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;
514 * We add the notion of a root-domain which will be used to define per-domain
515 * variables. Each exclusive cpuset essentially defines an island domain by
516 * fully partitioning the member cpus from any other cpuset. Whenever a new
517 * exclusive cpuset is created, we also create and attach a new root-domain
526 cpumask_var_t online;
528 /* Indicate more than one runnable task for any CPU */
532 * The bit corresponding to a CPU gets set here if such CPU has more
533 * than one runnable -deadline task (as it is below for RT tasks).
535 cpumask_var_t dlo_mask;
541 * The "RT overload" flag: it gets set if a CPU has more than
542 * one runnable RT task.
544 cpumask_var_t rto_mask;
545 struct cpupri cpupri;
548 extern struct root_domain def_root_domain;
550 #endif /* CONFIG_SMP */
553 * This is the main, per-CPU runqueue data structure.
555 * Locking rule: those places that want to lock multiple runqueues
556 * (such as the load balancing or the thread migration code), lock
557 * acquire operations must be ordered by ascending &runqueue.
564 * nr_running and cpu_load should be in the same cacheline because
565 * remote CPUs use both these fields when doing load calculation.
567 unsigned int nr_running;
568 #ifdef CONFIG_NUMA_BALANCING
569 unsigned int nr_numa_running;
570 unsigned int nr_preferred_running;
572 #define CPU_LOAD_IDX_MAX 5
573 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
574 unsigned long last_load_update_tick;
575 #ifdef CONFIG_NO_HZ_COMMON
577 unsigned long nohz_flags;
579 #ifdef CONFIG_NO_HZ_FULL
580 unsigned long last_sched_tick;
582 /* capture load from *all* tasks on this cpu: */
583 struct load_weight load;
584 unsigned long nr_load_updates;
591 #ifdef CONFIG_FAIR_GROUP_SCHED
592 /* list of leaf cfs_rq on this cpu: */
593 struct list_head leaf_cfs_rq_list;
594 #endif /* CONFIG_FAIR_GROUP_SCHED */
597 * This is part of a global counter where only the total sum
598 * over all CPUs matters. A task can increase this counter on
599 * one CPU and if it got migrated afterwards it may decrease
600 * it on another CPU. Always updated under the runqueue lock:
602 unsigned long nr_uninterruptible;
604 struct task_struct *curr, *idle, *stop;
605 unsigned long next_balance;
606 struct mm_struct *prev_mm;
608 unsigned int clock_skip_update;
615 struct root_domain *rd;
616 struct sched_domain *sd;
618 unsigned long cpu_capacity;
619 unsigned long cpu_capacity_orig;
621 struct callback_head *balance_callback;
623 unsigned char idle_balance;
624 /* For active balancing */
627 struct cpu_stop_work active_balance_work;
628 /* cpu of this runqueue: */
632 struct list_head cfs_tasks;
639 /* This is used to determine avg_idle's max value */
640 u64 max_idle_balance_cost;
643 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
646 #ifdef CONFIG_PARAVIRT
649 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
650 u64 prev_steal_time_rq;
653 /* calc_load related fields */
654 unsigned long calc_load_update;
655 long calc_load_active;
657 #ifdef CONFIG_SCHED_HRTICK
659 int hrtick_csd_pending;
660 struct call_single_data hrtick_csd;
662 struct hrtimer hrtick_timer;
665 #ifdef CONFIG_SCHEDSTATS
667 struct sched_info rq_sched_info;
668 unsigned long long rq_cpu_time;
669 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
671 /* sys_sched_yield() stats */
672 unsigned int yld_count;
674 /* schedule() stats */
675 unsigned int sched_count;
676 unsigned int sched_goidle;
678 /* try_to_wake_up() stats */
679 unsigned int ttwu_count;
680 unsigned int ttwu_local;
684 struct llist_head wake_list;
687 #ifdef CONFIG_CPU_IDLE
688 /* Must be inspected within a rcu lock section */
689 struct cpuidle_state *idle_state;
693 static inline int cpu_of(struct rq *rq)
702 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
704 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
705 #define this_rq() this_cpu_ptr(&runqueues)
706 #define task_rq(p) cpu_rq(task_cpu(p))
707 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
708 #define raw_rq() raw_cpu_ptr(&runqueues)
710 static inline u64 __rq_clock_broken(struct rq *rq)
712 return READ_ONCE(rq->clock);
715 static inline u64 rq_clock(struct rq *rq)
717 lockdep_assert_held(&rq->lock);
721 static inline u64 rq_clock_task(struct rq *rq)
723 lockdep_assert_held(&rq->lock);
724 return rq->clock_task;
727 #define RQCF_REQ_SKIP 0x01
728 #define RQCF_ACT_SKIP 0x02
730 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
732 lockdep_assert_held(&rq->lock);
734 rq->clock_skip_update |= RQCF_REQ_SKIP;
736 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
740 enum numa_topology_type {
745 extern enum numa_topology_type sched_numa_topology_type;
746 extern int sched_max_numa_distance;
747 extern bool find_numa_distance(int distance);
750 #ifdef CONFIG_NUMA_BALANCING
751 /* The regions in numa_faults array from task_struct */
752 enum numa_faults_stats {
758 extern void sched_setnuma(struct task_struct *p, int node);
759 extern int migrate_task_to(struct task_struct *p, int cpu);
760 extern int migrate_swap(struct task_struct *, struct task_struct *);
761 #endif /* CONFIG_NUMA_BALANCING */
766 queue_balance_callback(struct rq *rq,
767 struct callback_head *head,
768 void (*func)(struct rq *rq))
770 lockdep_assert_held(&rq->lock);
772 if (unlikely(head->next))
775 head->func = (void (*)(struct callback_head *))func;
776 head->next = rq->balance_callback;
777 rq->balance_callback = head;
780 extern void sched_ttwu_pending(void);
782 #define rcu_dereference_check_sched_domain(p) \
783 rcu_dereference_check((p), \
784 lockdep_is_held(&sched_domains_mutex))
787 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
788 * See detach_destroy_domains: synchronize_sched for details.
790 * The domain tree of any CPU may only be accessed from within
791 * preempt-disabled sections.
793 #define for_each_domain(cpu, __sd) \
794 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
795 __sd; __sd = __sd->parent)
797 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
800 * highest_flag_domain - Return highest sched_domain containing flag.
801 * @cpu: The cpu whose highest level of sched domain is to
803 * @flag: The flag to check for the highest sched_domain
806 * Returns the highest sched_domain of a cpu which contains the given flag.
808 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
810 struct sched_domain *sd, *hsd = NULL;
812 for_each_domain(cpu, sd) {
813 if (!(sd->flags & flag))
821 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
823 struct sched_domain *sd;
825 for_each_domain(cpu, sd) {
826 if (sd->flags & flag)
833 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
834 DECLARE_PER_CPU(int, sd_llc_size);
835 DECLARE_PER_CPU(int, sd_llc_id);
836 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
837 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
838 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
840 struct sched_group_capacity {
843 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
846 unsigned int capacity;
847 unsigned long next_update;
848 int imbalance; /* XXX unrelated to capacity but shared group state */
850 * Number of busy cpus in this group.
852 atomic_t nr_busy_cpus;
854 unsigned long cpumask[0]; /* iteration mask */
858 struct sched_group *next; /* Must be a circular list */
861 unsigned int group_weight;
862 struct sched_group_capacity *sgc;
865 * The CPUs this group covers.
867 * NOTE: this field is variable length. (Allocated dynamically
868 * by attaching extra space to the end of the structure,
869 * depending on how many CPUs the kernel has booted up with)
871 unsigned long cpumask[0];
874 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
876 return to_cpumask(sg->cpumask);
880 * cpumask masking which cpus in the group are allowed to iterate up the domain
883 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
885 return to_cpumask(sg->sgc->cpumask);
889 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
890 * @group: The group whose first cpu is to be returned.
892 static inline unsigned int group_first_cpu(struct sched_group *group)
894 return cpumask_first(sched_group_cpus(group));
897 extern int group_balance_cpu(struct sched_group *sg);
901 static inline void sched_ttwu_pending(void) { }
903 #endif /* CONFIG_SMP */
906 #include "auto_group.h"
908 #ifdef CONFIG_CGROUP_SCHED
911 * Return the group to which this tasks belongs.
913 * We cannot use task_css() and friends because the cgroup subsystem
914 * changes that value before the cgroup_subsys::attach() method is called,
915 * therefore we cannot pin it and might observe the wrong value.
917 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
918 * core changes this before calling sched_move_task().
920 * Instead we use a 'copy' which is updated from sched_move_task() while
921 * holding both task_struct::pi_lock and rq::lock.
923 static inline struct task_group *task_group(struct task_struct *p)
925 return p->sched_task_group;
928 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
929 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
931 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
932 struct task_group *tg = task_group(p);
935 #ifdef CONFIG_FAIR_GROUP_SCHED
936 p->se.cfs_rq = tg->cfs_rq[cpu];
937 p->se.parent = tg->se[cpu];
940 #ifdef CONFIG_RT_GROUP_SCHED
941 p->rt.rt_rq = tg->rt_rq[cpu];
942 p->rt.parent = tg->rt_se[cpu];
946 #else /* CONFIG_CGROUP_SCHED */
948 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
949 static inline struct task_group *task_group(struct task_struct *p)
954 #endif /* CONFIG_CGROUP_SCHED */
956 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
961 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
962 * successfuly executed on another CPU. We must ensure that updates of
963 * per-task data have been completed by this moment.
966 task_thread_info(p)->cpu = cpu;
972 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
974 #ifdef CONFIG_SCHED_DEBUG
975 # include <linux/static_key.h>
976 # define const_debug __read_mostly
978 # define const_debug const
981 extern const_debug unsigned int sysctl_sched_features;
983 #define SCHED_FEAT(name, enabled) \
984 __SCHED_FEAT_##name ,
987 #include "features.h"
993 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
994 #define SCHED_FEAT(name, enabled) \
995 static __always_inline bool static_branch_##name(struct static_key *key) \
997 return static_key_##enabled(key); \
1000 #include "features.h"
1004 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1005 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1006 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1007 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1008 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1010 extern struct static_key_false sched_numa_balancing;
1012 static inline u64 global_rt_period(void)
1014 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1017 static inline u64 global_rt_runtime(void)
1019 if (sysctl_sched_rt_runtime < 0)
1022 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1025 static inline int task_current(struct rq *rq, struct task_struct *p)
1027 return rq->curr == p;
1030 static inline int task_running(struct rq *rq, struct task_struct *p)
1035 return task_current(rq, p);
1039 static inline int task_on_rq_queued(struct task_struct *p)
1041 return p->on_rq == TASK_ON_RQ_QUEUED;
1044 static inline int task_on_rq_migrating(struct task_struct *p)
1046 return p->on_rq == TASK_ON_RQ_MIGRATING;
1049 #ifndef prepare_arch_switch
1050 # define prepare_arch_switch(next) do { } while (0)
1052 #ifndef finish_arch_post_lock_switch
1053 # define finish_arch_post_lock_switch() do { } while (0)
1056 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1060 * We can optimise this out completely for !SMP, because the
1061 * SMP rebalancing from interrupt is the only thing that cares
1068 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1072 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1073 * We must ensure this doesn't happen until the switch is completely
1076 * Pairs with the control dependency and rmb in try_to_wake_up().
1078 smp_store_release(&prev->on_cpu, 0);
1080 #ifdef CONFIG_DEBUG_SPINLOCK
1081 /* this is a valid case when another task releases the spinlock */
1082 rq->lock.owner = current;
1085 * If we are tracking spinlock dependencies then we have to
1086 * fix up the runqueue lock - which gets 'carried over' from
1087 * prev into current:
1089 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1091 raw_spin_unlock_irq(&rq->lock);
1097 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1098 #define WF_FORK 0x02 /* child wakeup after fork */
1099 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1102 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1103 * of tasks with abnormal "nice" values across CPUs the contribution that
1104 * each task makes to its run queue's load is weighted according to its
1105 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1106 * scaled version of the new time slice allocation that they receive on time
1110 #define WEIGHT_IDLEPRIO 3
1111 #define WMULT_IDLEPRIO 1431655765
1114 * Nice levels are multiplicative, with a gentle 10% change for every
1115 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1116 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1117 * that remained on nice 0.
1119 * The "10% effect" is relative and cumulative: from _any_ nice level,
1120 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1121 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1122 * If a task goes up by ~10% and another task goes down by ~10% then
1123 * the relative distance between them is ~25%.)
1125 static const int prio_to_weight[40] = {
1126 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1127 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1128 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1129 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1130 /* 0 */ 1024, 820, 655, 526, 423,
1131 /* 5 */ 335, 272, 215, 172, 137,
1132 /* 10 */ 110, 87, 70, 56, 45,
1133 /* 15 */ 36, 29, 23, 18, 15,
1137 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1139 * In cases where the weight does not change often, we can use the
1140 * precalculated inverse to speed up arithmetics by turning divisions
1141 * into multiplications:
1143 static const u32 prio_to_wmult[40] = {
1144 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1145 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1146 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1147 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1148 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1149 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1150 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1151 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1154 #define ENQUEUE_WAKEUP 0x01
1155 #define ENQUEUE_HEAD 0x02
1157 #define ENQUEUE_WAKING 0x04 /* sched_class::task_waking was called */
1159 #define ENQUEUE_WAKING 0x00
1161 #define ENQUEUE_REPLENISH 0x08
1162 #define ENQUEUE_RESTORE 0x10
1164 #define DEQUEUE_SLEEP 0x01
1165 #define DEQUEUE_SAVE 0x02
1167 #define RETRY_TASK ((void *)-1UL)
1169 struct sched_class {
1170 const struct sched_class *next;
1172 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1173 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1174 void (*yield_task) (struct rq *rq);
1175 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1177 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1180 * It is the responsibility of the pick_next_task() method that will
1181 * return the next task to call put_prev_task() on the @prev task or
1182 * something equivalent.
1184 * May return RETRY_TASK when it finds a higher prio class has runnable
1187 struct task_struct * (*pick_next_task) (struct rq *rq,
1188 struct task_struct *prev);
1189 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1192 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1193 void (*migrate_task_rq)(struct task_struct *p);
1195 void (*task_waking) (struct task_struct *task);
1196 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1198 void (*set_cpus_allowed)(struct task_struct *p,
1199 const struct cpumask *newmask);
1201 void (*rq_online)(struct rq *rq);
1202 void (*rq_offline)(struct rq *rq);
1205 void (*set_curr_task) (struct rq *rq);
1206 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1207 void (*task_fork) (struct task_struct *p);
1208 void (*task_dead) (struct task_struct *p);
1211 * The switched_from() call is allowed to drop rq->lock, therefore we
1212 * cannot assume the switched_from/switched_to pair is serliazed by
1213 * rq->lock. They are however serialized by p->pi_lock.
1215 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1216 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1217 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1220 unsigned int (*get_rr_interval) (struct rq *rq,
1221 struct task_struct *task);
1223 void (*update_curr) (struct rq *rq);
1225 #ifdef CONFIG_FAIR_GROUP_SCHED
1226 void (*task_move_group) (struct task_struct *p);
1230 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1232 prev->sched_class->put_prev_task(rq, prev);
1235 #define sched_class_highest (&stop_sched_class)
1236 #define for_each_class(class) \
1237 for (class = sched_class_highest; class; class = class->next)
1239 extern const struct sched_class stop_sched_class;
1240 extern const struct sched_class dl_sched_class;
1241 extern const struct sched_class rt_sched_class;
1242 extern const struct sched_class fair_sched_class;
1243 extern const struct sched_class idle_sched_class;
1248 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1250 extern void trigger_load_balance(struct rq *rq);
1252 extern void idle_enter_fair(struct rq *this_rq);
1253 extern void idle_exit_fair(struct rq *this_rq);
1255 extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask);
1259 static inline void idle_enter_fair(struct rq *rq) { }
1260 static inline void idle_exit_fair(struct rq *rq) { }
1264 #ifdef CONFIG_CPU_IDLE
1265 static inline void idle_set_state(struct rq *rq,
1266 struct cpuidle_state *idle_state)
1268 rq->idle_state = idle_state;
1271 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1273 WARN_ON(!rcu_read_lock_held());
1274 return rq->idle_state;
1277 static inline void idle_set_state(struct rq *rq,
1278 struct cpuidle_state *idle_state)
1282 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1288 extern void sysrq_sched_debug_show(void);
1289 extern void sched_init_granularity(void);
1290 extern void update_max_interval(void);
1292 extern void init_sched_dl_class(void);
1293 extern void init_sched_rt_class(void);
1294 extern void init_sched_fair_class(void);
1296 extern void resched_curr(struct rq *rq);
1297 extern void resched_cpu(int cpu);
1299 extern struct rt_bandwidth def_rt_bandwidth;
1300 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1302 extern struct dl_bandwidth def_dl_bandwidth;
1303 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1304 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1306 unsigned long to_ratio(u64 period, u64 runtime);
1308 extern void init_entity_runnable_average(struct sched_entity *se);
1310 static inline void add_nr_running(struct rq *rq, unsigned count)
1312 unsigned prev_nr = rq->nr_running;
1314 rq->nr_running = prev_nr + count;
1316 if (prev_nr < 2 && rq->nr_running >= 2) {
1318 if (!rq->rd->overload)
1319 rq->rd->overload = true;
1322 #ifdef CONFIG_NO_HZ_FULL
1323 if (tick_nohz_full_cpu(rq->cpu)) {
1325 * Tick is needed if more than one task runs on a CPU.
1326 * Send the target an IPI to kick it out of nohz mode.
1328 * We assume that IPI implies full memory barrier and the
1329 * new value of rq->nr_running is visible on reception
1332 tick_nohz_full_kick_cpu(rq->cpu);
1338 static inline void sub_nr_running(struct rq *rq, unsigned count)
1340 rq->nr_running -= count;
1343 static inline void rq_last_tick_reset(struct rq *rq)
1345 #ifdef CONFIG_NO_HZ_FULL
1346 rq->last_sched_tick = jiffies;
1350 extern void update_rq_clock(struct rq *rq);
1352 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1353 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1355 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1357 extern const_debug unsigned int sysctl_sched_time_avg;
1358 extern const_debug unsigned int sysctl_sched_nr_migrate;
1359 extern const_debug unsigned int sysctl_sched_migration_cost;
1361 static inline u64 sched_avg_period(void)
1363 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1366 #ifdef CONFIG_SCHED_HRTICK
1370 * - enabled by features
1371 * - hrtimer is actually high res
1373 static inline int hrtick_enabled(struct rq *rq)
1375 if (!sched_feat(HRTICK))
1377 if (!cpu_active(cpu_of(rq)))
1379 return hrtimer_is_hres_active(&rq->hrtick_timer);
1382 void hrtick_start(struct rq *rq, u64 delay);
1386 static inline int hrtick_enabled(struct rq *rq)
1391 #endif /* CONFIG_SCHED_HRTICK */
1394 extern void sched_avg_update(struct rq *rq);
1396 #ifndef arch_scale_freq_capacity
1397 static __always_inline
1398 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1400 return SCHED_CAPACITY_SCALE;
1404 #ifndef arch_scale_cpu_capacity
1405 static __always_inline
1406 unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
1408 if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
1409 return sd->smt_gain / sd->span_weight;
1411 return SCHED_CAPACITY_SCALE;
1415 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1417 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1418 sched_avg_update(rq);
1421 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1422 static inline void sched_avg_update(struct rq *rq) { }
1426 * __task_rq_lock - lock the rq @p resides on.
1428 static inline struct rq *__task_rq_lock(struct task_struct *p)
1429 __acquires(rq->lock)
1433 lockdep_assert_held(&p->pi_lock);
1437 raw_spin_lock(&rq->lock);
1438 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1439 lockdep_pin_lock(&rq->lock);
1442 raw_spin_unlock(&rq->lock);
1444 while (unlikely(task_on_rq_migrating(p)))
1450 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1452 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1453 __acquires(p->pi_lock)
1454 __acquires(rq->lock)
1459 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1461 raw_spin_lock(&rq->lock);
1463 * move_queued_task() task_rq_lock()
1465 * ACQUIRE (rq->lock)
1466 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1467 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1468 * [S] ->cpu = new_cpu [L] task_rq()
1470 * RELEASE (rq->lock)
1472 * If we observe the old cpu in task_rq_lock, the acquire of
1473 * the old rq->lock will fully serialize against the stores.
1475 * If we observe the new cpu in task_rq_lock, the acquire will
1476 * pair with the WMB to ensure we must then also see migrating.
1478 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1479 lockdep_pin_lock(&rq->lock);
1482 raw_spin_unlock(&rq->lock);
1483 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1485 while (unlikely(task_on_rq_migrating(p)))
1490 static inline void __task_rq_unlock(struct rq *rq)
1491 __releases(rq->lock)
1493 lockdep_unpin_lock(&rq->lock);
1494 raw_spin_unlock(&rq->lock);
1498 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1499 __releases(rq->lock)
1500 __releases(p->pi_lock)
1502 lockdep_unpin_lock(&rq->lock);
1503 raw_spin_unlock(&rq->lock);
1504 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1508 #ifdef CONFIG_PREEMPT
1510 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1513 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1514 * way at the expense of forcing extra atomic operations in all
1515 * invocations. This assures that the double_lock is acquired using the
1516 * same underlying policy as the spinlock_t on this architecture, which
1517 * reduces latency compared to the unfair variant below. However, it
1518 * also adds more overhead and therefore may reduce throughput.
1520 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1521 __releases(this_rq->lock)
1522 __acquires(busiest->lock)
1523 __acquires(this_rq->lock)
1525 raw_spin_unlock(&this_rq->lock);
1526 double_rq_lock(this_rq, busiest);
1533 * Unfair double_lock_balance: Optimizes throughput at the expense of
1534 * latency by eliminating extra atomic operations when the locks are
1535 * already in proper order on entry. This favors lower cpu-ids and will
1536 * grant the double lock to lower cpus over higher ids under contention,
1537 * regardless of entry order into the function.
1539 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1540 __releases(this_rq->lock)
1541 __acquires(busiest->lock)
1542 __acquires(this_rq->lock)
1546 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1547 if (busiest < this_rq) {
1548 raw_spin_unlock(&this_rq->lock);
1549 raw_spin_lock(&busiest->lock);
1550 raw_spin_lock_nested(&this_rq->lock,
1551 SINGLE_DEPTH_NESTING);
1554 raw_spin_lock_nested(&busiest->lock,
1555 SINGLE_DEPTH_NESTING);
1560 #endif /* CONFIG_PREEMPT */
1563 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1565 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1567 if (unlikely(!irqs_disabled())) {
1568 /* printk() doesn't work good under rq->lock */
1569 raw_spin_unlock(&this_rq->lock);
1573 return _double_lock_balance(this_rq, busiest);
1576 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1577 __releases(busiest->lock)
1579 raw_spin_unlock(&busiest->lock);
1580 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1583 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1589 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1592 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1598 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1601 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1607 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1611 * double_rq_lock - safely lock two runqueues
1613 * Note this does not disable interrupts like task_rq_lock,
1614 * you need to do so manually before calling.
1616 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1617 __acquires(rq1->lock)
1618 __acquires(rq2->lock)
1620 BUG_ON(!irqs_disabled());
1622 raw_spin_lock(&rq1->lock);
1623 __acquire(rq2->lock); /* Fake it out ;) */
1626 raw_spin_lock(&rq1->lock);
1627 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1629 raw_spin_lock(&rq2->lock);
1630 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1636 * double_rq_unlock - safely unlock two runqueues
1638 * Note this does not restore interrupts like task_rq_unlock,
1639 * you need to do so manually after calling.
1641 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1642 __releases(rq1->lock)
1643 __releases(rq2->lock)
1645 raw_spin_unlock(&rq1->lock);
1647 raw_spin_unlock(&rq2->lock);
1649 __release(rq2->lock);
1652 #else /* CONFIG_SMP */
1655 * double_rq_lock - safely lock two runqueues
1657 * Note this does not disable interrupts like task_rq_lock,
1658 * you need to do so manually before calling.
1660 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1661 __acquires(rq1->lock)
1662 __acquires(rq2->lock)
1664 BUG_ON(!irqs_disabled());
1666 raw_spin_lock(&rq1->lock);
1667 __acquire(rq2->lock); /* Fake it out ;) */
1671 * double_rq_unlock - safely unlock two runqueues
1673 * Note this does not restore interrupts like task_rq_unlock,
1674 * you need to do so manually after calling.
1676 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1677 __releases(rq1->lock)
1678 __releases(rq2->lock)
1681 raw_spin_unlock(&rq1->lock);
1682 __release(rq2->lock);
1687 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1688 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1690 #ifdef CONFIG_SCHED_DEBUG
1691 extern void print_cfs_stats(struct seq_file *m, int cpu);
1692 extern void print_rt_stats(struct seq_file *m, int cpu);
1693 extern void print_dl_stats(struct seq_file *m, int cpu);
1695 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1697 #ifdef CONFIG_NUMA_BALANCING
1699 show_numa_stats(struct task_struct *p, struct seq_file *m);
1701 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1702 unsigned long tpf, unsigned long gsf, unsigned long gpf);
1703 #endif /* CONFIG_NUMA_BALANCING */
1704 #endif /* CONFIG_SCHED_DEBUG */
1706 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1707 extern void init_rt_rq(struct rt_rq *rt_rq);
1708 extern void init_dl_rq(struct dl_rq *dl_rq);
1710 extern void cfs_bandwidth_usage_inc(void);
1711 extern void cfs_bandwidth_usage_dec(void);
1713 #ifdef CONFIG_NO_HZ_COMMON
1714 enum rq_nohz_flag_bits {
1719 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1722 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1724 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1725 DECLARE_PER_CPU(u64, cpu_softirq_time);
1727 #ifndef CONFIG_64BIT
1728 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1730 static inline void irq_time_write_begin(void)
1732 __this_cpu_inc(irq_time_seq.sequence);
1736 static inline void irq_time_write_end(void)
1739 __this_cpu_inc(irq_time_seq.sequence);
1742 static inline u64 irq_time_read(int cpu)
1748 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1749 irq_time = per_cpu(cpu_softirq_time, cpu) +
1750 per_cpu(cpu_hardirq_time, cpu);
1751 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1755 #else /* CONFIG_64BIT */
1756 static inline void irq_time_write_begin(void)
1760 static inline void irq_time_write_end(void)
1764 static inline u64 irq_time_read(int cpu)
1766 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1768 #endif /* CONFIG_64BIT */
1769 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */