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/irq_work.h>
10 #include <linux/tick.h>
11 #include <linux/slab.h>
14 #include "cpudeadline.h"
20 /* task_struct::on_rq states: */
21 #define TASK_ON_RQ_QUEUED 1
22 #define TASK_ON_RQ_MIGRATING 2
24 extern __read_mostly int scheduler_running;
26 extern unsigned long calc_load_update;
27 extern atomic_long_t calc_load_tasks;
29 extern void calc_global_load_tick(struct rq *this_rq);
30 extern long calc_load_fold_active(struct rq *this_rq);
33 extern void update_cpu_load_active(struct rq *this_rq);
35 static inline void update_cpu_load_active(struct rq *this_rq) { }
39 * Helpers for converting nanosecond timing to jiffy resolution
41 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
44 * Increase resolution of nice-level calculations for 64-bit architectures.
45 * The extra resolution improves shares distribution and load balancing of
46 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
47 * hierarchies, especially on larger systems. This is not a user-visible change
48 * and does not change the user-interface for setting shares/weights.
50 * We increase resolution only if we have enough bits to allow this increased
51 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
52 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
55 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
56 # define SCHED_LOAD_RESOLUTION 10
57 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
58 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
60 # define SCHED_LOAD_RESOLUTION 0
61 # define scale_load(w) (w)
62 # define scale_load_down(w) (w)
65 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
66 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
68 #define NICE_0_LOAD SCHED_LOAD_SCALE
69 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
72 * Single value that decides SCHED_DEADLINE internal math precision.
73 * 10 -> just above 1us
74 * 9 -> just above 0.5us
79 * These are the 'tuning knobs' of the scheduler:
83 * single value that denotes runtime == period, ie unlimited time.
85 #define RUNTIME_INF ((u64)~0ULL)
87 static inline int idle_policy(int policy)
89 return policy == SCHED_IDLE;
91 static inline int fair_policy(int policy)
93 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
96 static inline int rt_policy(int policy)
98 return policy == SCHED_FIFO || policy == SCHED_RR;
101 static inline int dl_policy(int policy)
103 return policy == SCHED_DEADLINE;
105 static inline bool valid_policy(int policy)
107 return idle_policy(policy) || fair_policy(policy) ||
108 rt_policy(policy) || dl_policy(policy);
111 static inline int task_has_rt_policy(struct task_struct *p)
113 return rt_policy(p->policy);
116 static inline int task_has_dl_policy(struct task_struct *p)
118 return dl_policy(p->policy);
122 * Tells if entity @a should preempt entity @b.
125 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
127 return dl_time_before(a->deadline, b->deadline);
131 * This is the priority-queue data structure of the RT scheduling class:
133 struct rt_prio_array {
134 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
135 struct list_head queue[MAX_RT_PRIO];
138 struct rt_bandwidth {
139 /* nests inside the rq lock: */
140 raw_spinlock_t rt_runtime_lock;
143 struct hrtimer rt_period_timer;
144 unsigned int rt_period_active;
147 void __dl_clear_params(struct task_struct *p);
150 * To keep the bandwidth of -deadline tasks and groups under control
151 * we need some place where:
152 * - store the maximum -deadline bandwidth of the system (the group);
153 * - cache the fraction of that bandwidth that is currently allocated.
155 * This is all done in the data structure below. It is similar to the
156 * one used for RT-throttling (rt_bandwidth), with the main difference
157 * that, since here we are only interested in admission control, we
158 * do not decrease any runtime while the group "executes", neither we
159 * need a timer to replenish it.
161 * With respect to SMP, the bandwidth is given on a per-CPU basis,
163 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
164 * - dl_total_bw array contains, in the i-eth element, the currently
165 * allocated bandwidth on the i-eth CPU.
166 * Moreover, groups consume bandwidth on each CPU, while tasks only
167 * consume bandwidth on the CPU they're running on.
168 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
169 * that will be shown the next time the proc or cgroup controls will
170 * be red. It on its turn can be changed by writing on its own
173 struct dl_bandwidth {
174 raw_spinlock_t dl_runtime_lock;
179 static inline int dl_bandwidth_enabled(void)
181 return sysctl_sched_rt_runtime >= 0;
184 extern struct dl_bw *dl_bw_of(int i);
192 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
194 dl_b->total_bw -= tsk_bw;
198 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
200 dl_b->total_bw += tsk_bw;
204 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
206 return dl_b->bw != -1 &&
207 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
210 extern struct mutex sched_domains_mutex;
212 #ifdef CONFIG_CGROUP_SCHED
214 #include <linux/cgroup.h>
219 extern struct list_head task_groups;
221 struct cfs_bandwidth {
222 #ifdef CONFIG_CFS_BANDWIDTH
226 s64 hierarchical_quota;
229 int idle, period_active;
230 struct hrtimer period_timer, slack_timer;
231 struct list_head throttled_cfs_rq;
234 int nr_periods, nr_throttled;
239 /* task group related information */
241 struct cgroup_subsys_state css;
243 #ifdef CONFIG_FAIR_GROUP_SCHED
244 /* schedulable entities of this group on each cpu */
245 struct sched_entity **se;
246 /* runqueue "owned" by this group on each cpu */
247 struct cfs_rq **cfs_rq;
248 unsigned long shares;
251 atomic_long_t load_avg;
255 #ifdef CONFIG_RT_GROUP_SCHED
256 struct sched_rt_entity **rt_se;
257 struct rt_rq **rt_rq;
259 struct rt_bandwidth rt_bandwidth;
263 struct list_head list;
265 struct task_group *parent;
266 struct list_head siblings;
267 struct list_head children;
269 #ifdef CONFIG_SCHED_AUTOGROUP
270 struct autogroup *autogroup;
273 struct cfs_bandwidth cfs_bandwidth;
276 #ifdef CONFIG_FAIR_GROUP_SCHED
277 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
280 * A weight of 0 or 1 can cause arithmetics problems.
281 * A weight of a cfs_rq is the sum of weights of which entities
282 * are queued on this cfs_rq, so a weight of a entity should not be
283 * too large, so as the shares value of a task group.
284 * (The default weight is 1024 - so there's no practical
285 * limitation from this.)
287 #define MIN_SHARES (1UL << 1)
288 #define MAX_SHARES (1UL << 18)
291 typedef int (*tg_visitor)(struct task_group *, void *);
293 extern int walk_tg_tree_from(struct task_group *from,
294 tg_visitor down, tg_visitor up, void *data);
297 * Iterate the full tree, calling @down when first entering a node and @up when
298 * leaving it for the final time.
300 * Caller must hold rcu_lock or sufficient equivalent.
302 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
304 return walk_tg_tree_from(&root_task_group, down, up, data);
307 extern int tg_nop(struct task_group *tg, void *data);
309 extern void free_fair_sched_group(struct task_group *tg);
310 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
311 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
312 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
313 struct sched_entity *se, int cpu,
314 struct sched_entity *parent);
315 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
316 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
318 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
319 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
320 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
322 extern void free_rt_sched_group(struct task_group *tg);
323 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
324 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
325 struct sched_rt_entity *rt_se, int cpu,
326 struct sched_rt_entity *parent);
328 extern struct task_group *sched_create_group(struct task_group *parent);
329 extern void sched_online_group(struct task_group *tg,
330 struct task_group *parent);
331 extern void sched_destroy_group(struct task_group *tg);
332 extern void sched_offline_group(struct task_group *tg);
334 extern void sched_move_task(struct task_struct *tsk);
336 #ifdef CONFIG_FAIR_GROUP_SCHED
337 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
340 #else /* CONFIG_CGROUP_SCHED */
342 struct cfs_bandwidth { };
344 #endif /* CONFIG_CGROUP_SCHED */
346 /* CFS-related fields in a runqueue */
348 struct load_weight load;
349 unsigned int nr_running, h_nr_running;
354 u64 min_vruntime_copy;
357 struct rb_root tasks_timeline;
358 struct rb_node *rb_leftmost;
361 * 'curr' points to currently running entity on this cfs_rq.
362 * It is set to NULL otherwise (i.e when none are currently running).
364 struct sched_entity *curr, *next, *last, *skip;
366 #ifdef CONFIG_SCHED_DEBUG
367 unsigned int nr_spread_over;
374 struct sched_avg avg;
375 u64 runnable_load_sum;
376 unsigned long runnable_load_avg;
377 #ifdef CONFIG_FAIR_GROUP_SCHED
378 unsigned long tg_load_avg_contrib;
380 atomic_long_t removed_load_avg, removed_util_avg;
382 u64 load_last_update_time_copy;
385 #ifdef CONFIG_FAIR_GROUP_SCHED
387 * h_load = weight * f(tg)
389 * Where f(tg) is the recursive weight fraction assigned to
392 unsigned long h_load;
393 u64 last_h_load_update;
394 struct sched_entity *h_load_next;
395 #endif /* CONFIG_FAIR_GROUP_SCHED */
396 #endif /* CONFIG_SMP */
398 #ifdef CONFIG_FAIR_GROUP_SCHED
399 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
402 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
403 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
404 * (like users, containers etc.)
406 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
407 * list is used during load balance.
410 struct list_head leaf_cfs_rq_list;
411 struct task_group *tg; /* group that "owns" this runqueue */
413 #ifdef CONFIG_SCHED_WALT
414 u64 cumulative_runnable_avg;
417 #ifdef CONFIG_CFS_BANDWIDTH
420 s64 runtime_remaining;
422 u64 throttled_clock, throttled_clock_task;
423 u64 throttled_clock_task_time;
424 int throttled, throttle_count, throttle_uptodate;
425 struct list_head throttled_list;
426 #endif /* CONFIG_CFS_BANDWIDTH */
427 #endif /* CONFIG_FAIR_GROUP_SCHED */
430 static inline int rt_bandwidth_enabled(void)
432 return sysctl_sched_rt_runtime >= 0;
435 /* RT IPI pull logic requires IRQ_WORK */
436 #ifdef CONFIG_IRQ_WORK
437 # define HAVE_RT_PUSH_IPI
440 /* Real-Time classes' related field in a runqueue: */
442 struct rt_prio_array active;
443 unsigned int rt_nr_running;
444 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
446 int curr; /* highest queued rt task prio */
448 int next; /* next highest */
453 unsigned long rt_nr_migratory;
454 unsigned long rt_nr_total;
456 struct plist_head pushable_tasks;
457 #ifdef HAVE_RT_PUSH_IPI
460 struct irq_work push_work;
461 raw_spinlock_t push_lock;
463 #endif /* CONFIG_SMP */
469 /* Nests inside the rq lock: */
470 raw_spinlock_t rt_runtime_lock;
472 #ifdef CONFIG_RT_GROUP_SCHED
473 unsigned long rt_nr_boosted;
476 struct task_group *tg;
480 /* Deadline class' related fields in a runqueue */
482 /* runqueue is an rbtree, ordered by deadline */
483 struct rb_root rb_root;
484 struct rb_node *rb_leftmost;
486 unsigned long dl_nr_running;
490 * Deadline values of the currently executing and the
491 * earliest ready task on this rq. Caching these facilitates
492 * the decision wether or not a ready but not running task
493 * should migrate somewhere else.
500 unsigned long dl_nr_migratory;
504 * Tasks on this rq that can be pushed away. They are kept in
505 * an rb-tree, ordered by tasks' deadlines, with caching
506 * of the leftmost (earliest deadline) element.
508 struct rb_root pushable_dl_tasks_root;
509 struct rb_node *pushable_dl_tasks_leftmost;
513 /* This is the "average utilization" for this runqueue */
519 struct max_cpu_capacity {
526 * We add the notion of a root-domain which will be used to define per-domain
527 * variables. Each exclusive cpuset essentially defines an island domain by
528 * fully partitioning the member cpus from any other cpuset. Whenever a new
529 * exclusive cpuset is created, we also create and attach a new root-domain
538 cpumask_var_t online;
540 /* Indicate more than one runnable task for any CPU */
543 /* Indicate one or more cpus over-utilized (tipping point) */
547 * The bit corresponding to a CPU gets set here if such CPU has more
548 * than one runnable -deadline task (as it is below for RT tasks).
550 cpumask_var_t dlo_mask;
556 * The "RT overload" flag: it gets set if a CPU has more than
557 * one runnable RT task.
559 cpumask_var_t rto_mask;
560 struct cpupri cpupri;
562 /* Maximum cpu capacity in the system. */
563 struct max_cpu_capacity max_cpu_capacity;
565 /* First cpu with maximum and minimum original capacity */
566 int max_cap_orig_cpu, min_cap_orig_cpu;
569 extern struct root_domain def_root_domain;
571 #endif /* CONFIG_SMP */
574 * This is the main, per-CPU runqueue data structure.
576 * Locking rule: those places that want to lock multiple runqueues
577 * (such as the load balancing or the thread migration code), lock
578 * acquire operations must be ordered by ascending &runqueue.
585 * nr_running and cpu_load should be in the same cacheline because
586 * remote CPUs use both these fields when doing load calculation.
588 unsigned int nr_running;
589 #ifdef CONFIG_NUMA_BALANCING
590 unsigned int nr_numa_running;
591 unsigned int nr_preferred_running;
593 #define CPU_LOAD_IDX_MAX 5
594 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
595 unsigned long last_load_update_tick;
596 unsigned int misfit_task;
597 #ifdef CONFIG_NO_HZ_COMMON
599 unsigned long nohz_flags;
601 #ifdef CONFIG_NO_HZ_FULL
602 unsigned long last_sched_tick;
605 #ifdef CONFIG_CPU_QUIET
606 /* time-based average load */
608 u64 nr_running_integral;
609 seqcount_t ave_seqcnt;
612 /* capture load from *all* tasks on this cpu: */
613 struct load_weight load;
614 unsigned long nr_load_updates;
621 #ifdef CONFIG_FAIR_GROUP_SCHED
622 /* list of leaf cfs_rq on this cpu: */
623 struct list_head leaf_cfs_rq_list;
624 #endif /* CONFIG_FAIR_GROUP_SCHED */
627 * This is part of a global counter where only the total sum
628 * over all CPUs matters. A task can increase this counter on
629 * one CPU and if it got migrated afterwards it may decrease
630 * it on another CPU. Always updated under the runqueue lock:
632 unsigned long nr_uninterruptible;
634 struct task_struct *curr, *idle, *stop;
635 unsigned long next_balance;
636 struct mm_struct *prev_mm;
638 unsigned int clock_skip_update;
645 struct root_domain *rd;
646 struct sched_domain *sd;
648 unsigned long cpu_capacity;
649 unsigned long cpu_capacity_orig;
651 struct callback_head *balance_callback;
653 unsigned char idle_balance;
654 /* For active balancing */
657 struct cpu_stop_work active_balance_work;
658 /* cpu of this runqueue: */
662 struct list_head cfs_tasks;
669 /* This is used to determine avg_idle's max value */
670 u64 max_idle_balance_cost;
673 #ifdef CONFIG_SCHED_WALT
675 * max_freq = user or thermal defined maximum
676 * max_possible_freq = maximum supported by hardware
678 unsigned int cur_freq, max_freq, min_freq, max_possible_freq;
679 struct cpumask freq_domain_cpumask;
681 u64 cumulative_runnable_avg;
682 int efficiency; /* Differentiate cpus with different IPC capability */
683 int load_scale_factor;
685 int max_possible_capacity;
687 u64 curr_runnable_sum;
688 u64 prev_runnable_sum;
689 u64 nt_curr_runnable_sum;
690 u64 nt_prev_runnable_sum;
694 #endif /* CONFIG_SCHED_WALT */
697 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
700 #ifdef CONFIG_PARAVIRT
703 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
704 u64 prev_steal_time_rq;
707 /* calc_load related fields */
708 unsigned long calc_load_update;
709 long calc_load_active;
711 #ifdef CONFIG_SCHED_HRTICK
713 int hrtick_csd_pending;
714 struct call_single_data hrtick_csd;
716 struct hrtimer hrtick_timer;
719 #ifdef CONFIG_SCHEDSTATS
721 struct sched_info rq_sched_info;
722 unsigned long long rq_cpu_time;
723 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
725 /* sys_sched_yield() stats */
726 unsigned int yld_count;
728 /* schedule() stats */
729 unsigned int sched_count;
730 unsigned int sched_goidle;
732 /* try_to_wake_up() stats */
733 unsigned int ttwu_count;
734 unsigned int ttwu_local;
738 struct llist_head wake_list;
741 #ifdef CONFIG_CPU_IDLE
742 /* Must be inspected within a rcu lock section */
743 struct cpuidle_state *idle_state;
748 static inline int cpu_of(struct rq *rq)
757 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
759 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
760 #define this_rq() this_cpu_ptr(&runqueues)
761 #define task_rq(p) cpu_rq(task_cpu(p))
762 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
763 #define raw_rq() raw_cpu_ptr(&runqueues)
765 static inline u64 __rq_clock_broken(struct rq *rq)
767 return READ_ONCE(rq->clock);
770 static inline u64 rq_clock(struct rq *rq)
772 lockdep_assert_held(&rq->lock);
776 static inline u64 rq_clock_task(struct rq *rq)
778 lockdep_assert_held(&rq->lock);
779 return rq->clock_task;
782 #define RQCF_REQ_SKIP 0x01
783 #define RQCF_ACT_SKIP 0x02
785 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
787 lockdep_assert_held(&rq->lock);
789 rq->clock_skip_update |= RQCF_REQ_SKIP;
791 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
795 enum numa_topology_type {
800 extern enum numa_topology_type sched_numa_topology_type;
801 extern int sched_max_numa_distance;
802 extern bool find_numa_distance(int distance);
805 #ifdef CONFIG_NUMA_BALANCING
806 /* The regions in numa_faults array from task_struct */
807 enum numa_faults_stats {
813 extern void sched_setnuma(struct task_struct *p, int node);
814 extern int migrate_task_to(struct task_struct *p, int cpu);
815 extern int migrate_swap(struct task_struct *, struct task_struct *);
816 #endif /* CONFIG_NUMA_BALANCING */
821 queue_balance_callback(struct rq *rq,
822 struct callback_head *head,
823 void (*func)(struct rq *rq))
825 lockdep_assert_held(&rq->lock);
827 if (unlikely(head->next))
830 head->func = (void (*)(struct callback_head *))func;
831 head->next = rq->balance_callback;
832 rq->balance_callback = head;
835 extern void sched_ttwu_pending(void);
837 #define rcu_dereference_check_sched_domain(p) \
838 rcu_dereference_check((p), \
839 lockdep_is_held(&sched_domains_mutex))
842 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
843 * See detach_destroy_domains: synchronize_sched for details.
845 * The domain tree of any CPU may only be accessed from within
846 * preempt-disabled sections.
848 #define for_each_domain(cpu, __sd) \
849 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
850 __sd; __sd = __sd->parent)
852 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
855 * highest_flag_domain - Return highest sched_domain containing flag.
856 * @cpu: The cpu whose highest level of sched domain is to
858 * @flag: The flag to check for the highest sched_domain
861 * Returns the highest sched_domain of a cpu which contains the given flag.
863 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
865 struct sched_domain *sd, *hsd = NULL;
867 for_each_domain(cpu, sd) {
868 if (!(sd->flags & flag))
876 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
878 struct sched_domain *sd;
880 for_each_domain(cpu, sd) {
881 if (sd->flags & flag)
888 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
889 DECLARE_PER_CPU(int, sd_llc_size);
890 DECLARE_PER_CPU(int, sd_llc_id);
891 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
892 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
893 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
894 DECLARE_PER_CPU(struct sched_domain *, sd_ea);
895 DECLARE_PER_CPU(struct sched_domain *, sd_scs);
897 struct sched_group_capacity {
900 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
903 unsigned long capacity;
904 unsigned long max_capacity; /* Max per-cpu capacity in group */
905 unsigned long min_capacity; /* Min per-CPU capacity in group */
906 unsigned long next_update;
907 int imbalance; /* XXX unrelated to capacity but shared group state */
909 * Number of busy cpus in this group.
911 atomic_t nr_busy_cpus;
913 unsigned long cpumask[0]; /* iteration mask */
917 struct sched_group *next; /* Must be a circular list */
920 unsigned int group_weight;
921 struct sched_group_capacity *sgc;
922 const struct sched_group_energy *sge;
925 * The CPUs this group covers.
927 * NOTE: this field is variable length. (Allocated dynamically
928 * by attaching extra space to the end of the structure,
929 * depending on how many CPUs the kernel has booted up with)
931 unsigned long cpumask[0];
934 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
936 return to_cpumask(sg->cpumask);
940 * cpumask masking which cpus in the group are allowed to iterate up the domain
943 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
945 return to_cpumask(sg->sgc->cpumask);
949 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
950 * @group: The group whose first cpu is to be returned.
952 static inline unsigned int group_first_cpu(struct sched_group *group)
954 return cpumask_first(sched_group_cpus(group));
957 extern int group_balance_cpu(struct sched_group *sg);
961 static inline void sched_ttwu_pending(void) { }
963 #endif /* CONFIG_SMP */
966 #include "auto_group.h"
968 #ifdef CONFIG_CGROUP_SCHED
971 * Return the group to which this tasks belongs.
973 * We cannot use task_css() and friends because the cgroup subsystem
974 * changes that value before the cgroup_subsys::attach() method is called,
975 * therefore we cannot pin it and might observe the wrong value.
977 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
978 * core changes this before calling sched_move_task().
980 * Instead we use a 'copy' which is updated from sched_move_task() while
981 * holding both task_struct::pi_lock and rq::lock.
983 static inline struct task_group *task_group(struct task_struct *p)
985 return p->sched_task_group;
988 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
989 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
991 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
992 struct task_group *tg = task_group(p);
995 #ifdef CONFIG_FAIR_GROUP_SCHED
996 p->se.cfs_rq = tg->cfs_rq[cpu];
997 p->se.parent = tg->se[cpu];
1000 #ifdef CONFIG_RT_GROUP_SCHED
1001 p->rt.rt_rq = tg->rt_rq[cpu];
1002 p->rt.parent = tg->rt_se[cpu];
1006 #else /* CONFIG_CGROUP_SCHED */
1008 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1009 static inline struct task_group *task_group(struct task_struct *p)
1014 #endif /* CONFIG_CGROUP_SCHED */
1016 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1018 set_task_rq(p, cpu);
1021 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1022 * successfuly executed on another CPU. We must ensure that updates of
1023 * per-task data have been completed by this moment.
1026 task_thread_info(p)->cpu = cpu;
1032 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1034 #ifdef CONFIG_SCHED_DEBUG
1035 # include <linux/static_key.h>
1036 # define const_debug __read_mostly
1038 # define const_debug const
1041 extern const_debug unsigned int sysctl_sched_features;
1043 #define SCHED_FEAT(name, enabled) \
1044 __SCHED_FEAT_##name ,
1047 #include "features.h"
1053 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1054 #define SCHED_FEAT(name, enabled) \
1055 static __always_inline bool static_branch_##name(struct static_key *key) \
1057 return static_key_##enabled(key); \
1060 #include "features.h"
1064 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1065 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1066 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1067 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1068 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1070 extern struct static_key_false sched_numa_balancing;
1072 static inline u64 global_rt_period(void)
1074 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1077 static inline u64 global_rt_runtime(void)
1079 if (sysctl_sched_rt_runtime < 0)
1082 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1085 static inline int task_current(struct rq *rq, struct task_struct *p)
1087 return rq->curr == p;
1090 static inline int task_running(struct rq *rq, struct task_struct *p)
1095 return task_current(rq, p);
1099 static inline int task_on_rq_queued(struct task_struct *p)
1101 return p->on_rq == TASK_ON_RQ_QUEUED;
1104 static inline int task_on_rq_migrating(struct task_struct *p)
1106 return p->on_rq == TASK_ON_RQ_MIGRATING;
1109 #ifndef prepare_arch_switch
1110 # define prepare_arch_switch(next) do { } while (0)
1112 #ifndef finish_arch_post_lock_switch
1113 # define finish_arch_post_lock_switch() do { } while (0)
1116 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1120 * We can optimise this out completely for !SMP, because the
1121 * SMP rebalancing from interrupt is the only thing that cares
1128 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1132 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1133 * We must ensure this doesn't happen until the switch is completely
1136 * In particular, the load of prev->state in finish_task_switch() must
1137 * happen before this.
1139 * Pairs with the control dependency and rmb in try_to_wake_up().
1141 smp_store_release(&prev->on_cpu, 0);
1143 #ifdef CONFIG_DEBUG_SPINLOCK
1144 /* this is a valid case when another task releases the spinlock */
1145 rq->lock.owner = current;
1148 * If we are tracking spinlock dependencies then we have to
1149 * fix up the runqueue lock - which gets 'carried over' from
1150 * prev into current:
1152 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1154 raw_spin_unlock_irq(&rq->lock);
1160 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1161 #define WF_FORK 0x02 /* child wakeup after fork */
1162 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1165 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1166 * of tasks with abnormal "nice" values across CPUs the contribution that
1167 * each task makes to its run queue's load is weighted according to its
1168 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1169 * scaled version of the new time slice allocation that they receive on time
1173 #define WEIGHT_IDLEPRIO 3
1174 #define WMULT_IDLEPRIO 1431655765
1177 * Nice levels are multiplicative, with a gentle 10% change for every
1178 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1179 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1180 * that remained on nice 0.
1182 * The "10% effect" is relative and cumulative: from _any_ nice level,
1183 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1184 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1185 * If a task goes up by ~10% and another task goes down by ~10% then
1186 * the relative distance between them is ~25%.)
1188 static const int prio_to_weight[40] = {
1189 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1190 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1191 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1192 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1193 /* 0 */ 1024, 820, 655, 526, 423,
1194 /* 5 */ 335, 272, 215, 172, 137,
1195 /* 10 */ 110, 87, 70, 56, 45,
1196 /* 15 */ 36, 29, 23, 18, 15,
1200 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1202 * In cases where the weight does not change often, we can use the
1203 * precalculated inverse to speed up arithmetics by turning divisions
1204 * into multiplications:
1206 static const u32 prio_to_wmult[40] = {
1207 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1208 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1209 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1210 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1211 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1212 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1213 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1214 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1217 #define ENQUEUE_WAKEUP 0x01
1218 #define ENQUEUE_HEAD 0x02
1220 #define ENQUEUE_WAKING 0x04 /* sched_class::task_waking was called */
1222 #define ENQUEUE_WAKING 0x00
1224 #define ENQUEUE_REPLENISH 0x08
1225 #define ENQUEUE_RESTORE 0x10
1226 #define ENQUEUE_WAKEUP_NEW 0x20
1228 #define DEQUEUE_SLEEP 0x01
1229 #define DEQUEUE_SAVE 0x02
1231 #define RETRY_TASK ((void *)-1UL)
1233 struct sched_class {
1234 const struct sched_class *next;
1236 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1237 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1238 void (*yield_task) (struct rq *rq);
1239 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1241 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1244 * It is the responsibility of the pick_next_task() method that will
1245 * return the next task to call put_prev_task() on the @prev task or
1246 * something equivalent.
1248 * May return RETRY_TASK when it finds a higher prio class has runnable
1251 struct task_struct * (*pick_next_task) (struct rq *rq,
1252 struct task_struct *prev);
1253 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1256 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1257 void (*migrate_task_rq)(struct task_struct *p);
1259 void (*task_waking) (struct task_struct *task);
1260 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1262 void (*set_cpus_allowed)(struct task_struct *p,
1263 const struct cpumask *newmask);
1265 void (*rq_online)(struct rq *rq);
1266 void (*rq_offline)(struct rq *rq);
1269 void (*set_curr_task) (struct rq *rq);
1270 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1271 void (*task_fork) (struct task_struct *p);
1272 void (*task_dead) (struct task_struct *p);
1275 * The switched_from() call is allowed to drop rq->lock, therefore we
1276 * cannot assume the switched_from/switched_to pair is serliazed by
1277 * rq->lock. They are however serialized by p->pi_lock.
1279 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1280 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1281 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1284 unsigned int (*get_rr_interval) (struct rq *rq,
1285 struct task_struct *task);
1287 void (*update_curr) (struct rq *rq);
1289 #ifdef CONFIG_FAIR_GROUP_SCHED
1290 void (*task_move_group) (struct task_struct *p);
1294 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1296 prev->sched_class->put_prev_task(rq, prev);
1299 #define sched_class_highest (&stop_sched_class)
1300 #define for_each_class(class) \
1301 for (class = sched_class_highest; class; class = class->next)
1303 extern const struct sched_class stop_sched_class;
1304 extern const struct sched_class dl_sched_class;
1305 extern const struct sched_class rt_sched_class;
1306 extern const struct sched_class fair_sched_class;
1307 extern const struct sched_class idle_sched_class;
1312 extern void init_max_cpu_capacity(struct max_cpu_capacity *mcc);
1313 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1315 extern void trigger_load_balance(struct rq *rq);
1317 extern void idle_enter_fair(struct rq *this_rq);
1318 extern void idle_exit_fair(struct rq *this_rq);
1320 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1324 static inline void idle_enter_fair(struct rq *rq) { }
1325 static inline void idle_exit_fair(struct rq *rq) { }
1329 #ifdef CONFIG_CPU_IDLE
1330 static inline void idle_set_state(struct rq *rq,
1331 struct cpuidle_state *idle_state)
1333 rq->idle_state = idle_state;
1336 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1338 WARN_ON(!rcu_read_lock_held());
1339 return rq->idle_state;
1342 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1344 rq->idle_state_idx = idle_state_idx;
1347 static inline int idle_get_state_idx(struct rq *rq)
1349 WARN_ON(!rcu_read_lock_held());
1350 return rq->idle_state_idx;
1353 static inline void idle_set_state(struct rq *rq,
1354 struct cpuidle_state *idle_state)
1358 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1363 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1367 static inline int idle_get_state_idx(struct rq *rq)
1373 extern void sysrq_sched_debug_show(void);
1374 extern void sched_init_granularity(void);
1375 extern void update_max_interval(void);
1377 extern void init_sched_dl_class(void);
1378 extern void init_sched_rt_class(void);
1379 extern void init_sched_fair_class(void);
1381 extern void resched_curr(struct rq *rq);
1382 extern void resched_cpu(int cpu);
1384 extern struct rt_bandwidth def_rt_bandwidth;
1385 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1387 extern struct dl_bandwidth def_dl_bandwidth;
1388 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1389 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1391 unsigned long to_ratio(u64 period, u64 runtime);
1393 extern void init_entity_runnable_average(struct sched_entity *se);
1394 extern void post_init_entity_util_avg(struct sched_entity *se);
1396 static inline void __add_nr_running(struct rq *rq, unsigned count)
1398 unsigned prev_nr = rq->nr_running;
1400 rq->nr_running = prev_nr + count;
1402 if (prev_nr < 2 && rq->nr_running >= 2) {
1404 if (!rq->rd->overload)
1405 rq->rd->overload = true;
1408 #ifdef CONFIG_NO_HZ_FULL
1409 if (tick_nohz_full_cpu(rq->cpu)) {
1411 * Tick is needed if more than one task runs on a CPU.
1412 * Send the target an IPI to kick it out of nohz mode.
1414 * We assume that IPI implies full memory barrier and the
1415 * new value of rq->nr_running is visible on reception
1418 tick_nohz_full_kick_cpu(rq->cpu);
1424 static inline void __sub_nr_running(struct rq *rq, unsigned count)
1426 rq->nr_running -= count;
1429 #ifdef CONFIG_CPU_QUIET
1430 #define NR_AVE_SCALE(x) ((x) << FSHIFT)
1431 static inline u64 do_nr_running_integral(struct rq *rq)
1434 u64 nr_running_integral = rq->nr_running_integral;
1436 deltax = rq->clock_task - rq->nr_last_stamp;
1437 nr = NR_AVE_SCALE(rq->nr_running);
1439 nr_running_integral += nr * deltax;
1441 return nr_running_integral;
1444 static inline void add_nr_running(struct rq *rq, unsigned count)
1446 write_seqcount_begin(&rq->ave_seqcnt);
1447 rq->nr_running_integral = do_nr_running_integral(rq);
1448 rq->nr_last_stamp = rq->clock_task;
1449 __add_nr_running(rq, count);
1450 write_seqcount_end(&rq->ave_seqcnt);
1453 static inline void sub_nr_running(struct rq *rq, unsigned count)
1455 write_seqcount_begin(&rq->ave_seqcnt);
1456 rq->nr_running_integral = do_nr_running_integral(rq);
1457 rq->nr_last_stamp = rq->clock_task;
1458 __sub_nr_running(rq, count);
1459 write_seqcount_end(&rq->ave_seqcnt);
1462 #define add_nr_running __add_nr_running
1463 #define sub_nr_running __sub_nr_running
1466 static inline void rq_last_tick_reset(struct rq *rq)
1468 #ifdef CONFIG_NO_HZ_FULL
1469 rq->last_sched_tick = jiffies;
1473 extern void update_rq_clock(struct rq *rq);
1475 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1476 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1478 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1480 extern const_debug unsigned int sysctl_sched_time_avg;
1481 extern const_debug unsigned int sysctl_sched_nr_migrate;
1482 extern const_debug unsigned int sysctl_sched_migration_cost;
1484 static inline u64 sched_avg_period(void)
1486 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1489 #ifdef CONFIG_SCHED_HRTICK
1493 * - enabled by features
1494 * - hrtimer is actually high res
1496 static inline int hrtick_enabled(struct rq *rq)
1498 if (!sched_feat(HRTICK))
1500 if (!cpu_active(cpu_of(rq)))
1502 return hrtimer_is_hres_active(&rq->hrtick_timer);
1505 void hrtick_start(struct rq *rq, u64 delay);
1509 static inline int hrtick_enabled(struct rq *rq)
1514 #endif /* CONFIG_SCHED_HRTICK */
1517 extern void sched_avg_update(struct rq *rq);
1519 #ifndef arch_scale_freq_capacity
1520 static __always_inline
1521 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1523 return SCHED_CAPACITY_SCALE;
1527 #ifndef arch_scale_cpu_capacity
1528 static __always_inline
1529 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1531 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1532 return sd->smt_gain / sd->span_weight;
1534 return SCHED_CAPACITY_SCALE;
1539 static inline unsigned long capacity_of(int cpu)
1541 return cpu_rq(cpu)->cpu_capacity;
1544 static inline unsigned long capacity_orig_of(int cpu)
1546 return cpu_rq(cpu)->cpu_capacity_orig;
1549 extern unsigned int sysctl_sched_use_walt_cpu_util;
1550 extern unsigned int walt_ravg_window;
1551 extern unsigned int walt_disabled;
1554 * cpu_util returns the amount of capacity of a CPU that is used by CFS
1555 * tasks. The unit of the return value must be the one of capacity so we can
1556 * compare the utilization with the capacity of the CPU that is available for
1557 * CFS task (ie cpu_capacity).
1559 * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the
1560 * recent utilization of currently non-runnable tasks on a CPU. It represents
1561 * the amount of utilization of a CPU in the range [0..capacity_orig] where
1562 * capacity_orig is the cpu_capacity available at the highest frequency
1563 * (arch_scale_freq_capacity()).
1564 * The utilization of a CPU converges towards a sum equal to or less than the
1565 * current capacity (capacity_curr <= capacity_orig) of the CPU because it is
1566 * the running time on this CPU scaled by capacity_curr.
1568 * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even
1569 * higher than capacity_orig because of unfortunate rounding in
1570 * cfs.avg.util_avg or just after migrating tasks and new task wakeups until
1571 * the average stabilizes with the new running time. We need to check that the
1572 * utilization stays within the range of [0..capacity_orig] and cap it if
1573 * necessary. Without utilization capping, a group could be seen as overloaded
1574 * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of
1575 * available capacity. We allow utilization to overshoot capacity_curr (but not
1576 * capacity_orig) as it useful for predicting the capacity required after task
1577 * migrations (scheduler-driven DVFS).
1579 static inline unsigned long __cpu_util(int cpu, int delta)
1581 unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
1582 unsigned long capacity = capacity_orig_of(cpu);
1584 #ifdef CONFIG_SCHED_WALT
1585 if (!walt_disabled && sysctl_sched_use_walt_cpu_util) {
1586 util = cpu_rq(cpu)->prev_runnable_sum << SCHED_LOAD_SHIFT;
1587 do_div(util, walt_ravg_window);
1594 return (delta >= capacity) ? capacity : delta;
1597 static inline unsigned long cpu_util(int cpu)
1599 return __cpu_util(cpu, 0);
1604 #ifdef CONFIG_CPU_FREQ_GOV_SCHED
1605 #define capacity_max SCHED_CAPACITY_SCALE
1606 extern unsigned int capacity_margin;
1607 extern struct static_key __sched_freq;
1609 static inline bool sched_freq(void)
1611 return static_key_false(&__sched_freq);
1614 DECLARE_PER_CPU(struct sched_capacity_reqs, cpu_sched_capacity_reqs);
1615 void update_cpu_capacity_request(int cpu, bool request);
1617 static inline void set_cfs_cpu_capacity(int cpu, bool request,
1618 unsigned long capacity)
1620 struct sched_capacity_reqs *scr = &per_cpu(cpu_sched_capacity_reqs, cpu);
1622 #ifdef CONFIG_SCHED_WALT
1623 if (!walt_disabled && sysctl_sched_use_walt_cpu_util) {
1624 int rtdl = scr->rt + scr->dl;
1626 * WALT tracks the utilization of a CPU considering the load
1627 * generated by all the scheduling classes.
1628 * Since the following call to:
1629 * update_cpu_capacity
1630 * is already adding the RT and DL utilizations let's remove
1631 * these contributions from the WALT signal.
1633 if (capacity > rtdl)
1639 if (scr->cfs != capacity) {
1640 scr->cfs = capacity;
1641 update_cpu_capacity_request(cpu, request);
1645 static inline void set_rt_cpu_capacity(int cpu, bool request,
1646 unsigned long capacity)
1648 if (per_cpu(cpu_sched_capacity_reqs, cpu).rt != capacity) {
1649 per_cpu(cpu_sched_capacity_reqs, cpu).rt = capacity;
1650 update_cpu_capacity_request(cpu, request);
1654 static inline void set_dl_cpu_capacity(int cpu, bool request,
1655 unsigned long capacity)
1657 if (per_cpu(cpu_sched_capacity_reqs, cpu).dl != capacity) {
1658 per_cpu(cpu_sched_capacity_reqs, cpu).dl = capacity;
1659 update_cpu_capacity_request(cpu, request);
1663 static inline bool sched_freq(void) { return false; }
1664 static inline void set_cfs_cpu_capacity(int cpu, bool request,
1665 unsigned long capacity)
1667 static inline void set_rt_cpu_capacity(int cpu, bool request,
1668 unsigned long capacity)
1670 static inline void set_dl_cpu_capacity(int cpu, bool request,
1671 unsigned long capacity)
1675 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1677 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1680 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1681 static inline void sched_avg_update(struct rq *rq) { }
1685 * __task_rq_lock - lock the rq @p resides on.
1687 static inline struct rq *__task_rq_lock(struct task_struct *p)
1688 __acquires(rq->lock)
1692 lockdep_assert_held(&p->pi_lock);
1696 raw_spin_lock(&rq->lock);
1697 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1698 lockdep_pin_lock(&rq->lock);
1701 raw_spin_unlock(&rq->lock);
1703 while (unlikely(task_on_rq_migrating(p)))
1709 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1711 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1712 __acquires(p->pi_lock)
1713 __acquires(rq->lock)
1718 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1720 raw_spin_lock(&rq->lock);
1722 * move_queued_task() task_rq_lock()
1724 * ACQUIRE (rq->lock)
1725 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1726 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1727 * [S] ->cpu = new_cpu [L] task_rq()
1729 * RELEASE (rq->lock)
1731 * If we observe the old cpu in task_rq_lock, the acquire of
1732 * the old rq->lock will fully serialize against the stores.
1734 * If we observe the new cpu in task_rq_lock, the acquire will
1735 * pair with the WMB to ensure we must then also see migrating.
1737 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1738 lockdep_pin_lock(&rq->lock);
1741 raw_spin_unlock(&rq->lock);
1742 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1744 while (unlikely(task_on_rq_migrating(p)))
1749 static inline void __task_rq_unlock(struct rq *rq)
1750 __releases(rq->lock)
1752 lockdep_unpin_lock(&rq->lock);
1753 raw_spin_unlock(&rq->lock);
1757 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1758 __releases(rq->lock)
1759 __releases(p->pi_lock)
1761 lockdep_unpin_lock(&rq->lock);
1762 raw_spin_unlock(&rq->lock);
1763 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1766 extern struct rq *lock_rq_of(struct task_struct *p, unsigned long *flags);
1767 extern void unlock_rq_of(struct rq *rq, struct task_struct *p, unsigned long *flags);
1770 #ifdef CONFIG_PREEMPT
1772 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1775 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1776 * way at the expense of forcing extra atomic operations in all
1777 * invocations. This assures that the double_lock is acquired using the
1778 * same underlying policy as the spinlock_t on this architecture, which
1779 * reduces latency compared to the unfair variant below. However, it
1780 * also adds more overhead and therefore may reduce throughput.
1782 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1783 __releases(this_rq->lock)
1784 __acquires(busiest->lock)
1785 __acquires(this_rq->lock)
1787 raw_spin_unlock(&this_rq->lock);
1788 double_rq_lock(this_rq, busiest);
1795 * Unfair double_lock_balance: Optimizes throughput at the expense of
1796 * latency by eliminating extra atomic operations when the locks are
1797 * already in proper order on entry. This favors lower cpu-ids and will
1798 * grant the double lock to lower cpus over higher ids under contention,
1799 * regardless of entry order into the function.
1801 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1802 __releases(this_rq->lock)
1803 __acquires(busiest->lock)
1804 __acquires(this_rq->lock)
1808 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1809 if (busiest < this_rq) {
1810 raw_spin_unlock(&this_rq->lock);
1811 raw_spin_lock(&busiest->lock);
1812 raw_spin_lock_nested(&this_rq->lock,
1813 SINGLE_DEPTH_NESTING);
1816 raw_spin_lock_nested(&busiest->lock,
1817 SINGLE_DEPTH_NESTING);
1822 #endif /* CONFIG_PREEMPT */
1825 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1827 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1829 if (unlikely(!irqs_disabled())) {
1830 /* printk() doesn't work good under rq->lock */
1831 raw_spin_unlock(&this_rq->lock);
1835 return _double_lock_balance(this_rq, busiest);
1838 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1839 __releases(busiest->lock)
1841 if (this_rq != busiest)
1842 raw_spin_unlock(&busiest->lock);
1843 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1846 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1852 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1855 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1861 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1864 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1870 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1874 * double_rq_lock - safely lock two runqueues
1876 * Note this does not disable interrupts like task_rq_lock,
1877 * you need to do so manually before calling.
1879 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1880 __acquires(rq1->lock)
1881 __acquires(rq2->lock)
1883 BUG_ON(!irqs_disabled());
1885 raw_spin_lock(&rq1->lock);
1886 __acquire(rq2->lock); /* Fake it out ;) */
1889 raw_spin_lock(&rq1->lock);
1890 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1892 raw_spin_lock(&rq2->lock);
1893 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1899 * double_rq_unlock - safely unlock two runqueues
1901 * Note this does not restore interrupts like task_rq_unlock,
1902 * you need to do so manually after calling.
1904 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1905 __releases(rq1->lock)
1906 __releases(rq2->lock)
1908 raw_spin_unlock(&rq1->lock);
1910 raw_spin_unlock(&rq2->lock);
1912 __release(rq2->lock);
1915 #else /* CONFIG_SMP */
1918 * double_rq_lock - safely lock two runqueues
1920 * Note this does not disable interrupts like task_rq_lock,
1921 * you need to do so manually before calling.
1923 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1924 __acquires(rq1->lock)
1925 __acquires(rq2->lock)
1927 BUG_ON(!irqs_disabled());
1929 raw_spin_lock(&rq1->lock);
1930 __acquire(rq2->lock); /* Fake it out ;) */
1934 * double_rq_unlock - safely unlock two runqueues
1936 * Note this does not restore interrupts like task_rq_unlock,
1937 * you need to do so manually after calling.
1939 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1940 __releases(rq1->lock)
1941 __releases(rq2->lock)
1944 raw_spin_unlock(&rq1->lock);
1945 __release(rq2->lock);
1950 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1951 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1953 #ifdef CONFIG_SCHED_DEBUG
1954 extern void print_cfs_stats(struct seq_file *m, int cpu);
1955 extern void print_rt_stats(struct seq_file *m, int cpu);
1956 extern void print_dl_stats(struct seq_file *m, int cpu);
1958 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1960 #ifdef CONFIG_NUMA_BALANCING
1962 show_numa_stats(struct task_struct *p, struct seq_file *m);
1964 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1965 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1966 #endif /* CONFIG_NUMA_BALANCING */
1967 #endif /* CONFIG_SCHED_DEBUG */
1969 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1970 extern void init_rt_rq(struct rt_rq *rt_rq);
1971 extern void init_dl_rq(struct dl_rq *dl_rq);
1973 extern void cfs_bandwidth_usage_inc(void);
1974 extern void cfs_bandwidth_usage_dec(void);
1976 #ifdef CONFIG_NO_HZ_COMMON
1977 enum rq_nohz_flag_bits {
1982 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1985 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1987 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1988 DECLARE_PER_CPU(u64, cpu_softirq_time);
1990 #ifndef CONFIG_64BIT
1991 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1993 static inline void irq_time_write_begin(void)
1995 __this_cpu_inc(irq_time_seq.sequence);
1999 static inline void irq_time_write_end(void)
2002 __this_cpu_inc(irq_time_seq.sequence);
2005 static inline u64 irq_time_read(int cpu)
2011 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
2012 irq_time = per_cpu(cpu_softirq_time, cpu) +
2013 per_cpu(cpu_hardirq_time, cpu);
2014 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
2018 #else /* CONFIG_64BIT */
2019 static inline void irq_time_write_begin(void)
2023 static inline void irq_time_write_end(void)
2027 static inline u64 irq_time_read(int cpu)
2029 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
2031 #endif /* CONFIG_64BIT */
2032 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2034 static inline void account_reset_rq(struct rq *rq)
2036 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2037 rq->prev_irq_time = 0;
2039 #ifdef CONFIG_PARAVIRT
2040 rq->prev_steal_time = 0;
2042 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
2043 rq->prev_steal_time_rq = 0;
2047 #ifdef CONFIG_CPU_FREQ
2048 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
2051 * cpufreq_update_util - Take a note about CPU utilization changes.
2052 * @rq: Runqueue to carry out the update for.
2053 * @flags: Update reason flags.
2055 * This function is called by the scheduler on the CPU whose utilization is
2058 * It can only be called from RCU-sched read-side critical sections.
2060 * The way cpufreq is currently arranged requires it to evaluate the CPU
2061 * performance state (frequency/voltage) on a regular basis to prevent it from
2062 * being stuck in a completely inadequate performance level for too long.
2063 * That is not guaranteed to happen if the updates are only triggered from CFS,
2064 * though, because they may not be coming in if RT or deadline tasks are active
2065 * all the time (or there are RT and DL tasks only).
2067 * As a workaround for that issue, this function is called by the RT and DL
2068 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
2069 * but that really is a band-aid. Going forward it should be replaced with
2070 * solutions targeted more specifically at RT and DL tasks.
2072 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)
2074 struct update_util_data *data;
2076 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
2078 data->func(data, rq_clock(rq), flags);
2081 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags)
2083 if (cpu_of(rq) == smp_processor_id())
2084 cpufreq_update_util(rq, flags);
2087 static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {}
2088 static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) {}
2089 #endif /* CONFIG_CPU_FREQ */
2091 #ifdef arch_scale_freq_capacity
2092 #ifndef arch_scale_freq_invariant
2093 #define arch_scale_freq_invariant() (true)
2095 #else /* arch_scale_freq_capacity */
2096 #define arch_scale_freq_invariant() (false)