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/deadline.h>
7 #include <linux/sched/clock.h>
8 #include <linux/sched/wake_q.h>
9 #include <linux/sched/signal.h>
10 #include <linux/sched/numa_balancing.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/cpufreq.h>
13 #include <linux/sched/stat.h>
14 #include <linux/sched/nohz.h>
15 #include <linux/sched/debug.h>
16 #include <linux/sched/hotplug.h>
18 #include <linux/u64_stats_sync.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/binfmts.h>
21 #include <linux/mutex.h>
22 #include <linux/spinlock.h>
23 #include <linux/stop_machine.h>
24 #include <linux/irq_work.h>
25 #include <linux/tick.h>
26 #include <linux/slab.h>
28 #ifdef CONFIG_PARAVIRT
29 #include <asm/paravirt.h>
33 #include "cpudeadline.h"
36 #ifdef CONFIG_SCHED_DEBUG
37 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
39 #define SCHED_WARN_ON(x) ((void)(x))
45 /* task_struct::on_rq states: */
46 #define TASK_ON_RQ_QUEUED 1
47 #define TASK_ON_RQ_MIGRATING 2
49 extern __read_mostly
int scheduler_running
;
51 extern unsigned long calc_load_update
;
52 extern atomic_long_t calc_load_tasks
;
54 extern void calc_global_load_tick(struct rq
*this_rq
);
55 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
58 extern void cpu_load_update_active(struct rq
*this_rq
);
60 static inline void cpu_load_update_active(struct rq
*this_rq
) { }
64 * Helpers for converting nanosecond timing to jiffy resolution
66 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
69 * Increase resolution of nice-level calculations for 64-bit architectures.
70 * The extra resolution improves shares distribution and load balancing of
71 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
72 * hierarchies, especially on larger systems. This is not a user-visible change
73 * and does not change the user-interface for setting shares/weights.
75 * We increase resolution only if we have enough bits to allow this increased
76 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
77 * pretty high and the returns do not justify the increased costs.
79 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
80 * increase coverage and consistency always enable it on 64bit platforms.
83 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
84 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
85 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
87 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
88 # define scale_load(w) (w)
89 # define scale_load_down(w) (w)
93 * Task weight (visible to users) and its load (invisible to users) have
94 * independent resolution, but they should be well calibrated. We use
95 * scale_load() and scale_load_down(w) to convert between them. The
96 * following must be true:
98 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
101 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
104 * Single value that decides SCHED_DEADLINE internal math precision.
105 * 10 -> just above 1us
106 * 9 -> just above 0.5us
108 #define DL_SCALE (10)
111 * These are the 'tuning knobs' of the scheduler:
115 * single value that denotes runtime == period, ie unlimited time.
117 #define RUNTIME_INF ((u64)~0ULL)
119 static inline int idle_policy(int policy
)
121 return policy
== SCHED_IDLE
;
123 static inline int fair_policy(int policy
)
125 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
128 static inline int rt_policy(int policy
)
130 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
133 static inline int dl_policy(int policy
)
135 return policy
== SCHED_DEADLINE
;
137 static inline bool valid_policy(int policy
)
139 return idle_policy(policy
) || fair_policy(policy
) ||
140 rt_policy(policy
) || dl_policy(policy
);
143 static inline int task_has_rt_policy(struct task_struct
*p
)
145 return rt_policy(p
->policy
);
148 static inline int task_has_dl_policy(struct task_struct
*p
)
150 return dl_policy(p
->policy
);
154 * Tells if entity @a should preempt entity @b.
157 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
159 return dl_time_before(a
->deadline
, b
->deadline
);
163 * This is the priority-queue data structure of the RT scheduling class:
165 struct rt_prio_array
{
166 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
167 struct list_head queue
[MAX_RT_PRIO
];
170 struct rt_bandwidth
{
171 /* nests inside the rq lock: */
172 raw_spinlock_t rt_runtime_lock
;
175 struct hrtimer rt_period_timer
;
176 unsigned int rt_period_active
;
179 void __dl_clear_params(struct task_struct
*p
);
182 * To keep the bandwidth of -deadline tasks and groups under control
183 * we need some place where:
184 * - store the maximum -deadline bandwidth of the system (the group);
185 * - cache the fraction of that bandwidth that is currently allocated.
187 * This is all done in the data structure below. It is similar to the
188 * one used for RT-throttling (rt_bandwidth), with the main difference
189 * that, since here we are only interested in admission control, we
190 * do not decrease any runtime while the group "executes", neither we
191 * need a timer to replenish it.
193 * With respect to SMP, the bandwidth is given on a per-CPU basis,
195 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
196 * - dl_total_bw array contains, in the i-eth element, the currently
197 * allocated bandwidth on the i-eth CPU.
198 * Moreover, groups consume bandwidth on each CPU, while tasks only
199 * consume bandwidth on the CPU they're running on.
200 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
201 * that will be shown the next time the proc or cgroup controls will
202 * be red. It on its turn can be changed by writing on its own
205 struct dl_bandwidth
{
206 raw_spinlock_t dl_runtime_lock
;
211 static inline int dl_bandwidth_enabled(void)
213 return sysctl_sched_rt_runtime
>= 0;
216 extern struct dl_bw
*dl_bw_of(int i
);
224 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
226 dl_b
->total_bw
-= tsk_bw
;
230 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
232 dl_b
->total_bw
+= tsk_bw
;
236 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
238 return dl_b
->bw
!= -1 &&
239 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
242 extern void init_dl_bw(struct dl_bw
*dl_b
);
244 #ifdef CONFIG_CGROUP_SCHED
246 #include <linux/cgroup.h>
251 extern struct list_head task_groups
;
253 struct cfs_bandwidth
{
254 #ifdef CONFIG_CFS_BANDWIDTH
258 s64 hierarchical_quota
;
261 int idle
, period_active
;
262 struct hrtimer period_timer
, slack_timer
;
263 struct list_head throttled_cfs_rq
;
266 int nr_periods
, nr_throttled
;
271 /* task group related information */
273 struct cgroup_subsys_state css
;
275 #ifdef CONFIG_FAIR_GROUP_SCHED
276 /* schedulable entities of this group on each cpu */
277 struct sched_entity
**se
;
278 /* runqueue "owned" by this group on each cpu */
279 struct cfs_rq
**cfs_rq
;
280 unsigned long shares
;
284 * load_avg can be heavily contended at clock tick time, so put
285 * it in its own cacheline separated from the fields above which
286 * will also be accessed at each tick.
288 atomic_long_t load_avg ____cacheline_aligned
;
292 #ifdef CONFIG_RT_GROUP_SCHED
293 struct sched_rt_entity
**rt_se
;
294 struct rt_rq
**rt_rq
;
296 struct rt_bandwidth rt_bandwidth
;
300 struct list_head list
;
302 struct task_group
*parent
;
303 struct list_head siblings
;
304 struct list_head children
;
306 #ifdef CONFIG_SCHED_AUTOGROUP
307 struct autogroup
*autogroup
;
310 struct cfs_bandwidth cfs_bandwidth
;
313 #ifdef CONFIG_FAIR_GROUP_SCHED
314 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
317 * A weight of 0 or 1 can cause arithmetics problems.
318 * A weight of a cfs_rq is the sum of weights of which entities
319 * are queued on this cfs_rq, so a weight of a entity should not be
320 * too large, so as the shares value of a task group.
321 * (The default weight is 1024 - so there's no practical
322 * limitation from this.)
324 #define MIN_SHARES (1UL << 1)
325 #define MAX_SHARES (1UL << 18)
328 typedef int (*tg_visitor
)(struct task_group
*, void *);
330 extern int walk_tg_tree_from(struct task_group
*from
,
331 tg_visitor down
, tg_visitor up
, void *data
);
334 * Iterate the full tree, calling @down when first entering a node and @up when
335 * leaving it for the final time.
337 * Caller must hold rcu_lock or sufficient equivalent.
339 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
341 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
344 extern int tg_nop(struct task_group
*tg
, void *data
);
346 extern void free_fair_sched_group(struct task_group
*tg
);
347 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
348 extern void online_fair_sched_group(struct task_group
*tg
);
349 extern void unregister_fair_sched_group(struct task_group
*tg
);
350 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
351 struct sched_entity
*se
, int cpu
,
352 struct sched_entity
*parent
);
353 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
355 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
356 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
357 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
359 extern void free_rt_sched_group(struct task_group
*tg
);
360 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
361 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
362 struct sched_rt_entity
*rt_se
, int cpu
,
363 struct sched_rt_entity
*parent
);
365 extern struct task_group
*sched_create_group(struct task_group
*parent
);
366 extern void sched_online_group(struct task_group
*tg
,
367 struct task_group
*parent
);
368 extern void sched_destroy_group(struct task_group
*tg
);
369 extern void sched_offline_group(struct task_group
*tg
);
371 extern void sched_move_task(struct task_struct
*tsk
);
373 #ifdef CONFIG_FAIR_GROUP_SCHED
374 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
377 extern void set_task_rq_fair(struct sched_entity
*se
,
378 struct cfs_rq
*prev
, struct cfs_rq
*next
);
379 #else /* !CONFIG_SMP */
380 static inline void set_task_rq_fair(struct sched_entity
*se
,
381 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
382 #endif /* CONFIG_SMP */
383 #endif /* CONFIG_FAIR_GROUP_SCHED */
385 #else /* CONFIG_CGROUP_SCHED */
387 struct cfs_bandwidth
{ };
389 #endif /* CONFIG_CGROUP_SCHED */
391 /* CFS-related fields in a runqueue */
393 struct load_weight load
;
394 unsigned int nr_running
, h_nr_running
;
399 u64 min_vruntime_copy
;
402 struct rb_root tasks_timeline
;
403 struct rb_node
*rb_leftmost
;
406 * 'curr' points to currently running entity on this cfs_rq.
407 * It is set to NULL otherwise (i.e when none are currently running).
409 struct sched_entity
*curr
, *next
, *last
, *skip
;
411 #ifdef CONFIG_SCHED_DEBUG
412 unsigned int nr_spread_over
;
419 struct sched_avg avg
;
420 u64 runnable_load_sum
;
421 unsigned long runnable_load_avg
;
422 #ifdef CONFIG_FAIR_GROUP_SCHED
423 unsigned long tg_load_avg_contrib
;
424 unsigned long propagate_avg
;
426 atomic_long_t removed_load_avg
, removed_util_avg
;
428 u64 load_last_update_time_copy
;
431 #ifdef CONFIG_FAIR_GROUP_SCHED
433 * h_load = weight * f(tg)
435 * Where f(tg) is the recursive weight fraction assigned to
438 unsigned long h_load
;
439 u64 last_h_load_update
;
440 struct sched_entity
*h_load_next
;
441 #endif /* CONFIG_FAIR_GROUP_SCHED */
442 #endif /* CONFIG_SMP */
444 #ifdef CONFIG_FAIR_GROUP_SCHED
445 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
448 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
449 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
450 * (like users, containers etc.)
452 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
453 * list is used during load balance.
456 struct list_head leaf_cfs_rq_list
;
457 struct task_group
*tg
; /* group that "owns" this runqueue */
459 #ifdef CONFIG_CFS_BANDWIDTH
462 s64 runtime_remaining
;
464 u64 throttled_clock
, throttled_clock_task
;
465 u64 throttled_clock_task_time
;
466 int throttled
, throttle_count
;
467 struct list_head throttled_list
;
468 #endif /* CONFIG_CFS_BANDWIDTH */
469 #endif /* CONFIG_FAIR_GROUP_SCHED */
472 static inline int rt_bandwidth_enabled(void)
474 return sysctl_sched_rt_runtime
>= 0;
477 /* RT IPI pull logic requires IRQ_WORK */
478 #ifdef CONFIG_IRQ_WORK
479 # define HAVE_RT_PUSH_IPI
482 /* Real-Time classes' related field in a runqueue: */
484 struct rt_prio_array active
;
485 unsigned int rt_nr_running
;
486 unsigned int rr_nr_running
;
487 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
489 int curr
; /* highest queued rt task prio */
491 int next
; /* next highest */
496 unsigned long rt_nr_migratory
;
497 unsigned long rt_nr_total
;
499 struct plist_head pushable_tasks
;
500 #ifdef HAVE_RT_PUSH_IPI
503 struct irq_work push_work
;
504 raw_spinlock_t push_lock
;
506 #endif /* CONFIG_SMP */
512 /* Nests inside the rq lock: */
513 raw_spinlock_t rt_runtime_lock
;
515 #ifdef CONFIG_RT_GROUP_SCHED
516 unsigned long rt_nr_boosted
;
519 struct task_group
*tg
;
523 /* Deadline class' related fields in a runqueue */
525 /* runqueue is an rbtree, ordered by deadline */
526 struct rb_root rb_root
;
527 struct rb_node
*rb_leftmost
;
529 unsigned long dl_nr_running
;
533 * Deadline values of the currently executing and the
534 * earliest ready task on this rq. Caching these facilitates
535 * the decision wether or not a ready but not running task
536 * should migrate somewhere else.
543 unsigned long dl_nr_migratory
;
547 * Tasks on this rq that can be pushed away. They are kept in
548 * an rb-tree, ordered by tasks' deadlines, with caching
549 * of the leftmost (earliest deadline) element.
551 struct rb_root pushable_dl_tasks_root
;
552 struct rb_node
*pushable_dl_tasks_leftmost
;
560 static inline bool sched_asym_prefer(int a
, int b
)
562 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
566 * We add the notion of a root-domain which will be used to define per-domain
567 * variables. Each exclusive cpuset essentially defines an island domain by
568 * fully partitioning the member cpus from any other cpuset. Whenever a new
569 * exclusive cpuset is created, we also create and attach a new root-domain
578 cpumask_var_t online
;
580 /* Indicate more than one runnable task for any CPU */
584 * The bit corresponding to a CPU gets set here if such CPU has more
585 * than one runnable -deadline task (as it is below for RT tasks).
587 cpumask_var_t dlo_mask
;
593 * The "RT overload" flag: it gets set if a CPU has more than
594 * one runnable RT task.
596 cpumask_var_t rto_mask
;
597 struct cpupri cpupri
;
599 unsigned long max_cpu_capacity
;
602 extern struct root_domain def_root_domain
;
603 extern struct mutex sched_domains_mutex
;
604 extern cpumask_var_t fallback_doms
;
605 extern cpumask_var_t sched_domains_tmpmask
;
607 extern void init_defrootdomain(void);
608 extern int init_sched_domains(const struct cpumask
*cpu_map
);
609 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
611 #endif /* CONFIG_SMP */
614 * This is the main, per-CPU runqueue data structure.
616 * Locking rule: those places that want to lock multiple runqueues
617 * (such as the load balancing or the thread migration code), lock
618 * acquire operations must be ordered by ascending &runqueue.
625 * nr_running and cpu_load should be in the same cacheline because
626 * remote CPUs use both these fields when doing load calculation.
628 unsigned int nr_running
;
629 #ifdef CONFIG_NUMA_BALANCING
630 unsigned int nr_numa_running
;
631 unsigned int nr_preferred_running
;
633 #define CPU_LOAD_IDX_MAX 5
634 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
635 #ifdef CONFIG_NO_HZ_COMMON
637 unsigned long last_load_update_tick
;
638 #endif /* CONFIG_SMP */
639 unsigned long nohz_flags
;
640 #endif /* CONFIG_NO_HZ_COMMON */
641 #ifdef CONFIG_NO_HZ_FULL
642 unsigned long last_sched_tick
;
644 /* capture load from *all* tasks on this cpu: */
645 struct load_weight load
;
646 unsigned long nr_load_updates
;
653 #ifdef CONFIG_FAIR_GROUP_SCHED
654 /* list of leaf cfs_rq on this cpu: */
655 struct list_head leaf_cfs_rq_list
;
656 struct list_head
*tmp_alone_branch
;
657 #endif /* CONFIG_FAIR_GROUP_SCHED */
660 * This is part of a global counter where only the total sum
661 * over all CPUs matters. A task can increase this counter on
662 * one CPU and if it got migrated afterwards it may decrease
663 * it on another CPU. Always updated under the runqueue lock:
665 unsigned long nr_uninterruptible
;
667 struct task_struct
*curr
, *idle
, *stop
;
668 unsigned long next_balance
;
669 struct mm_struct
*prev_mm
;
671 unsigned int clock_update_flags
;
678 struct root_domain
*rd
;
679 struct sched_domain
*sd
;
681 unsigned long cpu_capacity
;
682 unsigned long cpu_capacity_orig
;
684 struct callback_head
*balance_callback
;
686 unsigned char idle_balance
;
687 /* For active balancing */
690 struct cpu_stop_work active_balance_work
;
691 /* cpu of this runqueue: */
695 struct list_head cfs_tasks
;
702 /* This is used to determine avg_idle's max value */
703 u64 max_idle_balance_cost
;
706 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
709 #ifdef CONFIG_PARAVIRT
712 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
713 u64 prev_steal_time_rq
;
716 /* calc_load related fields */
717 unsigned long calc_load_update
;
718 long calc_load_active
;
720 #ifdef CONFIG_SCHED_HRTICK
722 int hrtick_csd_pending
;
723 struct call_single_data hrtick_csd
;
725 struct hrtimer hrtick_timer
;
728 #ifdef CONFIG_SCHEDSTATS
730 struct sched_info rq_sched_info
;
731 unsigned long long rq_cpu_time
;
732 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
734 /* sys_sched_yield() stats */
735 unsigned int yld_count
;
737 /* schedule() stats */
738 unsigned int sched_count
;
739 unsigned int sched_goidle
;
741 /* try_to_wake_up() stats */
742 unsigned int ttwu_count
;
743 unsigned int ttwu_local
;
747 struct llist_head wake_list
;
750 #ifdef CONFIG_CPU_IDLE
751 /* Must be inspected within a rcu lock section */
752 struct cpuidle_state
*idle_state
;
756 static inline int cpu_of(struct rq
*rq
)
766 #ifdef CONFIG_SCHED_SMT
768 extern struct static_key_false sched_smt_present
;
770 extern void __update_idle_core(struct rq
*rq
);
772 static inline void update_idle_core(struct rq
*rq
)
774 if (static_branch_unlikely(&sched_smt_present
))
775 __update_idle_core(rq
);
779 static inline void update_idle_core(struct rq
*rq
) { }
782 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
784 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
785 #define this_rq() this_cpu_ptr(&runqueues)
786 #define task_rq(p) cpu_rq(task_cpu(p))
787 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
788 #define raw_rq() raw_cpu_ptr(&runqueues)
790 static inline u64
__rq_clock_broken(struct rq
*rq
)
792 return READ_ONCE(rq
->clock
);
796 * rq::clock_update_flags bits
798 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
799 * call to __schedule(). This is an optimisation to avoid
800 * neighbouring rq clock updates.
802 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
803 * in effect and calls to update_rq_clock() are being ignored.
805 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
806 * made to update_rq_clock() since the last time rq::lock was pinned.
808 * If inside of __schedule(), clock_update_flags will have been
809 * shifted left (a left shift is a cheap operation for the fast path
810 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
812 * if (rq-clock_update_flags >= RQCF_UPDATED)
814 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
815 * one position though, because the next rq_unpin_lock() will shift it
818 #define RQCF_REQ_SKIP 0x01
819 #define RQCF_ACT_SKIP 0x02
820 #define RQCF_UPDATED 0x04
822 static inline void assert_clock_updated(struct rq
*rq
)
825 * The only reason for not seeing a clock update since the
826 * last rq_pin_lock() is if we're currently skipping updates.
828 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
831 static inline u64
rq_clock(struct rq
*rq
)
833 lockdep_assert_held(&rq
->lock
);
834 assert_clock_updated(rq
);
839 static inline u64
rq_clock_task(struct rq
*rq
)
841 lockdep_assert_held(&rq
->lock
);
842 assert_clock_updated(rq
);
844 return rq
->clock_task
;
847 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
849 lockdep_assert_held(&rq
->lock
);
851 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
853 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
858 struct pin_cookie cookie
;
859 #ifdef CONFIG_SCHED_DEBUG
861 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
862 * current pin context is stashed here in case it needs to be
863 * restored in rq_repin_lock().
865 unsigned int clock_update_flags
;
869 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
871 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
873 #ifdef CONFIG_SCHED_DEBUG
874 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
875 rf
->clock_update_flags
= 0;
879 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
881 #ifdef CONFIG_SCHED_DEBUG
882 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
883 rf
->clock_update_flags
= RQCF_UPDATED
;
886 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
889 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
891 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
893 #ifdef CONFIG_SCHED_DEBUG
895 * Restore the value we stashed in @rf for this pin context.
897 rq
->clock_update_flags
|= rf
->clock_update_flags
;
902 enum numa_topology_type
{
907 extern enum numa_topology_type sched_numa_topology_type
;
908 extern int sched_max_numa_distance
;
909 extern bool find_numa_distance(int distance
);
913 extern void sched_init_numa(void);
914 extern void sched_domains_numa_masks_set(unsigned int cpu
);
915 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
917 static inline void sched_init_numa(void) { }
918 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
919 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
922 #ifdef CONFIG_NUMA_BALANCING
923 /* The regions in numa_faults array from task_struct */
924 enum numa_faults_stats
{
930 extern void sched_setnuma(struct task_struct
*p
, int node
);
931 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
932 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
933 #endif /* CONFIG_NUMA_BALANCING */
938 queue_balance_callback(struct rq
*rq
,
939 struct callback_head
*head
,
940 void (*func
)(struct rq
*rq
))
942 lockdep_assert_held(&rq
->lock
);
944 if (unlikely(head
->next
))
947 head
->func
= (void (*)(struct callback_head
*))func
;
948 head
->next
= rq
->balance_callback
;
949 rq
->balance_callback
= head
;
952 extern void sched_ttwu_pending(void);
954 #define rcu_dereference_check_sched_domain(p) \
955 rcu_dereference_check((p), \
956 lockdep_is_held(&sched_domains_mutex))
959 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
960 * See detach_destroy_domains: synchronize_sched for details.
962 * The domain tree of any CPU may only be accessed from within
963 * preempt-disabled sections.
965 #define for_each_domain(cpu, __sd) \
966 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
967 __sd; __sd = __sd->parent)
969 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
972 * highest_flag_domain - Return highest sched_domain containing flag.
973 * @cpu: The cpu whose highest level of sched domain is to
975 * @flag: The flag to check for the highest sched_domain
978 * Returns the highest sched_domain of a cpu which contains the given flag.
980 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
982 struct sched_domain
*sd
, *hsd
= NULL
;
984 for_each_domain(cpu
, sd
) {
985 if (!(sd
->flags
& flag
))
993 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
995 struct sched_domain
*sd
;
997 for_each_domain(cpu
, sd
) {
998 if (sd
->flags
& flag
)
1005 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
1006 DECLARE_PER_CPU(int, sd_llc_size
);
1007 DECLARE_PER_CPU(int, sd_llc_id
);
1008 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
1009 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
1010 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
1012 struct sched_group_capacity
{
1015 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1018 unsigned long capacity
;
1019 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1020 unsigned long next_update
;
1021 int imbalance
; /* XXX unrelated to capacity but shared group state */
1023 unsigned long cpumask
[0]; /* iteration mask */
1026 struct sched_group
{
1027 struct sched_group
*next
; /* Must be a circular list */
1030 unsigned int group_weight
;
1031 struct sched_group_capacity
*sgc
;
1032 int asym_prefer_cpu
; /* cpu of highest priority in group */
1035 * The CPUs this group covers.
1037 * NOTE: this field is variable length. (Allocated dynamically
1038 * by attaching extra space to the end of the structure,
1039 * depending on how many CPUs the kernel has booted up with)
1041 unsigned long cpumask
[0];
1044 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
1046 return to_cpumask(sg
->cpumask
);
1050 * cpumask masking which cpus in the group are allowed to iterate up the domain
1053 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
1055 return to_cpumask(sg
->sgc
->cpumask
);
1059 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1060 * @group: The group whose first cpu is to be returned.
1062 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1064 return cpumask_first(sched_group_cpus(group
));
1067 extern int group_balance_cpu(struct sched_group
*sg
);
1069 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1070 void register_sched_domain_sysctl(void);
1071 void unregister_sched_domain_sysctl(void);
1073 static inline void register_sched_domain_sysctl(void)
1076 static inline void unregister_sched_domain_sysctl(void)
1083 static inline void sched_ttwu_pending(void) { }
1085 #endif /* CONFIG_SMP */
1088 #include "autogroup.h"
1090 #ifdef CONFIG_CGROUP_SCHED
1093 * Return the group to which this tasks belongs.
1095 * We cannot use task_css() and friends because the cgroup subsystem
1096 * changes that value before the cgroup_subsys::attach() method is called,
1097 * therefore we cannot pin it and might observe the wrong value.
1099 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1100 * core changes this before calling sched_move_task().
1102 * Instead we use a 'copy' which is updated from sched_move_task() while
1103 * holding both task_struct::pi_lock and rq::lock.
1105 static inline struct task_group
*task_group(struct task_struct
*p
)
1107 return p
->sched_task_group
;
1110 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1111 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1113 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1114 struct task_group
*tg
= task_group(p
);
1117 #ifdef CONFIG_FAIR_GROUP_SCHED
1118 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1119 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1120 p
->se
.parent
= tg
->se
[cpu
];
1123 #ifdef CONFIG_RT_GROUP_SCHED
1124 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1125 p
->rt
.parent
= tg
->rt_se
[cpu
];
1129 #else /* CONFIG_CGROUP_SCHED */
1131 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1132 static inline struct task_group
*task_group(struct task_struct
*p
)
1137 #endif /* CONFIG_CGROUP_SCHED */
1139 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1141 set_task_rq(p
, cpu
);
1144 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1145 * successfuly executed on another CPU. We must ensure that updates of
1146 * per-task data have been completed by this moment.
1149 #ifdef CONFIG_THREAD_INFO_IN_TASK
1152 task_thread_info(p
)->cpu
= cpu
;
1159 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1161 #ifdef CONFIG_SCHED_DEBUG
1162 # include <linux/static_key.h>
1163 # define const_debug __read_mostly
1165 # define const_debug const
1168 extern const_debug
unsigned int sysctl_sched_features
;
1170 #define SCHED_FEAT(name, enabled) \
1171 __SCHED_FEAT_##name ,
1174 #include "features.h"
1180 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1181 #define SCHED_FEAT(name, enabled) \
1182 static __always_inline bool static_branch_##name(struct static_key *key) \
1184 return static_key_##enabled(key); \
1187 #include "features.h"
1191 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1192 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1193 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1194 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1195 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1197 extern struct static_key_false sched_numa_balancing
;
1198 extern struct static_key_false sched_schedstats
;
1200 static inline u64
global_rt_period(void)
1202 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1205 static inline u64
global_rt_runtime(void)
1207 if (sysctl_sched_rt_runtime
< 0)
1210 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1213 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1215 return rq
->curr
== p
;
1218 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1223 return task_current(rq
, p
);
1227 static inline int task_on_rq_queued(struct task_struct
*p
)
1229 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1232 static inline int task_on_rq_migrating(struct task_struct
*p
)
1234 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1237 #ifndef prepare_arch_switch
1238 # define prepare_arch_switch(next) do { } while (0)
1240 #ifndef finish_arch_post_lock_switch
1241 # define finish_arch_post_lock_switch() do { } while (0)
1244 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1248 * We can optimise this out completely for !SMP, because the
1249 * SMP rebalancing from interrupt is the only thing that cares
1256 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1260 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1261 * We must ensure this doesn't happen until the switch is completely
1264 * In particular, the load of prev->state in finish_task_switch() must
1265 * happen before this.
1267 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1269 smp_store_release(&prev
->on_cpu
, 0);
1271 #ifdef CONFIG_DEBUG_SPINLOCK
1272 /* this is a valid case when another task releases the spinlock */
1273 rq
->lock
.owner
= current
;
1276 * If we are tracking spinlock dependencies then we have to
1277 * fix up the runqueue lock - which gets 'carried over' from
1278 * prev into current:
1280 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1282 raw_spin_unlock_irq(&rq
->lock
);
1288 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1289 #define WF_FORK 0x02 /* child wakeup after fork */
1290 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1293 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1294 * of tasks with abnormal "nice" values across CPUs the contribution that
1295 * each task makes to its run queue's load is weighted according to its
1296 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1297 * scaled version of the new time slice allocation that they receive on time
1301 #define WEIGHT_IDLEPRIO 3
1302 #define WMULT_IDLEPRIO 1431655765
1304 extern const int sched_prio_to_weight
[40];
1305 extern const u32 sched_prio_to_wmult
[40];
1308 * {de,en}queue flags:
1310 * DEQUEUE_SLEEP - task is no longer runnable
1311 * ENQUEUE_WAKEUP - task just became runnable
1313 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1314 * are in a known state which allows modification. Such pairs
1315 * should preserve as much state as possible.
1317 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1320 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1321 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1322 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1326 #define DEQUEUE_SLEEP 0x01
1327 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1328 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1330 #define ENQUEUE_WAKEUP 0x01
1331 #define ENQUEUE_RESTORE 0x02
1332 #define ENQUEUE_MOVE 0x04
1334 #define ENQUEUE_HEAD 0x08
1335 #define ENQUEUE_REPLENISH 0x10
1337 #define ENQUEUE_MIGRATED 0x20
1339 #define ENQUEUE_MIGRATED 0x00
1342 #define RETRY_TASK ((void *)-1UL)
1344 struct sched_class
{
1345 const struct sched_class
*next
;
1347 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1348 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1349 void (*yield_task
) (struct rq
*rq
);
1350 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1352 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1355 * It is the responsibility of the pick_next_task() method that will
1356 * return the next task to call put_prev_task() on the @prev task or
1357 * something equivalent.
1359 * May return RETRY_TASK when it finds a higher prio class has runnable
1362 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1363 struct task_struct
*prev
,
1364 struct rq_flags
*rf
);
1365 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1368 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1369 void (*migrate_task_rq
)(struct task_struct
*p
);
1371 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1373 void (*set_cpus_allowed
)(struct task_struct
*p
,
1374 const struct cpumask
*newmask
);
1376 void (*rq_online
)(struct rq
*rq
);
1377 void (*rq_offline
)(struct rq
*rq
);
1380 void (*set_curr_task
) (struct rq
*rq
);
1381 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1382 void (*task_fork
) (struct task_struct
*p
);
1383 void (*task_dead
) (struct task_struct
*p
);
1386 * The switched_from() call is allowed to drop rq->lock, therefore we
1387 * cannot assume the switched_from/switched_to pair is serliazed by
1388 * rq->lock. They are however serialized by p->pi_lock.
1390 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1391 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1392 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1395 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1396 struct task_struct
*task
);
1398 void (*update_curr
) (struct rq
*rq
);
1400 #define TASK_SET_GROUP 0
1401 #define TASK_MOVE_GROUP 1
1403 #ifdef CONFIG_FAIR_GROUP_SCHED
1404 void (*task_change_group
) (struct task_struct
*p
, int type
);
1408 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1410 prev
->sched_class
->put_prev_task(rq
, prev
);
1413 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1415 curr
->sched_class
->set_curr_task(rq
);
1418 #define sched_class_highest (&stop_sched_class)
1419 #define for_each_class(class) \
1420 for (class = sched_class_highest; class; class = class->next)
1422 extern const struct sched_class stop_sched_class
;
1423 extern const struct sched_class dl_sched_class
;
1424 extern const struct sched_class rt_sched_class
;
1425 extern const struct sched_class fair_sched_class
;
1426 extern const struct sched_class idle_sched_class
;
1431 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1433 extern void trigger_load_balance(struct rq
*rq
);
1435 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1439 #ifdef CONFIG_CPU_IDLE
1440 static inline void idle_set_state(struct rq
*rq
,
1441 struct cpuidle_state
*idle_state
)
1443 rq
->idle_state
= idle_state
;
1446 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1448 SCHED_WARN_ON(!rcu_read_lock_held());
1449 return rq
->idle_state
;
1452 static inline void idle_set_state(struct rq
*rq
,
1453 struct cpuidle_state
*idle_state
)
1457 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1463 extern void sysrq_sched_debug_show(void);
1464 extern void sched_init_granularity(void);
1465 extern void update_max_interval(void);
1467 extern void init_sched_dl_class(void);
1468 extern void init_sched_rt_class(void);
1469 extern void init_sched_fair_class(void);
1471 extern void resched_curr(struct rq
*rq
);
1472 extern void resched_cpu(int cpu
);
1474 extern struct rt_bandwidth def_rt_bandwidth
;
1475 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1477 extern struct dl_bandwidth def_dl_bandwidth
;
1478 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1479 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1481 unsigned long to_ratio(u64 period
, u64 runtime
);
1483 extern void init_entity_runnable_average(struct sched_entity
*se
);
1484 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1486 #ifdef CONFIG_NO_HZ_FULL
1487 extern bool sched_can_stop_tick(struct rq
*rq
);
1490 * Tick may be needed by tasks in the runqueue depending on their policy and
1491 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1492 * nohz mode if necessary.
1494 static inline void sched_update_tick_dependency(struct rq
*rq
)
1498 if (!tick_nohz_full_enabled())
1503 if (!tick_nohz_full_cpu(cpu
))
1506 if (sched_can_stop_tick(rq
))
1507 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1509 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1512 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1515 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1517 unsigned prev_nr
= rq
->nr_running
;
1519 rq
->nr_running
= prev_nr
+ count
;
1521 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1523 if (!rq
->rd
->overload
)
1524 rq
->rd
->overload
= true;
1528 sched_update_tick_dependency(rq
);
1531 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1533 rq
->nr_running
-= count
;
1534 /* Check if we still need preemption */
1535 sched_update_tick_dependency(rq
);
1538 static inline void rq_last_tick_reset(struct rq
*rq
)
1540 #ifdef CONFIG_NO_HZ_FULL
1541 rq
->last_sched_tick
= jiffies
;
1545 extern void update_rq_clock(struct rq
*rq
);
1547 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1548 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1550 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1552 extern const_debug
unsigned int sysctl_sched_time_avg
;
1553 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1554 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1556 static inline u64
sched_avg_period(void)
1558 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1561 #ifdef CONFIG_SCHED_HRTICK
1565 * - enabled by features
1566 * - hrtimer is actually high res
1568 static inline int hrtick_enabled(struct rq
*rq
)
1570 if (!sched_feat(HRTICK
))
1572 if (!cpu_active(cpu_of(rq
)))
1574 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1577 void hrtick_start(struct rq
*rq
, u64 delay
);
1581 static inline int hrtick_enabled(struct rq
*rq
)
1586 #endif /* CONFIG_SCHED_HRTICK */
1589 extern void sched_avg_update(struct rq
*rq
);
1591 #ifndef arch_scale_freq_capacity
1592 static __always_inline
1593 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1595 return SCHED_CAPACITY_SCALE
;
1599 #ifndef arch_scale_cpu_capacity
1600 static __always_inline
1601 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1603 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1604 return sd
->smt_gain
/ sd
->span_weight
;
1606 return SCHED_CAPACITY_SCALE
;
1610 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1612 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1613 sched_avg_update(rq
);
1616 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1617 static inline void sched_avg_update(struct rq
*rq
) { }
1620 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1621 __acquires(rq
->lock
);
1622 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1623 __acquires(p
->pi_lock
)
1624 __acquires(rq
->lock
);
1626 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1627 __releases(rq
->lock
)
1629 rq_unpin_lock(rq
, rf
);
1630 raw_spin_unlock(&rq
->lock
);
1634 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1635 __releases(rq
->lock
)
1636 __releases(p
->pi_lock
)
1638 rq_unpin_lock(rq
, rf
);
1639 raw_spin_unlock(&rq
->lock
);
1640 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1644 #ifdef CONFIG_PREEMPT
1646 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1649 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1650 * way at the expense of forcing extra atomic operations in all
1651 * invocations. This assures that the double_lock is acquired using the
1652 * same underlying policy as the spinlock_t on this architecture, which
1653 * reduces latency compared to the unfair variant below. However, it
1654 * also adds more overhead and therefore may reduce throughput.
1656 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1657 __releases(this_rq
->lock
)
1658 __acquires(busiest
->lock
)
1659 __acquires(this_rq
->lock
)
1661 raw_spin_unlock(&this_rq
->lock
);
1662 double_rq_lock(this_rq
, busiest
);
1669 * Unfair double_lock_balance: Optimizes throughput at the expense of
1670 * latency by eliminating extra atomic operations when the locks are
1671 * already in proper order on entry. This favors lower cpu-ids and will
1672 * grant the double lock to lower cpus over higher ids under contention,
1673 * regardless of entry order into the function.
1675 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1676 __releases(this_rq
->lock
)
1677 __acquires(busiest
->lock
)
1678 __acquires(this_rq
->lock
)
1682 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1683 if (busiest
< this_rq
) {
1684 raw_spin_unlock(&this_rq
->lock
);
1685 raw_spin_lock(&busiest
->lock
);
1686 raw_spin_lock_nested(&this_rq
->lock
,
1687 SINGLE_DEPTH_NESTING
);
1690 raw_spin_lock_nested(&busiest
->lock
,
1691 SINGLE_DEPTH_NESTING
);
1696 #endif /* CONFIG_PREEMPT */
1699 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1701 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1703 if (unlikely(!irqs_disabled())) {
1704 /* printk() doesn't work good under rq->lock */
1705 raw_spin_unlock(&this_rq
->lock
);
1709 return _double_lock_balance(this_rq
, busiest
);
1712 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1713 __releases(busiest
->lock
)
1715 raw_spin_unlock(&busiest
->lock
);
1716 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1719 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1725 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1728 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1734 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1737 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1743 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1747 * double_rq_lock - safely lock two runqueues
1749 * Note this does not disable interrupts like task_rq_lock,
1750 * you need to do so manually before calling.
1752 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1753 __acquires(rq1
->lock
)
1754 __acquires(rq2
->lock
)
1756 BUG_ON(!irqs_disabled());
1758 raw_spin_lock(&rq1
->lock
);
1759 __acquire(rq2
->lock
); /* Fake it out ;) */
1762 raw_spin_lock(&rq1
->lock
);
1763 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1765 raw_spin_lock(&rq2
->lock
);
1766 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1772 * double_rq_unlock - safely unlock two runqueues
1774 * Note this does not restore interrupts like task_rq_unlock,
1775 * you need to do so manually after calling.
1777 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1778 __releases(rq1
->lock
)
1779 __releases(rq2
->lock
)
1781 raw_spin_unlock(&rq1
->lock
);
1783 raw_spin_unlock(&rq2
->lock
);
1785 __release(rq2
->lock
);
1788 extern void set_rq_online (struct rq
*rq
);
1789 extern void set_rq_offline(struct rq
*rq
);
1790 extern bool sched_smp_initialized
;
1792 #else /* CONFIG_SMP */
1795 * double_rq_lock - safely lock two runqueues
1797 * Note this does not disable interrupts like task_rq_lock,
1798 * you need to do so manually before calling.
1800 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1801 __acquires(rq1
->lock
)
1802 __acquires(rq2
->lock
)
1804 BUG_ON(!irqs_disabled());
1806 raw_spin_lock(&rq1
->lock
);
1807 __acquire(rq2
->lock
); /* Fake it out ;) */
1811 * double_rq_unlock - safely unlock two runqueues
1813 * Note this does not restore interrupts like task_rq_unlock,
1814 * you need to do so manually after calling.
1816 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1817 __releases(rq1
->lock
)
1818 __releases(rq2
->lock
)
1821 raw_spin_unlock(&rq1
->lock
);
1822 __release(rq2
->lock
);
1827 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1828 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1830 #ifdef CONFIG_SCHED_DEBUG
1831 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1832 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1833 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1835 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1836 #ifdef CONFIG_NUMA_BALANCING
1838 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1840 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1841 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1842 #endif /* CONFIG_NUMA_BALANCING */
1843 #endif /* CONFIG_SCHED_DEBUG */
1845 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1846 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1847 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1849 extern void cfs_bandwidth_usage_inc(void);
1850 extern void cfs_bandwidth_usage_dec(void);
1852 #ifdef CONFIG_NO_HZ_COMMON
1853 enum rq_nohz_flag_bits
{
1858 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1860 extern void nohz_balance_exit_idle(unsigned int cpu
);
1862 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1865 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1869 struct u64_stats_sync sync
;
1872 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
1874 static inline u64
irq_time_read(int cpu
)
1876 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
1877 u64
*cpustat
= kcpustat_cpu(cpu
).cpustat
;
1882 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
1883 total
= cpustat
[CPUTIME_SOFTIRQ
] + cpustat
[CPUTIME_IRQ
];
1884 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
1888 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1890 #ifdef CONFIG_CPU_FREQ
1891 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
1894 * cpufreq_update_util - Take a note about CPU utilization changes.
1895 * @rq: Runqueue to carry out the update for.
1896 * @flags: Update reason flags.
1898 * This function is called by the scheduler on the CPU whose utilization is
1901 * It can only be called from RCU-sched read-side critical sections.
1903 * The way cpufreq is currently arranged requires it to evaluate the CPU
1904 * performance state (frequency/voltage) on a regular basis to prevent it from
1905 * being stuck in a completely inadequate performance level for too long.
1906 * That is not guaranteed to happen if the updates are only triggered from CFS,
1907 * though, because they may not be coming in if RT or deadline tasks are active
1908 * all the time (or there are RT and DL tasks only).
1910 * As a workaround for that issue, this function is called by the RT and DL
1911 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1912 * but that really is a band-aid. Going forward it should be replaced with
1913 * solutions targeted more specifically at RT and DL tasks.
1915 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
1917 struct update_util_data
*data
;
1919 data
= rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data
));
1921 data
->func(data
, rq_clock(rq
), flags
);
1924 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
)
1926 if (cpu_of(rq
) == smp_processor_id())
1927 cpufreq_update_util(rq
, flags
);
1930 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
1931 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
) {}
1932 #endif /* CONFIG_CPU_FREQ */
1934 #ifdef arch_scale_freq_capacity
1935 #ifndef arch_scale_freq_invariant
1936 #define arch_scale_freq_invariant() (true)
1938 #else /* arch_scale_freq_capacity */
1939 #define arch_scale_freq_invariant() (false)