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 tasks_timeline
;
430 struct rb_node
*rb_leftmost
;
433 * 'curr' points to currently running entity on this cfs_rq.
434 * It is set to NULL otherwise (i.e when none are currently running).
436 struct sched_entity
*curr
, *next
, *last
, *skip
;
438 #ifdef CONFIG_SCHED_DEBUG
439 unsigned int nr_spread_over
;
446 struct sched_avg avg
;
447 u64 runnable_load_sum
;
448 unsigned long runnable_load_avg
;
449 #ifdef CONFIG_FAIR_GROUP_SCHED
450 unsigned long tg_load_avg_contrib
;
451 unsigned long propagate_avg
;
453 atomic_long_t removed_load_avg
, removed_util_avg
;
455 u64 load_last_update_time_copy
;
458 #ifdef CONFIG_FAIR_GROUP_SCHED
460 * h_load = weight * f(tg)
462 * Where f(tg) is the recursive weight fraction assigned to
465 unsigned long h_load
;
466 u64 last_h_load_update
;
467 struct sched_entity
*h_load_next
;
468 #endif /* CONFIG_FAIR_GROUP_SCHED */
469 #endif /* CONFIG_SMP */
471 #ifdef CONFIG_FAIR_GROUP_SCHED
472 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
475 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
476 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
477 * (like users, containers etc.)
479 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
480 * list is used during load balance.
483 struct list_head leaf_cfs_rq_list
;
484 struct task_group
*tg
; /* group that "owns" this runqueue */
486 #ifdef CONFIG_CFS_BANDWIDTH
489 s64 runtime_remaining
;
491 u64 throttled_clock
, throttled_clock_task
;
492 u64 throttled_clock_task_time
;
493 int throttled
, throttle_count
;
494 struct list_head throttled_list
;
495 #endif /* CONFIG_CFS_BANDWIDTH */
496 #endif /* CONFIG_FAIR_GROUP_SCHED */
499 static inline int rt_bandwidth_enabled(void)
501 return sysctl_sched_rt_runtime
>= 0;
504 /* RT IPI pull logic requires IRQ_WORK */
505 #ifdef CONFIG_IRQ_WORK
506 # define HAVE_RT_PUSH_IPI
509 /* Real-Time classes' related field in a runqueue: */
511 struct rt_prio_array active
;
512 unsigned int rt_nr_running
;
513 unsigned int rr_nr_running
;
514 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
516 int curr
; /* highest queued rt task prio */
518 int next
; /* next highest */
523 unsigned long rt_nr_migratory
;
524 unsigned long rt_nr_total
;
526 struct plist_head pushable_tasks
;
527 #ifdef HAVE_RT_PUSH_IPI
530 struct irq_work push_work
;
531 raw_spinlock_t push_lock
;
533 #endif /* CONFIG_SMP */
539 /* Nests inside the rq lock: */
540 raw_spinlock_t rt_runtime_lock
;
542 #ifdef CONFIG_RT_GROUP_SCHED
543 unsigned long rt_nr_boosted
;
546 struct task_group
*tg
;
550 /* Deadline class' related fields in a runqueue */
552 /* runqueue is an rbtree, ordered by deadline */
553 struct rb_root rb_root
;
554 struct rb_node
*rb_leftmost
;
556 unsigned long dl_nr_running
;
560 * Deadline values of the currently executing and the
561 * earliest ready task on this rq. Caching these facilitates
562 * the decision wether or not a ready but not running task
563 * should migrate somewhere else.
570 unsigned long dl_nr_migratory
;
574 * Tasks on this rq that can be pushed away. They are kept in
575 * an rb-tree, ordered by tasks' deadlines, with caching
576 * of the leftmost (earliest deadline) element.
578 struct rb_root pushable_dl_tasks_root
;
579 struct rb_node
*pushable_dl_tasks_leftmost
;
584 * "Active utilization" for this runqueue: increased when a
585 * task wakes up (becomes TASK_RUNNING) and decreased when a
591 * Utilization of the tasks "assigned" to this runqueue (including
592 * the tasks that are in runqueue and the tasks that executed on this
593 * CPU and blocked). Increased when a task moves to this runqueue, and
594 * decreased when the task moves away (migrates, changes scheduling
595 * policy, or terminates).
596 * This is needed to compute the "inactive utilization" for the
597 * runqueue (inactive utilization = this_bw - running_bw).
603 * Inverse of the fraction of CPU utilization that can be reclaimed
604 * by the GRUB algorithm.
611 static inline bool sched_asym_prefer(int a
, int b
)
613 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
617 * We add the notion of a root-domain which will be used to define per-domain
618 * variables. Each exclusive cpuset essentially defines an island domain by
619 * fully partitioning the member cpus from any other cpuset. Whenever a new
620 * exclusive cpuset is created, we also create and attach a new root-domain
629 cpumask_var_t online
;
631 /* Indicate more than one runnable task for any CPU */
635 * The bit corresponding to a CPU gets set here if such CPU has more
636 * than one runnable -deadline task (as it is below for RT tasks).
638 cpumask_var_t dlo_mask
;
644 * The "RT overload" flag: it gets set if a CPU has more than
645 * one runnable RT task.
647 cpumask_var_t rto_mask
;
648 struct cpupri cpupri
;
650 unsigned long max_cpu_capacity
;
653 extern struct root_domain def_root_domain
;
654 extern struct mutex sched_domains_mutex
;
656 extern void init_defrootdomain(void);
657 extern int sched_init_domains(const struct cpumask
*cpu_map
);
658 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
660 #endif /* CONFIG_SMP */
663 * This is the main, per-CPU runqueue data structure.
665 * Locking rule: those places that want to lock multiple runqueues
666 * (such as the load balancing or the thread migration code), lock
667 * acquire operations must be ordered by ascending &runqueue.
674 * nr_running and cpu_load should be in the same cacheline because
675 * remote CPUs use both these fields when doing load calculation.
677 unsigned int nr_running
;
678 #ifdef CONFIG_NUMA_BALANCING
679 unsigned int nr_numa_running
;
680 unsigned int nr_preferred_running
;
682 #define CPU_LOAD_IDX_MAX 5
683 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
684 #ifdef CONFIG_NO_HZ_COMMON
686 unsigned long last_load_update_tick
;
687 #endif /* CONFIG_SMP */
688 unsigned long nohz_flags
;
689 #endif /* CONFIG_NO_HZ_COMMON */
690 #ifdef CONFIG_NO_HZ_FULL
691 unsigned long last_sched_tick
;
693 /* capture load from *all* tasks on this cpu: */
694 struct load_weight load
;
695 unsigned long nr_load_updates
;
702 #ifdef CONFIG_FAIR_GROUP_SCHED
703 /* list of leaf cfs_rq on this cpu: */
704 struct list_head leaf_cfs_rq_list
;
705 struct list_head
*tmp_alone_branch
;
706 #endif /* CONFIG_FAIR_GROUP_SCHED */
709 * This is part of a global counter where only the total sum
710 * over all CPUs matters. A task can increase this counter on
711 * one CPU and if it got migrated afterwards it may decrease
712 * it on another CPU. Always updated under the runqueue lock:
714 unsigned long nr_uninterruptible
;
716 struct task_struct
*curr
, *idle
, *stop
;
717 unsigned long next_balance
;
718 struct mm_struct
*prev_mm
;
720 unsigned int clock_update_flags
;
727 struct root_domain
*rd
;
728 struct sched_domain
*sd
;
730 unsigned long cpu_capacity
;
731 unsigned long cpu_capacity_orig
;
733 struct callback_head
*balance_callback
;
735 unsigned char idle_balance
;
736 /* For active balancing */
739 struct cpu_stop_work active_balance_work
;
740 /* cpu of this runqueue: */
744 struct list_head cfs_tasks
;
751 /* This is used to determine avg_idle's max value */
752 u64 max_idle_balance_cost
;
755 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
758 #ifdef CONFIG_PARAVIRT
761 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
762 u64 prev_steal_time_rq
;
765 /* calc_load related fields */
766 unsigned long calc_load_update
;
767 long calc_load_active
;
769 #ifdef CONFIG_SCHED_HRTICK
771 int hrtick_csd_pending
;
772 struct call_single_data hrtick_csd
;
774 struct hrtimer hrtick_timer
;
777 #ifdef CONFIG_SCHEDSTATS
779 struct sched_info rq_sched_info
;
780 unsigned long long rq_cpu_time
;
781 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
783 /* sys_sched_yield() stats */
784 unsigned int yld_count
;
786 /* schedule() stats */
787 unsigned int sched_count
;
788 unsigned int sched_goidle
;
790 /* try_to_wake_up() stats */
791 unsigned int ttwu_count
;
792 unsigned int ttwu_local
;
796 struct llist_head wake_list
;
799 #ifdef CONFIG_CPU_IDLE
800 /* Must be inspected within a rcu lock section */
801 struct cpuidle_state
*idle_state
;
805 static inline int cpu_of(struct rq
*rq
)
815 #ifdef CONFIG_SCHED_SMT
817 extern struct static_key_false sched_smt_present
;
819 extern void __update_idle_core(struct rq
*rq
);
821 static inline void update_idle_core(struct rq
*rq
)
823 if (static_branch_unlikely(&sched_smt_present
))
824 __update_idle_core(rq
);
828 static inline void update_idle_core(struct rq
*rq
) { }
831 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
833 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
834 #define this_rq() this_cpu_ptr(&runqueues)
835 #define task_rq(p) cpu_rq(task_cpu(p))
836 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
837 #define raw_rq() raw_cpu_ptr(&runqueues)
839 static inline u64
__rq_clock_broken(struct rq
*rq
)
841 return READ_ONCE(rq
->clock
);
845 * rq::clock_update_flags bits
847 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
848 * call to __schedule(). This is an optimisation to avoid
849 * neighbouring rq clock updates.
851 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
852 * in effect and calls to update_rq_clock() are being ignored.
854 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
855 * made to update_rq_clock() since the last time rq::lock was pinned.
857 * If inside of __schedule(), clock_update_flags will have been
858 * shifted left (a left shift is a cheap operation for the fast path
859 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
861 * if (rq-clock_update_flags >= RQCF_UPDATED)
863 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
864 * one position though, because the next rq_unpin_lock() will shift it
867 #define RQCF_REQ_SKIP 0x01
868 #define RQCF_ACT_SKIP 0x02
869 #define RQCF_UPDATED 0x04
871 static inline void assert_clock_updated(struct rq
*rq
)
874 * The only reason for not seeing a clock update since the
875 * last rq_pin_lock() is if we're currently skipping updates.
877 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
880 static inline u64
rq_clock(struct rq
*rq
)
882 lockdep_assert_held(&rq
->lock
);
883 assert_clock_updated(rq
);
888 static inline u64
rq_clock_task(struct rq
*rq
)
890 lockdep_assert_held(&rq
->lock
);
891 assert_clock_updated(rq
);
893 return rq
->clock_task
;
896 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
898 lockdep_assert_held(&rq
->lock
);
900 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
902 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
907 struct pin_cookie cookie
;
908 #ifdef CONFIG_SCHED_DEBUG
910 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
911 * current pin context is stashed here in case it needs to be
912 * restored in rq_repin_lock().
914 unsigned int clock_update_flags
;
918 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
920 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
922 #ifdef CONFIG_SCHED_DEBUG
923 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
924 rf
->clock_update_flags
= 0;
928 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
930 #ifdef CONFIG_SCHED_DEBUG
931 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
932 rf
->clock_update_flags
= RQCF_UPDATED
;
935 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
938 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
940 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
942 #ifdef CONFIG_SCHED_DEBUG
944 * Restore the value we stashed in @rf for this pin context.
946 rq
->clock_update_flags
|= rf
->clock_update_flags
;
951 enum numa_topology_type
{
956 extern enum numa_topology_type sched_numa_topology_type
;
957 extern int sched_max_numa_distance
;
958 extern bool find_numa_distance(int distance
);
962 extern void sched_init_numa(void);
963 extern void sched_domains_numa_masks_set(unsigned int cpu
);
964 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
966 static inline void sched_init_numa(void) { }
967 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
968 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
971 #ifdef CONFIG_NUMA_BALANCING
972 /* The regions in numa_faults array from task_struct */
973 enum numa_faults_stats
{
979 extern void sched_setnuma(struct task_struct
*p
, int node
);
980 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
981 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
982 #endif /* CONFIG_NUMA_BALANCING */
987 queue_balance_callback(struct rq
*rq
,
988 struct callback_head
*head
,
989 void (*func
)(struct rq
*rq
))
991 lockdep_assert_held(&rq
->lock
);
993 if (unlikely(head
->next
))
996 head
->func
= (void (*)(struct callback_head
*))func
;
997 head
->next
= rq
->balance_callback
;
998 rq
->balance_callback
= head
;
1001 extern void sched_ttwu_pending(void);
1003 #define rcu_dereference_check_sched_domain(p) \
1004 rcu_dereference_check((p), \
1005 lockdep_is_held(&sched_domains_mutex))
1008 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1009 * See detach_destroy_domains: synchronize_sched for details.
1011 * The domain tree of any CPU may only be accessed from within
1012 * preempt-disabled sections.
1014 #define for_each_domain(cpu, __sd) \
1015 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1016 __sd; __sd = __sd->parent)
1018 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1021 * highest_flag_domain - Return highest sched_domain containing flag.
1022 * @cpu: The cpu whose highest level of sched domain is to
1024 * @flag: The flag to check for the highest sched_domain
1025 * for the given cpu.
1027 * Returns the highest sched_domain of a cpu which contains the given flag.
1029 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
1031 struct sched_domain
*sd
, *hsd
= NULL
;
1033 for_each_domain(cpu
, sd
) {
1034 if (!(sd
->flags
& flag
))
1042 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
1044 struct sched_domain
*sd
;
1046 for_each_domain(cpu
, sd
) {
1047 if (sd
->flags
& flag
)
1054 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
1055 DECLARE_PER_CPU(int, sd_llc_size
);
1056 DECLARE_PER_CPU(int, sd_llc_id
);
1057 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
1058 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
1059 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
1061 struct sched_group_capacity
{
1064 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1067 unsigned long capacity
;
1068 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1069 unsigned long next_update
;
1070 int imbalance
; /* XXX unrelated to capacity but shared group state */
1072 #ifdef CONFIG_SCHED_DEBUG
1076 unsigned long cpumask
[0]; /* balance mask */
1079 struct sched_group
{
1080 struct sched_group
*next
; /* Must be a circular list */
1083 unsigned int group_weight
;
1084 struct sched_group_capacity
*sgc
;
1085 int asym_prefer_cpu
; /* cpu of highest priority in group */
1088 * The CPUs this group covers.
1090 * NOTE: this field is variable length. (Allocated dynamically
1091 * by attaching extra space to the end of the structure,
1092 * depending on how many CPUs the kernel has booted up with)
1094 unsigned long cpumask
[0];
1097 static inline struct cpumask
*sched_group_span(struct sched_group
*sg
)
1099 return to_cpumask(sg
->cpumask
);
1103 * See build_balance_mask().
1105 static inline struct cpumask
*group_balance_mask(struct sched_group
*sg
)
1107 return to_cpumask(sg
->sgc
->cpumask
);
1111 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1112 * @group: The group whose first cpu is to be returned.
1114 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1116 return cpumask_first(sched_group_span(group
));
1119 extern int group_balance_cpu(struct sched_group
*sg
);
1121 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1122 void register_sched_domain_sysctl(void);
1123 void unregister_sched_domain_sysctl(void);
1125 static inline void register_sched_domain_sysctl(void)
1128 static inline void unregister_sched_domain_sysctl(void)
1135 static inline void sched_ttwu_pending(void) { }
1137 #endif /* CONFIG_SMP */
1140 #include "autogroup.h"
1142 #ifdef CONFIG_CGROUP_SCHED
1145 * Return the group to which this tasks belongs.
1147 * We cannot use task_css() and friends because the cgroup subsystem
1148 * changes that value before the cgroup_subsys::attach() method is called,
1149 * therefore we cannot pin it and might observe the wrong value.
1151 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1152 * core changes this before calling sched_move_task().
1154 * Instead we use a 'copy' which is updated from sched_move_task() while
1155 * holding both task_struct::pi_lock and rq::lock.
1157 static inline struct task_group
*task_group(struct task_struct
*p
)
1159 return p
->sched_task_group
;
1162 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1163 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1165 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1166 struct task_group
*tg
= task_group(p
);
1169 #ifdef CONFIG_FAIR_GROUP_SCHED
1170 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1171 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1172 p
->se
.parent
= tg
->se
[cpu
];
1175 #ifdef CONFIG_RT_GROUP_SCHED
1176 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1177 p
->rt
.parent
= tg
->rt_se
[cpu
];
1181 #else /* CONFIG_CGROUP_SCHED */
1183 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1184 static inline struct task_group
*task_group(struct task_struct
*p
)
1189 #endif /* CONFIG_CGROUP_SCHED */
1191 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1193 set_task_rq(p
, cpu
);
1196 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1197 * successfuly executed on another CPU. We must ensure that updates of
1198 * per-task data have been completed by this moment.
1201 #ifdef CONFIG_THREAD_INFO_IN_TASK
1204 task_thread_info(p
)->cpu
= cpu
;
1211 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1213 #ifdef CONFIG_SCHED_DEBUG
1214 # include <linux/static_key.h>
1215 # define const_debug __read_mostly
1217 # define const_debug const
1220 extern const_debug
unsigned int sysctl_sched_features
;
1222 #define SCHED_FEAT(name, enabled) \
1223 __SCHED_FEAT_##name ,
1226 #include "features.h"
1232 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1233 #define SCHED_FEAT(name, enabled) \
1234 static __always_inline bool static_branch_##name(struct static_key *key) \
1236 return static_key_##enabled(key); \
1239 #include "features.h"
1243 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1244 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1245 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1246 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1247 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1249 extern struct static_key_false sched_numa_balancing
;
1250 extern struct static_key_false sched_schedstats
;
1252 static inline u64
global_rt_period(void)
1254 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1257 static inline u64
global_rt_runtime(void)
1259 if (sysctl_sched_rt_runtime
< 0)
1262 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1265 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1267 return rq
->curr
== p
;
1270 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1275 return task_current(rq
, p
);
1279 static inline int task_on_rq_queued(struct task_struct
*p
)
1281 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1284 static inline int task_on_rq_migrating(struct task_struct
*p
)
1286 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1289 #ifndef prepare_arch_switch
1290 # define prepare_arch_switch(next) do { } while (0)
1292 #ifndef finish_arch_post_lock_switch
1293 # define finish_arch_post_lock_switch() do { } while (0)
1296 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1300 * We can optimise this out completely for !SMP, because the
1301 * SMP rebalancing from interrupt is the only thing that cares
1308 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1312 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1313 * We must ensure this doesn't happen until the switch is completely
1316 * In particular, the load of prev->state in finish_task_switch() must
1317 * happen before this.
1319 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1321 smp_store_release(&prev
->on_cpu
, 0);
1323 #ifdef CONFIG_DEBUG_SPINLOCK
1324 /* this is a valid case when another task releases the spinlock */
1325 rq
->lock
.owner
= current
;
1328 * If we are tracking spinlock dependencies then we have to
1329 * fix up the runqueue lock - which gets 'carried over' from
1330 * prev into current:
1332 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1334 raw_spin_unlock_irq(&rq
->lock
);
1340 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1341 #define WF_FORK 0x02 /* child wakeup after fork */
1342 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1345 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1346 * of tasks with abnormal "nice" values across CPUs the contribution that
1347 * each task makes to its run queue's load is weighted according to its
1348 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1349 * scaled version of the new time slice allocation that they receive on time
1353 #define WEIGHT_IDLEPRIO 3
1354 #define WMULT_IDLEPRIO 1431655765
1356 extern const int sched_prio_to_weight
[40];
1357 extern const u32 sched_prio_to_wmult
[40];
1360 * {de,en}queue flags:
1362 * DEQUEUE_SLEEP - task is no longer runnable
1363 * ENQUEUE_WAKEUP - task just became runnable
1365 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1366 * are in a known state which allows modification. Such pairs
1367 * should preserve as much state as possible.
1369 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1372 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1373 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1374 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1378 #define DEQUEUE_SLEEP 0x01
1379 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1380 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1381 #define DEQUEUE_NOCLOCK 0x08 /* matches ENQUEUE_NOCLOCK */
1383 #define ENQUEUE_WAKEUP 0x01
1384 #define ENQUEUE_RESTORE 0x02
1385 #define ENQUEUE_MOVE 0x04
1386 #define ENQUEUE_NOCLOCK 0x08
1388 #define ENQUEUE_HEAD 0x10
1389 #define ENQUEUE_REPLENISH 0x20
1391 #define ENQUEUE_MIGRATED 0x40
1393 #define ENQUEUE_MIGRATED 0x00
1396 #define RETRY_TASK ((void *)-1UL)
1398 struct sched_class
{
1399 const struct sched_class
*next
;
1401 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1402 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1403 void (*yield_task
) (struct rq
*rq
);
1404 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1406 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1409 * It is the responsibility of the pick_next_task() method that will
1410 * return the next task to call put_prev_task() on the @prev task or
1411 * something equivalent.
1413 * May return RETRY_TASK when it finds a higher prio class has runnable
1416 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1417 struct task_struct
*prev
,
1418 struct rq_flags
*rf
);
1419 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1422 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1423 void (*migrate_task_rq
)(struct task_struct
*p
);
1425 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1427 void (*set_cpus_allowed
)(struct task_struct
*p
,
1428 const struct cpumask
*newmask
);
1430 void (*rq_online
)(struct rq
*rq
);
1431 void (*rq_offline
)(struct rq
*rq
);
1434 void (*set_curr_task
) (struct rq
*rq
);
1435 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1436 void (*task_fork
) (struct task_struct
*p
);
1437 void (*task_dead
) (struct task_struct
*p
);
1440 * The switched_from() call is allowed to drop rq->lock, therefore we
1441 * cannot assume the switched_from/switched_to pair is serliazed by
1442 * rq->lock. They are however serialized by p->pi_lock.
1444 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1445 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1446 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1449 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1450 struct task_struct
*task
);
1452 void (*update_curr
) (struct rq
*rq
);
1454 #define TASK_SET_GROUP 0
1455 #define TASK_MOVE_GROUP 1
1457 #ifdef CONFIG_FAIR_GROUP_SCHED
1458 void (*task_change_group
) (struct task_struct
*p
, int type
);
1462 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1464 prev
->sched_class
->put_prev_task(rq
, prev
);
1467 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1469 curr
->sched_class
->set_curr_task(rq
);
1473 #define sched_class_highest (&stop_sched_class)
1475 #define sched_class_highest (&dl_sched_class)
1477 #define for_each_class(class) \
1478 for (class = sched_class_highest; class; class = class->next)
1480 extern const struct sched_class stop_sched_class
;
1481 extern const struct sched_class dl_sched_class
;
1482 extern const struct sched_class rt_sched_class
;
1483 extern const struct sched_class fair_sched_class
;
1484 extern const struct sched_class idle_sched_class
;
1489 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1491 extern void trigger_load_balance(struct rq
*rq
);
1493 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1497 #ifdef CONFIG_CPU_IDLE
1498 static inline void idle_set_state(struct rq
*rq
,
1499 struct cpuidle_state
*idle_state
)
1501 rq
->idle_state
= idle_state
;
1504 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1506 SCHED_WARN_ON(!rcu_read_lock_held());
1507 return rq
->idle_state
;
1510 static inline void idle_set_state(struct rq
*rq
,
1511 struct cpuidle_state
*idle_state
)
1515 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1521 extern void schedule_idle(void);
1523 extern void sysrq_sched_debug_show(void);
1524 extern void sched_init_granularity(void);
1525 extern void update_max_interval(void);
1527 extern void init_sched_dl_class(void);
1528 extern void init_sched_rt_class(void);
1529 extern void init_sched_fair_class(void);
1531 extern void resched_curr(struct rq
*rq
);
1532 extern void resched_cpu(int cpu
);
1534 extern struct rt_bandwidth def_rt_bandwidth
;
1535 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1537 extern struct dl_bandwidth def_dl_bandwidth
;
1538 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1539 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1540 extern void init_dl_inactive_task_timer(struct sched_dl_entity
*dl_se
);
1541 extern void init_dl_rq_bw_ratio(struct dl_rq
*dl_rq
);
1544 #define BW_UNIT (1 << BW_SHIFT)
1545 #define RATIO_SHIFT 8
1546 unsigned long to_ratio(u64 period
, u64 runtime
);
1548 extern void init_entity_runnable_average(struct sched_entity
*se
);
1549 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1551 #ifdef CONFIG_NO_HZ_FULL
1552 extern bool sched_can_stop_tick(struct rq
*rq
);
1555 * Tick may be needed by tasks in the runqueue depending on their policy and
1556 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1557 * nohz mode if necessary.
1559 static inline void sched_update_tick_dependency(struct rq
*rq
)
1563 if (!tick_nohz_full_enabled())
1568 if (!tick_nohz_full_cpu(cpu
))
1571 if (sched_can_stop_tick(rq
))
1572 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1574 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1577 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1580 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1582 unsigned prev_nr
= rq
->nr_running
;
1584 rq
->nr_running
= prev_nr
+ count
;
1586 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1588 if (!rq
->rd
->overload
)
1589 rq
->rd
->overload
= true;
1593 sched_update_tick_dependency(rq
);
1596 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1598 rq
->nr_running
-= count
;
1599 /* Check if we still need preemption */
1600 sched_update_tick_dependency(rq
);
1603 static inline void rq_last_tick_reset(struct rq
*rq
)
1605 #ifdef CONFIG_NO_HZ_FULL
1606 rq
->last_sched_tick
= jiffies
;
1610 extern void update_rq_clock(struct rq
*rq
);
1612 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1613 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1615 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1617 extern const_debug
unsigned int sysctl_sched_time_avg
;
1618 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1619 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1621 static inline u64
sched_avg_period(void)
1623 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1626 #ifdef CONFIG_SCHED_HRTICK
1630 * - enabled by features
1631 * - hrtimer is actually high res
1633 static inline int hrtick_enabled(struct rq
*rq
)
1635 if (!sched_feat(HRTICK
))
1637 if (!cpu_active(cpu_of(rq
)))
1639 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1642 void hrtick_start(struct rq
*rq
, u64 delay
);
1646 static inline int hrtick_enabled(struct rq
*rq
)
1651 #endif /* CONFIG_SCHED_HRTICK */
1654 extern void sched_avg_update(struct rq
*rq
);
1656 #ifndef arch_scale_freq_capacity
1657 static __always_inline
1658 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1660 return SCHED_CAPACITY_SCALE
;
1664 #ifndef arch_scale_cpu_capacity
1665 static __always_inline
1666 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1668 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1669 return sd
->smt_gain
/ sd
->span_weight
;
1671 return SCHED_CAPACITY_SCALE
;
1675 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1677 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1678 sched_avg_update(rq
);
1681 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1682 static inline void sched_avg_update(struct rq
*rq
) { }
1685 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1686 __acquires(rq
->lock
);
1688 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1689 __acquires(p
->pi_lock
)
1690 __acquires(rq
->lock
);
1692 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1693 __releases(rq
->lock
)
1695 rq_unpin_lock(rq
, rf
);
1696 raw_spin_unlock(&rq
->lock
);
1700 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1701 __releases(rq
->lock
)
1702 __releases(p
->pi_lock
)
1704 rq_unpin_lock(rq
, rf
);
1705 raw_spin_unlock(&rq
->lock
);
1706 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1710 rq_lock_irqsave(struct rq
*rq
, struct rq_flags
*rf
)
1711 __acquires(rq
->lock
)
1713 raw_spin_lock_irqsave(&rq
->lock
, rf
->flags
);
1714 rq_pin_lock(rq
, rf
);
1718 rq_lock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1719 __acquires(rq
->lock
)
1721 raw_spin_lock_irq(&rq
->lock
);
1722 rq_pin_lock(rq
, rf
);
1726 rq_lock(struct rq
*rq
, struct rq_flags
*rf
)
1727 __acquires(rq
->lock
)
1729 raw_spin_lock(&rq
->lock
);
1730 rq_pin_lock(rq
, rf
);
1734 rq_relock(struct rq
*rq
, struct rq_flags
*rf
)
1735 __acquires(rq
->lock
)
1737 raw_spin_lock(&rq
->lock
);
1738 rq_repin_lock(rq
, rf
);
1742 rq_unlock_irqrestore(struct rq
*rq
, struct rq_flags
*rf
)
1743 __releases(rq
->lock
)
1745 rq_unpin_lock(rq
, rf
);
1746 raw_spin_unlock_irqrestore(&rq
->lock
, rf
->flags
);
1750 rq_unlock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1751 __releases(rq
->lock
)
1753 rq_unpin_lock(rq
, rf
);
1754 raw_spin_unlock_irq(&rq
->lock
);
1758 rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1759 __releases(rq
->lock
)
1761 rq_unpin_lock(rq
, rf
);
1762 raw_spin_unlock(&rq
->lock
);
1766 #ifdef CONFIG_PREEMPT
1768 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1771 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1772 * way at the expense of forcing extra atomic operations in all
1773 * invocations. This assures that the double_lock is acquired using the
1774 * same underlying policy as the spinlock_t on this architecture, which
1775 * reduces latency compared to the unfair variant below. However, it
1776 * also adds more overhead and therefore may reduce throughput.
1778 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1779 __releases(this_rq
->lock
)
1780 __acquires(busiest
->lock
)
1781 __acquires(this_rq
->lock
)
1783 raw_spin_unlock(&this_rq
->lock
);
1784 double_rq_lock(this_rq
, busiest
);
1791 * Unfair double_lock_balance: Optimizes throughput at the expense of
1792 * latency by eliminating extra atomic operations when the locks are
1793 * already in proper order on entry. This favors lower cpu-ids and will
1794 * grant the double lock to lower cpus over higher ids under contention,
1795 * regardless of entry order into the function.
1797 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1798 __releases(this_rq
->lock
)
1799 __acquires(busiest
->lock
)
1800 __acquires(this_rq
->lock
)
1804 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1805 if (busiest
< this_rq
) {
1806 raw_spin_unlock(&this_rq
->lock
);
1807 raw_spin_lock(&busiest
->lock
);
1808 raw_spin_lock_nested(&this_rq
->lock
,
1809 SINGLE_DEPTH_NESTING
);
1812 raw_spin_lock_nested(&busiest
->lock
,
1813 SINGLE_DEPTH_NESTING
);
1818 #endif /* CONFIG_PREEMPT */
1821 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1823 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1825 if (unlikely(!irqs_disabled())) {
1826 /* printk() doesn't work good under rq->lock */
1827 raw_spin_unlock(&this_rq
->lock
);
1831 return _double_lock_balance(this_rq
, busiest
);
1834 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1835 __releases(busiest
->lock
)
1837 raw_spin_unlock(&busiest
->lock
);
1838 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1841 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1847 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1850 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1856 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1859 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1865 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1869 * double_rq_lock - safely lock two runqueues
1871 * Note this does not disable interrupts like task_rq_lock,
1872 * you need to do so manually before calling.
1874 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1875 __acquires(rq1
->lock
)
1876 __acquires(rq2
->lock
)
1878 BUG_ON(!irqs_disabled());
1880 raw_spin_lock(&rq1
->lock
);
1881 __acquire(rq2
->lock
); /* Fake it out ;) */
1884 raw_spin_lock(&rq1
->lock
);
1885 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1887 raw_spin_lock(&rq2
->lock
);
1888 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1894 * double_rq_unlock - safely unlock two runqueues
1896 * Note this does not restore interrupts like task_rq_unlock,
1897 * you need to do so manually after calling.
1899 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1900 __releases(rq1
->lock
)
1901 __releases(rq2
->lock
)
1903 raw_spin_unlock(&rq1
->lock
);
1905 raw_spin_unlock(&rq2
->lock
);
1907 __release(rq2
->lock
);
1910 extern void set_rq_online (struct rq
*rq
);
1911 extern void set_rq_offline(struct rq
*rq
);
1912 extern bool sched_smp_initialized
;
1914 #else /* CONFIG_SMP */
1917 * double_rq_lock - safely lock two runqueues
1919 * Note this does not disable interrupts like task_rq_lock,
1920 * you need to do so manually before calling.
1922 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1923 __acquires(rq1
->lock
)
1924 __acquires(rq2
->lock
)
1926 BUG_ON(!irqs_disabled());
1928 raw_spin_lock(&rq1
->lock
);
1929 __acquire(rq2
->lock
); /* Fake it out ;) */
1933 * double_rq_unlock - safely unlock two runqueues
1935 * Note this does not restore interrupts like task_rq_unlock,
1936 * you need to do so manually after calling.
1938 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1939 __releases(rq1
->lock
)
1940 __releases(rq2
->lock
)
1943 raw_spin_unlock(&rq1
->lock
);
1944 __release(rq2
->lock
);
1949 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1950 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1952 #ifdef CONFIG_SCHED_DEBUG
1953 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1954 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1955 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1957 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1958 #ifdef CONFIG_NUMA_BALANCING
1960 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1962 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1963 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1964 #endif /* CONFIG_NUMA_BALANCING */
1965 #endif /* CONFIG_SCHED_DEBUG */
1967 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1968 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1969 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1971 extern void cfs_bandwidth_usage_inc(void);
1972 extern void cfs_bandwidth_usage_dec(void);
1974 #ifdef CONFIG_NO_HZ_COMMON
1975 enum rq_nohz_flag_bits
{
1980 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1982 extern void nohz_balance_exit_idle(unsigned int cpu
);
1984 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1990 void __dl_update(struct dl_bw
*dl_b
, s64 bw
)
1992 struct root_domain
*rd
= container_of(dl_b
, struct root_domain
, dl_bw
);
1995 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
1996 "sched RCU must be held");
1997 for_each_cpu_and(i
, rd
->span
, cpu_active_mask
) {
1998 struct rq
*rq
= cpu_rq(i
);
2000 rq
->dl
.extra_bw
+= bw
;
2005 void __dl_update(struct dl_bw
*dl_b
, s64 bw
)
2007 struct dl_rq
*dl
= container_of(dl_b
, struct dl_rq
, dl_bw
);
2014 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2019 struct u64_stats_sync sync
;
2022 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
2025 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2026 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2027 * and never move forward.
2029 static inline u64
irq_time_read(int cpu
)
2031 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
2036 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
2037 total
= irqtime
->total
;
2038 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
2042 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2044 #ifdef CONFIG_CPU_FREQ
2045 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
2048 * cpufreq_update_util - Take a note about CPU utilization changes.
2049 * @rq: Runqueue to carry out the update for.
2050 * @flags: Update reason flags.
2052 * This function is called by the scheduler on the CPU whose utilization is
2055 * It can only be called from RCU-sched read-side critical sections.
2057 * The way cpufreq is currently arranged requires it to evaluate the CPU
2058 * performance state (frequency/voltage) on a regular basis to prevent it from
2059 * being stuck in a completely inadequate performance level for too long.
2060 * That is not guaranteed to happen if the updates are only triggered from CFS,
2061 * though, because they may not be coming in if RT or deadline tasks are active
2062 * all the time (or there are RT and DL tasks only).
2064 * As a workaround for that issue, this function is called by the RT and DL
2065 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
2066 * but that really is a band-aid. Going forward it should be replaced with
2067 * solutions targeted more specifically at RT and DL tasks.
2069 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
2071 struct update_util_data
*data
;
2073 data
= rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data
));
2075 data
->func(data
, rq_clock(rq
), flags
);
2078 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
)
2080 if (cpu_of(rq
) == smp_processor_id())
2081 cpufreq_update_util(rq
, flags
);
2084 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
2085 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
) {}
2086 #endif /* CONFIG_CPU_FREQ */
2088 #ifdef arch_scale_freq_capacity
2089 #ifndef arch_scale_freq_invariant
2090 #define arch_scale_freq_invariant() (true)
2092 #else /* arch_scale_freq_capacity */
2093 #define arch_scale_freq_invariant() (false)