2 #include <linux/sched.h>
3 #include <linux/mutex.h>
4 #include <linux/spinlock.h>
5 #include <linux/stop_machine.h>
9 extern __read_mostly int scheduler_running;
12 * Convert user-nice values [ -20 ... 0 ... 19 ]
13 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
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)
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.
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))
30 * Helpers for converting nanosecond timing to jiffy resolution
32 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
34 #define NICE_0_LOAD SCHED_LOAD_SCALE
35 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
38 * These are the 'tuning knobs' of the scheduler:
42 * single value that denotes runtime == period, ie unlimited time.
44 #define RUNTIME_INF ((u64)~0ULL)
46 static inline int rt_policy(int policy)
48 if (policy == SCHED_FIFO || policy == SCHED_RR)
53 static inline int task_has_rt_policy(struct task_struct *p)
55 return rt_policy(p->policy);
59 * This is the priority-queue data structure of the RT scheduling class:
61 struct rt_prio_array {
62 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
63 struct list_head queue[MAX_RT_PRIO];
67 /* nests inside the rq lock: */
68 raw_spinlock_t rt_runtime_lock;
71 struct hrtimer rt_period_timer;
74 extern struct mutex sched_domains_mutex;
76 #ifdef CONFIG_CGROUP_SCHED
78 #include <linux/cgroup.h>
83 extern struct list_head task_groups;
85 struct cfs_bandwidth {
86 #ifdef CONFIG_CFS_BANDWIDTH
93 int idle, timer_active;
94 struct hrtimer period_timer, slack_timer;
95 struct list_head throttled_cfs_rq;
98 int nr_periods, nr_throttled;
103 /* task group related information */
105 struct cgroup_subsys_state css;
107 #ifdef CONFIG_FAIR_GROUP_SCHED
108 /* schedulable entities of this group on each cpu */
109 struct sched_entity **se;
110 /* runqueue "owned" by this group on each cpu */
111 struct cfs_rq **cfs_rq;
112 unsigned long shares;
114 atomic_t load_weight;
116 atomic_t runnable_avg;
119 #ifdef CONFIG_RT_GROUP_SCHED
120 struct sched_rt_entity **rt_se;
121 struct rt_rq **rt_rq;
123 struct rt_bandwidth rt_bandwidth;
127 struct list_head list;
129 struct task_group *parent;
130 struct list_head siblings;
131 struct list_head children;
133 #ifdef CONFIG_SCHED_AUTOGROUP
134 struct autogroup *autogroup;
137 struct cfs_bandwidth cfs_bandwidth;
140 #ifdef CONFIG_FAIR_GROUP_SCHED
141 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
144 * A weight of 0 or 1 can cause arithmetics problems.
145 * A weight of a cfs_rq is the sum of weights of which entities
146 * are queued on this cfs_rq, so a weight of a entity should not be
147 * too large, so as the shares value of a task group.
148 * (The default weight is 1024 - so there's no practical
149 * limitation from this.)
151 #define MIN_SHARES (1UL << 1)
152 #define MAX_SHARES (1UL << 18)
155 /* Default task group.
156 * Every task in system belong to this group at bootup.
158 extern struct task_group root_task_group;
160 typedef int (*tg_visitor)(struct task_group *, void *);
162 extern int walk_tg_tree_from(struct task_group *from,
163 tg_visitor down, tg_visitor up, void *data);
166 * Iterate the full tree, calling @down when first entering a node and @up when
167 * leaving it for the final time.
169 * Caller must hold rcu_lock or sufficient equivalent.
171 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
173 return walk_tg_tree_from(&root_task_group, down, up, data);
176 extern int tg_nop(struct task_group *tg, void *data);
178 extern void free_fair_sched_group(struct task_group *tg);
179 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
180 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
181 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
182 struct sched_entity *se, int cpu,
183 struct sched_entity *parent);
184 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
185 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
187 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
188 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
189 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
191 extern void free_rt_sched_group(struct task_group *tg);
192 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
193 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
194 struct sched_rt_entity *rt_se, int cpu,
195 struct sched_rt_entity *parent);
197 #else /* CONFIG_CGROUP_SCHED */
199 struct cfs_bandwidth { };
201 #endif /* CONFIG_CGROUP_SCHED */
203 /* CFS-related fields in a runqueue */
205 struct load_weight load;
206 unsigned int nr_running, h_nr_running;
211 u64 min_vruntime_copy;
214 struct rb_root tasks_timeline;
215 struct rb_node *rb_leftmost;
218 * 'curr' points to currently running entity on this cfs_rq.
219 * It is set to NULL otherwise (i.e when none are currently running).
221 struct sched_entity *curr, *next, *last, *skip;
223 #ifdef CONFIG_SCHED_DEBUG
224 unsigned int nr_spread_over;
229 * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
230 * removed when useful for applications beyond shares distribution (e.g.
233 #ifdef CONFIG_FAIR_GROUP_SCHED
236 * Under CFS, load is tracked on a per-entity basis and aggregated up.
237 * This allows for the description of both thread and group usage (in
238 * the FAIR_GROUP_SCHED case).
240 u64 runnable_load_avg, blocked_load_avg;
241 atomic64_t decay_counter, removed_load;
243 #endif /* CONFIG_FAIR_GROUP_SCHED */
244 /* These always depend on CONFIG_FAIR_GROUP_SCHED */
245 #ifdef CONFIG_FAIR_GROUP_SCHED
246 u32 tg_runnable_contrib;
248 #endif /* CONFIG_FAIR_GROUP_SCHED */
251 * h_load = weight * f(tg)
253 * Where f(tg) is the recursive weight fraction assigned to
256 unsigned long h_load;
257 #endif /* CONFIG_SMP */
259 #ifdef CONFIG_FAIR_GROUP_SCHED
260 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
263 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
264 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
265 * (like users, containers etc.)
267 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
268 * list is used during load balance.
271 struct list_head leaf_cfs_rq_list;
272 struct task_group *tg; /* group that "owns" this runqueue */
274 #ifdef CONFIG_CFS_BANDWIDTH
277 s64 runtime_remaining;
279 u64 throttled_clock, throttled_clock_task;
280 u64 throttled_clock_task_time;
281 int throttled, throttle_count;
282 struct list_head throttled_list;
283 #endif /* CONFIG_CFS_BANDWIDTH */
284 #endif /* CONFIG_FAIR_GROUP_SCHED */
287 static inline int rt_bandwidth_enabled(void)
289 return sysctl_sched_rt_runtime >= 0;
292 /* Real-Time classes' related field in a runqueue: */
294 struct rt_prio_array active;
295 unsigned int rt_nr_running;
296 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
298 int curr; /* highest queued rt task prio */
300 int next; /* next highest */
305 unsigned long rt_nr_migratory;
306 unsigned long rt_nr_total;
308 struct plist_head pushable_tasks;
313 /* Nests inside the rq lock: */
314 raw_spinlock_t rt_runtime_lock;
316 #ifdef CONFIG_RT_GROUP_SCHED
317 unsigned long rt_nr_boosted;
320 struct list_head leaf_rt_rq_list;
321 struct task_group *tg;
328 * We add the notion of a root-domain which will be used to define per-domain
329 * variables. Each exclusive cpuset essentially defines an island domain by
330 * fully partitioning the member cpus from any other cpuset. Whenever a new
331 * exclusive cpuset is created, we also create and attach a new root-domain
340 cpumask_var_t online;
343 * The "RT overload" flag: it gets set if a CPU has more than
344 * one runnable RT task.
346 cpumask_var_t rto_mask;
347 struct cpupri cpupri;
350 extern struct root_domain def_root_domain;
352 #endif /* CONFIG_SMP */
355 * This is the main, per-CPU runqueue data structure.
357 * Locking rule: those places that want to lock multiple runqueues
358 * (such as the load balancing or the thread migration code), lock
359 * acquire operations must be ordered by ascending &runqueue.
366 * nr_running and cpu_load should be in the same cacheline because
367 * remote CPUs use both these fields when doing load calculation.
369 unsigned int nr_running;
370 #define CPU_LOAD_IDX_MAX 5
371 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
372 unsigned long last_load_update_tick;
375 unsigned long nohz_flags;
377 int skip_clock_update;
379 /* capture load from *all* tasks on this cpu: */
380 struct load_weight load;
381 unsigned long nr_load_updates;
387 #ifdef CONFIG_FAIR_GROUP_SCHED
388 /* list of leaf cfs_rq on this cpu: */
389 struct list_head leaf_cfs_rq_list;
391 unsigned long h_load_throttle;
392 #endif /* CONFIG_SMP */
393 #endif /* CONFIG_FAIR_GROUP_SCHED */
395 #ifdef CONFIG_RT_GROUP_SCHED
396 struct list_head leaf_rt_rq_list;
400 * This is part of a global counter where only the total sum
401 * over all CPUs matters. A task can increase this counter on
402 * one CPU and if it got migrated afterwards it may decrease
403 * it on another CPU. Always updated under the runqueue lock:
405 unsigned long nr_uninterruptible;
407 struct task_struct *curr, *idle, *stop;
408 unsigned long next_balance;
409 struct mm_struct *prev_mm;
417 struct root_domain *rd;
418 struct sched_domain *sd;
420 unsigned long cpu_power;
422 unsigned char idle_balance;
423 /* For active balancing */
427 struct cpu_stop_work active_balance_work;
428 /* cpu of this runqueue: */
432 struct list_head cfs_tasks;
440 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
443 #ifdef CONFIG_PARAVIRT
446 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
447 u64 prev_steal_time_rq;
450 /* calc_load related fields */
451 unsigned long calc_load_update;
452 long calc_load_active;
454 #ifdef CONFIG_SCHED_HRTICK
456 int hrtick_csd_pending;
457 struct call_single_data hrtick_csd;
459 struct hrtimer hrtick_timer;
462 #ifdef CONFIG_SCHEDSTATS
464 struct sched_info rq_sched_info;
465 unsigned long long rq_cpu_time;
466 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
468 /* sys_sched_yield() stats */
469 unsigned int yld_count;
471 /* schedule() stats */
472 unsigned int sched_count;
473 unsigned int sched_goidle;
475 /* try_to_wake_up() stats */
476 unsigned int ttwu_count;
477 unsigned int ttwu_local;
481 struct llist_head wake_list;
484 struct sched_avg avg;
487 static inline int cpu_of(struct rq *rq)
496 DECLARE_PER_CPU(struct rq, runqueues);
498 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
499 #define this_rq() (&__get_cpu_var(runqueues))
500 #define task_rq(p) cpu_rq(task_cpu(p))
501 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
502 #define raw_rq() (&__raw_get_cpu_var(runqueues))
506 #define rcu_dereference_check_sched_domain(p) \
507 rcu_dereference_check((p), \
508 lockdep_is_held(&sched_domains_mutex))
511 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
512 * See detach_destroy_domains: synchronize_sched for details.
514 * The domain tree of any CPU may only be accessed from within
515 * preempt-disabled sections.
517 #define for_each_domain(cpu, __sd) \
518 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
519 __sd; __sd = __sd->parent)
521 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
524 * highest_flag_domain - Return highest sched_domain containing flag.
525 * @cpu: The cpu whose highest level of sched domain is to
527 * @flag: The flag to check for the highest sched_domain
530 * Returns the highest sched_domain of a cpu which contains the given flag.
532 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
534 struct sched_domain *sd, *hsd = NULL;
536 for_each_domain(cpu, sd) {
537 if (!(sd->flags & flag))
545 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
546 DECLARE_PER_CPU(int, sd_llc_id);
548 extern int group_balance_cpu(struct sched_group *sg);
550 #endif /* CONFIG_SMP */
553 #include "auto_group.h"
555 #ifdef CONFIG_CGROUP_SCHED
558 * Return the group to which this tasks belongs.
560 * We cannot use task_subsys_state() and friends because the cgroup
561 * subsystem changes that value before the cgroup_subsys::attach() method
562 * is called, therefore we cannot pin it and might observe the wrong value.
564 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
565 * core changes this before calling sched_move_task().
567 * Instead we use a 'copy' which is updated from sched_move_task() while
568 * holding both task_struct::pi_lock and rq::lock.
570 static inline struct task_group *task_group(struct task_struct *p)
572 return p->sched_task_group;
575 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
576 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
578 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
579 struct task_group *tg = task_group(p);
582 #ifdef CONFIG_FAIR_GROUP_SCHED
583 p->se.cfs_rq = tg->cfs_rq[cpu];
584 p->se.parent = tg->se[cpu];
587 #ifdef CONFIG_RT_GROUP_SCHED
588 p->rt.rt_rq = tg->rt_rq[cpu];
589 p->rt.parent = tg->rt_se[cpu];
593 #else /* CONFIG_CGROUP_SCHED */
595 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
596 static inline struct task_group *task_group(struct task_struct *p)
601 #endif /* CONFIG_CGROUP_SCHED */
603 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
608 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
609 * successfuly executed on another CPU. We must ensure that updates of
610 * per-task data have been completed by this moment.
613 task_thread_info(p)->cpu = cpu;
618 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
620 #ifdef CONFIG_SCHED_DEBUG
621 # include <linux/static_key.h>
622 # define const_debug __read_mostly
624 # define const_debug const
627 extern const_debug unsigned int sysctl_sched_features;
629 #define SCHED_FEAT(name, enabled) \
630 __SCHED_FEAT_##name ,
633 #include "features.h"
639 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
640 static __always_inline bool static_branch__true(struct static_key *key)
642 return static_key_true(key); /* Not out of line branch. */
645 static __always_inline bool static_branch__false(struct static_key *key)
647 return static_key_false(key); /* Out of line branch. */
650 #define SCHED_FEAT(name, enabled) \
651 static __always_inline bool static_branch_##name(struct static_key *key) \
653 return static_branch__##enabled(key); \
656 #include "features.h"
660 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
661 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
662 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
663 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
664 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
666 static inline u64 global_rt_period(void)
668 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
671 static inline u64 global_rt_runtime(void)
673 if (sysctl_sched_rt_runtime < 0)
676 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
681 static inline int task_current(struct rq *rq, struct task_struct *p)
683 return rq->curr == p;
686 static inline int task_running(struct rq *rq, struct task_struct *p)
691 return task_current(rq, p);
696 #ifndef prepare_arch_switch
697 # define prepare_arch_switch(next) do { } while (0)
699 #ifndef finish_arch_switch
700 # define finish_arch_switch(prev) do { } while (0)
702 #ifndef finish_arch_post_lock_switch
703 # define finish_arch_post_lock_switch() do { } while (0)
706 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
707 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
711 * We can optimise this out completely for !SMP, because the
712 * SMP rebalancing from interrupt is the only thing that cares
719 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
723 * After ->on_cpu is cleared, the task can be moved to a different CPU.
724 * We must ensure this doesn't happen until the switch is completely
730 #ifdef CONFIG_DEBUG_SPINLOCK
731 /* this is a valid case when another task releases the spinlock */
732 rq->lock.owner = current;
735 * If we are tracking spinlock dependencies then we have to
736 * fix up the runqueue lock - which gets 'carried over' from
739 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
741 raw_spin_unlock_irq(&rq->lock);
744 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
745 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
749 * We can optimise this out completely for !SMP, because the
750 * SMP rebalancing from interrupt is the only thing that cares
755 raw_spin_unlock(&rq->lock);
758 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
762 * After ->on_cpu is cleared, the task can be moved to a different CPU.
763 * We must ensure this doesn't happen until the switch is completely
771 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
774 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
780 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
786 static inline void update_load_set(struct load_weight *lw, unsigned long w)
793 * To aid in avoiding the subversion of "niceness" due to uneven distribution
794 * of tasks with abnormal "nice" values across CPUs the contribution that
795 * each task makes to its run queue's load is weighted according to its
796 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
797 * scaled version of the new time slice allocation that they receive on time
801 #define WEIGHT_IDLEPRIO 3
802 #define WMULT_IDLEPRIO 1431655765
805 * Nice levels are multiplicative, with a gentle 10% change for every
806 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
807 * nice 1, it will get ~10% less CPU time than another CPU-bound task
808 * that remained on nice 0.
810 * The "10% effect" is relative and cumulative: from _any_ nice level,
811 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
812 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
813 * If a task goes up by ~10% and another task goes down by ~10% then
814 * the relative distance between them is ~25%.)
816 static const int prio_to_weight[40] = {
817 /* -20 */ 88761, 71755, 56483, 46273, 36291,
818 /* -15 */ 29154, 23254, 18705, 14949, 11916,
819 /* -10 */ 9548, 7620, 6100, 4904, 3906,
820 /* -5 */ 3121, 2501, 1991, 1586, 1277,
821 /* 0 */ 1024, 820, 655, 526, 423,
822 /* 5 */ 335, 272, 215, 172, 137,
823 /* 10 */ 110, 87, 70, 56, 45,
824 /* 15 */ 36, 29, 23, 18, 15,
828 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
830 * In cases where the weight does not change often, we can use the
831 * precalculated inverse to speed up arithmetics by turning divisions
832 * into multiplications:
834 static const u32 prio_to_wmult[40] = {
835 /* -20 */ 48388, 59856, 76040, 92818, 118348,
836 /* -15 */ 147320, 184698, 229616, 287308, 360437,
837 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
838 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
839 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
840 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
841 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
842 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
845 /* Time spent by the tasks of the cpu accounting group executing in ... */
846 enum cpuacct_stat_index {
847 CPUACCT_STAT_USER, /* ... user mode */
848 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
854 #define sched_class_highest (&stop_sched_class)
855 #define for_each_class(class) \
856 for (class = sched_class_highest; class; class = class->next)
858 extern const struct sched_class stop_sched_class;
859 extern const struct sched_class rt_sched_class;
860 extern const struct sched_class fair_sched_class;
861 extern const struct sched_class idle_sched_class;
866 extern void trigger_load_balance(struct rq *rq, int cpu);
867 extern void idle_balance(int this_cpu, struct rq *this_rq);
869 #else /* CONFIG_SMP */
871 static inline void idle_balance(int cpu, struct rq *rq)
877 extern void sysrq_sched_debug_show(void);
878 extern void sched_init_granularity(void);
879 extern void update_max_interval(void);
880 extern void update_group_power(struct sched_domain *sd, int cpu);
881 extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
882 extern void init_sched_rt_class(void);
883 extern void init_sched_fair_class(void);
885 extern void resched_task(struct task_struct *p);
886 extern void resched_cpu(int cpu);
888 extern struct rt_bandwidth def_rt_bandwidth;
889 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
891 extern void update_idle_cpu_load(struct rq *this_rq);
893 #ifdef CONFIG_CGROUP_CPUACCT
894 #include <linux/cgroup.h>
895 /* track cpu usage of a group of tasks and its child groups */
897 struct cgroup_subsys_state css;
898 /* cpuusage holds pointer to a u64-type object on every cpu */
899 u64 __percpu *cpuusage;
900 struct kernel_cpustat __percpu *cpustat;
903 extern struct cgroup_subsys cpuacct_subsys;
904 extern struct cpuacct root_cpuacct;
906 /* return cpu accounting group corresponding to this container */
907 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
909 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
910 struct cpuacct, css);
913 /* return cpu accounting group to which this task belongs */
914 static inline struct cpuacct *task_ca(struct task_struct *tsk)
916 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
917 struct cpuacct, css);
920 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
922 if (!ca || !ca->css.cgroup->parent)
924 return cgroup_ca(ca->css.cgroup->parent);
927 extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
929 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
932 #ifdef CONFIG_PARAVIRT
933 static inline u64 steal_ticks(u64 steal)
935 if (unlikely(steal > NSEC_PER_SEC))
936 return div_u64(steal, TICK_NSEC);
938 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
942 static inline void inc_nr_running(struct rq *rq)
947 static inline void dec_nr_running(struct rq *rq)
952 extern void update_rq_clock(struct rq *rq);
954 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
955 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
957 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
959 extern const_debug unsigned int sysctl_sched_time_avg;
960 extern const_debug unsigned int sysctl_sched_nr_migrate;
961 extern const_debug unsigned int sysctl_sched_migration_cost;
963 static inline u64 sched_avg_period(void)
965 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
968 #ifdef CONFIG_SCHED_HRTICK
972 * - enabled by features
973 * - hrtimer is actually high res
975 static inline int hrtick_enabled(struct rq *rq)
977 if (!sched_feat(HRTICK))
979 if (!cpu_active(cpu_of(rq)))
981 return hrtimer_is_hres_active(&rq->hrtick_timer);
984 void hrtick_start(struct rq *rq, u64 delay);
988 static inline int hrtick_enabled(struct rq *rq)
993 #endif /* CONFIG_SCHED_HRTICK */
996 extern void sched_avg_update(struct rq *rq);
997 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
999 rq->rt_avg += rt_delta;
1000 sched_avg_update(rq);
1003 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1004 static inline void sched_avg_update(struct rq *rq) { }
1007 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1010 #ifdef CONFIG_PREEMPT
1012 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1015 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1016 * way at the expense of forcing extra atomic operations in all
1017 * invocations. This assures that the double_lock is acquired using the
1018 * same underlying policy as the spinlock_t on this architecture, which
1019 * reduces latency compared to the unfair variant below. However, it
1020 * also adds more overhead and therefore may reduce throughput.
1022 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1023 __releases(this_rq->lock)
1024 __acquires(busiest->lock)
1025 __acquires(this_rq->lock)
1027 raw_spin_unlock(&this_rq->lock);
1028 double_rq_lock(this_rq, busiest);
1035 * Unfair double_lock_balance: Optimizes throughput at the expense of
1036 * latency by eliminating extra atomic operations when the locks are
1037 * already in proper order on entry. This favors lower cpu-ids and will
1038 * grant the double lock to lower cpus over higher ids under contention,
1039 * regardless of entry order into the function.
1041 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1042 __releases(this_rq->lock)
1043 __acquires(busiest->lock)
1044 __acquires(this_rq->lock)
1048 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1049 if (busiest < this_rq) {
1050 raw_spin_unlock(&this_rq->lock);
1051 raw_spin_lock(&busiest->lock);
1052 raw_spin_lock_nested(&this_rq->lock,
1053 SINGLE_DEPTH_NESTING);
1056 raw_spin_lock_nested(&busiest->lock,
1057 SINGLE_DEPTH_NESTING);
1062 #endif /* CONFIG_PREEMPT */
1065 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1067 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1069 if (unlikely(!irqs_disabled())) {
1070 /* printk() doesn't work good under rq->lock */
1071 raw_spin_unlock(&this_rq->lock);
1075 return _double_lock_balance(this_rq, busiest);
1078 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1079 __releases(busiest->lock)
1081 raw_spin_unlock(&busiest->lock);
1082 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1086 * double_rq_lock - safely lock two runqueues
1088 * Note this does not disable interrupts like task_rq_lock,
1089 * you need to do so manually before calling.
1091 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1092 __acquires(rq1->lock)
1093 __acquires(rq2->lock)
1095 BUG_ON(!irqs_disabled());
1097 raw_spin_lock(&rq1->lock);
1098 __acquire(rq2->lock); /* Fake it out ;) */
1101 raw_spin_lock(&rq1->lock);
1102 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1104 raw_spin_lock(&rq2->lock);
1105 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1111 * double_rq_unlock - safely unlock two runqueues
1113 * Note this does not restore interrupts like task_rq_unlock,
1114 * you need to do so manually after calling.
1116 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1117 __releases(rq1->lock)
1118 __releases(rq2->lock)
1120 raw_spin_unlock(&rq1->lock);
1122 raw_spin_unlock(&rq2->lock);
1124 __release(rq2->lock);
1127 #else /* CONFIG_SMP */
1130 * double_rq_lock - safely lock two runqueues
1132 * Note this does not disable interrupts like task_rq_lock,
1133 * you need to do so manually before calling.
1135 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1136 __acquires(rq1->lock)
1137 __acquires(rq2->lock)
1139 BUG_ON(!irqs_disabled());
1141 raw_spin_lock(&rq1->lock);
1142 __acquire(rq2->lock); /* Fake it out ;) */
1146 * double_rq_unlock - safely unlock two runqueues
1148 * Note this does not restore interrupts like task_rq_unlock,
1149 * you need to do so manually after calling.
1151 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1152 __releases(rq1->lock)
1153 __releases(rq2->lock)
1156 raw_spin_unlock(&rq1->lock);
1157 __release(rq2->lock);
1162 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1163 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1164 extern void print_cfs_stats(struct seq_file *m, int cpu);
1165 extern void print_rt_stats(struct seq_file *m, int cpu);
1167 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1168 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1170 extern void account_cfs_bandwidth_used(int enabled, int was_enabled);
1173 enum rq_nohz_flag_bits {
1179 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1182 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1184 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1185 DECLARE_PER_CPU(u64, cpu_softirq_time);
1187 #ifndef CONFIG_64BIT
1188 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1190 static inline void irq_time_write_begin(void)
1192 __this_cpu_inc(irq_time_seq.sequence);
1196 static inline void irq_time_write_end(void)
1199 __this_cpu_inc(irq_time_seq.sequence);
1202 static inline u64 irq_time_read(int cpu)
1208 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1209 irq_time = per_cpu(cpu_softirq_time, cpu) +
1210 per_cpu(cpu_hardirq_time, cpu);
1211 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1215 #else /* CONFIG_64BIT */
1216 static inline void irq_time_write_begin(void)
1220 static inline void irq_time_write_end(void)
1224 static inline u64 irq_time_read(int cpu)
1226 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1228 #endif /* CONFIG_64BIT */
1229 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */