2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/sched/deadline.h>
6 #include <linux/mutex.h>
7 #include <linux/spinlock.h>
8 #include <linux/stop_machine.h>
9 #include <linux/tick.h>
10 #include <linux/slab.h>
13 #include "cpudeadline.h"
18 extern __read_mostly int scheduler_running;
20 extern unsigned long calc_load_update;
21 extern atomic_long_t calc_load_tasks;
23 extern long calc_load_fold_active(struct rq *this_rq);
24 extern void update_cpu_load_active(struct rq *this_rq);
27 * Helpers for converting nanosecond timing to jiffy resolution
29 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
32 * Increase resolution of nice-level calculations for 64-bit architectures.
33 * The extra resolution improves shares distribution and load balancing of
34 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
35 * hierarchies, especially on larger systems. This is not a user-visible change
36 * and does not change the user-interface for setting shares/weights.
38 * We increase resolution only if we have enough bits to allow this increased
39 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
40 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
43 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
44 # define SCHED_LOAD_RESOLUTION 10
45 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
46 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
48 # define SCHED_LOAD_RESOLUTION 0
49 # define scale_load(w) (w)
50 # define scale_load_down(w) (w)
53 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
54 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
56 #define NICE_0_LOAD SCHED_LOAD_SCALE
57 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
60 * Single value that decides SCHED_DEADLINE internal math precision.
61 * 10 -> just above 1us
62 * 9 -> just above 0.5us
67 * These are the 'tuning knobs' of the scheduler:
71 * single value that denotes runtime == period, ie unlimited time.
73 #define RUNTIME_INF ((u64)~0ULL)
75 static inline int fair_policy(int policy)
77 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
80 static inline int rt_policy(int policy)
82 return policy == SCHED_FIFO || policy == SCHED_RR;
85 static inline int dl_policy(int policy)
87 return policy == SCHED_DEADLINE;
90 static inline int task_has_rt_policy(struct task_struct *p)
92 return rt_policy(p->policy);
95 static inline int task_has_dl_policy(struct task_struct *p)
97 return dl_policy(p->policy);
100 static inline bool dl_time_before(u64 a, u64 b)
102 return (s64)(a - b) < 0;
106 * Tells if entity @a should preempt entity @b.
109 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
111 return dl_time_before(a->deadline, b->deadline);
115 * This is the priority-queue data structure of the RT scheduling class:
117 struct rt_prio_array {
118 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
119 struct list_head queue[MAX_RT_PRIO];
122 struct rt_bandwidth {
123 /* nests inside the rq lock: */
124 raw_spinlock_t rt_runtime_lock;
127 struct hrtimer rt_period_timer;
130 * To keep the bandwidth of -deadline tasks and groups under control
131 * we need some place where:
132 * - store the maximum -deadline bandwidth of the system (the group);
133 * - cache the fraction of that bandwidth that is currently allocated.
135 * This is all done in the data structure below. It is similar to the
136 * one used for RT-throttling (rt_bandwidth), with the main difference
137 * that, since here we are only interested in admission control, we
138 * do not decrease any runtime while the group "executes", neither we
139 * need a timer to replenish it.
141 * With respect to SMP, the bandwidth is given on a per-CPU basis,
143 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
144 * - dl_total_bw array contains, in the i-eth element, the currently
145 * allocated bandwidth on the i-eth CPU.
146 * Moreover, groups consume bandwidth on each CPU, while tasks only
147 * consume bandwidth on the CPU they're running on.
148 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
149 * that will be shown the next time the proc or cgroup controls will
150 * be red. It on its turn can be changed by writing on its own
153 struct dl_bandwidth {
154 raw_spinlock_t dl_runtime_lock;
159 static inline int dl_bandwidth_enabled(void)
161 return sysctl_sched_rt_runtime >= 0;
164 extern struct dl_bw *dl_bw_of(int i);
171 extern struct mutex sched_domains_mutex;
173 #ifdef CONFIG_CGROUP_SCHED
175 #include <linux/cgroup.h>
180 extern struct list_head task_groups;
182 struct cfs_bandwidth {
183 #ifdef CONFIG_CFS_BANDWIDTH
187 s64 hierarchal_quota;
190 int idle, timer_active;
191 struct hrtimer period_timer, slack_timer;
192 struct list_head throttled_cfs_rq;
195 int nr_periods, nr_throttled;
200 /* task group related information */
202 struct cgroup_subsys_state css;
204 #ifdef CONFIG_FAIR_GROUP_SCHED
205 /* schedulable entities of this group on each cpu */
206 struct sched_entity **se;
207 /* runqueue "owned" by this group on each cpu */
208 struct cfs_rq **cfs_rq;
209 unsigned long shares;
212 atomic_long_t load_avg;
213 atomic_t runnable_avg;
217 #ifdef CONFIG_RT_GROUP_SCHED
218 struct sched_rt_entity **rt_se;
219 struct rt_rq **rt_rq;
221 struct rt_bandwidth rt_bandwidth;
225 struct list_head list;
227 struct task_group *parent;
228 struct list_head siblings;
229 struct list_head children;
231 #ifdef CONFIG_SCHED_AUTOGROUP
232 struct autogroup *autogroup;
235 struct cfs_bandwidth cfs_bandwidth;
238 #ifdef CONFIG_FAIR_GROUP_SCHED
239 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
242 * A weight of 0 or 1 can cause arithmetics problems.
243 * A weight of a cfs_rq is the sum of weights of which entities
244 * are queued on this cfs_rq, so a weight of a entity should not be
245 * too large, so as the shares value of a task group.
246 * (The default weight is 1024 - so there's no practical
247 * limitation from this.)
249 #define MIN_SHARES (1UL << 1)
250 #define MAX_SHARES (1UL << 18)
253 typedef int (*tg_visitor)(struct task_group *, void *);
255 extern int walk_tg_tree_from(struct task_group *from,
256 tg_visitor down, tg_visitor up, void *data);
259 * Iterate the full tree, calling @down when first entering a node and @up when
260 * leaving it for the final time.
262 * Caller must hold rcu_lock or sufficient equivalent.
264 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
266 return walk_tg_tree_from(&root_task_group, down, up, data);
269 extern int tg_nop(struct task_group *tg, void *data);
271 extern void free_fair_sched_group(struct task_group *tg);
272 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
273 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
274 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
275 struct sched_entity *se, int cpu,
276 struct sched_entity *parent);
277 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
278 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
280 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
281 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
282 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
284 extern void free_rt_sched_group(struct task_group *tg);
285 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
286 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
287 struct sched_rt_entity *rt_se, int cpu,
288 struct sched_rt_entity *parent);
290 extern struct task_group *sched_create_group(struct task_group *parent);
291 extern void sched_online_group(struct task_group *tg,
292 struct task_group *parent);
293 extern void sched_destroy_group(struct task_group *tg);
294 extern void sched_offline_group(struct task_group *tg);
296 extern void sched_move_task(struct task_struct *tsk);
298 #ifdef CONFIG_FAIR_GROUP_SCHED
299 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
302 #else /* CONFIG_CGROUP_SCHED */
304 struct cfs_bandwidth { };
306 #endif /* CONFIG_CGROUP_SCHED */
308 /* CFS-related fields in a runqueue */
310 struct load_weight load;
311 unsigned int nr_running, h_nr_running;
316 u64 min_vruntime_copy;
319 struct rb_root tasks_timeline;
320 struct rb_node *rb_leftmost;
323 * 'curr' points to currently running entity on this cfs_rq.
324 * It is set to NULL otherwise (i.e when none are currently running).
326 struct sched_entity *curr, *next, *last, *skip;
328 #ifdef CONFIG_SCHED_DEBUG
329 unsigned int nr_spread_over;
335 * Under CFS, load is tracked on a per-entity basis and aggregated up.
336 * This allows for the description of both thread and group usage (in
337 * the FAIR_GROUP_SCHED case).
339 unsigned long runnable_load_avg, blocked_load_avg;
340 atomic64_t decay_counter;
342 atomic_long_t removed_load;
344 #ifdef CONFIG_FAIR_GROUP_SCHED
345 /* Required to track per-cpu representation of a task_group */
346 u32 tg_runnable_contrib;
347 unsigned long tg_load_contrib;
350 * h_load = weight * f(tg)
352 * Where f(tg) is the recursive weight fraction assigned to
355 unsigned long h_load;
356 u64 last_h_load_update;
357 struct sched_entity *h_load_next;
358 #endif /* CONFIG_FAIR_GROUP_SCHED */
359 #endif /* CONFIG_SMP */
361 #ifdef CONFIG_FAIR_GROUP_SCHED
362 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
365 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
366 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
367 * (like users, containers etc.)
369 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
370 * list is used during load balance.
373 struct list_head leaf_cfs_rq_list;
374 struct task_group *tg; /* group that "owns" this runqueue */
376 #ifdef CONFIG_CFS_BANDWIDTH
379 s64 runtime_remaining;
381 u64 throttled_clock, throttled_clock_task;
382 u64 throttled_clock_task_time;
383 int throttled, throttle_count;
384 struct list_head throttled_list;
385 #endif /* CONFIG_CFS_BANDWIDTH */
386 #endif /* CONFIG_FAIR_GROUP_SCHED */
389 static inline int rt_bandwidth_enabled(void)
391 return sysctl_sched_rt_runtime >= 0;
394 /* Real-Time classes' related field in a runqueue: */
396 struct rt_prio_array active;
397 unsigned int rt_nr_running;
398 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
400 int curr; /* highest queued rt task prio */
402 int next; /* next highest */
407 unsigned long rt_nr_migratory;
408 unsigned long rt_nr_total;
410 struct plist_head pushable_tasks;
415 /* Nests inside the rq lock: */
416 raw_spinlock_t rt_runtime_lock;
418 #ifdef CONFIG_RT_GROUP_SCHED
419 unsigned long rt_nr_boosted;
422 struct task_group *tg;
426 /* Deadline class' related fields in a runqueue */
428 /* runqueue is an rbtree, ordered by deadline */
429 struct rb_root rb_root;
430 struct rb_node *rb_leftmost;
432 unsigned long dl_nr_running;
436 * Deadline values of the currently executing and the
437 * earliest ready task on this rq. Caching these facilitates
438 * the decision wether or not a ready but not running task
439 * should migrate somewhere else.
446 unsigned long dl_nr_migratory;
447 unsigned long dl_nr_total;
451 * Tasks on this rq that can be pushed away. They are kept in
452 * an rb-tree, ordered by tasks' deadlines, with caching
453 * of the leftmost (earliest deadline) element.
455 struct rb_root pushable_dl_tasks_root;
456 struct rb_node *pushable_dl_tasks_leftmost;
465 * We add the notion of a root-domain which will be used to define per-domain
466 * variables. Each exclusive cpuset essentially defines an island domain by
467 * fully partitioning the member cpus from any other cpuset. Whenever a new
468 * exclusive cpuset is created, we also create and attach a new root-domain
477 cpumask_var_t online;
480 * The bit corresponding to a CPU gets set here if such CPU has more
481 * than one runnable -deadline task (as it is below for RT tasks).
483 cpumask_var_t dlo_mask;
489 * The "RT overload" flag: it gets set if a CPU has more than
490 * one runnable RT task.
492 cpumask_var_t rto_mask;
493 struct cpupri cpupri;
496 extern struct root_domain def_root_domain;
498 #endif /* CONFIG_SMP */
501 * This is the main, per-CPU runqueue data structure.
503 * Locking rule: those places that want to lock multiple runqueues
504 * (such as the load balancing or the thread migration code), lock
505 * acquire operations must be ordered by ascending &runqueue.
512 * nr_running and cpu_load should be in the same cacheline because
513 * remote CPUs use both these fields when doing load calculation.
515 unsigned int nr_running;
516 #ifdef CONFIG_NUMA_BALANCING
517 unsigned int nr_numa_running;
518 unsigned int nr_preferred_running;
520 #define CPU_LOAD_IDX_MAX 5
521 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
522 unsigned long last_load_update_tick;
523 #ifdef CONFIG_NO_HZ_COMMON
525 unsigned long nohz_flags;
527 #ifdef CONFIG_NO_HZ_FULL
528 unsigned long last_sched_tick;
530 int skip_clock_update;
532 /* capture load from *all* tasks on this cpu: */
533 struct load_weight load;
534 unsigned long nr_load_updates;
541 #ifdef CONFIG_FAIR_GROUP_SCHED
542 /* list of leaf cfs_rq on this cpu: */
543 struct list_head leaf_cfs_rq_list;
544 #endif /* CONFIG_FAIR_GROUP_SCHED */
546 #ifdef CONFIG_RT_GROUP_SCHED
547 struct list_head leaf_rt_rq_list;
551 * This is part of a global counter where only the total sum
552 * over all CPUs matters. A task can increase this counter on
553 * one CPU and if it got migrated afterwards it may decrease
554 * it on another CPU. Always updated under the runqueue lock:
556 unsigned long nr_uninterruptible;
558 struct task_struct *curr, *idle, *stop;
559 unsigned long next_balance;
560 struct mm_struct *prev_mm;
568 struct root_domain *rd;
569 struct sched_domain *sd;
571 unsigned long cpu_power;
573 unsigned char idle_balance;
574 /* For active balancing */
578 struct cpu_stop_work active_balance_work;
579 /* cpu of this runqueue: */
583 struct list_head cfs_tasks;
590 /* This is used to determine avg_idle's max value */
591 u64 max_idle_balance_cost;
594 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
597 #ifdef CONFIG_PARAVIRT
600 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
601 u64 prev_steal_time_rq;
604 /* calc_load related fields */
605 unsigned long calc_load_update;
606 long calc_load_active;
608 #ifdef CONFIG_SCHED_HRTICK
610 int hrtick_csd_pending;
611 struct call_single_data hrtick_csd;
613 struct hrtimer hrtick_timer;
616 #ifdef CONFIG_SCHEDSTATS
618 struct sched_info rq_sched_info;
619 unsigned long long rq_cpu_time;
620 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
622 /* sys_sched_yield() stats */
623 unsigned int yld_count;
625 /* schedule() stats */
626 unsigned int sched_count;
627 unsigned int sched_goidle;
629 /* try_to_wake_up() stats */
630 unsigned int ttwu_count;
631 unsigned int ttwu_local;
635 struct llist_head wake_list;
638 struct sched_avg avg;
641 static inline int cpu_of(struct rq *rq)
650 DECLARE_PER_CPU(struct rq, runqueues);
652 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
653 #define this_rq() (&__get_cpu_var(runqueues))
654 #define task_rq(p) cpu_rq(task_cpu(p))
655 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
656 #define raw_rq() (&__raw_get_cpu_var(runqueues))
658 static inline u64 rq_clock(struct rq *rq)
663 static inline u64 rq_clock_task(struct rq *rq)
665 return rq->clock_task;
668 #ifdef CONFIG_NUMA_BALANCING
669 extern void sched_setnuma(struct task_struct *p, int node);
670 extern int migrate_task_to(struct task_struct *p, int cpu);
671 extern int migrate_swap(struct task_struct *, struct task_struct *);
672 #endif /* CONFIG_NUMA_BALANCING */
676 #define rcu_dereference_check_sched_domain(p) \
677 rcu_dereference_check((p), \
678 lockdep_is_held(&sched_domains_mutex))
681 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
682 * See detach_destroy_domains: synchronize_sched for details.
684 * The domain tree of any CPU may only be accessed from within
685 * preempt-disabled sections.
687 #define for_each_domain(cpu, __sd) \
688 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
689 __sd; __sd = __sd->parent)
691 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
694 * highest_flag_domain - Return highest sched_domain containing flag.
695 * @cpu: The cpu whose highest level of sched domain is to
697 * @flag: The flag to check for the highest sched_domain
700 * Returns the highest sched_domain of a cpu which contains the given flag.
702 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
704 struct sched_domain *sd, *hsd = NULL;
706 for_each_domain(cpu, sd) {
707 if (!(sd->flags & flag))
715 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
717 struct sched_domain *sd;
719 for_each_domain(cpu, sd) {
720 if (sd->flags & flag)
727 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
728 DECLARE_PER_CPU(int, sd_llc_size);
729 DECLARE_PER_CPU(int, sd_llc_id);
730 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
731 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
732 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
734 struct sched_group_power {
737 * CPU power of this group, SCHED_LOAD_SCALE being max power for a
740 unsigned int power, power_orig;
741 unsigned long next_update;
742 int imbalance; /* XXX unrelated to power but shared group state */
744 * Number of busy cpus in this group.
746 atomic_t nr_busy_cpus;
748 unsigned long cpumask[0]; /* iteration mask */
752 struct sched_group *next; /* Must be a circular list */
755 unsigned int group_weight;
756 struct sched_group_power *sgp;
759 * The CPUs this group covers.
761 * NOTE: this field is variable length. (Allocated dynamically
762 * by attaching extra space to the end of the structure,
763 * depending on how many CPUs the kernel has booted up with)
765 unsigned long cpumask[0];
768 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
770 return to_cpumask(sg->cpumask);
774 * cpumask masking which cpus in the group are allowed to iterate up the domain
777 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
779 return to_cpumask(sg->sgp->cpumask);
783 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
784 * @group: The group whose first cpu is to be returned.
786 static inline unsigned int group_first_cpu(struct sched_group *group)
788 return cpumask_first(sched_group_cpus(group));
791 extern int group_balance_cpu(struct sched_group *sg);
793 #endif /* CONFIG_SMP */
796 #include "auto_group.h"
798 #ifdef CONFIG_CGROUP_SCHED
801 * Return the group to which this tasks belongs.
803 * We cannot use task_css() and friends because the cgroup subsystem
804 * changes that value before the cgroup_subsys::attach() method is called,
805 * therefore we cannot pin it and might observe the wrong value.
807 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
808 * core changes this before calling sched_move_task().
810 * Instead we use a 'copy' which is updated from sched_move_task() while
811 * holding both task_struct::pi_lock and rq::lock.
813 static inline struct task_group *task_group(struct task_struct *p)
815 return p->sched_task_group;
818 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
819 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
821 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
822 struct task_group *tg = task_group(p);
825 #ifdef CONFIG_FAIR_GROUP_SCHED
826 p->se.cfs_rq = tg->cfs_rq[cpu];
827 p->se.parent = tg->se[cpu];
830 #ifdef CONFIG_RT_GROUP_SCHED
831 p->rt.rt_rq = tg->rt_rq[cpu];
832 p->rt.parent = tg->rt_se[cpu];
836 #else /* CONFIG_CGROUP_SCHED */
838 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
839 static inline struct task_group *task_group(struct task_struct *p)
844 #endif /* CONFIG_CGROUP_SCHED */
846 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
851 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
852 * successfuly executed on another CPU. We must ensure that updates of
853 * per-task data have been completed by this moment.
856 task_thread_info(p)->cpu = cpu;
862 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
864 #ifdef CONFIG_SCHED_DEBUG
865 # include <linux/static_key.h>
866 # define const_debug __read_mostly
868 # define const_debug const
871 extern const_debug unsigned int sysctl_sched_features;
873 #define SCHED_FEAT(name, enabled) \
874 __SCHED_FEAT_##name ,
877 #include "features.h"
883 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
884 static __always_inline bool static_branch__true(struct static_key *key)
886 return static_key_true(key); /* Not out of line branch. */
889 static __always_inline bool static_branch__false(struct static_key *key)
891 return static_key_false(key); /* Out of line branch. */
894 #define SCHED_FEAT(name, enabled) \
895 static __always_inline bool static_branch_##name(struct static_key *key) \
897 return static_branch__##enabled(key); \
900 #include "features.h"
904 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
905 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
906 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
907 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
908 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
910 #ifdef CONFIG_NUMA_BALANCING
911 #define sched_feat_numa(x) sched_feat(x)
912 #ifdef CONFIG_SCHED_DEBUG
913 #define numabalancing_enabled sched_feat_numa(NUMA)
915 extern bool numabalancing_enabled;
916 #endif /* CONFIG_SCHED_DEBUG */
918 #define sched_feat_numa(x) (0)
919 #define numabalancing_enabled (0)
920 #endif /* CONFIG_NUMA_BALANCING */
922 static inline u64 global_rt_period(void)
924 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
927 static inline u64 global_rt_runtime(void)
929 if (sysctl_sched_rt_runtime < 0)
932 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
935 static inline int task_current(struct rq *rq, struct task_struct *p)
937 return rq->curr == p;
940 static inline int task_running(struct rq *rq, struct task_struct *p)
945 return task_current(rq, p);
950 #ifndef prepare_arch_switch
951 # define prepare_arch_switch(next) do { } while (0)
953 #ifndef finish_arch_switch
954 # define finish_arch_switch(prev) do { } while (0)
956 #ifndef finish_arch_post_lock_switch
957 # define finish_arch_post_lock_switch() do { } while (0)
960 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
961 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
965 * We can optimise this out completely for !SMP, because the
966 * SMP rebalancing from interrupt is the only thing that cares
973 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
977 * After ->on_cpu is cleared, the task can be moved to a different CPU.
978 * We must ensure this doesn't happen until the switch is completely
984 #ifdef CONFIG_DEBUG_SPINLOCK
985 /* this is a valid case when another task releases the spinlock */
986 rq->lock.owner = current;
989 * If we are tracking spinlock dependencies then we have to
990 * fix up the runqueue lock - which gets 'carried over' from
993 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
995 raw_spin_unlock_irq(&rq->lock);
998 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
999 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1003 * We can optimise this out completely for !SMP, because the
1004 * SMP rebalancing from interrupt is the only thing that cares
1009 raw_spin_unlock(&rq->lock);
1012 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1016 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1017 * We must ensure this doesn't happen until the switch is completely
1025 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
1030 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1031 #define WF_FORK 0x02 /* child wakeup after fork */
1032 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1035 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1036 * of tasks with abnormal "nice" values across CPUs the contribution that
1037 * each task makes to its run queue's load is weighted according to its
1038 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1039 * scaled version of the new time slice allocation that they receive on time
1043 #define WEIGHT_IDLEPRIO 3
1044 #define WMULT_IDLEPRIO 1431655765
1047 * Nice levels are multiplicative, with a gentle 10% change for every
1048 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1049 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1050 * that remained on nice 0.
1052 * The "10% effect" is relative and cumulative: from _any_ nice level,
1053 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1054 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1055 * If a task goes up by ~10% and another task goes down by ~10% then
1056 * the relative distance between them is ~25%.)
1058 static const int prio_to_weight[40] = {
1059 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1060 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1061 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1062 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1063 /* 0 */ 1024, 820, 655, 526, 423,
1064 /* 5 */ 335, 272, 215, 172, 137,
1065 /* 10 */ 110, 87, 70, 56, 45,
1066 /* 15 */ 36, 29, 23, 18, 15,
1070 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1072 * In cases where the weight does not change often, we can use the
1073 * precalculated inverse to speed up arithmetics by turning divisions
1074 * into multiplications:
1076 static const u32 prio_to_wmult[40] = {
1077 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1078 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1079 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1080 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1081 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1082 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1083 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1084 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1087 #define ENQUEUE_WAKEUP 1
1088 #define ENQUEUE_HEAD 2
1090 #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */
1092 #define ENQUEUE_WAKING 0
1094 #define ENQUEUE_REPLENISH 8
1096 #define DEQUEUE_SLEEP 1
1098 struct sched_class {
1099 const struct sched_class *next;
1101 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1102 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1103 void (*yield_task) (struct rq *rq);
1104 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1106 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1109 * It is the responsibility of the pick_next_task() method that will
1110 * return the next task to call put_prev_task() on the @prev task or
1111 * something equivalent.
1113 struct task_struct * (*pick_next_task) (struct rq *rq,
1114 struct task_struct *prev);
1115 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1118 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1119 void (*migrate_task_rq)(struct task_struct *p, int next_cpu);
1121 void (*pre_schedule) (struct rq *this_rq, struct task_struct *task);
1122 void (*post_schedule) (struct rq *this_rq);
1123 void (*task_waking) (struct task_struct *task);
1124 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1126 void (*set_cpus_allowed)(struct task_struct *p,
1127 const struct cpumask *newmask);
1129 void (*rq_online)(struct rq *rq);
1130 void (*rq_offline)(struct rq *rq);
1133 void (*set_curr_task) (struct rq *rq);
1134 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1135 void (*task_fork) (struct task_struct *p);
1136 void (*task_dead) (struct task_struct *p);
1138 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1139 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1140 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1143 unsigned int (*get_rr_interval) (struct rq *rq,
1144 struct task_struct *task);
1146 #ifdef CONFIG_FAIR_GROUP_SCHED
1147 void (*task_move_group) (struct task_struct *p, int on_rq);
1151 #define sched_class_highest (&stop_sched_class)
1152 #define for_each_class(class) \
1153 for (class = sched_class_highest; class; class = class->next)
1155 extern const struct sched_class stop_sched_class;
1156 extern const struct sched_class dl_sched_class;
1157 extern const struct sched_class rt_sched_class;
1158 extern const struct sched_class fair_sched_class;
1159 extern const struct sched_class idle_sched_class;
1164 extern void update_group_power(struct sched_domain *sd, int cpu);
1166 extern void trigger_load_balance(struct rq *rq);
1167 extern int idle_balance(struct rq *this_rq);
1169 extern void idle_enter_fair(struct rq *this_rq);
1170 extern void idle_exit_fair(struct rq *this_rq);
1172 #else /* CONFIG_SMP */
1174 static inline void idle_balance(int cpu, struct rq *rq)
1180 extern void sysrq_sched_debug_show(void);
1181 extern void sched_init_granularity(void);
1182 extern void update_max_interval(void);
1184 extern void init_sched_dl_class(void);
1185 extern void init_sched_rt_class(void);
1186 extern void init_sched_fair_class(void);
1187 extern void init_sched_dl_class(void);
1189 extern void resched_task(struct task_struct *p);
1190 extern void resched_cpu(int cpu);
1192 extern struct rt_bandwidth def_rt_bandwidth;
1193 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1195 extern struct dl_bandwidth def_dl_bandwidth;
1196 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1197 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1199 unsigned long to_ratio(u64 period, u64 runtime);
1201 extern void update_idle_cpu_load(struct rq *this_rq);
1203 extern void init_task_runnable_average(struct task_struct *p);
1205 #ifdef CONFIG_PARAVIRT
1206 static inline u64 steal_ticks(u64 steal)
1208 if (unlikely(steal > NSEC_PER_SEC))
1209 return div_u64(steal, TICK_NSEC);
1211 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
1215 static inline void inc_nr_running(struct rq *rq)
1219 #ifdef CONFIG_NO_HZ_FULL
1220 if (rq->nr_running == 2) {
1221 if (tick_nohz_full_cpu(rq->cpu)) {
1222 /* Order rq->nr_running write against the IPI */
1224 smp_send_reschedule(rq->cpu);
1230 static inline void dec_nr_running(struct rq *rq)
1235 static inline void rq_last_tick_reset(struct rq *rq)
1237 #ifdef CONFIG_NO_HZ_FULL
1238 rq->last_sched_tick = jiffies;
1242 extern void update_rq_clock(struct rq *rq);
1244 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1245 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1247 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1249 extern const_debug unsigned int sysctl_sched_time_avg;
1250 extern const_debug unsigned int sysctl_sched_nr_migrate;
1251 extern const_debug unsigned int sysctl_sched_migration_cost;
1253 static inline u64 sched_avg_period(void)
1255 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1258 #ifdef CONFIG_SCHED_HRTICK
1262 * - enabled by features
1263 * - hrtimer is actually high res
1265 static inline int hrtick_enabled(struct rq *rq)
1267 if (!sched_feat(HRTICK))
1269 if (!cpu_active(cpu_of(rq)))
1271 return hrtimer_is_hres_active(&rq->hrtick_timer);
1274 void hrtick_start(struct rq *rq, u64 delay);
1278 static inline int hrtick_enabled(struct rq *rq)
1283 #endif /* CONFIG_SCHED_HRTICK */
1286 extern void sched_avg_update(struct rq *rq);
1287 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1289 rq->rt_avg += rt_delta;
1290 sched_avg_update(rq);
1293 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1294 static inline void sched_avg_update(struct rq *rq) { }
1297 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1300 #ifdef CONFIG_PREEMPT
1302 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1305 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1306 * way at the expense of forcing extra atomic operations in all
1307 * invocations. This assures that the double_lock is acquired using the
1308 * same underlying policy as the spinlock_t on this architecture, which
1309 * reduces latency compared to the unfair variant below. However, it
1310 * also adds more overhead and therefore may reduce throughput.
1312 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1313 __releases(this_rq->lock)
1314 __acquires(busiest->lock)
1315 __acquires(this_rq->lock)
1317 raw_spin_unlock(&this_rq->lock);
1318 double_rq_lock(this_rq, busiest);
1325 * Unfair double_lock_balance: Optimizes throughput at the expense of
1326 * latency by eliminating extra atomic operations when the locks are
1327 * already in proper order on entry. This favors lower cpu-ids and will
1328 * grant the double lock to lower cpus over higher ids under contention,
1329 * regardless of entry order into the function.
1331 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1332 __releases(this_rq->lock)
1333 __acquires(busiest->lock)
1334 __acquires(this_rq->lock)
1338 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1339 if (busiest < this_rq) {
1340 raw_spin_unlock(&this_rq->lock);
1341 raw_spin_lock(&busiest->lock);
1342 raw_spin_lock_nested(&this_rq->lock,
1343 SINGLE_DEPTH_NESTING);
1346 raw_spin_lock_nested(&busiest->lock,
1347 SINGLE_DEPTH_NESTING);
1352 #endif /* CONFIG_PREEMPT */
1355 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1357 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1359 if (unlikely(!irqs_disabled())) {
1360 /* printk() doesn't work good under rq->lock */
1361 raw_spin_unlock(&this_rq->lock);
1365 return _double_lock_balance(this_rq, busiest);
1368 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1369 __releases(busiest->lock)
1371 raw_spin_unlock(&busiest->lock);
1372 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1375 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1381 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1384 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1390 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1394 * double_rq_lock - safely lock two runqueues
1396 * Note this does not disable interrupts like task_rq_lock,
1397 * you need to do so manually before calling.
1399 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1400 __acquires(rq1->lock)
1401 __acquires(rq2->lock)
1403 BUG_ON(!irqs_disabled());
1405 raw_spin_lock(&rq1->lock);
1406 __acquire(rq2->lock); /* Fake it out ;) */
1409 raw_spin_lock(&rq1->lock);
1410 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1412 raw_spin_lock(&rq2->lock);
1413 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1419 * double_rq_unlock - safely unlock two runqueues
1421 * Note this does not restore interrupts like task_rq_unlock,
1422 * you need to do so manually after calling.
1424 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1425 __releases(rq1->lock)
1426 __releases(rq2->lock)
1428 raw_spin_unlock(&rq1->lock);
1430 raw_spin_unlock(&rq2->lock);
1432 __release(rq2->lock);
1435 #else /* CONFIG_SMP */
1438 * double_rq_lock - safely lock two runqueues
1440 * Note this does not disable interrupts like task_rq_lock,
1441 * you need to do so manually before calling.
1443 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1444 __acquires(rq1->lock)
1445 __acquires(rq2->lock)
1447 BUG_ON(!irqs_disabled());
1449 raw_spin_lock(&rq1->lock);
1450 __acquire(rq2->lock); /* Fake it out ;) */
1454 * double_rq_unlock - safely unlock two runqueues
1456 * Note this does not restore interrupts like task_rq_unlock,
1457 * you need to do so manually after calling.
1459 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1460 __releases(rq1->lock)
1461 __releases(rq2->lock)
1464 raw_spin_unlock(&rq1->lock);
1465 __release(rq2->lock);
1470 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1471 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1472 extern void print_cfs_stats(struct seq_file *m, int cpu);
1473 extern void print_rt_stats(struct seq_file *m, int cpu);
1475 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1476 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1477 extern void init_dl_rq(struct dl_rq *dl_rq, struct rq *rq);
1479 extern void cfs_bandwidth_usage_inc(void);
1480 extern void cfs_bandwidth_usage_dec(void);
1482 #ifdef CONFIG_NO_HZ_COMMON
1483 enum rq_nohz_flag_bits {
1488 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1491 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1493 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1494 DECLARE_PER_CPU(u64, cpu_softirq_time);
1496 #ifndef CONFIG_64BIT
1497 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1499 static inline void irq_time_write_begin(void)
1501 __this_cpu_inc(irq_time_seq.sequence);
1505 static inline void irq_time_write_end(void)
1508 __this_cpu_inc(irq_time_seq.sequence);
1511 static inline u64 irq_time_read(int cpu)
1517 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1518 irq_time = per_cpu(cpu_softirq_time, cpu) +
1519 per_cpu(cpu_hardirq_time, cpu);
1520 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1524 #else /* CONFIG_64BIT */
1525 static inline void irq_time_write_begin(void)
1529 static inline void irq_time_write_end(void)
1533 static inline u64 irq_time_read(int cpu)
1535 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1537 #endif /* CONFIG_64BIT */
1538 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */