2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/mutex.h>
6 #include <linux/spinlock.h>
7 #include <linux/stop_machine.h>
11 extern __read_mostly int scheduler_running;
14 * Convert user-nice values [ -20 ... 0 ... 19 ]
15 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
18 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
19 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
20 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
23 * 'User priority' is the nice value converted to something we
24 * can work with better when scaling various scheduler parameters,
25 * it's a [ 0 ... 39 ] range.
27 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
28 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
29 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
32 * Helpers for converting nanosecond timing to jiffy resolution
34 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
37 * Increase resolution of nice-level calculations for 64-bit architectures.
38 * The extra resolution improves shares distribution and load balancing of
39 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
40 * hierarchies, especially on larger systems. This is not a user-visible change
41 * and does not change the user-interface for setting shares/weights.
43 * We increase resolution only if we have enough bits to allow this increased
44 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
45 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
48 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
49 # define SCHED_LOAD_RESOLUTION 10
50 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
51 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
53 # define SCHED_LOAD_RESOLUTION 0
54 # define scale_load(w) (w)
55 # define scale_load_down(w) (w)
58 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
59 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
61 #define NICE_0_LOAD SCHED_LOAD_SCALE
62 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
65 * These are the 'tuning knobs' of the scheduler:
69 * single value that denotes runtime == period, ie unlimited time.
71 #define RUNTIME_INF ((u64)~0ULL)
73 static inline int rt_policy(int policy)
75 if (policy == SCHED_FIFO || policy == SCHED_RR)
80 static inline int task_has_rt_policy(struct task_struct *p)
82 return rt_policy(p->policy);
86 * This is the priority-queue data structure of the RT scheduling class:
88 struct rt_prio_array {
89 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
90 struct list_head queue[MAX_RT_PRIO];
94 /* nests inside the rq lock: */
95 raw_spinlock_t rt_runtime_lock;
98 struct hrtimer rt_period_timer;
101 extern struct mutex sched_domains_mutex;
103 #ifdef CONFIG_CGROUP_SCHED
105 #include <linux/cgroup.h>
110 extern struct list_head task_groups;
112 struct cfs_bandwidth {
113 #ifdef CONFIG_CFS_BANDWIDTH
117 s64 hierarchal_quota;
120 int idle, timer_active;
121 struct hrtimer period_timer, slack_timer;
122 struct list_head throttled_cfs_rq;
125 int nr_periods, nr_throttled;
130 /* task group related information */
132 struct cgroup_subsys_state css;
134 #ifdef CONFIG_FAIR_GROUP_SCHED
135 /* schedulable entities of this group on each cpu */
136 struct sched_entity **se;
137 /* runqueue "owned" by this group on each cpu */
138 struct cfs_rq **cfs_rq;
139 unsigned long shares;
141 atomic_t load_weight;
143 atomic_t runnable_avg;
146 #ifdef CONFIG_RT_GROUP_SCHED
147 struct sched_rt_entity **rt_se;
148 struct rt_rq **rt_rq;
150 struct rt_bandwidth rt_bandwidth;
154 struct list_head list;
156 struct task_group *parent;
157 struct list_head siblings;
158 struct list_head children;
160 #ifdef CONFIG_SCHED_AUTOGROUP
161 struct autogroup *autogroup;
164 struct cfs_bandwidth cfs_bandwidth;
167 #ifdef CONFIG_FAIR_GROUP_SCHED
168 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
171 * A weight of 0 or 1 can cause arithmetics problems.
172 * A weight of a cfs_rq is the sum of weights of which entities
173 * are queued on this cfs_rq, so a weight of a entity should not be
174 * too large, so as the shares value of a task group.
175 * (The default weight is 1024 - so there's no practical
176 * limitation from this.)
178 #define MIN_SHARES (1UL << 1)
179 #define MAX_SHARES (1UL << 18)
182 /* Default task group.
183 * Every task in system belong to this group at bootup.
185 extern struct task_group root_task_group;
187 typedef int (*tg_visitor)(struct task_group *, void *);
189 extern int walk_tg_tree_from(struct task_group *from,
190 tg_visitor down, tg_visitor up, void *data);
193 * Iterate the full tree, calling @down when first entering a node and @up when
194 * leaving it for the final time.
196 * Caller must hold rcu_lock or sufficient equivalent.
198 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
200 return walk_tg_tree_from(&root_task_group, down, up, data);
203 extern int tg_nop(struct task_group *tg, void *data);
205 extern void free_fair_sched_group(struct task_group *tg);
206 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
207 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
208 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
209 struct sched_entity *se, int cpu,
210 struct sched_entity *parent);
211 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
212 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
214 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
215 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
216 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
218 extern void free_rt_sched_group(struct task_group *tg);
219 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
220 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
221 struct sched_rt_entity *rt_se, int cpu,
222 struct sched_rt_entity *parent);
224 #else /* CONFIG_CGROUP_SCHED */
226 struct cfs_bandwidth { };
228 #endif /* CONFIG_CGROUP_SCHED */
230 /* CFS-related fields in a runqueue */
232 struct load_weight load;
233 unsigned int nr_running, h_nr_running;
238 u64 min_vruntime_copy;
241 struct rb_root tasks_timeline;
242 struct rb_node *rb_leftmost;
245 * 'curr' points to currently running entity on this cfs_rq.
246 * It is set to NULL otherwise (i.e when none are currently running).
248 struct sched_entity *curr, *next, *last, *skip;
250 #ifdef CONFIG_SCHED_DEBUG
251 unsigned int nr_spread_over;
256 * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
257 * removed when useful for applications beyond shares distribution (e.g.
260 #ifdef CONFIG_FAIR_GROUP_SCHED
263 * Under CFS, load is tracked on a per-entity basis and aggregated up.
264 * This allows for the description of both thread and group usage (in
265 * the FAIR_GROUP_SCHED case).
267 u64 runnable_load_avg, blocked_load_avg;
268 atomic64_t decay_counter, removed_load;
270 #endif /* CONFIG_FAIR_GROUP_SCHED */
271 /* These always depend on CONFIG_FAIR_GROUP_SCHED */
272 #ifdef CONFIG_FAIR_GROUP_SCHED
273 u32 tg_runnable_contrib;
275 #endif /* CONFIG_FAIR_GROUP_SCHED */
278 * h_load = weight * f(tg)
280 * Where f(tg) is the recursive weight fraction assigned to
283 unsigned long h_load;
284 #endif /* CONFIG_SMP */
286 #ifdef CONFIG_FAIR_GROUP_SCHED
287 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
290 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
291 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
292 * (like users, containers etc.)
294 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
295 * list is used during load balance.
298 struct list_head leaf_cfs_rq_list;
299 struct task_group *tg; /* group that "owns" this runqueue */
301 #ifdef CONFIG_CFS_BANDWIDTH
304 s64 runtime_remaining;
306 u64 throttled_clock, throttled_clock_task;
307 u64 throttled_clock_task_time;
308 int throttled, throttle_count;
309 struct list_head throttled_list;
310 #endif /* CONFIG_CFS_BANDWIDTH */
311 #endif /* CONFIG_FAIR_GROUP_SCHED */
314 static inline int rt_bandwidth_enabled(void)
316 return sysctl_sched_rt_runtime >= 0;
319 /* Real-Time classes' related field in a runqueue: */
321 struct rt_prio_array active;
322 unsigned int rt_nr_running;
323 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
325 int curr; /* highest queued rt task prio */
327 int next; /* next highest */
332 unsigned long rt_nr_migratory;
333 unsigned long rt_nr_total;
335 struct plist_head pushable_tasks;
340 /* Nests inside the rq lock: */
341 raw_spinlock_t rt_runtime_lock;
343 #ifdef CONFIG_RT_GROUP_SCHED
344 unsigned long rt_nr_boosted;
347 struct list_head leaf_rt_rq_list;
348 struct task_group *tg;
355 * We add the notion of a root-domain which will be used to define per-domain
356 * variables. Each exclusive cpuset essentially defines an island domain by
357 * fully partitioning the member cpus from any other cpuset. Whenever a new
358 * exclusive cpuset is created, we also create and attach a new root-domain
367 cpumask_var_t online;
370 * The "RT overload" flag: it gets set if a CPU has more than
371 * one runnable RT task.
373 cpumask_var_t rto_mask;
374 struct cpupri cpupri;
377 extern struct root_domain def_root_domain;
379 #endif /* CONFIG_SMP */
382 * This is the main, per-CPU runqueue data structure.
384 * Locking rule: those places that want to lock multiple runqueues
385 * (such as the load balancing or the thread migration code), lock
386 * acquire operations must be ordered by ascending &runqueue.
393 * nr_running and cpu_load should be in the same cacheline because
394 * remote CPUs use both these fields when doing load calculation.
396 unsigned int nr_running;
397 #define CPU_LOAD_IDX_MAX 5
398 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
399 unsigned long last_load_update_tick;
402 unsigned long nohz_flags;
404 int skip_clock_update;
406 /* capture load from *all* tasks on this cpu: */
407 struct load_weight load;
408 unsigned long nr_load_updates;
414 #ifdef CONFIG_FAIR_GROUP_SCHED
415 /* list of leaf cfs_rq on this cpu: */
416 struct list_head leaf_cfs_rq_list;
418 unsigned long h_load_throttle;
419 #endif /* CONFIG_SMP */
420 #endif /* CONFIG_FAIR_GROUP_SCHED */
422 #ifdef CONFIG_RT_GROUP_SCHED
423 struct list_head leaf_rt_rq_list;
427 * This is part of a global counter where only the total sum
428 * over all CPUs matters. A task can increase this counter on
429 * one CPU and if it got migrated afterwards it may decrease
430 * it on another CPU. Always updated under the runqueue lock:
432 unsigned long nr_uninterruptible;
434 struct task_struct *curr, *idle, *stop;
435 unsigned long next_balance;
436 struct mm_struct *prev_mm;
444 struct root_domain *rd;
445 struct sched_domain *sd;
447 unsigned long cpu_power;
449 unsigned char idle_balance;
450 /* For active balancing */
454 struct cpu_stop_work active_balance_work;
455 /* cpu of this runqueue: */
459 struct list_head cfs_tasks;
467 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
470 #ifdef CONFIG_PARAVIRT
473 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
474 u64 prev_steal_time_rq;
477 /* calc_load related fields */
478 unsigned long calc_load_update;
479 long calc_load_active;
481 #ifdef CONFIG_SCHED_HRTICK
483 int hrtick_csd_pending;
484 struct call_single_data hrtick_csd;
486 struct hrtimer hrtick_timer;
489 #ifdef CONFIG_SCHEDSTATS
491 struct sched_info rq_sched_info;
492 unsigned long long rq_cpu_time;
493 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
495 /* sys_sched_yield() stats */
496 unsigned int yld_count;
498 /* schedule() stats */
499 unsigned int sched_count;
500 unsigned int sched_goidle;
502 /* try_to_wake_up() stats */
503 unsigned int ttwu_count;
504 unsigned int ttwu_local;
508 struct llist_head wake_list;
511 struct sched_avg avg;
514 static inline int cpu_of(struct rq *rq)
523 DECLARE_PER_CPU(struct rq, runqueues);
525 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
526 #define this_rq() (&__get_cpu_var(runqueues))
527 #define task_rq(p) cpu_rq(task_cpu(p))
528 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
529 #define raw_rq() (&__raw_get_cpu_var(runqueues))
533 #define rcu_dereference_check_sched_domain(p) \
534 rcu_dereference_check((p), \
535 lockdep_is_held(&sched_domains_mutex))
538 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
539 * See detach_destroy_domains: synchronize_sched for details.
541 * The domain tree of any CPU may only be accessed from within
542 * preempt-disabled sections.
544 #define for_each_domain(cpu, __sd) \
545 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
546 __sd; __sd = __sd->parent)
548 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
551 * highest_flag_domain - Return highest sched_domain containing flag.
552 * @cpu: The cpu whose highest level of sched domain is to
554 * @flag: The flag to check for the highest sched_domain
557 * Returns the highest sched_domain of a cpu which contains the given flag.
559 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
561 struct sched_domain *sd, *hsd = NULL;
563 for_each_domain(cpu, sd) {
564 if (!(sd->flags & flag))
572 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
573 DECLARE_PER_CPU(int, sd_llc_id);
575 struct sched_group_power {
578 * CPU power of this group, SCHED_LOAD_SCALE being max power for a
581 unsigned int power, power_orig;
582 unsigned long next_update;
584 * Number of busy cpus in this group.
586 atomic_t nr_busy_cpus;
588 unsigned long cpumask[0]; /* iteration mask */
592 struct sched_group *next; /* Must be a circular list */
595 unsigned int group_weight;
596 struct sched_group_power *sgp;
599 * The CPUs this group covers.
601 * NOTE: this field is variable length. (Allocated dynamically
602 * by attaching extra space to the end of the structure,
603 * depending on how many CPUs the kernel has booted up with)
605 unsigned long cpumask[0];
608 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
610 return to_cpumask(sg->cpumask);
614 * cpumask masking which cpus in the group are allowed to iterate up the domain
617 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
619 return to_cpumask(sg->sgp->cpumask);
623 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
624 * @group: The group whose first cpu is to be returned.
626 static inline unsigned int group_first_cpu(struct sched_group *group)
628 return cpumask_first(sched_group_cpus(group));
631 extern int group_balance_cpu(struct sched_group *sg);
633 #endif /* CONFIG_SMP */
636 #include "auto_group.h"
638 #ifdef CONFIG_CGROUP_SCHED
641 * Return the group to which this tasks belongs.
643 * We cannot use task_subsys_state() and friends because the cgroup
644 * subsystem changes that value before the cgroup_subsys::attach() method
645 * is called, therefore we cannot pin it and might observe the wrong value.
647 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
648 * core changes this before calling sched_move_task().
650 * Instead we use a 'copy' which is updated from sched_move_task() while
651 * holding both task_struct::pi_lock and rq::lock.
653 static inline struct task_group *task_group(struct task_struct *p)
655 return p->sched_task_group;
658 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
659 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
661 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
662 struct task_group *tg = task_group(p);
665 #ifdef CONFIG_FAIR_GROUP_SCHED
666 p->se.cfs_rq = tg->cfs_rq[cpu];
667 p->se.parent = tg->se[cpu];
670 #ifdef CONFIG_RT_GROUP_SCHED
671 p->rt.rt_rq = tg->rt_rq[cpu];
672 p->rt.parent = tg->rt_se[cpu];
676 #else /* CONFIG_CGROUP_SCHED */
678 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
679 static inline struct task_group *task_group(struct task_struct *p)
684 #endif /* CONFIG_CGROUP_SCHED */
686 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
691 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
692 * successfuly executed on another CPU. We must ensure that updates of
693 * per-task data have been completed by this moment.
696 task_thread_info(p)->cpu = cpu;
701 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
703 #ifdef CONFIG_SCHED_DEBUG
704 # include <linux/static_key.h>
705 # define const_debug __read_mostly
707 # define const_debug const
710 extern const_debug unsigned int sysctl_sched_features;
712 #define SCHED_FEAT(name, enabled) \
713 __SCHED_FEAT_##name ,
716 #include "features.h"
722 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
723 static __always_inline bool static_branch__true(struct static_key *key)
725 return static_key_true(key); /* Not out of line branch. */
728 static __always_inline bool static_branch__false(struct static_key *key)
730 return static_key_false(key); /* Out of line branch. */
733 #define SCHED_FEAT(name, enabled) \
734 static __always_inline bool static_branch_##name(struct static_key *key) \
736 return static_branch__##enabled(key); \
739 #include "features.h"
743 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
744 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
745 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
746 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
747 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
749 #ifdef CONFIG_NUMA_BALANCING
750 #define sched_feat_numa(x) sched_feat(x)
751 #ifdef CONFIG_SCHED_DEBUG
752 #define numabalancing_enabled sched_feat_numa(NUMA)
754 extern bool numabalancing_enabled;
755 #endif /* CONFIG_SCHED_DEBUG */
757 #define sched_feat_numa(x) (0)
758 #define numabalancing_enabled (0)
759 #endif /* CONFIG_NUMA_BALANCING */
761 static inline u64 global_rt_period(void)
763 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
766 static inline u64 global_rt_runtime(void)
768 if (sysctl_sched_rt_runtime < 0)
771 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
776 static inline int task_current(struct rq *rq, struct task_struct *p)
778 return rq->curr == p;
781 static inline int task_running(struct rq *rq, struct task_struct *p)
786 return task_current(rq, p);
791 #ifndef prepare_arch_switch
792 # define prepare_arch_switch(next) do { } while (0)
794 #ifndef finish_arch_switch
795 # define finish_arch_switch(prev) do { } while (0)
797 #ifndef finish_arch_post_lock_switch
798 # define finish_arch_post_lock_switch() do { } while (0)
801 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
802 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
806 * We can optimise this out completely for !SMP, because the
807 * SMP rebalancing from interrupt is the only thing that cares
814 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
818 * After ->on_cpu is cleared, the task can be moved to a different CPU.
819 * We must ensure this doesn't happen until the switch is completely
825 #ifdef CONFIG_DEBUG_SPINLOCK
826 /* this is a valid case when another task releases the spinlock */
827 rq->lock.owner = current;
830 * If we are tracking spinlock dependencies then we have to
831 * fix up the runqueue lock - which gets 'carried over' from
834 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
836 raw_spin_unlock_irq(&rq->lock);
839 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
840 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
844 * We can optimise this out completely for !SMP, because the
845 * SMP rebalancing from interrupt is the only thing that cares
850 raw_spin_unlock(&rq->lock);
853 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
857 * After ->on_cpu is cleared, the task can be moved to a different CPU.
858 * We must ensure this doesn't happen until the switch is completely
866 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
871 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
872 #define WF_FORK 0x02 /* child wakeup after fork */
873 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
875 static inline void update_load_add(struct load_weight *lw, unsigned long inc)
881 static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
887 static inline void update_load_set(struct load_weight *lw, unsigned long w)
894 * To aid in avoiding the subversion of "niceness" due to uneven distribution
895 * of tasks with abnormal "nice" values across CPUs the contribution that
896 * each task makes to its run queue's load is weighted according to its
897 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
898 * scaled version of the new time slice allocation that they receive on time
902 #define WEIGHT_IDLEPRIO 3
903 #define WMULT_IDLEPRIO 1431655765
906 * Nice levels are multiplicative, with a gentle 10% change for every
907 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
908 * nice 1, it will get ~10% less CPU time than another CPU-bound task
909 * that remained on nice 0.
911 * The "10% effect" is relative and cumulative: from _any_ nice level,
912 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
913 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
914 * If a task goes up by ~10% and another task goes down by ~10% then
915 * the relative distance between them is ~25%.)
917 static const int prio_to_weight[40] = {
918 /* -20 */ 88761, 71755, 56483, 46273, 36291,
919 /* -15 */ 29154, 23254, 18705, 14949, 11916,
920 /* -10 */ 9548, 7620, 6100, 4904, 3906,
921 /* -5 */ 3121, 2501, 1991, 1586, 1277,
922 /* 0 */ 1024, 820, 655, 526, 423,
923 /* 5 */ 335, 272, 215, 172, 137,
924 /* 10 */ 110, 87, 70, 56, 45,
925 /* 15 */ 36, 29, 23, 18, 15,
929 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
931 * In cases where the weight does not change often, we can use the
932 * precalculated inverse to speed up arithmetics by turning divisions
933 * into multiplications:
935 static const u32 prio_to_wmult[40] = {
936 /* -20 */ 48388, 59856, 76040, 92818, 118348,
937 /* -15 */ 147320, 184698, 229616, 287308, 360437,
938 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
939 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
940 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
941 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
942 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
943 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
946 /* Time spent by the tasks of the cpu accounting group executing in ... */
947 enum cpuacct_stat_index {
948 CPUACCT_STAT_USER, /* ... user mode */
949 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
955 #define sched_class_highest (&stop_sched_class)
956 #define for_each_class(class) \
957 for (class = sched_class_highest; class; class = class->next)
959 extern const struct sched_class stop_sched_class;
960 extern const struct sched_class rt_sched_class;
961 extern const struct sched_class fair_sched_class;
962 extern const struct sched_class idle_sched_class;
967 extern void trigger_load_balance(struct rq *rq, int cpu);
968 extern void idle_balance(int this_cpu, struct rq *this_rq);
970 #else /* CONFIG_SMP */
972 static inline void idle_balance(int cpu, struct rq *rq)
978 extern void sysrq_sched_debug_show(void);
979 extern void sched_init_granularity(void);
980 extern void update_max_interval(void);
981 extern void update_group_power(struct sched_domain *sd, int cpu);
982 extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu);
983 extern void init_sched_rt_class(void);
984 extern void init_sched_fair_class(void);
986 extern void resched_task(struct task_struct *p);
987 extern void resched_cpu(int cpu);
989 extern struct rt_bandwidth def_rt_bandwidth;
990 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
992 extern void update_idle_cpu_load(struct rq *this_rq);
994 #ifdef CONFIG_CGROUP_CPUACCT
995 #include <linux/cgroup.h>
996 /* track cpu usage of a group of tasks and its child groups */
998 struct cgroup_subsys_state css;
999 /* cpuusage holds pointer to a u64-type object on every cpu */
1000 u64 __percpu *cpuusage;
1001 struct kernel_cpustat __percpu *cpustat;
1004 extern struct cgroup_subsys cpuacct_subsys;
1005 extern struct cpuacct root_cpuacct;
1007 /* return cpu accounting group corresponding to this container */
1008 static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
1010 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
1011 struct cpuacct, css);
1014 /* return cpu accounting group to which this task belongs */
1015 static inline struct cpuacct *task_ca(struct task_struct *tsk)
1017 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
1018 struct cpuacct, css);
1021 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
1023 if (!ca || !ca->css.cgroup->parent)
1025 return cgroup_ca(ca->css.cgroup->parent);
1028 extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
1030 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
1033 #ifdef CONFIG_PARAVIRT
1034 static inline u64 steal_ticks(u64 steal)
1036 if (unlikely(steal > NSEC_PER_SEC))
1037 return div_u64(steal, TICK_NSEC);
1039 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
1043 static inline void inc_nr_running(struct rq *rq)
1048 static inline void dec_nr_running(struct rq *rq)
1053 extern void update_rq_clock(struct rq *rq);
1055 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1056 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1058 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1060 extern const_debug unsigned int sysctl_sched_time_avg;
1061 extern const_debug unsigned int sysctl_sched_nr_migrate;
1062 extern const_debug unsigned int sysctl_sched_migration_cost;
1064 static inline u64 sched_avg_period(void)
1066 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1069 #ifdef CONFIG_SCHED_HRTICK
1073 * - enabled by features
1074 * - hrtimer is actually high res
1076 static inline int hrtick_enabled(struct rq *rq)
1078 if (!sched_feat(HRTICK))
1080 if (!cpu_active(cpu_of(rq)))
1082 return hrtimer_is_hres_active(&rq->hrtick_timer);
1085 void hrtick_start(struct rq *rq, u64 delay);
1089 static inline int hrtick_enabled(struct rq *rq)
1094 #endif /* CONFIG_SCHED_HRTICK */
1097 extern void sched_avg_update(struct rq *rq);
1098 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1100 rq->rt_avg += rt_delta;
1101 sched_avg_update(rq);
1104 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1105 static inline void sched_avg_update(struct rq *rq) { }
1108 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1111 #ifdef CONFIG_PREEMPT
1113 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1116 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1117 * way at the expense of forcing extra atomic operations in all
1118 * invocations. This assures that the double_lock is acquired using the
1119 * same underlying policy as the spinlock_t on this architecture, which
1120 * reduces latency compared to the unfair variant below. However, it
1121 * also adds more overhead and therefore may reduce throughput.
1123 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1124 __releases(this_rq->lock)
1125 __acquires(busiest->lock)
1126 __acquires(this_rq->lock)
1128 raw_spin_unlock(&this_rq->lock);
1129 double_rq_lock(this_rq, busiest);
1136 * Unfair double_lock_balance: Optimizes throughput at the expense of
1137 * latency by eliminating extra atomic operations when the locks are
1138 * already in proper order on entry. This favors lower cpu-ids and will
1139 * grant the double lock to lower cpus over higher ids under contention,
1140 * regardless of entry order into the function.
1142 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1143 __releases(this_rq->lock)
1144 __acquires(busiest->lock)
1145 __acquires(this_rq->lock)
1149 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1150 if (busiest < this_rq) {
1151 raw_spin_unlock(&this_rq->lock);
1152 raw_spin_lock(&busiest->lock);
1153 raw_spin_lock_nested(&this_rq->lock,
1154 SINGLE_DEPTH_NESTING);
1157 raw_spin_lock_nested(&busiest->lock,
1158 SINGLE_DEPTH_NESTING);
1163 #endif /* CONFIG_PREEMPT */
1166 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1168 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1170 if (unlikely(!irqs_disabled())) {
1171 /* printk() doesn't work good under rq->lock */
1172 raw_spin_unlock(&this_rq->lock);
1176 return _double_lock_balance(this_rq, busiest);
1179 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1180 __releases(busiest->lock)
1182 raw_spin_unlock(&busiest->lock);
1183 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1187 * double_rq_lock - safely lock two runqueues
1189 * Note this does not disable interrupts like task_rq_lock,
1190 * you need to do so manually before calling.
1192 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1193 __acquires(rq1->lock)
1194 __acquires(rq2->lock)
1196 BUG_ON(!irqs_disabled());
1198 raw_spin_lock(&rq1->lock);
1199 __acquire(rq2->lock); /* Fake it out ;) */
1202 raw_spin_lock(&rq1->lock);
1203 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1205 raw_spin_lock(&rq2->lock);
1206 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1212 * double_rq_unlock - safely unlock two runqueues
1214 * Note this does not restore interrupts like task_rq_unlock,
1215 * you need to do so manually after calling.
1217 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1218 __releases(rq1->lock)
1219 __releases(rq2->lock)
1221 raw_spin_unlock(&rq1->lock);
1223 raw_spin_unlock(&rq2->lock);
1225 __release(rq2->lock);
1228 #else /* CONFIG_SMP */
1231 * double_rq_lock - safely lock two runqueues
1233 * Note this does not disable interrupts like task_rq_lock,
1234 * you need to do so manually before calling.
1236 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1237 __acquires(rq1->lock)
1238 __acquires(rq2->lock)
1240 BUG_ON(!irqs_disabled());
1242 raw_spin_lock(&rq1->lock);
1243 __acquire(rq2->lock); /* Fake it out ;) */
1247 * double_rq_unlock - safely unlock two runqueues
1249 * Note this does not restore interrupts like task_rq_unlock,
1250 * you need to do so manually after calling.
1252 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1253 __releases(rq1->lock)
1254 __releases(rq2->lock)
1257 raw_spin_unlock(&rq1->lock);
1258 __release(rq2->lock);
1263 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1264 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1265 extern void print_cfs_stats(struct seq_file *m, int cpu);
1266 extern void print_rt_stats(struct seq_file *m, int cpu);
1268 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1269 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1271 extern void account_cfs_bandwidth_used(int enabled, int was_enabled);
1274 enum rq_nohz_flag_bits {
1280 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1283 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1285 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1286 DECLARE_PER_CPU(u64, cpu_softirq_time);
1288 #ifndef CONFIG_64BIT
1289 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1291 static inline void irq_time_write_begin(void)
1293 __this_cpu_inc(irq_time_seq.sequence);
1297 static inline void irq_time_write_end(void)
1300 __this_cpu_inc(irq_time_seq.sequence);
1303 static inline u64 irq_time_read(int cpu)
1309 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1310 irq_time = per_cpu(cpu_softirq_time, cpu) +
1311 per_cpu(cpu_hardirq_time, cpu);
1312 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1316 #else /* CONFIG_64BIT */
1317 static inline void irq_time_write_begin(void)
1321 static inline void irq_time_write_end(void)
1325 static inline u64 irq_time_read(int cpu)
1327 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1329 #endif /* CONFIG_64BIT */
1330 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */