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;
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;
566 extern struct root_domain def_root_domain;
568 #endif /* CONFIG_SMP */
571 * This is the main, per-CPU runqueue data structure.
573 * Locking rule: those places that want to lock multiple runqueues
574 * (such as the load balancing or the thread migration code), lock
575 * acquire operations must be ordered by ascending &runqueue.
582 * nr_running and cpu_load should be in the same cacheline because
583 * remote CPUs use both these fields when doing load calculation.
585 unsigned int nr_running;
586 #ifdef CONFIG_NUMA_BALANCING
587 unsigned int nr_numa_running;
588 unsigned int nr_preferred_running;
590 #define CPU_LOAD_IDX_MAX 5
591 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
592 unsigned long last_load_update_tick;
593 unsigned int misfit_task;
594 #ifdef CONFIG_NO_HZ_COMMON
596 unsigned long nohz_flags;
598 #ifdef CONFIG_NO_HZ_FULL
599 unsigned long last_sched_tick;
602 #ifdef CONFIG_CPU_QUIET
603 /* time-based average load */
605 u64 nr_running_integral;
606 seqcount_t ave_seqcnt;
609 /* capture load from *all* tasks on this cpu: */
610 struct load_weight load;
611 unsigned long nr_load_updates;
618 #ifdef CONFIG_FAIR_GROUP_SCHED
619 /* list of leaf cfs_rq on this cpu: */
620 struct list_head leaf_cfs_rq_list;
621 #endif /* CONFIG_FAIR_GROUP_SCHED */
624 * This is part of a global counter where only the total sum
625 * over all CPUs matters. A task can increase this counter on
626 * one CPU and if it got migrated afterwards it may decrease
627 * it on another CPU. Always updated under the runqueue lock:
629 unsigned long nr_uninterruptible;
631 struct task_struct *curr, *idle, *stop;
632 unsigned long next_balance;
633 struct mm_struct *prev_mm;
635 unsigned int clock_skip_update;
642 struct root_domain *rd;
643 struct sched_domain *sd;
645 unsigned long cpu_capacity;
646 unsigned long cpu_capacity_orig;
648 struct callback_head *balance_callback;
650 unsigned char idle_balance;
651 /* For active balancing */
654 struct cpu_stop_work active_balance_work;
655 /* cpu of this runqueue: */
659 struct list_head cfs_tasks;
666 /* This is used to determine avg_idle's max value */
667 u64 max_idle_balance_cost;
670 #ifdef CONFIG_SCHED_WALT
672 * max_freq = user or thermal defined maximum
673 * max_possible_freq = maximum supported by hardware
675 unsigned int cur_freq, max_freq, min_freq, max_possible_freq;
676 struct cpumask freq_domain_cpumask;
678 u64 cumulative_runnable_avg;
679 int efficiency; /* Differentiate cpus with different IPC capability */
680 int load_scale_factor;
682 int max_possible_capacity;
684 u64 curr_runnable_sum;
685 u64 prev_runnable_sum;
686 u64 nt_curr_runnable_sum;
687 u64 nt_prev_runnable_sum;
691 #endif /* CONFIG_SCHED_WALT */
694 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
697 #ifdef CONFIG_PARAVIRT
700 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
701 u64 prev_steal_time_rq;
704 /* calc_load related fields */
705 unsigned long calc_load_update;
706 long calc_load_active;
708 #ifdef CONFIG_SCHED_HRTICK
710 int hrtick_csd_pending;
711 struct call_single_data hrtick_csd;
713 struct hrtimer hrtick_timer;
716 #ifdef CONFIG_SCHEDSTATS
718 struct sched_info rq_sched_info;
719 unsigned long long rq_cpu_time;
720 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
722 /* sys_sched_yield() stats */
723 unsigned int yld_count;
725 /* schedule() stats */
726 unsigned int sched_count;
727 unsigned int sched_goidle;
729 /* try_to_wake_up() stats */
730 unsigned int ttwu_count;
731 unsigned int ttwu_local;
735 struct llist_head wake_list;
738 #ifdef CONFIG_CPU_IDLE
739 /* Must be inspected within a rcu lock section */
740 struct cpuidle_state *idle_state;
745 static inline int cpu_of(struct rq *rq)
754 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
756 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
757 #define this_rq() this_cpu_ptr(&runqueues)
758 #define task_rq(p) cpu_rq(task_cpu(p))
759 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
760 #define raw_rq() raw_cpu_ptr(&runqueues)
762 static inline u64 __rq_clock_broken(struct rq *rq)
764 return READ_ONCE(rq->clock);
767 static inline u64 rq_clock(struct rq *rq)
769 lockdep_assert_held(&rq->lock);
773 static inline u64 rq_clock_task(struct rq *rq)
775 lockdep_assert_held(&rq->lock);
776 return rq->clock_task;
779 #define RQCF_REQ_SKIP 0x01
780 #define RQCF_ACT_SKIP 0x02
782 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
784 lockdep_assert_held(&rq->lock);
786 rq->clock_skip_update |= RQCF_REQ_SKIP;
788 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
792 enum numa_topology_type {
797 extern enum numa_topology_type sched_numa_topology_type;
798 extern int sched_max_numa_distance;
799 extern bool find_numa_distance(int distance);
802 #ifdef CONFIG_NUMA_BALANCING
803 /* The regions in numa_faults array from task_struct */
804 enum numa_faults_stats {
810 extern void sched_setnuma(struct task_struct *p, int node);
811 extern int migrate_task_to(struct task_struct *p, int cpu);
812 extern int migrate_swap(struct task_struct *, struct task_struct *);
813 #endif /* CONFIG_NUMA_BALANCING */
818 queue_balance_callback(struct rq *rq,
819 struct callback_head *head,
820 void (*func)(struct rq *rq))
822 lockdep_assert_held(&rq->lock);
824 if (unlikely(head->next))
827 head->func = (void (*)(struct callback_head *))func;
828 head->next = rq->balance_callback;
829 rq->balance_callback = head;
832 extern void sched_ttwu_pending(void);
834 #define rcu_dereference_check_sched_domain(p) \
835 rcu_dereference_check((p), \
836 lockdep_is_held(&sched_domains_mutex))
839 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
840 * See detach_destroy_domains: synchronize_sched for details.
842 * The domain tree of any CPU may only be accessed from within
843 * preempt-disabled sections.
845 #define for_each_domain(cpu, __sd) \
846 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
847 __sd; __sd = __sd->parent)
849 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
852 * highest_flag_domain - Return highest sched_domain containing flag.
853 * @cpu: The cpu whose highest level of sched domain is to
855 * @flag: The flag to check for the highest sched_domain
858 * Returns the highest sched_domain of a cpu which contains the given flag.
860 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
862 struct sched_domain *sd, *hsd = NULL;
864 for_each_domain(cpu, sd) {
865 if (!(sd->flags & flag))
873 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
875 struct sched_domain *sd;
877 for_each_domain(cpu, sd) {
878 if (sd->flags & flag)
885 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
886 DECLARE_PER_CPU(int, sd_llc_size);
887 DECLARE_PER_CPU(int, sd_llc_id);
888 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
889 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
890 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
891 DECLARE_PER_CPU(struct sched_domain *, sd_ea);
892 DECLARE_PER_CPU(struct sched_domain *, sd_scs);
894 struct sched_group_capacity {
897 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
900 unsigned long capacity;
901 unsigned long max_capacity; /* Max per-cpu capacity in group */
902 unsigned long next_update;
903 int imbalance; /* XXX unrelated to capacity but shared group state */
905 * Number of busy cpus in this group.
907 atomic_t nr_busy_cpus;
909 unsigned long cpumask[0]; /* iteration mask */
913 struct sched_group *next; /* Must be a circular list */
916 unsigned int group_weight;
917 struct sched_group_capacity *sgc;
918 const struct sched_group_energy const *sge;
921 * The CPUs this group covers.
923 * NOTE: this field is variable length. (Allocated dynamically
924 * by attaching extra space to the end of the structure,
925 * depending on how many CPUs the kernel has booted up with)
927 unsigned long cpumask[0];
930 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
932 return to_cpumask(sg->cpumask);
936 * cpumask masking which cpus in the group are allowed to iterate up the domain
939 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
941 return to_cpumask(sg->sgc->cpumask);
945 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
946 * @group: The group whose first cpu is to be returned.
948 static inline unsigned int group_first_cpu(struct sched_group *group)
950 return cpumask_first(sched_group_cpus(group));
953 extern int group_balance_cpu(struct sched_group *sg);
957 static inline void sched_ttwu_pending(void) { }
959 #endif /* CONFIG_SMP */
962 #include "auto_group.h"
964 #ifdef CONFIG_CGROUP_SCHED
967 * Return the group to which this tasks belongs.
969 * We cannot use task_css() and friends because the cgroup subsystem
970 * changes that value before the cgroup_subsys::attach() method is called,
971 * therefore we cannot pin it and might observe the wrong value.
973 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
974 * core changes this before calling sched_move_task().
976 * Instead we use a 'copy' which is updated from sched_move_task() while
977 * holding both task_struct::pi_lock and rq::lock.
979 static inline struct task_group *task_group(struct task_struct *p)
981 return p->sched_task_group;
984 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
985 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
987 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
988 struct task_group *tg = task_group(p);
991 #ifdef CONFIG_FAIR_GROUP_SCHED
992 p->se.cfs_rq = tg->cfs_rq[cpu];
993 p->se.parent = tg->se[cpu];
996 #ifdef CONFIG_RT_GROUP_SCHED
997 p->rt.rt_rq = tg->rt_rq[cpu];
998 p->rt.parent = tg->rt_se[cpu];
1002 #else /* CONFIG_CGROUP_SCHED */
1004 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
1005 static inline struct task_group *task_group(struct task_struct *p)
1010 #endif /* CONFIG_CGROUP_SCHED */
1012 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1014 set_task_rq(p, cpu);
1017 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1018 * successfuly executed on another CPU. We must ensure that updates of
1019 * per-task data have been completed by this moment.
1022 task_thread_info(p)->cpu = cpu;
1028 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1030 #ifdef CONFIG_SCHED_DEBUG
1031 # include <linux/static_key.h>
1032 # define const_debug __read_mostly
1034 # define const_debug const
1037 extern const_debug unsigned int sysctl_sched_features;
1039 #define SCHED_FEAT(name, enabled) \
1040 __SCHED_FEAT_##name ,
1043 #include "features.h"
1049 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1050 #define SCHED_FEAT(name, enabled) \
1051 static __always_inline bool static_branch_##name(struct static_key *key) \
1053 return static_key_##enabled(key); \
1056 #include "features.h"
1060 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1061 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1062 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1063 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1064 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1066 extern struct static_key_false sched_numa_balancing;
1068 static inline u64 global_rt_period(void)
1070 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1073 static inline u64 global_rt_runtime(void)
1075 if (sysctl_sched_rt_runtime < 0)
1078 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1081 static inline int task_current(struct rq *rq, struct task_struct *p)
1083 return rq->curr == p;
1086 static inline int task_running(struct rq *rq, struct task_struct *p)
1091 return task_current(rq, p);
1095 static inline int task_on_rq_queued(struct task_struct *p)
1097 return p->on_rq == TASK_ON_RQ_QUEUED;
1100 static inline int task_on_rq_migrating(struct task_struct *p)
1102 return p->on_rq == TASK_ON_RQ_MIGRATING;
1105 #ifndef prepare_arch_switch
1106 # define prepare_arch_switch(next) do { } while (0)
1108 #ifndef finish_arch_post_lock_switch
1109 # define finish_arch_post_lock_switch() do { } while (0)
1112 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1116 * We can optimise this out completely for !SMP, because the
1117 * SMP rebalancing from interrupt is the only thing that cares
1124 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1128 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1129 * We must ensure this doesn't happen until the switch is completely
1132 * In particular, the load of prev->state in finish_task_switch() must
1133 * happen before this.
1135 * Pairs with the control dependency and rmb in try_to_wake_up().
1137 smp_store_release(&prev->on_cpu, 0);
1139 #ifdef CONFIG_DEBUG_SPINLOCK
1140 /* this is a valid case when another task releases the spinlock */
1141 rq->lock.owner = current;
1144 * If we are tracking spinlock dependencies then we have to
1145 * fix up the runqueue lock - which gets 'carried over' from
1146 * prev into current:
1148 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1150 raw_spin_unlock_irq(&rq->lock);
1156 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1157 #define WF_FORK 0x02 /* child wakeup after fork */
1158 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1161 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1162 * of tasks with abnormal "nice" values across CPUs the contribution that
1163 * each task makes to its run queue's load is weighted according to its
1164 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1165 * scaled version of the new time slice allocation that they receive on time
1169 #define WEIGHT_IDLEPRIO 3
1170 #define WMULT_IDLEPRIO 1431655765
1173 * Nice levels are multiplicative, with a gentle 10% change for every
1174 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1175 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1176 * that remained on nice 0.
1178 * The "10% effect" is relative and cumulative: from _any_ nice level,
1179 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1180 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1181 * If a task goes up by ~10% and another task goes down by ~10% then
1182 * the relative distance between them is ~25%.)
1184 static const int prio_to_weight[40] = {
1185 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1186 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1187 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1188 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1189 /* 0 */ 1024, 820, 655, 526, 423,
1190 /* 5 */ 335, 272, 215, 172, 137,
1191 /* 10 */ 110, 87, 70, 56, 45,
1192 /* 15 */ 36, 29, 23, 18, 15,
1196 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1198 * In cases where the weight does not change often, we can use the
1199 * precalculated inverse to speed up arithmetics by turning divisions
1200 * into multiplications:
1202 static const u32 prio_to_wmult[40] = {
1203 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1204 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1205 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1206 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1207 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1208 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1209 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1210 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1213 #define ENQUEUE_WAKEUP 0x01
1214 #define ENQUEUE_HEAD 0x02
1216 #define ENQUEUE_WAKING 0x04 /* sched_class::task_waking was called */
1218 #define ENQUEUE_WAKING 0x00
1220 #define ENQUEUE_REPLENISH 0x08
1221 #define ENQUEUE_RESTORE 0x10
1222 #define ENQUEUE_WAKEUP_NEW 0x20
1224 #define DEQUEUE_SLEEP 0x01
1225 #define DEQUEUE_SAVE 0x02
1227 #define RETRY_TASK ((void *)-1UL)
1229 struct sched_class {
1230 const struct sched_class *next;
1232 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1233 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1234 void (*yield_task) (struct rq *rq);
1235 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1237 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1240 * It is the responsibility of the pick_next_task() method that will
1241 * return the next task to call put_prev_task() on the @prev task or
1242 * something equivalent.
1244 * May return RETRY_TASK when it finds a higher prio class has runnable
1247 struct task_struct * (*pick_next_task) (struct rq *rq,
1248 struct task_struct *prev);
1249 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1252 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1253 void (*migrate_task_rq)(struct task_struct *p);
1255 void (*task_waking) (struct task_struct *task);
1256 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1258 void (*set_cpus_allowed)(struct task_struct *p,
1259 const struct cpumask *newmask);
1261 void (*rq_online)(struct rq *rq);
1262 void (*rq_offline)(struct rq *rq);
1265 void (*set_curr_task) (struct rq *rq);
1266 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1267 void (*task_fork) (struct task_struct *p);
1268 void (*task_dead) (struct task_struct *p);
1271 * The switched_from() call is allowed to drop rq->lock, therefore we
1272 * cannot assume the switched_from/switched_to pair is serliazed by
1273 * rq->lock. They are however serialized by p->pi_lock.
1275 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1276 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1277 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1280 unsigned int (*get_rr_interval) (struct rq *rq,
1281 struct task_struct *task);
1283 void (*update_curr) (struct rq *rq);
1285 #ifdef CONFIG_FAIR_GROUP_SCHED
1286 void (*task_move_group) (struct task_struct *p);
1290 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1292 prev->sched_class->put_prev_task(rq, prev);
1295 #define sched_class_highest (&stop_sched_class)
1296 #define for_each_class(class) \
1297 for (class = sched_class_highest; class; class = class->next)
1299 extern const struct sched_class stop_sched_class;
1300 extern const struct sched_class dl_sched_class;
1301 extern const struct sched_class rt_sched_class;
1302 extern const struct sched_class fair_sched_class;
1303 extern const struct sched_class idle_sched_class;
1308 extern void init_max_cpu_capacity(struct max_cpu_capacity *mcc);
1309 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1311 extern void trigger_load_balance(struct rq *rq);
1313 extern void idle_enter_fair(struct rq *this_rq);
1314 extern void idle_exit_fair(struct rq *this_rq);
1316 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1320 static inline void idle_enter_fair(struct rq *rq) { }
1321 static inline void idle_exit_fair(struct rq *rq) { }
1325 #ifdef CONFIG_CPU_IDLE
1326 static inline void idle_set_state(struct rq *rq,
1327 struct cpuidle_state *idle_state)
1329 rq->idle_state = idle_state;
1332 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1334 WARN_ON(!rcu_read_lock_held());
1335 return rq->idle_state;
1338 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1340 rq->idle_state_idx = idle_state_idx;
1343 static inline int idle_get_state_idx(struct rq *rq)
1345 WARN_ON(!rcu_read_lock_held());
1346 return rq->idle_state_idx;
1349 static inline void idle_set_state(struct rq *rq,
1350 struct cpuidle_state *idle_state)
1354 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1359 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1363 static inline int idle_get_state_idx(struct rq *rq)
1369 extern void sysrq_sched_debug_show(void);
1370 extern void sched_init_granularity(void);
1371 extern void update_max_interval(void);
1373 extern void init_sched_dl_class(void);
1374 extern void init_sched_rt_class(void);
1375 extern void init_sched_fair_class(void);
1377 extern void resched_curr(struct rq *rq);
1378 extern void resched_cpu(int cpu);
1380 extern struct rt_bandwidth def_rt_bandwidth;
1381 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1383 extern struct dl_bandwidth def_dl_bandwidth;
1384 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1385 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1387 unsigned long to_ratio(u64 period, u64 runtime);
1389 extern void init_entity_runnable_average(struct sched_entity *se);
1391 static inline void __add_nr_running(struct rq *rq, unsigned count)
1393 unsigned prev_nr = rq->nr_running;
1395 rq->nr_running = prev_nr + count;
1397 if (prev_nr < 2 && rq->nr_running >= 2) {
1399 if (!rq->rd->overload)
1400 rq->rd->overload = true;
1403 #ifdef CONFIG_NO_HZ_FULL
1404 if (tick_nohz_full_cpu(rq->cpu)) {
1406 * Tick is needed if more than one task runs on a CPU.
1407 * Send the target an IPI to kick it out of nohz mode.
1409 * We assume that IPI implies full memory barrier and the
1410 * new value of rq->nr_running is visible on reception
1413 tick_nohz_full_kick_cpu(rq->cpu);
1419 static inline void __sub_nr_running(struct rq *rq, unsigned count)
1421 rq->nr_running -= count;
1424 #ifdef CONFIG_CPU_QUIET
1425 #define NR_AVE_SCALE(x) ((x) << FSHIFT)
1426 static inline u64 do_nr_running_integral(struct rq *rq)
1429 u64 nr_running_integral = rq->nr_running_integral;
1431 deltax = rq->clock_task - rq->nr_last_stamp;
1432 nr = NR_AVE_SCALE(rq->nr_running);
1434 nr_running_integral += nr * deltax;
1436 return nr_running_integral;
1439 static inline void add_nr_running(struct rq *rq, unsigned count)
1441 write_seqcount_begin(&rq->ave_seqcnt);
1442 rq->nr_running_integral = do_nr_running_integral(rq);
1443 rq->nr_last_stamp = rq->clock_task;
1444 __add_nr_running(rq, count);
1445 write_seqcount_end(&rq->ave_seqcnt);
1448 static inline void sub_nr_running(struct rq *rq, unsigned count)
1450 write_seqcount_begin(&rq->ave_seqcnt);
1451 rq->nr_running_integral = do_nr_running_integral(rq);
1452 rq->nr_last_stamp = rq->clock_task;
1453 __sub_nr_running(rq, count);
1454 write_seqcount_end(&rq->ave_seqcnt);
1457 #define add_nr_running __add_nr_running
1458 #define sub_nr_running __sub_nr_running
1461 static inline void rq_last_tick_reset(struct rq *rq)
1463 #ifdef CONFIG_NO_HZ_FULL
1464 rq->last_sched_tick = jiffies;
1468 extern void update_rq_clock(struct rq *rq);
1470 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1471 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1473 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1475 extern const_debug unsigned int sysctl_sched_time_avg;
1476 extern const_debug unsigned int sysctl_sched_nr_migrate;
1477 extern const_debug unsigned int sysctl_sched_migration_cost;
1479 static inline u64 sched_avg_period(void)
1481 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1484 #ifdef CONFIG_SCHED_HRTICK
1488 * - enabled by features
1489 * - hrtimer is actually high res
1491 static inline int hrtick_enabled(struct rq *rq)
1493 if (!sched_feat(HRTICK))
1495 if (!cpu_active(cpu_of(rq)))
1497 return hrtimer_is_hres_active(&rq->hrtick_timer);
1500 void hrtick_start(struct rq *rq, u64 delay);
1504 static inline int hrtick_enabled(struct rq *rq)
1509 #endif /* CONFIG_SCHED_HRTICK */
1512 extern void sched_avg_update(struct rq *rq);
1514 #ifndef arch_scale_freq_capacity
1515 static __always_inline
1516 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1518 return SCHED_CAPACITY_SCALE;
1522 #ifndef arch_scale_cpu_capacity
1523 static __always_inline
1524 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1526 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1527 return sd->smt_gain / sd->span_weight;
1529 return SCHED_CAPACITY_SCALE;
1534 static inline unsigned long capacity_of(int cpu)
1536 return cpu_rq(cpu)->cpu_capacity;
1539 static inline unsigned long capacity_orig_of(int cpu)
1541 return cpu_rq(cpu)->cpu_capacity_orig;
1544 extern unsigned int sysctl_sched_use_walt_cpu_util;
1545 extern unsigned int walt_ravg_window;
1546 extern unsigned int walt_disabled;
1549 * cpu_util returns the amount of capacity of a CPU that is used by CFS
1550 * tasks. The unit of the return value must be the one of capacity so we can
1551 * compare the utilization with the capacity of the CPU that is available for
1552 * CFS task (ie cpu_capacity).
1554 * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the
1555 * recent utilization of currently non-runnable tasks on a CPU. It represents
1556 * the amount of utilization of a CPU in the range [0..capacity_orig] where
1557 * capacity_orig is the cpu_capacity available at the highest frequency
1558 * (arch_scale_freq_capacity()).
1559 * The utilization of a CPU converges towards a sum equal to or less than the
1560 * current capacity (capacity_curr <= capacity_orig) of the CPU because it is
1561 * the running time on this CPU scaled by capacity_curr.
1563 * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even
1564 * higher than capacity_orig because of unfortunate rounding in
1565 * cfs.avg.util_avg or just after migrating tasks and new task wakeups until
1566 * the average stabilizes with the new running time. We need to check that the
1567 * utilization stays within the range of [0..capacity_orig] and cap it if
1568 * necessary. Without utilization capping, a group could be seen as overloaded
1569 * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of
1570 * available capacity. We allow utilization to overshoot capacity_curr (but not
1571 * capacity_orig) as it useful for predicting the capacity required after task
1572 * migrations (scheduler-driven DVFS).
1574 static inline unsigned long __cpu_util(int cpu, int delta)
1576 unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
1577 unsigned long capacity = capacity_orig_of(cpu);
1579 #ifdef CONFIG_SCHED_WALT
1580 if (!walt_disabled && sysctl_sched_use_walt_cpu_util) {
1581 util = cpu_rq(cpu)->prev_runnable_sum << SCHED_LOAD_SHIFT;
1582 do_div(util, walt_ravg_window);
1589 return (delta >= capacity) ? capacity : delta;
1592 static inline unsigned long cpu_util(int cpu)
1594 return __cpu_util(cpu, 0);
1599 #ifdef CONFIG_CPU_FREQ_GOV_SCHED
1600 #define capacity_max SCHED_CAPACITY_SCALE
1601 extern unsigned int capacity_margin;
1602 extern struct static_key __sched_freq;
1604 static inline bool sched_freq(void)
1606 return static_key_false(&__sched_freq);
1609 DECLARE_PER_CPU(struct sched_capacity_reqs, cpu_sched_capacity_reqs);
1610 void update_cpu_capacity_request(int cpu, bool request);
1612 static inline void set_cfs_cpu_capacity(int cpu, bool request,
1613 unsigned long capacity)
1615 struct sched_capacity_reqs *scr = &per_cpu(cpu_sched_capacity_reqs, cpu);
1617 #ifdef CONFIG_SCHED_WALT
1618 if (!walt_disabled && sysctl_sched_use_walt_cpu_util) {
1619 int rtdl = scr->rt + scr->dl;
1621 * WALT tracks the utilization of a CPU considering the load
1622 * generated by all the scheduling classes.
1623 * Since the following call to:
1624 * update_cpu_capacity
1625 * is already adding the RT and DL utilizations let's remove
1626 * these contributions from the WALT signal.
1628 if (capacity > rtdl)
1634 if (scr->cfs != capacity) {
1635 scr->cfs = capacity;
1636 update_cpu_capacity_request(cpu, request);
1640 static inline void set_rt_cpu_capacity(int cpu, bool request,
1641 unsigned long capacity)
1643 if (per_cpu(cpu_sched_capacity_reqs, cpu).rt != capacity) {
1644 per_cpu(cpu_sched_capacity_reqs, cpu).rt = capacity;
1645 update_cpu_capacity_request(cpu, request);
1649 static inline void set_dl_cpu_capacity(int cpu, bool request,
1650 unsigned long capacity)
1652 if (per_cpu(cpu_sched_capacity_reqs, cpu).dl != capacity) {
1653 per_cpu(cpu_sched_capacity_reqs, cpu).dl = capacity;
1654 update_cpu_capacity_request(cpu, request);
1658 static inline bool sched_freq(void) { return false; }
1659 static inline void set_cfs_cpu_capacity(int cpu, bool request,
1660 unsigned long capacity)
1662 static inline void set_rt_cpu_capacity(int cpu, bool request,
1663 unsigned long capacity)
1665 static inline void set_dl_cpu_capacity(int cpu, bool request,
1666 unsigned long capacity)
1670 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1672 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1675 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1676 static inline void sched_avg_update(struct rq *rq) { }
1680 * __task_rq_lock - lock the rq @p resides on.
1682 static inline struct rq *__task_rq_lock(struct task_struct *p)
1683 __acquires(rq->lock)
1687 lockdep_assert_held(&p->pi_lock);
1691 raw_spin_lock(&rq->lock);
1692 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1693 lockdep_pin_lock(&rq->lock);
1696 raw_spin_unlock(&rq->lock);
1698 while (unlikely(task_on_rq_migrating(p)))
1704 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1706 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1707 __acquires(p->pi_lock)
1708 __acquires(rq->lock)
1713 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1715 raw_spin_lock(&rq->lock);
1717 * move_queued_task() task_rq_lock()
1719 * ACQUIRE (rq->lock)
1720 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1721 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1722 * [S] ->cpu = new_cpu [L] task_rq()
1724 * RELEASE (rq->lock)
1726 * If we observe the old cpu in task_rq_lock, the acquire of
1727 * the old rq->lock will fully serialize against the stores.
1729 * If we observe the new cpu in task_rq_lock, the acquire will
1730 * pair with the WMB to ensure we must then also see migrating.
1732 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1733 lockdep_pin_lock(&rq->lock);
1736 raw_spin_unlock(&rq->lock);
1737 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1739 while (unlikely(task_on_rq_migrating(p)))
1744 static inline void __task_rq_unlock(struct rq *rq)
1745 __releases(rq->lock)
1747 lockdep_unpin_lock(&rq->lock);
1748 raw_spin_unlock(&rq->lock);
1752 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1753 __releases(rq->lock)
1754 __releases(p->pi_lock)
1756 lockdep_unpin_lock(&rq->lock);
1757 raw_spin_unlock(&rq->lock);
1758 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1761 extern struct rq *lock_rq_of(struct task_struct *p, unsigned long *flags);
1762 extern void unlock_rq_of(struct rq *rq, struct task_struct *p, unsigned long *flags);
1765 #ifdef CONFIG_PREEMPT
1767 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1770 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1771 * way at the expense of forcing extra atomic operations in all
1772 * invocations. This assures that the double_lock is acquired using the
1773 * same underlying policy as the spinlock_t on this architecture, which
1774 * reduces latency compared to the unfair variant below. However, it
1775 * also adds more overhead and therefore may reduce throughput.
1777 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1778 __releases(this_rq->lock)
1779 __acquires(busiest->lock)
1780 __acquires(this_rq->lock)
1782 raw_spin_unlock(&this_rq->lock);
1783 double_rq_lock(this_rq, busiest);
1790 * Unfair double_lock_balance: Optimizes throughput at the expense of
1791 * latency by eliminating extra atomic operations when the locks are
1792 * already in proper order on entry. This favors lower cpu-ids and will
1793 * grant the double lock to lower cpus over higher ids under contention,
1794 * regardless of entry order into the function.
1796 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1797 __releases(this_rq->lock)
1798 __acquires(busiest->lock)
1799 __acquires(this_rq->lock)
1803 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1804 if (busiest < this_rq) {
1805 raw_spin_unlock(&this_rq->lock);
1806 raw_spin_lock(&busiest->lock);
1807 raw_spin_lock_nested(&this_rq->lock,
1808 SINGLE_DEPTH_NESTING);
1811 raw_spin_lock_nested(&busiest->lock,
1812 SINGLE_DEPTH_NESTING);
1817 #endif /* CONFIG_PREEMPT */
1820 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1822 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1824 if (unlikely(!irqs_disabled())) {
1825 /* printk() doesn't work good under rq->lock */
1826 raw_spin_unlock(&this_rq->lock);
1830 return _double_lock_balance(this_rq, busiest);
1833 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1834 __releases(busiest->lock)
1836 if (this_rq != busiest)
1837 raw_spin_unlock(&busiest->lock);
1838 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1841 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1847 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1850 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1856 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1859 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1865 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1869 * double_rq_lock - safely lock two runqueues
1871 * Note this does not disable interrupts like task_rq_lock,
1872 * you need to do so manually before calling.
1874 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1875 __acquires(rq1->lock)
1876 __acquires(rq2->lock)
1878 BUG_ON(!irqs_disabled());
1880 raw_spin_lock(&rq1->lock);
1881 __acquire(rq2->lock); /* Fake it out ;) */
1884 raw_spin_lock(&rq1->lock);
1885 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1887 raw_spin_lock(&rq2->lock);
1888 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1894 * double_rq_unlock - safely unlock two runqueues
1896 * Note this does not restore interrupts like task_rq_unlock,
1897 * you need to do so manually after calling.
1899 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1900 __releases(rq1->lock)
1901 __releases(rq2->lock)
1903 raw_spin_unlock(&rq1->lock);
1905 raw_spin_unlock(&rq2->lock);
1907 __release(rq2->lock);
1910 #else /* CONFIG_SMP */
1913 * double_rq_lock - safely lock two runqueues
1915 * Note this does not disable interrupts like task_rq_lock,
1916 * you need to do so manually before calling.
1918 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1919 __acquires(rq1->lock)
1920 __acquires(rq2->lock)
1922 BUG_ON(!irqs_disabled());
1924 raw_spin_lock(&rq1->lock);
1925 __acquire(rq2->lock); /* Fake it out ;) */
1929 * double_rq_unlock - safely unlock two runqueues
1931 * Note this does not restore interrupts like task_rq_unlock,
1932 * you need to do so manually after calling.
1934 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1935 __releases(rq1->lock)
1936 __releases(rq2->lock)
1939 raw_spin_unlock(&rq1->lock);
1940 __release(rq2->lock);
1945 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1946 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1948 #ifdef CONFIG_SCHED_DEBUG
1949 extern void print_cfs_stats(struct seq_file *m, int cpu);
1950 extern void print_rt_stats(struct seq_file *m, int cpu);
1951 extern void print_dl_stats(struct seq_file *m, int cpu);
1953 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1955 #ifdef CONFIG_NUMA_BALANCING
1957 show_numa_stats(struct task_struct *p, struct seq_file *m);
1959 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1960 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1961 #endif /* CONFIG_NUMA_BALANCING */
1962 #endif /* CONFIG_SCHED_DEBUG */
1964 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1965 extern void init_rt_rq(struct rt_rq *rt_rq);
1966 extern void init_dl_rq(struct dl_rq *dl_rq);
1968 extern void cfs_bandwidth_usage_inc(void);
1969 extern void cfs_bandwidth_usage_dec(void);
1971 #ifdef CONFIG_NO_HZ_COMMON
1972 enum rq_nohz_flag_bits {
1977 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1980 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1982 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1983 DECLARE_PER_CPU(u64, cpu_softirq_time);
1985 #ifndef CONFIG_64BIT
1986 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1988 static inline void irq_time_write_begin(void)
1990 __this_cpu_inc(irq_time_seq.sequence);
1994 static inline void irq_time_write_end(void)
1997 __this_cpu_inc(irq_time_seq.sequence);
2000 static inline u64 irq_time_read(int cpu)
2006 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
2007 irq_time = per_cpu(cpu_softirq_time, cpu) +
2008 per_cpu(cpu_hardirq_time, cpu);
2009 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
2013 #else /* CONFIG_64BIT */
2014 static inline void irq_time_write_begin(void)
2018 static inline void irq_time_write_end(void)
2022 static inline u64 irq_time_read(int cpu)
2024 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
2026 #endif /* CONFIG_64BIT */
2027 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2029 static inline void account_reset_rq(struct rq *rq)
2031 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2032 rq->prev_irq_time = 0;
2034 #ifdef CONFIG_PARAVIRT
2035 rq->prev_steal_time = 0;
2037 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
2038 rq->prev_steal_time_rq = 0;