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;
117 #ifdef CONFIG_RT_GROUP_SCHED
118 struct sched_rt_entity **rt_se;
119 struct rt_rq **rt_rq;
121 struct rt_bandwidth rt_bandwidth;
125 struct list_head list;
127 struct task_group *parent;
128 struct list_head siblings;
129 struct list_head children;
131 #ifdef CONFIG_SCHED_AUTOGROUP
132 struct autogroup *autogroup;
135 struct cfs_bandwidth cfs_bandwidth;
138 #ifdef CONFIG_FAIR_GROUP_SCHED
139 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
142 * A weight of 0 or 1 can cause arithmetics problems.
143 * A weight of a cfs_rq is the sum of weights of which entities
144 * are queued on this cfs_rq, so a weight of a entity should not be
145 * too large, so as the shares value of a task group.
146 * (The default weight is 1024 - so there's no practical
147 * limitation from this.)
149 #define MIN_SHARES (1UL << 1)
150 #define MAX_SHARES (1UL << 18)
153 /* Default task group.
154 * Every task in system belong to this group at bootup.
156 extern struct task_group root_task_group;
158 typedef int (*tg_visitor)(struct task_group *, void *);
160 extern int walk_tg_tree_from(struct task_group *from,
161 tg_visitor down, tg_visitor up, void *data);
164 * Iterate the full tree, calling @down when first entering a node and @up when
165 * leaving it for the final time.
167 * Caller must hold rcu_lock or sufficient equivalent.
169 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
171 return walk_tg_tree_from(&root_task_group, down, up, data);
174 extern int tg_nop(struct task_group *tg, void *data);
176 extern void free_fair_sched_group(struct task_group *tg);
177 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
178 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
179 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
180 struct sched_entity *se, int cpu,
181 struct sched_entity *parent);
182 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
183 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
185 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
186 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
187 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
189 extern void free_rt_sched_group(struct task_group *tg);
190 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
191 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
192 struct sched_rt_entity *rt_se, int cpu,
193 struct sched_rt_entity *parent);
195 #else /* CONFIG_CGROUP_SCHED */
197 struct cfs_bandwidth { };
199 #endif /* CONFIG_CGROUP_SCHED */
201 /* CFS-related fields in a runqueue */
203 struct load_weight load;
204 unsigned int nr_running, h_nr_running;
209 u64 min_vruntime_copy;
212 struct rb_root tasks_timeline;
213 struct rb_node *rb_leftmost;
216 * 'curr' points to currently running entity on this cfs_rq.
217 * It is set to NULL otherwise (i.e when none are currently running).
219 struct sched_entity *curr, *next, *last, *skip;
221 #ifdef CONFIG_SCHED_DEBUG
222 unsigned int nr_spread_over;
225 #ifdef CONFIG_FAIR_GROUP_SCHED
226 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
229 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
230 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
231 * (like users, containers etc.)
233 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
234 * list is used during load balance.
237 struct list_head leaf_cfs_rq_list;
238 struct task_group *tg; /* group that "owns" this runqueue */
242 * h_load = weight * f(tg)
244 * Where f(tg) is the recursive weight fraction assigned to
247 unsigned long h_load;
250 * Maintaining per-cpu shares distribution for group scheduling
252 * load_stamp is the last time we updated the load average
253 * load_last is the last time we updated the load average and saw load
254 * load_unacc_exec_time is currently unaccounted execution time
258 u64 load_stamp, load_last, load_unacc_exec_time;
260 unsigned long load_contribution;
261 #endif /* CONFIG_SMP */
262 #ifdef CONFIG_CFS_BANDWIDTH
265 s64 runtime_remaining;
267 u64 throttled_timestamp;
268 int throttled, throttle_count;
269 struct list_head throttled_list;
270 #endif /* CONFIG_CFS_BANDWIDTH */
271 #endif /* CONFIG_FAIR_GROUP_SCHED */
274 static inline int rt_bandwidth_enabled(void)
276 return sysctl_sched_rt_runtime >= 0;
279 /* Real-Time classes' related field in a runqueue: */
281 struct rt_prio_array active;
282 unsigned int rt_nr_running;
283 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
285 int curr; /* highest queued rt task prio */
287 int next; /* next highest */
292 unsigned long rt_nr_migratory;
293 unsigned long rt_nr_total;
295 struct plist_head pushable_tasks;
300 /* Nests inside the rq lock: */
301 raw_spinlock_t rt_runtime_lock;
303 #ifdef CONFIG_RT_GROUP_SCHED
304 unsigned long rt_nr_boosted;
307 struct list_head leaf_rt_rq_list;
308 struct task_group *tg;
315 * We add the notion of a root-domain which will be used to define per-domain
316 * variables. Each exclusive cpuset essentially defines an island domain by
317 * fully partitioning the member cpus from any other cpuset. Whenever a new
318 * exclusive cpuset is created, we also create and attach a new root-domain
327 cpumask_var_t online;
330 * The "RT overload" flag: it gets set if a CPU has more than
331 * one runnable RT task.
333 cpumask_var_t rto_mask;
334 struct cpupri cpupri;
337 extern struct root_domain def_root_domain;
339 #endif /* CONFIG_SMP */
342 * This is the main, per-CPU runqueue data structure.
344 * Locking rule: those places that want to lock multiple runqueues
345 * (such as the load balancing or the thread migration code), lock
346 * acquire operations must be ordered by ascending &runqueue.
353 * nr_running and cpu_load should be in the same cacheline because
354 * remote CPUs use both these fields when doing load calculation.
356 unsigned int nr_running;
357 #define CPU_LOAD_IDX_MAX 5
358 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
359 unsigned long last_load_update_tick;
362 unsigned long nohz_flags;
364 int skip_clock_update;
366 /* capture load from *all* tasks on this cpu: */
367 struct load_weight load;
368 unsigned long nr_load_updates;
374 #ifdef CONFIG_FAIR_GROUP_SCHED
375 /* list of leaf cfs_rq on this cpu: */
376 struct list_head leaf_cfs_rq_list;
378 unsigned long h_load_throttle;
379 #endif /* CONFIG_SMP */
380 #endif /* CONFIG_FAIR_GROUP_SCHED */
382 #ifdef CONFIG_RT_GROUP_SCHED
383 struct list_head leaf_rt_rq_list;
387 * This is part of a global counter where only the total sum
388 * over all CPUs matters. A task can increase this counter on
389 * one CPU and if it got migrated afterwards it may decrease
390 * it on another CPU. Always updated under the runqueue lock:
392 unsigned long nr_uninterruptible;
394 struct task_struct *curr, *idle, *stop;
395 unsigned long next_balance;
396 struct mm_struct *prev_mm;
404 struct root_domain *rd;
405 struct sched_domain *sd;
407 unsigned long cpu_power;
409 unsigned char idle_balance;
410 /* For active balancing */
414 struct cpu_stop_work active_balance_work;
415 /* cpu of this runqueue: */
419 struct list_head cfs_tasks;
427 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
430 #ifdef CONFIG_PARAVIRT
433 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
434 u64 prev_steal_time_rq;
437 /* calc_load related fields */
438 unsigned long calc_load_update;
439 long calc_load_active;
441 #ifdef CONFIG_SCHED_HRTICK
443 int hrtick_csd_pending;
444 struct call_single_data hrtick_csd;
446 struct hrtimer hrtick_timer;
449 #ifdef CONFIG_SCHEDSTATS
451 struct sched_info rq_sched_info;
452 unsigned long long rq_cpu_time;
453 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
455 /* sys_sched_yield() stats */
456 unsigned int yld_count;
458 /* schedule() stats */
459 unsigned int sched_count;
460 unsigned int sched_goidle;
462 /* try_to_wake_up() stats */
463 unsigned int ttwu_count;
464 unsigned int ttwu_local;
468 struct llist_head wake_list;
472 static inline int cpu_of(struct rq *rq)
481 DECLARE_PER_CPU(struct rq, runqueues);
483 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
484 #define this_rq() (&__get_cpu_var(runqueues))
485 #define task_rq(p) cpu_rq(task_cpu(p))
486 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
487 #define raw_rq() (&__raw_get_cpu_var(runqueues))
491 #define rcu_dereference_check_sched_domain(p) \
492 rcu_dereference_check((p), \
493 lockdep_is_held(&sched_domains_mutex))
496 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
497 * See detach_destroy_domains: synchronize_sched for details.
499 * The domain tree of any CPU may only be accessed from within
500 * preempt-disabled sections.
502 #define for_each_domain(cpu, __sd) \
503 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
504 __sd; __sd = __sd->parent)
506 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
509 * highest_flag_domain - Return highest sched_domain containing flag.
510 * @cpu: The cpu whose highest level of sched domain is to
512 * @flag: The flag to check for the highest sched_domain
515 * Returns the highest sched_domain of a cpu which contains the given flag.
517 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
519 struct sched_domain *sd, *hsd = NULL;
521 for_each_domain(cpu, sd) {
522 if (!(sd->flags & flag))
530 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
531 DECLARE_PER_CPU(int, sd_llc_id);
533 extern int group_balance_cpu(struct sched_group *sg);
535 #endif /* CONFIG_SMP */
538 #include "auto_group.h"
540 #ifdef CONFIG_CGROUP_SCHED
543 * Return the group to which this tasks belongs.
545 * We cannot use task_subsys_state() and friends because the cgroup
546 * subsystem changes that value before the cgroup_subsys::attach() method
547 * is called, therefore we cannot pin it and might observe the wrong value.
549 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
550 * core changes this before calling sched_move_task().
552 * Instead we use a 'copy' which is updated from sched_move_task() while
553 * holding both task_struct::pi_lock and rq::lock.
555 static inline struct task_group *task_group(struct task_struct *p)
557 return p->sched_task_group;
560 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
561 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
563 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
564 struct task_group *tg = task_group(p);
567 #ifdef CONFIG_FAIR_GROUP_SCHED
568 p->se.cfs_rq = tg->cfs_rq[cpu];
569 p->se.parent = tg->se[cpu];
572 #ifdef CONFIG_RT_GROUP_SCHED
573 p->rt.rt_rq = tg->rt_rq[cpu];
574 p->rt.parent = tg->rt_se[cpu];
578 #else /* CONFIG_CGROUP_SCHED */
580 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
581 static inline struct task_group *task_group(struct task_struct *p)
586 #endif /* CONFIG_CGROUP_SCHED */
588 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
593 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
594 * successfuly executed on another CPU. We must ensure that updates of
595 * per-task data have been completed by this moment.
598 task_thread_info(p)->cpu = cpu;
603 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
605 #ifdef CONFIG_SCHED_DEBUG
606 # include <linux/static_key.h>
607 # define const_debug __read_mostly
609 # define const_debug const
612 extern const_debug unsigned int sysctl_sched_features;
614 #define SCHED_FEAT(name, enabled) \
615 __SCHED_FEAT_##name ,
618 #include "features.h"
624 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
625 static __always_inline bool static_branch__true(struct static_key *key)
627 return static_key_true(key); /* Not out of line branch. */
630 static __always_inline bool static_branch__false(struct static_key *key)
632 return static_key_false(key); /* Out of line branch. */
635 #define SCHED_FEAT(name, enabled) \
636 static __always_inline bool static_branch_##name(struct static_key *key) \
638 return static_branch__##enabled(key); \
641 #include "features.h"
645 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
646 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
647 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
648 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
649 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
651 #ifdef CONFIG_NUMA_BALANCING
652 #define sched_feat_numa(x) sched_feat(x)
654 #define sched_feat_numa(x) (0)
657 static inline u64 global_rt_period(void)
659 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
662 static inline u64 global_rt_runtime(void)
664 if (sysctl_sched_rt_runtime < 0)
667 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
672 static inline int task_current(struct rq *rq, struct task_struct *p)
674 return rq->curr == p;
677 static inline int task_running(struct rq *rq, struct task_struct *p)
682 return task_current(rq, p);
687 #ifndef prepare_arch_switch
688 # define prepare_arch_switch(next) do { } while (0)
690 #ifndef finish_arch_switch
691 # define finish_arch_switch(prev) do { } while (0)
693 #ifndef finish_arch_post_lock_switch
694 # define finish_arch_post_lock_switch() do { } while (0)
697 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
698 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
702 * We can optimise this out completely for !SMP, because the
703 * SMP rebalancing from interrupt is the only thing that cares
710 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
714 * After ->on_cpu is cleared, the task can be moved to a different CPU.
715 * We must ensure this doesn't happen until the switch is completely
721 #ifdef CONFIG_DEBUG_SPINLOCK
722 /* this is a valid case when another task releases the spinlock */
723 rq->lock.owner = current;
726 * If we are tracking spinlock dependencies then we have to
727 * fix up the runqueue lock - which gets 'carried over' from
730 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
732 raw_spin_unlock_irq(&rq->lock);
735 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
736 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
740 * We can optimise this out completely for !SMP, because the
741 * SMP rebalancing from interrupt is the only thing that cares
746 raw_spin_unlock(&rq->lock);
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
762 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
765 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
771 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
777 static inline void update_load_set(struct load_weight *lw, unsigned long w)
784 * To aid in avoiding the subversion of "niceness" due to uneven distribution
785 * of tasks with abnormal "nice" values across CPUs the contribution that
786 * each task makes to its run queue's load is weighted according to its
787 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
788 * scaled version of the new time slice allocation that they receive on time
792 #define WEIGHT_IDLEPRIO 3
793 #define WMULT_IDLEPRIO 1431655765
796 * Nice levels are multiplicative, with a gentle 10% change for every
797 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
798 * nice 1, it will get ~10% less CPU time than another CPU-bound task
799 * that remained on nice 0.
801 * The "10% effect" is relative and cumulative: from _any_ nice level,
802 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
803 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
804 * If a task goes up by ~10% and another task goes down by ~10% then
805 * the relative distance between them is ~25%.)
807 static const int prio_to_weight[40] = {
808 /* -20 */ 88761, 71755, 56483, 46273, 36291,
809 /* -15 */ 29154, 23254, 18705, 14949, 11916,
810 /* -10 */ 9548, 7620, 6100, 4904, 3906,
811 /* -5 */ 3121, 2501, 1991, 1586, 1277,
812 /* 0 */ 1024, 820, 655, 526, 423,
813 /* 5 */ 335, 272, 215, 172, 137,
814 /* 10 */ 110, 87, 70, 56, 45,
815 /* 15 */ 36, 29, 23, 18, 15,
819 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
821 * In cases where the weight does not change often, we can use the
822 * precalculated inverse to speed up arithmetics by turning divisions
823 * into multiplications:
825 static const u32 prio_to_wmult[40] = {
826 /* -20 */ 48388, 59856, 76040, 92818, 118348,
827 /* -15 */ 147320, 184698, 229616, 287308, 360437,
828 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
829 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
830 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
831 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
832 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
833 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
836 /* Time spent by the tasks of the cpu accounting group executing in ... */
837 enum cpuacct_stat_index {
838 CPUACCT_STAT_USER, /* ... user mode */
839 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
845 #define sched_class_highest (&stop_sched_class)
846 #define for_each_class(class) \
847 for (class = sched_class_highest; class; class = class->next)
849 extern const struct sched_class stop_sched_class;
850 extern const struct sched_class rt_sched_class;
851 extern const struct sched_class fair_sched_class;
852 extern const struct sched_class idle_sched_class;
857 extern void trigger_load_balance(struct rq *rq, int cpu);
858 extern void idle_balance(int this_cpu, struct rq *this_rq);
860 #else /* CONFIG_SMP */
862 static inline void idle_balance(int cpu, struct rq *rq)
868 extern void sysrq_sched_debug_show(void);
869 extern void sched_init_granularity(void);
870 extern void update_max_interval(void);
871 extern void update_group_power(struct sched_domain *sd, int cpu);
872 extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
873 extern void init_sched_rt_class(void);
874 extern void init_sched_fair_class(void);
876 extern void resched_task(struct task_struct *p);
877 extern void resched_cpu(int cpu);
879 extern struct rt_bandwidth def_rt_bandwidth;
880 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
882 extern void update_idle_cpu_load(struct rq *this_rq);
884 #ifdef CONFIG_CGROUP_CPUACCT
885 #include <linux/cgroup.h>
886 /* track cpu usage of a group of tasks and its child groups */
888 struct cgroup_subsys_state css;
889 /* cpuusage holds pointer to a u64-type object on every cpu */
890 u64 __percpu *cpuusage;
891 struct kernel_cpustat __percpu *cpustat;
894 extern struct cgroup_subsys cpuacct_subsys;
895 extern struct cpuacct root_cpuacct;
897 /* return cpu accounting group corresponding to this container */
898 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
900 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
901 struct cpuacct, css);
904 /* return cpu accounting group to which this task belongs */
905 static inline struct cpuacct *task_ca(struct task_struct *tsk)
907 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
908 struct cpuacct, css);
911 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
913 if (!ca || !ca->css.cgroup->parent)
915 return cgroup_ca(ca->css.cgroup->parent);
918 extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
920 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
923 #ifdef CONFIG_PARAVIRT
924 static inline u64 steal_ticks(u64 steal)
926 if (unlikely(steal > NSEC_PER_SEC))
927 return div_u64(steal, TICK_NSEC);
929 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
933 static inline void inc_nr_running(struct rq *rq)
938 static inline void dec_nr_running(struct rq *rq)
943 extern void update_rq_clock(struct rq *rq);
945 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
946 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
948 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
950 extern const_debug unsigned int sysctl_sched_time_avg;
951 extern const_debug unsigned int sysctl_sched_nr_migrate;
952 extern const_debug unsigned int sysctl_sched_migration_cost;
954 static inline u64 sched_avg_period(void)
956 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
959 #ifdef CONFIG_SCHED_HRTICK
963 * - enabled by features
964 * - hrtimer is actually high res
966 static inline int hrtick_enabled(struct rq *rq)
968 if (!sched_feat(HRTICK))
970 if (!cpu_active(cpu_of(rq)))
972 return hrtimer_is_hres_active(&rq->hrtick_timer);
975 void hrtick_start(struct rq *rq, u64 delay);
979 static inline int hrtick_enabled(struct rq *rq)
984 #endif /* CONFIG_SCHED_HRTICK */
987 extern void sched_avg_update(struct rq *rq);
988 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
990 rq->rt_avg += rt_delta;
991 sched_avg_update(rq);
994 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
995 static inline void sched_avg_update(struct rq *rq) { }
998 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1001 #ifdef CONFIG_PREEMPT
1003 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1006 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1007 * way at the expense of forcing extra atomic operations in all
1008 * invocations. This assures that the double_lock is acquired using the
1009 * same underlying policy as the spinlock_t on this architecture, which
1010 * reduces latency compared to the unfair variant below. However, it
1011 * also adds more overhead and therefore may reduce throughput.
1013 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1014 __releases(this_rq->lock)
1015 __acquires(busiest->lock)
1016 __acquires(this_rq->lock)
1018 raw_spin_unlock(&this_rq->lock);
1019 double_rq_lock(this_rq, busiest);
1026 * Unfair double_lock_balance: Optimizes throughput at the expense of
1027 * latency by eliminating extra atomic operations when the locks are
1028 * already in proper order on entry. This favors lower cpu-ids and will
1029 * grant the double lock to lower cpus over higher ids under contention,
1030 * regardless of entry order into the function.
1032 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1033 __releases(this_rq->lock)
1034 __acquires(busiest->lock)
1035 __acquires(this_rq->lock)
1039 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1040 if (busiest < this_rq) {
1041 raw_spin_unlock(&this_rq->lock);
1042 raw_spin_lock(&busiest->lock);
1043 raw_spin_lock_nested(&this_rq->lock,
1044 SINGLE_DEPTH_NESTING);
1047 raw_spin_lock_nested(&busiest->lock,
1048 SINGLE_DEPTH_NESTING);
1053 #endif /* CONFIG_PREEMPT */
1056 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1058 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1060 if (unlikely(!irqs_disabled())) {
1061 /* printk() doesn't work good under rq->lock */
1062 raw_spin_unlock(&this_rq->lock);
1066 return _double_lock_balance(this_rq, busiest);
1069 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1070 __releases(busiest->lock)
1072 raw_spin_unlock(&busiest->lock);
1073 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1077 * double_rq_lock - safely lock two runqueues
1079 * Note this does not disable interrupts like task_rq_lock,
1080 * you need to do so manually before calling.
1082 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1083 __acquires(rq1->lock)
1084 __acquires(rq2->lock)
1086 BUG_ON(!irqs_disabled());
1088 raw_spin_lock(&rq1->lock);
1089 __acquire(rq2->lock); /* Fake it out ;) */
1092 raw_spin_lock(&rq1->lock);
1093 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1095 raw_spin_lock(&rq2->lock);
1096 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1102 * double_rq_unlock - safely unlock two runqueues
1104 * Note this does not restore interrupts like task_rq_unlock,
1105 * you need to do so manually after calling.
1107 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1108 __releases(rq1->lock)
1109 __releases(rq2->lock)
1111 raw_spin_unlock(&rq1->lock);
1113 raw_spin_unlock(&rq2->lock);
1115 __release(rq2->lock);
1118 #else /* CONFIG_SMP */
1121 * double_rq_lock - safely lock two runqueues
1123 * Note this does not disable interrupts like task_rq_lock,
1124 * you need to do so manually before calling.
1126 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1127 __acquires(rq1->lock)
1128 __acquires(rq2->lock)
1130 BUG_ON(!irqs_disabled());
1132 raw_spin_lock(&rq1->lock);
1133 __acquire(rq2->lock); /* Fake it out ;) */
1137 * double_rq_unlock - safely unlock two runqueues
1139 * Note this does not restore interrupts like task_rq_unlock,
1140 * you need to do so manually after calling.
1142 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1143 __releases(rq1->lock)
1144 __releases(rq2->lock)
1147 raw_spin_unlock(&rq1->lock);
1148 __release(rq2->lock);
1153 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1154 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1155 extern void print_cfs_stats(struct seq_file *m, int cpu);
1156 extern void print_rt_stats(struct seq_file *m, int cpu);
1158 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1159 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1161 extern void account_cfs_bandwidth_used(int enabled, int was_enabled);
1164 enum rq_nohz_flag_bits {
1170 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1173 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1175 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1176 DECLARE_PER_CPU(u64, cpu_softirq_time);
1178 #ifndef CONFIG_64BIT
1179 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1181 static inline void irq_time_write_begin(void)
1183 __this_cpu_inc(irq_time_seq.sequence);
1187 static inline void irq_time_write_end(void)
1190 __this_cpu_inc(irq_time_seq.sequence);
1193 static inline u64 irq_time_read(int cpu)
1199 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1200 irq_time = per_cpu(cpu_softirq_time, cpu) +
1201 per_cpu(cpu_hardirq_time, cpu);
1202 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1206 #else /* CONFIG_64BIT */
1207 static inline void irq_time_write_begin(void)
1211 static inline void irq_time_write_end(void)
1215 static inline u64 irq_time_read(int cpu)
1217 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1219 #endif /* CONFIG_64BIT */
1220 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */