2 #include <linux/sched.h>
3 #include <linux/mutex.h>
4 #include <linux/spinlock.h>
5 #include <linux/stop_machine.h>
6 #include <linux/slab.h>
10 extern __read_mostly int scheduler_running;
13 * Convert user-nice values [ -20 ... 0 ... 19 ]
14 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
17 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
18 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
19 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
22 * 'User priority' is the nice value converted to something we
23 * can work with better when scaling various scheduler parameters,
24 * it's a [ 0 ... 39 ] range.
26 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
27 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
28 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
31 * Helpers for converting nanosecond timing to jiffy resolution
33 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
35 #define NICE_0_LOAD SCHED_LOAD_SCALE
36 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
39 * These are the 'tuning knobs' of the scheduler:
43 * single value that denotes runtime == period, ie unlimited time.
45 #define RUNTIME_INF ((u64)~0ULL)
47 static inline int rt_policy(int policy)
49 if (policy == SCHED_FIFO || policy == SCHED_RR)
54 static inline int task_has_rt_policy(struct task_struct *p)
56 return rt_policy(p->policy);
60 * This is the priority-queue data structure of the RT scheduling class:
62 struct rt_prio_array {
63 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
64 struct list_head queue[MAX_RT_PRIO];
68 /* nests inside the rq lock: */
69 raw_spinlock_t rt_runtime_lock;
72 struct hrtimer rt_period_timer;
75 extern struct mutex sched_domains_mutex;
77 #ifdef CONFIG_CGROUP_SCHED
79 #include <linux/cgroup.h>
84 extern struct list_head task_groups;
86 struct cfs_bandwidth {
87 #ifdef CONFIG_CFS_BANDWIDTH
94 int idle, timer_active;
95 struct hrtimer period_timer, slack_timer;
96 struct list_head throttled_cfs_rq;
99 int nr_periods, nr_throttled;
104 /* task group related information */
106 struct cgroup_subsys_state css;
108 #ifdef CONFIG_FAIR_GROUP_SCHED
109 /* schedulable entities of this group on each cpu */
110 struct sched_entity **se;
111 /* runqueue "owned" by this group on each cpu */
112 struct cfs_rq **cfs_rq;
113 unsigned long shares;
115 atomic_t load_weight;
118 #ifdef CONFIG_RT_GROUP_SCHED
119 struct sched_rt_entity **rt_se;
120 struct rt_rq **rt_rq;
122 struct rt_bandwidth rt_bandwidth;
126 struct list_head list;
128 struct task_group *parent;
129 struct list_head siblings;
130 struct list_head children;
132 #ifdef CONFIG_SCHED_AUTOGROUP
133 struct autogroup *autogroup;
136 struct cfs_bandwidth cfs_bandwidth;
139 #ifdef CONFIG_FAIR_GROUP_SCHED
140 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
143 * A weight of 0 or 1 can cause arithmetics problems.
144 * A weight of a cfs_rq is the sum of weights of which entities
145 * are queued on this cfs_rq, so a weight of a entity should not be
146 * too large, so as the shares value of a task group.
147 * (The default weight is 1024 - so there's no practical
148 * limitation from this.)
150 #define MIN_SHARES (1UL << 1)
151 #define MAX_SHARES (1UL << 18)
154 /* Default task group.
155 * Every task in system belong to this group at bootup.
157 extern struct task_group root_task_group;
159 typedef int (*tg_visitor)(struct task_group *, void *);
161 extern int walk_tg_tree_from(struct task_group *from,
162 tg_visitor down, tg_visitor up, void *data);
165 * Iterate the full tree, calling @down when first entering a node and @up when
166 * leaving it for the final time.
168 * Caller must hold rcu_lock or sufficient equivalent.
170 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
172 return walk_tg_tree_from(&root_task_group, down, up, data);
175 extern int tg_nop(struct task_group *tg, void *data);
177 extern void free_fair_sched_group(struct task_group *tg);
178 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
179 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
180 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
181 struct sched_entity *se, int cpu,
182 struct sched_entity *parent);
183 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
184 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
186 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
187 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
188 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
190 extern void free_rt_sched_group(struct task_group *tg);
191 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
192 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
193 struct sched_rt_entity *rt_se, int cpu,
194 struct sched_rt_entity *parent);
196 #else /* CONFIG_CGROUP_SCHED */
198 struct cfs_bandwidth { };
200 #endif /* CONFIG_CGROUP_SCHED */
202 /* CFS-related fields in a runqueue */
204 struct load_weight load;
205 unsigned int nr_running, h_nr_running;
210 u64 min_vruntime_copy;
213 struct rb_root tasks_timeline;
214 struct rb_node *rb_leftmost;
217 * 'curr' points to currently running entity on this cfs_rq.
218 * It is set to NULL otherwise (i.e when none are currently running).
220 struct sched_entity *curr, *next, *last, *skip;
222 #ifdef CONFIG_SCHED_DEBUG
223 unsigned int nr_spread_over;
226 #ifdef CONFIG_FAIR_GROUP_SCHED
227 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
230 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
231 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
232 * (like users, containers etc.)
234 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
235 * list is used during load balance.
238 struct list_head leaf_cfs_rq_list;
239 struct task_group *tg; /* group that "owns" this runqueue */
243 * h_load = weight * f(tg)
245 * Where f(tg) is the recursive weight fraction assigned to
248 unsigned long h_load;
251 * Maintaining per-cpu shares distribution for group scheduling
253 * load_stamp is the last time we updated the load average
254 * load_last is the last time we updated the load average and saw load
255 * load_unacc_exec_time is currently unaccounted execution time
259 u64 load_stamp, load_last, load_unacc_exec_time;
261 unsigned long load_contribution;
262 #endif /* CONFIG_SMP */
263 #ifdef CONFIG_CFS_BANDWIDTH
266 s64 runtime_remaining;
268 u64 throttled_timestamp;
269 int throttled, throttle_count;
270 struct list_head throttled_list;
271 #endif /* CONFIG_CFS_BANDWIDTH */
272 #endif /* CONFIG_FAIR_GROUP_SCHED */
275 static inline int rt_bandwidth_enabled(void)
277 return sysctl_sched_rt_runtime >= 0;
280 /* Real-Time classes' related field in a runqueue: */
282 struct rt_prio_array active;
283 unsigned int rt_nr_running;
284 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
286 int curr; /* highest queued rt task prio */
288 int next; /* next highest */
293 unsigned long rt_nr_migratory;
294 unsigned long rt_nr_total;
296 struct plist_head pushable_tasks;
301 /* Nests inside the rq lock: */
302 raw_spinlock_t rt_runtime_lock;
304 #ifdef CONFIG_RT_GROUP_SCHED
305 unsigned long rt_nr_boosted;
308 struct list_head leaf_rt_rq_list;
309 struct task_group *tg;
316 * We add the notion of a root-domain which will be used to define per-domain
317 * variables. Each exclusive cpuset essentially defines an island domain by
318 * fully partitioning the member cpus from any other cpuset. Whenever a new
319 * exclusive cpuset is created, we also create and attach a new root-domain
328 cpumask_var_t online;
331 * The "RT overload" flag: it gets set if a CPU has more than
332 * one runnable RT task.
334 cpumask_var_t rto_mask;
335 struct cpupri cpupri;
338 extern struct root_domain def_root_domain;
340 #endif /* CONFIG_SMP */
343 * This is the main, per-CPU runqueue data structure.
345 * Locking rule: those places that want to lock multiple runqueues
346 * (such as the load balancing or the thread migration code), lock
347 * acquire operations must be ordered by ascending &runqueue.
354 * nr_running and cpu_load should be in the same cacheline because
355 * remote CPUs use both these fields when doing load calculation.
357 unsigned int nr_running;
358 #define CPU_LOAD_IDX_MAX 5
359 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
360 unsigned long last_load_update_tick;
363 unsigned long nohz_flags;
365 int skip_clock_update;
367 /* capture load from *all* tasks on this cpu: */
368 struct load_weight load;
369 unsigned long nr_load_updates;
375 #ifdef CONFIG_FAIR_GROUP_SCHED
376 /* list of leaf cfs_rq on this cpu: */
377 struct list_head leaf_cfs_rq_list;
379 unsigned long h_load_throttle;
380 #endif /* CONFIG_SMP */
381 #endif /* CONFIG_FAIR_GROUP_SCHED */
383 #ifdef CONFIG_RT_GROUP_SCHED
384 struct list_head leaf_rt_rq_list;
388 * This is part of a global counter where only the total sum
389 * over all CPUs matters. A task can increase this counter on
390 * one CPU and if it got migrated afterwards it may decrease
391 * it on another CPU. Always updated under the runqueue lock:
393 unsigned long nr_uninterruptible;
395 struct task_struct *curr, *idle, *stop;
396 unsigned long next_balance;
397 struct mm_struct *prev_mm;
405 struct root_domain *rd;
406 struct sched_domain *sd;
408 unsigned long cpu_power;
410 unsigned char idle_balance;
411 /* For active balancing */
415 struct cpu_stop_work active_balance_work;
416 /* cpu of this runqueue: */
420 struct list_head cfs_tasks;
422 #ifdef CONFIG_SCHED_NUMA
423 unsigned long offnode_running;
424 unsigned long offnode_weight;
425 struct list_head offnode_tasks;
434 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
437 #ifdef CONFIG_PARAVIRT
440 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
441 u64 prev_steal_time_rq;
444 /* calc_load related fields */
445 unsigned long calc_load_update;
446 long calc_load_active;
448 #ifdef CONFIG_SCHED_HRTICK
450 int hrtick_csd_pending;
451 struct call_single_data hrtick_csd;
453 struct hrtimer hrtick_timer;
456 #ifdef CONFIG_SCHEDSTATS
458 struct sched_info rq_sched_info;
459 unsigned long long rq_cpu_time;
460 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
462 /* sys_sched_yield() stats */
463 unsigned int yld_count;
465 /* schedule() stats */
466 unsigned int sched_count;
467 unsigned int sched_goidle;
469 /* try_to_wake_up() stats */
470 unsigned int ttwu_count;
471 unsigned int ttwu_local;
475 struct llist_head wake_list;
479 static inline int cpu_of(struct rq *rq)
488 DECLARE_PER_CPU(struct rq, runqueues);
490 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
491 #define this_rq() (&__get_cpu_var(runqueues))
492 #define task_rq(p) cpu_rq(task_cpu(p))
493 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
494 #define raw_rq() (&__raw_get_cpu_var(runqueues))
496 #ifdef CONFIG_SCHED_NUMA
497 static inline bool offnode_task(struct task_struct *t)
499 return t->node != -1 && t->node != cpu_to_node(task_cpu(t));
502 static inline struct list_head *offnode_tasks(struct rq *rq)
504 return &rq->offnode_tasks;
507 static inline void task_numa_free(struct task_struct *p)
509 kfree(p->numa_faults);
511 #else /* CONFIG_SCHED_NUMA */
512 static inline bool offnode_task(struct task_struct *t)
517 static inline struct list_head *offnode_tasks(struct rq *rq)
522 static inline void task_numa_free(struct task_struct *p)
525 #endif /* CONFIG_SCHED_NUMA */
529 #define rcu_dereference_check_sched_domain(p) \
530 rcu_dereference_check((p), \
531 lockdep_is_held(&sched_domains_mutex))
534 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
535 * See detach_destroy_domains: synchronize_sched for details.
537 * The domain tree of any CPU may only be accessed from within
538 * preempt-disabled sections.
540 #define for_each_domain(cpu, __sd) \
541 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
542 __sd; __sd = __sd->parent)
544 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
547 * highest_flag_domain - Return highest sched_domain containing flag.
548 * @cpu: The cpu whose highest level of sched domain is to
550 * @flag: The flag to check for the highest sched_domain
553 * Returns the highest sched_domain of a cpu which contains the given flag.
555 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
557 struct sched_domain *sd, *hsd = NULL;
559 for_each_domain(cpu, sd) {
560 if (!(sd->flags & flag))
568 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
569 DECLARE_PER_CPU(int, sd_llc_id);
570 DECLARE_PER_CPU(struct sched_domain *, sd_node);
572 extern int group_balance_cpu(struct sched_group *sg);
574 #endif /* CONFIG_SMP */
577 #include "auto_group.h"
579 #ifdef CONFIG_CGROUP_SCHED
582 * Return the group to which this tasks belongs.
584 * We cannot use task_subsys_state() and friends because the cgroup
585 * subsystem changes that value before the cgroup_subsys::attach() method
586 * is called, therefore we cannot pin it and might observe the wrong value.
588 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
589 * core changes this before calling sched_move_task().
591 * Instead we use a 'copy' which is updated from sched_move_task() while
592 * holding both task_struct::pi_lock and rq::lock.
594 static inline struct task_group *task_group(struct task_struct *p)
596 return p->sched_task_group;
599 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
600 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
602 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
603 struct task_group *tg = task_group(p);
606 #ifdef CONFIG_FAIR_GROUP_SCHED
607 p->se.cfs_rq = tg->cfs_rq[cpu];
608 p->se.parent = tg->se[cpu];
611 #ifdef CONFIG_RT_GROUP_SCHED
612 p->rt.rt_rq = tg->rt_rq[cpu];
613 p->rt.parent = tg->rt_se[cpu];
617 #else /* CONFIG_CGROUP_SCHED */
619 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
620 static inline struct task_group *task_group(struct task_struct *p)
625 #endif /* CONFIG_CGROUP_SCHED */
627 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
632 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
633 * successfuly executed on another CPU. We must ensure that updates of
634 * per-task data have been completed by this moment.
637 task_thread_info(p)->cpu = cpu;
642 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
644 #ifdef CONFIG_SCHED_DEBUG
645 # include <linux/static_key.h>
646 # define const_debug __read_mostly
648 # define const_debug const
651 extern const_debug unsigned int sysctl_sched_features;
653 #define SCHED_FEAT(name, enabled) \
654 __SCHED_FEAT_##name ,
657 #include "features.h"
663 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
664 static __always_inline bool static_branch__true(struct static_key *key)
666 return static_key_true(key); /* Not out of line branch. */
669 static __always_inline bool static_branch__false(struct static_key *key)
671 return static_key_false(key); /* Out of line branch. */
674 #define SCHED_FEAT(name, enabled) \
675 static __always_inline bool static_branch_##name(struct static_key *key) \
677 return static_branch__##enabled(key); \
680 #include "features.h"
684 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
685 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
686 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
687 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
688 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
690 #ifdef CONFIG_SCHED_NUMA
691 #define sched_feat_numa(x) sched_feat(x)
693 #define sched_feat_numa(x) (0)
696 static inline u64 global_rt_period(void)
698 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
701 static inline u64 global_rt_runtime(void)
703 if (sysctl_sched_rt_runtime < 0)
706 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
711 static inline int task_current(struct rq *rq, struct task_struct *p)
713 return rq->curr == p;
716 static inline int task_running(struct rq *rq, struct task_struct *p)
721 return task_current(rq, p);
726 #ifndef prepare_arch_switch
727 # define prepare_arch_switch(next) do { } while (0)
729 #ifndef finish_arch_switch
730 # define finish_arch_switch(prev) do { } while (0)
732 #ifndef finish_arch_post_lock_switch
733 # define finish_arch_post_lock_switch() do { } while (0)
736 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
737 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
741 * We can optimise this out completely for !SMP, because the
742 * SMP rebalancing from interrupt is the only thing that cares
749 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
753 * After ->on_cpu is cleared, the task can be moved to a different CPU.
754 * We must ensure this doesn't happen until the switch is completely
760 #ifdef CONFIG_DEBUG_SPINLOCK
761 /* this is a valid case when another task releases the spinlock */
762 rq->lock.owner = current;
765 * If we are tracking spinlock dependencies then we have to
766 * fix up the runqueue lock - which gets 'carried over' from
769 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
771 raw_spin_unlock_irq(&rq->lock);
774 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
775 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
779 * We can optimise this out completely for !SMP, because the
780 * SMP rebalancing from interrupt is the only thing that cares
785 raw_spin_unlock(&rq->lock);
788 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
792 * After ->on_cpu is cleared, the task can be moved to a different CPU.
793 * We must ensure this doesn't happen until the switch is completely
801 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
804 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
810 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
816 static inline void update_load_set(struct load_weight *lw, unsigned long w)
823 * To aid in avoiding the subversion of "niceness" due to uneven distribution
824 * of tasks with abnormal "nice" values across CPUs the contribution that
825 * each task makes to its run queue's load is weighted according to its
826 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
827 * scaled version of the new time slice allocation that they receive on time
831 #define WEIGHT_IDLEPRIO 3
832 #define WMULT_IDLEPRIO 1431655765
835 * Nice levels are multiplicative, with a gentle 10% change for every
836 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
837 * nice 1, it will get ~10% less CPU time than another CPU-bound task
838 * that remained on nice 0.
840 * The "10% effect" is relative and cumulative: from _any_ nice level,
841 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
842 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
843 * If a task goes up by ~10% and another task goes down by ~10% then
844 * the relative distance between them is ~25%.)
846 static const int prio_to_weight[40] = {
847 /* -20 */ 88761, 71755, 56483, 46273, 36291,
848 /* -15 */ 29154, 23254, 18705, 14949, 11916,
849 /* -10 */ 9548, 7620, 6100, 4904, 3906,
850 /* -5 */ 3121, 2501, 1991, 1586, 1277,
851 /* 0 */ 1024, 820, 655, 526, 423,
852 /* 5 */ 335, 272, 215, 172, 137,
853 /* 10 */ 110, 87, 70, 56, 45,
854 /* 15 */ 36, 29, 23, 18, 15,
858 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
860 * In cases where the weight does not change often, we can use the
861 * precalculated inverse to speed up arithmetics by turning divisions
862 * into multiplications:
864 static const u32 prio_to_wmult[40] = {
865 /* -20 */ 48388, 59856, 76040, 92818, 118348,
866 /* -15 */ 147320, 184698, 229616, 287308, 360437,
867 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
868 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
869 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
870 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
871 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
872 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
875 /* Time spent by the tasks of the cpu accounting group executing in ... */
876 enum cpuacct_stat_index {
877 CPUACCT_STAT_USER, /* ... user mode */
878 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
884 #define sched_class_highest (&stop_sched_class)
885 #define for_each_class(class) \
886 for (class = sched_class_highest; class; class = class->next)
888 extern const struct sched_class stop_sched_class;
889 extern const struct sched_class rt_sched_class;
890 extern const struct sched_class fair_sched_class;
891 extern const struct sched_class idle_sched_class;
896 extern void trigger_load_balance(struct rq *rq, int cpu);
897 extern void idle_balance(int this_cpu, struct rq *this_rq);
899 #else /* CONFIG_SMP */
901 static inline void idle_balance(int cpu, struct rq *rq)
907 extern void sysrq_sched_debug_show(void);
908 extern void sched_init_granularity(void);
909 extern void update_max_interval(void);
910 extern void update_group_power(struct sched_domain *sd, int cpu);
911 extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
912 extern void init_sched_rt_class(void);
913 extern void init_sched_fair_class(void);
915 extern void resched_task(struct task_struct *p);
916 extern void resched_cpu(int cpu);
918 extern struct rt_bandwidth def_rt_bandwidth;
919 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
921 extern void update_idle_cpu_load(struct rq *this_rq);
923 #ifdef CONFIG_CGROUP_CPUACCT
924 #include <linux/cgroup.h>
925 /* track cpu usage of a group of tasks and its child groups */
927 struct cgroup_subsys_state css;
928 /* cpuusage holds pointer to a u64-type object on every cpu */
929 u64 __percpu *cpuusage;
930 struct kernel_cpustat __percpu *cpustat;
933 extern struct cgroup_subsys cpuacct_subsys;
934 extern struct cpuacct root_cpuacct;
936 /* return cpu accounting group corresponding to this container */
937 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
939 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
940 struct cpuacct, css);
943 /* return cpu accounting group to which this task belongs */
944 static inline struct cpuacct *task_ca(struct task_struct *tsk)
946 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
947 struct cpuacct, css);
950 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
952 if (!ca || !ca->css.cgroup->parent)
954 return cgroup_ca(ca->css.cgroup->parent);
957 extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
959 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
962 #ifdef CONFIG_PARAVIRT
963 static inline u64 steal_ticks(u64 steal)
965 if (unlikely(steal > NSEC_PER_SEC))
966 return div_u64(steal, TICK_NSEC);
968 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
972 static inline void inc_nr_running(struct rq *rq)
977 static inline void dec_nr_running(struct rq *rq)
982 extern void update_rq_clock(struct rq *rq);
984 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
985 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
987 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
989 extern const_debug unsigned int sysctl_sched_time_avg;
990 extern const_debug unsigned int sysctl_sched_nr_migrate;
991 extern const_debug unsigned int sysctl_sched_migration_cost;
993 static inline u64 sched_avg_period(void)
995 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
998 #ifdef CONFIG_SCHED_HRTICK
1002 * - enabled by features
1003 * - hrtimer is actually high res
1005 static inline int hrtick_enabled(struct rq *rq)
1007 if (!sched_feat(HRTICK))
1009 if (!cpu_active(cpu_of(rq)))
1011 return hrtimer_is_hres_active(&rq->hrtick_timer);
1014 void hrtick_start(struct rq *rq, u64 delay);
1018 static inline int hrtick_enabled(struct rq *rq)
1023 #endif /* CONFIG_SCHED_HRTICK */
1026 extern void sched_avg_update(struct rq *rq);
1027 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1029 rq->rt_avg += rt_delta;
1030 sched_avg_update(rq);
1033 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1034 static inline void sched_avg_update(struct rq *rq) { }
1037 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1040 #ifdef CONFIG_PREEMPT
1042 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1045 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1046 * way at the expense of forcing extra atomic operations in all
1047 * invocations. This assures that the double_lock is acquired using the
1048 * same underlying policy as the spinlock_t on this architecture, which
1049 * reduces latency compared to the unfair variant below. However, it
1050 * also adds more overhead and therefore may reduce throughput.
1052 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1053 __releases(this_rq->lock)
1054 __acquires(busiest->lock)
1055 __acquires(this_rq->lock)
1057 raw_spin_unlock(&this_rq->lock);
1058 double_rq_lock(this_rq, busiest);
1065 * Unfair double_lock_balance: Optimizes throughput at the expense of
1066 * latency by eliminating extra atomic operations when the locks are
1067 * already in proper order on entry. This favors lower cpu-ids and will
1068 * grant the double lock to lower cpus over higher ids under contention,
1069 * regardless of entry order into the function.
1071 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1072 __releases(this_rq->lock)
1073 __acquires(busiest->lock)
1074 __acquires(this_rq->lock)
1078 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1079 if (busiest < this_rq) {
1080 raw_spin_unlock(&this_rq->lock);
1081 raw_spin_lock(&busiest->lock);
1082 raw_spin_lock_nested(&this_rq->lock,
1083 SINGLE_DEPTH_NESTING);
1086 raw_spin_lock_nested(&busiest->lock,
1087 SINGLE_DEPTH_NESTING);
1092 #endif /* CONFIG_PREEMPT */
1095 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1097 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1099 if (unlikely(!irqs_disabled())) {
1100 /* printk() doesn't work good under rq->lock */
1101 raw_spin_unlock(&this_rq->lock);
1105 return _double_lock_balance(this_rq, busiest);
1108 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1109 __releases(busiest->lock)
1111 raw_spin_unlock(&busiest->lock);
1112 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1116 * double_rq_lock - safely lock two runqueues
1118 * Note this does not disable interrupts like task_rq_lock,
1119 * you need to do so manually before calling.
1121 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1122 __acquires(rq1->lock)
1123 __acquires(rq2->lock)
1125 BUG_ON(!irqs_disabled());
1127 raw_spin_lock(&rq1->lock);
1128 __acquire(rq2->lock); /* Fake it out ;) */
1131 raw_spin_lock(&rq1->lock);
1132 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1134 raw_spin_lock(&rq2->lock);
1135 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1141 * double_rq_unlock - safely unlock two runqueues
1143 * Note this does not restore interrupts like task_rq_unlock,
1144 * you need to do so manually after calling.
1146 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1147 __releases(rq1->lock)
1148 __releases(rq2->lock)
1150 raw_spin_unlock(&rq1->lock);
1152 raw_spin_unlock(&rq2->lock);
1154 __release(rq2->lock);
1157 #else /* CONFIG_SMP */
1160 * double_rq_lock - safely lock two runqueues
1162 * Note this does not disable interrupts like task_rq_lock,
1163 * you need to do so manually before calling.
1165 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1166 __acquires(rq1->lock)
1167 __acquires(rq2->lock)
1169 BUG_ON(!irqs_disabled());
1171 raw_spin_lock(&rq1->lock);
1172 __acquire(rq2->lock); /* Fake it out ;) */
1176 * double_rq_unlock - safely unlock two runqueues
1178 * Note this does not restore interrupts like task_rq_unlock,
1179 * you need to do so manually after calling.
1181 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1182 __releases(rq1->lock)
1183 __releases(rq2->lock)
1186 raw_spin_unlock(&rq1->lock);
1187 __release(rq2->lock);
1192 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1193 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1194 extern void print_cfs_stats(struct seq_file *m, int cpu);
1195 extern void print_rt_stats(struct seq_file *m, int cpu);
1197 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1198 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1200 extern void account_cfs_bandwidth_used(int enabled, int was_enabled);
1203 enum rq_nohz_flag_bits {
1209 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1212 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1214 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1215 DECLARE_PER_CPU(u64, cpu_softirq_time);
1217 #ifndef CONFIG_64BIT
1218 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1220 static inline void irq_time_write_begin(void)
1222 __this_cpu_inc(irq_time_seq.sequence);
1226 static inline void irq_time_write_end(void)
1229 __this_cpu_inc(irq_time_seq.sequence);
1232 static inline u64 irq_time_read(int cpu)
1238 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1239 irq_time = per_cpu(cpu_softirq_time, cpu) +
1240 per_cpu(cpu_hardirq_time, cpu);
1241 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1245 #else /* CONFIG_64BIT */
1246 static inline void irq_time_write_begin(void)
1250 static inline void irq_time_write_end(void)
1254 static inline u64 irq_time_read(int cpu)
1256 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1258 #endif /* CONFIG_64BIT */
1259 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */