2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/u64_stats_sync.h>
6 #include <linux/sched/deadline.h>
7 #include <linux/kernel_stat.h>
8 #include <linux/binfmts.h>
9 #include <linux/mutex.h>
10 #include <linux/spinlock.h>
11 #include <linux/stop_machine.h>
12 #include <linux/irq_work.h>
13 #include <linux/tick.h>
14 #include <linux/slab.h>
17 #include "cpudeadline.h"
20 #ifdef CONFIG_SCHED_DEBUG
21 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
23 #define SCHED_WARN_ON(x) ((void)(x))
29 /* task_struct::on_rq states: */
30 #define TASK_ON_RQ_QUEUED 1
31 #define TASK_ON_RQ_MIGRATING 2
33 extern __read_mostly
int scheduler_running
;
35 extern unsigned long calc_load_update
;
36 extern atomic_long_t calc_load_tasks
;
38 extern void calc_global_load_tick(struct rq
*this_rq
);
39 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
42 extern void cpu_load_update_active(struct rq
*this_rq
);
44 static inline void cpu_load_update_active(struct rq
*this_rq
) { }
48 * Helpers for converting nanosecond timing to jiffy resolution
50 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
53 * Increase resolution of nice-level calculations for 64-bit architectures.
54 * The extra resolution improves shares distribution and load balancing of
55 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
56 * hierarchies, especially on larger systems. This is not a user-visible change
57 * and does not change the user-interface for setting shares/weights.
59 * We increase resolution only if we have enough bits to allow this increased
60 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
61 * pretty high and the returns do not justify the increased costs.
63 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
64 * increase coverage and consistency always enable it on 64bit platforms.
67 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
68 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
69 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
71 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
72 # define scale_load(w) (w)
73 # define scale_load_down(w) (w)
77 * Task weight (visible to users) and its load (invisible to users) have
78 * independent resolution, but they should be well calibrated. We use
79 * scale_load() and scale_load_down(w) to convert between them. The
80 * following must be true:
82 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
85 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
88 * Single value that decides SCHED_DEADLINE internal math precision.
89 * 10 -> just above 1us
90 * 9 -> just above 0.5us
95 * These are the 'tuning knobs' of the scheduler:
99 * single value that denotes runtime == period, ie unlimited time.
101 #define RUNTIME_INF ((u64)~0ULL)
103 static inline int idle_policy(int policy
)
105 return policy
== SCHED_IDLE
;
107 static inline int fair_policy(int policy
)
109 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
112 static inline int rt_policy(int policy
)
114 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
117 static inline int dl_policy(int policy
)
119 return policy
== SCHED_DEADLINE
;
121 static inline bool valid_policy(int policy
)
123 return idle_policy(policy
) || fair_policy(policy
) ||
124 rt_policy(policy
) || dl_policy(policy
);
127 static inline int task_has_rt_policy(struct task_struct
*p
)
129 return rt_policy(p
->policy
);
132 static inline int task_has_dl_policy(struct task_struct
*p
)
134 return dl_policy(p
->policy
);
138 * Tells if entity @a should preempt entity @b.
141 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
143 return dl_time_before(a
->deadline
, b
->deadline
);
147 * This is the priority-queue data structure of the RT scheduling class:
149 struct rt_prio_array
{
150 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
151 struct list_head queue
[MAX_RT_PRIO
];
154 struct rt_bandwidth
{
155 /* nests inside the rq lock: */
156 raw_spinlock_t rt_runtime_lock
;
159 struct hrtimer rt_period_timer
;
160 unsigned int rt_period_active
;
163 void __dl_clear_params(struct task_struct
*p
);
166 * To keep the bandwidth of -deadline tasks and groups under control
167 * we need some place where:
168 * - store the maximum -deadline bandwidth of the system (the group);
169 * - cache the fraction of that bandwidth that is currently allocated.
171 * This is all done in the data structure below. It is similar to the
172 * one used for RT-throttling (rt_bandwidth), with the main difference
173 * that, since here we are only interested in admission control, we
174 * do not decrease any runtime while the group "executes", neither we
175 * need a timer to replenish it.
177 * With respect to SMP, the bandwidth is given on a per-CPU basis,
179 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
180 * - dl_total_bw array contains, in the i-eth element, the currently
181 * allocated bandwidth on the i-eth CPU.
182 * Moreover, groups consume bandwidth on each CPU, while tasks only
183 * consume bandwidth on the CPU they're running on.
184 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
185 * that will be shown the next time the proc or cgroup controls will
186 * be red. It on its turn can be changed by writing on its own
189 struct dl_bandwidth
{
190 raw_spinlock_t dl_runtime_lock
;
195 static inline int dl_bandwidth_enabled(void)
197 return sysctl_sched_rt_runtime
>= 0;
200 extern struct dl_bw
*dl_bw_of(int i
);
208 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
210 dl_b
->total_bw
-= tsk_bw
;
214 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
216 dl_b
->total_bw
+= tsk_bw
;
220 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
222 return dl_b
->bw
!= -1 &&
223 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
226 extern void init_dl_bw(struct dl_bw
*dl_b
);
228 #ifdef CONFIG_CGROUP_SCHED
230 #include <linux/cgroup.h>
235 extern struct list_head task_groups
;
237 struct cfs_bandwidth
{
238 #ifdef CONFIG_CFS_BANDWIDTH
242 s64 hierarchical_quota
;
245 int idle
, period_active
;
246 struct hrtimer period_timer
, slack_timer
;
247 struct list_head throttled_cfs_rq
;
250 int nr_periods
, nr_throttled
;
255 /* task group related information */
257 struct cgroup_subsys_state css
;
259 #ifdef CONFIG_FAIR_GROUP_SCHED
260 /* schedulable entities of this group on each cpu */
261 struct sched_entity
**se
;
262 /* runqueue "owned" by this group on each cpu */
263 struct cfs_rq
**cfs_rq
;
264 unsigned long shares
;
268 * load_avg can be heavily contended at clock tick time, so put
269 * it in its own cacheline separated from the fields above which
270 * will also be accessed at each tick.
272 atomic_long_t load_avg ____cacheline_aligned
;
276 #ifdef CONFIG_RT_GROUP_SCHED
277 struct sched_rt_entity
**rt_se
;
278 struct rt_rq
**rt_rq
;
280 struct rt_bandwidth rt_bandwidth
;
284 struct list_head list
;
286 struct task_group
*parent
;
287 struct list_head siblings
;
288 struct list_head children
;
290 #ifdef CONFIG_SCHED_AUTOGROUP
291 struct autogroup
*autogroup
;
294 struct cfs_bandwidth cfs_bandwidth
;
297 #ifdef CONFIG_FAIR_GROUP_SCHED
298 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
301 * A weight of 0 or 1 can cause arithmetics problems.
302 * A weight of a cfs_rq is the sum of weights of which entities
303 * are queued on this cfs_rq, so a weight of a entity should not be
304 * too large, so as the shares value of a task group.
305 * (The default weight is 1024 - so there's no practical
306 * limitation from this.)
308 #define MIN_SHARES (1UL << 1)
309 #define MAX_SHARES (1UL << 18)
312 typedef int (*tg_visitor
)(struct task_group
*, void *);
314 extern int walk_tg_tree_from(struct task_group
*from
,
315 tg_visitor down
, tg_visitor up
, void *data
);
318 * Iterate the full tree, calling @down when first entering a node and @up when
319 * leaving it for the final time.
321 * Caller must hold rcu_lock or sufficient equivalent.
323 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
325 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
328 extern int tg_nop(struct task_group
*tg
, void *data
);
330 extern void free_fair_sched_group(struct task_group
*tg
);
331 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
332 extern void online_fair_sched_group(struct task_group
*tg
);
333 extern void unregister_fair_sched_group(struct task_group
*tg
);
334 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
335 struct sched_entity
*se
, int cpu
,
336 struct sched_entity
*parent
);
337 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
339 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
340 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
341 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
343 extern void free_rt_sched_group(struct task_group
*tg
);
344 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
345 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
346 struct sched_rt_entity
*rt_se
, int cpu
,
347 struct sched_rt_entity
*parent
);
349 extern struct task_group
*sched_create_group(struct task_group
*parent
);
350 extern void sched_online_group(struct task_group
*tg
,
351 struct task_group
*parent
);
352 extern void sched_destroy_group(struct task_group
*tg
);
353 extern void sched_offline_group(struct task_group
*tg
);
355 extern void sched_move_task(struct task_struct
*tsk
);
357 #ifdef CONFIG_FAIR_GROUP_SCHED
358 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
361 extern void set_task_rq_fair(struct sched_entity
*se
,
362 struct cfs_rq
*prev
, struct cfs_rq
*next
);
363 #else /* !CONFIG_SMP */
364 static inline void set_task_rq_fair(struct sched_entity
*se
,
365 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
366 #endif /* CONFIG_SMP */
367 #endif /* CONFIG_FAIR_GROUP_SCHED */
369 #else /* CONFIG_CGROUP_SCHED */
371 struct cfs_bandwidth
{ };
373 #endif /* CONFIG_CGROUP_SCHED */
375 /* CFS-related fields in a runqueue */
377 struct load_weight load
;
378 unsigned int nr_running
, h_nr_running
;
383 u64 min_vruntime_copy
;
386 struct rb_root tasks_timeline
;
387 struct rb_node
*rb_leftmost
;
390 * 'curr' points to currently running entity on this cfs_rq.
391 * It is set to NULL otherwise (i.e when none are currently running).
393 struct sched_entity
*curr
, *next
, *last
, *skip
;
395 #ifdef CONFIG_SCHED_DEBUG
396 unsigned int nr_spread_over
;
403 struct sched_avg avg
;
404 u64 runnable_load_sum
;
405 unsigned long runnable_load_avg
;
406 #ifdef CONFIG_FAIR_GROUP_SCHED
407 unsigned long tg_load_avg_contrib
;
408 unsigned long propagate_avg
;
410 atomic_long_t removed_load_avg
, removed_util_avg
;
412 u64 load_last_update_time_copy
;
415 #ifdef CONFIG_FAIR_GROUP_SCHED
417 * h_load = weight * f(tg)
419 * Where f(tg) is the recursive weight fraction assigned to
422 unsigned long h_load
;
423 u64 last_h_load_update
;
424 struct sched_entity
*h_load_next
;
425 #endif /* CONFIG_FAIR_GROUP_SCHED */
426 #endif /* CONFIG_SMP */
428 #ifdef CONFIG_FAIR_GROUP_SCHED
429 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
432 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
433 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
434 * (like users, containers etc.)
436 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
437 * list is used during load balance.
440 struct list_head leaf_cfs_rq_list
;
441 struct task_group
*tg
; /* group that "owns" this runqueue */
443 #ifdef CONFIG_CFS_BANDWIDTH
446 s64 runtime_remaining
;
448 u64 throttled_clock
, throttled_clock_task
;
449 u64 throttled_clock_task_time
;
450 int throttled
, throttle_count
;
451 struct list_head throttled_list
;
452 #endif /* CONFIG_CFS_BANDWIDTH */
453 #endif /* CONFIG_FAIR_GROUP_SCHED */
456 static inline int rt_bandwidth_enabled(void)
458 return sysctl_sched_rt_runtime
>= 0;
461 /* RT IPI pull logic requires IRQ_WORK */
462 #ifdef CONFIG_IRQ_WORK
463 # define HAVE_RT_PUSH_IPI
466 /* Real-Time classes' related field in a runqueue: */
468 struct rt_prio_array active
;
469 unsigned int rt_nr_running
;
470 unsigned int rr_nr_running
;
471 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
473 int curr
; /* highest queued rt task prio */
475 int next
; /* next highest */
480 unsigned long rt_nr_migratory
;
481 unsigned long rt_nr_total
;
483 struct plist_head pushable_tasks
;
484 #ifdef HAVE_RT_PUSH_IPI
487 struct irq_work push_work
;
488 raw_spinlock_t push_lock
;
490 #endif /* CONFIG_SMP */
496 /* Nests inside the rq lock: */
497 raw_spinlock_t rt_runtime_lock
;
499 #ifdef CONFIG_RT_GROUP_SCHED
500 unsigned long rt_nr_boosted
;
503 struct task_group
*tg
;
507 /* Deadline class' related fields in a runqueue */
509 /* runqueue is an rbtree, ordered by deadline */
510 struct rb_root rb_root
;
511 struct rb_node
*rb_leftmost
;
513 unsigned long dl_nr_running
;
517 * Deadline values of the currently executing and the
518 * earliest ready task on this rq. Caching these facilitates
519 * the decision wether or not a ready but not running task
520 * should migrate somewhere else.
527 unsigned long dl_nr_migratory
;
531 * Tasks on this rq that can be pushed away. They are kept in
532 * an rb-tree, ordered by tasks' deadlines, with caching
533 * of the leftmost (earliest deadline) element.
535 struct rb_root pushable_dl_tasks_root
;
536 struct rb_node
*pushable_dl_tasks_leftmost
;
544 static inline bool sched_asym_prefer(int a
, int b
)
546 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
550 * We add the notion of a root-domain which will be used to define per-domain
551 * variables. Each exclusive cpuset essentially defines an island domain by
552 * fully partitioning the member cpus from any other cpuset. Whenever a new
553 * exclusive cpuset is created, we also create and attach a new root-domain
562 cpumask_var_t online
;
564 /* Indicate more than one runnable task for any CPU */
568 * The bit corresponding to a CPU gets set here if such CPU has more
569 * than one runnable -deadline task (as it is below for RT tasks).
571 cpumask_var_t dlo_mask
;
577 * The "RT overload" flag: it gets set if a CPU has more than
578 * one runnable RT task.
580 cpumask_var_t rto_mask
;
581 struct cpupri cpupri
;
583 unsigned long max_cpu_capacity
;
586 extern struct root_domain def_root_domain
;
587 extern struct mutex sched_domains_mutex
;
588 extern cpumask_var_t fallback_doms
;
589 extern cpumask_var_t sched_domains_tmpmask
;
591 extern void init_defrootdomain(void);
592 extern int init_sched_domains(const struct cpumask
*cpu_map
);
593 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
595 #endif /* CONFIG_SMP */
598 * This is the main, per-CPU runqueue data structure.
600 * Locking rule: those places that want to lock multiple runqueues
601 * (such as the load balancing or the thread migration code), lock
602 * acquire operations must be ordered by ascending &runqueue.
609 * nr_running and cpu_load should be in the same cacheline because
610 * remote CPUs use both these fields when doing load calculation.
612 unsigned int nr_running
;
613 #ifdef CONFIG_NUMA_BALANCING
614 unsigned int nr_numa_running
;
615 unsigned int nr_preferred_running
;
617 #define CPU_LOAD_IDX_MAX 5
618 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
619 #ifdef CONFIG_NO_HZ_COMMON
621 unsigned long last_load_update_tick
;
622 #endif /* CONFIG_SMP */
623 unsigned long nohz_flags
;
624 #endif /* CONFIG_NO_HZ_COMMON */
625 #ifdef CONFIG_NO_HZ_FULL
626 unsigned long last_sched_tick
;
628 /* capture load from *all* tasks on this cpu: */
629 struct load_weight load
;
630 unsigned long nr_load_updates
;
637 #ifdef CONFIG_FAIR_GROUP_SCHED
638 /* list of leaf cfs_rq on this cpu: */
639 struct list_head leaf_cfs_rq_list
;
640 struct list_head
*tmp_alone_branch
;
641 #endif /* CONFIG_FAIR_GROUP_SCHED */
644 * This is part of a global counter where only the total sum
645 * over all CPUs matters. A task can increase this counter on
646 * one CPU and if it got migrated afterwards it may decrease
647 * it on another CPU. Always updated under the runqueue lock:
649 unsigned long nr_uninterruptible
;
651 struct task_struct
*curr
, *idle
, *stop
;
652 unsigned long next_balance
;
653 struct mm_struct
*prev_mm
;
655 unsigned int clock_update_flags
;
662 struct root_domain
*rd
;
663 struct sched_domain
*sd
;
665 unsigned long cpu_capacity
;
666 unsigned long cpu_capacity_orig
;
668 struct callback_head
*balance_callback
;
670 unsigned char idle_balance
;
671 /* For active balancing */
674 struct cpu_stop_work active_balance_work
;
675 /* cpu of this runqueue: */
679 struct list_head cfs_tasks
;
686 /* This is used to determine avg_idle's max value */
687 u64 max_idle_balance_cost
;
690 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
693 #ifdef CONFIG_PARAVIRT
696 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
697 u64 prev_steal_time_rq
;
700 /* calc_load related fields */
701 unsigned long calc_load_update
;
702 long calc_load_active
;
704 #ifdef CONFIG_SCHED_HRTICK
706 int hrtick_csd_pending
;
707 struct call_single_data hrtick_csd
;
709 struct hrtimer hrtick_timer
;
712 #ifdef CONFIG_SCHEDSTATS
714 struct sched_info rq_sched_info
;
715 unsigned long long rq_cpu_time
;
716 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
718 /* sys_sched_yield() stats */
719 unsigned int yld_count
;
721 /* schedule() stats */
722 unsigned int sched_count
;
723 unsigned int sched_goidle
;
725 /* try_to_wake_up() stats */
726 unsigned int ttwu_count
;
727 unsigned int ttwu_local
;
731 struct llist_head wake_list
;
734 #ifdef CONFIG_CPU_IDLE
735 /* Must be inspected within a rcu lock section */
736 struct cpuidle_state
*idle_state
;
740 static inline int cpu_of(struct rq
*rq
)
750 #ifdef CONFIG_SCHED_SMT
752 extern struct static_key_false sched_smt_present
;
754 extern void __update_idle_core(struct rq
*rq
);
756 static inline void update_idle_core(struct rq
*rq
)
758 if (static_branch_unlikely(&sched_smt_present
))
759 __update_idle_core(rq
);
763 static inline void update_idle_core(struct rq
*rq
) { }
766 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
768 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
769 #define this_rq() this_cpu_ptr(&runqueues)
770 #define task_rq(p) cpu_rq(task_cpu(p))
771 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
772 #define raw_rq() raw_cpu_ptr(&runqueues)
774 static inline u64
__rq_clock_broken(struct rq
*rq
)
776 return READ_ONCE(rq
->clock
);
780 * rq::clock_update_flags bits
782 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
783 * call to __schedule(). This is an optimisation to avoid
784 * neighbouring rq clock updates.
786 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
787 * in effect and calls to update_rq_clock() are being ignored.
789 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
790 * made to update_rq_clock() since the last time rq::lock was pinned.
792 * If inside of __schedule(), clock_update_flags will have been
793 * shifted left (a left shift is a cheap operation for the fast path
794 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
796 * if (rq-clock_update_flags >= RQCF_UPDATED)
798 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
799 * one position though, because the next rq_unpin_lock() will shift it
802 #define RQCF_REQ_SKIP 0x01
803 #define RQCF_ACT_SKIP 0x02
804 #define RQCF_UPDATED 0x04
806 static inline void assert_clock_updated(struct rq
*rq
)
809 * The only reason for not seeing a clock update since the
810 * last rq_pin_lock() is if we're currently skipping updates.
812 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
815 static inline u64
rq_clock(struct rq
*rq
)
817 lockdep_assert_held(&rq
->lock
);
818 assert_clock_updated(rq
);
823 static inline u64
rq_clock_task(struct rq
*rq
)
825 lockdep_assert_held(&rq
->lock
);
826 assert_clock_updated(rq
);
828 return rq
->clock_task
;
831 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
833 lockdep_assert_held(&rq
->lock
);
835 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
837 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
842 struct pin_cookie cookie
;
843 #ifdef CONFIG_SCHED_DEBUG
845 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
846 * current pin context is stashed here in case it needs to be
847 * restored in rq_repin_lock().
849 unsigned int clock_update_flags
;
853 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
855 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
857 #ifdef CONFIG_SCHED_DEBUG
858 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
859 rf
->clock_update_flags
= 0;
863 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
865 #ifdef CONFIG_SCHED_DEBUG
866 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
867 rf
->clock_update_flags
= RQCF_UPDATED
;
870 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
873 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
875 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
877 #ifdef CONFIG_SCHED_DEBUG
879 * Restore the value we stashed in @rf for this pin context.
881 rq
->clock_update_flags
|= rf
->clock_update_flags
;
886 enum numa_topology_type
{
891 extern enum numa_topology_type sched_numa_topology_type
;
892 extern int sched_max_numa_distance
;
893 extern bool find_numa_distance(int distance
);
897 extern void sched_init_numa(void);
898 extern void sched_domains_numa_masks_set(unsigned int cpu
);
899 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
901 static inline void sched_init_numa(void) { }
902 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
903 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
906 #ifdef CONFIG_NUMA_BALANCING
907 /* The regions in numa_faults array from task_struct */
908 enum numa_faults_stats
{
914 extern void sched_setnuma(struct task_struct
*p
, int node
);
915 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
916 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
917 #endif /* CONFIG_NUMA_BALANCING */
922 queue_balance_callback(struct rq
*rq
,
923 struct callback_head
*head
,
924 void (*func
)(struct rq
*rq
))
926 lockdep_assert_held(&rq
->lock
);
928 if (unlikely(head
->next
))
931 head
->func
= (void (*)(struct callback_head
*))func
;
932 head
->next
= rq
->balance_callback
;
933 rq
->balance_callback
= head
;
936 extern void sched_ttwu_pending(void);
938 #define rcu_dereference_check_sched_domain(p) \
939 rcu_dereference_check((p), \
940 lockdep_is_held(&sched_domains_mutex))
943 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
944 * See detach_destroy_domains: synchronize_sched for details.
946 * The domain tree of any CPU may only be accessed from within
947 * preempt-disabled sections.
949 #define for_each_domain(cpu, __sd) \
950 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
951 __sd; __sd = __sd->parent)
953 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
956 * highest_flag_domain - Return highest sched_domain containing flag.
957 * @cpu: The cpu whose highest level of sched domain is to
959 * @flag: The flag to check for the highest sched_domain
962 * Returns the highest sched_domain of a cpu which contains the given flag.
964 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
966 struct sched_domain
*sd
, *hsd
= NULL
;
968 for_each_domain(cpu
, sd
) {
969 if (!(sd
->flags
& flag
))
977 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
979 struct sched_domain
*sd
;
981 for_each_domain(cpu
, sd
) {
982 if (sd
->flags
& flag
)
989 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
990 DECLARE_PER_CPU(int, sd_llc_size
);
991 DECLARE_PER_CPU(int, sd_llc_id
);
992 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
993 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
994 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
996 struct sched_group_capacity
{
999 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1002 unsigned long capacity
;
1003 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1004 unsigned long next_update
;
1005 int imbalance
; /* XXX unrelated to capacity but shared group state */
1007 unsigned long cpumask
[0]; /* iteration mask */
1010 struct sched_group
{
1011 struct sched_group
*next
; /* Must be a circular list */
1014 unsigned int group_weight
;
1015 struct sched_group_capacity
*sgc
;
1016 int asym_prefer_cpu
; /* cpu of highest priority in group */
1019 * The CPUs this group covers.
1021 * NOTE: this field is variable length. (Allocated dynamically
1022 * by attaching extra space to the end of the structure,
1023 * depending on how many CPUs the kernel has booted up with)
1025 unsigned long cpumask
[0];
1028 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
1030 return to_cpumask(sg
->cpumask
);
1034 * cpumask masking which cpus in the group are allowed to iterate up the domain
1037 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
1039 return to_cpumask(sg
->sgc
->cpumask
);
1043 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1044 * @group: The group whose first cpu is to be returned.
1046 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1048 return cpumask_first(sched_group_cpus(group
));
1051 extern int group_balance_cpu(struct sched_group
*sg
);
1053 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1054 void register_sched_domain_sysctl(void);
1055 void unregister_sched_domain_sysctl(void);
1057 static inline void register_sched_domain_sysctl(void)
1060 static inline void unregister_sched_domain_sysctl(void)
1067 static inline void sched_ttwu_pending(void) { }
1069 #endif /* CONFIG_SMP */
1072 #include "autogroup.h"
1074 #ifdef CONFIG_CGROUP_SCHED
1077 * Return the group to which this tasks belongs.
1079 * We cannot use task_css() and friends because the cgroup subsystem
1080 * changes that value before the cgroup_subsys::attach() method is called,
1081 * therefore we cannot pin it and might observe the wrong value.
1083 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1084 * core changes this before calling sched_move_task().
1086 * Instead we use a 'copy' which is updated from sched_move_task() while
1087 * holding both task_struct::pi_lock and rq::lock.
1089 static inline struct task_group
*task_group(struct task_struct
*p
)
1091 return p
->sched_task_group
;
1094 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1095 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1097 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1098 struct task_group
*tg
= task_group(p
);
1101 #ifdef CONFIG_FAIR_GROUP_SCHED
1102 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1103 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1104 p
->se
.parent
= tg
->se
[cpu
];
1107 #ifdef CONFIG_RT_GROUP_SCHED
1108 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1109 p
->rt
.parent
= tg
->rt_se
[cpu
];
1113 #else /* CONFIG_CGROUP_SCHED */
1115 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1116 static inline struct task_group
*task_group(struct task_struct
*p
)
1121 #endif /* CONFIG_CGROUP_SCHED */
1123 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1125 set_task_rq(p
, cpu
);
1128 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1129 * successfuly executed on another CPU. We must ensure that updates of
1130 * per-task data have been completed by this moment.
1133 #ifdef CONFIG_THREAD_INFO_IN_TASK
1136 task_thread_info(p
)->cpu
= cpu
;
1143 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1145 #ifdef CONFIG_SCHED_DEBUG
1146 # include <linux/static_key.h>
1147 # define const_debug __read_mostly
1149 # define const_debug const
1152 extern const_debug
unsigned int sysctl_sched_features
;
1154 #define SCHED_FEAT(name, enabled) \
1155 __SCHED_FEAT_##name ,
1158 #include "features.h"
1164 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1165 #define SCHED_FEAT(name, enabled) \
1166 static __always_inline bool static_branch_##name(struct static_key *key) \
1168 return static_key_##enabled(key); \
1171 #include "features.h"
1175 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1176 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1177 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1178 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1179 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1181 extern struct static_key_false sched_numa_balancing
;
1182 extern struct static_key_false sched_schedstats
;
1184 static inline u64
global_rt_period(void)
1186 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1189 static inline u64
global_rt_runtime(void)
1191 if (sysctl_sched_rt_runtime
< 0)
1194 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1197 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1199 return rq
->curr
== p
;
1202 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1207 return task_current(rq
, p
);
1211 static inline int task_on_rq_queued(struct task_struct
*p
)
1213 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1216 static inline int task_on_rq_migrating(struct task_struct
*p
)
1218 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1221 #ifndef prepare_arch_switch
1222 # define prepare_arch_switch(next) do { } while (0)
1224 #ifndef finish_arch_post_lock_switch
1225 # define finish_arch_post_lock_switch() do { } while (0)
1228 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1232 * We can optimise this out completely for !SMP, because the
1233 * SMP rebalancing from interrupt is the only thing that cares
1240 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1244 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1245 * We must ensure this doesn't happen until the switch is completely
1248 * In particular, the load of prev->state in finish_task_switch() must
1249 * happen before this.
1251 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1253 smp_store_release(&prev
->on_cpu
, 0);
1255 #ifdef CONFIG_DEBUG_SPINLOCK
1256 /* this is a valid case when another task releases the spinlock */
1257 rq
->lock
.owner
= current
;
1260 * If we are tracking spinlock dependencies then we have to
1261 * fix up the runqueue lock - which gets 'carried over' from
1262 * prev into current:
1264 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1266 raw_spin_unlock_irq(&rq
->lock
);
1272 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1273 #define WF_FORK 0x02 /* child wakeup after fork */
1274 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1277 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1278 * of tasks with abnormal "nice" values across CPUs the contribution that
1279 * each task makes to its run queue's load is weighted according to its
1280 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1281 * scaled version of the new time slice allocation that they receive on time
1285 #define WEIGHT_IDLEPRIO 3
1286 #define WMULT_IDLEPRIO 1431655765
1288 extern const int sched_prio_to_weight
[40];
1289 extern const u32 sched_prio_to_wmult
[40];
1292 * {de,en}queue flags:
1294 * DEQUEUE_SLEEP - task is no longer runnable
1295 * ENQUEUE_WAKEUP - task just became runnable
1297 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1298 * are in a known state which allows modification. Such pairs
1299 * should preserve as much state as possible.
1301 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1304 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1305 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1306 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1310 #define DEQUEUE_SLEEP 0x01
1311 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1312 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1314 #define ENQUEUE_WAKEUP 0x01
1315 #define ENQUEUE_RESTORE 0x02
1316 #define ENQUEUE_MOVE 0x04
1318 #define ENQUEUE_HEAD 0x08
1319 #define ENQUEUE_REPLENISH 0x10
1321 #define ENQUEUE_MIGRATED 0x20
1323 #define ENQUEUE_MIGRATED 0x00
1326 #define RETRY_TASK ((void *)-1UL)
1328 struct sched_class
{
1329 const struct sched_class
*next
;
1331 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1332 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1333 void (*yield_task
) (struct rq
*rq
);
1334 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1336 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1339 * It is the responsibility of the pick_next_task() method that will
1340 * return the next task to call put_prev_task() on the @prev task or
1341 * something equivalent.
1343 * May return RETRY_TASK when it finds a higher prio class has runnable
1346 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1347 struct task_struct
*prev
,
1348 struct rq_flags
*rf
);
1349 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1352 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1353 void (*migrate_task_rq
)(struct task_struct
*p
);
1355 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1357 void (*set_cpus_allowed
)(struct task_struct
*p
,
1358 const struct cpumask
*newmask
);
1360 void (*rq_online
)(struct rq
*rq
);
1361 void (*rq_offline
)(struct rq
*rq
);
1364 void (*set_curr_task
) (struct rq
*rq
);
1365 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1366 void (*task_fork
) (struct task_struct
*p
);
1367 void (*task_dead
) (struct task_struct
*p
);
1370 * The switched_from() call is allowed to drop rq->lock, therefore we
1371 * cannot assume the switched_from/switched_to pair is serliazed by
1372 * rq->lock. They are however serialized by p->pi_lock.
1374 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1375 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1376 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1379 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1380 struct task_struct
*task
);
1382 void (*update_curr
) (struct rq
*rq
);
1384 #define TASK_SET_GROUP 0
1385 #define TASK_MOVE_GROUP 1
1387 #ifdef CONFIG_FAIR_GROUP_SCHED
1388 void (*task_change_group
) (struct task_struct
*p
, int type
);
1392 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1394 prev
->sched_class
->put_prev_task(rq
, prev
);
1397 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1399 curr
->sched_class
->set_curr_task(rq
);
1402 #define sched_class_highest (&stop_sched_class)
1403 #define for_each_class(class) \
1404 for (class = sched_class_highest; class; class = class->next)
1406 extern const struct sched_class stop_sched_class
;
1407 extern const struct sched_class dl_sched_class
;
1408 extern const struct sched_class rt_sched_class
;
1409 extern const struct sched_class fair_sched_class
;
1410 extern const struct sched_class idle_sched_class
;
1415 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1417 extern void trigger_load_balance(struct rq
*rq
);
1419 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1423 #ifdef CONFIG_CPU_IDLE
1424 static inline void idle_set_state(struct rq
*rq
,
1425 struct cpuidle_state
*idle_state
)
1427 rq
->idle_state
= idle_state
;
1430 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1432 SCHED_WARN_ON(!rcu_read_lock_held());
1433 return rq
->idle_state
;
1436 static inline void idle_set_state(struct rq
*rq
,
1437 struct cpuidle_state
*idle_state
)
1441 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1447 extern void sysrq_sched_debug_show(void);
1448 extern void sched_init_granularity(void);
1449 extern void update_max_interval(void);
1451 extern void init_sched_dl_class(void);
1452 extern void init_sched_rt_class(void);
1453 extern void init_sched_fair_class(void);
1455 extern void resched_curr(struct rq
*rq
);
1456 extern void resched_cpu(int cpu
);
1458 extern struct rt_bandwidth def_rt_bandwidth
;
1459 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1461 extern struct dl_bandwidth def_dl_bandwidth
;
1462 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1463 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1465 unsigned long to_ratio(u64 period
, u64 runtime
);
1467 extern void init_entity_runnable_average(struct sched_entity
*se
);
1468 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1470 #ifdef CONFIG_NO_HZ_FULL
1471 extern bool sched_can_stop_tick(struct rq
*rq
);
1474 * Tick may be needed by tasks in the runqueue depending on their policy and
1475 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1476 * nohz mode if necessary.
1478 static inline void sched_update_tick_dependency(struct rq
*rq
)
1482 if (!tick_nohz_full_enabled())
1487 if (!tick_nohz_full_cpu(cpu
))
1490 if (sched_can_stop_tick(rq
))
1491 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1493 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1496 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1499 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1501 unsigned prev_nr
= rq
->nr_running
;
1503 rq
->nr_running
= prev_nr
+ count
;
1505 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1507 if (!rq
->rd
->overload
)
1508 rq
->rd
->overload
= true;
1512 sched_update_tick_dependency(rq
);
1515 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1517 rq
->nr_running
-= count
;
1518 /* Check if we still need preemption */
1519 sched_update_tick_dependency(rq
);
1522 static inline void rq_last_tick_reset(struct rq
*rq
)
1524 #ifdef CONFIG_NO_HZ_FULL
1525 rq
->last_sched_tick
= jiffies
;
1529 extern void update_rq_clock(struct rq
*rq
);
1531 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1532 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1534 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1536 extern const_debug
unsigned int sysctl_sched_time_avg
;
1537 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1538 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1540 static inline u64
sched_avg_period(void)
1542 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1545 #ifdef CONFIG_SCHED_HRTICK
1549 * - enabled by features
1550 * - hrtimer is actually high res
1552 static inline int hrtick_enabled(struct rq
*rq
)
1554 if (!sched_feat(HRTICK
))
1556 if (!cpu_active(cpu_of(rq
)))
1558 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1561 void hrtick_start(struct rq
*rq
, u64 delay
);
1565 static inline int hrtick_enabled(struct rq
*rq
)
1570 #endif /* CONFIG_SCHED_HRTICK */
1573 extern void sched_avg_update(struct rq
*rq
);
1575 #ifndef arch_scale_freq_capacity
1576 static __always_inline
1577 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1579 return SCHED_CAPACITY_SCALE
;
1583 #ifndef arch_scale_cpu_capacity
1584 static __always_inline
1585 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1587 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1588 return sd
->smt_gain
/ sd
->span_weight
;
1590 return SCHED_CAPACITY_SCALE
;
1594 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1596 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1597 sched_avg_update(rq
);
1600 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1601 static inline void sched_avg_update(struct rq
*rq
) { }
1604 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1605 __acquires(rq
->lock
);
1606 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1607 __acquires(p
->pi_lock
)
1608 __acquires(rq
->lock
);
1610 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1611 __releases(rq
->lock
)
1613 rq_unpin_lock(rq
, rf
);
1614 raw_spin_unlock(&rq
->lock
);
1618 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1619 __releases(rq
->lock
)
1620 __releases(p
->pi_lock
)
1622 rq_unpin_lock(rq
, rf
);
1623 raw_spin_unlock(&rq
->lock
);
1624 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1628 #ifdef CONFIG_PREEMPT
1630 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1633 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1634 * way at the expense of forcing extra atomic operations in all
1635 * invocations. This assures that the double_lock is acquired using the
1636 * same underlying policy as the spinlock_t on this architecture, which
1637 * reduces latency compared to the unfair variant below. However, it
1638 * also adds more overhead and therefore may reduce throughput.
1640 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1641 __releases(this_rq
->lock
)
1642 __acquires(busiest
->lock
)
1643 __acquires(this_rq
->lock
)
1645 raw_spin_unlock(&this_rq
->lock
);
1646 double_rq_lock(this_rq
, busiest
);
1653 * Unfair double_lock_balance: Optimizes throughput at the expense of
1654 * latency by eliminating extra atomic operations when the locks are
1655 * already in proper order on entry. This favors lower cpu-ids and will
1656 * grant the double lock to lower cpus over higher ids under contention,
1657 * regardless of entry order into the function.
1659 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1660 __releases(this_rq
->lock
)
1661 __acquires(busiest
->lock
)
1662 __acquires(this_rq
->lock
)
1666 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1667 if (busiest
< this_rq
) {
1668 raw_spin_unlock(&this_rq
->lock
);
1669 raw_spin_lock(&busiest
->lock
);
1670 raw_spin_lock_nested(&this_rq
->lock
,
1671 SINGLE_DEPTH_NESTING
);
1674 raw_spin_lock_nested(&busiest
->lock
,
1675 SINGLE_DEPTH_NESTING
);
1680 #endif /* CONFIG_PREEMPT */
1683 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1685 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1687 if (unlikely(!irqs_disabled())) {
1688 /* printk() doesn't work good under rq->lock */
1689 raw_spin_unlock(&this_rq
->lock
);
1693 return _double_lock_balance(this_rq
, busiest
);
1696 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1697 __releases(busiest
->lock
)
1699 raw_spin_unlock(&busiest
->lock
);
1700 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1703 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1709 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1712 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1718 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1721 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1727 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1731 * double_rq_lock - safely lock two runqueues
1733 * Note this does not disable interrupts like task_rq_lock,
1734 * you need to do so manually before calling.
1736 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1737 __acquires(rq1
->lock
)
1738 __acquires(rq2
->lock
)
1740 BUG_ON(!irqs_disabled());
1742 raw_spin_lock(&rq1
->lock
);
1743 __acquire(rq2
->lock
); /* Fake it out ;) */
1746 raw_spin_lock(&rq1
->lock
);
1747 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1749 raw_spin_lock(&rq2
->lock
);
1750 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1756 * double_rq_unlock - safely unlock two runqueues
1758 * Note this does not restore interrupts like task_rq_unlock,
1759 * you need to do so manually after calling.
1761 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1762 __releases(rq1
->lock
)
1763 __releases(rq2
->lock
)
1765 raw_spin_unlock(&rq1
->lock
);
1767 raw_spin_unlock(&rq2
->lock
);
1769 __release(rq2
->lock
);
1772 extern void set_rq_online (struct rq
*rq
);
1773 extern void set_rq_offline(struct rq
*rq
);
1774 extern bool sched_smp_initialized
;
1776 #else /* CONFIG_SMP */
1779 * double_rq_lock - safely lock two runqueues
1781 * Note this does not disable interrupts like task_rq_lock,
1782 * you need to do so manually before calling.
1784 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1785 __acquires(rq1
->lock
)
1786 __acquires(rq2
->lock
)
1788 BUG_ON(!irqs_disabled());
1790 raw_spin_lock(&rq1
->lock
);
1791 __acquire(rq2
->lock
); /* Fake it out ;) */
1795 * double_rq_unlock - safely unlock two runqueues
1797 * Note this does not restore interrupts like task_rq_unlock,
1798 * you need to do so manually after calling.
1800 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1801 __releases(rq1
->lock
)
1802 __releases(rq2
->lock
)
1805 raw_spin_unlock(&rq1
->lock
);
1806 __release(rq2
->lock
);
1811 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1812 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1814 #ifdef CONFIG_SCHED_DEBUG
1815 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1816 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1817 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1819 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1821 #ifdef CONFIG_NUMA_BALANCING
1823 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1825 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1826 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1827 #endif /* CONFIG_NUMA_BALANCING */
1828 #endif /* CONFIG_SCHED_DEBUG */
1830 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1831 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1832 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1834 extern void cfs_bandwidth_usage_inc(void);
1835 extern void cfs_bandwidth_usage_dec(void);
1837 #ifdef CONFIG_NO_HZ_COMMON
1838 enum rq_nohz_flag_bits
{
1843 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1845 extern void nohz_balance_exit_idle(unsigned int cpu
);
1847 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1850 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1854 struct u64_stats_sync sync
;
1857 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
1859 static inline u64
irq_time_read(int cpu
)
1861 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
1862 u64
*cpustat
= kcpustat_cpu(cpu
).cpustat
;
1867 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
1868 total
= cpustat
[CPUTIME_SOFTIRQ
] + cpustat
[CPUTIME_IRQ
];
1869 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
1873 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1875 #ifdef CONFIG_CPU_FREQ
1876 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
1879 * cpufreq_update_util - Take a note about CPU utilization changes.
1880 * @rq: Runqueue to carry out the update for.
1881 * @flags: Update reason flags.
1883 * This function is called by the scheduler on the CPU whose utilization is
1886 * It can only be called from RCU-sched read-side critical sections.
1888 * The way cpufreq is currently arranged requires it to evaluate the CPU
1889 * performance state (frequency/voltage) on a regular basis to prevent it from
1890 * being stuck in a completely inadequate performance level for too long.
1891 * That is not guaranteed to happen if the updates are only triggered from CFS,
1892 * though, because they may not be coming in if RT or deadline tasks are active
1893 * all the time (or there are RT and DL tasks only).
1895 * As a workaround for that issue, this function is called by the RT and DL
1896 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1897 * but that really is a band-aid. Going forward it should be replaced with
1898 * solutions targeted more specifically at RT and DL tasks.
1900 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
1902 struct update_util_data
*data
;
1904 data
= rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data
));
1906 data
->func(data
, rq_clock(rq
), flags
);
1909 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
)
1911 if (cpu_of(rq
) == smp_processor_id())
1912 cpufreq_update_util(rq
, flags
);
1915 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
1916 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
) {}
1917 #endif /* CONFIG_CPU_FREQ */
1919 #ifdef arch_scale_freq_capacity
1920 #ifndef arch_scale_freq_invariant
1921 #define arch_scale_freq_invariant() (true)
1923 #else /* arch_scale_freq_capacity */
1924 #define arch_scale_freq_invariant() (false)