2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/topology.h>
5 #include <linux/sched/rt.h>
6 #include <linux/sched/clock.h>
7 #include <linux/sched/wake_q.h>
8 #include <linux/sched/signal.h>
9 #include <linux/sched/numa_balancing.h>
10 #include <linux/sched/mm.h>
11 #include <linux/sched/cpufreq.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/nohz.h>
14 #include <linux/sched/debug.h>
15 #include <linux/u64_stats_sync.h>
16 #include <linux/sched/deadline.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/binfmts.h>
19 #include <linux/mutex.h>
20 #include <linux/spinlock.h>
21 #include <linux/stop_machine.h>
22 #include <linux/irq_work.h>
23 #include <linux/tick.h>
24 #include <linux/slab.h>
26 #ifdef CONFIG_PARAVIRT
27 #include <asm/paravirt.h>
31 #include "cpudeadline.h"
34 #ifdef CONFIG_SCHED_DEBUG
35 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
37 #define SCHED_WARN_ON(x) ((void)(x))
43 /* task_struct::on_rq states: */
44 #define TASK_ON_RQ_QUEUED 1
45 #define TASK_ON_RQ_MIGRATING 2
47 extern __read_mostly
int scheduler_running
;
49 extern unsigned long calc_load_update
;
50 extern atomic_long_t calc_load_tasks
;
52 extern void calc_global_load_tick(struct rq
*this_rq
);
53 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
56 extern void cpu_load_update_active(struct rq
*this_rq
);
58 static inline void cpu_load_update_active(struct rq
*this_rq
) { }
62 * Helpers for converting nanosecond timing to jiffy resolution
64 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
67 * Increase resolution of nice-level calculations for 64-bit architectures.
68 * The extra resolution improves shares distribution and load balancing of
69 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
70 * hierarchies, especially on larger systems. This is not a user-visible change
71 * and does not change the user-interface for setting shares/weights.
73 * We increase resolution only if we have enough bits to allow this increased
74 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
75 * pretty high and the returns do not justify the increased costs.
77 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
78 * increase coverage and consistency always enable it on 64bit platforms.
81 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
82 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
83 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
85 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
86 # define scale_load(w) (w)
87 # define scale_load_down(w) (w)
91 * Task weight (visible to users) and its load (invisible to users) have
92 * independent resolution, but they should be well calibrated. We use
93 * scale_load() and scale_load_down(w) to convert between them. The
94 * following must be true:
96 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
99 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
102 * Single value that decides SCHED_DEADLINE internal math precision.
103 * 10 -> just above 1us
104 * 9 -> just above 0.5us
106 #define DL_SCALE (10)
109 * These are the 'tuning knobs' of the scheduler:
113 * single value that denotes runtime == period, ie unlimited time.
115 #define RUNTIME_INF ((u64)~0ULL)
117 static inline int idle_policy(int policy
)
119 return policy
== SCHED_IDLE
;
121 static inline int fair_policy(int policy
)
123 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
126 static inline int rt_policy(int policy
)
128 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
131 static inline int dl_policy(int policy
)
133 return policy
== SCHED_DEADLINE
;
135 static inline bool valid_policy(int policy
)
137 return idle_policy(policy
) || fair_policy(policy
) ||
138 rt_policy(policy
) || dl_policy(policy
);
141 static inline int task_has_rt_policy(struct task_struct
*p
)
143 return rt_policy(p
->policy
);
146 static inline int task_has_dl_policy(struct task_struct
*p
)
148 return dl_policy(p
->policy
);
152 * Tells if entity @a should preempt entity @b.
155 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
157 return dl_time_before(a
->deadline
, b
->deadline
);
161 * This is the priority-queue data structure of the RT scheduling class:
163 struct rt_prio_array
{
164 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
165 struct list_head queue
[MAX_RT_PRIO
];
168 struct rt_bandwidth
{
169 /* nests inside the rq lock: */
170 raw_spinlock_t rt_runtime_lock
;
173 struct hrtimer rt_period_timer
;
174 unsigned int rt_period_active
;
177 void __dl_clear_params(struct task_struct
*p
);
180 * To keep the bandwidth of -deadline tasks and groups under control
181 * we need some place where:
182 * - store the maximum -deadline bandwidth of the system (the group);
183 * - cache the fraction of that bandwidth that is currently allocated.
185 * This is all done in the data structure below. It is similar to the
186 * one used for RT-throttling (rt_bandwidth), with the main difference
187 * that, since here we are only interested in admission control, we
188 * do not decrease any runtime while the group "executes", neither we
189 * need a timer to replenish it.
191 * With respect to SMP, the bandwidth is given on a per-CPU basis,
193 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
194 * - dl_total_bw array contains, in the i-eth element, the currently
195 * allocated bandwidth on the i-eth CPU.
196 * Moreover, groups consume bandwidth on each CPU, while tasks only
197 * consume bandwidth on the CPU they're running on.
198 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
199 * that will be shown the next time the proc or cgroup controls will
200 * be red. It on its turn can be changed by writing on its own
203 struct dl_bandwidth
{
204 raw_spinlock_t dl_runtime_lock
;
209 static inline int dl_bandwidth_enabled(void)
211 return sysctl_sched_rt_runtime
>= 0;
214 extern struct dl_bw
*dl_bw_of(int i
);
222 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
224 dl_b
->total_bw
-= tsk_bw
;
228 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
230 dl_b
->total_bw
+= tsk_bw
;
234 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
236 return dl_b
->bw
!= -1 &&
237 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
240 extern void init_dl_bw(struct dl_bw
*dl_b
);
242 #ifdef CONFIG_CGROUP_SCHED
244 #include <linux/cgroup.h>
249 extern struct list_head task_groups
;
251 struct cfs_bandwidth
{
252 #ifdef CONFIG_CFS_BANDWIDTH
256 s64 hierarchical_quota
;
259 int idle
, period_active
;
260 struct hrtimer period_timer
, slack_timer
;
261 struct list_head throttled_cfs_rq
;
264 int nr_periods
, nr_throttled
;
269 /* task group related information */
271 struct cgroup_subsys_state css
;
273 #ifdef CONFIG_FAIR_GROUP_SCHED
274 /* schedulable entities of this group on each cpu */
275 struct sched_entity
**se
;
276 /* runqueue "owned" by this group on each cpu */
277 struct cfs_rq
**cfs_rq
;
278 unsigned long shares
;
282 * load_avg can be heavily contended at clock tick time, so put
283 * it in its own cacheline separated from the fields above which
284 * will also be accessed at each tick.
286 atomic_long_t load_avg ____cacheline_aligned
;
290 #ifdef CONFIG_RT_GROUP_SCHED
291 struct sched_rt_entity
**rt_se
;
292 struct rt_rq
**rt_rq
;
294 struct rt_bandwidth rt_bandwidth
;
298 struct list_head list
;
300 struct task_group
*parent
;
301 struct list_head siblings
;
302 struct list_head children
;
304 #ifdef CONFIG_SCHED_AUTOGROUP
305 struct autogroup
*autogroup
;
308 struct cfs_bandwidth cfs_bandwidth
;
311 #ifdef CONFIG_FAIR_GROUP_SCHED
312 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
315 * A weight of 0 or 1 can cause arithmetics problems.
316 * A weight of a cfs_rq is the sum of weights of which entities
317 * are queued on this cfs_rq, so a weight of a entity should not be
318 * too large, so as the shares value of a task group.
319 * (The default weight is 1024 - so there's no practical
320 * limitation from this.)
322 #define MIN_SHARES (1UL << 1)
323 #define MAX_SHARES (1UL << 18)
326 typedef int (*tg_visitor
)(struct task_group
*, void *);
328 extern int walk_tg_tree_from(struct task_group
*from
,
329 tg_visitor down
, tg_visitor up
, void *data
);
332 * Iterate the full tree, calling @down when first entering a node and @up when
333 * leaving it for the final time.
335 * Caller must hold rcu_lock or sufficient equivalent.
337 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
339 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
342 extern int tg_nop(struct task_group
*tg
, void *data
);
344 extern void free_fair_sched_group(struct task_group
*tg
);
345 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
346 extern void online_fair_sched_group(struct task_group
*tg
);
347 extern void unregister_fair_sched_group(struct task_group
*tg
);
348 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
349 struct sched_entity
*se
, int cpu
,
350 struct sched_entity
*parent
);
351 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
353 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
354 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
355 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
357 extern void free_rt_sched_group(struct task_group
*tg
);
358 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
359 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
360 struct sched_rt_entity
*rt_se
, int cpu
,
361 struct sched_rt_entity
*parent
);
363 extern struct task_group
*sched_create_group(struct task_group
*parent
);
364 extern void sched_online_group(struct task_group
*tg
,
365 struct task_group
*parent
);
366 extern void sched_destroy_group(struct task_group
*tg
);
367 extern void sched_offline_group(struct task_group
*tg
);
369 extern void sched_move_task(struct task_struct
*tsk
);
371 #ifdef CONFIG_FAIR_GROUP_SCHED
372 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
375 extern void set_task_rq_fair(struct sched_entity
*se
,
376 struct cfs_rq
*prev
, struct cfs_rq
*next
);
377 #else /* !CONFIG_SMP */
378 static inline void set_task_rq_fair(struct sched_entity
*se
,
379 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
380 #endif /* CONFIG_SMP */
381 #endif /* CONFIG_FAIR_GROUP_SCHED */
383 #else /* CONFIG_CGROUP_SCHED */
385 struct cfs_bandwidth
{ };
387 #endif /* CONFIG_CGROUP_SCHED */
389 /* CFS-related fields in a runqueue */
391 struct load_weight load
;
392 unsigned int nr_running
, h_nr_running
;
397 u64 min_vruntime_copy
;
400 struct rb_root tasks_timeline
;
401 struct rb_node
*rb_leftmost
;
404 * 'curr' points to currently running entity on this cfs_rq.
405 * It is set to NULL otherwise (i.e when none are currently running).
407 struct sched_entity
*curr
, *next
, *last
, *skip
;
409 #ifdef CONFIG_SCHED_DEBUG
410 unsigned int nr_spread_over
;
417 struct sched_avg avg
;
418 u64 runnable_load_sum
;
419 unsigned long runnable_load_avg
;
420 #ifdef CONFIG_FAIR_GROUP_SCHED
421 unsigned long tg_load_avg_contrib
;
422 unsigned long propagate_avg
;
424 atomic_long_t removed_load_avg
, removed_util_avg
;
426 u64 load_last_update_time_copy
;
429 #ifdef CONFIG_FAIR_GROUP_SCHED
431 * h_load = weight * f(tg)
433 * Where f(tg) is the recursive weight fraction assigned to
436 unsigned long h_load
;
437 u64 last_h_load_update
;
438 struct sched_entity
*h_load_next
;
439 #endif /* CONFIG_FAIR_GROUP_SCHED */
440 #endif /* CONFIG_SMP */
442 #ifdef CONFIG_FAIR_GROUP_SCHED
443 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
446 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
447 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
448 * (like users, containers etc.)
450 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
451 * list is used during load balance.
454 struct list_head leaf_cfs_rq_list
;
455 struct task_group
*tg
; /* group that "owns" this runqueue */
457 #ifdef CONFIG_CFS_BANDWIDTH
460 s64 runtime_remaining
;
462 u64 throttled_clock
, throttled_clock_task
;
463 u64 throttled_clock_task_time
;
464 int throttled
, throttle_count
;
465 struct list_head throttled_list
;
466 #endif /* CONFIG_CFS_BANDWIDTH */
467 #endif /* CONFIG_FAIR_GROUP_SCHED */
470 static inline int rt_bandwidth_enabled(void)
472 return sysctl_sched_rt_runtime
>= 0;
475 /* RT IPI pull logic requires IRQ_WORK */
476 #ifdef CONFIG_IRQ_WORK
477 # define HAVE_RT_PUSH_IPI
480 /* Real-Time classes' related field in a runqueue: */
482 struct rt_prio_array active
;
483 unsigned int rt_nr_running
;
484 unsigned int rr_nr_running
;
485 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
487 int curr
; /* highest queued rt task prio */
489 int next
; /* next highest */
494 unsigned long rt_nr_migratory
;
495 unsigned long rt_nr_total
;
497 struct plist_head pushable_tasks
;
498 #ifdef HAVE_RT_PUSH_IPI
501 struct irq_work push_work
;
502 raw_spinlock_t push_lock
;
504 #endif /* CONFIG_SMP */
510 /* Nests inside the rq lock: */
511 raw_spinlock_t rt_runtime_lock
;
513 #ifdef CONFIG_RT_GROUP_SCHED
514 unsigned long rt_nr_boosted
;
517 struct task_group
*tg
;
521 /* Deadline class' related fields in a runqueue */
523 /* runqueue is an rbtree, ordered by deadline */
524 struct rb_root rb_root
;
525 struct rb_node
*rb_leftmost
;
527 unsigned long dl_nr_running
;
531 * Deadline values of the currently executing and the
532 * earliest ready task on this rq. Caching these facilitates
533 * the decision wether or not a ready but not running task
534 * should migrate somewhere else.
541 unsigned long dl_nr_migratory
;
545 * Tasks on this rq that can be pushed away. They are kept in
546 * an rb-tree, ordered by tasks' deadlines, with caching
547 * of the leftmost (earliest deadline) element.
549 struct rb_root pushable_dl_tasks_root
;
550 struct rb_node
*pushable_dl_tasks_leftmost
;
558 static inline bool sched_asym_prefer(int a
, int b
)
560 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
564 * We add the notion of a root-domain which will be used to define per-domain
565 * variables. Each exclusive cpuset essentially defines an island domain by
566 * fully partitioning the member cpus from any other cpuset. Whenever a new
567 * exclusive cpuset is created, we also create and attach a new root-domain
576 cpumask_var_t online
;
578 /* Indicate more than one runnable task for any CPU */
582 * The bit corresponding to a CPU gets set here if such CPU has more
583 * than one runnable -deadline task (as it is below for RT tasks).
585 cpumask_var_t dlo_mask
;
591 * The "RT overload" flag: it gets set if a CPU has more than
592 * one runnable RT task.
594 cpumask_var_t rto_mask
;
595 struct cpupri cpupri
;
597 unsigned long max_cpu_capacity
;
600 extern struct root_domain def_root_domain
;
601 extern struct mutex sched_domains_mutex
;
602 extern cpumask_var_t fallback_doms
;
603 extern cpumask_var_t sched_domains_tmpmask
;
605 extern void init_defrootdomain(void);
606 extern int init_sched_domains(const struct cpumask
*cpu_map
);
607 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
609 #endif /* CONFIG_SMP */
612 * This is the main, per-CPU runqueue data structure.
614 * Locking rule: those places that want to lock multiple runqueues
615 * (such as the load balancing or the thread migration code), lock
616 * acquire operations must be ordered by ascending &runqueue.
623 * nr_running and cpu_load should be in the same cacheline because
624 * remote CPUs use both these fields when doing load calculation.
626 unsigned int nr_running
;
627 #ifdef CONFIG_NUMA_BALANCING
628 unsigned int nr_numa_running
;
629 unsigned int nr_preferred_running
;
631 #define CPU_LOAD_IDX_MAX 5
632 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
633 #ifdef CONFIG_NO_HZ_COMMON
635 unsigned long last_load_update_tick
;
636 #endif /* CONFIG_SMP */
637 unsigned long nohz_flags
;
638 #endif /* CONFIG_NO_HZ_COMMON */
639 #ifdef CONFIG_NO_HZ_FULL
640 unsigned long last_sched_tick
;
642 /* capture load from *all* tasks on this cpu: */
643 struct load_weight load
;
644 unsigned long nr_load_updates
;
651 #ifdef CONFIG_FAIR_GROUP_SCHED
652 /* list of leaf cfs_rq on this cpu: */
653 struct list_head leaf_cfs_rq_list
;
654 struct list_head
*tmp_alone_branch
;
655 #endif /* CONFIG_FAIR_GROUP_SCHED */
658 * This is part of a global counter where only the total sum
659 * over all CPUs matters. A task can increase this counter on
660 * one CPU and if it got migrated afterwards it may decrease
661 * it on another CPU. Always updated under the runqueue lock:
663 unsigned long nr_uninterruptible
;
665 struct task_struct
*curr
, *idle
, *stop
;
666 unsigned long next_balance
;
667 struct mm_struct
*prev_mm
;
669 unsigned int clock_update_flags
;
676 struct root_domain
*rd
;
677 struct sched_domain
*sd
;
679 unsigned long cpu_capacity
;
680 unsigned long cpu_capacity_orig
;
682 struct callback_head
*balance_callback
;
684 unsigned char idle_balance
;
685 /* For active balancing */
688 struct cpu_stop_work active_balance_work
;
689 /* cpu of this runqueue: */
693 struct list_head cfs_tasks
;
700 /* This is used to determine avg_idle's max value */
701 u64 max_idle_balance_cost
;
704 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
707 #ifdef CONFIG_PARAVIRT
710 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
711 u64 prev_steal_time_rq
;
714 /* calc_load related fields */
715 unsigned long calc_load_update
;
716 long calc_load_active
;
718 #ifdef CONFIG_SCHED_HRTICK
720 int hrtick_csd_pending
;
721 struct call_single_data hrtick_csd
;
723 struct hrtimer hrtick_timer
;
726 #ifdef CONFIG_SCHEDSTATS
728 struct sched_info rq_sched_info
;
729 unsigned long long rq_cpu_time
;
730 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
732 /* sys_sched_yield() stats */
733 unsigned int yld_count
;
735 /* schedule() stats */
736 unsigned int sched_count
;
737 unsigned int sched_goidle
;
739 /* try_to_wake_up() stats */
740 unsigned int ttwu_count
;
741 unsigned int ttwu_local
;
745 struct llist_head wake_list
;
748 #ifdef CONFIG_CPU_IDLE
749 /* Must be inspected within a rcu lock section */
750 struct cpuidle_state
*idle_state
;
754 static inline int cpu_of(struct rq
*rq
)
764 #ifdef CONFIG_SCHED_SMT
766 extern struct static_key_false sched_smt_present
;
768 extern void __update_idle_core(struct rq
*rq
);
770 static inline void update_idle_core(struct rq
*rq
)
772 if (static_branch_unlikely(&sched_smt_present
))
773 __update_idle_core(rq
);
777 static inline void update_idle_core(struct rq
*rq
) { }
780 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
782 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
783 #define this_rq() this_cpu_ptr(&runqueues)
784 #define task_rq(p) cpu_rq(task_cpu(p))
785 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
786 #define raw_rq() raw_cpu_ptr(&runqueues)
788 static inline u64
__rq_clock_broken(struct rq
*rq
)
790 return READ_ONCE(rq
->clock
);
794 * rq::clock_update_flags bits
796 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
797 * call to __schedule(). This is an optimisation to avoid
798 * neighbouring rq clock updates.
800 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
801 * in effect and calls to update_rq_clock() are being ignored.
803 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
804 * made to update_rq_clock() since the last time rq::lock was pinned.
806 * If inside of __schedule(), clock_update_flags will have been
807 * shifted left (a left shift is a cheap operation for the fast path
808 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
810 * if (rq-clock_update_flags >= RQCF_UPDATED)
812 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
813 * one position though, because the next rq_unpin_lock() will shift it
816 #define RQCF_REQ_SKIP 0x01
817 #define RQCF_ACT_SKIP 0x02
818 #define RQCF_UPDATED 0x04
820 static inline void assert_clock_updated(struct rq
*rq
)
823 * The only reason for not seeing a clock update since the
824 * last rq_pin_lock() is if we're currently skipping updates.
826 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
829 static inline u64
rq_clock(struct rq
*rq
)
831 lockdep_assert_held(&rq
->lock
);
832 assert_clock_updated(rq
);
837 static inline u64
rq_clock_task(struct rq
*rq
)
839 lockdep_assert_held(&rq
->lock
);
840 assert_clock_updated(rq
);
842 return rq
->clock_task
;
845 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
847 lockdep_assert_held(&rq
->lock
);
849 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
851 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
856 struct pin_cookie cookie
;
857 #ifdef CONFIG_SCHED_DEBUG
859 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
860 * current pin context is stashed here in case it needs to be
861 * restored in rq_repin_lock().
863 unsigned int clock_update_flags
;
867 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
869 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
871 #ifdef CONFIG_SCHED_DEBUG
872 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
873 rf
->clock_update_flags
= 0;
877 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
879 #ifdef CONFIG_SCHED_DEBUG
880 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
881 rf
->clock_update_flags
= RQCF_UPDATED
;
884 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
887 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
889 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
891 #ifdef CONFIG_SCHED_DEBUG
893 * Restore the value we stashed in @rf for this pin context.
895 rq
->clock_update_flags
|= rf
->clock_update_flags
;
900 enum numa_topology_type
{
905 extern enum numa_topology_type sched_numa_topology_type
;
906 extern int sched_max_numa_distance
;
907 extern bool find_numa_distance(int distance
);
911 extern void sched_init_numa(void);
912 extern void sched_domains_numa_masks_set(unsigned int cpu
);
913 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
915 static inline void sched_init_numa(void) { }
916 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
917 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
920 #ifdef CONFIG_NUMA_BALANCING
921 /* The regions in numa_faults array from task_struct */
922 enum numa_faults_stats
{
928 extern void sched_setnuma(struct task_struct
*p
, int node
);
929 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
930 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
931 #endif /* CONFIG_NUMA_BALANCING */
936 queue_balance_callback(struct rq
*rq
,
937 struct callback_head
*head
,
938 void (*func
)(struct rq
*rq
))
940 lockdep_assert_held(&rq
->lock
);
942 if (unlikely(head
->next
))
945 head
->func
= (void (*)(struct callback_head
*))func
;
946 head
->next
= rq
->balance_callback
;
947 rq
->balance_callback
= head
;
950 extern void sched_ttwu_pending(void);
952 #define rcu_dereference_check_sched_domain(p) \
953 rcu_dereference_check((p), \
954 lockdep_is_held(&sched_domains_mutex))
957 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
958 * See detach_destroy_domains: synchronize_sched for details.
960 * The domain tree of any CPU may only be accessed from within
961 * preempt-disabled sections.
963 #define for_each_domain(cpu, __sd) \
964 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
965 __sd; __sd = __sd->parent)
967 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
970 * highest_flag_domain - Return highest sched_domain containing flag.
971 * @cpu: The cpu whose highest level of sched domain is to
973 * @flag: The flag to check for the highest sched_domain
976 * Returns the highest sched_domain of a cpu which contains the given flag.
978 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
980 struct sched_domain
*sd
, *hsd
= NULL
;
982 for_each_domain(cpu
, sd
) {
983 if (!(sd
->flags
& flag
))
991 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
993 struct sched_domain
*sd
;
995 for_each_domain(cpu
, sd
) {
996 if (sd
->flags
& flag
)
1003 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
1004 DECLARE_PER_CPU(int, sd_llc_size
);
1005 DECLARE_PER_CPU(int, sd_llc_id
);
1006 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
1007 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
1008 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
1010 struct sched_group_capacity
{
1013 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1016 unsigned long capacity
;
1017 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1018 unsigned long next_update
;
1019 int imbalance
; /* XXX unrelated to capacity but shared group state */
1021 unsigned long cpumask
[0]; /* iteration mask */
1024 struct sched_group
{
1025 struct sched_group
*next
; /* Must be a circular list */
1028 unsigned int group_weight
;
1029 struct sched_group_capacity
*sgc
;
1030 int asym_prefer_cpu
; /* cpu of highest priority in group */
1033 * The CPUs this group covers.
1035 * NOTE: this field is variable length. (Allocated dynamically
1036 * by attaching extra space to the end of the structure,
1037 * depending on how many CPUs the kernel has booted up with)
1039 unsigned long cpumask
[0];
1042 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
1044 return to_cpumask(sg
->cpumask
);
1048 * cpumask masking which cpus in the group are allowed to iterate up the domain
1051 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
1053 return to_cpumask(sg
->sgc
->cpumask
);
1057 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1058 * @group: The group whose first cpu is to be returned.
1060 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1062 return cpumask_first(sched_group_cpus(group
));
1065 extern int group_balance_cpu(struct sched_group
*sg
);
1067 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1068 void register_sched_domain_sysctl(void);
1069 void unregister_sched_domain_sysctl(void);
1071 static inline void register_sched_domain_sysctl(void)
1074 static inline void unregister_sched_domain_sysctl(void)
1081 static inline void sched_ttwu_pending(void) { }
1083 #endif /* CONFIG_SMP */
1086 #include "autogroup.h"
1088 #ifdef CONFIG_CGROUP_SCHED
1091 * Return the group to which this tasks belongs.
1093 * We cannot use task_css() and friends because the cgroup subsystem
1094 * changes that value before the cgroup_subsys::attach() method is called,
1095 * therefore we cannot pin it and might observe the wrong value.
1097 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1098 * core changes this before calling sched_move_task().
1100 * Instead we use a 'copy' which is updated from sched_move_task() while
1101 * holding both task_struct::pi_lock and rq::lock.
1103 static inline struct task_group
*task_group(struct task_struct
*p
)
1105 return p
->sched_task_group
;
1108 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1109 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1111 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1112 struct task_group
*tg
= task_group(p
);
1115 #ifdef CONFIG_FAIR_GROUP_SCHED
1116 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1117 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1118 p
->se
.parent
= tg
->se
[cpu
];
1121 #ifdef CONFIG_RT_GROUP_SCHED
1122 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1123 p
->rt
.parent
= tg
->rt_se
[cpu
];
1127 #else /* CONFIG_CGROUP_SCHED */
1129 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1130 static inline struct task_group
*task_group(struct task_struct
*p
)
1135 #endif /* CONFIG_CGROUP_SCHED */
1137 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1139 set_task_rq(p
, cpu
);
1142 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1143 * successfuly executed on another CPU. We must ensure that updates of
1144 * per-task data have been completed by this moment.
1147 #ifdef CONFIG_THREAD_INFO_IN_TASK
1150 task_thread_info(p
)->cpu
= cpu
;
1157 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1159 #ifdef CONFIG_SCHED_DEBUG
1160 # include <linux/static_key.h>
1161 # define const_debug __read_mostly
1163 # define const_debug const
1166 extern const_debug
unsigned int sysctl_sched_features
;
1168 #define SCHED_FEAT(name, enabled) \
1169 __SCHED_FEAT_##name ,
1172 #include "features.h"
1178 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1179 #define SCHED_FEAT(name, enabled) \
1180 static __always_inline bool static_branch_##name(struct static_key *key) \
1182 return static_key_##enabled(key); \
1185 #include "features.h"
1189 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1190 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1191 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1192 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1193 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1195 extern struct static_key_false sched_numa_balancing
;
1196 extern struct static_key_false sched_schedstats
;
1198 static inline u64
global_rt_period(void)
1200 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1203 static inline u64
global_rt_runtime(void)
1205 if (sysctl_sched_rt_runtime
< 0)
1208 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1211 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1213 return rq
->curr
== p
;
1216 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1221 return task_current(rq
, p
);
1225 static inline int task_on_rq_queued(struct task_struct
*p
)
1227 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1230 static inline int task_on_rq_migrating(struct task_struct
*p
)
1232 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1235 #ifndef prepare_arch_switch
1236 # define prepare_arch_switch(next) do { } while (0)
1238 #ifndef finish_arch_post_lock_switch
1239 # define finish_arch_post_lock_switch() do { } while (0)
1242 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1246 * We can optimise this out completely for !SMP, because the
1247 * SMP rebalancing from interrupt is the only thing that cares
1254 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1258 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1259 * We must ensure this doesn't happen until the switch is completely
1262 * In particular, the load of prev->state in finish_task_switch() must
1263 * happen before this.
1265 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1267 smp_store_release(&prev
->on_cpu
, 0);
1269 #ifdef CONFIG_DEBUG_SPINLOCK
1270 /* this is a valid case when another task releases the spinlock */
1271 rq
->lock
.owner
= current
;
1274 * If we are tracking spinlock dependencies then we have to
1275 * fix up the runqueue lock - which gets 'carried over' from
1276 * prev into current:
1278 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1280 raw_spin_unlock_irq(&rq
->lock
);
1286 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1287 #define WF_FORK 0x02 /* child wakeup after fork */
1288 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1291 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1292 * of tasks with abnormal "nice" values across CPUs the contribution that
1293 * each task makes to its run queue's load is weighted according to its
1294 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1295 * scaled version of the new time slice allocation that they receive on time
1299 #define WEIGHT_IDLEPRIO 3
1300 #define WMULT_IDLEPRIO 1431655765
1302 extern const int sched_prio_to_weight
[40];
1303 extern const u32 sched_prio_to_wmult
[40];
1306 * {de,en}queue flags:
1308 * DEQUEUE_SLEEP - task is no longer runnable
1309 * ENQUEUE_WAKEUP - task just became runnable
1311 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1312 * are in a known state which allows modification. Such pairs
1313 * should preserve as much state as possible.
1315 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1318 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1319 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1320 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1324 #define DEQUEUE_SLEEP 0x01
1325 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1326 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1328 #define ENQUEUE_WAKEUP 0x01
1329 #define ENQUEUE_RESTORE 0x02
1330 #define ENQUEUE_MOVE 0x04
1332 #define ENQUEUE_HEAD 0x08
1333 #define ENQUEUE_REPLENISH 0x10
1335 #define ENQUEUE_MIGRATED 0x20
1337 #define ENQUEUE_MIGRATED 0x00
1340 #define RETRY_TASK ((void *)-1UL)
1342 struct sched_class
{
1343 const struct sched_class
*next
;
1345 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1346 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1347 void (*yield_task
) (struct rq
*rq
);
1348 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1350 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1353 * It is the responsibility of the pick_next_task() method that will
1354 * return the next task to call put_prev_task() on the @prev task or
1355 * something equivalent.
1357 * May return RETRY_TASK when it finds a higher prio class has runnable
1360 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1361 struct task_struct
*prev
,
1362 struct rq_flags
*rf
);
1363 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1366 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1367 void (*migrate_task_rq
)(struct task_struct
*p
);
1369 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1371 void (*set_cpus_allowed
)(struct task_struct
*p
,
1372 const struct cpumask
*newmask
);
1374 void (*rq_online
)(struct rq
*rq
);
1375 void (*rq_offline
)(struct rq
*rq
);
1378 void (*set_curr_task
) (struct rq
*rq
);
1379 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1380 void (*task_fork
) (struct task_struct
*p
);
1381 void (*task_dead
) (struct task_struct
*p
);
1384 * The switched_from() call is allowed to drop rq->lock, therefore we
1385 * cannot assume the switched_from/switched_to pair is serliazed by
1386 * rq->lock. They are however serialized by p->pi_lock.
1388 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1389 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1390 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1393 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1394 struct task_struct
*task
);
1396 void (*update_curr
) (struct rq
*rq
);
1398 #define TASK_SET_GROUP 0
1399 #define TASK_MOVE_GROUP 1
1401 #ifdef CONFIG_FAIR_GROUP_SCHED
1402 void (*task_change_group
) (struct task_struct
*p
, int type
);
1406 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1408 prev
->sched_class
->put_prev_task(rq
, prev
);
1411 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1413 curr
->sched_class
->set_curr_task(rq
);
1416 #define sched_class_highest (&stop_sched_class)
1417 #define for_each_class(class) \
1418 for (class = sched_class_highest; class; class = class->next)
1420 extern const struct sched_class stop_sched_class
;
1421 extern const struct sched_class dl_sched_class
;
1422 extern const struct sched_class rt_sched_class
;
1423 extern const struct sched_class fair_sched_class
;
1424 extern const struct sched_class idle_sched_class
;
1429 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1431 extern void trigger_load_balance(struct rq
*rq
);
1433 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1437 #ifdef CONFIG_CPU_IDLE
1438 static inline void idle_set_state(struct rq
*rq
,
1439 struct cpuidle_state
*idle_state
)
1441 rq
->idle_state
= idle_state
;
1444 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1446 SCHED_WARN_ON(!rcu_read_lock_held());
1447 return rq
->idle_state
;
1450 static inline void idle_set_state(struct rq
*rq
,
1451 struct cpuidle_state
*idle_state
)
1455 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1461 extern void sysrq_sched_debug_show(void);
1462 extern void sched_init_granularity(void);
1463 extern void update_max_interval(void);
1465 extern void init_sched_dl_class(void);
1466 extern void init_sched_rt_class(void);
1467 extern void init_sched_fair_class(void);
1469 extern void resched_curr(struct rq
*rq
);
1470 extern void resched_cpu(int cpu
);
1472 extern struct rt_bandwidth def_rt_bandwidth
;
1473 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1475 extern struct dl_bandwidth def_dl_bandwidth
;
1476 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1477 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1479 unsigned long to_ratio(u64 period
, u64 runtime
);
1481 extern void init_entity_runnable_average(struct sched_entity
*se
);
1482 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1484 #ifdef CONFIG_NO_HZ_FULL
1485 extern bool sched_can_stop_tick(struct rq
*rq
);
1488 * Tick may be needed by tasks in the runqueue depending on their policy and
1489 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1490 * nohz mode if necessary.
1492 static inline void sched_update_tick_dependency(struct rq
*rq
)
1496 if (!tick_nohz_full_enabled())
1501 if (!tick_nohz_full_cpu(cpu
))
1504 if (sched_can_stop_tick(rq
))
1505 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1507 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1510 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1513 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1515 unsigned prev_nr
= rq
->nr_running
;
1517 rq
->nr_running
= prev_nr
+ count
;
1519 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1521 if (!rq
->rd
->overload
)
1522 rq
->rd
->overload
= true;
1526 sched_update_tick_dependency(rq
);
1529 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1531 rq
->nr_running
-= count
;
1532 /* Check if we still need preemption */
1533 sched_update_tick_dependency(rq
);
1536 static inline void rq_last_tick_reset(struct rq
*rq
)
1538 #ifdef CONFIG_NO_HZ_FULL
1539 rq
->last_sched_tick
= jiffies
;
1543 extern void update_rq_clock(struct rq
*rq
);
1545 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1546 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1548 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1550 extern const_debug
unsigned int sysctl_sched_time_avg
;
1551 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1552 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1554 static inline u64
sched_avg_period(void)
1556 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1559 #ifdef CONFIG_SCHED_HRTICK
1563 * - enabled by features
1564 * - hrtimer is actually high res
1566 static inline int hrtick_enabled(struct rq
*rq
)
1568 if (!sched_feat(HRTICK
))
1570 if (!cpu_active(cpu_of(rq
)))
1572 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1575 void hrtick_start(struct rq
*rq
, u64 delay
);
1579 static inline int hrtick_enabled(struct rq
*rq
)
1584 #endif /* CONFIG_SCHED_HRTICK */
1587 extern void sched_avg_update(struct rq
*rq
);
1589 #ifndef arch_scale_freq_capacity
1590 static __always_inline
1591 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1593 return SCHED_CAPACITY_SCALE
;
1597 #ifndef arch_scale_cpu_capacity
1598 static __always_inline
1599 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1601 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1602 return sd
->smt_gain
/ sd
->span_weight
;
1604 return SCHED_CAPACITY_SCALE
;
1608 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1610 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1611 sched_avg_update(rq
);
1614 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1615 static inline void sched_avg_update(struct rq
*rq
) { }
1618 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1619 __acquires(rq
->lock
);
1620 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1621 __acquires(p
->pi_lock
)
1622 __acquires(rq
->lock
);
1624 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1625 __releases(rq
->lock
)
1627 rq_unpin_lock(rq
, rf
);
1628 raw_spin_unlock(&rq
->lock
);
1632 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1633 __releases(rq
->lock
)
1634 __releases(p
->pi_lock
)
1636 rq_unpin_lock(rq
, rf
);
1637 raw_spin_unlock(&rq
->lock
);
1638 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1642 #ifdef CONFIG_PREEMPT
1644 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1647 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1648 * way at the expense of forcing extra atomic operations in all
1649 * invocations. This assures that the double_lock is acquired using the
1650 * same underlying policy as the spinlock_t on this architecture, which
1651 * reduces latency compared to the unfair variant below. However, it
1652 * also adds more overhead and therefore may reduce throughput.
1654 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1655 __releases(this_rq
->lock
)
1656 __acquires(busiest
->lock
)
1657 __acquires(this_rq
->lock
)
1659 raw_spin_unlock(&this_rq
->lock
);
1660 double_rq_lock(this_rq
, busiest
);
1667 * Unfair double_lock_balance: Optimizes throughput at the expense of
1668 * latency by eliminating extra atomic operations when the locks are
1669 * already in proper order on entry. This favors lower cpu-ids and will
1670 * grant the double lock to lower cpus over higher ids under contention,
1671 * regardless of entry order into the function.
1673 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1674 __releases(this_rq
->lock
)
1675 __acquires(busiest
->lock
)
1676 __acquires(this_rq
->lock
)
1680 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1681 if (busiest
< this_rq
) {
1682 raw_spin_unlock(&this_rq
->lock
);
1683 raw_spin_lock(&busiest
->lock
);
1684 raw_spin_lock_nested(&this_rq
->lock
,
1685 SINGLE_DEPTH_NESTING
);
1688 raw_spin_lock_nested(&busiest
->lock
,
1689 SINGLE_DEPTH_NESTING
);
1694 #endif /* CONFIG_PREEMPT */
1697 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1699 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1701 if (unlikely(!irqs_disabled())) {
1702 /* printk() doesn't work good under rq->lock */
1703 raw_spin_unlock(&this_rq
->lock
);
1707 return _double_lock_balance(this_rq
, busiest
);
1710 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1711 __releases(busiest
->lock
)
1713 raw_spin_unlock(&busiest
->lock
);
1714 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1717 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1723 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1726 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1732 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1735 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1741 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1745 * double_rq_lock - safely lock two runqueues
1747 * Note this does not disable interrupts like task_rq_lock,
1748 * you need to do so manually before calling.
1750 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1751 __acquires(rq1
->lock
)
1752 __acquires(rq2
->lock
)
1754 BUG_ON(!irqs_disabled());
1756 raw_spin_lock(&rq1
->lock
);
1757 __acquire(rq2
->lock
); /* Fake it out ;) */
1760 raw_spin_lock(&rq1
->lock
);
1761 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1763 raw_spin_lock(&rq2
->lock
);
1764 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1770 * double_rq_unlock - safely unlock two runqueues
1772 * Note this does not restore interrupts like task_rq_unlock,
1773 * you need to do so manually after calling.
1775 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1776 __releases(rq1
->lock
)
1777 __releases(rq2
->lock
)
1779 raw_spin_unlock(&rq1
->lock
);
1781 raw_spin_unlock(&rq2
->lock
);
1783 __release(rq2
->lock
);
1786 extern void set_rq_online (struct rq
*rq
);
1787 extern void set_rq_offline(struct rq
*rq
);
1788 extern bool sched_smp_initialized
;
1790 #else /* CONFIG_SMP */
1793 * double_rq_lock - safely lock two runqueues
1795 * Note this does not disable interrupts like task_rq_lock,
1796 * you need to do so manually before calling.
1798 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1799 __acquires(rq1
->lock
)
1800 __acquires(rq2
->lock
)
1802 BUG_ON(!irqs_disabled());
1804 raw_spin_lock(&rq1
->lock
);
1805 __acquire(rq2
->lock
); /* Fake it out ;) */
1809 * double_rq_unlock - safely unlock two runqueues
1811 * Note this does not restore interrupts like task_rq_unlock,
1812 * you need to do so manually after calling.
1814 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1815 __releases(rq1
->lock
)
1816 __releases(rq2
->lock
)
1819 raw_spin_unlock(&rq1
->lock
);
1820 __release(rq2
->lock
);
1825 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1826 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1828 #ifdef CONFIG_SCHED_DEBUG
1829 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1830 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1831 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1833 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1834 #ifdef CONFIG_NUMA_BALANCING
1836 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1838 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1839 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1840 #endif /* CONFIG_NUMA_BALANCING */
1841 #endif /* CONFIG_SCHED_DEBUG */
1843 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1844 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1845 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1847 extern void cfs_bandwidth_usage_inc(void);
1848 extern void cfs_bandwidth_usage_dec(void);
1850 #ifdef CONFIG_NO_HZ_COMMON
1851 enum rq_nohz_flag_bits
{
1856 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1858 extern void nohz_balance_exit_idle(unsigned int cpu
);
1860 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1863 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1867 struct u64_stats_sync sync
;
1870 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
1872 static inline u64
irq_time_read(int cpu
)
1874 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
1875 u64
*cpustat
= kcpustat_cpu(cpu
).cpustat
;
1880 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
1881 total
= cpustat
[CPUTIME_SOFTIRQ
] + cpustat
[CPUTIME_IRQ
];
1882 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
1886 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1888 #ifdef CONFIG_CPU_FREQ
1889 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
1892 * cpufreq_update_util - Take a note about CPU utilization changes.
1893 * @rq: Runqueue to carry out the update for.
1894 * @flags: Update reason flags.
1896 * This function is called by the scheduler on the CPU whose utilization is
1899 * It can only be called from RCU-sched read-side critical sections.
1901 * The way cpufreq is currently arranged requires it to evaluate the CPU
1902 * performance state (frequency/voltage) on a regular basis to prevent it from
1903 * being stuck in a completely inadequate performance level for too long.
1904 * That is not guaranteed to happen if the updates are only triggered from CFS,
1905 * though, because they may not be coming in if RT or deadline tasks are active
1906 * all the time (or there are RT and DL tasks only).
1908 * As a workaround for that issue, this function is called by the RT and DL
1909 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1910 * but that really is a band-aid. Going forward it should be replaced with
1911 * solutions targeted more specifically at RT and DL tasks.
1913 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
1915 struct update_util_data
*data
;
1917 data
= rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data
));
1919 data
->func(data
, rq_clock(rq
), flags
);
1922 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
)
1924 if (cpu_of(rq
) == smp_processor_id())
1925 cpufreq_update_util(rq
, flags
);
1928 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
1929 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
) {}
1930 #endif /* CONFIG_CPU_FREQ */
1932 #ifdef arch_scale_freq_capacity
1933 #ifndef arch_scale_freq_invariant
1934 #define arch_scale_freq_invariant() (true)
1936 #else /* arch_scale_freq_capacity */
1937 #define arch_scale_freq_invariant() (false)