sched/rt: Code cleanup, remove a redundant function call
[linux-2.6-block.git] / kernel / sched / sched.h
CommitLineData
029632fb
PZ
1
2#include <linux/sched.h>
3#include <linux/mutex.h>
4#include <linux/spinlock.h>
5#include <linux/stop_machine.h>
6
391e43da 7#include "cpupri.h"
029632fb
PZ
8
9extern __read_mostly int scheduler_running;
10
11/*
12 * Convert user-nice values [ -20 ... 0 ... 19 ]
13 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
14 * and back.
15 */
16#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
17#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
18#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
19
20/*
21 * 'User priority' is the nice value converted to something we
22 * can work with better when scaling various scheduler parameters,
23 * it's a [ 0 ... 39 ] range.
24 */
25#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
26#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
27#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
28
29/*
30 * Helpers for converting nanosecond timing to jiffy resolution
31 */
32#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
33
34#define NICE_0_LOAD SCHED_LOAD_SCALE
35#define NICE_0_SHIFT SCHED_LOAD_SHIFT
36
37/*
38 * These are the 'tuning knobs' of the scheduler:
39 *
40 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
41 * Timeslices get refilled after they expire.
42 */
43#define DEF_TIMESLICE (100 * HZ / 1000)
44
45/*
46 * single value that denotes runtime == period, ie unlimited time.
47 */
48#define RUNTIME_INF ((u64)~0ULL)
49
50static inline int rt_policy(int policy)
51{
52 if (policy == SCHED_FIFO || policy == SCHED_RR)
53 return 1;
54 return 0;
55}
56
57static inline int task_has_rt_policy(struct task_struct *p)
58{
59 return rt_policy(p->policy);
60}
61
62/*
63 * This is the priority-queue data structure of the RT scheduling class:
64 */
65struct rt_prio_array {
66 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
67 struct list_head queue[MAX_RT_PRIO];
68};
69
70struct rt_bandwidth {
71 /* nests inside the rq lock: */
72 raw_spinlock_t rt_runtime_lock;
73 ktime_t rt_period;
74 u64 rt_runtime;
75 struct hrtimer rt_period_timer;
76};
77
78extern struct mutex sched_domains_mutex;
79
80#ifdef CONFIG_CGROUP_SCHED
81
82#include <linux/cgroup.h>
83
84struct cfs_rq;
85struct rt_rq;
86
87static LIST_HEAD(task_groups);
88
89struct cfs_bandwidth {
90#ifdef CONFIG_CFS_BANDWIDTH
91 raw_spinlock_t lock;
92 ktime_t period;
93 u64 quota, runtime;
94 s64 hierarchal_quota;
95 u64 runtime_expires;
96
97 int idle, timer_active;
98 struct hrtimer period_timer, slack_timer;
99 struct list_head throttled_cfs_rq;
100
101 /* statistics */
102 int nr_periods, nr_throttled;
103 u64 throttled_time;
104#endif
105};
106
107/* task group related information */
108struct task_group {
109 struct cgroup_subsys_state css;
110
111#ifdef CONFIG_FAIR_GROUP_SCHED
112 /* schedulable entities of this group on each cpu */
113 struct sched_entity **se;
114 /* runqueue "owned" by this group on each cpu */
115 struct cfs_rq **cfs_rq;
116 unsigned long shares;
117
118 atomic_t load_weight;
119#endif
120
121#ifdef CONFIG_RT_GROUP_SCHED
122 struct sched_rt_entity **rt_se;
123 struct rt_rq **rt_rq;
124
125 struct rt_bandwidth rt_bandwidth;
126#endif
127
128 struct rcu_head rcu;
129 struct list_head list;
130
131 struct task_group *parent;
132 struct list_head siblings;
133 struct list_head children;
134
135#ifdef CONFIG_SCHED_AUTOGROUP
136 struct autogroup *autogroup;
137#endif
138
139 struct cfs_bandwidth cfs_bandwidth;
140};
141
142#ifdef CONFIG_FAIR_GROUP_SCHED
143#define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
144
145/*
146 * A weight of 0 or 1 can cause arithmetics problems.
147 * A weight of a cfs_rq is the sum of weights of which entities
148 * are queued on this cfs_rq, so a weight of a entity should not be
149 * too large, so as the shares value of a task group.
150 * (The default weight is 1024 - so there's no practical
151 * limitation from this.)
152 */
153#define MIN_SHARES (1UL << 1)
154#define MAX_SHARES (1UL << 18)
155#endif
156
157/* Default task group.
158 * Every task in system belong to this group at bootup.
159 */
160extern struct task_group root_task_group;
161
162typedef int (*tg_visitor)(struct task_group *, void *);
163
164extern int walk_tg_tree_from(struct task_group *from,
165 tg_visitor down, tg_visitor up, void *data);
166
167/*
168 * Iterate the full tree, calling @down when first entering a node and @up when
169 * leaving it for the final time.
170 *
171 * Caller must hold rcu_lock or sufficient equivalent.
172 */
173static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
174{
175 return walk_tg_tree_from(&root_task_group, down, up, data);
176}
177
178extern int tg_nop(struct task_group *tg, void *data);
179
180extern void free_fair_sched_group(struct task_group *tg);
181extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
182extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
183extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
184 struct sched_entity *se, int cpu,
185 struct sched_entity *parent);
186extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
187extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
188
189extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
190extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
191extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
192
193extern void free_rt_sched_group(struct task_group *tg);
194extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
195extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
196 struct sched_rt_entity *rt_se, int cpu,
197 struct sched_rt_entity *parent);
198
199#else /* CONFIG_CGROUP_SCHED */
200
201struct cfs_bandwidth { };
202
203#endif /* CONFIG_CGROUP_SCHED */
204
205/* CFS-related fields in a runqueue */
206struct cfs_rq {
207 struct load_weight load;
208 unsigned long nr_running, h_nr_running;
209
210 u64 exec_clock;
211 u64 min_vruntime;
212#ifndef CONFIG_64BIT
213 u64 min_vruntime_copy;
214#endif
215
216 struct rb_root tasks_timeline;
217 struct rb_node *rb_leftmost;
218
219 struct list_head tasks;
220 struct list_head *balance_iterator;
221
222 /*
223 * 'curr' points to currently running entity on this cfs_rq.
224 * It is set to NULL otherwise (i.e when none are currently running).
225 */
226 struct sched_entity *curr, *next, *last, *skip;
227
228#ifdef CONFIG_SCHED_DEBUG
229 unsigned int nr_spread_over;
230#endif
231
232#ifdef CONFIG_FAIR_GROUP_SCHED
233 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
234
235 /*
236 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
237 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
238 * (like users, containers etc.)
239 *
240 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
241 * list is used during load balance.
242 */
243 int on_list;
244 struct list_head leaf_cfs_rq_list;
245 struct task_group *tg; /* group that "owns" this runqueue */
246
247#ifdef CONFIG_SMP
248 /*
249 * the part of load.weight contributed by tasks
250 */
251 unsigned long task_weight;
252
253 /*
254 * h_load = weight * f(tg)
255 *
256 * Where f(tg) is the recursive weight fraction assigned to
257 * this group.
258 */
259 unsigned long h_load;
260
261 /*
262 * Maintaining per-cpu shares distribution for group scheduling
263 *
264 * load_stamp is the last time we updated the load average
265 * load_last is the last time we updated the load average and saw load
266 * load_unacc_exec_time is currently unaccounted execution time
267 */
268 u64 load_avg;
269 u64 load_period;
270 u64 load_stamp, load_last, load_unacc_exec_time;
271
272 unsigned long load_contribution;
273#endif /* CONFIG_SMP */
274#ifdef CONFIG_CFS_BANDWIDTH
275 int runtime_enabled;
276 u64 runtime_expires;
277 s64 runtime_remaining;
278
279 u64 throttled_timestamp;
280 int throttled, throttle_count;
281 struct list_head throttled_list;
282#endif /* CONFIG_CFS_BANDWIDTH */
283#endif /* CONFIG_FAIR_GROUP_SCHED */
284};
285
286static inline int rt_bandwidth_enabled(void)
287{
288 return sysctl_sched_rt_runtime >= 0;
289}
290
291/* Real-Time classes' related field in a runqueue: */
292struct rt_rq {
293 struct rt_prio_array active;
294 unsigned long rt_nr_running;
295#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
296 struct {
297 int curr; /* highest queued rt task prio */
298#ifdef CONFIG_SMP
299 int next; /* next highest */
300#endif
301 } highest_prio;
302#endif
303#ifdef CONFIG_SMP
304 unsigned long rt_nr_migratory;
305 unsigned long rt_nr_total;
306 int overloaded;
307 struct plist_head pushable_tasks;
308#endif
309 int rt_throttled;
310 u64 rt_time;
311 u64 rt_runtime;
312 /* Nests inside the rq lock: */
313 raw_spinlock_t rt_runtime_lock;
314
315#ifdef CONFIG_RT_GROUP_SCHED
316 unsigned long rt_nr_boosted;
317
318 struct rq *rq;
319 struct list_head leaf_rt_rq_list;
320 struct task_group *tg;
321#endif
322};
323
324#ifdef CONFIG_SMP
325
326/*
327 * We add the notion of a root-domain which will be used to define per-domain
328 * variables. Each exclusive cpuset essentially defines an island domain by
329 * fully partitioning the member cpus from any other cpuset. Whenever a new
330 * exclusive cpuset is created, we also create and attach a new root-domain
331 * object.
332 *
333 */
334struct root_domain {
335 atomic_t refcount;
336 atomic_t rto_count;
337 struct rcu_head rcu;
338 cpumask_var_t span;
339 cpumask_var_t online;
340
341 /*
342 * The "RT overload" flag: it gets set if a CPU has more than
343 * one runnable RT task.
344 */
345 cpumask_var_t rto_mask;
346 struct cpupri cpupri;
347};
348
349extern struct root_domain def_root_domain;
350
351#endif /* CONFIG_SMP */
352
353/*
354 * This is the main, per-CPU runqueue data structure.
355 *
356 * Locking rule: those places that want to lock multiple runqueues
357 * (such as the load balancing or the thread migration code), lock
358 * acquire operations must be ordered by ascending &runqueue.
359 */
360struct rq {
361 /* runqueue lock: */
362 raw_spinlock_t lock;
363
364 /*
365 * nr_running and cpu_load should be in the same cacheline because
366 * remote CPUs use both these fields when doing load calculation.
367 */
368 unsigned long nr_running;
369 #define CPU_LOAD_IDX_MAX 5
370 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
371 unsigned long last_load_update_tick;
372#ifdef CONFIG_NO_HZ
373 u64 nohz_stamp;
374 unsigned char nohz_balance_kick;
375#endif
376 int skip_clock_update;
377
378 /* capture load from *all* tasks on this cpu: */
379 struct load_weight load;
380 unsigned long nr_load_updates;
381 u64 nr_switches;
382
383 struct cfs_rq cfs;
384 struct rt_rq rt;
385
386#ifdef CONFIG_FAIR_GROUP_SCHED
387 /* list of leaf cfs_rq on this cpu: */
388 struct list_head leaf_cfs_rq_list;
389#endif
390#ifdef CONFIG_RT_GROUP_SCHED
391 struct list_head leaf_rt_rq_list;
392#endif
393
394 /*
395 * This is part of a global counter where only the total sum
396 * over all CPUs matters. A task can increase this counter on
397 * one CPU and if it got migrated afterwards it may decrease
398 * it on another CPU. Always updated under the runqueue lock:
399 */
400 unsigned long nr_uninterruptible;
401
402 struct task_struct *curr, *idle, *stop;
403 unsigned long next_balance;
404 struct mm_struct *prev_mm;
405
406 u64 clock;
407 u64 clock_task;
408
409 atomic_t nr_iowait;
410
411#ifdef CONFIG_SMP
412 struct root_domain *rd;
413 struct sched_domain *sd;
414
415 unsigned long cpu_power;
416
417 unsigned char idle_balance;
418 /* For active balancing */
419 int post_schedule;
420 int active_balance;
421 int push_cpu;
422 struct cpu_stop_work active_balance_work;
423 /* cpu of this runqueue: */
424 int cpu;
425 int online;
426
427 u64 rt_avg;
428 u64 age_stamp;
429 u64 idle_stamp;
430 u64 avg_idle;
431#endif
432
433#ifdef CONFIG_IRQ_TIME_ACCOUNTING
434 u64 prev_irq_time;
435#endif
436#ifdef CONFIG_PARAVIRT
437 u64 prev_steal_time;
438#endif
439#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
440 u64 prev_steal_time_rq;
441#endif
442
443 /* calc_load related fields */
444 unsigned long calc_load_update;
445 long calc_load_active;
446
447#ifdef CONFIG_SCHED_HRTICK
448#ifdef CONFIG_SMP
449 int hrtick_csd_pending;
450 struct call_single_data hrtick_csd;
451#endif
452 struct hrtimer hrtick_timer;
453#endif
454
455#ifdef CONFIG_SCHEDSTATS
456 /* latency stats */
457 struct sched_info rq_sched_info;
458 unsigned long long rq_cpu_time;
459 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
460
461 /* sys_sched_yield() stats */
462 unsigned int yld_count;
463
464 /* schedule() stats */
465 unsigned int sched_switch;
466 unsigned int sched_count;
467 unsigned int sched_goidle;
468
469 /* try_to_wake_up() stats */
470 unsigned int ttwu_count;
471 unsigned int ttwu_local;
472#endif
473
474#ifdef CONFIG_SMP
475 struct llist_head wake_list;
476#endif
477};
478
479static inline int cpu_of(struct rq *rq)
480{
481#ifdef CONFIG_SMP
482 return rq->cpu;
483#else
484 return 0;
485#endif
486}
487
488DECLARE_PER_CPU(struct rq, runqueues);
489
490#define rcu_dereference_check_sched_domain(p) \
491 rcu_dereference_check((p), \
492 lockdep_is_held(&sched_domains_mutex))
493
494/*
495 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
496 * See detach_destroy_domains: synchronize_sched for details.
497 *
498 * The domain tree of any CPU may only be accessed from within
499 * preempt-disabled sections.
500 */
501#define for_each_domain(cpu, __sd) \
502 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
503
77e81365
SS
504#define for_each_lower_domain(sd) for (; sd; sd = sd->child)
505
029632fb
PZ
506#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
507#define this_rq() (&__get_cpu_var(runqueues))
508#define task_rq(p) cpu_rq(task_cpu(p))
509#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
510#define raw_rq() (&__raw_get_cpu_var(runqueues))
511
391e43da
PZ
512#include "stats.h"
513#include "auto_group.h"
029632fb
PZ
514
515#ifdef CONFIG_CGROUP_SCHED
516
517/*
518 * Return the group to which this tasks belongs.
519 *
520 * We use task_subsys_state_check() and extend the RCU verification with
521 * pi->lock and rq->lock because cpu_cgroup_attach() holds those locks for each
522 * task it moves into the cgroup. Therefore by holding either of those locks,
523 * we pin the task to the current cgroup.
524 */
525static inline struct task_group *task_group(struct task_struct *p)
526{
527 struct task_group *tg;
528 struct cgroup_subsys_state *css;
529
530 css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
531 lockdep_is_held(&p->pi_lock) ||
532 lockdep_is_held(&task_rq(p)->lock));
533 tg = container_of(css, struct task_group, css);
534
535 return autogroup_task_group(p, tg);
536}
537
538/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
539static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
540{
541#if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
542 struct task_group *tg = task_group(p);
543#endif
544
545#ifdef CONFIG_FAIR_GROUP_SCHED
546 p->se.cfs_rq = tg->cfs_rq[cpu];
547 p->se.parent = tg->se[cpu];
548#endif
549
550#ifdef CONFIG_RT_GROUP_SCHED
551 p->rt.rt_rq = tg->rt_rq[cpu];
552 p->rt.parent = tg->rt_se[cpu];
553#endif
554}
555
556#else /* CONFIG_CGROUP_SCHED */
557
558static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
559static inline struct task_group *task_group(struct task_struct *p)
560{
561 return NULL;
562}
563
564#endif /* CONFIG_CGROUP_SCHED */
565
566static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
567{
568 set_task_rq(p, cpu);
569#ifdef CONFIG_SMP
570 /*
571 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
572 * successfuly executed on another CPU. We must ensure that updates of
573 * per-task data have been completed by this moment.
574 */
575 smp_wmb();
576 task_thread_info(p)->cpu = cpu;
577#endif
578}
579
580/*
581 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
582 */
583#ifdef CONFIG_SCHED_DEBUG
584# define const_debug __read_mostly
585#else
586# define const_debug const
587#endif
588
589extern const_debug unsigned int sysctl_sched_features;
590
591#define SCHED_FEAT(name, enabled) \
592 __SCHED_FEAT_##name ,
593
594enum {
391e43da 595#include "features.h"
029632fb
PZ
596};
597
598#undef SCHED_FEAT
599
600#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
601
602static inline u64 global_rt_period(void)
603{
604 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
605}
606
607static inline u64 global_rt_runtime(void)
608{
609 if (sysctl_sched_rt_runtime < 0)
610 return RUNTIME_INF;
611
612 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
613}
614
615
616
617static inline int task_current(struct rq *rq, struct task_struct *p)
618{
619 return rq->curr == p;
620}
621
622static inline int task_running(struct rq *rq, struct task_struct *p)
623{
624#ifdef CONFIG_SMP
625 return p->on_cpu;
626#else
627 return task_current(rq, p);
628#endif
629}
630
631
632#ifndef prepare_arch_switch
633# define prepare_arch_switch(next) do { } while (0)
634#endif
635#ifndef finish_arch_switch
636# define finish_arch_switch(prev) do { } while (0)
637#endif
638
639#ifndef __ARCH_WANT_UNLOCKED_CTXSW
640static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
641{
642#ifdef CONFIG_SMP
643 /*
644 * We can optimise this out completely for !SMP, because the
645 * SMP rebalancing from interrupt is the only thing that cares
646 * here.
647 */
648 next->on_cpu = 1;
649#endif
650}
651
652static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
653{
654#ifdef CONFIG_SMP
655 /*
656 * After ->on_cpu is cleared, the task can be moved to a different CPU.
657 * We must ensure this doesn't happen until the switch is completely
658 * finished.
659 */
660 smp_wmb();
661 prev->on_cpu = 0;
662#endif
663#ifdef CONFIG_DEBUG_SPINLOCK
664 /* this is a valid case when another task releases the spinlock */
665 rq->lock.owner = current;
666#endif
667 /*
668 * If we are tracking spinlock dependencies then we have to
669 * fix up the runqueue lock - which gets 'carried over' from
670 * prev into current:
671 */
672 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
673
674 raw_spin_unlock_irq(&rq->lock);
675}
676
677#else /* __ARCH_WANT_UNLOCKED_CTXSW */
678static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
679{
680#ifdef CONFIG_SMP
681 /*
682 * We can optimise this out completely for !SMP, because the
683 * SMP rebalancing from interrupt is the only thing that cares
684 * here.
685 */
686 next->on_cpu = 1;
687#endif
688#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
689 raw_spin_unlock_irq(&rq->lock);
690#else
691 raw_spin_unlock(&rq->lock);
692#endif
693}
694
695static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
696{
697#ifdef CONFIG_SMP
698 /*
699 * After ->on_cpu is cleared, the task can be moved to a different CPU.
700 * We must ensure this doesn't happen until the switch is completely
701 * finished.
702 */
703 smp_wmb();
704 prev->on_cpu = 0;
705#endif
706#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
707 local_irq_enable();
708#endif
709}
710#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
711
712
713static inline void update_load_add(struct load_weight *lw, unsigned long inc)
714{
715 lw->weight += inc;
716 lw->inv_weight = 0;
717}
718
719static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
720{
721 lw->weight -= dec;
722 lw->inv_weight = 0;
723}
724
725static inline void update_load_set(struct load_weight *lw, unsigned long w)
726{
727 lw->weight = w;
728 lw->inv_weight = 0;
729}
730
731/*
732 * To aid in avoiding the subversion of "niceness" due to uneven distribution
733 * of tasks with abnormal "nice" values across CPUs the contribution that
734 * each task makes to its run queue's load is weighted according to its
735 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
736 * scaled version of the new time slice allocation that they receive on time
737 * slice expiry etc.
738 */
739
740#define WEIGHT_IDLEPRIO 3
741#define WMULT_IDLEPRIO 1431655765
742
743/*
744 * Nice levels are multiplicative, with a gentle 10% change for every
745 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
746 * nice 1, it will get ~10% less CPU time than another CPU-bound task
747 * that remained on nice 0.
748 *
749 * The "10% effect" is relative and cumulative: from _any_ nice level,
750 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
751 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
752 * If a task goes up by ~10% and another task goes down by ~10% then
753 * the relative distance between them is ~25%.)
754 */
755static const int prio_to_weight[40] = {
756 /* -20 */ 88761, 71755, 56483, 46273, 36291,
757 /* -15 */ 29154, 23254, 18705, 14949, 11916,
758 /* -10 */ 9548, 7620, 6100, 4904, 3906,
759 /* -5 */ 3121, 2501, 1991, 1586, 1277,
760 /* 0 */ 1024, 820, 655, 526, 423,
761 /* 5 */ 335, 272, 215, 172, 137,
762 /* 10 */ 110, 87, 70, 56, 45,
763 /* 15 */ 36, 29, 23, 18, 15,
764};
765
766/*
767 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
768 *
769 * In cases where the weight does not change often, we can use the
770 * precalculated inverse to speed up arithmetics by turning divisions
771 * into multiplications:
772 */
773static const u32 prio_to_wmult[40] = {
774 /* -20 */ 48388, 59856, 76040, 92818, 118348,
775 /* -15 */ 147320, 184698, 229616, 287308, 360437,
776 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
777 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
778 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
779 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
780 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
781 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
782};
783
784/* Time spent by the tasks of the cpu accounting group executing in ... */
785enum cpuacct_stat_index {
786 CPUACCT_STAT_USER, /* ... user mode */
787 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
788
789 CPUACCT_STAT_NSTATS,
790};
791
792
793#define sched_class_highest (&stop_sched_class)
794#define for_each_class(class) \
795 for (class = sched_class_highest; class; class = class->next)
796
797extern const struct sched_class stop_sched_class;
798extern const struct sched_class rt_sched_class;
799extern const struct sched_class fair_sched_class;
800extern const struct sched_class idle_sched_class;
801
802
803#ifdef CONFIG_SMP
804
805extern void trigger_load_balance(struct rq *rq, int cpu);
806extern void idle_balance(int this_cpu, struct rq *this_rq);
807
808#else /* CONFIG_SMP */
809
810static inline void idle_balance(int cpu, struct rq *rq)
811{
812}
813
814#endif
815
816extern void sysrq_sched_debug_show(void);
817extern void sched_init_granularity(void);
818extern void update_max_interval(void);
819extern void update_group_power(struct sched_domain *sd, int cpu);
820extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
821extern void init_sched_rt_class(void);
822extern void init_sched_fair_class(void);
823
824extern void resched_task(struct task_struct *p);
825extern void resched_cpu(int cpu);
826
827extern struct rt_bandwidth def_rt_bandwidth;
828extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
829
830extern void update_cpu_load(struct rq *this_rq);
831
832#ifdef CONFIG_CGROUP_CPUACCT
833extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
834extern void cpuacct_update_stats(struct task_struct *tsk,
835 enum cpuacct_stat_index idx, cputime_t val);
836#else
837static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
838static inline void cpuacct_update_stats(struct task_struct *tsk,
839 enum cpuacct_stat_index idx, cputime_t val) {}
840#endif
841
842static inline void inc_nr_running(struct rq *rq)
843{
844 rq->nr_running++;
845}
846
847static inline void dec_nr_running(struct rq *rq)
848{
849 rq->nr_running--;
850}
851
852extern void update_rq_clock(struct rq *rq);
853
854extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
855extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
856
857extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
858
859extern const_debug unsigned int sysctl_sched_time_avg;
860extern const_debug unsigned int sysctl_sched_nr_migrate;
861extern const_debug unsigned int sysctl_sched_migration_cost;
862
863static inline u64 sched_avg_period(void)
864{
865 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
866}
867
868void calc_load_account_idle(struct rq *this_rq);
869
870#ifdef CONFIG_SCHED_HRTICK
871
872/*
873 * Use hrtick when:
874 * - enabled by features
875 * - hrtimer is actually high res
876 */
877static inline int hrtick_enabled(struct rq *rq)
878{
879 if (!sched_feat(HRTICK))
880 return 0;
881 if (!cpu_active(cpu_of(rq)))
882 return 0;
883 return hrtimer_is_hres_active(&rq->hrtick_timer);
884}
885
886void hrtick_start(struct rq *rq, u64 delay);
887
888#endif /* CONFIG_SCHED_HRTICK */
889
890#ifdef CONFIG_SMP
891extern void sched_avg_update(struct rq *rq);
892static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
893{
894 rq->rt_avg += rt_delta;
895 sched_avg_update(rq);
896}
897#else
898static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
899static inline void sched_avg_update(struct rq *rq) { }
900#endif
901
902extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
903
904#ifdef CONFIG_SMP
905#ifdef CONFIG_PREEMPT
906
907static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
908
909/*
910 * fair double_lock_balance: Safely acquires both rq->locks in a fair
911 * way at the expense of forcing extra atomic operations in all
912 * invocations. This assures that the double_lock is acquired using the
913 * same underlying policy as the spinlock_t on this architecture, which
914 * reduces latency compared to the unfair variant below. However, it
915 * also adds more overhead and therefore may reduce throughput.
916 */
917static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
918 __releases(this_rq->lock)
919 __acquires(busiest->lock)
920 __acquires(this_rq->lock)
921{
922 raw_spin_unlock(&this_rq->lock);
923 double_rq_lock(this_rq, busiest);
924
925 return 1;
926}
927
928#else
929/*
930 * Unfair double_lock_balance: Optimizes throughput at the expense of
931 * latency by eliminating extra atomic operations when the locks are
932 * already in proper order on entry. This favors lower cpu-ids and will
933 * grant the double lock to lower cpus over higher ids under contention,
934 * regardless of entry order into the function.
935 */
936static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
937 __releases(this_rq->lock)
938 __acquires(busiest->lock)
939 __acquires(this_rq->lock)
940{
941 int ret = 0;
942
943 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
944 if (busiest < this_rq) {
945 raw_spin_unlock(&this_rq->lock);
946 raw_spin_lock(&busiest->lock);
947 raw_spin_lock_nested(&this_rq->lock,
948 SINGLE_DEPTH_NESTING);
949 ret = 1;
950 } else
951 raw_spin_lock_nested(&busiest->lock,
952 SINGLE_DEPTH_NESTING);
953 }
954 return ret;
955}
956
957#endif /* CONFIG_PREEMPT */
958
959/*
960 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
961 */
962static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
963{
964 if (unlikely(!irqs_disabled())) {
965 /* printk() doesn't work good under rq->lock */
966 raw_spin_unlock(&this_rq->lock);
967 BUG_ON(1);
968 }
969
970 return _double_lock_balance(this_rq, busiest);
971}
972
973static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
974 __releases(busiest->lock)
975{
976 raw_spin_unlock(&busiest->lock);
977 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
978}
979
980/*
981 * double_rq_lock - safely lock two runqueues
982 *
983 * Note this does not disable interrupts like task_rq_lock,
984 * you need to do so manually before calling.
985 */
986static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
987 __acquires(rq1->lock)
988 __acquires(rq2->lock)
989{
990 BUG_ON(!irqs_disabled());
991 if (rq1 == rq2) {
992 raw_spin_lock(&rq1->lock);
993 __acquire(rq2->lock); /* Fake it out ;) */
994 } else {
995 if (rq1 < rq2) {
996 raw_spin_lock(&rq1->lock);
997 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
998 } else {
999 raw_spin_lock(&rq2->lock);
1000 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1001 }
1002 }
1003}
1004
1005/*
1006 * double_rq_unlock - safely unlock two runqueues
1007 *
1008 * Note this does not restore interrupts like task_rq_unlock,
1009 * you need to do so manually after calling.
1010 */
1011static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1012 __releases(rq1->lock)
1013 __releases(rq2->lock)
1014{
1015 raw_spin_unlock(&rq1->lock);
1016 if (rq1 != rq2)
1017 raw_spin_unlock(&rq2->lock);
1018 else
1019 __release(rq2->lock);
1020}
1021
1022#else /* CONFIG_SMP */
1023
1024/*
1025 * double_rq_lock - safely lock two runqueues
1026 *
1027 * Note this does not disable interrupts like task_rq_lock,
1028 * you need to do so manually before calling.
1029 */
1030static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1031 __acquires(rq1->lock)
1032 __acquires(rq2->lock)
1033{
1034 BUG_ON(!irqs_disabled());
1035 BUG_ON(rq1 != rq2);
1036 raw_spin_lock(&rq1->lock);
1037 __acquire(rq2->lock); /* Fake it out ;) */
1038}
1039
1040/*
1041 * double_rq_unlock - safely unlock two runqueues
1042 *
1043 * Note this does not restore interrupts like task_rq_unlock,
1044 * you need to do so manually after calling.
1045 */
1046static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1047 __releases(rq1->lock)
1048 __releases(rq2->lock)
1049{
1050 BUG_ON(rq1 != rq2);
1051 raw_spin_unlock(&rq1->lock);
1052 __release(rq2->lock);
1053}
1054
1055#endif
1056
1057extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1058extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1059extern void print_cfs_stats(struct seq_file *m, int cpu);
1060extern void print_rt_stats(struct seq_file *m, int cpu);
1061
1062extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1063extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1064extern void unthrottle_offline_cfs_rqs(struct rq *rq);
1065
1066extern void account_cfs_bandwidth_used(int enabled, int was_enabled);