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), 0; })
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;
226 static inline void __dl_update(struct dl_bw
*dl_b
, s64 bw
);
229 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
, int cpus
)
231 dl_b
->total_bw
-= tsk_bw
;
232 __dl_update(dl_b
, (s32
)tsk_bw
/ cpus
);
236 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
, int cpus
)
238 dl_b
->total_bw
+= tsk_bw
;
239 __dl_update(dl_b
, -((s32
)tsk_bw
/ cpus
));
243 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
245 return dl_b
->bw
!= -1 &&
246 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
249 void dl_change_utilization(struct task_struct
*p
, u64 new_bw
);
250 extern void init_dl_bw(struct dl_bw
*dl_b
);
251 extern int sched_dl_global_validate(void);
252 extern void sched_dl_do_global(void);
253 extern int sched_dl_overflow(struct task_struct
*p
, int policy
,
254 const struct sched_attr
*attr
);
255 extern void __setparam_dl(struct task_struct
*p
, const struct sched_attr
*attr
);
256 extern void __getparam_dl(struct task_struct
*p
, struct sched_attr
*attr
);
257 extern bool __checkparam_dl(const struct sched_attr
*attr
);
258 extern void __dl_clear_params(struct task_struct
*p
);
259 extern bool dl_param_changed(struct task_struct
*p
, const struct sched_attr
*attr
);
260 extern int dl_task_can_attach(struct task_struct
*p
,
261 const struct cpumask
*cs_cpus_allowed
);
262 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask
*cur
,
263 const struct cpumask
*trial
);
264 extern bool dl_cpu_busy(unsigned int cpu
);
266 #ifdef CONFIG_CGROUP_SCHED
268 #include <linux/cgroup.h>
273 extern struct list_head task_groups
;
275 struct cfs_bandwidth
{
276 #ifdef CONFIG_CFS_BANDWIDTH
280 s64 hierarchical_quota
;
283 int idle
, period_active
;
284 struct hrtimer period_timer
, slack_timer
;
285 struct list_head throttled_cfs_rq
;
288 int nr_periods
, nr_throttled
;
293 /* task group related information */
295 struct cgroup_subsys_state css
;
297 #ifdef CONFIG_FAIR_GROUP_SCHED
298 /* schedulable entities of this group on each cpu */
299 struct sched_entity
**se
;
300 /* runqueue "owned" by this group on each cpu */
301 struct cfs_rq
**cfs_rq
;
302 unsigned long shares
;
306 * load_avg can be heavily contended at clock tick time, so put
307 * it in its own cacheline separated from the fields above which
308 * will also be accessed at each tick.
310 atomic_long_t load_avg ____cacheline_aligned
;
314 #ifdef CONFIG_RT_GROUP_SCHED
315 struct sched_rt_entity
**rt_se
;
316 struct rt_rq
**rt_rq
;
318 struct rt_bandwidth rt_bandwidth
;
322 struct list_head list
;
324 struct task_group
*parent
;
325 struct list_head siblings
;
326 struct list_head children
;
328 #ifdef CONFIG_SCHED_AUTOGROUP
329 struct autogroup
*autogroup
;
332 struct cfs_bandwidth cfs_bandwidth
;
335 #ifdef CONFIG_FAIR_GROUP_SCHED
336 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
339 * A weight of 0 or 1 can cause arithmetics problems.
340 * A weight of a cfs_rq is the sum of weights of which entities
341 * are queued on this cfs_rq, so a weight of a entity should not be
342 * too large, so as the shares value of a task group.
343 * (The default weight is 1024 - so there's no practical
344 * limitation from this.)
346 #define MIN_SHARES (1UL << 1)
347 #define MAX_SHARES (1UL << 18)
350 typedef int (*tg_visitor
)(struct task_group
*, void *);
352 extern int walk_tg_tree_from(struct task_group
*from
,
353 tg_visitor down
, tg_visitor up
, void *data
);
356 * Iterate the full tree, calling @down when first entering a node and @up when
357 * leaving it for the final time.
359 * Caller must hold rcu_lock or sufficient equivalent.
361 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
363 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
366 extern int tg_nop(struct task_group
*tg
, void *data
);
368 extern void free_fair_sched_group(struct task_group
*tg
);
369 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
370 extern void online_fair_sched_group(struct task_group
*tg
);
371 extern void unregister_fair_sched_group(struct task_group
*tg
);
372 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
373 struct sched_entity
*se
, int cpu
,
374 struct sched_entity
*parent
);
375 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
377 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
378 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
379 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
381 extern void free_rt_sched_group(struct task_group
*tg
);
382 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
383 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
384 struct sched_rt_entity
*rt_se
, int cpu
,
385 struct sched_rt_entity
*parent
);
386 extern int sched_group_set_rt_runtime(struct task_group
*tg
, long rt_runtime_us
);
387 extern int sched_group_set_rt_period(struct task_group
*tg
, u64 rt_period_us
);
388 extern long sched_group_rt_runtime(struct task_group
*tg
);
389 extern long sched_group_rt_period(struct task_group
*tg
);
390 extern int sched_rt_can_attach(struct task_group
*tg
, struct task_struct
*tsk
);
392 extern struct task_group
*sched_create_group(struct task_group
*parent
);
393 extern void sched_online_group(struct task_group
*tg
,
394 struct task_group
*parent
);
395 extern void sched_destroy_group(struct task_group
*tg
);
396 extern void sched_offline_group(struct task_group
*tg
);
398 extern void sched_move_task(struct task_struct
*tsk
);
400 #ifdef CONFIG_FAIR_GROUP_SCHED
401 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
404 extern void set_task_rq_fair(struct sched_entity
*se
,
405 struct cfs_rq
*prev
, struct cfs_rq
*next
);
406 #else /* !CONFIG_SMP */
407 static inline void set_task_rq_fair(struct sched_entity
*se
,
408 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
409 #endif /* CONFIG_SMP */
410 #endif /* CONFIG_FAIR_GROUP_SCHED */
412 #else /* CONFIG_CGROUP_SCHED */
414 struct cfs_bandwidth
{ };
416 #endif /* CONFIG_CGROUP_SCHED */
418 /* CFS-related fields in a runqueue */
420 struct load_weight load
;
421 unsigned int nr_running
, h_nr_running
;
426 u64 min_vruntime_copy
;
429 struct rb_root_cached tasks_timeline
;
432 * 'curr' points to currently running entity on this cfs_rq.
433 * It is set to NULL otherwise (i.e when none are currently running).
435 struct sched_entity
*curr
, *next
, *last
, *skip
;
437 #ifdef CONFIG_SCHED_DEBUG
438 unsigned int nr_spread_over
;
445 struct sched_avg avg
;
446 u64 runnable_load_sum
;
447 unsigned long runnable_load_avg
;
448 #ifdef CONFIG_FAIR_GROUP_SCHED
449 unsigned long tg_load_avg_contrib
;
450 unsigned long propagate_avg
;
452 atomic_long_t removed_load_avg
, removed_util_avg
;
454 u64 load_last_update_time_copy
;
457 #ifdef CONFIG_FAIR_GROUP_SCHED
459 * h_load = weight * f(tg)
461 * Where f(tg) is the recursive weight fraction assigned to
464 unsigned long h_load
;
465 u64 last_h_load_update
;
466 struct sched_entity
*h_load_next
;
467 #endif /* CONFIG_FAIR_GROUP_SCHED */
468 #endif /* CONFIG_SMP */
470 #ifdef CONFIG_FAIR_GROUP_SCHED
471 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
474 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
475 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
476 * (like users, containers etc.)
478 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
479 * list is used during load balance.
482 struct list_head leaf_cfs_rq_list
;
483 struct task_group
*tg
; /* group that "owns" this runqueue */
485 #ifdef CONFIG_CFS_BANDWIDTH
488 s64 runtime_remaining
;
490 u64 throttled_clock
, throttled_clock_task
;
491 u64 throttled_clock_task_time
;
492 int throttled
, throttle_count
;
493 struct list_head throttled_list
;
494 #endif /* CONFIG_CFS_BANDWIDTH */
495 #endif /* CONFIG_FAIR_GROUP_SCHED */
498 static inline int rt_bandwidth_enabled(void)
500 return sysctl_sched_rt_runtime
>= 0;
503 /* RT IPI pull logic requires IRQ_WORK */
504 #ifdef CONFIG_IRQ_WORK
505 # define HAVE_RT_PUSH_IPI
508 /* Real-Time classes' related field in a runqueue: */
510 struct rt_prio_array active
;
511 unsigned int rt_nr_running
;
512 unsigned int rr_nr_running
;
513 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
515 int curr
; /* highest queued rt task prio */
517 int next
; /* next highest */
522 unsigned long rt_nr_migratory
;
523 unsigned long rt_nr_total
;
525 struct plist_head pushable_tasks
;
526 #ifdef HAVE_RT_PUSH_IPI
529 struct irq_work push_work
;
530 raw_spinlock_t push_lock
;
532 #endif /* CONFIG_SMP */
538 /* Nests inside the rq lock: */
539 raw_spinlock_t rt_runtime_lock
;
541 #ifdef CONFIG_RT_GROUP_SCHED
542 unsigned long rt_nr_boosted
;
545 struct task_group
*tg
;
549 /* Deadline class' related fields in a runqueue */
551 /* runqueue is an rbtree, ordered by deadline */
552 struct rb_root_cached root
;
554 unsigned long dl_nr_running
;
558 * Deadline values of the currently executing and the
559 * earliest ready task on this rq. Caching these facilitates
560 * the decision wether or not a ready but not running task
561 * should migrate somewhere else.
568 unsigned long dl_nr_migratory
;
572 * Tasks on this rq that can be pushed away. They are kept in
573 * an rb-tree, ordered by tasks' deadlines, with caching
574 * of the leftmost (earliest deadline) element.
576 struct rb_root_cached pushable_dl_tasks_root
;
581 * "Active utilization" for this runqueue: increased when a
582 * task wakes up (becomes TASK_RUNNING) and decreased when a
588 * Utilization of the tasks "assigned" to this runqueue (including
589 * the tasks that are in runqueue and the tasks that executed on this
590 * CPU and blocked). Increased when a task moves to this runqueue, and
591 * decreased when the task moves away (migrates, changes scheduling
592 * policy, or terminates).
593 * This is needed to compute the "inactive utilization" for the
594 * runqueue (inactive utilization = this_bw - running_bw).
600 * Inverse of the fraction of CPU utilization that can be reclaimed
601 * by the GRUB algorithm.
608 static inline bool sched_asym_prefer(int a
, int b
)
610 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
614 * We add the notion of a root-domain which will be used to define per-domain
615 * variables. Each exclusive cpuset essentially defines an island domain by
616 * fully partitioning the member cpus from any other cpuset. Whenever a new
617 * exclusive cpuset is created, we also create and attach a new root-domain
626 cpumask_var_t online
;
628 /* Indicate more than one runnable task for any CPU */
632 * The bit corresponding to a CPU gets set here if such CPU has more
633 * than one runnable -deadline task (as it is below for RT tasks).
635 cpumask_var_t dlo_mask
;
641 * The "RT overload" flag: it gets set if a CPU has more than
642 * one runnable RT task.
644 cpumask_var_t rto_mask
;
645 struct cpupri cpupri
;
647 unsigned long max_cpu_capacity
;
650 extern struct root_domain def_root_domain
;
651 extern struct mutex sched_domains_mutex
;
653 extern void init_defrootdomain(void);
654 extern int sched_init_domains(const struct cpumask
*cpu_map
);
655 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
657 #endif /* CONFIG_SMP */
660 * This is the main, per-CPU runqueue data structure.
662 * Locking rule: those places that want to lock multiple runqueues
663 * (such as the load balancing or the thread migration code), lock
664 * acquire operations must be ordered by ascending &runqueue.
671 * nr_running and cpu_load should be in the same cacheline because
672 * remote CPUs use both these fields when doing load calculation.
674 unsigned int nr_running
;
675 #ifdef CONFIG_NUMA_BALANCING
676 unsigned int nr_numa_running
;
677 unsigned int nr_preferred_running
;
679 #define CPU_LOAD_IDX_MAX 5
680 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
681 #ifdef CONFIG_NO_HZ_COMMON
683 unsigned long last_load_update_tick
;
684 #endif /* CONFIG_SMP */
685 unsigned long nohz_flags
;
686 #endif /* CONFIG_NO_HZ_COMMON */
687 #ifdef CONFIG_NO_HZ_FULL
688 unsigned long last_sched_tick
;
690 /* capture load from *all* tasks on this cpu: */
691 struct load_weight load
;
692 unsigned long nr_load_updates
;
699 #ifdef CONFIG_FAIR_GROUP_SCHED
700 /* list of leaf cfs_rq on this cpu: */
701 struct list_head leaf_cfs_rq_list
;
702 struct list_head
*tmp_alone_branch
;
703 #endif /* CONFIG_FAIR_GROUP_SCHED */
706 * This is part of a global counter where only the total sum
707 * over all CPUs matters. A task can increase this counter on
708 * one CPU and if it got migrated afterwards it may decrease
709 * it on another CPU. Always updated under the runqueue lock:
711 unsigned long nr_uninterruptible
;
713 struct task_struct
*curr
, *idle
, *stop
;
714 unsigned long next_balance
;
715 struct mm_struct
*prev_mm
;
717 unsigned int clock_update_flags
;
724 struct root_domain
*rd
;
725 struct sched_domain
*sd
;
727 unsigned long cpu_capacity
;
728 unsigned long cpu_capacity_orig
;
730 struct callback_head
*balance_callback
;
732 unsigned char idle_balance
;
733 /* For active balancing */
736 struct cpu_stop_work active_balance_work
;
737 /* cpu of this runqueue: */
741 struct list_head cfs_tasks
;
748 /* This is used to determine avg_idle's max value */
749 u64 max_idle_balance_cost
;
752 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
755 #ifdef CONFIG_PARAVIRT
758 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
759 u64 prev_steal_time_rq
;
762 /* calc_load related fields */
763 unsigned long calc_load_update
;
764 long calc_load_active
;
766 #ifdef CONFIG_SCHED_HRTICK
768 int hrtick_csd_pending
;
769 call_single_data_t hrtick_csd
;
771 struct hrtimer hrtick_timer
;
774 #ifdef CONFIG_SCHEDSTATS
776 struct sched_info rq_sched_info
;
777 unsigned long long rq_cpu_time
;
778 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
780 /* sys_sched_yield() stats */
781 unsigned int yld_count
;
783 /* schedule() stats */
784 unsigned int sched_count
;
785 unsigned int sched_goidle
;
787 /* try_to_wake_up() stats */
788 unsigned int ttwu_count
;
789 unsigned int ttwu_local
;
793 struct llist_head wake_list
;
796 #ifdef CONFIG_CPU_IDLE
797 /* Must be inspected within a rcu lock section */
798 struct cpuidle_state
*idle_state
;
802 static inline int cpu_of(struct rq
*rq
)
812 #ifdef CONFIG_SCHED_SMT
814 extern struct static_key_false sched_smt_present
;
816 extern void __update_idle_core(struct rq
*rq
);
818 static inline void update_idle_core(struct rq
*rq
)
820 if (static_branch_unlikely(&sched_smt_present
))
821 __update_idle_core(rq
);
825 static inline void update_idle_core(struct rq
*rq
) { }
828 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
830 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
831 #define this_rq() this_cpu_ptr(&runqueues)
832 #define task_rq(p) cpu_rq(task_cpu(p))
833 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
834 #define raw_rq() raw_cpu_ptr(&runqueues)
836 static inline u64
__rq_clock_broken(struct rq
*rq
)
838 return READ_ONCE(rq
->clock
);
842 * rq::clock_update_flags bits
844 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
845 * call to __schedule(). This is an optimisation to avoid
846 * neighbouring rq clock updates.
848 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
849 * in effect and calls to update_rq_clock() are being ignored.
851 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
852 * made to update_rq_clock() since the last time rq::lock was pinned.
854 * If inside of __schedule(), clock_update_flags will have been
855 * shifted left (a left shift is a cheap operation for the fast path
856 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
858 * if (rq-clock_update_flags >= RQCF_UPDATED)
860 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
861 * one position though, because the next rq_unpin_lock() will shift it
864 #define RQCF_REQ_SKIP 0x01
865 #define RQCF_ACT_SKIP 0x02
866 #define RQCF_UPDATED 0x04
868 static inline void assert_clock_updated(struct rq
*rq
)
871 * The only reason for not seeing a clock update since the
872 * last rq_pin_lock() is if we're currently skipping updates.
874 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
877 static inline u64
rq_clock(struct rq
*rq
)
879 lockdep_assert_held(&rq
->lock
);
880 assert_clock_updated(rq
);
885 static inline u64
rq_clock_task(struct rq
*rq
)
887 lockdep_assert_held(&rq
->lock
);
888 assert_clock_updated(rq
);
890 return rq
->clock_task
;
893 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
895 lockdep_assert_held(&rq
->lock
);
897 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
899 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
904 struct pin_cookie cookie
;
905 #ifdef CONFIG_SCHED_DEBUG
907 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
908 * current pin context is stashed here in case it needs to be
909 * restored in rq_repin_lock().
911 unsigned int clock_update_flags
;
915 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
917 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
919 #ifdef CONFIG_SCHED_DEBUG
920 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
921 rf
->clock_update_flags
= 0;
925 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
927 #ifdef CONFIG_SCHED_DEBUG
928 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
929 rf
->clock_update_flags
= RQCF_UPDATED
;
932 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
935 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
937 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
939 #ifdef CONFIG_SCHED_DEBUG
941 * Restore the value we stashed in @rf for this pin context.
943 rq
->clock_update_flags
|= rf
->clock_update_flags
;
948 enum numa_topology_type
{
953 extern enum numa_topology_type sched_numa_topology_type
;
954 extern int sched_max_numa_distance
;
955 extern bool find_numa_distance(int distance
);
959 extern void sched_init_numa(void);
960 extern void sched_domains_numa_masks_set(unsigned int cpu
);
961 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
963 static inline void sched_init_numa(void) { }
964 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
965 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
968 #ifdef CONFIG_NUMA_BALANCING
969 /* The regions in numa_faults array from task_struct */
970 enum numa_faults_stats
{
976 extern void sched_setnuma(struct task_struct
*p
, int node
);
977 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
978 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
979 #endif /* CONFIG_NUMA_BALANCING */
984 queue_balance_callback(struct rq
*rq
,
985 struct callback_head
*head
,
986 void (*func
)(struct rq
*rq
))
988 lockdep_assert_held(&rq
->lock
);
990 if (unlikely(head
->next
))
993 head
->func
= (void (*)(struct callback_head
*))func
;
994 head
->next
= rq
->balance_callback
;
995 rq
->balance_callback
= head
;
998 extern void sched_ttwu_pending(void);
1000 #define rcu_dereference_check_sched_domain(p) \
1001 rcu_dereference_check((p), \
1002 lockdep_is_held(&sched_domains_mutex))
1005 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1006 * See detach_destroy_domains: synchronize_sched for details.
1008 * The domain tree of any CPU may only be accessed from within
1009 * preempt-disabled sections.
1011 #define for_each_domain(cpu, __sd) \
1012 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1013 __sd; __sd = __sd->parent)
1015 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1018 * highest_flag_domain - Return highest sched_domain containing flag.
1019 * @cpu: The cpu whose highest level of sched domain is to
1021 * @flag: The flag to check for the highest sched_domain
1022 * for the given cpu.
1024 * Returns the highest sched_domain of a cpu which contains the given flag.
1026 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
1028 struct sched_domain
*sd
, *hsd
= NULL
;
1030 for_each_domain(cpu
, sd
) {
1031 if (!(sd
->flags
& flag
))
1039 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
1041 struct sched_domain
*sd
;
1043 for_each_domain(cpu
, sd
) {
1044 if (sd
->flags
& flag
)
1051 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
1052 DECLARE_PER_CPU(int, sd_llc_size
);
1053 DECLARE_PER_CPU(int, sd_llc_id
);
1054 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
1055 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
1056 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
1058 struct sched_group_capacity
{
1061 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1064 unsigned long capacity
;
1065 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1066 unsigned long next_update
;
1067 int imbalance
; /* XXX unrelated to capacity but shared group state */
1069 #ifdef CONFIG_SCHED_DEBUG
1073 unsigned long cpumask
[0]; /* balance mask */
1076 struct sched_group
{
1077 struct sched_group
*next
; /* Must be a circular list */
1080 unsigned int group_weight
;
1081 struct sched_group_capacity
*sgc
;
1082 int asym_prefer_cpu
; /* cpu of highest priority in group */
1085 * The CPUs this group covers.
1087 * NOTE: this field is variable length. (Allocated dynamically
1088 * by attaching extra space to the end of the structure,
1089 * depending on how many CPUs the kernel has booted up with)
1091 unsigned long cpumask
[0];
1094 static inline struct cpumask
*sched_group_span(struct sched_group
*sg
)
1096 return to_cpumask(sg
->cpumask
);
1100 * See build_balance_mask().
1102 static inline struct cpumask
*group_balance_mask(struct sched_group
*sg
)
1104 return to_cpumask(sg
->sgc
->cpumask
);
1108 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1109 * @group: The group whose first cpu is to be returned.
1111 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1113 return cpumask_first(sched_group_span(group
));
1116 extern int group_balance_cpu(struct sched_group
*sg
);
1118 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1119 void register_sched_domain_sysctl(void);
1120 void dirty_sched_domain_sysctl(int cpu
);
1121 void unregister_sched_domain_sysctl(void);
1123 static inline void register_sched_domain_sysctl(void)
1126 static inline void dirty_sched_domain_sysctl(int cpu
)
1129 static inline void unregister_sched_domain_sysctl(void)
1136 static inline void sched_ttwu_pending(void) { }
1138 #endif /* CONFIG_SMP */
1141 #include "autogroup.h"
1143 #ifdef CONFIG_CGROUP_SCHED
1146 * Return the group to which this tasks belongs.
1148 * We cannot use task_css() and friends because the cgroup subsystem
1149 * changes that value before the cgroup_subsys::attach() method is called,
1150 * therefore we cannot pin it and might observe the wrong value.
1152 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1153 * core changes this before calling sched_move_task().
1155 * Instead we use a 'copy' which is updated from sched_move_task() while
1156 * holding both task_struct::pi_lock and rq::lock.
1158 static inline struct task_group
*task_group(struct task_struct
*p
)
1160 return p
->sched_task_group
;
1163 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1164 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1166 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1167 struct task_group
*tg
= task_group(p
);
1170 #ifdef CONFIG_FAIR_GROUP_SCHED
1171 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1172 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1173 p
->se
.parent
= tg
->se
[cpu
];
1176 #ifdef CONFIG_RT_GROUP_SCHED
1177 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1178 p
->rt
.parent
= tg
->rt_se
[cpu
];
1182 #else /* CONFIG_CGROUP_SCHED */
1184 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1185 static inline struct task_group
*task_group(struct task_struct
*p
)
1190 #endif /* CONFIG_CGROUP_SCHED */
1192 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1194 set_task_rq(p
, cpu
);
1197 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1198 * successfuly executed on another CPU. We must ensure that updates of
1199 * per-task data have been completed by this moment.
1202 #ifdef CONFIG_THREAD_INFO_IN_TASK
1205 task_thread_info(p
)->cpu
= cpu
;
1212 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1214 #ifdef CONFIG_SCHED_DEBUG
1215 # include <linux/static_key.h>
1216 # define const_debug __read_mostly
1218 # define const_debug const
1221 extern const_debug
unsigned int sysctl_sched_features
;
1223 #define SCHED_FEAT(name, enabled) \
1224 __SCHED_FEAT_##name ,
1227 #include "features.h"
1233 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1234 #define SCHED_FEAT(name, enabled) \
1235 static __always_inline bool static_branch_##name(struct static_key *key) \
1237 return static_key_##enabled(key); \
1240 #include "features.h"
1244 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1245 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1246 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1247 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1248 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1250 extern struct static_key_false sched_numa_balancing
;
1251 extern struct static_key_false sched_schedstats
;
1253 static inline u64
global_rt_period(void)
1255 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1258 static inline u64
global_rt_runtime(void)
1260 if (sysctl_sched_rt_runtime
< 0)
1263 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1266 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1268 return rq
->curr
== p
;
1271 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1276 return task_current(rq
, p
);
1280 static inline int task_on_rq_queued(struct task_struct
*p
)
1282 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1285 static inline int task_on_rq_migrating(struct task_struct
*p
)
1287 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1290 #ifndef prepare_arch_switch
1291 # define prepare_arch_switch(next) do { } while (0)
1293 #ifndef finish_arch_post_lock_switch
1294 # define finish_arch_post_lock_switch() do { } while (0)
1297 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1301 * We can optimise this out completely for !SMP, because the
1302 * SMP rebalancing from interrupt is the only thing that cares
1309 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1313 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1314 * We must ensure this doesn't happen until the switch is completely
1317 * In particular, the load of prev->state in finish_task_switch() must
1318 * happen before this.
1320 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1322 smp_store_release(&prev
->on_cpu
, 0);
1324 #ifdef CONFIG_DEBUG_SPINLOCK
1325 /* this is a valid case when another task releases the spinlock */
1326 rq
->lock
.owner
= current
;
1329 * If we are tracking spinlock dependencies then we have to
1330 * fix up the runqueue lock - which gets 'carried over' from
1331 * prev into current:
1333 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1335 raw_spin_unlock_irq(&rq
->lock
);
1341 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1342 #define WF_FORK 0x02 /* child wakeup after fork */
1343 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1346 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1347 * of tasks with abnormal "nice" values across CPUs the contribution that
1348 * each task makes to its run queue's load is weighted according to its
1349 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1350 * scaled version of the new time slice allocation that they receive on time
1354 #define WEIGHT_IDLEPRIO 3
1355 #define WMULT_IDLEPRIO 1431655765
1357 extern const int sched_prio_to_weight
[40];
1358 extern const u32 sched_prio_to_wmult
[40];
1361 * {de,en}queue flags:
1363 * DEQUEUE_SLEEP - task is no longer runnable
1364 * ENQUEUE_WAKEUP - task just became runnable
1366 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1367 * are in a known state which allows modification. Such pairs
1368 * should preserve as much state as possible.
1370 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1373 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1374 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1375 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1379 #define DEQUEUE_SLEEP 0x01
1380 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1381 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1382 #define DEQUEUE_NOCLOCK 0x08 /* matches ENQUEUE_NOCLOCK */
1384 #define ENQUEUE_WAKEUP 0x01
1385 #define ENQUEUE_RESTORE 0x02
1386 #define ENQUEUE_MOVE 0x04
1387 #define ENQUEUE_NOCLOCK 0x08
1389 #define ENQUEUE_HEAD 0x10
1390 #define ENQUEUE_REPLENISH 0x20
1392 #define ENQUEUE_MIGRATED 0x40
1394 #define ENQUEUE_MIGRATED 0x00
1397 #define RETRY_TASK ((void *)-1UL)
1399 struct sched_class
{
1400 const struct sched_class
*next
;
1402 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1403 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1404 void (*yield_task
) (struct rq
*rq
);
1405 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1407 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1410 * It is the responsibility of the pick_next_task() method that will
1411 * return the next task to call put_prev_task() on the @prev task or
1412 * something equivalent.
1414 * May return RETRY_TASK when it finds a higher prio class has runnable
1417 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1418 struct task_struct
*prev
,
1419 struct rq_flags
*rf
);
1420 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1423 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1424 void (*migrate_task_rq
)(struct task_struct
*p
);
1426 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1428 void (*set_cpus_allowed
)(struct task_struct
*p
,
1429 const struct cpumask
*newmask
);
1431 void (*rq_online
)(struct rq
*rq
);
1432 void (*rq_offline
)(struct rq
*rq
);
1435 void (*set_curr_task
) (struct rq
*rq
);
1436 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1437 void (*task_fork
) (struct task_struct
*p
);
1438 void (*task_dead
) (struct task_struct
*p
);
1441 * The switched_from() call is allowed to drop rq->lock, therefore we
1442 * cannot assume the switched_from/switched_to pair is serliazed by
1443 * rq->lock. They are however serialized by p->pi_lock.
1445 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1446 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1447 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1450 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1451 struct task_struct
*task
);
1453 void (*update_curr
) (struct rq
*rq
);
1455 #define TASK_SET_GROUP 0
1456 #define TASK_MOVE_GROUP 1
1458 #ifdef CONFIG_FAIR_GROUP_SCHED
1459 void (*task_change_group
) (struct task_struct
*p
, int type
);
1463 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1465 prev
->sched_class
->put_prev_task(rq
, prev
);
1468 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1470 curr
->sched_class
->set_curr_task(rq
);
1474 #define sched_class_highest (&stop_sched_class)
1476 #define sched_class_highest (&dl_sched_class)
1478 #define for_each_class(class) \
1479 for (class = sched_class_highest; class; class = class->next)
1481 extern const struct sched_class stop_sched_class
;
1482 extern const struct sched_class dl_sched_class
;
1483 extern const struct sched_class rt_sched_class
;
1484 extern const struct sched_class fair_sched_class
;
1485 extern const struct sched_class idle_sched_class
;
1490 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1492 extern void trigger_load_balance(struct rq
*rq
);
1494 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1498 #ifdef CONFIG_CPU_IDLE
1499 static inline void idle_set_state(struct rq
*rq
,
1500 struct cpuidle_state
*idle_state
)
1502 rq
->idle_state
= idle_state
;
1505 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1507 SCHED_WARN_ON(!rcu_read_lock_held());
1508 return rq
->idle_state
;
1511 static inline void idle_set_state(struct rq
*rq
,
1512 struct cpuidle_state
*idle_state
)
1516 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1522 extern void schedule_idle(void);
1524 extern void sysrq_sched_debug_show(void);
1525 extern void sched_init_granularity(void);
1526 extern void update_max_interval(void);
1528 extern void init_sched_dl_class(void);
1529 extern void init_sched_rt_class(void);
1530 extern void init_sched_fair_class(void);
1532 extern void resched_curr(struct rq
*rq
);
1533 extern void resched_cpu(int cpu
);
1535 extern struct rt_bandwidth def_rt_bandwidth
;
1536 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1538 extern struct dl_bandwidth def_dl_bandwidth
;
1539 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1540 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1541 extern void init_dl_inactive_task_timer(struct sched_dl_entity
*dl_se
);
1542 extern void init_dl_rq_bw_ratio(struct dl_rq
*dl_rq
);
1545 #define BW_UNIT (1 << BW_SHIFT)
1546 #define RATIO_SHIFT 8
1547 unsigned long to_ratio(u64 period
, u64 runtime
);
1549 extern void init_entity_runnable_average(struct sched_entity
*se
);
1550 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1552 #ifdef CONFIG_NO_HZ_FULL
1553 extern bool sched_can_stop_tick(struct rq
*rq
);
1556 * Tick may be needed by tasks in the runqueue depending on their policy and
1557 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1558 * nohz mode if necessary.
1560 static inline void sched_update_tick_dependency(struct rq
*rq
)
1564 if (!tick_nohz_full_enabled())
1569 if (!tick_nohz_full_cpu(cpu
))
1572 if (sched_can_stop_tick(rq
))
1573 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1575 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1578 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1581 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1583 unsigned prev_nr
= rq
->nr_running
;
1585 rq
->nr_running
= prev_nr
+ count
;
1587 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1589 if (!rq
->rd
->overload
)
1590 rq
->rd
->overload
= true;
1594 sched_update_tick_dependency(rq
);
1597 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1599 rq
->nr_running
-= count
;
1600 /* Check if we still need preemption */
1601 sched_update_tick_dependency(rq
);
1604 static inline void rq_last_tick_reset(struct rq
*rq
)
1606 #ifdef CONFIG_NO_HZ_FULL
1607 rq
->last_sched_tick
= jiffies
;
1611 extern void update_rq_clock(struct rq
*rq
);
1613 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1614 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1616 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1618 extern const_debug
unsigned int sysctl_sched_time_avg
;
1619 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1620 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1622 static inline u64
sched_avg_period(void)
1624 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1627 #ifdef CONFIG_SCHED_HRTICK
1631 * - enabled by features
1632 * - hrtimer is actually high res
1634 static inline int hrtick_enabled(struct rq
*rq
)
1636 if (!sched_feat(HRTICK
))
1638 if (!cpu_active(cpu_of(rq
)))
1640 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1643 void hrtick_start(struct rq
*rq
, u64 delay
);
1647 static inline int hrtick_enabled(struct rq
*rq
)
1652 #endif /* CONFIG_SCHED_HRTICK */
1655 extern void sched_avg_update(struct rq
*rq
);
1657 #ifndef arch_scale_freq_capacity
1658 static __always_inline
1659 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1661 return SCHED_CAPACITY_SCALE
;
1665 #ifndef arch_scale_cpu_capacity
1666 static __always_inline
1667 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1669 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1670 return sd
->smt_gain
/ sd
->span_weight
;
1672 return SCHED_CAPACITY_SCALE
;
1676 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1678 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1679 sched_avg_update(rq
);
1682 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1683 static inline void sched_avg_update(struct rq
*rq
) { }
1686 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1687 __acquires(rq
->lock
);
1689 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1690 __acquires(p
->pi_lock
)
1691 __acquires(rq
->lock
);
1693 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1694 __releases(rq
->lock
)
1696 rq_unpin_lock(rq
, rf
);
1697 raw_spin_unlock(&rq
->lock
);
1701 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1702 __releases(rq
->lock
)
1703 __releases(p
->pi_lock
)
1705 rq_unpin_lock(rq
, rf
);
1706 raw_spin_unlock(&rq
->lock
);
1707 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1711 rq_lock_irqsave(struct rq
*rq
, struct rq_flags
*rf
)
1712 __acquires(rq
->lock
)
1714 raw_spin_lock_irqsave(&rq
->lock
, rf
->flags
);
1715 rq_pin_lock(rq
, rf
);
1719 rq_lock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1720 __acquires(rq
->lock
)
1722 raw_spin_lock_irq(&rq
->lock
);
1723 rq_pin_lock(rq
, rf
);
1727 rq_lock(struct rq
*rq
, struct rq_flags
*rf
)
1728 __acquires(rq
->lock
)
1730 raw_spin_lock(&rq
->lock
);
1731 rq_pin_lock(rq
, rf
);
1735 rq_relock(struct rq
*rq
, struct rq_flags
*rf
)
1736 __acquires(rq
->lock
)
1738 raw_spin_lock(&rq
->lock
);
1739 rq_repin_lock(rq
, rf
);
1743 rq_unlock_irqrestore(struct rq
*rq
, struct rq_flags
*rf
)
1744 __releases(rq
->lock
)
1746 rq_unpin_lock(rq
, rf
);
1747 raw_spin_unlock_irqrestore(&rq
->lock
, rf
->flags
);
1751 rq_unlock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1752 __releases(rq
->lock
)
1754 rq_unpin_lock(rq
, rf
);
1755 raw_spin_unlock_irq(&rq
->lock
);
1759 rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1760 __releases(rq
->lock
)
1762 rq_unpin_lock(rq
, rf
);
1763 raw_spin_unlock(&rq
->lock
);
1767 #ifdef CONFIG_PREEMPT
1769 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1772 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1773 * way at the expense of forcing extra atomic operations in all
1774 * invocations. This assures that the double_lock is acquired using the
1775 * same underlying policy as the spinlock_t on this architecture, which
1776 * reduces latency compared to the unfair variant below. However, it
1777 * also adds more overhead and therefore may reduce throughput.
1779 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1780 __releases(this_rq
->lock
)
1781 __acquires(busiest
->lock
)
1782 __acquires(this_rq
->lock
)
1784 raw_spin_unlock(&this_rq
->lock
);
1785 double_rq_lock(this_rq
, busiest
);
1792 * Unfair double_lock_balance: Optimizes throughput at the expense of
1793 * latency by eliminating extra atomic operations when the locks are
1794 * already in proper order on entry. This favors lower cpu-ids and will
1795 * grant the double lock to lower cpus over higher ids under contention,
1796 * regardless of entry order into the function.
1798 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1799 __releases(this_rq
->lock
)
1800 __acquires(busiest
->lock
)
1801 __acquires(this_rq
->lock
)
1805 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1806 if (busiest
< this_rq
) {
1807 raw_spin_unlock(&this_rq
->lock
);
1808 raw_spin_lock(&busiest
->lock
);
1809 raw_spin_lock_nested(&this_rq
->lock
,
1810 SINGLE_DEPTH_NESTING
);
1813 raw_spin_lock_nested(&busiest
->lock
,
1814 SINGLE_DEPTH_NESTING
);
1819 #endif /* CONFIG_PREEMPT */
1822 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1824 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1826 if (unlikely(!irqs_disabled())) {
1827 /* printk() doesn't work good under rq->lock */
1828 raw_spin_unlock(&this_rq
->lock
);
1832 return _double_lock_balance(this_rq
, busiest
);
1835 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1836 __releases(busiest
->lock
)
1838 raw_spin_unlock(&busiest
->lock
);
1839 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1842 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1848 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1851 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1857 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1860 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1866 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1870 * double_rq_lock - safely lock two runqueues
1872 * Note this does not disable interrupts like task_rq_lock,
1873 * you need to do so manually before calling.
1875 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1876 __acquires(rq1
->lock
)
1877 __acquires(rq2
->lock
)
1879 BUG_ON(!irqs_disabled());
1881 raw_spin_lock(&rq1
->lock
);
1882 __acquire(rq2
->lock
); /* Fake it out ;) */
1885 raw_spin_lock(&rq1
->lock
);
1886 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1888 raw_spin_lock(&rq2
->lock
);
1889 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1895 * double_rq_unlock - safely unlock two runqueues
1897 * Note this does not restore interrupts like task_rq_unlock,
1898 * you need to do so manually after calling.
1900 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1901 __releases(rq1
->lock
)
1902 __releases(rq2
->lock
)
1904 raw_spin_unlock(&rq1
->lock
);
1906 raw_spin_unlock(&rq2
->lock
);
1908 __release(rq2
->lock
);
1911 extern void set_rq_online (struct rq
*rq
);
1912 extern void set_rq_offline(struct rq
*rq
);
1913 extern bool sched_smp_initialized
;
1915 #else /* CONFIG_SMP */
1918 * double_rq_lock - safely lock two runqueues
1920 * Note this does not disable interrupts like task_rq_lock,
1921 * you need to do so manually before calling.
1923 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1924 __acquires(rq1
->lock
)
1925 __acquires(rq2
->lock
)
1927 BUG_ON(!irqs_disabled());
1929 raw_spin_lock(&rq1
->lock
);
1930 __acquire(rq2
->lock
); /* Fake it out ;) */
1934 * double_rq_unlock - safely unlock two runqueues
1936 * Note this does not restore interrupts like task_rq_unlock,
1937 * you need to do so manually after calling.
1939 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1940 __releases(rq1
->lock
)
1941 __releases(rq2
->lock
)
1944 raw_spin_unlock(&rq1
->lock
);
1945 __release(rq2
->lock
);
1950 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1951 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1953 #ifdef CONFIG_SCHED_DEBUG
1954 extern bool sched_debug_enabled
;
1956 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1957 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1958 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1960 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1961 #ifdef CONFIG_NUMA_BALANCING
1963 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1965 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1966 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1967 #endif /* CONFIG_NUMA_BALANCING */
1968 #endif /* CONFIG_SCHED_DEBUG */
1970 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1971 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1972 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1974 extern void cfs_bandwidth_usage_inc(void);
1975 extern void cfs_bandwidth_usage_dec(void);
1977 #ifdef CONFIG_NO_HZ_COMMON
1978 enum rq_nohz_flag_bits
{
1983 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1985 extern void nohz_balance_exit_idle(unsigned int cpu
);
1987 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1993 void __dl_update(struct dl_bw
*dl_b
, s64 bw
)
1995 struct root_domain
*rd
= container_of(dl_b
, struct root_domain
, dl_bw
);
1998 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
1999 "sched RCU must be held");
2000 for_each_cpu_and(i
, rd
->span
, cpu_active_mask
) {
2001 struct rq
*rq
= cpu_rq(i
);
2003 rq
->dl
.extra_bw
+= bw
;
2008 void __dl_update(struct dl_bw
*dl_b
, s64 bw
)
2010 struct dl_rq
*dl
= container_of(dl_b
, struct dl_rq
, dl_bw
);
2017 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2022 struct u64_stats_sync sync
;
2025 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
2028 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2029 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2030 * and never move forward.
2032 static inline u64
irq_time_read(int cpu
)
2034 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
2039 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
2040 total
= irqtime
->total
;
2041 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
2045 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2047 #ifdef CONFIG_CPU_FREQ
2048 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
2051 * cpufreq_update_util - Take a note about CPU utilization changes.
2052 * @rq: Runqueue to carry out the update for.
2053 * @flags: Update reason flags.
2055 * This function is called by the scheduler on the CPU whose utilization is
2058 * It can only be called from RCU-sched read-side critical sections.
2060 * The way cpufreq is currently arranged requires it to evaluate the CPU
2061 * performance state (frequency/voltage) on a regular basis to prevent it from
2062 * being stuck in a completely inadequate performance level for too long.
2063 * That is not guaranteed to happen if the updates are only triggered from CFS,
2064 * though, because they may not be coming in if RT or deadline tasks are active
2065 * all the time (or there are RT and DL tasks only).
2067 * As a workaround for that issue, this function is called by the RT and DL
2068 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
2069 * but that really is a band-aid. Going forward it should be replaced with
2070 * solutions targeted more specifically at RT and DL tasks.
2072 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
2074 struct update_util_data
*data
;
2076 data
= rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data
,
2079 data
->func(data
, rq_clock(rq
), flags
);
2082 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
2083 #endif /* CONFIG_CPU_FREQ */
2085 #ifdef arch_scale_freq_capacity
2086 #ifndef arch_scale_freq_invariant
2087 #define arch_scale_freq_invariant() (true)
2089 #else /* arch_scale_freq_capacity */
2090 #define arch_scale_freq_invariant() (false)