1 /* SPDX-License-Identifier: GPL-2.0 */
3 #include <linux/sched.h>
4 #include <linux/sched/autogroup.h>
5 #include <linux/sched/sysctl.h>
6 #include <linux/sched/topology.h>
7 #include <linux/sched/rt.h>
8 #include <linux/sched/deadline.h>
9 #include <linux/sched/clock.h>
10 #include <linux/sched/wake_q.h>
11 #include <linux/sched/signal.h>
12 #include <linux/sched/numa_balancing.h>
13 #include <linux/sched/mm.h>
14 #include <linux/sched/cpufreq.h>
15 #include <linux/sched/stat.h>
16 #include <linux/sched/nohz.h>
17 #include <linux/sched/debug.h>
18 #include <linux/sched/hotplug.h>
19 #include <linux/sched/task.h>
20 #include <linux/sched/task_stack.h>
21 #include <linux/sched/cputime.h>
22 #include <linux/sched/init.h>
24 #include <linux/u64_stats_sync.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/binfmts.h>
27 #include <linux/mutex.h>
28 #include <linux/spinlock.h>
29 #include <linux/stop_machine.h>
30 #include <linux/irq_work.h>
31 #include <linux/tick.h>
32 #include <linux/slab.h>
34 #ifdef CONFIG_PARAVIRT
35 #include <asm/paravirt.h>
39 #include "cpudeadline.h"
42 #ifdef CONFIG_SCHED_DEBUG
43 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
45 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
51 /* task_struct::on_rq states: */
52 #define TASK_ON_RQ_QUEUED 1
53 #define TASK_ON_RQ_MIGRATING 2
55 extern __read_mostly
int scheduler_running
;
57 extern unsigned long calc_load_update
;
58 extern atomic_long_t calc_load_tasks
;
60 extern void calc_global_load_tick(struct rq
*this_rq
);
61 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
64 extern void cpu_load_update_active(struct rq
*this_rq
);
66 static inline void cpu_load_update_active(struct rq
*this_rq
) { }
70 * Helpers for converting nanosecond timing to jiffy resolution
72 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
75 * Increase resolution of nice-level calculations for 64-bit architectures.
76 * The extra resolution improves shares distribution and load balancing of
77 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
78 * hierarchies, especially on larger systems. This is not a user-visible change
79 * and does not change the user-interface for setting shares/weights.
81 * We increase resolution only if we have enough bits to allow this increased
82 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
83 * pretty high and the returns do not justify the increased costs.
85 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
86 * increase coverage and consistency always enable it on 64bit platforms.
89 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
90 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
91 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
93 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
94 # define scale_load(w) (w)
95 # define scale_load_down(w) (w)
99 * Task weight (visible to users) and its load (invisible to users) have
100 * independent resolution, but they should be well calibrated. We use
101 * scale_load() and scale_load_down(w) to convert between them. The
102 * following must be true:
104 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
107 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
110 * Single value that decides SCHED_DEADLINE internal math precision.
111 * 10 -> just above 1us
112 * 9 -> just above 0.5us
114 #define DL_SCALE (10)
117 * These are the 'tuning knobs' of the scheduler:
121 * single value that denotes runtime == period, ie unlimited time.
123 #define RUNTIME_INF ((u64)~0ULL)
125 static inline int idle_policy(int policy
)
127 return policy
== SCHED_IDLE
;
129 static inline int fair_policy(int policy
)
131 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
134 static inline int rt_policy(int policy
)
136 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
139 static inline int dl_policy(int policy
)
141 return policy
== SCHED_DEADLINE
;
143 static inline bool valid_policy(int policy
)
145 return idle_policy(policy
) || fair_policy(policy
) ||
146 rt_policy(policy
) || dl_policy(policy
);
149 static inline int task_has_rt_policy(struct task_struct
*p
)
151 return rt_policy(p
->policy
);
154 static inline int task_has_dl_policy(struct task_struct
*p
)
156 return dl_policy(p
->policy
);
160 * Tells if entity @a should preempt entity @b.
163 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
165 return dl_time_before(a
->deadline
, b
->deadline
);
169 * This is the priority-queue data structure of the RT scheduling class:
171 struct rt_prio_array
{
172 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
173 struct list_head queue
[MAX_RT_PRIO
];
176 struct rt_bandwidth
{
177 /* nests inside the rq lock: */
178 raw_spinlock_t rt_runtime_lock
;
181 struct hrtimer rt_period_timer
;
182 unsigned int rt_period_active
;
185 void __dl_clear_params(struct task_struct
*p
);
188 * To keep the bandwidth of -deadline tasks and groups under control
189 * we need some place where:
190 * - store the maximum -deadline bandwidth of the system (the group);
191 * - cache the fraction of that bandwidth that is currently allocated.
193 * This is all done in the data structure below. It is similar to the
194 * one used for RT-throttling (rt_bandwidth), with the main difference
195 * that, since here we are only interested in admission control, we
196 * do not decrease any runtime while the group "executes", neither we
197 * need a timer to replenish it.
199 * With respect to SMP, the bandwidth is given on a per-CPU basis,
201 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
202 * - dl_total_bw array contains, in the i-eth element, the currently
203 * allocated bandwidth on the i-eth CPU.
204 * Moreover, groups consume bandwidth on each CPU, while tasks only
205 * consume bandwidth on the CPU they're running on.
206 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
207 * that will be shown the next time the proc or cgroup controls will
208 * be red. It on its turn can be changed by writing on its own
211 struct dl_bandwidth
{
212 raw_spinlock_t dl_runtime_lock
;
217 static inline int dl_bandwidth_enabled(void)
219 return sysctl_sched_rt_runtime
>= 0;
227 static inline void __dl_update(struct dl_bw
*dl_b
, s64 bw
);
230 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
, int cpus
)
232 dl_b
->total_bw
-= tsk_bw
;
233 __dl_update(dl_b
, (s32
)tsk_bw
/ cpus
);
237 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
, int cpus
)
239 dl_b
->total_bw
+= tsk_bw
;
240 __dl_update(dl_b
, -((s32
)tsk_bw
/ cpus
));
244 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
246 return dl_b
->bw
!= -1 &&
247 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
250 void dl_change_utilization(struct task_struct
*p
, u64 new_bw
);
251 extern void init_dl_bw(struct dl_bw
*dl_b
);
252 extern int sched_dl_global_validate(void);
253 extern void sched_dl_do_global(void);
254 extern int sched_dl_overflow(struct task_struct
*p
, int policy
,
255 const struct sched_attr
*attr
);
256 extern void __setparam_dl(struct task_struct
*p
, const struct sched_attr
*attr
);
257 extern void __getparam_dl(struct task_struct
*p
, struct sched_attr
*attr
);
258 extern bool __checkparam_dl(const struct sched_attr
*attr
);
259 extern void __dl_clear_params(struct task_struct
*p
);
260 extern bool dl_param_changed(struct task_struct
*p
, const struct sched_attr
*attr
);
261 extern int dl_task_can_attach(struct task_struct
*p
,
262 const struct cpumask
*cs_cpus_allowed
);
263 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask
*cur
,
264 const struct cpumask
*trial
);
265 extern bool dl_cpu_busy(unsigned int cpu
);
267 #ifdef CONFIG_CGROUP_SCHED
269 #include <linux/cgroup.h>
274 extern struct list_head task_groups
;
276 struct cfs_bandwidth
{
277 #ifdef CONFIG_CFS_BANDWIDTH
281 s64 hierarchical_quota
;
284 int idle
, period_active
;
285 struct hrtimer period_timer
, slack_timer
;
286 struct list_head throttled_cfs_rq
;
289 int nr_periods
, nr_throttled
;
294 /* task group related information */
296 struct cgroup_subsys_state css
;
298 #ifdef CONFIG_FAIR_GROUP_SCHED
299 /* schedulable entities of this group on each cpu */
300 struct sched_entity
**se
;
301 /* runqueue "owned" by this group on each cpu */
302 struct cfs_rq
**cfs_rq
;
303 unsigned long shares
;
307 * load_avg can be heavily contended at clock tick time, so put
308 * it in its own cacheline separated from the fields above which
309 * will also be accessed at each tick.
311 atomic_long_t load_avg ____cacheline_aligned
;
315 #ifdef CONFIG_RT_GROUP_SCHED
316 struct sched_rt_entity
**rt_se
;
317 struct rt_rq
**rt_rq
;
319 struct rt_bandwidth rt_bandwidth
;
323 struct list_head list
;
325 struct task_group
*parent
;
326 struct list_head siblings
;
327 struct list_head children
;
329 #ifdef CONFIG_SCHED_AUTOGROUP
330 struct autogroup
*autogroup
;
333 struct cfs_bandwidth cfs_bandwidth
;
336 #ifdef CONFIG_FAIR_GROUP_SCHED
337 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
340 * A weight of 0 or 1 can cause arithmetics problems.
341 * A weight of a cfs_rq is the sum of weights of which entities
342 * are queued on this cfs_rq, so a weight of a entity should not be
343 * too large, so as the shares value of a task group.
344 * (The default weight is 1024 - so there's no practical
345 * limitation from this.)
347 #define MIN_SHARES (1UL << 1)
348 #define MAX_SHARES (1UL << 18)
351 typedef int (*tg_visitor
)(struct task_group
*, void *);
353 extern int walk_tg_tree_from(struct task_group
*from
,
354 tg_visitor down
, tg_visitor up
, void *data
);
357 * Iterate the full tree, calling @down when first entering a node and @up when
358 * leaving it for the final time.
360 * Caller must hold rcu_lock or sufficient equivalent.
362 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
364 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
367 extern int tg_nop(struct task_group
*tg
, void *data
);
369 extern void free_fair_sched_group(struct task_group
*tg
);
370 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
371 extern void online_fair_sched_group(struct task_group
*tg
);
372 extern void unregister_fair_sched_group(struct task_group
*tg
);
373 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
374 struct sched_entity
*se
, int cpu
,
375 struct sched_entity
*parent
);
376 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
378 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
379 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
380 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
382 extern void free_rt_sched_group(struct task_group
*tg
);
383 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
384 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
385 struct sched_rt_entity
*rt_se
, int cpu
,
386 struct sched_rt_entity
*parent
);
387 extern int sched_group_set_rt_runtime(struct task_group
*tg
, long rt_runtime_us
);
388 extern int sched_group_set_rt_period(struct task_group
*tg
, u64 rt_period_us
);
389 extern long sched_group_rt_runtime(struct task_group
*tg
);
390 extern long sched_group_rt_period(struct task_group
*tg
);
391 extern int sched_rt_can_attach(struct task_group
*tg
, struct task_struct
*tsk
);
393 extern struct task_group
*sched_create_group(struct task_group
*parent
);
394 extern void sched_online_group(struct task_group
*tg
,
395 struct task_group
*parent
);
396 extern void sched_destroy_group(struct task_group
*tg
);
397 extern void sched_offline_group(struct task_group
*tg
);
399 extern void sched_move_task(struct task_struct
*tsk
);
401 #ifdef CONFIG_FAIR_GROUP_SCHED
402 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
405 extern void set_task_rq_fair(struct sched_entity
*se
,
406 struct cfs_rq
*prev
, struct cfs_rq
*next
);
407 #else /* !CONFIG_SMP */
408 static inline void set_task_rq_fair(struct sched_entity
*se
,
409 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
410 #endif /* CONFIG_SMP */
411 #endif /* CONFIG_FAIR_GROUP_SCHED */
413 #else /* CONFIG_CGROUP_SCHED */
415 struct cfs_bandwidth
{ };
417 #endif /* CONFIG_CGROUP_SCHED */
419 /* CFS-related fields in a runqueue */
421 struct load_weight load
;
422 unsigned int nr_running
, h_nr_running
;
427 u64 min_vruntime_copy
;
430 struct rb_root_cached tasks_timeline
;
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_cached root
;
555 unsigned long dl_nr_running
;
559 * Deadline values of the currently executing and the
560 * earliest ready task on this rq. Caching these facilitates
561 * the decision wether or not a ready but not running task
562 * should migrate somewhere else.
569 unsigned long dl_nr_migratory
;
573 * Tasks on this rq that can be pushed away. They are kept in
574 * an rb-tree, ordered by tasks' deadlines, with caching
575 * of the leftmost (earliest deadline) element.
577 struct rb_root_cached pushable_dl_tasks_root
;
582 * "Active utilization" for this runqueue: increased when a
583 * task wakes up (becomes TASK_RUNNING) and decreased when a
589 * Utilization of the tasks "assigned" to this runqueue (including
590 * the tasks that are in runqueue and the tasks that executed on this
591 * CPU and blocked). Increased when a task moves to this runqueue, and
592 * decreased when the task moves away (migrates, changes scheduling
593 * policy, or terminates).
594 * This is needed to compute the "inactive utilization" for the
595 * runqueue (inactive utilization = this_bw - running_bw).
601 * Inverse of the fraction of CPU utilization that can be reclaimed
602 * by the GRUB algorithm.
609 static inline bool sched_asym_prefer(int a
, int b
)
611 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
615 * We add the notion of a root-domain which will be used to define per-domain
616 * variables. Each exclusive cpuset essentially defines an island domain by
617 * fully partitioning the member cpus from any other cpuset. Whenever a new
618 * exclusive cpuset is created, we also create and attach a new root-domain
627 cpumask_var_t online
;
629 /* Indicate more than one runnable task for any CPU */
633 * The bit corresponding to a CPU gets set here if such CPU has more
634 * than one runnable -deadline task (as it is below for RT tasks).
636 cpumask_var_t dlo_mask
;
642 * The "RT overload" flag: it gets set if a CPU has more than
643 * one runnable RT task.
645 cpumask_var_t rto_mask
;
646 struct cpupri cpupri
;
648 unsigned long max_cpu_capacity
;
651 extern struct root_domain def_root_domain
;
652 extern struct mutex sched_domains_mutex
;
654 extern void init_defrootdomain(void);
655 extern int sched_init_domains(const struct cpumask
*cpu_map
);
656 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
658 #endif /* CONFIG_SMP */
661 * This is the main, per-CPU runqueue data structure.
663 * Locking rule: those places that want to lock multiple runqueues
664 * (such as the load balancing or the thread migration code), lock
665 * acquire operations must be ordered by ascending &runqueue.
672 * nr_running and cpu_load should be in the same cacheline because
673 * remote CPUs use both these fields when doing load calculation.
675 unsigned int nr_running
;
676 #ifdef CONFIG_NUMA_BALANCING
677 unsigned int nr_numa_running
;
678 unsigned int nr_preferred_running
;
680 #define CPU_LOAD_IDX_MAX 5
681 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
682 #ifdef CONFIG_NO_HZ_COMMON
684 unsigned long last_load_update_tick
;
685 #endif /* CONFIG_SMP */
686 unsigned long nohz_flags
;
687 #endif /* CONFIG_NO_HZ_COMMON */
688 #ifdef CONFIG_NO_HZ_FULL
689 unsigned long last_sched_tick
;
691 /* capture load from *all* tasks on this cpu: */
692 struct load_weight load
;
693 unsigned long nr_load_updates
;
700 #ifdef CONFIG_FAIR_GROUP_SCHED
701 /* list of leaf cfs_rq on this cpu: */
702 struct list_head leaf_cfs_rq_list
;
703 struct list_head
*tmp_alone_branch
;
704 #endif /* CONFIG_FAIR_GROUP_SCHED */
707 * This is part of a global counter where only the total sum
708 * over all CPUs matters. A task can increase this counter on
709 * one CPU and if it got migrated afterwards it may decrease
710 * it on another CPU. Always updated under the runqueue lock:
712 unsigned long nr_uninterruptible
;
714 struct task_struct
*curr
, *idle
, *stop
;
715 unsigned long next_balance
;
716 struct mm_struct
*prev_mm
;
718 unsigned int clock_update_flags
;
725 struct root_domain
*rd
;
726 struct sched_domain
*sd
;
728 unsigned long cpu_capacity
;
729 unsigned long cpu_capacity_orig
;
731 struct callback_head
*balance_callback
;
733 unsigned char idle_balance
;
734 /* For active balancing */
737 struct cpu_stop_work active_balance_work
;
738 /* cpu of this runqueue: */
742 struct list_head cfs_tasks
;
749 /* This is used to determine avg_idle's max value */
750 u64 max_idle_balance_cost
;
753 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
756 #ifdef CONFIG_PARAVIRT
759 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
760 u64 prev_steal_time_rq
;
763 /* calc_load related fields */
764 unsigned long calc_load_update
;
765 long calc_load_active
;
767 #ifdef CONFIG_SCHED_HRTICK
769 int hrtick_csd_pending
;
770 call_single_data_t hrtick_csd
;
772 struct hrtimer hrtick_timer
;
775 #ifdef CONFIG_SCHEDSTATS
777 struct sched_info rq_sched_info
;
778 unsigned long long rq_cpu_time
;
779 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
781 /* sys_sched_yield() stats */
782 unsigned int yld_count
;
784 /* schedule() stats */
785 unsigned int sched_count
;
786 unsigned int sched_goidle
;
788 /* try_to_wake_up() stats */
789 unsigned int ttwu_count
;
790 unsigned int ttwu_local
;
794 struct llist_head wake_list
;
797 #ifdef CONFIG_CPU_IDLE
798 /* Must be inspected within a rcu lock section */
799 struct cpuidle_state
*idle_state
;
803 static inline int cpu_of(struct rq
*rq
)
813 #ifdef CONFIG_SCHED_SMT
815 extern struct static_key_false sched_smt_present
;
817 extern void __update_idle_core(struct rq
*rq
);
819 static inline void update_idle_core(struct rq
*rq
)
821 if (static_branch_unlikely(&sched_smt_present
))
822 __update_idle_core(rq
);
826 static inline void update_idle_core(struct rq
*rq
) { }
829 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
831 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
832 #define this_rq() this_cpu_ptr(&runqueues)
833 #define task_rq(p) cpu_rq(task_cpu(p))
834 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
835 #define raw_rq() raw_cpu_ptr(&runqueues)
837 static inline u64
__rq_clock_broken(struct rq
*rq
)
839 return READ_ONCE(rq
->clock
);
843 * rq::clock_update_flags bits
845 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
846 * call to __schedule(). This is an optimisation to avoid
847 * neighbouring rq clock updates.
849 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
850 * in effect and calls to update_rq_clock() are being ignored.
852 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
853 * made to update_rq_clock() since the last time rq::lock was pinned.
855 * If inside of __schedule(), clock_update_flags will have been
856 * shifted left (a left shift is a cheap operation for the fast path
857 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
859 * if (rq-clock_update_flags >= RQCF_UPDATED)
861 * to check if %RQCF_UPADTED is set. It'll never be shifted more than
862 * one position though, because the next rq_unpin_lock() will shift it
865 #define RQCF_REQ_SKIP 0x01
866 #define RQCF_ACT_SKIP 0x02
867 #define RQCF_UPDATED 0x04
869 static inline void assert_clock_updated(struct rq
*rq
)
872 * The only reason for not seeing a clock update since the
873 * last rq_pin_lock() is if we're currently skipping updates.
875 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
878 static inline u64
rq_clock(struct rq
*rq
)
880 lockdep_assert_held(&rq
->lock
);
881 assert_clock_updated(rq
);
886 static inline u64
rq_clock_task(struct rq
*rq
)
888 lockdep_assert_held(&rq
->lock
);
889 assert_clock_updated(rq
);
891 return rq
->clock_task
;
894 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
896 lockdep_assert_held(&rq
->lock
);
898 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
900 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
905 struct pin_cookie cookie
;
906 #ifdef CONFIG_SCHED_DEBUG
908 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
909 * current pin context is stashed here in case it needs to be
910 * restored in rq_repin_lock().
912 unsigned int clock_update_flags
;
916 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
918 rf
->cookie
= lockdep_pin_lock(&rq
->lock
);
920 #ifdef CONFIG_SCHED_DEBUG
921 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
922 rf
->clock_update_flags
= 0;
926 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
928 #ifdef CONFIG_SCHED_DEBUG
929 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
930 rf
->clock_update_flags
= RQCF_UPDATED
;
933 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
936 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
938 lockdep_repin_lock(&rq
->lock
, rf
->cookie
);
940 #ifdef CONFIG_SCHED_DEBUG
942 * Restore the value we stashed in @rf for this pin context.
944 rq
->clock_update_flags
|= rf
->clock_update_flags
;
949 enum numa_topology_type
{
954 extern enum numa_topology_type sched_numa_topology_type
;
955 extern int sched_max_numa_distance
;
956 extern bool find_numa_distance(int distance
);
960 extern void sched_init_numa(void);
961 extern void sched_domains_numa_masks_set(unsigned int cpu
);
962 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
964 static inline void sched_init_numa(void) { }
965 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
966 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
969 #ifdef CONFIG_NUMA_BALANCING
970 /* The regions in numa_faults array from task_struct */
971 enum numa_faults_stats
{
977 extern void sched_setnuma(struct task_struct
*p
, int node
);
978 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
979 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
980 #endif /* CONFIG_NUMA_BALANCING */
985 queue_balance_callback(struct rq
*rq
,
986 struct callback_head
*head
,
987 void (*func
)(struct rq
*rq
))
989 lockdep_assert_held(&rq
->lock
);
991 if (unlikely(head
->next
))
994 head
->func
= (void (*)(struct callback_head
*))func
;
995 head
->next
= rq
->balance_callback
;
996 rq
->balance_callback
= head
;
999 extern void sched_ttwu_pending(void);
1001 #define rcu_dereference_check_sched_domain(p) \
1002 rcu_dereference_check((p), \
1003 lockdep_is_held(&sched_domains_mutex))
1006 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1007 * See detach_destroy_domains: synchronize_sched for details.
1009 * The domain tree of any CPU may only be accessed from within
1010 * preempt-disabled sections.
1012 #define for_each_domain(cpu, __sd) \
1013 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1014 __sd; __sd = __sd->parent)
1016 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
1019 * highest_flag_domain - Return highest sched_domain containing flag.
1020 * @cpu: The cpu whose highest level of sched domain is to
1022 * @flag: The flag to check for the highest sched_domain
1023 * for the given cpu.
1025 * Returns the highest sched_domain of a cpu which contains the given flag.
1027 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
1029 struct sched_domain
*sd
, *hsd
= NULL
;
1031 for_each_domain(cpu
, sd
) {
1032 if (!(sd
->flags
& flag
))
1040 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
1042 struct sched_domain
*sd
;
1044 for_each_domain(cpu
, sd
) {
1045 if (sd
->flags
& flag
)
1052 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
1053 DECLARE_PER_CPU(int, sd_llc_size
);
1054 DECLARE_PER_CPU(int, sd_llc_id
);
1055 DECLARE_PER_CPU(struct sched_domain_shared
*, sd_llc_shared
);
1056 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
1057 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
1059 struct sched_group_capacity
{
1062 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1065 unsigned long capacity
;
1066 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1067 unsigned long next_update
;
1068 int imbalance
; /* XXX unrelated to capacity but shared group state */
1070 #ifdef CONFIG_SCHED_DEBUG
1074 unsigned long cpumask
[0]; /* balance mask */
1077 struct sched_group
{
1078 struct sched_group
*next
; /* Must be a circular list */
1081 unsigned int group_weight
;
1082 struct sched_group_capacity
*sgc
;
1083 int asym_prefer_cpu
; /* cpu of highest priority in group */
1086 * The CPUs this group covers.
1088 * NOTE: this field is variable length. (Allocated dynamically
1089 * by attaching extra space to the end of the structure,
1090 * depending on how many CPUs the kernel has booted up with)
1092 unsigned long cpumask
[0];
1095 static inline struct cpumask
*sched_group_span(struct sched_group
*sg
)
1097 return to_cpumask(sg
->cpumask
);
1101 * See build_balance_mask().
1103 static inline struct cpumask
*group_balance_mask(struct sched_group
*sg
)
1105 return to_cpumask(sg
->sgc
->cpumask
);
1109 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
1110 * @group: The group whose first cpu is to be returned.
1112 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1114 return cpumask_first(sched_group_span(group
));
1117 extern int group_balance_cpu(struct sched_group
*sg
);
1119 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
1120 void register_sched_domain_sysctl(void);
1121 void dirty_sched_domain_sysctl(int cpu
);
1122 void unregister_sched_domain_sysctl(void);
1124 static inline void register_sched_domain_sysctl(void)
1127 static inline void dirty_sched_domain_sysctl(int cpu
)
1130 static inline void unregister_sched_domain_sysctl(void)
1137 static inline void sched_ttwu_pending(void) { }
1139 #endif /* CONFIG_SMP */
1142 #include "autogroup.h"
1144 #ifdef CONFIG_CGROUP_SCHED
1147 * Return the group to which this tasks belongs.
1149 * We cannot use task_css() and friends because the cgroup subsystem
1150 * changes that value before the cgroup_subsys::attach() method is called,
1151 * therefore we cannot pin it and might observe the wrong value.
1153 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1154 * core changes this before calling sched_move_task().
1156 * Instead we use a 'copy' which is updated from sched_move_task() while
1157 * holding both task_struct::pi_lock and rq::lock.
1159 static inline struct task_group
*task_group(struct task_struct
*p
)
1161 return p
->sched_task_group
;
1164 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1165 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1167 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1168 struct task_group
*tg
= task_group(p
);
1171 #ifdef CONFIG_FAIR_GROUP_SCHED
1172 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1173 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1174 p
->se
.parent
= tg
->se
[cpu
];
1177 #ifdef CONFIG_RT_GROUP_SCHED
1178 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1179 p
->rt
.parent
= tg
->rt_se
[cpu
];
1183 #else /* CONFIG_CGROUP_SCHED */
1185 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1186 static inline struct task_group
*task_group(struct task_struct
*p
)
1191 #endif /* CONFIG_CGROUP_SCHED */
1193 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1195 set_task_rq(p
, cpu
);
1198 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1199 * successfuly executed on another CPU. We must ensure that updates of
1200 * per-task data have been completed by this moment.
1203 #ifdef CONFIG_THREAD_INFO_IN_TASK
1206 task_thread_info(p
)->cpu
= cpu
;
1213 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1215 #ifdef CONFIG_SCHED_DEBUG
1216 # include <linux/static_key.h>
1217 # define const_debug __read_mostly
1219 # define const_debug const
1222 extern const_debug
unsigned int sysctl_sched_features
;
1224 #define SCHED_FEAT(name, enabled) \
1225 __SCHED_FEAT_##name ,
1228 #include "features.h"
1234 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1235 #define SCHED_FEAT(name, enabled) \
1236 static __always_inline bool static_branch_##name(struct static_key *key) \
1238 return static_key_##enabled(key); \
1241 #include "features.h"
1245 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1246 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1247 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1248 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1249 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1251 extern struct static_key_false sched_numa_balancing
;
1252 extern struct static_key_false sched_schedstats
;
1254 static inline u64
global_rt_period(void)
1256 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1259 static inline u64
global_rt_runtime(void)
1261 if (sysctl_sched_rt_runtime
< 0)
1264 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1267 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1269 return rq
->curr
== p
;
1272 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1277 return task_current(rq
, p
);
1281 static inline int task_on_rq_queued(struct task_struct
*p
)
1283 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1286 static inline int task_on_rq_migrating(struct task_struct
*p
)
1288 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1291 #ifndef prepare_arch_switch
1292 # define prepare_arch_switch(next) do { } while (0)
1294 #ifndef finish_arch_post_lock_switch
1295 # define finish_arch_post_lock_switch() do { } while (0)
1298 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1302 * We can optimise this out completely for !SMP, because the
1303 * SMP rebalancing from interrupt is the only thing that cares
1310 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1314 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1315 * We must ensure this doesn't happen until the switch is completely
1318 * In particular, the load of prev->state in finish_task_switch() must
1319 * happen before this.
1321 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1323 smp_store_release(&prev
->on_cpu
, 0);
1325 #ifdef CONFIG_DEBUG_SPINLOCK
1326 /* this is a valid case when another task releases the spinlock */
1327 rq
->lock
.owner
= current
;
1330 * If we are tracking spinlock dependencies then we have to
1331 * fix up the runqueue lock - which gets 'carried over' from
1332 * prev into current:
1334 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1336 raw_spin_unlock_irq(&rq
->lock
);
1342 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1343 #define WF_FORK 0x02 /* child wakeup after fork */
1344 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1347 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1348 * of tasks with abnormal "nice" values across CPUs the contribution that
1349 * each task makes to its run queue's load is weighted according to its
1350 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1351 * scaled version of the new time slice allocation that they receive on time
1355 #define WEIGHT_IDLEPRIO 3
1356 #define WMULT_IDLEPRIO 1431655765
1358 extern const int sched_prio_to_weight
[40];
1359 extern const u32 sched_prio_to_wmult
[40];
1362 * {de,en}queue flags:
1364 * DEQUEUE_SLEEP - task is no longer runnable
1365 * ENQUEUE_WAKEUP - task just became runnable
1367 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1368 * are in a known state which allows modification. Such pairs
1369 * should preserve as much state as possible.
1371 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1374 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1375 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1376 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1380 #define DEQUEUE_SLEEP 0x01
1381 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1382 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1383 #define DEQUEUE_NOCLOCK 0x08 /* matches ENQUEUE_NOCLOCK */
1385 #define ENQUEUE_WAKEUP 0x01
1386 #define ENQUEUE_RESTORE 0x02
1387 #define ENQUEUE_MOVE 0x04
1388 #define ENQUEUE_NOCLOCK 0x08
1390 #define ENQUEUE_HEAD 0x10
1391 #define ENQUEUE_REPLENISH 0x20
1393 #define ENQUEUE_MIGRATED 0x40
1395 #define ENQUEUE_MIGRATED 0x00
1398 #define RETRY_TASK ((void *)-1UL)
1400 struct sched_class
{
1401 const struct sched_class
*next
;
1403 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1404 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1405 void (*yield_task
) (struct rq
*rq
);
1406 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1408 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1411 * It is the responsibility of the pick_next_task() method that will
1412 * return the next task to call put_prev_task() on the @prev task or
1413 * something equivalent.
1415 * May return RETRY_TASK when it finds a higher prio class has runnable
1418 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1419 struct task_struct
*prev
,
1420 struct rq_flags
*rf
);
1421 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1424 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1425 void (*migrate_task_rq
)(struct task_struct
*p
);
1427 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1429 void (*set_cpus_allowed
)(struct task_struct
*p
,
1430 const struct cpumask
*newmask
);
1432 void (*rq_online
)(struct rq
*rq
);
1433 void (*rq_offline
)(struct rq
*rq
);
1436 void (*set_curr_task
) (struct rq
*rq
);
1437 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1438 void (*task_fork
) (struct task_struct
*p
);
1439 void (*task_dead
) (struct task_struct
*p
);
1442 * The switched_from() call is allowed to drop rq->lock, therefore we
1443 * cannot assume the switched_from/switched_to pair is serliazed by
1444 * rq->lock. They are however serialized by p->pi_lock.
1446 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1447 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1448 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1451 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1452 struct task_struct
*task
);
1454 void (*update_curr
) (struct rq
*rq
);
1456 #define TASK_SET_GROUP 0
1457 #define TASK_MOVE_GROUP 1
1459 #ifdef CONFIG_FAIR_GROUP_SCHED
1460 void (*task_change_group
) (struct task_struct
*p
, int type
);
1464 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1466 prev
->sched_class
->put_prev_task(rq
, prev
);
1469 static inline void set_curr_task(struct rq
*rq
, struct task_struct
*curr
)
1471 curr
->sched_class
->set_curr_task(rq
);
1475 #define sched_class_highest (&stop_sched_class)
1477 #define sched_class_highest (&dl_sched_class)
1479 #define for_each_class(class) \
1480 for (class = sched_class_highest; class; class = class->next)
1482 extern const struct sched_class stop_sched_class
;
1483 extern const struct sched_class dl_sched_class
;
1484 extern const struct sched_class rt_sched_class
;
1485 extern const struct sched_class fair_sched_class
;
1486 extern const struct sched_class idle_sched_class
;
1491 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1493 extern void trigger_load_balance(struct rq
*rq
);
1495 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1499 #ifdef CONFIG_CPU_IDLE
1500 static inline void idle_set_state(struct rq
*rq
,
1501 struct cpuidle_state
*idle_state
)
1503 rq
->idle_state
= idle_state
;
1506 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1508 SCHED_WARN_ON(!rcu_read_lock_held());
1509 return rq
->idle_state
;
1512 static inline void idle_set_state(struct rq
*rq
,
1513 struct cpuidle_state
*idle_state
)
1517 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1523 extern void schedule_idle(void);
1525 extern void sysrq_sched_debug_show(void);
1526 extern void sched_init_granularity(void);
1527 extern void update_max_interval(void);
1529 extern void init_sched_dl_class(void);
1530 extern void init_sched_rt_class(void);
1531 extern void init_sched_fair_class(void);
1533 extern void resched_curr(struct rq
*rq
);
1534 extern void resched_cpu(int cpu
);
1536 extern struct rt_bandwidth def_rt_bandwidth
;
1537 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1539 extern struct dl_bandwidth def_dl_bandwidth
;
1540 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1541 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1542 extern void init_dl_inactive_task_timer(struct sched_dl_entity
*dl_se
);
1543 extern void init_dl_rq_bw_ratio(struct dl_rq
*dl_rq
);
1546 #define BW_UNIT (1 << BW_SHIFT)
1547 #define RATIO_SHIFT 8
1548 unsigned long to_ratio(u64 period
, u64 runtime
);
1550 extern void init_entity_runnable_average(struct sched_entity
*se
);
1551 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1553 #ifdef CONFIG_NO_HZ_FULL
1554 extern bool sched_can_stop_tick(struct rq
*rq
);
1557 * Tick may be needed by tasks in the runqueue depending on their policy and
1558 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1559 * nohz mode if necessary.
1561 static inline void sched_update_tick_dependency(struct rq
*rq
)
1565 if (!tick_nohz_full_enabled())
1570 if (!tick_nohz_full_cpu(cpu
))
1573 if (sched_can_stop_tick(rq
))
1574 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1576 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1579 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1582 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1584 unsigned prev_nr
= rq
->nr_running
;
1586 rq
->nr_running
= prev_nr
+ count
;
1588 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1590 if (!rq
->rd
->overload
)
1591 rq
->rd
->overload
= true;
1595 sched_update_tick_dependency(rq
);
1598 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1600 rq
->nr_running
-= count
;
1601 /* Check if we still need preemption */
1602 sched_update_tick_dependency(rq
);
1605 static inline void rq_last_tick_reset(struct rq
*rq
)
1607 #ifdef CONFIG_NO_HZ_FULL
1608 rq
->last_sched_tick
= jiffies
;
1612 extern void update_rq_clock(struct rq
*rq
);
1614 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1615 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1617 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1619 extern const_debug
unsigned int sysctl_sched_time_avg
;
1620 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1621 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1623 static inline u64
sched_avg_period(void)
1625 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1628 #ifdef CONFIG_SCHED_HRTICK
1632 * - enabled by features
1633 * - hrtimer is actually high res
1635 static inline int hrtick_enabled(struct rq
*rq
)
1637 if (!sched_feat(HRTICK
))
1639 if (!cpu_active(cpu_of(rq
)))
1641 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1644 void hrtick_start(struct rq
*rq
, u64 delay
);
1648 static inline int hrtick_enabled(struct rq
*rq
)
1653 #endif /* CONFIG_SCHED_HRTICK */
1656 extern void sched_avg_update(struct rq
*rq
);
1658 #ifndef arch_scale_freq_capacity
1659 static __always_inline
1660 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1662 return SCHED_CAPACITY_SCALE
;
1666 #ifndef arch_scale_cpu_capacity
1667 static __always_inline
1668 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1670 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1671 return sd
->smt_gain
/ sd
->span_weight
;
1673 return SCHED_CAPACITY_SCALE
;
1677 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1679 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1680 sched_avg_update(rq
);
1683 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1684 static inline void sched_avg_update(struct rq
*rq
) { }
1687 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1688 __acquires(rq
->lock
);
1690 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1691 __acquires(p
->pi_lock
)
1692 __acquires(rq
->lock
);
1694 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1695 __releases(rq
->lock
)
1697 rq_unpin_lock(rq
, rf
);
1698 raw_spin_unlock(&rq
->lock
);
1702 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1703 __releases(rq
->lock
)
1704 __releases(p
->pi_lock
)
1706 rq_unpin_lock(rq
, rf
);
1707 raw_spin_unlock(&rq
->lock
);
1708 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1712 rq_lock_irqsave(struct rq
*rq
, struct rq_flags
*rf
)
1713 __acquires(rq
->lock
)
1715 raw_spin_lock_irqsave(&rq
->lock
, rf
->flags
);
1716 rq_pin_lock(rq
, rf
);
1720 rq_lock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1721 __acquires(rq
->lock
)
1723 raw_spin_lock_irq(&rq
->lock
);
1724 rq_pin_lock(rq
, rf
);
1728 rq_lock(struct rq
*rq
, struct rq_flags
*rf
)
1729 __acquires(rq
->lock
)
1731 raw_spin_lock(&rq
->lock
);
1732 rq_pin_lock(rq
, rf
);
1736 rq_relock(struct rq
*rq
, struct rq_flags
*rf
)
1737 __acquires(rq
->lock
)
1739 raw_spin_lock(&rq
->lock
);
1740 rq_repin_lock(rq
, rf
);
1744 rq_unlock_irqrestore(struct rq
*rq
, struct rq_flags
*rf
)
1745 __releases(rq
->lock
)
1747 rq_unpin_lock(rq
, rf
);
1748 raw_spin_unlock_irqrestore(&rq
->lock
, rf
->flags
);
1752 rq_unlock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1753 __releases(rq
->lock
)
1755 rq_unpin_lock(rq
, rf
);
1756 raw_spin_unlock_irq(&rq
->lock
);
1760 rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1761 __releases(rq
->lock
)
1763 rq_unpin_lock(rq
, rf
);
1764 raw_spin_unlock(&rq
->lock
);
1768 #ifdef CONFIG_PREEMPT
1770 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1773 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1774 * way at the expense of forcing extra atomic operations in all
1775 * invocations. This assures that the double_lock is acquired using the
1776 * same underlying policy as the spinlock_t on this architecture, which
1777 * reduces latency compared to the unfair variant below. However, it
1778 * also adds more overhead and therefore may reduce throughput.
1780 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1781 __releases(this_rq
->lock
)
1782 __acquires(busiest
->lock
)
1783 __acquires(this_rq
->lock
)
1785 raw_spin_unlock(&this_rq
->lock
);
1786 double_rq_lock(this_rq
, busiest
);
1793 * Unfair double_lock_balance: Optimizes throughput at the expense of
1794 * latency by eliminating extra atomic operations when the locks are
1795 * already in proper order on entry. This favors lower cpu-ids and will
1796 * grant the double lock to lower cpus over higher ids under contention,
1797 * regardless of entry order into the function.
1799 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1800 __releases(this_rq
->lock
)
1801 __acquires(busiest
->lock
)
1802 __acquires(this_rq
->lock
)
1806 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1807 if (busiest
< this_rq
) {
1808 raw_spin_unlock(&this_rq
->lock
);
1809 raw_spin_lock(&busiest
->lock
);
1810 raw_spin_lock_nested(&this_rq
->lock
,
1811 SINGLE_DEPTH_NESTING
);
1814 raw_spin_lock_nested(&busiest
->lock
,
1815 SINGLE_DEPTH_NESTING
);
1820 #endif /* CONFIG_PREEMPT */
1823 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1825 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1827 if (unlikely(!irqs_disabled())) {
1828 /* printk() doesn't work good under rq->lock */
1829 raw_spin_unlock(&this_rq
->lock
);
1833 return _double_lock_balance(this_rq
, busiest
);
1836 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1837 __releases(busiest
->lock
)
1839 raw_spin_unlock(&busiest
->lock
);
1840 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1843 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1849 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1852 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1858 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1861 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1867 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1871 * double_rq_lock - safely lock two runqueues
1873 * Note this does not disable interrupts like task_rq_lock,
1874 * you need to do so manually before calling.
1876 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1877 __acquires(rq1
->lock
)
1878 __acquires(rq2
->lock
)
1880 BUG_ON(!irqs_disabled());
1882 raw_spin_lock(&rq1
->lock
);
1883 __acquire(rq2
->lock
); /* Fake it out ;) */
1886 raw_spin_lock(&rq1
->lock
);
1887 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1889 raw_spin_lock(&rq2
->lock
);
1890 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1896 * double_rq_unlock - safely unlock two runqueues
1898 * Note this does not restore interrupts like task_rq_unlock,
1899 * you need to do so manually after calling.
1901 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1902 __releases(rq1
->lock
)
1903 __releases(rq2
->lock
)
1905 raw_spin_unlock(&rq1
->lock
);
1907 raw_spin_unlock(&rq2
->lock
);
1909 __release(rq2
->lock
);
1912 extern void set_rq_online (struct rq
*rq
);
1913 extern void set_rq_offline(struct rq
*rq
);
1914 extern bool sched_smp_initialized
;
1916 #else /* CONFIG_SMP */
1919 * double_rq_lock - safely lock two runqueues
1921 * Note this does not disable interrupts like task_rq_lock,
1922 * you need to do so manually before calling.
1924 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1925 __acquires(rq1
->lock
)
1926 __acquires(rq2
->lock
)
1928 BUG_ON(!irqs_disabled());
1930 raw_spin_lock(&rq1
->lock
);
1931 __acquire(rq2
->lock
); /* Fake it out ;) */
1935 * double_rq_unlock - safely unlock two runqueues
1937 * Note this does not restore interrupts like task_rq_unlock,
1938 * you need to do so manually after calling.
1940 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1941 __releases(rq1
->lock
)
1942 __releases(rq2
->lock
)
1945 raw_spin_unlock(&rq1
->lock
);
1946 __release(rq2
->lock
);
1951 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1952 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1954 #ifdef CONFIG_SCHED_DEBUG
1955 extern bool sched_debug_enabled
;
1957 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1958 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1959 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1961 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1962 #ifdef CONFIG_NUMA_BALANCING
1964 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1966 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1967 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1968 #endif /* CONFIG_NUMA_BALANCING */
1969 #endif /* CONFIG_SCHED_DEBUG */
1971 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1972 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1973 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1975 extern void cfs_bandwidth_usage_inc(void);
1976 extern void cfs_bandwidth_usage_dec(void);
1978 #ifdef CONFIG_NO_HZ_COMMON
1979 enum rq_nohz_flag_bits
{
1984 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1986 extern void nohz_balance_exit_idle(unsigned int cpu
);
1988 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1994 void __dl_update(struct dl_bw
*dl_b
, s64 bw
)
1996 struct root_domain
*rd
= container_of(dl_b
, struct root_domain
, dl_bw
);
1999 RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
2000 "sched RCU must be held");
2001 for_each_cpu_and(i
, rd
->span
, cpu_active_mask
) {
2002 struct rq
*rq
= cpu_rq(i
);
2004 rq
->dl
.extra_bw
+= bw
;
2009 void __dl_update(struct dl_bw
*dl_b
, s64 bw
)
2011 struct dl_rq
*dl
= container_of(dl_b
, struct dl_rq
, dl_bw
);
2018 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2023 struct u64_stats_sync sync
;
2026 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
2029 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2030 * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime
2031 * and never move forward.
2033 static inline u64
irq_time_read(int cpu
)
2035 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
2040 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
2041 total
= irqtime
->total
;
2042 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
2046 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2048 #ifdef CONFIG_CPU_FREQ
2049 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
2052 * cpufreq_update_util - Take a note about CPU utilization changes.
2053 * @rq: Runqueue to carry out the update for.
2054 * @flags: Update reason flags.
2056 * This function is called by the scheduler on the CPU whose utilization is
2059 * It can only be called from RCU-sched read-side critical sections.
2061 * The way cpufreq is currently arranged requires it to evaluate the CPU
2062 * performance state (frequency/voltage) on a regular basis to prevent it from
2063 * being stuck in a completely inadequate performance level for too long.
2064 * That is not guaranteed to happen if the updates are only triggered from CFS,
2065 * though, because they may not be coming in if RT or deadline tasks are active
2066 * all the time (or there are RT and DL tasks only).
2068 * As a workaround for that issue, this function is called by the RT and DL
2069 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
2070 * but that really is a band-aid. Going forward it should be replaced with
2071 * solutions targeted more specifically at RT and DL tasks.
2073 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
2075 struct update_util_data
*data
;
2077 data
= rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data
,
2080 data
->func(data
, rq_clock(rq
), flags
);
2083 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
2084 #endif /* CONFIG_CPU_FREQ */
2086 #ifdef arch_scale_freq_capacity
2087 #ifndef arch_scale_freq_invariant
2088 #define arch_scale_freq_invariant() (true)
2090 #else /* arch_scale_freq_capacity */
2091 #define arch_scale_freq_invariant() (false)