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_CFS_BANDWIDTH
416 s64 runtime_remaining;
418 u64 throttled_clock, throttled_clock_task;
419 u64 throttled_clock_task_time;
420 int throttled, throttle_count;
421 struct list_head throttled_list;
422 #endif /* CONFIG_CFS_BANDWIDTH */
423 #endif /* CONFIG_FAIR_GROUP_SCHED */
426 static inline int rt_bandwidth_enabled(void)
428 return sysctl_sched_rt_runtime >= 0;
431 /* RT IPI pull logic requires IRQ_WORK */
432 #ifdef CONFIG_IRQ_WORK
433 # define HAVE_RT_PUSH_IPI
436 /* Real-Time classes' related field in a runqueue: */
438 struct rt_prio_array active;
439 unsigned int rt_nr_running;
440 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
442 int curr; /* highest queued rt task prio */
444 int next; /* next highest */
449 unsigned long rt_nr_migratory;
450 unsigned long rt_nr_total;
452 struct plist_head pushable_tasks;
453 #ifdef HAVE_RT_PUSH_IPI
456 struct irq_work push_work;
457 raw_spinlock_t push_lock;
459 #endif /* CONFIG_SMP */
465 /* Nests inside the rq lock: */
466 raw_spinlock_t rt_runtime_lock;
468 #ifdef CONFIG_RT_GROUP_SCHED
469 unsigned long rt_nr_boosted;
472 struct task_group *tg;
476 /* Deadline class' related fields in a runqueue */
478 /* runqueue is an rbtree, ordered by deadline */
479 struct rb_root rb_root;
480 struct rb_node *rb_leftmost;
482 unsigned long dl_nr_running;
486 * Deadline values of the currently executing and the
487 * earliest ready task on this rq. Caching these facilitates
488 * the decision wether or not a ready but not running task
489 * should migrate somewhere else.
496 unsigned long dl_nr_migratory;
500 * Tasks on this rq that can be pushed away. They are kept in
501 * an rb-tree, ordered by tasks' deadlines, with caching
502 * of the leftmost (earliest deadline) element.
504 struct rb_root pushable_dl_tasks_root;
505 struct rb_node *pushable_dl_tasks_leftmost;
513 struct max_cpu_capacity {
520 * We add the notion of a root-domain which will be used to define per-domain
521 * variables. Each exclusive cpuset essentially defines an island domain by
522 * fully partitioning the member cpus from any other cpuset. Whenever a new
523 * exclusive cpuset is created, we also create and attach a new root-domain
532 cpumask_var_t online;
534 /* Indicate more than one runnable task for any CPU */
537 /* Indicate one or more cpus over-utilized (tipping point) */
541 * The bit corresponding to a CPU gets set here if such CPU has more
542 * than one runnable -deadline task (as it is below for RT tasks).
544 cpumask_var_t dlo_mask;
550 * The "RT overload" flag: it gets set if a CPU has more than
551 * one runnable RT task.
553 cpumask_var_t rto_mask;
554 struct cpupri cpupri;
556 /* Maximum cpu capacity in the system. */
557 struct max_cpu_capacity max_cpu_capacity;
560 extern struct root_domain def_root_domain;
562 #endif /* CONFIG_SMP */
565 * This is the main, per-CPU runqueue data structure.
567 * Locking rule: those places that want to lock multiple runqueues
568 * (such as the load balancing or the thread migration code), lock
569 * acquire operations must be ordered by ascending &runqueue.
576 * nr_running and cpu_load should be in the same cacheline because
577 * remote CPUs use both these fields when doing load calculation.
579 unsigned int nr_running;
580 #ifdef CONFIG_NUMA_BALANCING
581 unsigned int nr_numa_running;
582 unsigned int nr_preferred_running;
584 #define CPU_LOAD_IDX_MAX 5
585 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
586 unsigned long last_load_update_tick;
587 unsigned int misfit_task;
588 #ifdef CONFIG_NO_HZ_COMMON
590 unsigned long nohz_flags;
592 #ifdef CONFIG_NO_HZ_FULL
593 unsigned long last_sched_tick;
595 /* capture load from *all* tasks on this cpu: */
596 struct load_weight load;
597 unsigned long nr_load_updates;
604 #ifdef CONFIG_FAIR_GROUP_SCHED
605 /* list of leaf cfs_rq on this cpu: */
606 struct list_head leaf_cfs_rq_list;
607 #endif /* CONFIG_FAIR_GROUP_SCHED */
610 * This is part of a global counter where only the total sum
611 * over all CPUs matters. A task can increase this counter on
612 * one CPU and if it got migrated afterwards it may decrease
613 * it on another CPU. Always updated under the runqueue lock:
615 unsigned long nr_uninterruptible;
617 struct task_struct *curr, *idle, *stop;
618 unsigned long next_balance;
619 struct mm_struct *prev_mm;
621 unsigned int clock_skip_update;
628 struct root_domain *rd;
629 struct sched_domain *sd;
631 unsigned long cpu_capacity;
632 unsigned long cpu_capacity_orig;
634 struct callback_head *balance_callback;
636 unsigned char idle_balance;
637 /* For active balancing */
640 struct cpu_stop_work active_balance_work;
641 /* cpu of this runqueue: */
645 struct list_head cfs_tasks;
652 /* This is used to determine avg_idle's max value */
653 u64 max_idle_balance_cost;
656 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
659 #ifdef CONFIG_PARAVIRT
662 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
663 u64 prev_steal_time_rq;
666 /* calc_load related fields */
667 unsigned long calc_load_update;
668 long calc_load_active;
670 #ifdef CONFIG_SCHED_HRTICK
672 int hrtick_csd_pending;
673 struct call_single_data hrtick_csd;
675 struct hrtimer hrtick_timer;
678 #ifdef CONFIG_SCHEDSTATS
680 struct sched_info rq_sched_info;
681 unsigned long long rq_cpu_time;
682 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
684 /* sys_sched_yield() stats */
685 unsigned int yld_count;
687 /* schedule() stats */
688 unsigned int sched_count;
689 unsigned int sched_goidle;
691 /* try_to_wake_up() stats */
692 unsigned int ttwu_count;
693 unsigned int ttwu_local;
697 struct llist_head wake_list;
700 #ifdef CONFIG_CPU_IDLE
701 /* Must be inspected within a rcu lock section */
702 struct cpuidle_state *idle_state;
707 static inline int cpu_of(struct rq *rq)
716 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
718 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
719 #define this_rq() this_cpu_ptr(&runqueues)
720 #define task_rq(p) cpu_rq(task_cpu(p))
721 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
722 #define raw_rq() raw_cpu_ptr(&runqueues)
724 static inline u64 __rq_clock_broken(struct rq *rq)
726 return READ_ONCE(rq->clock);
729 static inline u64 rq_clock(struct rq *rq)
731 lockdep_assert_held(&rq->lock);
735 static inline u64 rq_clock_task(struct rq *rq)
737 lockdep_assert_held(&rq->lock);
738 return rq->clock_task;
741 #define RQCF_REQ_SKIP 0x01
742 #define RQCF_ACT_SKIP 0x02
744 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
746 lockdep_assert_held(&rq->lock);
748 rq->clock_skip_update |= RQCF_REQ_SKIP;
750 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
754 enum numa_topology_type {
759 extern enum numa_topology_type sched_numa_topology_type;
760 extern int sched_max_numa_distance;
761 extern bool find_numa_distance(int distance);
764 #ifdef CONFIG_NUMA_BALANCING
765 /* The regions in numa_faults array from task_struct */
766 enum numa_faults_stats {
772 extern void sched_setnuma(struct task_struct *p, int node);
773 extern int migrate_task_to(struct task_struct *p, int cpu);
774 extern int migrate_swap(struct task_struct *, struct task_struct *);
775 #endif /* CONFIG_NUMA_BALANCING */
780 queue_balance_callback(struct rq *rq,
781 struct callback_head *head,
782 void (*func)(struct rq *rq))
784 lockdep_assert_held(&rq->lock);
786 if (unlikely(head->next))
789 head->func = (void (*)(struct callback_head *))func;
790 head->next = rq->balance_callback;
791 rq->balance_callback = head;
794 extern void sched_ttwu_pending(void);
796 #define rcu_dereference_check_sched_domain(p) \
797 rcu_dereference_check((p), \
798 lockdep_is_held(&sched_domains_mutex))
801 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
802 * See detach_destroy_domains: synchronize_sched for details.
804 * The domain tree of any CPU may only be accessed from within
805 * preempt-disabled sections.
807 #define for_each_domain(cpu, __sd) \
808 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
809 __sd; __sd = __sd->parent)
811 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
814 * highest_flag_domain - Return highest sched_domain containing flag.
815 * @cpu: The cpu whose highest level of sched domain is to
817 * @flag: The flag to check for the highest sched_domain
820 * Returns the highest sched_domain of a cpu which contains the given flag.
822 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
824 struct sched_domain *sd, *hsd = NULL;
826 for_each_domain(cpu, sd) {
827 if (!(sd->flags & flag))
835 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
837 struct sched_domain *sd;
839 for_each_domain(cpu, sd) {
840 if (sd->flags & flag)
847 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
848 DECLARE_PER_CPU(int, sd_llc_size);
849 DECLARE_PER_CPU(int, sd_llc_id);
850 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
851 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
852 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
853 DECLARE_PER_CPU(struct sched_domain *, sd_ea);
854 DECLARE_PER_CPU(struct sched_domain *, sd_scs);
856 struct sched_group_capacity {
859 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
862 unsigned long capacity;
863 unsigned long max_capacity; /* Max per-cpu capacity in group */
864 unsigned long next_update;
865 int imbalance; /* XXX unrelated to capacity but shared group state */
867 * Number of busy cpus in this group.
869 atomic_t nr_busy_cpus;
871 unsigned long cpumask[0]; /* iteration mask */
875 struct sched_group *next; /* Must be a circular list */
878 unsigned int group_weight;
879 struct sched_group_capacity *sgc;
880 const struct sched_group_energy const *sge;
883 * The CPUs this group covers.
885 * NOTE: this field is variable length. (Allocated dynamically
886 * by attaching extra space to the end of the structure,
887 * depending on how many CPUs the kernel has booted up with)
889 unsigned long cpumask[0];
892 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
894 return to_cpumask(sg->cpumask);
898 * cpumask masking which cpus in the group are allowed to iterate up the domain
901 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
903 return to_cpumask(sg->sgc->cpumask);
907 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
908 * @group: The group whose first cpu is to be returned.
910 static inline unsigned int group_first_cpu(struct sched_group *group)
912 return cpumask_first(sched_group_cpus(group));
915 extern int group_balance_cpu(struct sched_group *sg);
919 static inline void sched_ttwu_pending(void) { }
921 #endif /* CONFIG_SMP */
924 #include "auto_group.h"
926 #ifdef CONFIG_CGROUP_SCHED
929 * Return the group to which this tasks belongs.
931 * We cannot use task_css() and friends because the cgroup subsystem
932 * changes that value before the cgroup_subsys::attach() method is called,
933 * therefore we cannot pin it and might observe the wrong value.
935 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
936 * core changes this before calling sched_move_task().
938 * Instead we use a 'copy' which is updated from sched_move_task() while
939 * holding both task_struct::pi_lock and rq::lock.
941 static inline struct task_group *task_group(struct task_struct *p)
943 return p->sched_task_group;
946 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
947 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
949 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
950 struct task_group *tg = task_group(p);
953 #ifdef CONFIG_FAIR_GROUP_SCHED
954 p->se.cfs_rq = tg->cfs_rq[cpu];
955 p->se.parent = tg->se[cpu];
958 #ifdef CONFIG_RT_GROUP_SCHED
959 p->rt.rt_rq = tg->rt_rq[cpu];
960 p->rt.parent = tg->rt_se[cpu];
964 #else /* CONFIG_CGROUP_SCHED */
966 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
967 static inline struct task_group *task_group(struct task_struct *p)
972 #endif /* CONFIG_CGROUP_SCHED */
974 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
979 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
980 * successfuly executed on another CPU. We must ensure that updates of
981 * per-task data have been completed by this moment.
984 task_thread_info(p)->cpu = cpu;
990 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
992 #ifdef CONFIG_SCHED_DEBUG
993 # include <linux/static_key.h>
994 # define const_debug __read_mostly
996 # define const_debug const
999 extern const_debug unsigned int sysctl_sched_features;
1001 #define SCHED_FEAT(name, enabled) \
1002 __SCHED_FEAT_##name ,
1005 #include "features.h"
1011 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1012 #define SCHED_FEAT(name, enabled) \
1013 static __always_inline bool static_branch_##name(struct static_key *key) \
1015 return static_key_##enabled(key); \
1018 #include "features.h"
1022 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1023 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1024 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1025 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1026 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1028 extern struct static_key_false sched_numa_balancing;
1030 static inline u64 global_rt_period(void)
1032 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1035 static inline u64 global_rt_runtime(void)
1037 if (sysctl_sched_rt_runtime < 0)
1040 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1043 static inline int task_current(struct rq *rq, struct task_struct *p)
1045 return rq->curr == p;
1048 static inline int task_running(struct rq *rq, struct task_struct *p)
1053 return task_current(rq, p);
1057 static inline int task_on_rq_queued(struct task_struct *p)
1059 return p->on_rq == TASK_ON_RQ_QUEUED;
1062 static inline int task_on_rq_migrating(struct task_struct *p)
1064 return p->on_rq == TASK_ON_RQ_MIGRATING;
1067 #ifndef prepare_arch_switch
1068 # define prepare_arch_switch(next) do { } while (0)
1070 #ifndef finish_arch_post_lock_switch
1071 # define finish_arch_post_lock_switch() do { } while (0)
1074 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1078 * We can optimise this out completely for !SMP, because the
1079 * SMP rebalancing from interrupt is the only thing that cares
1086 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1090 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1091 * We must ensure this doesn't happen until the switch is completely
1094 * In particular, the load of prev->state in finish_task_switch() must
1095 * happen before this.
1097 * Pairs with the control dependency and rmb in try_to_wake_up().
1099 smp_store_release(&prev->on_cpu, 0);
1101 #ifdef CONFIG_DEBUG_SPINLOCK
1102 /* this is a valid case when another task releases the spinlock */
1103 rq->lock.owner = current;
1106 * If we are tracking spinlock dependencies then we have to
1107 * fix up the runqueue lock - which gets 'carried over' from
1108 * prev into current:
1110 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1112 raw_spin_unlock_irq(&rq->lock);
1118 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1119 #define WF_FORK 0x02 /* child wakeup after fork */
1120 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1123 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1124 * of tasks with abnormal "nice" values across CPUs the contribution that
1125 * each task makes to its run queue's load is weighted according to its
1126 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1127 * scaled version of the new time slice allocation that they receive on time
1131 #define WEIGHT_IDLEPRIO 3
1132 #define WMULT_IDLEPRIO 1431655765
1135 * Nice levels are multiplicative, with a gentle 10% change for every
1136 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1137 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1138 * that remained on nice 0.
1140 * The "10% effect" is relative and cumulative: from _any_ nice level,
1141 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1142 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1143 * If a task goes up by ~10% and another task goes down by ~10% then
1144 * the relative distance between them is ~25%.)
1146 static const int prio_to_weight[40] = {
1147 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1148 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1149 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1150 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1151 /* 0 */ 1024, 820, 655, 526, 423,
1152 /* 5 */ 335, 272, 215, 172, 137,
1153 /* 10 */ 110, 87, 70, 56, 45,
1154 /* 15 */ 36, 29, 23, 18, 15,
1158 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1160 * In cases where the weight does not change often, we can use the
1161 * precalculated inverse to speed up arithmetics by turning divisions
1162 * into multiplications:
1164 static const u32 prio_to_wmult[40] = {
1165 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1166 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1167 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1168 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1169 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1170 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1171 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1172 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1175 #define ENQUEUE_WAKEUP 0x01
1176 #define ENQUEUE_HEAD 0x02
1178 #define ENQUEUE_WAKING 0x04 /* sched_class::task_waking was called */
1180 #define ENQUEUE_WAKING 0x00
1182 #define ENQUEUE_REPLENISH 0x08
1183 #define ENQUEUE_RESTORE 0x10
1184 #define ENQUEUE_WAKEUP_NEW 0x20
1186 #define DEQUEUE_SLEEP 0x01
1187 #define DEQUEUE_SAVE 0x02
1189 #define RETRY_TASK ((void *)-1UL)
1191 struct sched_class {
1192 const struct sched_class *next;
1194 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1195 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1196 void (*yield_task) (struct rq *rq);
1197 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1199 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1202 * It is the responsibility of the pick_next_task() method that will
1203 * return the next task to call put_prev_task() on the @prev task or
1204 * something equivalent.
1206 * May return RETRY_TASK when it finds a higher prio class has runnable
1209 struct task_struct * (*pick_next_task) (struct rq *rq,
1210 struct task_struct *prev);
1211 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1214 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1215 void (*migrate_task_rq)(struct task_struct *p);
1217 void (*task_waking) (struct task_struct *task);
1218 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1220 void (*set_cpus_allowed)(struct task_struct *p,
1221 const struct cpumask *newmask);
1223 void (*rq_online)(struct rq *rq);
1224 void (*rq_offline)(struct rq *rq);
1227 void (*set_curr_task) (struct rq *rq);
1228 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1229 void (*task_fork) (struct task_struct *p);
1230 void (*task_dead) (struct task_struct *p);
1233 * The switched_from() call is allowed to drop rq->lock, therefore we
1234 * cannot assume the switched_from/switched_to pair is serliazed by
1235 * rq->lock. They are however serialized by p->pi_lock.
1237 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1238 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1239 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1242 unsigned int (*get_rr_interval) (struct rq *rq,
1243 struct task_struct *task);
1245 void (*update_curr) (struct rq *rq);
1247 #ifdef CONFIG_FAIR_GROUP_SCHED
1248 void (*task_move_group) (struct task_struct *p);
1252 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1254 prev->sched_class->put_prev_task(rq, prev);
1257 #define sched_class_highest (&stop_sched_class)
1258 #define for_each_class(class) \
1259 for (class = sched_class_highest; class; class = class->next)
1261 extern const struct sched_class stop_sched_class;
1262 extern const struct sched_class dl_sched_class;
1263 extern const struct sched_class rt_sched_class;
1264 extern const struct sched_class fair_sched_class;
1265 extern const struct sched_class idle_sched_class;
1270 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1272 extern void trigger_load_balance(struct rq *rq);
1274 extern void idle_enter_fair(struct rq *this_rq);
1275 extern void idle_exit_fair(struct rq *this_rq);
1277 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1281 static inline void idle_enter_fair(struct rq *rq) { }
1282 static inline void idle_exit_fair(struct rq *rq) { }
1286 #ifdef CONFIG_CPU_IDLE
1287 static inline void idle_set_state(struct rq *rq,
1288 struct cpuidle_state *idle_state)
1290 rq->idle_state = idle_state;
1293 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1295 WARN_ON(!rcu_read_lock_held());
1296 return rq->idle_state;
1299 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1301 rq->idle_state_idx = idle_state_idx;
1304 static inline int idle_get_state_idx(struct rq *rq)
1306 WARN_ON(!rcu_read_lock_held());
1307 return rq->idle_state_idx;
1310 static inline void idle_set_state(struct rq *rq,
1311 struct cpuidle_state *idle_state)
1315 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1320 static inline void idle_set_state_idx(struct rq *rq, int idle_state_idx)
1324 static inline int idle_get_state_idx(struct rq *rq)
1330 extern void sysrq_sched_debug_show(void);
1331 extern void sched_init_granularity(void);
1332 extern void update_max_interval(void);
1334 extern void init_sched_dl_class(void);
1335 extern void init_sched_rt_class(void);
1336 extern void init_sched_fair_class(void);
1338 extern void resched_curr(struct rq *rq);
1339 extern void resched_cpu(int cpu);
1341 extern struct rt_bandwidth def_rt_bandwidth;
1342 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1344 extern struct dl_bandwidth def_dl_bandwidth;
1345 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1346 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1348 unsigned long to_ratio(u64 period, u64 runtime);
1350 extern void init_entity_runnable_average(struct sched_entity *se);
1352 extern void init_max_cpu_capacity(struct max_cpu_capacity *mcc);
1354 static inline void add_nr_running(struct rq *rq, unsigned count)
1356 unsigned prev_nr = rq->nr_running;
1358 rq->nr_running = prev_nr + count;
1360 if (prev_nr < 2 && rq->nr_running >= 2) {
1362 if (!rq->rd->overload)
1363 rq->rd->overload = true;
1366 #ifdef CONFIG_NO_HZ_FULL
1367 if (tick_nohz_full_cpu(rq->cpu)) {
1369 * Tick is needed if more than one task runs on a CPU.
1370 * Send the target an IPI to kick it out of nohz mode.
1372 * We assume that IPI implies full memory barrier and the
1373 * new value of rq->nr_running is visible on reception
1376 tick_nohz_full_kick_cpu(rq->cpu);
1382 static inline void sub_nr_running(struct rq *rq, unsigned count)
1384 rq->nr_running -= count;
1387 static inline void rq_last_tick_reset(struct rq *rq)
1389 #ifdef CONFIG_NO_HZ_FULL
1390 rq->last_sched_tick = jiffies;
1394 extern void update_rq_clock(struct rq *rq);
1396 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1397 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1399 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1401 extern const_debug unsigned int sysctl_sched_time_avg;
1402 extern const_debug unsigned int sysctl_sched_nr_migrate;
1403 extern const_debug unsigned int sysctl_sched_migration_cost;
1405 static inline u64 sched_avg_period(void)
1407 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1410 #ifdef CONFIG_SCHED_HRTICK
1414 * - enabled by features
1415 * - hrtimer is actually high res
1417 static inline int hrtick_enabled(struct rq *rq)
1419 if (!sched_feat(HRTICK))
1421 if (!cpu_active(cpu_of(rq)))
1423 return hrtimer_is_hres_active(&rq->hrtick_timer);
1426 void hrtick_start(struct rq *rq, u64 delay);
1430 static inline int hrtick_enabled(struct rq *rq)
1435 #endif /* CONFIG_SCHED_HRTICK */
1438 extern void sched_avg_update(struct rq *rq);
1440 #ifndef arch_scale_freq_capacity
1441 static __always_inline
1442 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1444 return SCHED_CAPACITY_SCALE;
1448 #ifndef arch_scale_cpu_capacity
1449 static __always_inline
1450 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1452 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1453 return sd->smt_gain / sd->span_weight;
1455 return SCHED_CAPACITY_SCALE;
1460 static inline unsigned long capacity_of(int cpu)
1462 return cpu_rq(cpu)->cpu_capacity;
1465 static inline unsigned long capacity_orig_of(int cpu)
1467 return cpu_rq(cpu)->cpu_capacity_orig;
1471 * cpu_util returns the amount of capacity of a CPU that is used by CFS
1472 * tasks. The unit of the return value must be the one of capacity so we can
1473 * compare the utilization with the capacity of the CPU that is available for
1474 * CFS task (ie cpu_capacity).
1476 * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the
1477 * recent utilization of currently non-runnable tasks on a CPU. It represents
1478 * the amount of utilization of a CPU in the range [0..capacity_orig] where
1479 * capacity_orig is the cpu_capacity available at the highest frequency
1480 * (arch_scale_freq_capacity()).
1481 * The utilization of a CPU converges towards a sum equal to or less than the
1482 * current capacity (capacity_curr <= capacity_orig) of the CPU because it is
1483 * the running time on this CPU scaled by capacity_curr.
1485 * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even
1486 * higher than capacity_orig because of unfortunate rounding in
1487 * cfs.avg.util_avg or just after migrating tasks and new task wakeups until
1488 * the average stabilizes with the new running time. We need to check that the
1489 * utilization stays within the range of [0..capacity_orig] and cap it if
1490 * necessary. Without utilization capping, a group could be seen as overloaded
1491 * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of
1492 * available capacity. We allow utilization to overshoot capacity_curr (but not
1493 * capacity_orig) as it useful for predicting the capacity required after task
1494 * migrations (scheduler-driven DVFS).
1496 static inline unsigned long __cpu_util(int cpu, int delta)
1498 unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
1499 unsigned long capacity = capacity_orig_of(cpu);
1505 return (delta >= capacity) ? capacity : delta;
1508 static inline unsigned long cpu_util(int cpu)
1510 return __cpu_util(cpu, 0);
1514 * Returns the current capacity of cpu after applying both
1515 * cpu and freq scaling.
1517 static inline unsigned long capacity_curr_of(int cpu)
1519 return cpu_rq(cpu)->cpu_capacity_orig *
1520 arch_scale_freq_capacity(NULL, cpu)
1521 >> SCHED_CAPACITY_SHIFT;
1526 #ifdef CONFIG_CPU_FREQ_GOV_SCHED
1527 #define capacity_max SCHED_CAPACITY_SCALE
1528 extern unsigned int capacity_margin;
1529 extern struct static_key __sched_freq;
1531 static inline bool sched_freq(void)
1533 return static_key_false(&__sched_freq);
1536 DECLARE_PER_CPU(struct sched_capacity_reqs, cpu_sched_capacity_reqs);
1537 void update_cpu_capacity_request(int cpu, bool request);
1539 static inline void set_cfs_cpu_capacity(int cpu, bool request,
1540 unsigned long capacity)
1542 if (per_cpu(cpu_sched_capacity_reqs, cpu).cfs != capacity) {
1543 per_cpu(cpu_sched_capacity_reqs, cpu).cfs = capacity;
1544 update_cpu_capacity_request(cpu, request);
1548 static inline void set_rt_cpu_capacity(int cpu, bool request,
1549 unsigned long capacity)
1551 if (per_cpu(cpu_sched_capacity_reqs, cpu).rt != capacity) {
1552 per_cpu(cpu_sched_capacity_reqs, cpu).rt = capacity;
1553 update_cpu_capacity_request(cpu, request);
1557 static inline void set_dl_cpu_capacity(int cpu, bool request,
1558 unsigned long capacity)
1560 if (per_cpu(cpu_sched_capacity_reqs, cpu).dl != capacity) {
1561 per_cpu(cpu_sched_capacity_reqs, cpu).dl = capacity;
1562 update_cpu_capacity_request(cpu, request);
1566 static inline bool sched_freq(void) { return false; }
1567 static inline void set_cfs_cpu_capacity(int cpu, bool request,
1568 unsigned long capacity)
1570 static inline void set_rt_cpu_capacity(int cpu, bool request,
1571 unsigned long capacity)
1573 static inline void set_dl_cpu_capacity(int cpu, bool request,
1574 unsigned long capacity)
1578 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1580 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1581 sched_avg_update(rq);
1584 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1585 static inline void sched_avg_update(struct rq *rq) { }
1589 * __task_rq_lock - lock the rq @p resides on.
1591 static inline struct rq *__task_rq_lock(struct task_struct *p)
1592 __acquires(rq->lock)
1596 lockdep_assert_held(&p->pi_lock);
1600 raw_spin_lock(&rq->lock);
1601 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1602 lockdep_pin_lock(&rq->lock);
1605 raw_spin_unlock(&rq->lock);
1607 while (unlikely(task_on_rq_migrating(p)))
1613 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1615 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1616 __acquires(p->pi_lock)
1617 __acquires(rq->lock)
1622 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1624 raw_spin_lock(&rq->lock);
1626 * move_queued_task() task_rq_lock()
1628 * ACQUIRE (rq->lock)
1629 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1630 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1631 * [S] ->cpu = new_cpu [L] task_rq()
1633 * RELEASE (rq->lock)
1635 * If we observe the old cpu in task_rq_lock, the acquire of
1636 * the old rq->lock will fully serialize against the stores.
1638 * If we observe the new cpu in task_rq_lock, the acquire will
1639 * pair with the WMB to ensure we must then also see migrating.
1641 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1642 lockdep_pin_lock(&rq->lock);
1645 raw_spin_unlock(&rq->lock);
1646 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1648 while (unlikely(task_on_rq_migrating(p)))
1653 static inline void __task_rq_unlock(struct rq *rq)
1654 __releases(rq->lock)
1656 lockdep_unpin_lock(&rq->lock);
1657 raw_spin_unlock(&rq->lock);
1661 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1662 __releases(rq->lock)
1663 __releases(p->pi_lock)
1665 lockdep_unpin_lock(&rq->lock);
1666 raw_spin_unlock(&rq->lock);
1667 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1671 #ifdef CONFIG_PREEMPT
1673 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1676 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1677 * way at the expense of forcing extra atomic operations in all
1678 * invocations. This assures that the double_lock is acquired using the
1679 * same underlying policy as the spinlock_t on this architecture, which
1680 * reduces latency compared to the unfair variant below. However, it
1681 * also adds more overhead and therefore may reduce throughput.
1683 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1684 __releases(this_rq->lock)
1685 __acquires(busiest->lock)
1686 __acquires(this_rq->lock)
1688 raw_spin_unlock(&this_rq->lock);
1689 double_rq_lock(this_rq, busiest);
1696 * Unfair double_lock_balance: Optimizes throughput at the expense of
1697 * latency by eliminating extra atomic operations when the locks are
1698 * already in proper order on entry. This favors lower cpu-ids and will
1699 * grant the double lock to lower cpus over higher ids under contention,
1700 * regardless of entry order into the function.
1702 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1703 __releases(this_rq->lock)
1704 __acquires(busiest->lock)
1705 __acquires(this_rq->lock)
1709 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1710 if (busiest < this_rq) {
1711 raw_spin_unlock(&this_rq->lock);
1712 raw_spin_lock(&busiest->lock);
1713 raw_spin_lock_nested(&this_rq->lock,
1714 SINGLE_DEPTH_NESTING);
1717 raw_spin_lock_nested(&busiest->lock,
1718 SINGLE_DEPTH_NESTING);
1723 #endif /* CONFIG_PREEMPT */
1726 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1728 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1730 if (unlikely(!irqs_disabled())) {
1731 /* printk() doesn't work good under rq->lock */
1732 raw_spin_unlock(&this_rq->lock);
1736 return _double_lock_balance(this_rq, busiest);
1739 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1740 __releases(busiest->lock)
1742 raw_spin_unlock(&busiest->lock);
1743 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1746 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1752 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1755 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1761 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1764 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1770 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1774 * double_rq_lock - safely lock two runqueues
1776 * Note this does not disable interrupts like task_rq_lock,
1777 * you need to do so manually before calling.
1779 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1780 __acquires(rq1->lock)
1781 __acquires(rq2->lock)
1783 BUG_ON(!irqs_disabled());
1785 raw_spin_lock(&rq1->lock);
1786 __acquire(rq2->lock); /* Fake it out ;) */
1789 raw_spin_lock(&rq1->lock);
1790 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1792 raw_spin_lock(&rq2->lock);
1793 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1799 * double_rq_unlock - safely unlock two runqueues
1801 * Note this does not restore interrupts like task_rq_unlock,
1802 * you need to do so manually after calling.
1804 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1805 __releases(rq1->lock)
1806 __releases(rq2->lock)
1808 raw_spin_unlock(&rq1->lock);
1810 raw_spin_unlock(&rq2->lock);
1812 __release(rq2->lock);
1815 #else /* CONFIG_SMP */
1818 * double_rq_lock - safely lock two runqueues
1820 * Note this does not disable interrupts like task_rq_lock,
1821 * you need to do so manually before calling.
1823 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1824 __acquires(rq1->lock)
1825 __acquires(rq2->lock)
1827 BUG_ON(!irqs_disabled());
1829 raw_spin_lock(&rq1->lock);
1830 __acquire(rq2->lock); /* Fake it out ;) */
1834 * double_rq_unlock - safely unlock two runqueues
1836 * Note this does not restore interrupts like task_rq_unlock,
1837 * you need to do so manually after calling.
1839 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1840 __releases(rq1->lock)
1841 __releases(rq2->lock)
1844 raw_spin_unlock(&rq1->lock);
1845 __release(rq2->lock);
1850 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1851 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1853 #ifdef CONFIG_SCHED_DEBUG
1854 extern void print_cfs_stats(struct seq_file *m, int cpu);
1855 extern void print_rt_stats(struct seq_file *m, int cpu);
1856 extern void print_dl_stats(struct seq_file *m, int cpu);
1858 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1860 #ifdef CONFIG_NUMA_BALANCING
1862 show_numa_stats(struct task_struct *p, struct seq_file *m);
1864 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1865 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1866 #endif /* CONFIG_NUMA_BALANCING */
1867 #endif /* CONFIG_SCHED_DEBUG */
1869 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1870 extern void init_rt_rq(struct rt_rq *rt_rq);
1871 extern void init_dl_rq(struct dl_rq *dl_rq);
1873 extern void cfs_bandwidth_usage_inc(void);
1874 extern void cfs_bandwidth_usage_dec(void);
1876 #ifdef CONFIG_NO_HZ_COMMON
1877 enum rq_nohz_flag_bits {
1882 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1885 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1887 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1888 DECLARE_PER_CPU(u64, cpu_softirq_time);
1890 #ifndef CONFIG_64BIT
1891 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1893 static inline void irq_time_write_begin(void)
1895 __this_cpu_inc(irq_time_seq.sequence);
1899 static inline void irq_time_write_end(void)
1902 __this_cpu_inc(irq_time_seq.sequence);
1905 static inline u64 irq_time_read(int cpu)
1911 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1912 irq_time = per_cpu(cpu_softirq_time, cpu) +
1913 per_cpu(cpu_hardirq_time, cpu);
1914 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1918 #else /* CONFIG_64BIT */
1919 static inline void irq_time_write_begin(void)
1923 static inline void irq_time_write_end(void)
1927 static inline u64 irq_time_read(int cpu)
1929 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1931 #endif /* CONFIG_64BIT */
1932 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1934 static inline void account_reset_rq(struct rq *rq)
1936 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1937 rq->prev_irq_time = 0;
1939 #ifdef CONFIG_PARAVIRT
1940 rq->prev_steal_time = 0;
1942 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1943 rq->prev_steal_time_rq = 0;