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 long calc_load_fold_active(struct rq *this_rq);
30 extern void update_cpu_load_active(struct rq *this_rq);
33 * Helpers for converting nanosecond timing to jiffy resolution
35 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
38 * Increase resolution of nice-level calculations for 64-bit architectures.
39 * The extra resolution improves shares distribution and load balancing of
40 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
41 * hierarchies, especially on larger systems. This is not a user-visible change
42 * and does not change the user-interface for setting shares/weights.
44 * We increase resolution only if we have enough bits to allow this increased
45 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
46 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
49 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
50 # define SCHED_LOAD_RESOLUTION 10
51 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
52 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
54 # define SCHED_LOAD_RESOLUTION 0
55 # define scale_load(w) (w)
56 # define scale_load_down(w) (w)
59 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
60 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
62 #define NICE_0_LOAD SCHED_LOAD_SCALE
63 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
66 * Single value that decides SCHED_DEADLINE internal math precision.
67 * 10 -> just above 1us
68 * 9 -> just above 0.5us
73 * These are the 'tuning knobs' of the scheduler:
77 * single value that denotes runtime == period, ie unlimited time.
79 #define RUNTIME_INF ((u64)~0ULL)
81 static inline int fair_policy(int policy)
83 return policy == SCHED_NORMAL || policy == SCHED_BATCH;
86 static inline int rt_policy(int policy)
88 return policy == SCHED_FIFO || policy == SCHED_RR;
91 static inline int dl_policy(int policy)
93 return policy == SCHED_DEADLINE;
96 static inline int task_has_rt_policy(struct task_struct *p)
98 return rt_policy(p->policy);
101 static inline int task_has_dl_policy(struct task_struct *p)
103 return dl_policy(p->policy);
106 static inline bool dl_time_before(u64 a, u64 b)
108 return (s64)(a - b) < 0;
112 * Tells if entity @a should preempt entity @b.
115 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
117 return dl_time_before(a->deadline, b->deadline);
121 * This is the priority-queue data structure of the RT scheduling class:
123 struct rt_prio_array {
124 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
125 struct list_head queue[MAX_RT_PRIO];
128 struct rt_bandwidth {
129 /* nests inside the rq lock: */
130 raw_spinlock_t rt_runtime_lock;
133 struct hrtimer rt_period_timer;
136 void __dl_clear_params(struct task_struct *p);
139 * To keep the bandwidth of -deadline tasks and groups under control
140 * we need some place where:
141 * - store the maximum -deadline bandwidth of the system (the group);
142 * - cache the fraction of that bandwidth that is currently allocated.
144 * This is all done in the data structure below. It is similar to the
145 * one used for RT-throttling (rt_bandwidth), with the main difference
146 * that, since here we are only interested in admission control, we
147 * do not decrease any runtime while the group "executes", neither we
148 * need a timer to replenish it.
150 * With respect to SMP, the bandwidth is given on a per-CPU basis,
152 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
153 * - dl_total_bw array contains, in the i-eth element, the currently
154 * allocated bandwidth on the i-eth CPU.
155 * Moreover, groups consume bandwidth on each CPU, while tasks only
156 * consume bandwidth on the CPU they're running on.
157 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
158 * that will be shown the next time the proc or cgroup controls will
159 * be red. It on its turn can be changed by writing on its own
162 struct dl_bandwidth {
163 raw_spinlock_t dl_runtime_lock;
168 static inline int dl_bandwidth_enabled(void)
170 return sysctl_sched_rt_runtime >= 0;
173 extern struct dl_bw *dl_bw_of(int i);
181 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
183 dl_b->total_bw -= tsk_bw;
187 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
189 dl_b->total_bw += tsk_bw;
193 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
195 return dl_b->bw != -1 &&
196 dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
199 extern struct mutex sched_domains_mutex;
201 #ifdef CONFIG_CGROUP_SCHED
203 #include <linux/cgroup.h>
208 extern struct list_head task_groups;
210 struct cfs_bandwidth {
211 #ifdef CONFIG_CFS_BANDWIDTH
215 s64 hierarchical_quota;
218 int idle, timer_active;
219 struct hrtimer period_timer, slack_timer;
220 struct list_head throttled_cfs_rq;
223 int nr_periods, nr_throttled;
228 /* task group related information */
230 struct cgroup_subsys_state css;
232 #ifdef CONFIG_FAIR_GROUP_SCHED
233 /* schedulable entities of this group on each cpu */
234 struct sched_entity **se;
235 /* runqueue "owned" by this group on each cpu */
236 struct cfs_rq **cfs_rq;
237 unsigned long shares;
240 atomic_long_t load_avg;
241 atomic_t runnable_avg;
245 #ifdef CONFIG_RT_GROUP_SCHED
246 struct sched_rt_entity **rt_se;
247 struct rt_rq **rt_rq;
249 struct rt_bandwidth rt_bandwidth;
253 struct list_head list;
255 struct task_group *parent;
256 struct list_head siblings;
257 struct list_head children;
259 #ifdef CONFIG_SCHED_AUTOGROUP
260 struct autogroup *autogroup;
263 struct cfs_bandwidth cfs_bandwidth;
266 #ifdef CONFIG_FAIR_GROUP_SCHED
267 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
270 * A weight of 0 or 1 can cause arithmetics problems.
271 * A weight of a cfs_rq is the sum of weights of which entities
272 * are queued on this cfs_rq, so a weight of a entity should not be
273 * too large, so as the shares value of a task group.
274 * (The default weight is 1024 - so there's no practical
275 * limitation from this.)
277 #define MIN_SHARES (1UL << 1)
278 #define MAX_SHARES (1UL << 18)
281 typedef int (*tg_visitor)(struct task_group *, void *);
283 extern int walk_tg_tree_from(struct task_group *from,
284 tg_visitor down, tg_visitor up, void *data);
287 * Iterate the full tree, calling @down when first entering a node and @up when
288 * leaving it for the final time.
290 * Caller must hold rcu_lock or sufficient equivalent.
292 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
294 return walk_tg_tree_from(&root_task_group, down, up, data);
297 extern int tg_nop(struct task_group *tg, void *data);
299 extern void free_fair_sched_group(struct task_group *tg);
300 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
301 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
302 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
303 struct sched_entity *se, int cpu,
304 struct sched_entity *parent);
305 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
306 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
308 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
309 extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force);
310 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
312 extern void free_rt_sched_group(struct task_group *tg);
313 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
314 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
315 struct sched_rt_entity *rt_se, int cpu,
316 struct sched_rt_entity *parent);
318 extern struct task_group *sched_create_group(struct task_group *parent);
319 extern void sched_online_group(struct task_group *tg,
320 struct task_group *parent);
321 extern void sched_destroy_group(struct task_group *tg);
322 extern void sched_offline_group(struct task_group *tg);
324 extern void sched_move_task(struct task_struct *tsk);
326 #ifdef CONFIG_FAIR_GROUP_SCHED
327 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
330 #else /* CONFIG_CGROUP_SCHED */
332 struct cfs_bandwidth { };
334 #endif /* CONFIG_CGROUP_SCHED */
336 /* CFS-related fields in a runqueue */
338 struct load_weight load;
339 unsigned int nr_running, h_nr_running;
344 u64 min_vruntime_copy;
347 struct rb_root tasks_timeline;
348 struct rb_node *rb_leftmost;
351 * 'curr' points to currently running entity on this cfs_rq.
352 * It is set to NULL otherwise (i.e when none are currently running).
354 struct sched_entity *curr, *next, *last, *skip;
356 #ifdef CONFIG_SCHED_DEBUG
357 unsigned int nr_spread_over;
363 * Under CFS, load is tracked on a per-entity basis and aggregated up.
364 * This allows for the description of both thread and group usage (in
365 * the FAIR_GROUP_SCHED case).
366 * runnable_load_avg is the sum of the load_avg_contrib of the
367 * sched_entities on the rq.
368 * blocked_load_avg is similar to runnable_load_avg except that its
369 * the blocked sched_entities on the rq.
370 * utilization_load_avg is the sum of the average running time of the
371 * sched_entities on the rq.
373 unsigned long runnable_load_avg, blocked_load_avg, utilization_load_avg;
374 atomic64_t decay_counter;
376 atomic_long_t removed_load;
378 #ifdef CONFIG_FAIR_GROUP_SCHED
379 /* Required to track per-cpu representation of a task_group */
380 u32 tg_runnable_contrib;
381 unsigned long tg_load_contrib;
384 * h_load = weight * f(tg)
386 * Where f(tg) is the recursive weight fraction assigned to
389 unsigned long h_load;
390 u64 last_h_load_update;
391 struct sched_entity *h_load_next;
392 #endif /* CONFIG_FAIR_GROUP_SCHED */
393 #endif /* CONFIG_SMP */
395 #ifdef CONFIG_FAIR_GROUP_SCHED
396 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
399 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
400 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
401 * (like users, containers etc.)
403 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
404 * list is used during load balance.
407 struct list_head leaf_cfs_rq_list;
408 struct task_group *tg; /* group that "owns" this runqueue */
410 #ifdef CONFIG_CFS_BANDWIDTH
413 s64 runtime_remaining;
415 u64 throttled_clock, throttled_clock_task;
416 u64 throttled_clock_task_time;
417 int throttled, throttle_count;
418 struct list_head throttled_list;
419 #endif /* CONFIG_CFS_BANDWIDTH */
420 #endif /* CONFIG_FAIR_GROUP_SCHED */
423 static inline int rt_bandwidth_enabled(void)
425 return sysctl_sched_rt_runtime >= 0;
428 /* RT IPI pull logic requires IRQ_WORK */
429 #ifdef CONFIG_IRQ_WORK
430 # define HAVE_RT_PUSH_IPI
433 /* Real-Time classes' related field in a runqueue: */
435 struct rt_prio_array active;
436 unsigned int rt_nr_running;
437 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
439 int curr; /* highest queued rt task prio */
441 int next; /* next highest */
446 unsigned long rt_nr_migratory;
447 unsigned long rt_nr_total;
449 struct plist_head pushable_tasks;
450 #ifdef HAVE_RT_PUSH_IPI
453 struct irq_work push_work;
454 raw_spinlock_t push_lock;
456 #endif /* CONFIG_SMP */
462 /* Nests inside the rq lock: */
463 raw_spinlock_t rt_runtime_lock;
465 #ifdef CONFIG_RT_GROUP_SCHED
466 unsigned long rt_nr_boosted;
469 struct task_group *tg;
473 /* Deadline class' related fields in a runqueue */
475 /* runqueue is an rbtree, ordered by deadline */
476 struct rb_root rb_root;
477 struct rb_node *rb_leftmost;
479 unsigned long dl_nr_running;
483 * Deadline values of the currently executing and the
484 * earliest ready task on this rq. Caching these facilitates
485 * the decision wether or not a ready but not running task
486 * should migrate somewhere else.
493 unsigned long dl_nr_migratory;
497 * Tasks on this rq that can be pushed away. They are kept in
498 * an rb-tree, ordered by tasks' deadlines, with caching
499 * of the leftmost (earliest deadline) element.
501 struct rb_root pushable_dl_tasks_root;
502 struct rb_node *pushable_dl_tasks_leftmost;
511 * We add the notion of a root-domain which will be used to define per-domain
512 * variables. Each exclusive cpuset essentially defines an island domain by
513 * fully partitioning the member cpus from any other cpuset. Whenever a new
514 * exclusive cpuset is created, we also create and attach a new root-domain
523 cpumask_var_t online;
525 /* Indicate more than one runnable task for any CPU */
529 * The bit corresponding to a CPU gets set here if such CPU has more
530 * than one runnable -deadline task (as it is below for RT tasks).
532 cpumask_var_t dlo_mask;
538 * The "RT overload" flag: it gets set if a CPU has more than
539 * one runnable RT task.
541 cpumask_var_t rto_mask;
542 struct cpupri cpupri;
545 extern struct root_domain def_root_domain;
547 #endif /* CONFIG_SMP */
550 * This is the main, per-CPU runqueue data structure.
552 * Locking rule: those places that want to lock multiple runqueues
553 * (such as the load balancing or the thread migration code), lock
554 * acquire operations must be ordered by ascending &runqueue.
561 * nr_running and cpu_load should be in the same cacheline because
562 * remote CPUs use both these fields when doing load calculation.
564 unsigned int nr_running;
565 #ifdef CONFIG_NUMA_BALANCING
566 unsigned int nr_numa_running;
567 unsigned int nr_preferred_running;
569 #define CPU_LOAD_IDX_MAX 5
570 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
571 unsigned long last_load_update_tick;
572 #ifdef CONFIG_NO_HZ_COMMON
574 unsigned long nohz_flags;
576 #ifdef CONFIG_NO_HZ_FULL
577 unsigned long last_sched_tick;
579 /* capture load from *all* tasks on this cpu: */
580 struct load_weight load;
581 unsigned long nr_load_updates;
588 #ifdef CONFIG_FAIR_GROUP_SCHED
589 /* list of leaf cfs_rq on this cpu: */
590 struct list_head leaf_cfs_rq_list;
592 struct sched_avg avg;
593 #endif /* CONFIG_FAIR_GROUP_SCHED */
596 * This is part of a global counter where only the total sum
597 * over all CPUs matters. A task can increase this counter on
598 * one CPU and if it got migrated afterwards it may decrease
599 * it on another CPU. Always updated under the runqueue lock:
601 unsigned long nr_uninterruptible;
603 struct task_struct *curr, *idle, *stop;
604 unsigned long next_balance;
605 struct mm_struct *prev_mm;
607 unsigned int clock_skip_update;
614 struct root_domain *rd;
615 struct sched_domain *sd;
617 unsigned long cpu_capacity;
618 unsigned long cpu_capacity_orig;
620 unsigned char idle_balance;
621 /* For active balancing */
625 struct cpu_stop_work active_balance_work;
626 /* cpu of this runqueue: */
630 struct list_head cfs_tasks;
637 /* This is used to determine avg_idle's max value */
638 u64 max_idle_balance_cost;
641 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
644 #ifdef CONFIG_PARAVIRT
647 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
648 u64 prev_steal_time_rq;
651 /* calc_load related fields */
652 unsigned long calc_load_update;
653 long calc_load_active;
655 #ifdef CONFIG_SCHED_HRTICK
657 int hrtick_csd_pending;
658 struct call_single_data hrtick_csd;
660 struct hrtimer hrtick_timer;
663 #ifdef CONFIG_SCHEDSTATS
665 struct sched_info rq_sched_info;
666 unsigned long long rq_cpu_time;
667 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
669 /* sys_sched_yield() stats */
670 unsigned int yld_count;
672 /* schedule() stats */
673 unsigned int sched_count;
674 unsigned int sched_goidle;
676 /* try_to_wake_up() stats */
677 unsigned int ttwu_count;
678 unsigned int ttwu_local;
682 struct llist_head wake_list;
685 #ifdef CONFIG_CPU_IDLE
686 /* Must be inspected within a rcu lock section */
687 struct cpuidle_state *idle_state;
691 static inline int cpu_of(struct rq *rq)
700 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
702 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
703 #define this_rq() this_cpu_ptr(&runqueues)
704 #define task_rq(p) cpu_rq(task_cpu(p))
705 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
706 #define raw_rq() raw_cpu_ptr(&runqueues)
708 static inline u64 __rq_clock_broken(struct rq *rq)
710 return ACCESS_ONCE(rq->clock);
713 static inline u64 rq_clock(struct rq *rq)
715 lockdep_assert_held(&rq->lock);
719 static inline u64 rq_clock_task(struct rq *rq)
721 lockdep_assert_held(&rq->lock);
722 return rq->clock_task;
725 #define RQCF_REQ_SKIP 0x01
726 #define RQCF_ACT_SKIP 0x02
728 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
730 lockdep_assert_held(&rq->lock);
732 rq->clock_skip_update |= RQCF_REQ_SKIP;
734 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
738 enum numa_topology_type {
743 extern enum numa_topology_type sched_numa_topology_type;
744 extern int sched_max_numa_distance;
745 extern bool find_numa_distance(int distance);
748 #ifdef CONFIG_NUMA_BALANCING
749 /* The regions in numa_faults array from task_struct */
750 enum numa_faults_stats {
756 extern void sched_setnuma(struct task_struct *p, int node);
757 extern int migrate_task_to(struct task_struct *p, int cpu);
758 extern int migrate_swap(struct task_struct *, struct task_struct *);
759 #endif /* CONFIG_NUMA_BALANCING */
763 extern void sched_ttwu_pending(void);
765 #define rcu_dereference_check_sched_domain(p) \
766 rcu_dereference_check((p), \
767 lockdep_is_held(&sched_domains_mutex))
770 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
771 * See detach_destroy_domains: synchronize_sched for details.
773 * The domain tree of any CPU may only be accessed from within
774 * preempt-disabled sections.
776 #define for_each_domain(cpu, __sd) \
777 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
778 __sd; __sd = __sd->parent)
780 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
783 * highest_flag_domain - Return highest sched_domain containing flag.
784 * @cpu: The cpu whose highest level of sched domain is to
786 * @flag: The flag to check for the highest sched_domain
789 * Returns the highest sched_domain of a cpu which contains the given flag.
791 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
793 struct sched_domain *sd, *hsd = NULL;
795 for_each_domain(cpu, sd) {
796 if (!(sd->flags & flag))
804 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
806 struct sched_domain *sd;
808 for_each_domain(cpu, sd) {
809 if (sd->flags & flag)
816 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
817 DECLARE_PER_CPU(int, sd_llc_size);
818 DECLARE_PER_CPU(int, sd_llc_id);
819 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
820 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
821 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
823 struct sched_group_capacity {
826 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
829 unsigned int capacity;
830 unsigned long next_update;
831 int imbalance; /* XXX unrelated to capacity but shared group state */
833 * Number of busy cpus in this group.
835 atomic_t nr_busy_cpus;
837 unsigned long cpumask[0]; /* iteration mask */
841 struct sched_group *next; /* Must be a circular list */
844 unsigned int group_weight;
845 struct sched_group_capacity *sgc;
848 * The CPUs this group covers.
850 * NOTE: this field is variable length. (Allocated dynamically
851 * by attaching extra space to the end of the structure,
852 * depending on how many CPUs the kernel has booted up with)
854 unsigned long cpumask[0];
857 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
859 return to_cpumask(sg->cpumask);
863 * cpumask masking which cpus in the group are allowed to iterate up the domain
866 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
868 return to_cpumask(sg->sgc->cpumask);
872 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
873 * @group: The group whose first cpu is to be returned.
875 static inline unsigned int group_first_cpu(struct sched_group *group)
877 return cpumask_first(sched_group_cpus(group));
880 extern int group_balance_cpu(struct sched_group *sg);
884 static inline void sched_ttwu_pending(void) { }
886 #endif /* CONFIG_SMP */
889 #include "auto_group.h"
891 #ifdef CONFIG_CGROUP_SCHED
894 * Return the group to which this tasks belongs.
896 * We cannot use task_css() and friends because the cgroup subsystem
897 * changes that value before the cgroup_subsys::attach() method is called,
898 * therefore we cannot pin it and might observe the wrong value.
900 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
901 * core changes this before calling sched_move_task().
903 * Instead we use a 'copy' which is updated from sched_move_task() while
904 * holding both task_struct::pi_lock and rq::lock.
906 static inline struct task_group *task_group(struct task_struct *p)
908 return p->sched_task_group;
911 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
912 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
914 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
915 struct task_group *tg = task_group(p);
918 #ifdef CONFIG_FAIR_GROUP_SCHED
919 p->se.cfs_rq = tg->cfs_rq[cpu];
920 p->se.parent = tg->se[cpu];
923 #ifdef CONFIG_RT_GROUP_SCHED
924 p->rt.rt_rq = tg->rt_rq[cpu];
925 p->rt.parent = tg->rt_se[cpu];
929 #else /* CONFIG_CGROUP_SCHED */
931 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
932 static inline struct task_group *task_group(struct task_struct *p)
937 #endif /* CONFIG_CGROUP_SCHED */
939 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
944 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
945 * successfuly executed on another CPU. We must ensure that updates of
946 * per-task data have been completed by this moment.
949 task_thread_info(p)->cpu = cpu;
955 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
957 #ifdef CONFIG_SCHED_DEBUG
958 # include <linux/static_key.h>
959 # define const_debug __read_mostly
961 # define const_debug const
964 extern const_debug unsigned int sysctl_sched_features;
966 #define SCHED_FEAT(name, enabled) \
967 __SCHED_FEAT_##name ,
970 #include "features.h"
976 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
977 #define SCHED_FEAT(name, enabled) \
978 static __always_inline bool static_branch_##name(struct static_key *key) \
980 return static_key_##enabled(key); \
983 #include "features.h"
987 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
988 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
989 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
990 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
991 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
993 #ifdef CONFIG_NUMA_BALANCING
994 #define sched_feat_numa(x) sched_feat(x)
995 #ifdef CONFIG_SCHED_DEBUG
996 #define numabalancing_enabled sched_feat_numa(NUMA)
998 extern bool numabalancing_enabled;
999 #endif /* CONFIG_SCHED_DEBUG */
1001 #define sched_feat_numa(x) (0)
1002 #define numabalancing_enabled (0)
1003 #endif /* CONFIG_NUMA_BALANCING */
1005 static inline u64 global_rt_period(void)
1007 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1010 static inline u64 global_rt_runtime(void)
1012 if (sysctl_sched_rt_runtime < 0)
1015 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1018 static inline int task_current(struct rq *rq, struct task_struct *p)
1020 return rq->curr == p;
1023 static inline int task_running(struct rq *rq, struct task_struct *p)
1028 return task_current(rq, p);
1032 static inline int task_on_rq_queued(struct task_struct *p)
1034 return p->on_rq == TASK_ON_RQ_QUEUED;
1037 static inline int task_on_rq_migrating(struct task_struct *p)
1039 return p->on_rq == TASK_ON_RQ_MIGRATING;
1042 #ifndef prepare_arch_switch
1043 # define prepare_arch_switch(next) do { } while (0)
1045 #ifndef finish_arch_switch
1046 # define finish_arch_switch(prev) do { } while (0)
1048 #ifndef finish_arch_post_lock_switch
1049 # define finish_arch_post_lock_switch() do { } while (0)
1052 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1056 * We can optimise this out completely for !SMP, because the
1057 * SMP rebalancing from interrupt is the only thing that cares
1064 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1068 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1069 * We must ensure this doesn't happen until the switch is completely
1075 #ifdef CONFIG_DEBUG_SPINLOCK
1076 /* this is a valid case when another task releases the spinlock */
1077 rq->lock.owner = current;
1080 * If we are tracking spinlock dependencies then we have to
1081 * fix up the runqueue lock - which gets 'carried over' from
1082 * prev into current:
1084 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1086 raw_spin_unlock_irq(&rq->lock);
1092 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1093 #define WF_FORK 0x02 /* child wakeup after fork */
1094 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1097 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1098 * of tasks with abnormal "nice" values across CPUs the contribution that
1099 * each task makes to its run queue's load is weighted according to its
1100 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1101 * scaled version of the new time slice allocation that they receive on time
1105 #define WEIGHT_IDLEPRIO 3
1106 #define WMULT_IDLEPRIO 1431655765
1109 * Nice levels are multiplicative, with a gentle 10% change for every
1110 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1111 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1112 * that remained on nice 0.
1114 * The "10% effect" is relative and cumulative: from _any_ nice level,
1115 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1116 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1117 * If a task goes up by ~10% and another task goes down by ~10% then
1118 * the relative distance between them is ~25%.)
1120 static const int prio_to_weight[40] = {
1121 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1122 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1123 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1124 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1125 /* 0 */ 1024, 820, 655, 526, 423,
1126 /* 5 */ 335, 272, 215, 172, 137,
1127 /* 10 */ 110, 87, 70, 56, 45,
1128 /* 15 */ 36, 29, 23, 18, 15,
1132 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1134 * In cases where the weight does not change often, we can use the
1135 * precalculated inverse to speed up arithmetics by turning divisions
1136 * into multiplications:
1138 static const u32 prio_to_wmult[40] = {
1139 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1140 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1141 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1142 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1143 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1144 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1145 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1146 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1149 #define ENQUEUE_WAKEUP 1
1150 #define ENQUEUE_HEAD 2
1152 #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */
1154 #define ENQUEUE_WAKING 0
1156 #define ENQUEUE_REPLENISH 8
1158 #define DEQUEUE_SLEEP 1
1160 #define RETRY_TASK ((void *)-1UL)
1162 struct sched_class {
1163 const struct sched_class *next;
1165 void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1166 void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1167 void (*yield_task) (struct rq *rq);
1168 bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1170 void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1173 * It is the responsibility of the pick_next_task() method that will
1174 * return the next task to call put_prev_task() on the @prev task or
1175 * something equivalent.
1177 * May return RETRY_TASK when it finds a higher prio class has runnable
1180 struct task_struct * (*pick_next_task) (struct rq *rq,
1181 struct task_struct *prev);
1182 void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1185 int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1186 void (*migrate_task_rq)(struct task_struct *p, int next_cpu);
1188 void (*post_schedule) (struct rq *this_rq);
1189 void (*task_waking) (struct task_struct *task);
1190 void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1192 void (*set_cpus_allowed)(struct task_struct *p,
1193 const struct cpumask *newmask);
1195 void (*rq_online)(struct rq *rq);
1196 void (*rq_offline)(struct rq *rq);
1199 void (*set_curr_task) (struct rq *rq);
1200 void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1201 void (*task_fork) (struct task_struct *p);
1202 void (*task_dead) (struct task_struct *p);
1205 * The switched_from() call is allowed to drop rq->lock, therefore we
1206 * cannot assume the switched_from/switched_to pair is serliazed by
1207 * rq->lock. They are however serialized by p->pi_lock.
1209 void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1210 void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1211 void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1214 unsigned int (*get_rr_interval) (struct rq *rq,
1215 struct task_struct *task);
1217 void (*update_curr) (struct rq *rq);
1219 #ifdef CONFIG_FAIR_GROUP_SCHED
1220 void (*task_move_group) (struct task_struct *p, int on_rq);
1224 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1226 prev->sched_class->put_prev_task(rq, prev);
1229 #define sched_class_highest (&stop_sched_class)
1230 #define for_each_class(class) \
1231 for (class = sched_class_highest; class; class = class->next)
1233 extern const struct sched_class stop_sched_class;
1234 extern const struct sched_class dl_sched_class;
1235 extern const struct sched_class rt_sched_class;
1236 extern const struct sched_class fair_sched_class;
1237 extern const struct sched_class idle_sched_class;
1242 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1244 extern void trigger_load_balance(struct rq *rq);
1246 extern void idle_enter_fair(struct rq *this_rq);
1247 extern void idle_exit_fair(struct rq *this_rq);
1251 static inline void idle_enter_fair(struct rq *rq) { }
1252 static inline void idle_exit_fair(struct rq *rq) { }
1256 #ifdef CONFIG_CPU_IDLE
1257 static inline void idle_set_state(struct rq *rq,
1258 struct cpuidle_state *idle_state)
1260 rq->idle_state = idle_state;
1263 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1265 WARN_ON(!rcu_read_lock_held());
1266 return rq->idle_state;
1269 static inline void idle_set_state(struct rq *rq,
1270 struct cpuidle_state *idle_state)
1274 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1280 extern void sysrq_sched_debug_show(void);
1281 extern void sched_init_granularity(void);
1282 extern void update_max_interval(void);
1284 extern void init_sched_dl_class(void);
1285 extern void init_sched_rt_class(void);
1286 extern void init_sched_fair_class(void);
1287 extern void init_sched_dl_class(void);
1289 extern void resched_curr(struct rq *rq);
1290 extern void resched_cpu(int cpu);
1292 extern struct rt_bandwidth def_rt_bandwidth;
1293 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1295 extern struct dl_bandwidth def_dl_bandwidth;
1296 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1297 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1299 unsigned long to_ratio(u64 period, u64 runtime);
1301 extern void update_idle_cpu_load(struct rq *this_rq);
1303 extern void init_task_runnable_average(struct task_struct *p);
1305 static inline void add_nr_running(struct rq *rq, unsigned count)
1307 unsigned prev_nr = rq->nr_running;
1309 rq->nr_running = prev_nr + count;
1311 if (prev_nr < 2 && rq->nr_running >= 2) {
1313 if (!rq->rd->overload)
1314 rq->rd->overload = true;
1317 #ifdef CONFIG_NO_HZ_FULL
1318 if (tick_nohz_full_cpu(rq->cpu)) {
1320 * Tick is needed if more than one task runs on a CPU.
1321 * Send the target an IPI to kick it out of nohz mode.
1323 * We assume that IPI implies full memory barrier and the
1324 * new value of rq->nr_running is visible on reception
1327 tick_nohz_full_kick_cpu(rq->cpu);
1333 static inline void sub_nr_running(struct rq *rq, unsigned count)
1335 rq->nr_running -= count;
1338 static inline void rq_last_tick_reset(struct rq *rq)
1340 #ifdef CONFIG_NO_HZ_FULL
1341 rq->last_sched_tick = jiffies;
1345 extern void update_rq_clock(struct rq *rq);
1347 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1348 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1350 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1352 extern const_debug unsigned int sysctl_sched_time_avg;
1353 extern const_debug unsigned int sysctl_sched_nr_migrate;
1354 extern const_debug unsigned int sysctl_sched_migration_cost;
1356 static inline u64 sched_avg_period(void)
1358 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1361 #ifdef CONFIG_SCHED_HRTICK
1365 * - enabled by features
1366 * - hrtimer is actually high res
1368 static inline int hrtick_enabled(struct rq *rq)
1370 if (!sched_feat(HRTICK))
1372 if (!cpu_active(cpu_of(rq)))
1374 return hrtimer_is_hres_active(&rq->hrtick_timer);
1377 void hrtick_start(struct rq *rq, u64 delay);
1381 static inline int hrtick_enabled(struct rq *rq)
1386 #endif /* CONFIG_SCHED_HRTICK */
1389 extern void sched_avg_update(struct rq *rq);
1391 #ifndef arch_scale_freq_capacity
1392 static __always_inline
1393 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1395 return SCHED_CAPACITY_SCALE;
1399 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1401 rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1402 sched_avg_update(rq);
1405 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1406 static inline void sched_avg_update(struct rq *rq) { }
1409 extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period);
1412 * __task_rq_lock - lock the rq @p resides on.
1414 static inline struct rq *__task_rq_lock(struct task_struct *p)
1415 __acquires(rq->lock)
1419 lockdep_assert_held(&p->pi_lock);
1423 raw_spin_lock(&rq->lock);
1424 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p)))
1426 raw_spin_unlock(&rq->lock);
1428 while (unlikely(task_on_rq_migrating(p)))
1434 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1436 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1437 __acquires(p->pi_lock)
1438 __acquires(rq->lock)
1443 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1445 raw_spin_lock(&rq->lock);
1447 * move_queued_task() task_rq_lock()
1449 * ACQUIRE (rq->lock)
1450 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1451 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1452 * [S] ->cpu = new_cpu [L] task_rq()
1454 * RELEASE (rq->lock)
1456 * If we observe the old cpu in task_rq_lock, the acquire of
1457 * the old rq->lock will fully serialize against the stores.
1459 * If we observe the new cpu in task_rq_lock, the acquire will
1460 * pair with the WMB to ensure we must then also see migrating.
1462 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p)))
1464 raw_spin_unlock(&rq->lock);
1465 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1467 while (unlikely(task_on_rq_migrating(p)))
1472 static inline void __task_rq_unlock(struct rq *rq)
1473 __releases(rq->lock)
1475 raw_spin_unlock(&rq->lock);
1479 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1480 __releases(rq->lock)
1481 __releases(p->pi_lock)
1483 raw_spin_unlock(&rq->lock);
1484 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1488 #ifdef CONFIG_PREEMPT
1490 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1493 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1494 * way at the expense of forcing extra atomic operations in all
1495 * invocations. This assures that the double_lock is acquired using the
1496 * same underlying policy as the spinlock_t on this architecture, which
1497 * reduces latency compared to the unfair variant below. However, it
1498 * also adds more overhead and therefore may reduce throughput.
1500 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1501 __releases(this_rq->lock)
1502 __acquires(busiest->lock)
1503 __acquires(this_rq->lock)
1505 raw_spin_unlock(&this_rq->lock);
1506 double_rq_lock(this_rq, busiest);
1513 * Unfair double_lock_balance: Optimizes throughput at the expense of
1514 * latency by eliminating extra atomic operations when the locks are
1515 * already in proper order on entry. This favors lower cpu-ids and will
1516 * grant the double lock to lower cpus over higher ids under contention,
1517 * regardless of entry order into the function.
1519 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1520 __releases(this_rq->lock)
1521 __acquires(busiest->lock)
1522 __acquires(this_rq->lock)
1526 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1527 if (busiest < this_rq) {
1528 raw_spin_unlock(&this_rq->lock);
1529 raw_spin_lock(&busiest->lock);
1530 raw_spin_lock_nested(&this_rq->lock,
1531 SINGLE_DEPTH_NESTING);
1534 raw_spin_lock_nested(&busiest->lock,
1535 SINGLE_DEPTH_NESTING);
1540 #endif /* CONFIG_PREEMPT */
1543 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1545 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1547 if (unlikely(!irqs_disabled())) {
1548 /* printk() doesn't work good under rq->lock */
1549 raw_spin_unlock(&this_rq->lock);
1553 return _double_lock_balance(this_rq, busiest);
1556 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1557 __releases(busiest->lock)
1559 raw_spin_unlock(&busiest->lock);
1560 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1563 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1569 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1572 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1578 spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1581 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1587 raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1591 * double_rq_lock - safely lock two runqueues
1593 * Note this does not disable interrupts like task_rq_lock,
1594 * you need to do so manually before calling.
1596 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1597 __acquires(rq1->lock)
1598 __acquires(rq2->lock)
1600 BUG_ON(!irqs_disabled());
1602 raw_spin_lock(&rq1->lock);
1603 __acquire(rq2->lock); /* Fake it out ;) */
1606 raw_spin_lock(&rq1->lock);
1607 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1609 raw_spin_lock(&rq2->lock);
1610 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1616 * double_rq_unlock - safely unlock two runqueues
1618 * Note this does not restore interrupts like task_rq_unlock,
1619 * you need to do so manually after calling.
1621 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1622 __releases(rq1->lock)
1623 __releases(rq2->lock)
1625 raw_spin_unlock(&rq1->lock);
1627 raw_spin_unlock(&rq2->lock);
1629 __release(rq2->lock);
1632 #else /* CONFIG_SMP */
1635 * double_rq_lock - safely lock two runqueues
1637 * Note this does not disable interrupts like task_rq_lock,
1638 * you need to do so manually before calling.
1640 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1641 __acquires(rq1->lock)
1642 __acquires(rq2->lock)
1644 BUG_ON(!irqs_disabled());
1646 raw_spin_lock(&rq1->lock);
1647 __acquire(rq2->lock); /* Fake it out ;) */
1651 * double_rq_unlock - safely unlock two runqueues
1653 * Note this does not restore interrupts like task_rq_unlock,
1654 * you need to do so manually after calling.
1656 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1657 __releases(rq1->lock)
1658 __releases(rq2->lock)
1661 raw_spin_unlock(&rq1->lock);
1662 __release(rq2->lock);
1667 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1668 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1669 extern void print_cfs_stats(struct seq_file *m, int cpu);
1670 extern void print_rt_stats(struct seq_file *m, int cpu);
1671 extern void print_dl_stats(struct seq_file *m, int cpu);
1673 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1674 extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq);
1675 extern void init_dl_rq(struct dl_rq *dl_rq, struct rq *rq);
1677 extern void cfs_bandwidth_usage_inc(void);
1678 extern void cfs_bandwidth_usage_dec(void);
1680 #ifdef CONFIG_NO_HZ_COMMON
1681 enum rq_nohz_flag_bits {
1686 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1689 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1691 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1692 DECLARE_PER_CPU(u64, cpu_softirq_time);
1694 #ifndef CONFIG_64BIT
1695 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1697 static inline void irq_time_write_begin(void)
1699 __this_cpu_inc(irq_time_seq.sequence);
1703 static inline void irq_time_write_end(void)
1706 __this_cpu_inc(irq_time_seq.sequence);
1709 static inline u64 irq_time_read(int cpu)
1715 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1716 irq_time = per_cpu(cpu_softirq_time, cpu) +
1717 per_cpu(cpu_hardirq_time, cpu);
1718 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1722 #else /* CONFIG_64BIT */
1723 static inline void irq_time_write_begin(void)
1727 static inline void irq_time_write_end(void)
1731 static inline u64 irq_time_read(int cpu)
1733 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1735 #endif /* CONFIG_64BIT */
1736 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */