt
,
2050 #ifdef CONFIG_SCHED_DEBUG
2051 extern void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
);
2053 void print_rt_stats(struct seq_file
*m
, int cpu
)
2056 struct rt_rq
*rt_rq
;
2059 for_each_rt_rq(rt_rq
, iter
, cpu_rq(cpu
))
2060 print_rt_rq(m
, cpu
, rt_rq
);
2063 #endif /* CONFIG_SCHED_DEBUG */