2 #include <linux/sched.h>
3 #include <linux/sched/autogroup.h>
4 #include <linux/sched/sysctl.h>
5 #include <linux/sched/topology.h>
6 #include <linux/sched/rt.h>
7 #include <linux/sched/deadline.h>
8 #include <linux/sched/clock.h>
9 #include <linux/sched/wake_q.h>
10 #include <linux/sched/signal.h>
11 #include <linux/sched/numa_balancing.h>
12 #include <linux/sched/mm.h>
13 #include <linux/sched/cpufreq.h>
14 #include <linux/sched/stat.h>
15 #include <linux/sched/nohz.h>
16 #include <linux/sched/debug.h>
17 #include <linux/sched/hotplug.h>
18 #include <linux/sched/task.h>
19 #include <linux/sched/task_stack.h>
20 #include <linux/sched/cputime.h>
21 #include <linux/sched/init.h>
23 #include <linux/u64_stats_sync.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/binfmts.h>
26 #include <linux/mutex.h>
27 #include <linux/spinlock.h>
28 #include <linux/stop_machine.h>
29 #include <linux/irq_work.h>
30 #include <linux/tick.h>
31 #include <linux/slab.h>
33 #ifdef CONFIG_PARAVIRT
34 #include <asm/paravirt.h>
38 #include "cpudeadline.h"
41 #ifdef CONFIG_SCHED_DEBUG
42 #define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
44 #define SCHED_WARN_ON(x) ((void)(x))
50 /* task_struct::on_rq states: */
51 #define TASK_ON_RQ_QUEUED 1
52 #define TASK_ON_RQ_MIGRATING 2
54 extern __read_mostly
int scheduler_running
;
56 extern unsigned long calc_load_update
;
57 extern atomic_long_t calc_load_tasks
;
59 extern void calc_global_load_tick(struct rq
*this_rq
);
60 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
63 extern void cpu_load_update_active(struct rq
*this_rq
);
65 static inline void cpu_load_update_active(struct rq
*this_rq
) { }
69 * Helpers for converting nanosecond timing to jiffy resolution
71 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
74 * Increase resolution of nice-level calculations for 64-bit architectures.
75 * The extra resolution improves shares distribution and load balancing of
76 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
77 * hierarchies, especially on larger systems. This is not a user-visible change
78 * and does not change the user-interface for setting shares/weights.
80 * We increase resolution only if we have enough bits to allow this increased
81 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
82 * pretty high and the returns do not justify the increased costs.
84 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
85 * increase coverage and consistency always enable it on 64bit platforms.
88 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
89 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
90 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
92 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
93 # define scale_load(w) (w)
94 # define scale_load_down(w) (w)
98 * Task weight (visible to users) and its load (invisible to users) have
99 * independent resolution, but they should be well calibrated. We use
100 * scale_load() and scale_load_down(w) to convert between them. The
101 * following must be true:
103 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
106 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
109 * Single value that decides SCHED_DEADLINE internal math precision.
110 * 10 -> just above 1us
111 * 9 -> just above 0.5us
113 #define DL_SCALE (10)
116 * These are the 'tuning knobs' of the scheduler:
120 * single value that denotes runtime == period, ie unlimited time.
122 #define RUNTIME_INF ((u64)~0ULL)
124 static inline int idle_policy(int policy
)
126 return policy
== SCHED_IDLE
;
128 static inline int fair_policy(int policy
)
130 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
133 static inline int rt_policy(int policy
)
135 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
138 static inline int dl_policy(int policy
)
140 return policy
== SCHED_DEADLINE
;
142 static inline bool valid_policy(int policy
)
144 return idle_policy(policy
) || fair_policy(policy
) ||
145 rt_policy(policy
) || dl_policy(policy
);
148 static inline int task_has_rt_policy(struct task_struct
*p
)
150 return rt_policy(p
->policy
);
153 static inline int task_has_dl_policy(struct task_struct
*p
)
155 return dl_policy(p
->policy
);
159 * Tells if entity @a should preempt entity @b.
162 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
164 return dl_time_before(a
->deadline
, b
->deadline
);
168 * This is the priority-queue data structure of the RT scheduling class:
170 struct rt_prio_array
{
171 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
172 struct list_head queue
[MAX_RT_PRIO
];
175 struct rt_bandwidth
{
176 /* nests inside the rq lock: */
177 raw_spinlock_t rt_runtime_lock
;
180 struct hrtimer rt_period_timer
;
181 unsigned int rt_period_active
;
184 void __dl_clear_params(struct task_struct
*p
);
187 * To keep the bandwidth of -deadline tasks and groups under control
188 * we need some place where:
189 * - store the maximum -deadline bandwidth of the system (the group);
190 * - cache the fraction of that bandwidth that is currently allocated.
192 * This is all done in the data structure below. It is similar to the
193 * one used for RT-throttling (rt_bandwidth), with the main difference
194 * that, since here we are only interested in admission control, we
195 * do not decrease any runtime while the group "executes", neither we
196 * need a timer to replenish it.
198 * With respect to SMP, the bandwidth is given on a per-CPU basis,
200 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
201 * - dl_total_bw array contains, in the i-eth element, the currently
202 * allocated bandwidth on the i-eth CPU.
203 * Moreover, groups consume bandwidth on each CPU, while tasks only
204 * consume bandwidth on the CPU they're running on.
205 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
206 * that will be shown the next time the proc or cgroup controls will
207 * be red. It on its turn can be changed by writing on its own
210 struct dl_bandwidth
{
211 raw_spinlock_t dl_runtime_lock
;
216 static inline int dl_bandwidth_enabled(void)
218 return sysctl_sched_rt_runtime
>= 0;
221 extern struct dl_bw
*dl_bw_of(int i
);
229 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
231 dl_b
->total_bw
-= tsk_bw
;
235 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
237 dl_b
->total_bw
+= tsk_bw
;
241 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
243 return dl_b
->bw
!= -1 &&
244 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
247 extern void init_dl_bw(struct dl_bw
*dl_b
);
249 #ifdef CONFIG_CGROUP_SCHED
251 #include <linux/cgroup.h>
256 extern struct list_head task_groups
;
258 struct cfs_bandwidth
{
259 #ifdef CONFIG_CFS_BANDWIDTH
263 s64 hierarchical_quota
;
266 int idle
, period_active
;
267 struct hrtimer period_timer
, slack_timer
;
268 struct list_head throttled_cfs_rq
;
271 int nr_periods
, nr_throttled
;
276 /* task group related information */
278 struct cgroup_subsys_state css
;
280 #ifdef CONFIG_FAIR_GROUP_SCHED
281 /* schedulable entities of this group on each cpu */
282 struct sched_entity
**se
;
283 /* runqueue "owned" by this group on each cpu */
284 struct cfs_rq
**cfs_rq
;
285 unsigned long shares
;
289 * load_avg can be heavily contended at clock tick time, so put
290 * it in its own cacheline separated from the fields above which
291 * will also be accessed at each tick.
293 atomic_long_t load_avg ____cacheline_aligned
;
297 #ifdef CONFIG_RT_GROUP_SCHED
298 struct sched_rt_entity
**rt_se
;
299 struct rt_rq
**rt_rq
;
301 struct rt_bandwidth rt_bandwidth
;
305 struct list_head list
;
307 struct task_group
*parent
;
308 struct list_head siblings
;
309 struct list_head children
;
311 #ifdef CONFIG_SCHED_AUTOGROUP
312 struct autogroup
*autogroup
;
315 struct cfs_bandwidth cfs_bandwidth
;
318 #ifdef CONFIG_FAIR_GROUP_SCHED
319 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
322 * A weight of 0 or 1 can cause arithmetics problems.
323 * A weight of a cfs_rq is the sum of weights of which entities
324 * are queued on this cfs_rq, so a weight of a entity should not be
325 * too large, so as the shares value of a task group.
326 * (The default weight is 1024 - so there's no practical
327 * limitation from this.)
329 #define MIN_SHARES (1UL << 1)
330 #define MAX_SHARES (1UL << 18)
333 typedef int (*tg_visitor
)(struct task_group
*, void *);
335 extern int walk_tg_tree_from(struct task_group
*from
,
336 tg_visitor down
, tg_visitor up
, void *data
);
339 * Iterate the full tree, calling @down when first entering a node and @up when
340 * leaving it for the final time.
342 * Caller must hold rcu_lock or sufficient equivalent.
344 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
346 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
349 extern int tg_nop(struct task_group
*tg
, void *data
);
351 extern void free_fair_sched_group(struct task_group
*tg
);
352 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
353 extern void online_fair_sched_group(struct task_group
*tg
);
354 extern void unregister_fair_sched_group(struct task_group
*tg
);
355 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
356 struct sched_entity
*se
, int cpu
,
357 struct sched_entity
*parent
);
358 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
360 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
361 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
362 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
364 extern void free_rt_sched_group(struct task_group
*tg
);
365 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
366 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
367 struct sched_rt_entity
*rt_se
, int cpu
,
368 struct sched_rt_entity
*parent
);
370 extern struct task_group
*sched_create_group(struct task_group
*parent
);
371 extern void sched_online_group(struct task_group
*tg
,
372 struct task_group
*parent
);
373 extern void sched_destroy_group(struct task_group
*tg
);
374 extern void sched_offline_group(struct task_group
*tg
);
376 extern void sched_move_task(struct task_struct
*tsk
);
378 #ifdef CONFIG_FAIR_GROUP_SCHED
379 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
382 extern void set_task_rq_fair(struct sched_entity
*se
,
383 struct cfs_rq
*prev
, struct cfs_rq
*next
);
384 #else /* !CONFIG_SMP */
385 static inline void set_task_rq_fair(struct sched_entity
*se
,
386 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
387 #endif /* CONFIG_SMP */
388 #endif /* CONFIG_FAIR_GROUP_SCHED */
390 #else /* CONFIG_CGROUP_SCHED */
392 struct cfs_bandwidth
{ };
394 #endif /* CONFIG_CGROUP_SCHED */
396 /* CFS-related fields in a runqueue */
398 struct load_weight load
;
399 unsigned int nr_running
, h_nr_running
;
404 u64 min_vruntime_copy
;
407 struct rb_root tasks_timeline
;
408 struct rb_node
*rb_leftmost
;
411 * 'curr' points to currently running entity on this cfs_rq.
412 * It is set to NULL otherwise (i.e when none are currently running).
414 struct sched_entity
*curr
, *next
, *last
, *skip
;
416 #ifdef CONFIG_SCHED_DEBUG
417 unsigned int nr_spread_over
;
424 struct sched_avg avg
;
425 u64 runnable_load_sum
;
426 unsigned long runnable_load_avg
;
427 #ifdef CONFIG_FAIR_GROUP_SCHED
428 unsigned long tg_load_avg_contrib
;
429 unsigned long propagate_avg
;
431 atomic_long_t removed_load_avg
, removed_util_avg
;
433 u64 load_last_update_time_copy
;
436 #ifdef CONFIG_FAIR_GROUP_SCHED
438 * h_load = weight * f(tg)
440 * Where f(tg) is the recursive weight fraction assigned to
443 unsigned long h_load
;
444 u64 last_h_load_update
;
445 struct sched_entity
*h_load_next
;
446 #endif /* CONFIG_FAIR_GROUP_SCHED */
447 #endif /* CONFIG_SMP */
449 #ifdef CONFIG_FAIR_GROUP_SCHED
450 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
453 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
454 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
455 * (like users, containers etc.)
457 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
458 * list is used during load balance.
461 struct list_head leaf_cfs_rq_list
;
462 struct task_group
*tg
; /* group that "owns" this runqueue */
464 #ifdef CONFIG_CFS_BANDWIDTH
467 s64 runtime_remaining
;
469 u64 throttled_clock
, throttled_clock_task
;
470 u64 throttled_clock_task_time
;
471 int throttled
, throttle_count
;
472 struct list_head throttled_list
;
473 #endif /* CONFIG_CFS_BANDWIDTH */
474 #endif /* CONFIG_FAIR_GROUP_SCHED */
477 static inline int rt_bandwidth_enabled(void)
479 return sysctl_sched_rt_runtime
>= 0;
482 /* RT IPI pull logic requires IRQ_WORK */
483 #ifdef CONFIG_IRQ_WORK
484 # define HAVE_RT_PUSH_IPI
487 /* Real-Time classes' related field in a runqueue: */
489 struct rt_prio_array active
;
490 unsigned int rt_nr_running
;
491 unsigned int rr_nr_running
;
492 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
494 int curr
; /* highest queued rt task prio */
496 int next
; /* next highest */
501 unsigned long rt_nr_migratory
;
502 unsigned long rt_nr_total
;
504 struct plist_head pushable_tasks
;
505 #ifdef HAVE_RT_PUSH_IPI
508 struct irq_work push_work
;
509 raw_spinlock_t push_lock
;
511 #endif /* CONFIG_SMP */
517 /* Nests inside the rq lock: */
518 raw_spinlock_t rt_runtime_lock
;
520 #ifdef CONFIG_RT_GROUP_SCHED
521 unsigned long rt_nr_boosted
;
524 struct task_group
*tg
;
528 /* Deadline class' related fields in a runqueue */
530 /* runqueue is an rbtree, ordered by deadline */
531 struct rb_root rb_root
;
532 struct rb_node
*rb_leftmost
;
534 unsigned long dl_nr_running
;
538 * Deadline values of the currently executing and the
539 * earliest ready task on this rq. Caching these facilitates
540 * the decision wether or not a ready but not running task
541 * should migrate somewhere else.
548 unsigned long dl_nr_migratory
;
552 * Tasks on this rq that can be pushed away. They are kept in
553 * an rb-tree, ordered by tasks' deadlines, with caching
554 * of the leftmost (earliest deadline) element.
556 struct rb_root pushable_dl_tasks_root
;
557 struct rb_node
*pushable_dl_tasks_leftmost
;
565 static inline bool sched_asym_prefer(int a
, int b
)
567 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
571 * We add the notion of a root-domain which will be used to define per-domain
572 * variables. Each exclusive cpuset essentially defines an island domain by
573 * fully partitioning the member cpus from any other cpuset. Whenever a new
574 * exclusive cpuset is created, we also create and attach a new root-domain
583 cpumask_var_t online
;
585 /* Indicate more than one runnable task for any CPU */
589 * The bit corresponding to a CPU gets set here if such CPU has more
590 * than one runnable -deadline task (as it is below for RT tasks).
592 cpumask_var_t dlo_mask
;
598 * The "RT overload" flag: it gets set if a CPU has more than
599 * one runnable RT task.
601 cpumask_var_t rto_mask
;
602 struct cpupri cpupri
;
604 unsigned long max_cpu_capacity
;
607 extern struct root_domain def_root_domain
;
608 extern struct mutex sched_domains_mutex
;
609 extern cpumask_var_t fallback_doms
;
610 extern cpumask_var_t sched_domains_tmpmask
;
612 extern void init_defrootdomain(void);
613 extern int init_sched_domains(const struct cpumask
*cpu_map
);
614 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
616 #endif /* CONFIG_SMP */
619 * This is the main, per-CPU runqueue data structure.
621 * Locking rule: those places that want to lock multiple runqueues
622 * (such as the load balancing or the thread migration code), lock
623 * acquire operations must be ordered by ascending &runqueue.
630 * nr_running and cpu_load should be in the same cacheline because
631 * remote CPUs use both these fields when doing load calculation.
633 unsigned int nr_running
;
634 #ifdef CONFIG_NUMA_BALANCING
635 unsigned int nr_numa_running
;
636 unsigned int nr_preferred_running
;
638 #define CPU_LOAD_IDX_MAX 5
639 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
640 #ifdef CONFIG_NO_HZ_COMMON
642 unsigned long last_load_update_tick
;
643 #endif /* CONFIG_SMP */
644 unsigned long nohz_flags
;
645 #endif /* CONFIG_NO_HZ_COMMON */
646 #ifdef CONFIG_NO_HZ_FULL
647 unsigned long last_sched_tick
;
649 /* capture load from *all* tasks on this cpu: */
650 struct load_weight load
;
651 unsigned long nr_load_updates
;
658 #ifdef CONFIG_FAIR_GROUP_SCHED
659 /* list of leaf cfs_rq on this cpu: */
660 struct list_head leaf_cfs_rq_list
;
661 struct list_head
*tmp_alone_branch
;
662 #endif /* CONFIG_FAIR_GROUP_SCHED */
665 * This is part of a global counter where only the total sum
666 * over all CPUs matters. A task can increase this counter on
667 * one CPU and if it got migrated afterwards it may decrease
668 * it on another CPU. Always updated under the runqueue lock:
670 unsigned long nr_uninterruptible
;
672 struct task_struct
*curr
, *idle
, *stop
;
673 unsigned long next_balance
;
674 struct mm_struct
*prev_mm
;
676 unsigned int clock_update_flags
;
683 struct root_domain
*rd
;
684 struct sched_domain
*sd
;
686 unsigned long cpu_capacity
;
687 unsigned long cpu_capacity_orig
;
689 struct callback_head
*balance_callback
;
691 unsigned char idle_balance
;
692 /* For active balancing */
695 struct cpu_stop_work active_balance_work
;
696 /* cpu of this runqueue: */
700 struct list_head cfs_tasks
;
707 /* This is used to determine avg_idle's max value */
708 u64 max_idle_balance_cost
;
711 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
714 #ifdef CONFIG_PARAVIRT
717 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
718 u64 prev_steal_time_rq
;
721 /* calc_load related fields */
722 unsigned long calc_load_update
;
723 long calc_load_active
;
725 #ifdef CONFIG_SCHED_HRTICK
727 int hrtick_csd_pending
;
728 struct call_single_data hrtick_csd
;
730 struct hrtimer hrtick_timer
;
733 #ifdef CONFIG_SCHEDSTATS
735 struct sched_info rq_sched_info
;
736 unsigned long long rq_cpu_time
;
737 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
739 /* sys_sched_yield() stats */
740 unsigned int yld_count
;
742 /* schedule() stats */
743 unsigned int sched_count
;
744 unsigned int sched_goidle
;
746 /* try_to_wake_up() stats */
747 unsigned int ttwu_count
;
748 unsigned int ttwu_local
;
752 struct llist_head wake_list
;
755 #ifdef CONFIG_CPU_IDLE
756 /* Must be inspected within a rcu lock section */
757 struct cpuidle_state
*idle_state
;
761 static inline int cpu_of(struct rq
*rq
)
771 #ifdef CONFIG_SCHED_SMT
773 extern struct static_key_false sched_smt_present
;
775 extern void __update_idle_core(struct rq
*rq
);
777 static inline void update_idle_core(struct rq
*rq
)
779 if (static_branch_unlikely(&sched_smt_present
))
780 __update_idle_core(rq
);
784 static inline void update_idle_core(struct rq
*rq
) { }
787 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
789 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
790 #define this_rq() this_cpu_ptr(&runqueues)
791 #define task_rq(p) cpu_rq(task_cpu(p))
792 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
793 #define raw_rq() raw_cpu_ptr(&runqueues)
795 static inline u64
__rq_clock_broken(struct rq
*rq
)
797 return READ_ONCE(rq
->clock
);
801 * rq::clock_update_flags bits
803 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
804 * call to __schedule(). This is an optimisation to avoid
805 * neighbouring rq clock updates.
807 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
808 * in effect and calls to update_rq_clock() are being ignored.
810 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
811 * made to update_rq_clock() since the last time rq::lock was pinned.
813 * If inside of __schedule(), clock_update_flags will have been
814 * shifted left (a left shift is a cheap operation for the fast path
815 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
817 * if (rq-clock_update_flags >= RQCF_UPDATED)
819 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
820 * one position though, because the next rq_unpin_lock() will shift it
823 #define RQCF_REQ_SKIP 0x01
824 #define RQCF_ACT_SKIP 0x02
825 #define RQCF_UPDATED 0x04
827 static inline void assert_clock_updated(struct rq
*rq
)
830 * The only reason for not seeing a clock update since the
831 * last rq_pin_lock() is if we're currently skipping updates.
833 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
836 static inline u64
rq_clock(struct rq
*rq
)
838 lockdep_assert_held(&rq
->lock
);
839 assert_clock_updated(rq
);
844 static inline u64
rq_clock_task(struct rq
*rq
)
846 lockdep_assert_held(&rq
->lock
);
847 assert_clock_updated(rq
);
849 return rq
->clock_task
;
852 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
854 lockdep_assert_held(&rq
->lock
);
856 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
858 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
863 struct pin_cookie cookie
;
864 #ifdef CONFIG_SCHED_DEBUG
866 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
867 * current pin context is stashed here in case it needs to be
868 * restored in rq_repin_lock().
870 unsigned int clock_update_flags
;
874 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
876 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
878 #ifdef CONFIG_SCHED_DEBUG
879 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
880 rf
->clock_update_flags
= 0;
884 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
886 #ifdef CONFIG_SCHED_DEBUG
887 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
888 rf
->clock_update_flags
= RQCF_UPDATED
;
891 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
894 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
896 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
898 #ifdef CONFIG_SCHED_DEBUG
900 * Restore the value we stashed in @rf for this pin context.
902 rq
->clock_update_flags
|= rf
->clock_update_flags
;
907 enum numa_topology_type
{
912 extern enum numa_topology_type sched_numa_topology_type
;
913 extern int sched_max_numa_distance
;
914 extern bool find_numa_distance(int distance
);
918 extern void sched_init_numa(void);
919 extern void sched_domains_numa_masks_set(unsigned int cpu
);
920 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
922 static inline void sched_init_numa(void) { }
923 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
924 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
927 #ifdef CONFIG_NUMA_BALANCING
928 /* The regions in numa_faults array from task_struct */
929 enum numa_faults_stats
{
935 extern void sched_setnuma(struct task_struct
*p
, int node
);
936 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
937 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
938 #endif /* CONFIG_NUMA_BALANCING */
943 queue_balance_callback(struct rq
*rq
,
944 struct callback_head
*head
,
945 void (*func
)(struct rq
*rq
))
947 lockdep_assert_held(&rq
->lock
);
949 if (unlikely(head
->next
))
952 head
->func
= (void (*)(struct callback_head
*))func
;
953 head
->next
= rq
->balance_callback
;
954 rq
->balance_callback
= head
;
957 extern void sched_ttwu_pending(void);
959 #define rcu_dereference_check_sched_domain(p) \
960 rcu_dereference_check((p), \
961 lockdep_is_held(&sched_domains_mutex))
964 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
965 * See detach_destroy_domains: synchronize_sched for details.
967 * The domain tree of any CPU may only be accessed from within
968 * preempt-disabled sections.
970 #define for_each_domain(cpu, __sd) \
971 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
972 __sd; __sd = __sd->parent)
974 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
977 * highest_flag_domain - Return highest sched_domain containing flag.
978 * @cpu: The cpu whose highest level of sched domain is to
980 * @flag: The flag to check for the highest sched_domain
983 * Returns the highest sched_domain of a cpu which contains the given flag.
985 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
987 struct sched_domain
*sd
, *hsd
= NULL
;
989 for_each_domain(cpu
, sd
) {
990 if (!(sd
->flags
& flag
))
998 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
1000 struct sched_domain
*sd
;
1002 for_each_domain(cpu
, sd
) {
1003 if (sd
->flags
& flag
)
1010 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
1011 DECLARE_PER_CPU(int, sd_llc_size
);
1012 DECLARE_PER_CPU(int, sd_llc_id
);
1013 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
1014 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
1015 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
1017 struct sched_group_capacity
{
1020 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1023 unsigned long capacity
;
1024 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1025 unsigned long next_update
;
1026 int imbalance
; /* XXX unrelated to capacity but shared group state */
1028 unsigned long cpumask
[0]; /* iteration mask */
1031 struct sched_group
{
1032 struct sched_group
*next
; /* Must be a circular list */
1035 unsigned int group_weight
;
1036 struct sched_group_capacity
*sgc
;
1037 int asym_prefer_cpu
; /* cpu of highest priority in group */
1040 * The CPUs this group covers.
1042 * NOTE: this field is variable length. (Allocated dynamically
1043 * by attaching extra space to the end of the structure,
1044 * depending on how many CPUs the kernel has booted up with)
1046 unsigned long cpumask
[0];
1049 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
1051 return to_cpumask(sg
->cpumask
);
1055 * cpumask masking which cpus in the group are allowed to iterate up the domain
1058 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
1060 return to_cpumask(sg
->sgc
->cpumask
);
1064 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1065 * @group: The group whose first cpu is to be returned.
1067 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1069 return cpumask_first(sched_group_cpus(group
));
1072 extern int group_balance_cpu(struct sched_group
*sg
);
1074 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1075 void register_sched_domain_sysctl(void);
1076 void unregister_sched_domain_sysctl(void);
1078 static inline void register_sched_domain_sysctl(void)
1081 static inline void unregister_sched_domain_sysctl(void)
1088 static inline void sched_ttwu_pending(void) { }
1090 #endif /* CONFIG_SMP */
1093 #include "autogroup.h"
1095 #ifdef CONFIG_CGROUP_SCHED
1098 * Return the group to which this tasks belongs.
1100 * We cannot use task_css() and friends because the cgroup subsystem
1101 * changes that value before the cgroup_subsys::attach() method is called,
1102 * therefore we cannot pin it and might observe the wrong value.
1104 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1105 * core changes this before calling sched_move_task().
1107 * Instead we use a 'copy' which is updated from sched_move_task() while
1108 * holding both task_struct::pi_lock and rq::lock.
1110 static inline struct task_group
*task_group(struct task_struct
*p
)
1112 return p
->sched_task_group
;
1115 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1116 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1118 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1119 struct task_group
*tg
= task_group(p
);
1122 #ifdef CONFIG_FAIR_GROUP_SCHED
1123 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1124 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1125 p
->se
.parent
= tg
->se
[cpu
];
1128 #ifdef CONFIG_RT_GROUP_SCHED
1129 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1130 p
->rt
.parent
= tg
->rt_se
[cpu
];
1134 #else /* CONFIG_CGROUP_SCHED */
1136 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1137 static inline struct task_group
*task_group(struct task_struct
*p
)
1142 #endif /* CONFIG_CGROUP_SCHED */
1144 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1146 set_task_rq(p
, cpu
);
1149 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1150 * successfuly executed on another CPU. We must ensure that updates of
1151 * per-task data have been completed by this moment.
1154 #ifdef CONFIG_THREAD_INFO_IN_TASK
1157 task_thread_info(p
)->cpu
= cpu
;
1164 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1166 #ifdef CONFIG_SCHED_DEBUG
1167 # include <linux/static_key.h>
1168 # define const_debug __read_mostly
1170 # define const_debug const
1173 extern const_debug
unsigned int sysctl_sched_features
;
1175 #define SCHED_FEAT(name, enabled) \
1176 __SCHED_FEAT_##name ,
1179 #include "features.h"
1185 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1186 #define SCHED_FEAT(name, enabled) \
1187 static __always_inline bool static_branch_##name(struct static_key *key) \
1189 return static_key_##enabled(key); \
1192 #include "features.h"
1196 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1197 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1198 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1199 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1200 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1202 extern struct static_key_false sched_numa_balancing
;
1203 extern struct static_key_false sched_schedstats
;
1205 static inline u64
global_rt_period(void)
1207 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1210 static inline u64
global_rt_runtime(void)
1212 if (sysctl_sched_rt_runtime
< 0)
1215 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1218 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1220 return rq
->curr
== p
;
1223 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1228 return task_current(rq
, p
);
1232 static inline int task_on_rq_queued(struct task_struct
*p
)
1234 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1237 static inline int task_on_rq_migrating(struct task_struct
*p
)
1239 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1242 #ifndef prepare_arch_switch
1243 # define prepare_arch_switch(next) do { } while (0)
1245 #ifndef finish_arch_post_lock_switch
1246 # define finish_arch_post_lock_switch() do { } while (0)
1249 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1253 * We can optimise this out completely for !SMP, because the
1254 * SMP rebalancing from interrupt is the only thing that cares
1261 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1265 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1266 * We must ensure this doesn't happen until the switch is completely
1269 * In particular, the load of prev->state in finish_task_switch() must
1270 * happen before this.
1272 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1274 smp_store_release(&prev
->on_cpu
, 0);
1276 #ifdef CONFIG_DEBUG_SPINLOCK
1277 /* this is a valid case when another task releases the spinlock */
1278 rq
->lock
.owner
= current
;
1281 * If we are tracking spinlock dependencies then we have to
1282 * fix up the runqueue lock - which gets 'carried over' from
1283 * prev into current:
1285 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1287 raw_spin_unlock_irq(&rq
->lock
);
1293 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1294 #define WF_FORK 0x02 /* child wakeup after fork */
1295 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1298 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1299 * of tasks with abnormal "nice" values across CPUs the contribution that
1300 * each task makes to its run queue's load is weighted according to its
1301 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1302 * scaled version of the new time slice allocation that they receive on time
1306 #define WEIGHT_IDLEPRIO 3
1307 #define WMULT_IDLEPRIO 1431655765
1309 extern const int sched_prio_to_weight
[40];
1310 extern const u32 sched_prio_to_wmult
[40];
1313 * {de,en}queue flags:
1315 * DEQUEUE_SLEEP - task is no longer runnable
1316 * ENQUEUE_WAKEUP - task just became runnable
1318 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1319 * are in a known state which allows modification. Such pairs
1320 * should preserve as much state as possible.
1322 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1325 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1326 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1327 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1331 #define DEQUEUE_SLEEP 0x01
1332 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1333 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1334 #define DEQUEUE_NOCLOCK 0x08 /* matches ENQUEUE_NOCLOCK */
1336 #define ENQUEUE_WAKEUP 0x01
1337 #define ENQUEUE_RESTORE 0x02
1338 #define ENQUEUE_MOVE 0x04
1339 #define ENQUEUE_NOCLOCK 0x08
1341 #define ENQUEUE_HEAD 0x10
1342 #define ENQUEUE_REPLENISH 0x20
1344 #define ENQUEUE_MIGRATED 0x40
1346 #define ENQUEUE_MIGRATED 0x00
1349 #define RETRY_TASK ((void *)-1UL)
1351 struct sched_class
{
1352 const struct sched_class
*next
;
1354 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1355 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1356 void (*yield_task
) (struct rq
*rq
);
1357 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1359 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1362 * It is the responsibility of the pick_next_task() method that will
1363 * return the next task to call put_prev_task() on the @prev task or
1364 * something equivalent.
1366 * May return RETRY_TASK when it finds a higher prio class has runnable
1369 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1370 struct task_struct
*prev
,
1371 struct rq_flags
*rf
);
1372 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1375 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1376 void (*migrate_task_rq
)(struct task_struct
*p
);
1378 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1380 void (*set_cpus_allowed
)(struct task_struct
*p
,
1381 const struct cpumask
*newmask
);
1383 void (*rq_online
)(struct rq
*rq
);
1384 void (*rq_offline
)(struct rq
*rq
);
1387 void (*set_curr_task
) (struct rq
*rq
);
1388 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1389 void (*task_fork
) (struct task_struct
*p
);
1390 void (*task_dead
) (struct task_struct
*p
);
1393 * The switched_from() call is allowed to drop rq->lock, therefore we
1394 * cannot assume the switched_from/switched_to pair is serliazed by
1395 * rq->lock. They are however serialized by p->pi_lock.
1397 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1398 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1399 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1402 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1403 struct task_struct
*task
);
1405 void (*update_curr
) (struct rq
*rq
);
1407 #define TASK_SET_GROUP 0
1408 #define TASK_MOVE_GROUP 1
1410 #ifdef CONFIG_FAIR_GROUP_SCHED
1411 void (*task_change_group
) (struct task_struct
*p
, int type
);
1415 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1417 prev
->sched_class
->put_prev_task(rq
, prev
);
1420 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1422 curr
->sched_class
->set_curr_task(rq
);
1425 #define sched_class_highest (&stop_sched_class)
1426 #define for_each_class(class) \
1427 for (class = sched_class_highest; class; class = class->next)
1429 extern const struct sched_class stop_sched_class
;
1430 extern const struct sched_class dl_sched_class
;
1431 extern const struct sched_class rt_sched_class
;
1432 extern const struct sched_class fair_sched_class
;
1433 extern const struct sched_class idle_sched_class
;
1438 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1440 extern void trigger_load_balance(struct rq
*rq
);
1442 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1446 #ifdef CONFIG_CPU_IDLE
1447 static inline void idle_set_state(struct rq
*rq
,
1448 struct cpuidle_state
*idle_state
)
1450 rq
->idle_state
= idle_state
;
1453 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1455 SCHED_WARN_ON(!rcu_read_lock_held());
1456 return rq
->idle_state
;
1459 static inline void idle_set_state(struct rq
*rq
,
1460 struct cpuidle_state
*idle_state
)
1464 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1470 extern void sysrq_sched_debug_show(void);
1471 extern void sched_init_granularity(void);
1472 extern void update_max_interval(void);
1474 extern void init_sched_dl_class(void);
1475 extern void init_sched_rt_class(void);
1476 extern void init_sched_fair_class(void);
1478 extern void resched_curr(struct rq
*rq
);
1479 extern void resched_cpu(int cpu
);
1481 extern struct rt_bandwidth def_rt_bandwidth
;
1482 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1484 extern struct dl_bandwidth def_dl_bandwidth
;
1485 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1486 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1488 unsigned long to_ratio(u64 period
, u64 runtime
);
1490 extern void init_entity_runnable_average(struct sched_entity
*se
);
1491 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1493 #ifdef CONFIG_NO_HZ_FULL
1494 extern bool sched_can_stop_tick(struct rq
*rq
);
1497 * Tick may be needed by tasks in the runqueue depending on their policy and
1498 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1499 * nohz mode if necessary.
1501 static inline void sched_update_tick_dependency(struct rq
*rq
)
1505 if (!tick_nohz_full_enabled())
1510 if (!tick_nohz_full_cpu(cpu
))
1513 if (sched_can_stop_tick(rq
))
1514 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1516 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1519 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1522 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1524 unsigned prev_nr
= rq
->nr_running
;
1526 rq
->nr_running
= prev_nr
+ count
;
1528 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1530 if (!rq
->rd
->overload
)
1531 rq
->rd
->overload
= true;
1535 sched_update_tick_dependency(rq
);
1538 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1540 rq
->nr_running
-= count
;
1541 /* Check if we still need preemption */
1542 sched_update_tick_dependency(rq
);
1545 static inline void rq_last_tick_reset(struct rq
*rq
)
1547 #ifdef CONFIG_NO_HZ_FULL
1548 rq
->last_sched_tick
= jiffies
;
1552 extern void update_rq_clock(struct rq
*rq
);
1554 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1555 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1557 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1559 extern const_debug
unsigned int sysctl_sched_time_avg
;
1560 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1561 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1563 static inline u64
sched_avg_period(void)
1565 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1568 #ifdef CONFIG_SCHED_HRTICK
1572 * - enabled by features
1573 * - hrtimer is actually high res
1575 static inline int hrtick_enabled(struct rq
*rq
)
1577 if (!sched_feat(HRTICK
))
1579 if (!cpu_active(cpu_of(rq
)))
1581 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1584 void hrtick_start(struct rq
*rq
, u64 delay
);
1588 static inline int hrtick_enabled(struct rq
*rq
)
1593 #endif /* CONFIG_SCHED_HRTICK */
1596 extern void sched_avg_update(struct rq
*rq
);
1598 #ifndef arch_scale_freq_capacity
1599 static __always_inline
1600 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1602 return SCHED_CAPACITY_SCALE
;
1606 #ifndef arch_scale_cpu_capacity
1607 static __always_inline
1608 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1610 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1611 return sd
->smt_gain
/ sd
->span_weight
;
1613 return SCHED_CAPACITY_SCALE
;
1617 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1619 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1620 sched_avg_update(rq
);
1623 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1624 static inline void sched_avg_update(struct rq
*rq
) { }
1627 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1628 __acquires(rq
->lock
);
1630 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1631 __acquires(p
->pi_lock
)
1632 __acquires(rq
->lock
);
1634 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1635 __releases(rq
->lock
)
1637 rq_unpin_lock(rq
, rf
);
1638 raw_spin_unlock(&rq
->lock
);
1642 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1643 __releases(rq
->lock
)
1644 __releases(p
->pi_lock
)
1646 rq_unpin_lock(rq
, rf
);
1647 raw_spin_unlock(&rq
->lock
);
1648 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1652 rq_lock_irqsave(struct rq
*rq
, struct rq_flags
*rf
)
1653 __acquires(rq
->lock
)
1655 raw_spin_lock_irqsave(&rq
->lock
, rf
->flags
);
1656 rq_pin_lock(rq
, rf
);
1660 rq_lock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1661 __acquires(rq
->lock
)
1663 raw_spin_lock_irq(&rq
->lock
);
1664 rq_pin_lock(rq
, rf
);
1668 rq_lock(struct rq
*rq
, struct rq_flags
*rf
)
1669 __acquires(rq
->lock
)
1671 raw_spin_lock(&rq
->lock
);
1672 rq_pin_lock(rq
, rf
);
1676 rq_relock(struct rq
*rq
, struct rq_flags
*rf
)
1677 __acquires(rq
->lock
)
1679 raw_spin_lock(&rq
->lock
);
1680 rq_repin_lock(rq
, rf
);
1684 rq_unlock_irqrestore(struct rq
*rq
, struct rq_flags
*rf
)
1685 __releases(rq
->lock
)
1687 rq_unpin_lock(rq
, rf
);
1688 raw_spin_unlock_irqrestore(&rq
->lock
, rf
->flags
);
1692 rq_unlock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1693 __releases(rq
->lock
)
1695 rq_unpin_lock(rq
, rf
);
1696 raw_spin_unlock_irq(&rq
->lock
);
1700 rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1701 __releases(rq
->lock
)
1703 rq_unpin_lock(rq
, rf
);
1704 raw_spin_unlock(&rq
->lock
);
1708 #ifdef CONFIG_PREEMPT
1710 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1713 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1714 * way at the expense of forcing extra atomic operations in all
1715 * invocations. This assures that the double_lock is acquired using the
1716 * same underlying policy as the spinlock_t on this architecture, which
1717 * reduces latency compared to the unfair variant below. However, it
1718 * also adds more overhead and therefore may reduce throughput.
1720 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1721 __releases(this_rq
->lock
)
1722 __acquires(busiest
->lock
)
1723 __acquires(this_rq
->lock
)
1725 raw_spin_unlock(&this_rq
->lock
);
1726 double_rq_lock(this_rq
, busiest
);
1733 * Unfair double_lock_balance: Optimizes throughput at the expense of
1734 * latency by eliminating extra atomic operations when the locks are
1735 * already in proper order on entry. This favors lower cpu-ids and will
1736 * grant the double lock to lower cpus over higher ids under contention,
1737 * regardless of entry order into the function.
1739 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1740 __releases(this_rq
->lock
)
1741 __acquires(busiest
->lock
)
1742 __acquires(this_rq
->lock
)
1746 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1747 if (busiest
< this_rq
) {
1748 raw_spin_unlock(&this_rq
->lock
);
1749 raw_spin_lock(&busiest
->lock
);
1750 raw_spin_lock_nested(&this_rq
->lock
,
1751 SINGLE_DEPTH_NESTING
);
1754 raw_spin_lock_nested(&busiest
->lock
,
1755 SINGLE_DEPTH_NESTING
);
1760 #endif /* CONFIG_PREEMPT */
1763 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1765 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1767 if (unlikely(!irqs_disabled())) {
1768 /* printk() doesn't work good under rq->lock */
1769 raw_spin_unlock(&this_rq
->lock
);
1773 return _double_lock_balance(this_rq
, busiest
);
1776 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1777 __releases(busiest
->lock
)
1779 raw_spin_unlock(&busiest
->lock
);
1780 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1783 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1789 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1792 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1798 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1801 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1807 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1811 * double_rq_lock - safely lock two runqueues
1813 * Note this does not disable interrupts like task_rq_lock,
1814 * you need to do so manually before calling.
1816 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1817 __acquires(rq1
->lock
)
1818 __acquires(rq2
->lock
)
1820 BUG_ON(!irqs_disabled());
1822 raw_spin_lock(&rq1
->lock
);
1823 __acquire(rq2
->lock
); /* Fake it out ;) */
1826 raw_spin_lock(&rq1
->lock
);
1827 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1829 raw_spin_lock(&rq2
->lock
);
1830 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1836 * double_rq_unlock - safely unlock two runqueues
1838 * Note this does not restore interrupts like task_rq_unlock,
1839 * you need to do so manually after calling.
1841 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1842 __releases(rq1
->lock
)
1843 __releases(rq2
->lock
)
1845 raw_spin_unlock(&rq1
->lock
);
1847 raw_spin_unlock(&rq2
->lock
);
1849 __release(rq2
->lock
);
1852 extern void set_rq_online (struct rq
*rq
);
1853 extern void set_rq_offline(struct rq
*rq
);
1854 extern bool sched_smp_initialized
;
1856 #else /* CONFIG_SMP */
1859 * double_rq_lock - safely lock two runqueues
1861 * Note this does not disable interrupts like task_rq_lock,
1862 * you need to do so manually before calling.
1864 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1865 __acquires(rq1
->lock
)
1866 __acquires(rq2
->lock
)
1868 BUG_ON(!irqs_disabled());
1870 raw_spin_lock(&rq1
->lock
);
1871 __acquire(rq2
->lock
); /* Fake it out ;) */
1875 * double_rq_unlock - safely unlock two runqueues
1877 * Note this does not restore interrupts like task_rq_unlock,
1878 * you need to do so manually after calling.
1880 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1881 __releases(rq1
->lock
)
1882 __releases(rq2
->lock
)
1885 raw_spin_unlock(&rq1
->lock
);
1886 __release(rq2
->lock
);
1891 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1892 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1894 #ifdef CONFIG_SCHED_DEBUG
1895 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1896 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1897 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1899 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1900 #ifdef CONFIG_NUMA_BALANCING
1902 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1904 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1905 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1906 #endif /* CONFIG_NUMA_BALANCING */
1907 #endif /* CONFIG_SCHED_DEBUG */
1909 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1910 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1911 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1913 extern void cfs_bandwidth_usage_inc(void);
1914 extern void cfs_bandwidth_usage_dec(void);
1916 #ifdef CONFIG_NO_HZ_COMMON
1917 enum rq_nohz_flag_bits
{
1922 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1924 extern void nohz_balance_exit_idle(unsigned int cpu
);
1926 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1929 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1934 struct u64_stats_sync sync
;
1937 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
1940 * Returns the irqtime minus the softirq time computed by ksoftirqd.
1941 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
1942 * and never move forward.
1944 static inline u64
irq_time_read(int cpu
)
1946 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
1951 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
1952 total
= irqtime
->total
;
1953 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
1957 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1959 #ifdef CONFIG_CPU_FREQ
1960 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
1963 * cpufreq_update_util - Take a note about CPU utilization changes.
1964 * @rq: Runqueue to carry out the update for.
1965 * @flags: Update reason flags.
1967 * This function is called by the scheduler on the CPU whose utilization is
1970 * It can only be called from RCU-sched read-side critical sections.
1972 * The way cpufreq is currently arranged requires it to evaluate the CPU
1973 * performance state (frequency/voltage) on a regular basis to prevent it from
1974 * being stuck in a completely inadequate performance level for too long.
1975 * That is not guaranteed to happen if the updates are only triggered from CFS,
1976 * though, because they may not be coming in if RT or deadline tasks are active
1977 * all the time (or there are RT and DL tasks only).
1979 * As a workaround for that issue, this function is called by the RT and DL
1980 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1981 * but that really is a band-aid. Going forward it should be replaced with
1982 * solutions targeted more specifically at RT and DL tasks.
1984 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
1986 struct update_util_data
*data
;
1988 data
= rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data
));
1990 data
->func(data
, rq_clock(rq
), flags
);
1993 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
)
1995 if (cpu_of(rq
) == smp_processor_id())
1996 cpufreq_update_util(rq
, flags
);
1999 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
2000 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
) {}
2001 #endif /* CONFIG_CPU_FREQ */
2003 #ifdef arch_scale_freq_capacity
2004 #ifndef arch_scale_freq_invariant
2005 #define arch_scale_freq_invariant() (true)
2007 #else /* arch_scale_freq_capacity */
2008 #define arch_scale_freq_invariant() (false)