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/binfmts.h>
7 #include <linux/mutex.h>
8 #include <linux/spinlock.h>
9 #include <linux/stop_machine.h>
10 #include <linux/irq_work.h>
11 #include <linux/tick.h>
12 #include <linux/slab.h>
15 #include "cpudeadline.h"
21 /* task_struct::on_rq states: */
22 #define TASK_ON_RQ_QUEUED 1
23 #define TASK_ON_RQ_MIGRATING 2
25 extern __read_mostly int scheduler_running;
27 extern unsigned long calc_load_update;
28 extern atomic_long_t calc_load_tasks;
30 extern void calc_global_load_tick(struct rq *this_rq);
31 extern long calc_load_fold_active(struct rq *this_rq, long adjust);
34 extern void cpu_load_update_active(struct rq *this_rq);
36 static inline void cpu_load_update_active(struct rq *this_rq) { }
40 * Helpers for converting nanosecond timing to jiffy resolution
42 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
45 * Increase resolution of nice-level calculations for 64-bit architectures.
46 * The extra resolution improves shares distribution and load balancing of
47 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
48 * hierarchies, especially on larger systems. This is not a user-visible change
49 * and does not change the user-interface for setting shares/weights.
51 * We increase resolution only if we have enough bits to allow this increased
52 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
53 * pretty high and the returns do not justify the increased costs.
55 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
56 * increase coverage and consistency always enable it on 64bit platforms.
59 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
60 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
61 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
63 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
64 # define scale_load(w) (w)
65 # define scale_load_down(w) (w)
69 * Task weight (visible to users) and its load (invisible to users) have
70 * independent resolution, but they should be well calibrated. We use
71 * scale_load() and scale_load_down(w) to convert between them. The
72 * following must be true:
74 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
77 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
80 * Single value that decides SCHED_DEADLINE internal math precision.
81 * 10 -> just above 1us
82 * 9 -> just above 0.5us
87 * These are the 'tuning knobs' of the scheduler:
91 * single value that denotes runtime == period, ie unlimited time.
93 #define RUNTIME_INF ((u64)~0ULL)
95 static inline int idle_policy(int policy)
97 return policy == SCHED_IDLE;
99 static inline int fair_policy(int policy)
101 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
104 static inline int rt_policy(int policy)
106 return policy == SCHED_FIFO || policy == SCHED_RR;
109 static inline int dl_policy(int policy)
111 return policy == SCHED_DEADLINE;
113 static inline bool valid_policy(int policy)
115 return idle_policy(policy) || fair_policy(policy) ||
116 rt_policy(policy) || dl_policy(policy);
119 static inline int task_has_rt_policy(struct task_struct *p)
121 return rt_policy(p->policy);
124 static inline int task_has_dl_policy(struct task_struct *p)
126 return dl_policy(p->policy);
130 * Tells if entity @a should preempt entity @b.
133 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
135 return dl_time_before(a->deadline, b->deadline);
139 * This is the priority-queue data structure of the RT scheduling class:
141 struct rt_prio_array {
142 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
143 struct list_head queue[MAX_RT_PRIO];
146 struct rt_bandwidth {
147 /* nests inside the rq lock: */
148 raw_spinlock_t rt_runtime_lock;
151 struct hrtimer rt_period_timer;
152 unsigned int rt_period_active;
155 void __dl_clear_params(struct task_struct *p);
158 * To keep the bandwidth of -deadline tasks and groups under control
159 * we need some place where:
160 * - store the maximum -deadline bandwidth of the system (the group);
161 * - cache the fraction of that bandwidth that is currently allocated.
163 * This is all done in the data structure below. It is similar to the
164 * one used for RT-throttling (rt_bandwidth), with the main difference
165 * that, since here we are only interested in admission control, we
166 * do not decrease any runtime while the group "executes", neither we
167 * need a timer to replenish it.
169 * With respect to SMP, the bandwidth is given on a per-CPU basis,
171 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
172 * - dl_total_bw array contains, in the i-eth element, the currently
173 * allocated bandwidth on the i-eth CPU.
174 * Moreover, groups consume bandwidth on each CPU, while tasks only
175 * consume bandwidth on the CPU they're running on.
176 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
177 * that will be shown the next time the proc or cgroup controls will
178 * be red. It on its turn can be changed by writing on its own
181 struct dl_bandwidth {
182 raw_spinlock_t dl_runtime_lock;
187 static inline int dl_bandwidth_enabled(void)
189 return sysctl_sched_rt_runtime >= 0;
192 extern struct dl_bw *dl_bw_of(int i);
200 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
202 dl_b->total_bw -= tsk_bw;
206 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
208 dl_b->total_bw += tsk_bw;
212 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
214 return dl_b->bw != -1 &&
215 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
218 extern struct mutex sched_domains_mutex;
220 #ifdef CONFIG_CGROUP_SCHED
222 #include <linux/cgroup.h>
227 extern struct list_head task_groups;
229 struct cfs_bandwidth {
230 #ifdef CONFIG_CFS_BANDWIDTH
234 s64 hierarchical_quota;
237 int idle, period_active;
238 struct hrtimer period_timer, slack_timer;
239 struct list_head throttled_cfs_rq;
242 int nr_periods, nr_throttled;
247 /* task group related information */
249 struct cgroup_subsys_state css;
251 #ifdef CONFIG_FAIR_GROUP_SCHED
252 /* schedulable entities of this group on each cpu */
253 struct sched_entity **se;
254 /* runqueue "owned" by this group on each cpu */
255 struct cfs_rq **cfs_rq;
256 unsigned long shares;
260 * load_avg can be heavily contended at clock tick time, so put
261 * it in its own cacheline separated from the fields above which
262 * will also be accessed at each tick.
264 atomic_long_t load_avg ____cacheline_aligned;
268 #ifdef CONFIG_RT_GROUP_SCHED
269 struct sched_rt_entity **rt_se;
270 struct rt_rq **rt_rq;
272 struct rt_bandwidth rt_bandwidth;
276 struct list_head list;
278 struct task_group *parent;
279 struct list_head siblings;
280 struct list_head children;
282 #ifdef CONFIG_SCHED_AUTOGROUP
283 struct autogroup *autogroup;
286 struct cfs_bandwidth cfs_bandwidth;
289 #ifdef CONFIG_FAIR_GROUP_SCHED
290 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
293 * A weight of 0 or 1 can cause arithmetics problems.
294 * A weight of a cfs_rq is the sum of weights of which entities
295 * are queued on this cfs_rq, so a weight of a entity should not be
296 * too large, so as the shares value of a task group.
297 * (The default weight is 1024 - so there's no practical
298 * limitation from this.)
300 #define MIN_SHARES (1UL << 1)
301 #define MAX_SHARES (1UL << 18)
304 typedef int (*tg_visitor)(struct task_group *, void *);
306 extern int walk_tg_tree_from(struct task_group *from,
307 tg_visitor down, tg_visitor up, void *data);
310 * Iterate the full tree, calling @down when first entering a node and @up when
311 * leaving it for the final time.
313 * Caller must hold rcu_lock or sufficient equivalent.
315 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
317 return walk_tg_tree_from(&root_task_group, down, up, data);
320 extern int tg_nop(struct task_group *tg, void *data);
322 extern void free_fair_sched_group(struct task_group *tg);
323 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
324 extern void online_fair_sched_group(struct task_group *tg);
325 extern void unregister_fair_sched_group(struct task_group *tg);
326 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
327 struct sched_entity *se, int cpu,
328 struct sched_entity *parent);
329 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
331 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
332 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
333 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
335 extern void free_rt_sched_group(struct task_group *tg);
336 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
337 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
338 struct sched_rt_entity *rt_se, int cpu,
339 struct sched_rt_entity *parent);
341 extern struct task_group *sched_create_group(struct task_group *parent);
342 extern void sched_online_group(struct task_group *tg,
343 struct task_group *parent);
344 extern void sched_destroy_group(struct task_group *tg);
345 extern void sched_offline_group(struct task_group *tg);
347 extern void sched_move_task(struct task_struct *tsk);
349 #ifdef CONFIG_FAIR_GROUP_SCHED
350 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
353 extern void set_task_rq_fair(struct sched_entity *se,
354 struct cfs_rq *prev, struct cfs_rq *next);
355 #else /* !CONFIG_SMP */
356 static inline void set_task_rq_fair(struct sched_entity *se,
357 struct cfs_rq *prev, struct cfs_rq *next) { }
358 #endif /* CONFIG_SMP */
359 #endif /* CONFIG_FAIR_GROUP_SCHED */
361 #else /* CONFIG_CGROUP_SCHED */
363 struct cfs_bandwidth { };
365 #endif /* CONFIG_CGROUP_SCHED */
367 /* CFS-related fields in a runqueue */
369 struct load_weight load;
370 unsigned int nr_running, h_nr_running;
375 u64 min_vruntime_copy;
378 struct rb_root tasks_timeline;
379 struct rb_node *rb_leftmost;
382 * 'curr' points to currently running entity on this cfs_rq.
383 * It is set to NULL otherwise (i.e when none are currently running).
385 struct sched_entity *curr, *next, *last, *skip;
387 #ifdef CONFIG_SCHED_DEBUG
388 unsigned int nr_spread_over;
395 struct sched_avg avg;
396 u64 runnable_load_sum;
397 unsigned long runnable_load_avg;
398 #ifdef CONFIG_FAIR_GROUP_SCHED
399 unsigned long tg_load_avg_contrib;
401 atomic_long_t removed_load_avg, removed_util_avg;
403 u64 load_last_update_time_copy;
406 #ifdef CONFIG_FAIR_GROUP_SCHED
408 * h_load = weight * f(tg)
410 * Where f(tg) is the recursive weight fraction assigned to
413 unsigned long h_load;
414 u64 last_h_load_update;
415 struct sched_entity *h_load_next;
416 #endif /* CONFIG_FAIR_GROUP_SCHED */
417 #endif /* CONFIG_SMP */
419 #ifdef CONFIG_FAIR_GROUP_SCHED
420 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
423 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
424 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
425 * (like users, containers etc.)
427 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
428 * list is used during load balance.
431 struct list_head leaf_cfs_rq_list;
432 struct task_group *tg; /* group that "owns" this runqueue */
434 #ifdef CONFIG_CFS_BANDWIDTH
437 s64 runtime_remaining;
439 u64 throttled_clock, throttled_clock_task;
440 u64 throttled_clock_task_time;
441 int throttled, throttle_count;
442 struct list_head throttled_list;
443 #endif /* CONFIG_CFS_BANDWIDTH */
444 #endif /* CONFIG_FAIR_GROUP_SCHED */
447 static inline int rt_bandwidth_enabled(void)
449 return sysctl_sched_rt_runtime >= 0;
452 /* RT IPI pull logic requires IRQ_WORK */
453 #ifdef CONFIG_IRQ_WORK
454 # define HAVE_RT_PUSH_IPI
457 /* Real-Time classes' related field in a runqueue: */
459 struct rt_prio_array active;
460 unsigned int rt_nr_running;
461 unsigned int rr_nr_running;
462 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
464 int curr; /* highest queued rt task prio */
466 int next; /* next highest */
471 unsigned long rt_nr_migratory;
472 unsigned long rt_nr_total;
474 struct plist_head pushable_tasks;
475 #ifdef HAVE_RT_PUSH_IPI
478 struct irq_work push_work;
479 raw_spinlock_t push_lock;
481 #endif /* CONFIG_SMP */
487 /* Nests inside the rq lock: */
488 raw_spinlock_t rt_runtime_lock;
490 #ifdef CONFIG_RT_GROUP_SCHED
491 unsigned long rt_nr_boosted;
494 struct task_group *tg;
498 /* Deadline class' related fields in a runqueue */
500 /* runqueue is an rbtree, ordered by deadline */
501 struct rb_root rb_root;
502 struct rb_node *rb_leftmost;
504 unsigned long dl_nr_running;
508 * Deadline values of the currently executing and the
509 * earliest ready task on this rq. Caching these facilitates
510 * the decision wether or not a ready but not running task
511 * should migrate somewhere else.
518 unsigned long dl_nr_migratory;
522 * Tasks on this rq that can be pushed away. They are kept in
523 * an rb-tree, ordered by tasks' deadlines, with caching
524 * of the leftmost (earliest deadline) element.
526 struct rb_root pushable_dl_tasks_root;
527 struct rb_node *pushable_dl_tasks_leftmost;
536 * We add the notion of a root-domain which will be used to define per-domain
537 * variables. Each exclusive cpuset essentially defines an island domain by
538 * fully partitioning the member cpus from any other cpuset. Whenever a new
539 * exclusive cpuset is created, we also create and attach a new root-domain
548 cpumask_var_t online;
550 /* Indicate more than one runnable task for any CPU */
554 * The bit corresponding to a CPU gets set here if such CPU has more
555 * than one runnable -deadline task (as it is below for RT tasks).
557 cpumask_var_t dlo_mask;
563 * The "RT overload" flag: it gets set if a CPU has more than
564 * one runnable RT task.
566 cpumask_var_t rto_mask;
567 struct cpupri cpupri;
569 unsigned long max_cpu_capacity;
572 extern struct root_domain def_root_domain;
574 #endif /* CONFIG_SMP */
577 * This is the main, per-CPU runqueue data structure.
579 * Locking rule: those places that want to lock multiple runqueues
580 * (such as the load balancing or the thread migration code), lock
581 * acquire operations must be ordered by ascending &runqueue.
588 * nr_running and cpu_load should be in the same cacheline because
589 * remote CPUs use both these fields when doing load calculation.
591 unsigned int nr_running;
592 #ifdef CONFIG_NUMA_BALANCING
593 unsigned int nr_numa_running;
594 unsigned int nr_preferred_running;
596 #define CPU_LOAD_IDX_MAX 5
597 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
598 #ifdef CONFIG_NO_HZ_COMMON
600 unsigned long last_load_update_tick;
601 #endif /* CONFIG_SMP */
602 unsigned long nohz_flags;
603 #endif /* CONFIG_NO_HZ_COMMON */
604 #ifdef CONFIG_NO_HZ_FULL
605 unsigned long last_sched_tick;
607 /* capture load from *all* tasks on this cpu: */
608 struct load_weight load;
609 unsigned long nr_load_updates;
616 #ifdef CONFIG_FAIR_GROUP_SCHED
617 /* list of leaf cfs_rq on this cpu: */
618 struct list_head leaf_cfs_rq_list;
619 #endif /* CONFIG_FAIR_GROUP_SCHED */
622 * This is part of a global counter where only the total sum
623 * over all CPUs matters. A task can increase this counter on
624 * one CPU and if it got migrated afterwards it may decrease
625 * it on another CPU. Always updated under the runqueue lock:
627 unsigned long nr_uninterruptible;
629 struct task_struct *curr, *idle, *stop;
630 unsigned long next_balance;
631 struct mm_struct *prev_mm;
633 unsigned int clock_skip_update;
640 struct root_domain *rd;
641 struct sched_domain *sd;
643 unsigned long cpu_capacity;
644 unsigned long cpu_capacity_orig;
646 struct callback_head *balance_callback;
648 unsigned char idle_balance;
649 /* For active balancing */
652 struct cpu_stop_work active_balance_work;
653 /* cpu of this runqueue: */
657 struct list_head cfs_tasks;
664 /* This is used to determine avg_idle's max value */
665 u64 max_idle_balance_cost;
668 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
671 #ifdef CONFIG_PARAVIRT
674 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
675 u64 prev_steal_time_rq;
678 /* calc_load related fields */
679 unsigned long calc_load_update;
680 long calc_load_active;
682 #ifdef CONFIG_SCHED_HRTICK
684 int hrtick_csd_pending;
685 struct call_single_data hrtick_csd;
687 struct hrtimer hrtick_timer;
690 #ifdef CONFIG_SCHEDSTATS
692 struct sched_info rq_sched_info;
693 unsigned long long rq_cpu_time;
694 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
696 /* sys_sched_yield() stats */
697 unsigned int yld_count;
699 /* schedule() stats */
700 unsigned int sched_count;
701 unsigned int sched_goidle;
703 /* try_to_wake_up() stats */
704 unsigned int ttwu_count;
705 unsigned int ttwu_local;
709 struct llist_head wake_list;
712 #ifdef CONFIG_CPU_IDLE
713 /* Must be inspected within a rcu lock section */
714 struct cpuidle_state *idle_state;
718 static inline int cpu_of(struct rq *rq)
727 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
729 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
730 #define this_rq() this_cpu_ptr(&runqueues)
731 #define task_rq(p) cpu_rq(task_cpu(p))
732 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
733 #define raw_rq() raw_cpu_ptr(&runqueues)
735 static inline u64 __rq_clock_broken(struct rq *rq)
737 return READ_ONCE(rq->clock);
740 static inline u64 rq_clock(struct rq *rq)
742 lockdep_assert_held(&rq->lock);
746 static inline u64 rq_clock_task(struct rq *rq)
748 lockdep_assert_held(&rq->lock);
749 return rq->clock_task;
752 #define RQCF_REQ_SKIP 0x01
753 #define RQCF_ACT_SKIP 0x02
755 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
757 lockdep_assert_held(&rq->lock);
759 rq->clock_skip_update |= RQCF_REQ_SKIP;
761 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
765 enum numa_topology_type {
770 extern enum numa_topology_type sched_numa_topology_type;
771 extern int sched_max_numa_distance;
772 extern bool find_numa_distance(int distance);
775 #ifdef CONFIG_NUMA_BALANCING
776 /* The regions in numa_faults array from task_struct */
777 enum numa_faults_stats {
783 extern void sched_setnuma(struct task_struct *p, int node);
784 extern int migrate_task_to(struct task_struct *p, int cpu);
785 extern int migrate_swap(struct task_struct *, struct task_struct *);
786 #endif /* CONFIG_NUMA_BALANCING */
791 queue_balance_callback(struct rq *rq,
792 struct callback_head *head,
793 void (*func)(struct rq *rq))
795 lockdep_assert_held(&rq->lock);
797 if (unlikely(head->next))
800 head->func = (void (*)(struct callback_head *))func;
801 head->next = rq->balance_callback;
802 rq->balance_callback = head;
805 extern void sched_ttwu_pending(void);
807 #define rcu_dereference_check_sched_domain(p) \
808 rcu_dereference_check((p), \
809 lockdep_is_held(&sched_domains_mutex))
812 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
813 * See detach_destroy_domains: synchronize_sched for details.
815 * The domain tree of any CPU may only be accessed from within
816 * preempt-disabled sections.
818 #define for_each_domain(cpu, __sd) \
819 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
820 __sd; __sd = __sd->parent)
822 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
825 * highest_flag_domain - Return highest sched_domain containing flag.
826 * @cpu: The cpu whose highest level of sched domain is to
828 * @flag: The flag to check for the highest sched_domain
831 * Returns the highest sched_domain of a cpu which contains the given flag.
833 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
835 struct sched_domain *sd, *hsd = NULL;
837 for_each_domain(cpu, sd) {
838 if (!(sd->flags & flag))
846 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
848 struct sched_domain *sd;
850 for_each_domain(cpu, sd) {
851 if (sd->flags & flag)
858 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
859 DECLARE_PER_CPU(int, sd_llc_size);
860 DECLARE_PER_CPU(int, sd_llc_id);
861 DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
862 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
863 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
865 struct sched_group_capacity {
868 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
871 unsigned int capacity;
872 unsigned long next_update;
873 int imbalance; /* XXX unrelated to capacity but shared group state */
875 unsigned long cpumask[0]; /* iteration mask */
879 struct sched_group *next; /* Must be a circular list */
882 unsigned int group_weight;
883 struct sched_group_capacity *sgc;
886 * The CPUs this group covers.
888 * NOTE: this field is variable length. (Allocated dynamically
889 * by attaching extra space to the end of the structure,
890 * depending on how many CPUs the kernel has booted up with)
892 unsigned long cpumask[0];
895 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
897 return to_cpumask(sg->cpumask);
901 * cpumask masking which cpus in the group are allowed to iterate up the domain
904 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
906 return to_cpumask(sg->sgc->cpumask);
910 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
911 * @group: The group whose first cpu is to be returned.
913 static inline unsigned int group_first_cpu(struct sched_group *group)
915 return cpumask_first(sched_group_cpus(group));
918 extern int group_balance_cpu(struct sched_group *sg);
920 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
921 void register_sched_domain_sysctl(void);
922 void unregister_sched_domain_sysctl(void);
924 static inline void register_sched_domain_sysctl(void)
927 static inline void unregister_sched_domain_sysctl(void)
934 static inline void sched_ttwu_pending(void) { }
936 #endif /* CONFIG_SMP */
939 #include "auto_group.h"
941 #ifdef CONFIG_CGROUP_SCHED
944 * Return the group to which this tasks belongs.
946 * We cannot use task_css() and friends because the cgroup subsystem
947 * changes that value before the cgroup_subsys::attach() method is called,
948 * therefore we cannot pin it and might observe the wrong value.
950 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
951 * core changes this before calling sched_move_task().
953 * Instead we use a 'copy' which is updated from sched_move_task() while
954 * holding both task_struct::pi_lock and rq::lock.
956 static inline struct task_group *task_group(struct task_struct *p)
958 return p->sched_task_group;
961 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
962 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
964 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
965 struct task_group *tg = task_group(p);
968 #ifdef CONFIG_FAIR_GROUP_SCHED
969 set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]);
970 p->se.cfs_rq = tg->cfs_rq[cpu];
971 p->se.parent = tg->se[cpu];
974 #ifdef CONFIG_RT_GROUP_SCHED
975 p->rt.rt_rq = tg->rt_rq[cpu];
976 p->rt.parent = tg->rt_se[cpu];
980 #else /* CONFIG_CGROUP_SCHED */
982 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
983 static inline struct task_group *task_group(struct task_struct *p)
988 #endif /* CONFIG_CGROUP_SCHED */
990 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
995 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
996 * successfuly executed on another CPU. We must ensure that updates of
997 * per-task data have been completed by this moment.
1000 task_thread_info(p)->cpu = cpu;
1006 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1008 #ifdef CONFIG_SCHED_DEBUG
1009 # include <linux/static_key.h>
1010 # define const_debug __read_mostly
1012 # define const_debug const
1015 extern const_debug unsigned int sysctl_sched_features;
1017 #define SCHED_FEAT(name, enabled) \
1018 __SCHED_FEAT_##name ,
1021 #include "features.h"
1027 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1028 #define SCHED_FEAT(name, enabled) \
1029 static __always_inline bool static_branch_##name(struct static_key *key) \
1031 return static_key_##enabled(key); \
1034 #include "features.h"
1038 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1039 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1040 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1041 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1042 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1044 extern struct static_key_false sched_numa_balancing;
1045 extern struct static_key_false sched_schedstats;
1047 static inline u64 global_rt_period(void)
1049 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1052 static inline u64 global_rt_runtime(void)
1054 if (sysctl_sched_rt_runtime < 0)
1057 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1060 static inline int task_current(struct rq *rq, struct task_struct *p)
1062 return rq->curr == p;
1065 static inline int task_running(struct rq *rq, struct task_struct *p)
1070 return task_current(rq, p);
1074 static inline int task_on_rq_queued(struct task_struct *p)
1076 return p->on_rq == TASK_ON_RQ_QUEUED;
1079 static inline int task_on_rq_migrating(struct task_struct *p)
1081 return p->on_rq == TASK_ON_RQ_MIGRATING;
1084 #ifndef prepare_arch_switch
1085 # define prepare_arch_switch(next) do { } while (0)
1087 #ifndef finish_arch_post_lock_switch
1088 # define finish_arch_post_lock_switch() do { } while (0)
1091 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1095 * We can optimise this out completely for !SMP, because the
1096 * SMP rebalancing from interrupt is the only thing that cares
1103 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1107 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1108 * We must ensure this doesn't happen until the switch is completely
1111 * In particular, the load of prev->state in finish_task_switch() must
1112 * happen before this.
1114 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1116 smp_store_release(&prev->on_cpu, 0);
1118 #ifdef CONFIG_DEBUG_SPINLOCK
1119 /* this is a valid case when another task releases the spinlock */
1120 rq->lock.owner = current;
1123 * If we are tracking spinlock dependencies then we have to
1124 * fix up the runqueue lock - which gets 'carried over' from
1125 * prev into current:
1127 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1129 raw_spin_unlock_irq(&rq->lock);
1135 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1136 #define WF_FORK 0x02 /* child wakeup after fork */
1137 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1140 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1141 * of tasks with abnormal "nice" values across CPUs the contribution that
1142 * each task makes to its run queue's load is weighted according to its
1143 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1144 * scaled version of the new time slice allocation that they receive on time
1148 #define WEIGHT_IDLEPRIO 3
1149 #define WMULT_IDLEPRIO 1431655765
1151 extern const int sched_prio_to_weight[40];
1152 extern const u32 sched_prio_to_wmult[40];
1155 * {de,en}queue flags:
1157 * DEQUEUE_SLEEP - task is no longer runnable
1158 * ENQUEUE_WAKEUP - task just became runnable
1160 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1161 * are in a known state which allows modification. Such pairs
1162 * should preserve as much state as possible.
1164 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1167 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1168 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1169 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1173 #define DEQUEUE_SLEEP 0x01
1174 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1175 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1177 #define ENQUEUE_WAKEUP 0x01
1178 #define ENQUEUE_RESTORE 0x02
1179 #define ENQUEUE_MOVE 0x04
1181 #define ENQUEUE_HEAD 0x08
1182 #define ENQUEUE_REPLENISH 0x10
1184 #define ENQUEUE_MIGRATED 0x20
1186 #define ENQUEUE_MIGRATED 0x00
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 struct pin_cookie cookie);
1212 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1215 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1216 void (*migrate_task_rq)(struct task_struct *p);
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 #define TASK_SET_GROUP 0
1248 #define TASK_MOVE_GROUP 1
1250 #ifdef CONFIG_FAIR_GROUP_SCHED
1251 void (*task_change_group) (struct task_struct *p, int type);
1255 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1257 prev->sched_class->put_prev_task(rq, prev);
1260 #define sched_class_highest (&stop_sched_class)
1261 #define for_each_class(class) \
1262 for (class = sched_class_highest; class; class = class->next)
1264 extern const struct sched_class stop_sched_class;
1265 extern const struct sched_class dl_sched_class;
1266 extern const struct sched_class rt_sched_class;
1267 extern const struct sched_class fair_sched_class;
1268 extern const struct sched_class idle_sched_class;
1273 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1275 extern void trigger_load_balance(struct rq *rq);
1277 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1281 #ifdef CONFIG_CPU_IDLE
1282 static inline void idle_set_state(struct rq *rq,
1283 struct cpuidle_state *idle_state)
1285 rq->idle_state = idle_state;
1288 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1290 WARN_ON(!rcu_read_lock_held());
1291 return rq->idle_state;
1294 static inline void idle_set_state(struct rq *rq,
1295 struct cpuidle_state *idle_state)
1299 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1305 extern void sysrq_sched_debug_show(void);
1306 extern void sched_init_granularity(void);
1307 extern void update_max_interval(void);
1309 extern void init_sched_dl_class(void);
1310 extern void init_sched_rt_class(void);
1311 extern void init_sched_fair_class(void);
1313 extern void resched_curr(struct rq *rq);
1314 extern void resched_cpu(int cpu);
1316 extern struct rt_bandwidth def_rt_bandwidth;
1317 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1319 extern struct dl_bandwidth def_dl_bandwidth;
1320 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1321 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1323 unsigned long to_ratio(u64 period, u64 runtime);
1325 extern void init_entity_runnable_average(struct sched_entity *se);
1326 extern void post_init_entity_util_avg(struct sched_entity *se);
1328 #ifdef CONFIG_NO_HZ_FULL
1329 extern bool sched_can_stop_tick(struct rq *rq);
1332 * Tick may be needed by tasks in the runqueue depending on their policy and
1333 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1334 * nohz mode if necessary.
1336 static inline void sched_update_tick_dependency(struct rq *rq)
1340 if (!tick_nohz_full_enabled())
1345 if (!tick_nohz_full_cpu(cpu))
1348 if (sched_can_stop_tick(rq))
1349 tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED);
1351 tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED);
1354 static inline void sched_update_tick_dependency(struct rq *rq) { }
1357 static inline void add_nr_running(struct rq *rq, unsigned count)
1359 unsigned prev_nr = rq->nr_running;
1361 rq->nr_running = prev_nr + count;
1363 if (prev_nr < 2 && rq->nr_running >= 2) {
1365 if (!rq->rd->overload)
1366 rq->rd->overload = true;
1370 sched_update_tick_dependency(rq);
1373 static inline void sub_nr_running(struct rq *rq, unsigned count)
1375 rq->nr_running -= count;
1376 /* Check if we still need preemption */
1377 sched_update_tick_dependency(rq);
1380 static inline void rq_last_tick_reset(struct rq *rq)
1382 #ifdef CONFIG_NO_HZ_FULL
1383 rq->last_sched_tick = jiffies;
1387 extern void update_rq_clock(struct rq *rq);
1389 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1390 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1392 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1394 extern const_debug unsigned int sysctl_sched_time_avg;
1395 extern const_debug unsigned int sysctl_sched_nr_migrate;
1396 extern const_debug unsigned int sysctl_sched_migration_cost;
1398 static inline u64 sched_avg_period(void)
1400 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1403 #ifdef CONFIG_SCHED_HRTICK
1407 * - enabled by features
1408 * - hrtimer is actually high res
1410 static inline int hrtick_enabled(struct rq *rq)
1412 if (!sched_feat(HRTICK))
1414 if (!cpu_active(cpu_of(rq)))
1416 return hrtimer_is_hres_active(&rq->hrtick_timer);
1419 void hrtick_start(struct rq *rq, u64 delay);
1423 static inline int hrtick_enabled(struct rq *rq)
1428 #endif /* CONFIG_SCHED_HRTICK */
1431 extern void sched_avg_update(struct rq *rq);
1433 #ifndef arch_scale_freq_capacity
1434 static __always_inline
1435 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1437 return SCHED_CAPACITY_SCALE;
1441 #ifndef arch_scale_cpu_capacity
1442 static __always_inline
1443 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1445 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1446 return sd->smt_gain / sd->span_weight;
1448 return SCHED_CAPACITY_SCALE;
1452 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1454 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1455 sched_avg_update(rq);
1458 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1459 static inline void sched_avg_update(struct rq *rq) { }
1463 unsigned long flags;
1464 struct pin_cookie cookie;
1467 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1468 __acquires(rq->lock);
1469 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
1470 __acquires(p->pi_lock)
1471 __acquires(rq->lock);
1473 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
1474 __releases(rq->lock)
1476 lockdep_unpin_lock(&rq->lock, rf->cookie);
1477 raw_spin_unlock(&rq->lock);
1481 task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
1482 __releases(rq->lock)
1483 __releases(p->pi_lock)
1485 lockdep_unpin_lock(&rq->lock, rf->cookie);
1486 raw_spin_unlock(&rq->lock);
1487 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
1491 #ifdef CONFIG_PREEMPT
1493 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1496 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1497 * way at the expense of forcing extra atomic operations in all
1498 * invocations. This assures that the double_lock is acquired using the
1499 * same underlying policy as the spinlock_t on this architecture, which
1500 * reduces latency compared to the unfair variant below. However, it
1501 * also adds more overhead and therefore may reduce throughput.
1503 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1504 __releases(this_rq->lock)
1505 __acquires(busiest->lock)
1506 __acquires(this_rq->lock)
1508 raw_spin_unlock(&this_rq->lock);
1509 double_rq_lock(this_rq, busiest);
1516 * Unfair double_lock_balance: Optimizes throughput at the expense of
1517 * latency by eliminating extra atomic operations when the locks are
1518 * already in proper order on entry. This favors lower cpu-ids and will
1519 * grant the double lock to lower cpus over higher ids under contention,
1520 * regardless of entry order into the function.
1522 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1523 __releases(this_rq->lock)
1524 __acquires(busiest->lock)
1525 __acquires(this_rq->lock)
1529 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1530 if (busiest < this_rq) {
1531 raw_spin_unlock(&this_rq->lock);
1532 raw_spin_lock(&busiest->lock);
1533 raw_spin_lock_nested(&this_rq->lock,
1534 SINGLE_DEPTH_NESTING);
1537 raw_spin_lock_nested(&busiest->lock,
1538 SINGLE_DEPTH_NESTING);
1543 #endif /* CONFIG_PREEMPT */
1546 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1548 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1550 if (unlikely(!irqs_disabled())) {
1551 /* printk() doesn't work good under rq->lock */
1552 raw_spin_unlock(&this_rq->lock);
1556 return _double_lock_balance(this_rq, busiest);
1559 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1560 __releases(busiest->lock)
1562 raw_spin_unlock(&busiest->lock);
1563 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1566 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1572 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1575 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1581 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1584 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1590 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1594 * double_rq_lock - safely lock two runqueues
1596 * Note this does not disable interrupts like task_rq_lock,
1597 * you need to do so manually before calling.
1599 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1600 __acquires(rq1->lock)
1601 __acquires(rq2->lock)
1603 BUG_ON(!irqs_disabled());
1605 raw_spin_lock(&rq1->lock);
1606 __acquire(rq2->lock); /* Fake it out ;) */
1609 raw_spin_lock(&rq1->lock);
1610 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1612 raw_spin_lock(&rq2->lock);
1613 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1619 * double_rq_unlock - safely unlock two runqueues
1621 * Note this does not restore interrupts like task_rq_unlock,
1622 * you need to do so manually after calling.
1624 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1625 __releases(rq1->lock)
1626 __releases(rq2->lock)
1628 raw_spin_unlock(&rq1->lock);
1630 raw_spin_unlock(&rq2->lock);
1632 __release(rq2->lock);
1635 #else /* CONFIG_SMP */
1638 * double_rq_lock - safely lock two runqueues
1640 * Note this does not disable interrupts like task_rq_lock,
1641 * you need to do so manually before calling.
1643 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1644 __acquires(rq1->lock)
1645 __acquires(rq2->lock)
1647 BUG_ON(!irqs_disabled());
1649 raw_spin_lock(&rq1->lock);
1650 __acquire(rq2->lock); /* Fake it out ;) */
1654 * double_rq_unlock - safely unlock two runqueues
1656 * Note this does not restore interrupts like task_rq_unlock,
1657 * you need to do so manually after calling.
1659 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1660 __releases(rq1->lock)
1661 __releases(rq2->lock)
1664 raw_spin_unlock(&rq1->lock);
1665 __release(rq2->lock);
1670 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1671 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1673 #ifdef CONFIG_SCHED_DEBUG
1674 extern void print_cfs_stats(struct seq_file *m, int cpu);
1675 extern void print_rt_stats(struct seq_file *m, int cpu);
1676 extern void print_dl_stats(struct seq_file *m, int cpu);
1678 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1680 #ifdef CONFIG_NUMA_BALANCING
1682 show_numa_stats(struct task_struct *p, struct seq_file *m);
1684 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1685 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1686 #endif /* CONFIG_NUMA_BALANCING */
1687 #endif /* CONFIG_SCHED_DEBUG */
1689 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1690 extern void init_rt_rq(struct rt_rq *rt_rq);
1691 extern void init_dl_rq(struct dl_rq *dl_rq);
1693 extern void cfs_bandwidth_usage_inc(void);
1694 extern void cfs_bandwidth_usage_dec(void);
1696 #ifdef CONFIG_NO_HZ_COMMON
1697 enum rq_nohz_flag_bits {
1702 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1704 extern void nohz_balance_exit_idle(unsigned int cpu);
1706 static inline void nohz_balance_exit_idle(unsigned int cpu) { }
1709 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1711 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1712 DECLARE_PER_CPU(u64, cpu_softirq_time);
1714 #ifndef CONFIG_64BIT
1715 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1717 static inline void irq_time_write_begin(void)
1719 __this_cpu_inc(irq_time_seq.sequence);
1723 static inline void irq_time_write_end(void)
1726 __this_cpu_inc(irq_time_seq.sequence);
1729 static inline u64 irq_time_read(int cpu)
1735 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1736 irq_time = per_cpu(cpu_softirq_time, cpu) +
1737 per_cpu(cpu_hardirq_time, cpu);
1738 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1742 #else /* CONFIG_64BIT */
1743 static inline void irq_time_write_begin(void)
1747 static inline void irq_time_write_end(void)
1751 static inline u64 irq_time_read(int cpu)
1753 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1755 #endif /* CONFIG_64BIT */
1756 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1758 #ifdef CONFIG_CPU_FREQ
1759 DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data);
1762 * cpufreq_update_util - Take a note about CPU utilization changes.
1763 * @time: Current time.
1764 * @util: Current utilization.
1765 * @max: Utilization ceiling.
1767 * This function is called by the scheduler on every invocation of
1768 * update_load_avg() on the CPU whose utilization is being updated.
1770 * It can only be called from RCU-sched read-side critical sections.
1772 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max)
1774 struct update_util_data *data;
1776 data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data));
1778 data->func(data, time, util, max);
1782 * cpufreq_trigger_update - Trigger CPU performance state evaluation if needed.
1783 * @time: Current time.
1785 * The way cpufreq is currently arranged requires it to evaluate the CPU
1786 * performance state (frequency/voltage) on a regular basis to prevent it from
1787 * being stuck in a completely inadequate performance level for too long.
1788 * That is not guaranteed to happen if the updates are only triggered from CFS,
1789 * though, because they may not be coming in if RT or deadline tasks are active
1790 * all the time (or there are RT and DL tasks only).
1792 * As a workaround for that issue, this function is called by the RT and DL
1793 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1794 * but that really is a band-aid. Going forward it should be replaced with
1795 * solutions targeted more specifically at RT and DL tasks.
1797 static inline void cpufreq_trigger_update(u64 time)
1799 cpufreq_update_util(time, ULONG_MAX, 0);
1802 static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) {}
1803 static inline void cpufreq_trigger_update(u64 time) {}
1804 #endif /* CONFIG_CPU_FREQ */
1806 #ifdef arch_scale_freq_capacity
1807 #ifndef arch_scale_freq_invariant
1808 #define arch_scale_freq_invariant() (true)
1810 #else /* arch_scale_freq_capacity */
1811 #define arch_scale_freq_invariant() (false)