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/wake_q.h>
7 #include <linux/u64_stats_sync.h>
8 #include <linux/sched/deadline.h>
9 #include <linux/kernel_stat.h>
10 #include <linux/binfmts.h>
11 #include <linux/mutex.h>
12 #include <linux/spinlock.h>
13 #include <linux/stop_machine.h>
14 #include <linux/irq_work.h>
15 #include <linux/tick.h>
16 #include <linux/slab.h>
19 #include "cpudeadline.h"
22 #ifdef CONFIG_SCHED_DEBUG
23 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
25 #define SCHED_WARN_ON(x) ((void)(x))
31 /* task_struct::on_rq states: */
32 #define TASK_ON_RQ_QUEUED 1
33 #define TASK_ON_RQ_MIGRATING 2
35 extern __read_mostly
int scheduler_running
;
37 extern unsigned long calc_load_update
;
38 extern atomic_long_t calc_load_tasks
;
40 extern void calc_global_load_tick(struct rq
*this_rq
);
41 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
44 extern void cpu_load_update_active(struct rq
*this_rq
);
46 static inline void cpu_load_update_active(struct rq
*this_rq
) { }
50 * Helpers for converting nanosecond timing to jiffy resolution
52 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
55 * Increase resolution of nice-level calculations for 64-bit architectures.
56 * The extra resolution improves shares distribution and load balancing of
57 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
58 * hierarchies, especially on larger systems. This is not a user-visible change
59 * and does not change the user-interface for setting shares/weights.
61 * We increase resolution only if we have enough bits to allow this increased
62 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
63 * pretty high and the returns do not justify the increased costs.
65 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
66 * increase coverage and consistency always enable it on 64bit platforms.
69 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
70 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
71 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
73 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
74 # define scale_load(w) (w)
75 # define scale_load_down(w) (w)
79 * Task weight (visible to users) and its load (invisible to users) have
80 * independent resolution, but they should be well calibrated. We use
81 * scale_load() and scale_load_down(w) to convert between them. The
82 * following must be true:
84 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
87 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
90 * Single value that decides SCHED_DEADLINE internal math precision.
91 * 10 -> just above 1us
92 * 9 -> just above 0.5us
97 * These are the 'tuning knobs' of the scheduler:
101 * single value that denotes runtime == period, ie unlimited time.
103 #define RUNTIME_INF ((u64)~0ULL)
105 static inline int idle_policy(int policy
)
107 return policy
== SCHED_IDLE
;
109 static inline int fair_policy(int policy
)
111 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
114 static inline int rt_policy(int policy
)
116 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
119 static inline int dl_policy(int policy
)
121 return policy
== SCHED_DEADLINE
;
123 static inline bool valid_policy(int policy
)
125 return idle_policy(policy
) || fair_policy(policy
) ||
126 rt_policy(policy
) || dl_policy(policy
);
129 static inline int task_has_rt_policy(struct task_struct
*p
)
131 return rt_policy(p
->policy
);
134 static inline int task_has_dl_policy(struct task_struct
*p
)
136 return dl_policy(p
->policy
);
140 * Tells if entity @a should preempt entity @b.
143 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
145 return dl_time_before(a
->deadline
, b
->deadline
);
149 * This is the priority-queue data structure of the RT scheduling class:
151 struct rt_prio_array
{
152 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
153 struct list_head queue
[MAX_RT_PRIO
];
156 struct rt_bandwidth
{
157 /* nests inside the rq lock: */
158 raw_spinlock_t rt_runtime_lock
;
161 struct hrtimer rt_period_timer
;
162 unsigned int rt_period_active
;
165 void __dl_clear_params(struct task_struct
*p
);
168 * To keep the bandwidth of -deadline tasks and groups under control
169 * we need some place where:
170 * - store the maximum -deadline bandwidth of the system (the group);
171 * - cache the fraction of that bandwidth that is currently allocated.
173 * This is all done in the data structure below. It is similar to the
174 * one used for RT-throttling (rt_bandwidth), with the main difference
175 * that, since here we are only interested in admission control, we
176 * do not decrease any runtime while the group "executes", neither we
177 * need a timer to replenish it.
179 * With respect to SMP, the bandwidth is given on a per-CPU basis,
181 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
182 * - dl_total_bw array contains, in the i-eth element, the currently
183 * allocated bandwidth on the i-eth CPU.
184 * Moreover, groups consume bandwidth on each CPU, while tasks only
185 * consume bandwidth on the CPU they're running on.
186 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
187 * that will be shown the next time the proc or cgroup controls will
188 * be red. It on its turn can be changed by writing on its own
191 struct dl_bandwidth
{
192 raw_spinlock_t dl_runtime_lock
;
197 static inline int dl_bandwidth_enabled(void)
199 return sysctl_sched_rt_runtime
>= 0;
202 extern struct dl_bw
*dl_bw_of(int i
);
210 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
212 dl_b
->total_bw
-= tsk_bw
;
216 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
218 dl_b
->total_bw
+= tsk_bw
;
222 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
224 return dl_b
->bw
!= -1 &&
225 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
228 extern void init_dl_bw(struct dl_bw
*dl_b
);
230 #ifdef CONFIG_CGROUP_SCHED
232 #include <linux/cgroup.h>
237 extern struct list_head task_groups
;
239 struct cfs_bandwidth
{
240 #ifdef CONFIG_CFS_BANDWIDTH
244 s64 hierarchical_quota
;
247 int idle
, period_active
;
248 struct hrtimer period_timer
, slack_timer
;
249 struct list_head throttled_cfs_rq
;
252 int nr_periods
, nr_throttled
;
257 /* task group related information */
259 struct cgroup_subsys_state css
;
261 #ifdef CONFIG_FAIR_GROUP_SCHED
262 /* schedulable entities of this group on each cpu */
263 struct sched_entity
**se
;
264 /* runqueue "owned" by this group on each cpu */
265 struct cfs_rq
**cfs_rq
;
266 unsigned long shares
;
270 * load_avg can be heavily contended at clock tick time, so put
271 * it in its own cacheline separated from the fields above which
272 * will also be accessed at each tick.
274 atomic_long_t load_avg ____cacheline_aligned
;
278 #ifdef CONFIG_RT_GROUP_SCHED
279 struct sched_rt_entity
**rt_se
;
280 struct rt_rq
**rt_rq
;
282 struct rt_bandwidth rt_bandwidth
;
286 struct list_head list
;
288 struct task_group
*parent
;
289 struct list_head siblings
;
290 struct list_head children
;
292 #ifdef CONFIG_SCHED_AUTOGROUP
293 struct autogroup
*autogroup
;
296 struct cfs_bandwidth cfs_bandwidth
;
299 #ifdef CONFIG_FAIR_GROUP_SCHED
300 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
303 * A weight of 0 or 1 can cause arithmetics problems.
304 * A weight of a cfs_rq is the sum of weights of which entities
305 * are queued on this cfs_rq, so a weight of a entity should not be
306 * too large, so as the shares value of a task group.
307 * (The default weight is 1024 - so there's no practical
308 * limitation from this.)
310 #define MIN_SHARES (1UL << 1)
311 #define MAX_SHARES (1UL << 18)
314 typedef int (*tg_visitor
)(struct task_group
*, void *);
316 extern int walk_tg_tree_from(struct task_group
*from
,
317 tg_visitor down
, tg_visitor up
, void *data
);
320 * Iterate the full tree, calling @down when first entering a node and @up when
321 * leaving it for the final time.
323 * Caller must hold rcu_lock or sufficient equivalent.
325 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
327 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
330 extern int tg_nop(struct task_group
*tg
, void *data
);
332 extern void free_fair_sched_group(struct task_group
*tg
);
333 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
334 extern void online_fair_sched_group(struct task_group
*tg
);
335 extern void unregister_fair_sched_group(struct task_group
*tg
);
336 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
337 struct sched_entity
*se
, int cpu
,
338 struct sched_entity
*parent
);
339 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
341 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
342 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
343 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
345 extern void free_rt_sched_group(struct task_group
*tg
);
346 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
347 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
348 struct sched_rt_entity
*rt_se
, int cpu
,
349 struct sched_rt_entity
*parent
);
351 extern struct task_group
*sched_create_group(struct task_group
*parent
);
352 extern void sched_online_group(struct task_group
*tg
,
353 struct task_group
*parent
);
354 extern void sched_destroy_group(struct task_group
*tg
);
355 extern void sched_offline_group(struct task_group
*tg
);
357 extern void sched_move_task(struct task_struct
*tsk
);
359 #ifdef CONFIG_FAIR_GROUP_SCHED
360 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
363 extern void set_task_rq_fair(struct sched_entity
*se
,
364 struct cfs_rq
*prev
, struct cfs_rq
*next
);
365 #else /* !CONFIG_SMP */
366 static inline void set_task_rq_fair(struct sched_entity
*se
,
367 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
368 #endif /* CONFIG_SMP */
369 #endif /* CONFIG_FAIR_GROUP_SCHED */
371 #else /* CONFIG_CGROUP_SCHED */
373 struct cfs_bandwidth
{ };
375 #endif /* CONFIG_CGROUP_SCHED */
377 /* CFS-related fields in a runqueue */
379 struct load_weight load
;
380 unsigned int nr_running
, h_nr_running
;
385 u64 min_vruntime_copy
;
388 struct rb_root tasks_timeline
;
389 struct rb_node
*rb_leftmost
;
392 * 'curr' points to currently running entity on this cfs_rq.
393 * It is set to NULL otherwise (i.e when none are currently running).
395 struct sched_entity
*curr
, *next
, *last
, *skip
;
397 #ifdef CONFIG_SCHED_DEBUG
398 unsigned int nr_spread_over
;
405 struct sched_avg avg
;
406 u64 runnable_load_sum
;
407 unsigned long runnable_load_avg
;
408 #ifdef CONFIG_FAIR_GROUP_SCHED
409 unsigned long tg_load_avg_contrib
;
410 unsigned long propagate_avg
;
412 atomic_long_t removed_load_avg
, removed_util_avg
;
414 u64 load_last_update_time_copy
;
417 #ifdef CONFIG_FAIR_GROUP_SCHED
419 * h_load = weight * f(tg)
421 * Where f(tg) is the recursive weight fraction assigned to
424 unsigned long h_load
;
425 u64 last_h_load_update
;
426 struct sched_entity
*h_load_next
;
427 #endif /* CONFIG_FAIR_GROUP_SCHED */
428 #endif /* CONFIG_SMP */
430 #ifdef CONFIG_FAIR_GROUP_SCHED
431 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
434 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
435 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
436 * (like users, containers etc.)
438 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
439 * list is used during load balance.
442 struct list_head leaf_cfs_rq_list
;
443 struct task_group
*tg
; /* group that "owns" this runqueue */
445 #ifdef CONFIG_CFS_BANDWIDTH
448 s64 runtime_remaining
;
450 u64 throttled_clock
, throttled_clock_task
;
451 u64 throttled_clock_task_time
;
452 int throttled
, throttle_count
;
453 struct list_head throttled_list
;
454 #endif /* CONFIG_CFS_BANDWIDTH */
455 #endif /* CONFIG_FAIR_GROUP_SCHED */
458 static inline int rt_bandwidth_enabled(void)
460 return sysctl_sched_rt_runtime
>= 0;
463 /* RT IPI pull logic requires IRQ_WORK */
464 #ifdef CONFIG_IRQ_WORK
465 # define HAVE_RT_PUSH_IPI
468 /* Real-Time classes' related field in a runqueue: */
470 struct rt_prio_array active
;
471 unsigned int rt_nr_running
;
472 unsigned int rr_nr_running
;
473 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
475 int curr
; /* highest queued rt task prio */
477 int next
; /* next highest */
482 unsigned long rt_nr_migratory
;
483 unsigned long rt_nr_total
;
485 struct plist_head pushable_tasks
;
486 #ifdef HAVE_RT_PUSH_IPI
489 struct irq_work push_work
;
490 raw_spinlock_t push_lock
;
492 #endif /* CONFIG_SMP */
498 /* Nests inside the rq lock: */
499 raw_spinlock_t rt_runtime_lock
;
501 #ifdef CONFIG_RT_GROUP_SCHED
502 unsigned long rt_nr_boosted
;
505 struct task_group
*tg
;
509 /* Deadline class' related fields in a runqueue */
511 /* runqueue is an rbtree, ordered by deadline */
512 struct rb_root rb_root
;
513 struct rb_node
*rb_leftmost
;
515 unsigned long dl_nr_running
;
519 * Deadline values of the currently executing and the
520 * earliest ready task on this rq. Caching these facilitates
521 * the decision wether or not a ready but not running task
522 * should migrate somewhere else.
529 unsigned long dl_nr_migratory
;
533 * Tasks on this rq that can be pushed away. They are kept in
534 * an rb-tree, ordered by tasks' deadlines, with caching
535 * of the leftmost (earliest deadline) element.
537 struct rb_root pushable_dl_tasks_root
;
538 struct rb_node
*pushable_dl_tasks_leftmost
;
546 static inline bool sched_asym_prefer(int a
, int b
)
548 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
552 * We add the notion of a root-domain which will be used to define per-domain
553 * variables. Each exclusive cpuset essentially defines an island domain by
554 * fully partitioning the member cpus from any other cpuset. Whenever a new
555 * exclusive cpuset is created, we also create and attach a new root-domain
564 cpumask_var_t online
;
566 /* Indicate more than one runnable task for any CPU */
570 * The bit corresponding to a CPU gets set here if such CPU has more
571 * than one runnable -deadline task (as it is below for RT tasks).
573 cpumask_var_t dlo_mask
;
579 * The "RT overload" flag: it gets set if a CPU has more than
580 * one runnable RT task.
582 cpumask_var_t rto_mask
;
583 struct cpupri cpupri
;
585 unsigned long max_cpu_capacity
;
588 extern struct root_domain def_root_domain
;
589 extern struct mutex sched_domains_mutex
;
590 extern cpumask_var_t fallback_doms
;
591 extern cpumask_var_t sched_domains_tmpmask
;
593 extern void init_defrootdomain(void);
594 extern int init_sched_domains(const struct cpumask
*cpu_map
);
595 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
597 #endif /* CONFIG_SMP */
600 * This is the main, per-CPU runqueue data structure.
602 * Locking rule: those places that want to lock multiple runqueues
603 * (such as the load balancing or the thread migration code), lock
604 * acquire operations must be ordered by ascending &runqueue.
611 * nr_running and cpu_load should be in the same cacheline because
612 * remote CPUs use both these fields when doing load calculation.
614 unsigned int nr_running
;
615 #ifdef CONFIG_NUMA_BALANCING
616 unsigned int nr_numa_running
;
617 unsigned int nr_preferred_running
;
619 #define CPU_LOAD_IDX_MAX 5
620 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
621 #ifdef CONFIG_NO_HZ_COMMON
623 unsigned long last_load_update_tick
;
624 #endif /* CONFIG_SMP */
625 unsigned long nohz_flags
;
626 #endif /* CONFIG_NO_HZ_COMMON */
627 #ifdef CONFIG_NO_HZ_FULL
628 unsigned long last_sched_tick
;
630 /* capture load from *all* tasks on this cpu: */
631 struct load_weight load
;
632 unsigned long nr_load_updates
;
639 #ifdef CONFIG_FAIR_GROUP_SCHED
640 /* list of leaf cfs_rq on this cpu: */
641 struct list_head leaf_cfs_rq_list
;
642 struct list_head
*tmp_alone_branch
;
643 #endif /* CONFIG_FAIR_GROUP_SCHED */
646 * This is part of a global counter where only the total sum
647 * over all CPUs matters. A task can increase this counter on
648 * one CPU and if it got migrated afterwards it may decrease
649 * it on another CPU. Always updated under the runqueue lock:
651 unsigned long nr_uninterruptible
;
653 struct task_struct
*curr
, *idle
, *stop
;
654 unsigned long next_balance
;
655 struct mm_struct
*prev_mm
;
657 unsigned int clock_update_flags
;
664 struct root_domain
*rd
;
665 struct sched_domain
*sd
;
667 unsigned long cpu_capacity
;
668 unsigned long cpu_capacity_orig
;
670 struct callback_head
*balance_callback
;
672 unsigned char idle_balance
;
673 /* For active balancing */
676 struct cpu_stop_work active_balance_work
;
677 /* cpu of this runqueue: */
681 struct list_head cfs_tasks
;
688 /* This is used to determine avg_idle's max value */
689 u64 max_idle_balance_cost
;
692 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
695 #ifdef CONFIG_PARAVIRT
698 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
699 u64 prev_steal_time_rq
;
702 /* calc_load related fields */
703 unsigned long calc_load_update
;
704 long calc_load_active
;
706 #ifdef CONFIG_SCHED_HRTICK
708 int hrtick_csd_pending
;
709 struct call_single_data hrtick_csd
;
711 struct hrtimer hrtick_timer
;
714 #ifdef CONFIG_SCHEDSTATS
716 struct sched_info rq_sched_info
;
717 unsigned long long rq_cpu_time
;
718 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
720 /* sys_sched_yield() stats */
721 unsigned int yld_count
;
723 /* schedule() stats */
724 unsigned int sched_count
;
725 unsigned int sched_goidle
;
727 /* try_to_wake_up() stats */
728 unsigned int ttwu_count
;
729 unsigned int ttwu_local
;
733 struct llist_head wake_list
;
736 #ifdef CONFIG_CPU_IDLE
737 /* Must be inspected within a rcu lock section */
738 struct cpuidle_state
*idle_state
;
742 static inline int cpu_of(struct rq
*rq
)
752 #ifdef CONFIG_SCHED_SMT
754 extern struct static_key_false sched_smt_present
;
756 extern void __update_idle_core(struct rq
*rq
);
758 static inline void update_idle_core(struct rq
*rq
)
760 if (static_branch_unlikely(&sched_smt_present
))
761 __update_idle_core(rq
);
765 static inline void update_idle_core(struct rq
*rq
) { }
768 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
770 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
771 #define this_rq() this_cpu_ptr(&runqueues)
772 #define task_rq(p) cpu_rq(task_cpu(p))
773 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
774 #define raw_rq() raw_cpu_ptr(&runqueues)
776 static inline u64
__rq_clock_broken(struct rq
*rq
)
778 return READ_ONCE(rq
->clock
);
782 * rq::clock_update_flags bits
784 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
785 * call to __schedule(). This is an optimisation to avoid
786 * neighbouring rq clock updates.
788 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
789 * in effect and calls to update_rq_clock() are being ignored.
791 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
792 * made to update_rq_clock() since the last time rq::lock was pinned.
794 * If inside of __schedule(), clock_update_flags will have been
795 * shifted left (a left shift is a cheap operation for the fast path
796 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
798 * if (rq-clock_update_flags >= RQCF_UPDATED)
800 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
801 * one position though, because the next rq_unpin_lock() will shift it
804 #define RQCF_REQ_SKIP 0x01
805 #define RQCF_ACT_SKIP 0x02
806 #define RQCF_UPDATED 0x04
808 static inline void assert_clock_updated(struct rq
*rq
)
811 * The only reason for not seeing a clock update since the
812 * last rq_pin_lock() is if we're currently skipping updates.
814 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
817 static inline u64
rq_clock(struct rq
*rq
)
819 lockdep_assert_held(&rq
->lock
);
820 assert_clock_updated(rq
);
825 static inline u64
rq_clock_task(struct rq
*rq
)
827 lockdep_assert_held(&rq
->lock
);
828 assert_clock_updated(rq
);
830 return rq
->clock_task
;
833 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
835 lockdep_assert_held(&rq
->lock
);
837 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
839 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
844 struct pin_cookie cookie
;
845 #ifdef CONFIG_SCHED_DEBUG
847 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
848 * current pin context is stashed here in case it needs to be
849 * restored in rq_repin_lock().
851 unsigned int clock_update_flags
;
855 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
857 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
859 #ifdef CONFIG_SCHED_DEBUG
860 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
861 rf
->clock_update_flags
= 0;
865 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
867 #ifdef CONFIG_SCHED_DEBUG
868 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
869 rf
->clock_update_flags
= RQCF_UPDATED
;
872 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
875 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
877 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
879 #ifdef CONFIG_SCHED_DEBUG
881 * Restore the value we stashed in @rf for this pin context.
883 rq
->clock_update_flags
|= rf
->clock_update_flags
;
888 enum numa_topology_type
{
893 extern enum numa_topology_type sched_numa_topology_type
;
894 extern int sched_max_numa_distance
;
895 extern bool find_numa_distance(int distance
);
899 extern void sched_init_numa(void);
900 extern void sched_domains_numa_masks_set(unsigned int cpu
);
901 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
903 static inline void sched_init_numa(void) { }
904 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
905 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
908 #ifdef CONFIG_NUMA_BALANCING
909 /* The regions in numa_faults array from task_struct */
910 enum numa_faults_stats
{
916 extern void sched_setnuma(struct task_struct
*p
, int node
);
917 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
918 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
919 #endif /* CONFIG_NUMA_BALANCING */
924 queue_balance_callback(struct rq
*rq
,
925 struct callback_head
*head
,
926 void (*func
)(struct rq
*rq
))
928 lockdep_assert_held(&rq
->lock
);
930 if (unlikely(head
->next
))
933 head
->func
= (void (*)(struct callback_head
*))func
;
934 head
->next
= rq
->balance_callback
;
935 rq
->balance_callback
= head
;
938 extern void sched_ttwu_pending(void);
940 #define rcu_dereference_check_sched_domain(p) \
941 rcu_dereference_check((p), \
942 lockdep_is_held(&sched_domains_mutex))
945 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
946 * See detach_destroy_domains: synchronize_sched for details.
948 * The domain tree of any CPU may only be accessed from within
949 * preempt-disabled sections.
951 #define for_each_domain(cpu, __sd) \
952 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
953 __sd; __sd = __sd->parent)
955 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
958 * highest_flag_domain - Return highest sched_domain containing flag.
959 * @cpu: The cpu whose highest level of sched domain is to
961 * @flag: The flag to check for the highest sched_domain
964 * Returns the highest sched_domain of a cpu which contains the given flag.
966 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
968 struct sched_domain
*sd
, *hsd
= NULL
;
970 for_each_domain(cpu
, sd
) {
971 if (!(sd
->flags
& flag
))
979 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
981 struct sched_domain
*sd
;
983 for_each_domain(cpu
, sd
) {
984 if (sd
->flags
& flag
)
991 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
992 DECLARE_PER_CPU(int, sd_llc_size
);
993 DECLARE_PER_CPU(int, sd_llc_id
);
994 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
995 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
996 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
998 struct sched_group_capacity
{
1001 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1004 unsigned long capacity
;
1005 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1006 unsigned long next_update
;
1007 int imbalance
; /* XXX unrelated to capacity but shared group state */
1009 unsigned long cpumask
[0]; /* iteration mask */
1012 struct sched_group
{
1013 struct sched_group
*next
; /* Must be a circular list */
1016 unsigned int group_weight
;
1017 struct sched_group_capacity
*sgc
;
1018 int asym_prefer_cpu
; /* cpu of highest priority in group */
1021 * The CPUs this group covers.
1023 * NOTE: this field is variable length. (Allocated dynamically
1024 * by attaching extra space to the end of the structure,
1025 * depending on how many CPUs the kernel has booted up with)
1027 unsigned long cpumask
[0];
1030 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
1032 return to_cpumask(sg
->cpumask
);
1036 * cpumask masking which cpus in the group are allowed to iterate up the domain
1039 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
1041 return to_cpumask(sg
->sgc
->cpumask
);
1045 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1046 * @group: The group whose first cpu is to be returned.
1048 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1050 return cpumask_first(sched_group_cpus(group
));
1053 extern int group_balance_cpu(struct sched_group
*sg
);
1055 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1056 void register_sched_domain_sysctl(void);
1057 void unregister_sched_domain_sysctl(void);
1059 static inline void register_sched_domain_sysctl(void)
1062 static inline void unregister_sched_domain_sysctl(void)
1069 static inline void sched_ttwu_pending(void) { }
1071 #endif /* CONFIG_SMP */
1074 #include "autogroup.h"
1076 #ifdef CONFIG_CGROUP_SCHED
1079 * Return the group to which this tasks belongs.
1081 * We cannot use task_css() and friends because the cgroup subsystem
1082 * changes that value before the cgroup_subsys::attach() method is called,
1083 * therefore we cannot pin it and might observe the wrong value.
1085 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1086 * core changes this before calling sched_move_task().
1088 * Instead we use a 'copy' which is updated from sched_move_task() while
1089 * holding both task_struct::pi_lock and rq::lock.
1091 static inline struct task_group
*task_group(struct task_struct
*p
)
1093 return p
->sched_task_group
;
1096 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1097 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1099 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1100 struct task_group
*tg
= task_group(p
);
1103 #ifdef CONFIG_FAIR_GROUP_SCHED
1104 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1105 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1106 p
->se
.parent
= tg
->se
[cpu
];
1109 #ifdef CONFIG_RT_GROUP_SCHED
1110 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1111 p
->rt
.parent
= tg
->rt_se
[cpu
];
1115 #else /* CONFIG_CGROUP_SCHED */
1117 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1118 static inline struct task_group
*task_group(struct task_struct
*p
)
1123 #endif /* CONFIG_CGROUP_SCHED */
1125 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1127 set_task_rq(p
, cpu
);
1130 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1131 * successfuly executed on another CPU. We must ensure that updates of
1132 * per-task data have been completed by this moment.
1135 #ifdef CONFIG_THREAD_INFO_IN_TASK
1138 task_thread_info(p
)->cpu
= cpu
;
1145 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1147 #ifdef CONFIG_SCHED_DEBUG
1148 # include <linux/static_key.h>
1149 # define const_debug __read_mostly
1151 # define const_debug const
1154 extern const_debug
unsigned int sysctl_sched_features
;
1156 #define SCHED_FEAT(name, enabled) \
1157 __SCHED_FEAT_##name ,
1160 #include "features.h"
1166 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1167 #define SCHED_FEAT(name, enabled) \
1168 static __always_inline bool static_branch_##name(struct static_key *key) \
1170 return static_key_##enabled(key); \
1173 #include "features.h"
1177 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1178 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1179 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1180 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1181 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1183 extern struct static_key_false sched_numa_balancing
;
1184 extern struct static_key_false sched_schedstats
;
1186 static inline u64
global_rt_period(void)
1188 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1191 static inline u64
global_rt_runtime(void)
1193 if (sysctl_sched_rt_runtime
< 0)
1196 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1199 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1201 return rq
->curr
== p
;
1204 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1209 return task_current(rq
, p
);
1213 static inline int task_on_rq_queued(struct task_struct
*p
)
1215 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1218 static inline int task_on_rq_migrating(struct task_struct
*p
)
1220 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1223 #ifndef prepare_arch_switch
1224 # define prepare_arch_switch(next) do { } while (0)
1226 #ifndef finish_arch_post_lock_switch
1227 # define finish_arch_post_lock_switch() do { } while (0)
1230 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1234 * We can optimise this out completely for !SMP, because the
1235 * SMP rebalancing from interrupt is the only thing that cares
1242 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1246 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1247 * We must ensure this doesn't happen until the switch is completely
1250 * In particular, the load of prev->state in finish_task_switch() must
1251 * happen before this.
1253 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1255 smp_store_release(&prev
->on_cpu
, 0);
1257 #ifdef CONFIG_DEBUG_SPINLOCK
1258 /* this is a valid case when another task releases the spinlock */
1259 rq
->lock
.owner
= current
;
1262 * If we are tracking spinlock dependencies then we have to
1263 * fix up the runqueue lock - which gets 'carried over' from
1264 * prev into current:
1266 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1268 raw_spin_unlock_irq(&rq
->lock
);
1274 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1275 #define WF_FORK 0x02 /* child wakeup after fork */
1276 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1279 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1280 * of tasks with abnormal "nice" values across CPUs the contribution that
1281 * each task makes to its run queue's load is weighted according to its
1282 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1283 * scaled version of the new time slice allocation that they receive on time
1287 #define WEIGHT_IDLEPRIO 3
1288 #define WMULT_IDLEPRIO 1431655765
1290 extern const int sched_prio_to_weight
[40];
1291 extern const u32 sched_prio_to_wmult
[40];
1294 * {de,en}queue flags:
1296 * DEQUEUE_SLEEP - task is no longer runnable
1297 * ENQUEUE_WAKEUP - task just became runnable
1299 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1300 * are in a known state which allows modification. Such pairs
1301 * should preserve as much state as possible.
1303 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1306 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1307 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1308 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1312 #define DEQUEUE_SLEEP 0x01
1313 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1314 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1316 #define ENQUEUE_WAKEUP 0x01
1317 #define ENQUEUE_RESTORE 0x02
1318 #define ENQUEUE_MOVE 0x04
1320 #define ENQUEUE_HEAD 0x08
1321 #define ENQUEUE_REPLENISH 0x10
1323 #define ENQUEUE_MIGRATED 0x20
1325 #define ENQUEUE_MIGRATED 0x00
1328 #define RETRY_TASK ((void *)-1UL)
1330 struct sched_class
{
1331 const struct sched_class
*next
;
1333 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1334 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1335 void (*yield_task
) (struct rq
*rq
);
1336 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1338 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1341 * It is the responsibility of the pick_next_task() method that will
1342 * return the next task to call put_prev_task() on the @prev task or
1343 * something equivalent.
1345 * May return RETRY_TASK when it finds a higher prio class has runnable
1348 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1349 struct task_struct
*prev
,
1350 struct rq_flags
*rf
);
1351 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1354 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1355 void (*migrate_task_rq
)(struct task_struct
*p
);
1357 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1359 void (*set_cpus_allowed
)(struct task_struct
*p
,
1360 const struct cpumask
*newmask
);
1362 void (*rq_online
)(struct rq
*rq
);
1363 void (*rq_offline
)(struct rq
*rq
);
1366 void (*set_curr_task
) (struct rq
*rq
);
1367 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1368 void (*task_fork
) (struct task_struct
*p
);
1369 void (*task_dead
) (struct task_struct
*p
);
1372 * The switched_from() call is allowed to drop rq->lock, therefore we
1373 * cannot assume the switched_from/switched_to pair is serliazed by
1374 * rq->lock. They are however serialized by p->pi_lock.
1376 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1377 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1378 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1381 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1382 struct task_struct
*task
);
1384 void (*update_curr
) (struct rq
*rq
);
1386 #define TASK_SET_GROUP 0
1387 #define TASK_MOVE_GROUP 1
1389 #ifdef CONFIG_FAIR_GROUP_SCHED
1390 void (*task_change_group
) (struct task_struct
*p
, int type
);
1394 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1396 prev
->sched_class
->put_prev_task(rq
, prev
);
1399 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1401 curr
->sched_class
->set_curr_task(rq
);
1404 #define sched_class_highest (&stop_sched_class)
1405 #define for_each_class(class) \
1406 for (class = sched_class_highest; class; class = class->next)
1408 extern const struct sched_class stop_sched_class
;
1409 extern const struct sched_class dl_sched_class
;
1410 extern const struct sched_class rt_sched_class
;
1411 extern const struct sched_class fair_sched_class
;
1412 extern const struct sched_class idle_sched_class
;
1417 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1419 extern void trigger_load_balance(struct rq
*rq
);
1421 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1425 #ifdef CONFIG_CPU_IDLE
1426 static inline void idle_set_state(struct rq
*rq
,
1427 struct cpuidle_state
*idle_state
)
1429 rq
->idle_state
= idle_state
;
1432 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1434 SCHED_WARN_ON(!rcu_read_lock_held());
1435 return rq
->idle_state
;
1438 static inline void idle_set_state(struct rq
*rq
,
1439 struct cpuidle_state
*idle_state
)
1443 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1449 extern void sysrq_sched_debug_show(void);
1450 extern void sched_init_granularity(void);
1451 extern void update_max_interval(void);
1453 extern void init_sched_dl_class(void);
1454 extern void init_sched_rt_class(void);
1455 extern void init_sched_fair_class(void);
1457 extern void resched_curr(struct rq
*rq
);
1458 extern void resched_cpu(int cpu
);
1460 extern struct rt_bandwidth def_rt_bandwidth
;
1461 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1463 extern struct dl_bandwidth def_dl_bandwidth
;
1464 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1465 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1467 unsigned long to_ratio(u64 period
, u64 runtime
);
1469 extern void init_entity_runnable_average(struct sched_entity
*se
);
1470 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1472 #ifdef CONFIG_NO_HZ_FULL
1473 extern bool sched_can_stop_tick(struct rq
*rq
);
1476 * Tick may be needed by tasks in the runqueue depending on their policy and
1477 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1478 * nohz mode if necessary.
1480 static inline void sched_update_tick_dependency(struct rq
*rq
)
1484 if (!tick_nohz_full_enabled())
1489 if (!tick_nohz_full_cpu(cpu
))
1492 if (sched_can_stop_tick(rq
))
1493 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1495 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1498 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1501 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1503 unsigned prev_nr
= rq
->nr_running
;
1505 rq
->nr_running
= prev_nr
+ count
;
1507 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1509 if (!rq
->rd
->overload
)
1510 rq
->rd
->overload
= true;
1514 sched_update_tick_dependency(rq
);
1517 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1519 rq
->nr_running
-= count
;
1520 /* Check if we still need preemption */
1521 sched_update_tick_dependency(rq
);
1524 static inline void rq_last_tick_reset(struct rq
*rq
)
1526 #ifdef CONFIG_NO_HZ_FULL
1527 rq
->last_sched_tick
= jiffies
;
1531 extern void update_rq_clock(struct rq
*rq
);
1533 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1534 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1536 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1538 extern const_debug
unsigned int sysctl_sched_time_avg
;
1539 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1540 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1542 static inline u64
sched_avg_period(void)
1544 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1547 #ifdef CONFIG_SCHED_HRTICK
1551 * - enabled by features
1552 * - hrtimer is actually high res
1554 static inline int hrtick_enabled(struct rq
*rq
)
1556 if (!sched_feat(HRTICK
))
1558 if (!cpu_active(cpu_of(rq
)))
1560 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1563 void hrtick_start(struct rq
*rq
, u64 delay
);
1567 static inline int hrtick_enabled(struct rq
*rq
)
1572 #endif /* CONFIG_SCHED_HRTICK */
1575 extern void sched_avg_update(struct rq
*rq
);
1577 #ifndef arch_scale_freq_capacity
1578 static __always_inline
1579 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1581 return SCHED_CAPACITY_SCALE
;
1585 #ifndef arch_scale_cpu_capacity
1586 static __always_inline
1587 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1589 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1590 return sd
->smt_gain
/ sd
->span_weight
;
1592 return SCHED_CAPACITY_SCALE
;
1596 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1598 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1599 sched_avg_update(rq
);
1602 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1603 static inline void sched_avg_update(struct rq
*rq
) { }
1606 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1607 __acquires(rq
->lock
);
1608 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1609 __acquires(p
->pi_lock
)
1610 __acquires(rq
->lock
);
1612 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1613 __releases(rq
->lock
)
1615 rq_unpin_lock(rq
, rf
);
1616 raw_spin_unlock(&rq
->lock
);
1620 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1621 __releases(rq
->lock
)
1622 __releases(p
->pi_lock
)
1624 rq_unpin_lock(rq
, rf
);
1625 raw_spin_unlock(&rq
->lock
);
1626 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1630 #ifdef CONFIG_PREEMPT
1632 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1635 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1636 * way at the expense of forcing extra atomic operations in all
1637 * invocations. This assures that the double_lock is acquired using the
1638 * same underlying policy as the spinlock_t on this architecture, which
1639 * reduces latency compared to the unfair variant below. However, it
1640 * also adds more overhead and therefore may reduce throughput.
1642 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1643 __releases(this_rq
->lock
)
1644 __acquires(busiest
->lock
)
1645 __acquires(this_rq
->lock
)
1647 raw_spin_unlock(&this_rq
->lock
);
1648 double_rq_lock(this_rq
, busiest
);
1655 * Unfair double_lock_balance: Optimizes throughput at the expense of
1656 * latency by eliminating extra atomic operations when the locks are
1657 * already in proper order on entry. This favors lower cpu-ids and will
1658 * grant the double lock to lower cpus over higher ids under contention,
1659 * regardless of entry order into the function.
1661 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1662 __releases(this_rq
->lock
)
1663 __acquires(busiest
->lock
)
1664 __acquires(this_rq
->lock
)
1668 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1669 if (busiest
< this_rq
) {
1670 raw_spin_unlock(&this_rq
->lock
);
1671 raw_spin_lock(&busiest
->lock
);
1672 raw_spin_lock_nested(&this_rq
->lock
,
1673 SINGLE_DEPTH_NESTING
);
1676 raw_spin_lock_nested(&busiest
->lock
,
1677 SINGLE_DEPTH_NESTING
);
1682 #endif /* CONFIG_PREEMPT */
1685 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1687 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1689 if (unlikely(!irqs_disabled())) {
1690 /* printk() doesn't work good under rq->lock */
1691 raw_spin_unlock(&this_rq
->lock
);
1695 return _double_lock_balance(this_rq
, busiest
);
1698 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1699 __releases(busiest
->lock
)
1701 raw_spin_unlock(&busiest
->lock
);
1702 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1705 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1711 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1714 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1720 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1723 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1729 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1733 * double_rq_lock - safely lock two runqueues
1735 * Note this does not disable interrupts like task_rq_lock,
1736 * you need to do so manually before calling.
1738 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1739 __acquires(rq1
->lock
)
1740 __acquires(rq2
->lock
)
1742 BUG_ON(!irqs_disabled());
1744 raw_spin_lock(&rq1
->lock
);
1745 __acquire(rq2
->lock
); /* Fake it out ;) */
1748 raw_spin_lock(&rq1
->lock
);
1749 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1751 raw_spin_lock(&rq2
->lock
);
1752 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1758 * double_rq_unlock - safely unlock two runqueues
1760 * Note this does not restore interrupts like task_rq_unlock,
1761 * you need to do so manually after calling.
1763 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1764 __releases(rq1
->lock
)
1765 __releases(rq2
->lock
)
1767 raw_spin_unlock(&rq1
->lock
);
1769 raw_spin_unlock(&rq2
->lock
);
1771 __release(rq2
->lock
);
1774 extern void set_rq_online (struct rq
*rq
);
1775 extern void set_rq_offline(struct rq
*rq
);
1776 extern bool sched_smp_initialized
;
1778 #else /* CONFIG_SMP */
1781 * double_rq_lock - safely lock two runqueues
1783 * Note this does not disable interrupts like task_rq_lock,
1784 * you need to do so manually before calling.
1786 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1787 __acquires(rq1
->lock
)
1788 __acquires(rq2
->lock
)
1790 BUG_ON(!irqs_disabled());
1792 raw_spin_lock(&rq1
->lock
);
1793 __acquire(rq2
->lock
); /* Fake it out ;) */
1797 * double_rq_unlock - safely unlock two runqueues
1799 * Note this does not restore interrupts like task_rq_unlock,
1800 * you need to do so manually after calling.
1802 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1803 __releases(rq1
->lock
)
1804 __releases(rq2
->lock
)
1807 raw_spin_unlock(&rq1
->lock
);
1808 __release(rq2
->lock
);
1813 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1814 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1816 #ifdef CONFIG_SCHED_DEBUG
1817 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1818 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1819 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1821 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1823 #ifdef CONFIG_NUMA_BALANCING
1825 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1827 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1828 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1829 #endif /* CONFIG_NUMA_BALANCING */
1830 #endif /* CONFIG_SCHED_DEBUG */
1832 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1833 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1834 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1836 extern void cfs_bandwidth_usage_inc(void);
1837 extern void cfs_bandwidth_usage_dec(void);
1839 #ifdef CONFIG_NO_HZ_COMMON
1840 enum rq_nohz_flag_bits
{
1845 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1847 extern void nohz_balance_exit_idle(unsigned int cpu
);
1849 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1852 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1856 struct u64_stats_sync sync
;
1859 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
1861 static inline u64
irq_time_read(int cpu
)
1863 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
1864 u64
*cpustat
= kcpustat_cpu(cpu
).cpustat
;
1869 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
1870 total
= cpustat
[CPUTIME_SOFTIRQ
] + cpustat
[CPUTIME_IRQ
];
1871 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
1875 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1877 #ifdef CONFIG_CPU_FREQ
1878 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
1881 * cpufreq_update_util - Take a note about CPU utilization changes.
1882 * @rq: Runqueue to carry out the update for.
1883 * @flags: Update reason flags.
1885 * This function is called by the scheduler on the CPU whose utilization is
1888 * It can only be called from RCU-sched read-side critical sections.
1890 * The way cpufreq is currently arranged requires it to evaluate the CPU
1891 * performance state (frequency/voltage) on a regular basis to prevent it from
1892 * being stuck in a completely inadequate performance level for too long.
1893 * That is not guaranteed to happen if the updates are only triggered from CFS,
1894 * though, because they may not be coming in if RT or deadline tasks are active
1895 * all the time (or there are RT and DL tasks only).
1897 * As a workaround for that issue, this function is called by the RT and DL
1898 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1899 * but that really is a band-aid. Going forward it should be replaced with
1900 * solutions targeted more specifically at RT and DL tasks.
1902 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
1904 struct update_util_data
*data
;
1906 data
= rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data
));
1908 data
->func(data
, rq_clock(rq
), flags
);
1911 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
)
1913 if (cpu_of(rq
) == smp_processor_id())
1914 cpufreq_update_util(rq
, flags
);
1917 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
1918 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
) {}
1919 #endif /* CONFIG_CPU_FREQ */
1921 #ifdef arch_scale_freq_capacity
1922 #ifndef arch_scale_freq_invariant
1923 #define arch_scale_freq_invariant() (true)
1925 #else /* arch_scale_freq_capacity */
1926 #define arch_scale_freq_invariant() (false)