2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/mutex.h>
5 #include <linux/spinlock.h>
6 #include <linux/stop_machine.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;
117 atomic_t runnable_avg;
120 #ifdef CONFIG_RT_GROUP_SCHED
121 struct sched_rt_entity **rt_se;
122 struct rt_rq **rt_rq;
124 struct rt_bandwidth rt_bandwidth;
128 struct list_head list;
130 struct task_group *parent;
131 struct list_head siblings;
132 struct list_head children;
134 #ifdef CONFIG_SCHED_AUTOGROUP
135 struct autogroup *autogroup;
138 struct cfs_bandwidth cfs_bandwidth;
141 #ifdef CONFIG_FAIR_GROUP_SCHED
142 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
145 * A weight of 0 or 1 can cause arithmetics problems.
146 * A weight of a cfs_rq is the sum of weights of which entities
147 * are queued on this cfs_rq, so a weight of a entity should not be
148 * too large, so as the shares value of a task group.
149 * (The default weight is 1024 - so there's no practical
150 * limitation from this.)
152 #define MIN_SHARES (1UL << 1)
153 #define MAX_SHARES (1UL << 18)
156 /* Default task group.
157 * Every task in system belong to this group at bootup.
159 extern struct task_group root_task_group;
161 typedef int (*tg_visitor)(struct task_group *, void *);
163 extern int walk_tg_tree_from(struct task_group *from,
164 tg_visitor down, tg_visitor up, void *data);
167 * Iterate the full tree, calling @down when first entering a node and @up when
168 * leaving it for the final time.
170 * Caller must hold rcu_lock or sufficient equivalent.
172 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
174 return walk_tg_tree_from(&root_task_group, down, up, data);
177 extern int tg_nop(struct task_group *tg, void *data);
179 extern void free_fair_sched_group(struct task_group *tg);
180 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
181 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
182 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
183 struct sched_entity *se, int cpu,
184 struct sched_entity *parent);
185 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
186 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
188 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
189 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
190 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
192 extern void free_rt_sched_group(struct task_group *tg);
193 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
194 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
195 struct sched_rt_entity *rt_se, int cpu,
196 struct sched_rt_entity *parent);
198 #else /* CONFIG_CGROUP_SCHED */
200 struct cfs_bandwidth { };
202 #endif /* CONFIG_CGROUP_SCHED */
204 /* CFS-related fields in a runqueue */
206 struct load_weight load;
207 unsigned int nr_running, h_nr_running;
212 u64 min_vruntime_copy;
215 struct rb_root tasks_timeline;
216 struct rb_node *rb_leftmost;
219 * 'curr' points to currently running entity on this cfs_rq.
220 * It is set to NULL otherwise (i.e when none are currently running).
222 struct sched_entity *curr, *next, *last, *skip;
224 #ifdef CONFIG_SCHED_DEBUG
225 unsigned int nr_spread_over;
230 * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
231 * removed when useful for applications beyond shares distribution (e.g.
234 #ifdef CONFIG_FAIR_GROUP_SCHED
237 * Under CFS, load is tracked on a per-entity basis and aggregated up.
238 * This allows for the description of both thread and group usage (in
239 * the FAIR_GROUP_SCHED case).
241 u64 runnable_load_avg, blocked_load_avg;
242 atomic64_t decay_counter, removed_load;
244 #endif /* CONFIG_FAIR_GROUP_SCHED */
245 /* These always depend on CONFIG_FAIR_GROUP_SCHED */
246 #ifdef CONFIG_FAIR_GROUP_SCHED
247 u32 tg_runnable_contrib;
249 #endif /* CONFIG_FAIR_GROUP_SCHED */
252 * h_load = weight * f(tg)
254 * Where f(tg) is the recursive weight fraction assigned to
257 unsigned long h_load;
258 #endif /* CONFIG_SMP */
260 #ifdef CONFIG_FAIR_GROUP_SCHED
261 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
264 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
265 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
266 * (like users, containers etc.)
268 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
269 * list is used during load balance.
272 struct list_head leaf_cfs_rq_list;
273 struct task_group *tg; /* group that "owns" this runqueue */
275 #ifdef CONFIG_CFS_BANDWIDTH
278 s64 runtime_remaining;
280 u64 throttled_clock, throttled_clock_task;
281 u64 throttled_clock_task_time;
282 int throttled, throttle_count;
283 struct list_head throttled_list;
284 #endif /* CONFIG_CFS_BANDWIDTH */
285 #endif /* CONFIG_FAIR_GROUP_SCHED */
288 static inline int rt_bandwidth_enabled(void)
290 return sysctl_sched_rt_runtime >= 0;
293 /* Real-Time classes' related field in a runqueue: */
295 struct rt_prio_array active;
296 unsigned int rt_nr_running;
297 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
299 int curr; /* highest queued rt task prio */
301 int next; /* next highest */
306 unsigned long rt_nr_migratory;
307 unsigned long rt_nr_total;
309 struct plist_head pushable_tasks;
314 /* Nests inside the rq lock: */
315 raw_spinlock_t rt_runtime_lock;
317 #ifdef CONFIG_RT_GROUP_SCHED
318 unsigned long rt_nr_boosted;
321 struct list_head leaf_rt_rq_list;
322 struct task_group *tg;
329 * We add the notion of a root-domain which will be used to define per-domain
330 * variables. Each exclusive cpuset essentially defines an island domain by
331 * fully partitioning the member cpus from any other cpuset. Whenever a new
332 * exclusive cpuset is created, we also create and attach a new root-domain
341 cpumask_var_t online;
344 * The "RT overload" flag: it gets set if a CPU has more than
345 * one runnable RT task.
347 cpumask_var_t rto_mask;
348 struct cpupri cpupri;
351 extern struct root_domain def_root_domain;
353 #endif /* CONFIG_SMP */
356 * This is the main, per-CPU runqueue data structure.
358 * Locking rule: those places that want to lock multiple runqueues
359 * (such as the load balancing or the thread migration code), lock
360 * acquire operations must be ordered by ascending &runqueue.
367 * nr_running and cpu_load should be in the same cacheline because
368 * remote CPUs use both these fields when doing load calculation.
370 unsigned int nr_running;
371 #define CPU_LOAD_IDX_MAX 5
372 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
373 unsigned long last_load_update_tick;
376 unsigned long nohz_flags;
378 int skip_clock_update;
380 /* capture load from *all* tasks on this cpu: */
381 struct load_weight load;
382 unsigned long nr_load_updates;
388 #ifdef CONFIG_FAIR_GROUP_SCHED
389 /* list of leaf cfs_rq on this cpu: */
390 struct list_head leaf_cfs_rq_list;
392 unsigned long h_load_throttle;
393 #endif /* CONFIG_SMP */
394 #endif /* CONFIG_FAIR_GROUP_SCHED */
396 #ifdef CONFIG_RT_GROUP_SCHED
397 struct list_head leaf_rt_rq_list;
401 * This is part of a global counter where only the total sum
402 * over all CPUs matters. A task can increase this counter on
403 * one CPU and if it got migrated afterwards it may decrease
404 * it on another CPU. Always updated under the runqueue lock:
406 unsigned long nr_uninterruptible;
408 struct task_struct *curr, *idle, *stop;
409 unsigned long next_balance;
410 struct mm_struct *prev_mm;
418 struct root_domain *rd;
419 struct sched_domain *sd;
421 unsigned long cpu_power;
423 unsigned char idle_balance;
424 /* For active balancing */
428 struct cpu_stop_work active_balance_work;
429 /* cpu of this runqueue: */
433 struct list_head cfs_tasks;
441 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
444 #ifdef CONFIG_PARAVIRT
447 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
448 u64 prev_steal_time_rq;
451 /* calc_load related fields */
452 unsigned long calc_load_update;
453 long calc_load_active;
455 #ifdef CONFIG_SCHED_HRTICK
457 int hrtick_csd_pending;
458 struct call_single_data hrtick_csd;
460 struct hrtimer hrtick_timer;
463 #ifdef CONFIG_SCHEDSTATS
465 struct sched_info rq_sched_info;
466 unsigned long long rq_cpu_time;
467 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
469 /* sys_sched_yield() stats */
470 unsigned int yld_count;
472 /* schedule() stats */
473 unsigned int sched_count;
474 unsigned int sched_goidle;
476 /* try_to_wake_up() stats */
477 unsigned int ttwu_count;
478 unsigned int ttwu_local;
482 struct llist_head wake_list;
485 struct sched_avg avg;
488 static inline int cpu_of(struct rq *rq)
497 DECLARE_PER_CPU(struct rq, runqueues);
499 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
500 #define this_rq() (&__get_cpu_var(runqueues))
501 #define task_rq(p) cpu_rq(task_cpu(p))
502 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
503 #define raw_rq() (&__raw_get_cpu_var(runqueues))
507 #define rcu_dereference_check_sched_domain(p) \
508 rcu_dereference_check((p), \
509 lockdep_is_held(&sched_domains_mutex))
512 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
513 * See detach_destroy_domains: synchronize_sched for details.
515 * The domain tree of any CPU may only be accessed from within
516 * preempt-disabled sections.
518 #define for_each_domain(cpu, __sd) \
519 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
520 __sd; __sd = __sd->parent)
522 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
525 * highest_flag_domain - Return highest sched_domain containing flag.
526 * @cpu: The cpu whose highest level of sched domain is to
528 * @flag: The flag to check for the highest sched_domain
531 * Returns the highest sched_domain of a cpu which contains the given flag.
533 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
535 struct sched_domain *sd, *hsd = NULL;
537 for_each_domain(cpu, sd) {
538 if (!(sd->flags & flag))
546 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
547 DECLARE_PER_CPU(int, sd_llc_id);
549 extern int group_balance_cpu(struct sched_group *sg);
551 #endif /* CONFIG_SMP */
554 #include "auto_group.h"
556 #ifdef CONFIG_CGROUP_SCHED
559 * Return the group to which this tasks belongs.
561 * We cannot use task_subsys_state() and friends because the cgroup
562 * subsystem changes that value before the cgroup_subsys::attach() method
563 * is called, therefore we cannot pin it and might observe the wrong value.
565 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
566 * core changes this before calling sched_move_task().
568 * Instead we use a 'copy' which is updated from sched_move_task() while
569 * holding both task_struct::pi_lock and rq::lock.
571 static inline struct task_group *task_group(struct task_struct *p)
573 return p->sched_task_group;
576 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
577 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
579 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
580 struct task_group *tg = task_group(p);
583 #ifdef CONFIG_FAIR_GROUP_SCHED
584 p->se.cfs_rq = tg->cfs_rq[cpu];
585 p->se.parent = tg->se[cpu];
588 #ifdef CONFIG_RT_GROUP_SCHED
589 p->rt.rt_rq = tg->rt_rq[cpu];
590 p->rt.parent = tg->rt_se[cpu];
594 #else /* CONFIG_CGROUP_SCHED */
596 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
597 static inline struct task_group *task_group(struct task_struct *p)
602 #endif /* CONFIG_CGROUP_SCHED */
604 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
609 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
610 * successfuly executed on another CPU. We must ensure that updates of
611 * per-task data have been completed by this moment.
614 task_thread_info(p)->cpu = cpu;
619 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
621 #ifdef CONFIG_SCHED_DEBUG
622 # include <linux/static_key.h>
623 # define const_debug __read_mostly
625 # define const_debug const
628 extern const_debug unsigned int sysctl_sched_features;
630 #define SCHED_FEAT(name, enabled) \
631 __SCHED_FEAT_##name ,
634 #include "features.h"
640 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
641 static __always_inline bool static_branch__true(struct static_key *key)
643 return static_key_true(key); /* Not out of line branch. */
646 static __always_inline bool static_branch__false(struct static_key *key)
648 return static_key_false(key); /* Out of line branch. */
651 #define SCHED_FEAT(name, enabled) \
652 static __always_inline bool static_branch_##name(struct static_key *key) \
654 return static_branch__##enabled(key); \
657 #include "features.h"
661 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
662 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
663 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
664 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
665 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
667 #ifdef CONFIG_NUMA_BALANCING
668 #define sched_feat_numa(x) sched_feat(x)
669 #ifdef CONFIG_SCHED_DEBUG
670 #define numabalancing_enabled sched_feat_numa(NUMA)
672 extern bool numabalancing_enabled;
673 #endif /* CONFIG_SCHED_DEBUG */
675 #define sched_feat_numa(x) (0)
676 #define numabalancing_enabled (0)
677 #endif /* CONFIG_NUMA_BALANCING */
679 static inline u64 global_rt_period(void)
681 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
684 static inline u64 global_rt_runtime(void)
686 if (sysctl_sched_rt_runtime < 0)
689 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
694 static inline int task_current(struct rq *rq, struct task_struct *p)
696 return rq->curr == p;
699 static inline int task_running(struct rq *rq, struct task_struct *p)
704 return task_current(rq, p);
709 #ifndef prepare_arch_switch
710 # define prepare_arch_switch(next) do { } while (0)
712 #ifndef finish_arch_switch
713 # define finish_arch_switch(prev) do { } while (0)
715 #ifndef finish_arch_post_lock_switch
716 # define finish_arch_post_lock_switch() do { } while (0)
719 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
720 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
724 * We can optimise this out completely for !SMP, because the
725 * SMP rebalancing from interrupt is the only thing that cares
732 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
736 * After ->on_cpu is cleared, the task can be moved to a different CPU.
737 * We must ensure this doesn't happen until the switch is completely
743 #ifdef CONFIG_DEBUG_SPINLOCK
744 /* this is a valid case when another task releases the spinlock */
745 rq->lock.owner = current;
748 * If we are tracking spinlock dependencies then we have to
749 * fix up the runqueue lock - which gets 'carried over' from
752 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
754 raw_spin_unlock_irq(&rq->lock);
757 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
758 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
762 * We can optimise this out completely for !SMP, because the
763 * SMP rebalancing from interrupt is the only thing that cares
768 raw_spin_unlock(&rq->lock);
771 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
775 * After ->on_cpu is cleared, the task can be moved to a different CPU.
776 * We must ensure this doesn't happen until the switch is completely
784 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
787 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
793 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
799 static inline void update_load_set(struct load_weight *lw, unsigned long w)
806 * To aid in avoiding the subversion of "niceness" due to uneven distribution
807 * of tasks with abnormal "nice" values across CPUs the contribution that
808 * each task makes to its run queue's load is weighted according to its
809 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
810 * scaled version of the new time slice allocation that they receive on time
814 #define WEIGHT_IDLEPRIO 3
815 #define WMULT_IDLEPRIO 1431655765
818 * Nice levels are multiplicative, with a gentle 10% change for every
819 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
820 * nice 1, it will get ~10% less CPU time than another CPU-bound task
821 * that remained on nice 0.
823 * The "10% effect" is relative and cumulative: from _any_ nice level,
824 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
825 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
826 * If a task goes up by ~10% and another task goes down by ~10% then
827 * the relative distance between them is ~25%.)
829 static const int prio_to_weight[40] = {
830 /* -20 */ 88761, 71755, 56483, 46273, 36291,
831 /* -15 */ 29154, 23254, 18705, 14949, 11916,
832 /* -10 */ 9548, 7620, 6100, 4904, 3906,
833 /* -5 */ 3121, 2501, 1991, 1586, 1277,
834 /* 0 */ 1024, 820, 655, 526, 423,
835 /* 5 */ 335, 272, 215, 172, 137,
836 /* 10 */ 110, 87, 70, 56, 45,
837 /* 15 */ 36, 29, 23, 18, 15,
841 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
843 * In cases where the weight does not change often, we can use the
844 * precalculated inverse to speed up arithmetics by turning divisions
845 * into multiplications:
847 static const u32 prio_to_wmult[40] = {
848 /* -20 */ 48388, 59856, 76040, 92818, 118348,
849 /* -15 */ 147320, 184698, 229616, 287308, 360437,
850 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
851 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
852 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
853 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
854 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
855 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
858 /* Time spent by the tasks of the cpu accounting group executing in ... */
859 enum cpuacct_stat_index {
860 CPUACCT_STAT_USER, /* ... user mode */
861 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
867 #define sched_class_highest (&stop_sched_class)
868 #define for_each_class(class) \
869 for (class = sched_class_highest; class; class = class->next)
871 extern const struct sched_class stop_sched_class;
872 extern const struct sched_class rt_sched_class;
873 extern const struct sched_class fair_sched_class;
874 extern const struct sched_class idle_sched_class;
879 extern void trigger_load_balance(struct rq *rq, int cpu);
880 extern void idle_balance(int this_cpu, struct rq *this_rq);
882 #else /* CONFIG_SMP */
884 static inline void idle_balance(int cpu, struct rq *rq)
890 extern void sysrq_sched_debug_show(void);
891 extern void sched_init_granularity(void);
892 extern void update_max_interval(void);
893 extern void update_group_power(struct sched_domain *sd, int cpu);
894 extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
895 extern void init_sched_rt_class(void);
896 extern void init_sched_fair_class(void);
898 extern void resched_task(struct task_struct *p);
899 extern void resched_cpu(int cpu);
901 extern struct rt_bandwidth def_rt_bandwidth;
902 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
904 extern void update_idle_cpu_load(struct rq *this_rq);
906 #ifdef CONFIG_CGROUP_CPUACCT
907 #include <linux/cgroup.h>
908 /* track cpu usage of a group of tasks and its child groups */
910 struct cgroup_subsys_state css;
911 /* cpuusage holds pointer to a u64-type object on every cpu */
912 u64 __percpu *cpuusage;
913 struct kernel_cpustat __percpu *cpustat;
916 extern struct cgroup_subsys cpuacct_subsys;
917 extern struct cpuacct root_cpuacct;
919 /* return cpu accounting group corresponding to this container */
920 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
922 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
923 struct cpuacct, css);
926 /* return cpu accounting group to which this task belongs */
927 static inline struct cpuacct *task_ca(struct task_struct *tsk)
929 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
930 struct cpuacct, css);
933 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
935 if (!ca || !ca->css.cgroup->parent)
937 return cgroup_ca(ca->css.cgroup->parent);
940 extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
942 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
945 #ifdef CONFIG_PARAVIRT
946 static inline u64 steal_ticks(u64 steal)
948 if (unlikely(steal > NSEC_PER_SEC))
949 return div_u64(steal, TICK_NSEC);
951 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
955 static inline void inc_nr_running(struct rq *rq)
960 static inline void dec_nr_running(struct rq *rq)
965 extern void update_rq_clock(struct rq *rq);
967 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
968 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
970 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
972 extern const_debug unsigned int sysctl_sched_time_avg;
973 extern const_debug unsigned int sysctl_sched_nr_migrate;
974 extern const_debug unsigned int sysctl_sched_migration_cost;
976 static inline u64 sched_avg_period(void)
978 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
981 #ifdef CONFIG_SCHED_HRTICK
985 * - enabled by features
986 * - hrtimer is actually high res
988 static inline int hrtick_enabled(struct rq *rq)
990 if (!sched_feat(HRTICK))
992 if (!cpu_active(cpu_of(rq)))
994 return hrtimer_is_hres_active(&rq->hrtick_timer);
997 void hrtick_start(struct rq *rq, u64 delay);
1001 static inline int hrtick_enabled(struct rq *rq)
1006 #endif /* CONFIG_SCHED_HRTICK */
1009 extern void sched_avg_update(struct rq *rq);
1010 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1012 rq->rt_avg += rt_delta;
1013 sched_avg_update(rq);
1016 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1017 static inline void sched_avg_update(struct rq *rq) { }
1020 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1023 #ifdef CONFIG_PREEMPT
1025 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1028 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1029 * way at the expense of forcing extra atomic operations in all
1030 * invocations. This assures that the double_lock is acquired using the
1031 * same underlying policy as the spinlock_t on this architecture, which
1032 * reduces latency compared to the unfair variant below. However, it
1033 * also adds more overhead and therefore may reduce throughput.
1035 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1036 __releases(this_rq->lock)
1037 __acquires(busiest->lock)
1038 __acquires(this_rq->lock)
1040 raw_spin_unlock(&this_rq->lock);
1041 double_rq_lock(this_rq, busiest);
1048 * Unfair double_lock_balance: Optimizes throughput at the expense of
1049 * latency by eliminating extra atomic operations when the locks are
1050 * already in proper order on entry. This favors lower cpu-ids and will
1051 * grant the double lock to lower cpus over higher ids under contention,
1052 * regardless of entry order into the function.
1054 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1055 __releases(this_rq->lock)
1056 __acquires(busiest->lock)
1057 __acquires(this_rq->lock)
1061 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1062 if (busiest < this_rq) {
1063 raw_spin_unlock(&this_rq->lock);
1064 raw_spin_lock(&busiest->lock);
1065 raw_spin_lock_nested(&this_rq->lock,
1066 SINGLE_DEPTH_NESTING);
1069 raw_spin_lock_nested(&busiest->lock,
1070 SINGLE_DEPTH_NESTING);
1075 #endif /* CONFIG_PREEMPT */
1078 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1080 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1082 if (unlikely(!irqs_disabled())) {
1083 /* printk() doesn't work good under rq->lock */
1084 raw_spin_unlock(&this_rq->lock);
1088 return _double_lock_balance(this_rq, busiest);
1091 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1092 __releases(busiest->lock)
1094 raw_spin_unlock(&busiest->lock);
1095 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1099 * double_rq_lock - safely lock two runqueues
1101 * Note this does not disable interrupts like task_rq_lock,
1102 * you need to do so manually before calling.
1104 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1105 __acquires(rq1->lock)
1106 __acquires(rq2->lock)
1108 BUG_ON(!irqs_disabled());
1110 raw_spin_lock(&rq1->lock);
1111 __acquire(rq2->lock); /* Fake it out ;) */
1114 raw_spin_lock(&rq1->lock);
1115 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1117 raw_spin_lock(&rq2->lock);
1118 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1124 * double_rq_unlock - safely unlock two runqueues
1126 * Note this does not restore interrupts like task_rq_unlock,
1127 * you need to do so manually after calling.
1129 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1130 __releases(rq1->lock)
1131 __releases(rq2->lock)
1133 raw_spin_unlock(&rq1->lock);
1135 raw_spin_unlock(&rq2->lock);
1137 __release(rq2->lock);
1140 #else /* CONFIG_SMP */
1143 * double_rq_lock - safely lock two runqueues
1145 * Note this does not disable interrupts like task_rq_lock,
1146 * you need to do so manually before calling.
1148 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1149 __acquires(rq1->lock)
1150 __acquires(rq2->lock)
1152 BUG_ON(!irqs_disabled());
1154 raw_spin_lock(&rq1->lock);
1155 __acquire(rq2->lock); /* Fake it out ;) */
1159 * double_rq_unlock - safely unlock two runqueues
1161 * Note this does not restore interrupts like task_rq_unlock,
1162 * you need to do so manually after calling.
1164 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1165 __releases(rq1->lock)
1166 __releases(rq2->lock)
1169 raw_spin_unlock(&rq1->lock);
1170 __release(rq2->lock);
1175 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1176 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1177 extern void print_cfs_stats(struct seq_file *m, int cpu);
1178 extern void print_rt_stats(struct seq_file *m, int cpu);
1180 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1181 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1183 extern void account_cfs_bandwidth_used(int enabled, int was_enabled);
1186 enum rq_nohz_flag_bits {
1192 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1195 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1197 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1198 DECLARE_PER_CPU(u64, cpu_softirq_time);
1200 #ifndef CONFIG_64BIT
1201 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1203 static inline void irq_time_write_begin(void)
1205 __this_cpu_inc(irq_time_seq.sequence);
1209 static inline void irq_time_write_end(void)
1212 __this_cpu_inc(irq_time_seq.sequence);
1215 static inline u64 irq_time_read(int cpu)
1221 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1222 irq_time = per_cpu(cpu_softirq_time, cpu) +
1223 per_cpu(cpu_hardirq_time, cpu);
1224 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1228 #else /* CONFIG_64BIT */
1229 static inline void irq_time_write_begin(void)
1233 static inline void irq_time_write_end(void)
1237 static inline u64 irq_time_read(int cpu)
1239 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1241 #endif /* CONFIG_64BIT */
1242 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */