1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Scheduler internal types and methods:
5 #ifndef _KERNEL_SCHED_SCHED_H
6 #define _KERNEL_SCHED_SCHED_H
8 #include <linux/sched/affinity.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/cpufreq.h>
11 #include <linux/sched/deadline.h>
12 #include <linux/sched.h>
13 #include <linux/sched/loadavg.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/rseq_api.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/smt.h>
18 #include <linux/sched/stat.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/sched/task_flags.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/topology.h>
24 #include <linux/atomic.h>
25 #include <linux/bitmap.h>
26 #include <linux/bug.h>
27 #include <linux/capability.h>
28 #include <linux/cgroup_api.h>
29 #include <linux/cgroup.h>
30 #include <linux/cpufreq.h>
31 #include <linux/cpumask_api.h>
32 #include <linux/ctype.h>
33 #include <linux/file.h>
34 #include <linux/fs_api.h>
35 #include <linux/hrtimer_api.h>
36 #include <linux/interrupt.h>
37 #include <linux/irq_work.h>
38 #include <linux/jiffies.h>
39 #include <linux/kref_api.h>
40 #include <linux/kthread.h>
41 #include <linux/ktime_api.h>
42 #include <linux/lockdep_api.h>
43 #include <linux/lockdep.h>
44 #include <linux/minmax.h>
46 #include <linux/module.h>
47 #include <linux/mutex_api.h>
48 #include <linux/plist.h>
49 #include <linux/poll.h>
50 #include <linux/proc_fs.h>
51 #include <linux/profile.h>
52 #include <linux/psi.h>
53 #include <linux/rcupdate.h>
54 #include <linux/seq_file.h>
55 #include <linux/seqlock.h>
56 #include <linux/softirq.h>
57 #include <linux/spinlock_api.h>
58 #include <linux/static_key.h>
59 #include <linux/stop_machine.h>
60 #include <linux/syscalls_api.h>
61 #include <linux/syscalls.h>
62 #include <linux/tick.h>
63 #include <linux/topology.h>
64 #include <linux/types.h>
65 #include <linux/u64_stats_sync_api.h>
66 #include <linux/uaccess.h>
67 #include <linux/wait_api.h>
68 #include <linux/wait_bit.h>
69 #include <linux/workqueue_api.h>
71 #include <trace/events/power.h>
72 #include <trace/events/sched.h>
74 #include "../workqueue_internal.h"
76 #ifdef CONFIG_CGROUP_SCHED
77 #include <linux/cgroup.h>
78 #include <linux/psi.h>
81 #ifdef CONFIG_SCHED_DEBUG
82 # include <linux/static_key.h>
85 #ifdef CONFIG_PARAVIRT
86 # include <asm/paravirt.h>
87 # include <asm/paravirt_api_clock.h>
91 #include "cpudeadline.h"
93 #ifdef CONFIG_SCHED_DEBUG
94 # define SCHED_WARN_ON(x) WARN_ONCE(x, #x)
96 # define SCHED_WARN_ON(x) ({ (void)(x), 0; })
100 struct cpuidle_state
;
102 /* task_struct::on_rq states: */
103 #define TASK_ON_RQ_QUEUED 1
104 #define TASK_ON_RQ_MIGRATING 2
106 extern __read_mostly
int scheduler_running
;
108 extern unsigned long calc_load_update
;
109 extern atomic_long_t calc_load_tasks
;
111 extern void calc_global_load_tick(struct rq
*this_rq
);
112 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
114 extern void call_trace_sched_update_nr_running(struct rq
*rq
, int count
);
116 * Helpers for converting nanosecond timing to jiffy resolution
118 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
121 * Increase resolution of nice-level calculations for 64-bit architectures.
122 * The extra resolution improves shares distribution and load balancing of
123 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
124 * hierarchies, especially on larger systems. This is not a user-visible change
125 * and does not change the user-interface for setting shares/weights.
127 * We increase resolution only if we have enough bits to allow this increased
128 * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit
129 * are pretty high and the returns do not justify the increased costs.
131 * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to
132 * increase coverage and consistency always enable it on 64-bit platforms.
135 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
136 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
137 # define scale_load_down(w) \
139 unsigned long __w = (w); \
141 __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
145 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
146 # define scale_load(w) (w)
147 # define scale_load_down(w) (w)
151 * Task weight (visible to users) and its load (invisible to users) have
152 * independent resolution, but they should be well calibrated. We use
153 * scale_load() and scale_load_down(w) to convert between them. The
154 * following must be true:
156 * scale_load(sched_prio_to_weight[NICE_TO_PRIO(0)-MAX_RT_PRIO]) == NICE_0_LOAD
159 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
162 * Single value that decides SCHED_DEADLINE internal math precision.
163 * 10 -> just above 1us
164 * 9 -> just above 0.5us
169 * Single value that denotes runtime == period, ie unlimited time.
171 #define RUNTIME_INF ((u64)~0ULL)
173 static inline int idle_policy(int policy
)
175 return policy
== SCHED_IDLE
;
177 static inline int fair_policy(int policy
)
179 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
182 static inline int rt_policy(int policy
)
184 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
187 static inline int dl_policy(int policy
)
189 return policy
== SCHED_DEADLINE
;
191 static inline bool valid_policy(int policy
)
193 return idle_policy(policy
) || fair_policy(policy
) ||
194 rt_policy(policy
) || dl_policy(policy
);
197 static inline int task_has_idle_policy(struct task_struct
*p
)
199 return idle_policy(p
->policy
);
202 static inline int task_has_rt_policy(struct task_struct
*p
)
204 return rt_policy(p
->policy
);
207 static inline int task_has_dl_policy(struct task_struct
*p
)
209 return dl_policy(p
->policy
);
212 #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
214 static inline void update_avg(u64
*avg
, u64 sample
)
216 s64 diff
= sample
- *avg
;
221 * Shifting a value by an exponent greater *or equal* to the size of said value
222 * is UB; cap at size-1.
224 #define shr_bound(val, shift) \
225 (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1))
228 * !! For sched_setattr_nocheck() (kernel) only !!
230 * This is actually gross. :(
232 * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE
233 * tasks, but still be able to sleep. We need this on platforms that cannot
234 * atomically change clock frequency. Remove once fast switching will be
235 * available on such platforms.
237 * SUGOV stands for SchedUtil GOVernor.
239 #define SCHED_FLAG_SUGOV 0x10000000
241 #define SCHED_DL_FLAGS (SCHED_FLAG_RECLAIM | SCHED_FLAG_DL_OVERRUN | SCHED_FLAG_SUGOV)
243 static inline bool dl_entity_is_special(struct sched_dl_entity
*dl_se
)
245 #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL
246 return unlikely(dl_se
->flags
& SCHED_FLAG_SUGOV
);
253 * Tells if entity @a should preempt entity @b.
256 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
258 return dl_entity_is_special(a
) ||
259 dl_time_before(a
->deadline
, b
->deadline
);
263 * This is the priority-queue data structure of the RT scheduling class:
265 struct rt_prio_array
{
266 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
267 struct list_head queue
[MAX_RT_PRIO
];
270 struct rt_bandwidth
{
271 /* nests inside the rq lock: */
272 raw_spinlock_t rt_runtime_lock
;
275 struct hrtimer rt_period_timer
;
276 unsigned int rt_period_active
;
279 void __dl_clear_params(struct task_struct
*p
);
281 struct dl_bandwidth
{
282 raw_spinlock_t dl_runtime_lock
;
287 static inline int dl_bandwidth_enabled(void)
289 return sysctl_sched_rt_runtime
>= 0;
293 * To keep the bandwidth of -deadline tasks under control
294 * we need some place where:
295 * - store the maximum -deadline bandwidth of each cpu;
296 * - cache the fraction of bandwidth that is currently allocated in
299 * This is all done in the data structure below. It is similar to the
300 * one used for RT-throttling (rt_bandwidth), with the main difference
301 * that, since here we are only interested in admission control, we
302 * do not decrease any runtime while the group "executes", neither we
303 * need a timer to replenish it.
305 * With respect to SMP, bandwidth is given on a per root domain basis,
307 * - bw (< 100%) is the deadline bandwidth of each CPU;
308 * - total_bw is the currently allocated bandwidth in each root domain;
317 * Verify the fitness of task @p to run on @cpu taking into account the
318 * CPU original capacity and the runtime/deadline ratio of the task.
320 * The function will return true if the CPU original capacity of the
321 * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the
322 * task and false otherwise.
324 static inline bool dl_task_fits_capacity(struct task_struct
*p
, int cpu
)
326 unsigned long cap
= arch_scale_cpu_capacity(cpu
);
328 return cap_scale(p
->dl
.dl_deadline
, cap
) >= p
->dl
.dl_runtime
;
331 extern void init_dl_bw(struct dl_bw
*dl_b
);
332 extern int sched_dl_global_validate(void);
333 extern void sched_dl_do_global(void);
334 extern int sched_dl_overflow(struct task_struct
*p
, int policy
, const struct sched_attr
*attr
);
335 extern void __setparam_dl(struct task_struct
*p
, const struct sched_attr
*attr
);
336 extern void __getparam_dl(struct task_struct
*p
, struct sched_attr
*attr
);
337 extern bool __checkparam_dl(const struct sched_attr
*attr
);
338 extern bool dl_param_changed(struct task_struct
*p
, const struct sched_attr
*attr
);
339 extern int dl_cpuset_cpumask_can_shrink(const struct cpumask
*cur
, const struct cpumask
*trial
);
340 extern int dl_cpu_busy(int cpu
, struct task_struct
*p
);
342 #ifdef CONFIG_CGROUP_SCHED
347 extern struct list_head task_groups
;
349 struct cfs_bandwidth
{
350 #ifdef CONFIG_CFS_BANDWIDTH
357 s64 hierarchical_quota
;
362 struct hrtimer period_timer
;
363 struct hrtimer slack_timer
;
364 struct list_head throttled_cfs_rq
;
375 /* Task group related information */
377 struct cgroup_subsys_state css
;
379 #ifdef CONFIG_FAIR_GROUP_SCHED
380 /* schedulable entities of this group on each CPU */
381 struct sched_entity
**se
;
382 /* runqueue "owned" by this group on each CPU */
383 struct cfs_rq
**cfs_rq
;
384 unsigned long shares
;
386 /* A positive value indicates that this is a SCHED_IDLE group. */
391 * load_avg can be heavily contended at clock tick time, so put
392 * it in its own cacheline separated from the fields above which
393 * will also be accessed at each tick.
395 atomic_long_t load_avg ____cacheline_aligned
;
399 #ifdef CONFIG_RT_GROUP_SCHED
400 struct sched_rt_entity
**rt_se
;
401 struct rt_rq
**rt_rq
;
403 struct rt_bandwidth rt_bandwidth
;
407 struct list_head list
;
409 struct task_group
*parent
;
410 struct list_head siblings
;
411 struct list_head children
;
413 #ifdef CONFIG_SCHED_AUTOGROUP
414 struct autogroup
*autogroup
;
417 struct cfs_bandwidth cfs_bandwidth
;
419 #ifdef CONFIG_UCLAMP_TASK_GROUP
420 /* The two decimal precision [%] value requested from user-space */
421 unsigned int uclamp_pct
[UCLAMP_CNT
];
422 /* Clamp values requested for a task group */
423 struct uclamp_se uclamp_req
[UCLAMP_CNT
];
424 /* Effective clamp values used for a task group */
425 struct uclamp_se uclamp
[UCLAMP_CNT
];
430 #ifdef CONFIG_FAIR_GROUP_SCHED
431 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
434 * A weight of 0 or 1 can cause arithmetics problems.
435 * A weight of a cfs_rq is the sum of weights of which entities
436 * are queued on this cfs_rq, so a weight of a entity should not be
437 * too large, so as the shares value of a task group.
438 * (The default weight is 1024 - so there's no practical
439 * limitation from this.)
441 #define MIN_SHARES (1UL << 1)
442 #define MAX_SHARES (1UL << 18)
445 typedef int (*tg_visitor
)(struct task_group
*, void *);
447 extern int walk_tg_tree_from(struct task_group
*from
,
448 tg_visitor down
, tg_visitor up
, void *data
);
451 * Iterate the full tree, calling @down when first entering a node and @up when
452 * leaving it for the final time.
454 * Caller must hold rcu_lock or sufficient equivalent.
456 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
458 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
461 extern int tg_nop(struct task_group
*tg
, void *data
);
463 extern void free_fair_sched_group(struct task_group
*tg
);
464 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
465 extern void online_fair_sched_group(struct task_group
*tg
);
466 extern void unregister_fair_sched_group(struct task_group
*tg
);
467 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
468 struct sched_entity
*se
, int cpu
,
469 struct sched_entity
*parent
);
470 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
472 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
473 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
474 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
476 extern void unregister_rt_sched_group(struct task_group
*tg
);
477 extern void free_rt_sched_group(struct task_group
*tg
);
478 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
479 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
480 struct sched_rt_entity
*rt_se
, int cpu
,
481 struct sched_rt_entity
*parent
);
482 extern int sched_group_set_rt_runtime(struct task_group
*tg
, long rt_runtime_us
);
483 extern int sched_group_set_rt_period(struct task_group
*tg
, u64 rt_period_us
);
484 extern long sched_group_rt_runtime(struct task_group
*tg
);
485 extern long sched_group_rt_period(struct task_group
*tg
);
486 extern int sched_rt_can_attach(struct task_group
*tg
, struct task_struct
*tsk
);
488 extern struct task_group
*sched_create_group(struct task_group
*parent
);
489 extern void sched_online_group(struct task_group
*tg
,
490 struct task_group
*parent
);
491 extern void sched_destroy_group(struct task_group
*tg
);
492 extern void sched_release_group(struct task_group
*tg
);
494 extern void sched_move_task(struct task_struct
*tsk
);
496 #ifdef CONFIG_FAIR_GROUP_SCHED
497 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
499 extern int sched_group_set_idle(struct task_group
*tg
, long idle
);
502 extern void set_task_rq_fair(struct sched_entity
*se
,
503 struct cfs_rq
*prev
, struct cfs_rq
*next
);
504 #else /* !CONFIG_SMP */
505 static inline void set_task_rq_fair(struct sched_entity
*se
,
506 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
507 #endif /* CONFIG_SMP */
508 #endif /* CONFIG_FAIR_GROUP_SCHED */
510 #else /* CONFIG_CGROUP_SCHED */
512 struct cfs_bandwidth
{ };
514 #endif /* CONFIG_CGROUP_SCHED */
516 /* CFS-related fields in a runqueue */
518 struct load_weight load
;
519 unsigned int nr_running
;
520 unsigned int h_nr_running
; /* SCHED_{NORMAL,BATCH,IDLE} */
521 unsigned int idle_nr_running
; /* SCHED_IDLE */
522 unsigned int idle_h_nr_running
; /* SCHED_IDLE */
526 #ifdef CONFIG_SCHED_CORE
527 unsigned int forceidle_seq
;
532 u64 min_vruntime_copy
;
535 struct rb_root_cached tasks_timeline
;
538 * 'curr' points to currently running entity on this cfs_rq.
539 * It is set to NULL otherwise (i.e when none are currently running).
541 struct sched_entity
*curr
;
542 struct sched_entity
*next
;
543 struct sched_entity
*last
;
544 struct sched_entity
*skip
;
546 #ifdef CONFIG_SCHED_DEBUG
547 unsigned int nr_spread_over
;
554 struct sched_avg avg
;
556 u64 load_last_update_time_copy
;
559 raw_spinlock_t lock ____cacheline_aligned
;
561 unsigned long load_avg
;
562 unsigned long util_avg
;
563 unsigned long runnable_avg
;
566 #ifdef CONFIG_FAIR_GROUP_SCHED
567 unsigned long tg_load_avg_contrib
;
569 long prop_runnable_sum
;
572 * h_load = weight * f(tg)
574 * Where f(tg) is the recursive weight fraction assigned to
577 unsigned long h_load
;
578 u64 last_h_load_update
;
579 struct sched_entity
*h_load_next
;
580 #endif /* CONFIG_FAIR_GROUP_SCHED */
581 #endif /* CONFIG_SMP */
583 #ifdef CONFIG_FAIR_GROUP_SCHED
584 struct rq
*rq
; /* CPU runqueue to which this cfs_rq is attached */
587 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
588 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
589 * (like users, containers etc.)
591 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU.
592 * This list is used during load balance.
595 struct list_head leaf_cfs_rq_list
;
596 struct task_group
*tg
; /* group that "owns" this runqueue */
598 /* Locally cached copy of our task_group's idle value */
601 #ifdef CONFIG_CFS_BANDWIDTH
603 s64 runtime_remaining
;
606 u64 throttled_clock_task
;
607 u64 throttled_clock_task_time
;
610 struct list_head throttled_list
;
611 #endif /* CONFIG_CFS_BANDWIDTH */
612 #endif /* CONFIG_FAIR_GROUP_SCHED */
615 static inline int rt_bandwidth_enabled(void)
617 return sysctl_sched_rt_runtime
>= 0;
620 /* RT IPI pull logic requires IRQ_WORK */
621 #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP)
622 # define HAVE_RT_PUSH_IPI
625 /* Real-Time classes' related field in a runqueue: */
627 struct rt_prio_array active
;
628 unsigned int rt_nr_running
;
629 unsigned int rr_nr_running
;
630 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
632 int curr
; /* highest queued rt task prio */
634 int next
; /* next highest */
639 unsigned int rt_nr_migratory
;
640 unsigned int rt_nr_total
;
642 struct plist_head pushable_tasks
;
644 #endif /* CONFIG_SMP */
650 /* Nests inside the rq lock: */
651 raw_spinlock_t rt_runtime_lock
;
653 #ifdef CONFIG_RT_GROUP_SCHED
654 unsigned int rt_nr_boosted
;
657 struct task_group
*tg
;
661 static inline bool rt_rq_is_runnable(struct rt_rq
*rt_rq
)
663 return rt_rq
->rt_queued
&& rt_rq
->rt_nr_running
;
666 /* Deadline class' related fields in a runqueue */
668 /* runqueue is an rbtree, ordered by deadline */
669 struct rb_root_cached root
;
671 unsigned int dl_nr_running
;
675 * Deadline values of the currently executing and the
676 * earliest ready task on this rq. Caching these facilitates
677 * the decision whether or not a ready but not running task
678 * should migrate somewhere else.
685 unsigned int dl_nr_migratory
;
689 * Tasks on this rq that can be pushed away. They are kept in
690 * an rb-tree, ordered by tasks' deadlines, with caching
691 * of the leftmost (earliest deadline) element.
693 struct rb_root_cached pushable_dl_tasks_root
;
698 * "Active utilization" for this runqueue: increased when a
699 * task wakes up (becomes TASK_RUNNING) and decreased when a
705 * Utilization of the tasks "assigned" to this runqueue (including
706 * the tasks that are in runqueue and the tasks that executed on this
707 * CPU and blocked). Increased when a task moves to this runqueue, and
708 * decreased when the task moves away (migrates, changes scheduling
709 * policy, or terminates).
710 * This is needed to compute the "inactive utilization" for the
711 * runqueue (inactive utilization = this_bw - running_bw).
717 * Inverse of the fraction of CPU utilization that can be reclaimed
718 * by the GRUB algorithm.
723 #ifdef CONFIG_FAIR_GROUP_SCHED
724 /* An entity is a task if it doesn't "own" a runqueue */
725 #define entity_is_task(se) (!se->my_q)
727 static inline void se_update_runnable(struct sched_entity
*se
)
729 if (!entity_is_task(se
))
730 se
->runnable_weight
= se
->my_q
->h_nr_running
;
733 static inline long se_runnable(struct sched_entity
*se
)
735 if (entity_is_task(se
))
738 return se
->runnable_weight
;
742 #define entity_is_task(se) 1
744 static inline void se_update_runnable(struct sched_entity
*se
) {}
746 static inline long se_runnable(struct sched_entity
*se
)
754 * XXX we want to get rid of these helpers and use the full load resolution.
756 static inline long se_weight(struct sched_entity
*se
)
758 return scale_load_down(se
->load
.weight
);
762 static inline bool sched_asym_prefer(int a
, int b
)
764 return arch_asym_cpu_priority(a
) > arch_asym_cpu_priority(b
);
768 struct em_perf_domain
*em_pd
;
769 struct perf_domain
*next
;
773 /* Scheduling group status flags */
774 #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */
775 #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */
778 * We add the notion of a root-domain which will be used to define per-domain
779 * variables. Each exclusive cpuset essentially defines an island domain by
780 * fully partitioning the member CPUs from any other cpuset. Whenever a new
781 * exclusive cpuset is created, we also create and attach a new root-domain
790 cpumask_var_t online
;
793 * Indicate pullable load on at least one CPU, e.g:
794 * - More than one runnable task
795 * - Running task is misfit
799 /* Indicate one or more cpus over-utilized (tipping point) */
803 * The bit corresponding to a CPU gets set here if such CPU has more
804 * than one runnable -deadline task (as it is below for RT tasks).
806 cpumask_var_t dlo_mask
;
812 * Indicate whether a root_domain's dl_bw has been checked or
813 * updated. It's monotonously increasing value.
815 * Also, some corner cases, like 'wrap around' is dangerous, but given
816 * that u64 is 'big enough'. So that shouldn't be a concern.
820 #ifdef HAVE_RT_PUSH_IPI
822 * For IPI pull requests, loop across the rto_mask.
824 struct irq_work rto_push_work
;
825 raw_spinlock_t rto_lock
;
826 /* These are only updated and read within rto_lock */
829 /* These atomics are updated outside of a lock */
830 atomic_t rto_loop_next
;
831 atomic_t rto_loop_start
;
834 * The "RT overload" flag: it gets set if a CPU has more than
835 * one runnable RT task.
837 cpumask_var_t rto_mask
;
838 struct cpupri cpupri
;
840 unsigned long max_cpu_capacity
;
843 * NULL-terminated list of performance domains intersecting with the
844 * CPUs of the rd. Protected by RCU.
846 struct perf_domain __rcu
*pd
;
849 extern void init_defrootdomain(void);
850 extern int sched_init_domains(const struct cpumask
*cpu_map
);
851 extern void rq_attach_root(struct rq
*rq
, struct root_domain
*rd
);
852 extern void sched_get_rd(struct root_domain
*rd
);
853 extern void sched_put_rd(struct root_domain
*rd
);
855 #ifdef HAVE_RT_PUSH_IPI
856 extern void rto_push_irq_work_func(struct irq_work
*work
);
858 #endif /* CONFIG_SMP */
860 #ifdef CONFIG_UCLAMP_TASK
862 * struct uclamp_bucket - Utilization clamp bucket
863 * @value: utilization clamp value for tasks on this clamp bucket
864 * @tasks: number of RUNNABLE tasks on this clamp bucket
866 * Keep track of how many tasks are RUNNABLE for a given utilization
869 struct uclamp_bucket
{
870 unsigned long value
: bits_per(SCHED_CAPACITY_SCALE
);
871 unsigned long tasks
: BITS_PER_LONG
- bits_per(SCHED_CAPACITY_SCALE
);
875 * struct uclamp_rq - rq's utilization clamp
876 * @value: currently active clamp values for a rq
877 * @bucket: utilization clamp buckets affecting a rq
879 * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values.
880 * A clamp value is affecting a rq when there is at least one task RUNNABLE
881 * (or actually running) with that value.
883 * There are up to UCLAMP_CNT possible different clamp values, currently there
884 * are only two: minimum utilization and maximum utilization.
886 * All utilization clamping values are MAX aggregated, since:
887 * - for util_min: we want to run the CPU at least at the max of the minimum
888 * utilization required by its currently RUNNABLE tasks.
889 * - for util_max: we want to allow the CPU to run up to the max of the
890 * maximum utilization allowed by its currently RUNNABLE tasks.
892 * Since on each system we expect only a limited number of different
893 * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track
894 * the metrics required to compute all the per-rq utilization clamp values.
898 struct uclamp_bucket bucket
[UCLAMP_BUCKETS
];
901 DECLARE_STATIC_KEY_FALSE(sched_uclamp_used
);
902 #endif /* CONFIG_UCLAMP_TASK */
905 * This is the main, per-CPU runqueue data structure.
907 * Locking rule: those places that want to lock multiple runqueues
908 * (such as the load balancing or the thread migration code), lock
909 * acquire operations must be ordered by ascending &runqueue.
913 raw_spinlock_t __lock
;
916 * nr_running and cpu_load should be in the same cacheline because
917 * remote CPUs use both these fields when doing load calculation.
919 unsigned int nr_running
;
920 #ifdef CONFIG_NUMA_BALANCING
921 unsigned int nr_numa_running
;
922 unsigned int nr_preferred_running
;
923 unsigned int numa_migrate_on
;
925 #ifdef CONFIG_NO_HZ_COMMON
927 unsigned long last_blocked_load_update_tick
;
928 unsigned int has_blocked_load
;
929 call_single_data_t nohz_csd
;
930 #endif /* CONFIG_SMP */
931 unsigned int nohz_tick_stopped
;
933 #endif /* CONFIG_NO_HZ_COMMON */
936 unsigned int ttwu_pending
;
940 #ifdef CONFIG_UCLAMP_TASK
941 /* Utilization clamp values based on CPU's RUNNABLE tasks */
942 struct uclamp_rq uclamp
[UCLAMP_CNT
] ____cacheline_aligned
;
943 unsigned int uclamp_flags
;
944 #define UCLAMP_FLAG_IDLE 0x01
951 #ifdef CONFIG_FAIR_GROUP_SCHED
952 /* list of leaf cfs_rq on this CPU: */
953 struct list_head leaf_cfs_rq_list
;
954 struct list_head
*tmp_alone_branch
;
955 #endif /* CONFIG_FAIR_GROUP_SCHED */
958 * This is part of a global counter where only the total sum
959 * over all CPUs matters. A task can increase this counter on
960 * one CPU and if it got migrated afterwards it may decrease
961 * it on another CPU. Always updated under the runqueue lock:
963 unsigned int nr_uninterruptible
;
965 struct task_struct __rcu
*curr
;
966 struct task_struct
*idle
;
967 struct task_struct
*stop
;
968 unsigned long next_balance
;
969 struct mm_struct
*prev_mm
;
971 unsigned int clock_update_flags
;
973 /* Ensure that all clocks are in the same cache line */
974 u64 clock_task ____cacheline_aligned
;
976 unsigned long lost_idle_time
;
980 #ifdef CONFIG_SCHED_DEBUG
981 u64 last_seen_need_resched_ns
;
982 int ticks_without_resched
;
985 #ifdef CONFIG_MEMBARRIER
986 int membarrier_state
;
990 struct root_domain
*rd
;
991 struct sched_domain __rcu
*sd
;
993 unsigned long cpu_capacity
;
994 unsigned long cpu_capacity_orig
;
996 struct callback_head
*balance_callback
;
998 unsigned char nohz_idle_balance
;
999 unsigned char idle_balance
;
1001 unsigned long misfit_task_load
;
1003 /* For active balancing */
1006 struct cpu_stop_work active_balance_work
;
1008 /* CPU of this runqueue: */
1012 struct list_head cfs_tasks
;
1014 struct sched_avg avg_rt
;
1015 struct sched_avg avg_dl
;
1016 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
1017 struct sched_avg avg_irq
;
1019 #ifdef CONFIG_SCHED_THERMAL_PRESSURE
1020 struct sched_avg avg_thermal
;
1025 unsigned long wake_stamp
;
1028 /* This is used to determine avg_idle's max value */
1029 u64 max_idle_balance_cost
;
1031 #ifdef CONFIG_HOTPLUG_CPU
1032 struct rcuwait hotplug_wait
;
1034 #endif /* CONFIG_SMP */
1036 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1039 #ifdef CONFIG_PARAVIRT
1040 u64 prev_steal_time
;
1042 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
1043 u64 prev_steal_time_rq
;
1046 /* calc_load related fields */
1047 unsigned long calc_load_update
;
1048 long calc_load_active
;
1050 #ifdef CONFIG_SCHED_HRTICK
1052 call_single_data_t hrtick_csd
;
1054 struct hrtimer hrtick_timer
;
1055 ktime_t hrtick_time
;
1058 #ifdef CONFIG_SCHEDSTATS
1060 struct sched_info rq_sched_info
;
1061 unsigned long long rq_cpu_time
;
1062 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1064 /* sys_sched_yield() stats */
1065 unsigned int yld_count
;
1067 /* schedule() stats */
1068 unsigned int sched_count
;
1069 unsigned int sched_goidle
;
1071 /* try_to_wake_up() stats */
1072 unsigned int ttwu_count
;
1073 unsigned int ttwu_local
;
1076 #ifdef CONFIG_CPU_IDLE
1077 /* Must be inspected within a rcu lock section */
1078 struct cpuidle_state
*idle_state
;
1082 unsigned int nr_pinned
;
1084 unsigned int push_busy
;
1085 struct cpu_stop_work push_work
;
1087 #ifdef CONFIG_SCHED_CORE
1090 struct task_struct
*core_pick
;
1091 unsigned int core_enabled
;
1092 unsigned int core_sched_seq
;
1093 struct rb_root core_tree
;
1095 /* shared state -- careful with sched_core_cpu_deactivate() */
1096 unsigned int core_task_seq
;
1097 unsigned int core_pick_seq
;
1098 unsigned long core_cookie
;
1099 unsigned int core_forceidle_count
;
1100 unsigned int core_forceidle_seq
;
1101 unsigned int core_forceidle_occupation
;
1102 u64 core_forceidle_start
;
1106 #ifdef CONFIG_FAIR_GROUP_SCHED
1108 /* CPU runqueue to which this cfs_rq is attached */
1109 static inline struct rq
*rq_of(struct cfs_rq
*cfs_rq
)
1116 static inline struct rq
*rq_of(struct cfs_rq
*cfs_rq
)
1118 return container_of(cfs_rq
, struct rq
, cfs
);
1122 static inline int cpu_of(struct rq
*rq
)
1131 #define MDF_PUSH 0x01
1133 static inline bool is_migration_disabled(struct task_struct
*p
)
1136 return p
->migration_disabled
;
1143 #ifdef CONFIG_SCHED_CORE
1144 static inline struct cpumask
*sched_group_span(struct sched_group
*sg
);
1146 DECLARE_STATIC_KEY_FALSE(__sched_core_enabled
);
1148 static inline bool sched_core_enabled(struct rq
*rq
)
1150 return static_branch_unlikely(&__sched_core_enabled
) && rq
->core_enabled
;
1153 static inline bool sched_core_disabled(void)
1155 return !static_branch_unlikely(&__sched_core_enabled
);
1159 * Be careful with this function; not for general use. The return value isn't
1160 * stable unless you actually hold a relevant rq->__lock.
1162 static inline raw_spinlock_t
*rq_lockp(struct rq
*rq
)
1164 if (sched_core_enabled(rq
))
1165 return &rq
->core
->__lock
;
1170 static inline raw_spinlock_t
*__rq_lockp(struct rq
*rq
)
1172 if (rq
->core_enabled
)
1173 return &rq
->core
->__lock
;
1178 bool cfs_prio_less(struct task_struct
*a
, struct task_struct
*b
, bool fi
);
1181 * Helpers to check if the CPU's core cookie matches with the task's cookie
1182 * when core scheduling is enabled.
1183 * A special case is that the task's cookie always matches with CPU's core
1184 * cookie if the CPU is in an idle core.
1186 static inline bool sched_cpu_cookie_match(struct rq
*rq
, struct task_struct
*p
)
1188 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1189 if (!sched_core_enabled(rq
))
1192 return rq
->core
->core_cookie
== p
->core_cookie
;
1195 static inline bool sched_core_cookie_match(struct rq
*rq
, struct task_struct
*p
)
1197 bool idle_core
= true;
1200 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1201 if (!sched_core_enabled(rq
))
1204 for_each_cpu(cpu
, cpu_smt_mask(cpu_of(rq
))) {
1205 if (!available_idle_cpu(cpu
)) {
1212 * A CPU in an idle core is always the best choice for tasks with
1215 return idle_core
|| rq
->core
->core_cookie
== p
->core_cookie
;
1218 static inline bool sched_group_cookie_match(struct rq
*rq
,
1219 struct task_struct
*p
,
1220 struct sched_group
*group
)
1224 /* Ignore cookie match if core scheduler is not enabled on the CPU. */
1225 if (!sched_core_enabled(rq
))
1228 for_each_cpu_and(cpu
, sched_group_span(group
), p
->cpus_ptr
) {
1229 if (sched_core_cookie_match(rq
, p
))
1235 extern void queue_core_balance(struct rq
*rq
);
1237 static inline bool sched_core_enqueued(struct task_struct
*p
)
1239 return !RB_EMPTY_NODE(&p
->core_node
);
1242 extern void sched_core_enqueue(struct rq
*rq
, struct task_struct
*p
);
1243 extern void sched_core_dequeue(struct rq
*rq
, struct task_struct
*p
, int flags
);
1245 extern void sched_core_get(void);
1246 extern void sched_core_put(void);
1248 #else /* !CONFIG_SCHED_CORE */
1250 static inline bool sched_core_enabled(struct rq
*rq
)
1255 static inline bool sched_core_disabled(void)
1260 static inline raw_spinlock_t
*rq_lockp(struct rq
*rq
)
1265 static inline raw_spinlock_t
*__rq_lockp(struct rq
*rq
)
1270 static inline void queue_core_balance(struct rq
*rq
)
1274 static inline bool sched_cpu_cookie_match(struct rq
*rq
, struct task_struct
*p
)
1279 static inline bool sched_core_cookie_match(struct rq
*rq
, struct task_struct
*p
)
1284 static inline bool sched_group_cookie_match(struct rq
*rq
,
1285 struct task_struct
*p
,
1286 struct sched_group
*group
)
1290 #endif /* CONFIG_SCHED_CORE */
1292 static inline void lockdep_assert_rq_held(struct rq
*rq
)
1294 lockdep_assert_held(__rq_lockp(rq
));
1297 extern void raw_spin_rq_lock_nested(struct rq
*rq
, int subclass
);
1298 extern bool raw_spin_rq_trylock(struct rq
*rq
);
1299 extern void raw_spin_rq_unlock(struct rq
*rq
);
1301 static inline void raw_spin_rq_lock(struct rq
*rq
)
1303 raw_spin_rq_lock_nested(rq
, 0);
1306 static inline void raw_spin_rq_lock_irq(struct rq
*rq
)
1308 local_irq_disable();
1309 raw_spin_rq_lock(rq
);
1312 static inline void raw_spin_rq_unlock_irq(struct rq
*rq
)
1314 raw_spin_rq_unlock(rq
);
1318 static inline unsigned long _raw_spin_rq_lock_irqsave(struct rq
*rq
)
1320 unsigned long flags
;
1321 local_irq_save(flags
);
1322 raw_spin_rq_lock(rq
);
1326 static inline void raw_spin_rq_unlock_irqrestore(struct rq
*rq
, unsigned long flags
)
1328 raw_spin_rq_unlock(rq
);
1329 local_irq_restore(flags
);
1332 #define raw_spin_rq_lock_irqsave(rq, flags) \
1334 flags = _raw_spin_rq_lock_irqsave(rq); \
1337 #ifdef CONFIG_SCHED_SMT
1338 extern void __update_idle_core(struct rq
*rq
);
1340 static inline void update_idle_core(struct rq
*rq
)
1342 if (static_branch_unlikely(&sched_smt_present
))
1343 __update_idle_core(rq
);
1347 static inline void update_idle_core(struct rq
*rq
) { }
1350 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
1352 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
1353 #define this_rq() this_cpu_ptr(&runqueues)
1354 #define task_rq(p) cpu_rq(task_cpu(p))
1355 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
1356 #define raw_rq() raw_cpu_ptr(&runqueues)
1358 #ifdef CONFIG_FAIR_GROUP_SCHED
1359 static inline struct task_struct
*task_of(struct sched_entity
*se
)
1361 SCHED_WARN_ON(!entity_is_task(se
));
1362 return container_of(se
, struct task_struct
, se
);
1365 static inline struct cfs_rq
*task_cfs_rq(struct task_struct
*p
)
1367 return p
->se
.cfs_rq
;
1370 /* runqueue on which this entity is (to be) queued */
1371 static inline struct cfs_rq
*cfs_rq_of(struct sched_entity
*se
)
1376 /* runqueue "owned" by this group */
1377 static inline struct cfs_rq
*group_cfs_rq(struct sched_entity
*grp
)
1384 static inline struct task_struct
*task_of(struct sched_entity
*se
)
1386 return container_of(se
, struct task_struct
, se
);
1389 static inline struct cfs_rq
*task_cfs_rq(struct task_struct
*p
)
1391 return &task_rq(p
)->cfs
;
1394 static inline struct cfs_rq
*cfs_rq_of(struct sched_entity
*se
)
1396 struct task_struct
*p
= task_of(se
);
1397 struct rq
*rq
= task_rq(p
);
1402 /* runqueue "owned" by this group */
1403 static inline struct cfs_rq
*group_cfs_rq(struct sched_entity
*grp
)
1409 extern void update_rq_clock(struct rq
*rq
);
1412 * rq::clock_update_flags bits
1414 * %RQCF_REQ_SKIP - will request skipping of clock update on the next
1415 * call to __schedule(). This is an optimisation to avoid
1416 * neighbouring rq clock updates.
1418 * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is
1419 * in effect and calls to update_rq_clock() are being ignored.
1421 * %RQCF_UPDATED - is a debug flag that indicates whether a call has been
1422 * made to update_rq_clock() since the last time rq::lock was pinned.
1424 * If inside of __schedule(), clock_update_flags will have been
1425 * shifted left (a left shift is a cheap operation for the fast path
1426 * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use,
1428 * if (rq-clock_update_flags >= RQCF_UPDATED)
1430 * to check if %RQCF_UPDATED is set. It'll never be shifted more than
1431 * one position though, because the next rq_unpin_lock() will shift it
1434 #define RQCF_REQ_SKIP 0x01
1435 #define RQCF_ACT_SKIP 0x02
1436 #define RQCF_UPDATED 0x04
1438 static inline void assert_clock_updated(struct rq
*rq
)
1441 * The only reason for not seeing a clock update since the
1442 * last rq_pin_lock() is if we're currently skipping updates.
1444 SCHED_WARN_ON(rq
->clock_update_flags
< RQCF_ACT_SKIP
);
1447 static inline u64
rq_clock(struct rq
*rq
)
1449 lockdep_assert_rq_held(rq
);
1450 assert_clock_updated(rq
);
1455 static inline u64
rq_clock_task(struct rq
*rq
)
1457 lockdep_assert_rq_held(rq
);
1458 assert_clock_updated(rq
);
1460 return rq
->clock_task
;
1464 * By default the decay is the default pelt decay period.
1465 * The decay shift can change the decay period in
1467 * Decay shift Decay period(ms)
1474 extern int sched_thermal_decay_shift
;
1476 static inline u64
rq_clock_thermal(struct rq
*rq
)
1478 return rq_clock_task(rq
) >> sched_thermal_decay_shift
;
1481 static inline void rq_clock_skip_update(struct rq
*rq
)
1483 lockdep_assert_rq_held(rq
);
1484 rq
->clock_update_flags
|= RQCF_REQ_SKIP
;
1488 * See rt task throttling, which is the only time a skip
1489 * request is canceled.
1491 static inline void rq_clock_cancel_skipupdate(struct rq
*rq
)
1493 lockdep_assert_rq_held(rq
);
1494 rq
->clock_update_flags
&= ~RQCF_REQ_SKIP
;
1498 unsigned long flags
;
1499 struct pin_cookie cookie
;
1500 #ifdef CONFIG_SCHED_DEBUG
1502 * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the
1503 * current pin context is stashed here in case it needs to be
1504 * restored in rq_repin_lock().
1506 unsigned int clock_update_flags
;
1510 extern struct callback_head balance_push_callback
;
1513 * Lockdep annotation that avoids accidental unlocks; it's like a
1514 * sticky/continuous lockdep_assert_held().
1516 * This avoids code that has access to 'struct rq *rq' (basically everything in
1517 * the scheduler) from accidentally unlocking the rq if they do not also have a
1518 * copy of the (on-stack) 'struct rq_flags rf'.
1520 * Also see Documentation/locking/lockdep-design.rst.
1522 static inline void rq_pin_lock(struct rq
*rq
, struct rq_flags
*rf
)
1524 rf
->cookie
= lockdep_pin_lock(__rq_lockp(rq
));
1526 #ifdef CONFIG_SCHED_DEBUG
1527 rq
->clock_update_flags
&= (RQCF_REQ_SKIP
|RQCF_ACT_SKIP
);
1528 rf
->clock_update_flags
= 0;
1530 SCHED_WARN_ON(rq
->balance_callback
&& rq
->balance_callback
!= &balance_push_callback
);
1535 static inline void rq_unpin_lock(struct rq
*rq
, struct rq_flags
*rf
)
1537 #ifdef CONFIG_SCHED_DEBUG
1538 if (rq
->clock_update_flags
> RQCF_ACT_SKIP
)
1539 rf
->clock_update_flags
= RQCF_UPDATED
;
1542 lockdep_unpin_lock(__rq_lockp(rq
), rf
->cookie
);
1545 static inline void rq_repin_lock(struct rq
*rq
, struct rq_flags
*rf
)
1547 lockdep_repin_lock(__rq_lockp(rq
), rf
->cookie
);
1549 #ifdef CONFIG_SCHED_DEBUG
1551 * Restore the value we stashed in @rf for this pin context.
1553 rq
->clock_update_flags
|= rf
->clock_update_flags
;
1557 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1558 __acquires(rq
->lock
);
1560 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1561 __acquires(p
->pi_lock
)
1562 __acquires(rq
->lock
);
1564 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1565 __releases(rq
->lock
)
1567 rq_unpin_lock(rq
, rf
);
1568 raw_spin_rq_unlock(rq
);
1572 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1573 __releases(rq
->lock
)
1574 __releases(p
->pi_lock
)
1576 rq_unpin_lock(rq
, rf
);
1577 raw_spin_rq_unlock(rq
);
1578 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1582 rq_lock_irqsave(struct rq
*rq
, struct rq_flags
*rf
)
1583 __acquires(rq
->lock
)
1585 raw_spin_rq_lock_irqsave(rq
, rf
->flags
);
1586 rq_pin_lock(rq
, rf
);
1590 rq_lock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1591 __acquires(rq
->lock
)
1593 raw_spin_rq_lock_irq(rq
);
1594 rq_pin_lock(rq
, rf
);
1598 rq_lock(struct rq
*rq
, struct rq_flags
*rf
)
1599 __acquires(rq
->lock
)
1601 raw_spin_rq_lock(rq
);
1602 rq_pin_lock(rq
, rf
);
1606 rq_unlock_irqrestore(struct rq
*rq
, struct rq_flags
*rf
)
1607 __releases(rq
->lock
)
1609 rq_unpin_lock(rq
, rf
);
1610 raw_spin_rq_unlock_irqrestore(rq
, rf
->flags
);
1614 rq_unlock_irq(struct rq
*rq
, struct rq_flags
*rf
)
1615 __releases(rq
->lock
)
1617 rq_unpin_lock(rq
, rf
);
1618 raw_spin_rq_unlock_irq(rq
);
1622 rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1623 __releases(rq
->lock
)
1625 rq_unpin_lock(rq
, rf
);
1626 raw_spin_rq_unlock(rq
);
1629 static inline struct rq
*
1630 this_rq_lock_irq(struct rq_flags
*rf
)
1631 __acquires(rq
->lock
)
1635 local_irq_disable();
1642 enum numa_topology_type
{
1647 extern enum numa_topology_type sched_numa_topology_type
;
1648 extern int sched_max_numa_distance
;
1649 extern bool find_numa_distance(int distance
);
1650 extern void sched_init_numa(int offline_node
);
1651 extern void sched_update_numa(int cpu
, bool online
);
1652 extern void sched_domains_numa_masks_set(unsigned int cpu
);
1653 extern void sched_domains_numa_masks_clear(unsigned int cpu
);
1654 extern int sched_numa_find_closest(const struct cpumask
*cpus
, int cpu
);
1656 static inline void sched_init_numa(int offline_node
) { }
1657 static inline void sched_update_numa(int cpu
, bool online
) { }
1658 static inline void sched_domains_numa_masks_set(unsigned int cpu
) { }
1659 static inline void sched_domains_numa_masks_clear(unsigned int cpu
) { }
1660 static inline int sched_numa_find_closest(const struct cpumask
*cpus
, int cpu
)
1666 #ifdef CONFIG_NUMA_BALANCING
1667 /* The regions in numa_faults array from task_struct */
1668 enum numa_faults_stats
{
1674 extern void sched_setnuma(struct task_struct
*p
, int node
);
1675 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
1676 extern int migrate_swap(struct task_struct
*p
, struct task_struct
*t
,
1678 extern void init_numa_balancing(unsigned long clone_flags
, struct task_struct
*p
);
1681 init_numa_balancing(unsigned long clone_flags
, struct task_struct
*p
)
1684 #endif /* CONFIG_NUMA_BALANCING */
1689 queue_balance_callback(struct rq
*rq
,
1690 struct callback_head
*head
,
1691 void (*func
)(struct rq
*rq
))
1693 lockdep_assert_rq_held(rq
);
1695 if (unlikely(head
->next
|| rq
->balance_callback
== &balance_push_callback
))
1698 head
->func
= (void (*)(struct callback_head
*))func
;
1699 head
->next
= rq
->balance_callback
;
1700 rq
->balance_callback
= head
;
1703 #define rcu_dereference_check_sched_domain(p) \
1704 rcu_dereference_check((p), \
1705 lockdep_is_held(&sched_domains_mutex))
1708 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1709 * See destroy_sched_domains: call_rcu for details.
1711 * The domain tree of any CPU may only be accessed from within
1712 * preempt-disabled sections.
1714 #define for_each_domain(cpu, __sd) \
1715 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
1716 __sd; __sd = __sd->parent)
1719 * highest_flag_domain - Return highest sched_domain containing flag.
1720 * @cpu: The CPU whose highest level of sched domain is to
1722 * @flag: The flag to check for the highest sched_domain
1723 * for the given CPU.
1725 * Returns the highest sched_domain of a CPU which contains the given flag.
1727 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
1729 struct sched_domain
*sd
, *hsd
= NULL
;
1731 for_each_domain(cpu
, sd
) {
1732 if (!(sd
->flags
& flag
))
1740 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
1742 struct sched_domain
*sd
;
1744 for_each_domain(cpu
, sd
) {
1745 if (sd
->flags
& flag
)
1752 DECLARE_PER_CPU(struct sched_domain __rcu
*, sd_llc
);
1753 DECLARE_PER_CPU(int, sd_llc_size
);
1754 DECLARE_PER_CPU(int, sd_llc_id
);
1755 DECLARE_PER_CPU(struct sched_domain_shared __rcu
*, sd_llc_shared
);
1756 DECLARE_PER_CPU(struct sched_domain __rcu
*, sd_numa
);
1757 DECLARE_PER_CPU(struct sched_domain __rcu
*, sd_asym_packing
);
1758 DECLARE_PER_CPU(struct sched_domain __rcu
*, sd_asym_cpucapacity
);
1759 extern struct static_key_false sched_asym_cpucapacity
;
1761 struct sched_group_capacity
{
1764 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
1767 unsigned long capacity
;
1768 unsigned long min_capacity
; /* Min per-CPU capacity in group */
1769 unsigned long max_capacity
; /* Max per-CPU capacity in group */
1770 unsigned long next_update
;
1771 int imbalance
; /* XXX unrelated to capacity but shared group state */
1773 #ifdef CONFIG_SCHED_DEBUG
1777 unsigned long cpumask
[]; /* Balance mask */
1780 struct sched_group
{
1781 struct sched_group
*next
; /* Must be a circular list */
1784 unsigned int group_weight
;
1785 struct sched_group_capacity
*sgc
;
1786 int asym_prefer_cpu
; /* CPU of highest priority in group */
1790 * The CPUs this group covers.
1792 * NOTE: this field is variable length. (Allocated dynamically
1793 * by attaching extra space to the end of the structure,
1794 * depending on how many CPUs the kernel has booted up with)
1796 unsigned long cpumask
[];
1799 static inline struct cpumask
*sched_group_span(struct sched_group
*sg
)
1801 return to_cpumask(sg
->cpumask
);
1805 * See build_balance_mask().
1807 static inline struct cpumask
*group_balance_mask(struct sched_group
*sg
)
1809 return to_cpumask(sg
->sgc
->cpumask
);
1813 * group_first_cpu - Returns the first CPU in the cpumask of a sched_group.
1814 * @group: The group whose first CPU is to be returned.
1816 static inline unsigned int group_first_cpu(struct sched_group
*group
)
1818 return cpumask_first(sched_group_span(group
));
1821 extern int group_balance_cpu(struct sched_group
*sg
);
1823 #ifdef CONFIG_SCHED_DEBUG
1824 void update_sched_domain_debugfs(void);
1825 void dirty_sched_domain_sysctl(int cpu
);
1827 static inline void update_sched_domain_debugfs(void)
1830 static inline void dirty_sched_domain_sysctl(int cpu
)
1835 extern int sched_update_scaling(void);
1837 extern void flush_smp_call_function_from_idle(void);
1839 #else /* !CONFIG_SMP: */
1840 static inline void flush_smp_call_function_from_idle(void) { }
1845 #if defined(CONFIG_SCHED_CORE) && defined(CONFIG_SCHEDSTATS)
1847 extern void __sched_core_account_forceidle(struct rq
*rq
);
1849 static inline void sched_core_account_forceidle(struct rq
*rq
)
1851 if (schedstat_enabled())
1852 __sched_core_account_forceidle(rq
);
1855 extern void __sched_core_tick(struct rq
*rq
);
1857 static inline void sched_core_tick(struct rq
*rq
)
1859 if (sched_core_enabled(rq
) && schedstat_enabled())
1860 __sched_core_tick(rq
);
1865 static inline void sched_core_account_forceidle(struct rq
*rq
) {}
1867 static inline void sched_core_tick(struct rq
*rq
) {}
1869 #endif /* CONFIG_SCHED_CORE && CONFIG_SCHEDSTATS */
1871 #ifdef CONFIG_CGROUP_SCHED
1874 * Return the group to which this tasks belongs.
1876 * We cannot use task_css() and friends because the cgroup subsystem
1877 * changes that value before the cgroup_subsys::attach() method is called,
1878 * therefore we cannot pin it and might observe the wrong value.
1880 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
1881 * core changes this before calling sched_move_task().
1883 * Instead we use a 'copy' which is updated from sched_move_task() while
1884 * holding both task_struct::pi_lock and rq::lock.
1886 static inline struct task_group
*task_group(struct task_struct
*p
)
1888 return p
->sched_task_group
;
1891 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
1892 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
1894 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
1895 struct task_group
*tg
= task_group(p
);
1898 #ifdef CONFIG_FAIR_GROUP_SCHED
1899 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
1900 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
1901 p
->se
.parent
= tg
->se
[cpu
];
1904 #ifdef CONFIG_RT_GROUP_SCHED
1905 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
1906 p
->rt
.parent
= tg
->rt_se
[cpu
];
1910 #else /* CONFIG_CGROUP_SCHED */
1912 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
1913 static inline struct task_group
*task_group(struct task_struct
*p
)
1918 #endif /* CONFIG_CGROUP_SCHED */
1920 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
1922 set_task_rq(p
, cpu
);
1925 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1926 * successfully executed on another CPU. We must ensure that updates of
1927 * per-task data have been completed by this moment.
1930 WRITE_ONCE(task_thread_info(p
)->cpu
, cpu
);
1936 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1938 #ifdef CONFIG_SCHED_DEBUG
1939 # define const_debug __read_mostly
1941 # define const_debug const
1944 #define SCHED_FEAT(name, enabled) \
1945 __SCHED_FEAT_##name ,
1948 #include "features.h"
1954 #ifdef CONFIG_SCHED_DEBUG
1957 * To support run-time toggling of sched features, all the translation units
1958 * (but core.c) reference the sysctl_sched_features defined in core.c.
1960 extern const_debug
unsigned int sysctl_sched_features
;
1962 #ifdef CONFIG_JUMP_LABEL
1963 #define SCHED_FEAT(name, enabled) \
1964 static __always_inline bool static_branch_##name(struct static_key *key) \
1966 return static_key_##enabled(key); \
1969 #include "features.h"
1972 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1973 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1975 #else /* !CONFIG_JUMP_LABEL */
1977 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1979 #endif /* CONFIG_JUMP_LABEL */
1981 #else /* !SCHED_DEBUG */
1984 * Each translation unit has its own copy of sysctl_sched_features to allow
1985 * constants propagation at compile time and compiler optimization based on
1988 #define SCHED_FEAT(name, enabled) \
1989 (1UL << __SCHED_FEAT_##name) * enabled |
1990 static const_debug __maybe_unused
unsigned int sysctl_sched_features
=
1991 #include "features.h"
1995 #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1997 #endif /* SCHED_DEBUG */
1999 extern struct static_key_false sched_numa_balancing
;
2000 extern struct static_key_false sched_schedstats
;
2002 static inline u64
global_rt_period(void)
2004 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
2007 static inline u64
global_rt_runtime(void)
2009 if (sysctl_sched_rt_runtime
< 0)
2012 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
2015 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
2017 return rq
->curr
== p
;
2020 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
2025 return task_current(rq
, p
);
2029 static inline int task_on_rq_queued(struct task_struct
*p
)
2031 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
2034 static inline int task_on_rq_migrating(struct task_struct
*p
)
2036 return READ_ONCE(p
->on_rq
) == TASK_ON_RQ_MIGRATING
;
2039 /* Wake flags. The first three directly map to some SD flag value */
2040 #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */
2041 #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */
2042 #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */
2044 #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */
2045 #define WF_MIGRATED 0x20 /* Internal use, task got migrated */
2046 #define WF_ON_CPU 0x40 /* Wakee is on_cpu */
2049 static_assert(WF_EXEC
== SD_BALANCE_EXEC
);
2050 static_assert(WF_FORK
== SD_BALANCE_FORK
);
2051 static_assert(WF_TTWU
== SD_BALANCE_WAKE
);
2055 * To aid in avoiding the subversion of "niceness" due to uneven distribution
2056 * of tasks with abnormal "nice" values across CPUs the contribution that
2057 * each task makes to its run queue's load is weighted according to its
2058 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2059 * scaled version of the new time slice allocation that they receive on time
2063 #define WEIGHT_IDLEPRIO 3
2064 #define WMULT_IDLEPRIO 1431655765
2066 extern const int sched_prio_to_weight
[40];
2067 extern const u32 sched_prio_to_wmult
[40];
2070 * {de,en}queue flags:
2072 * DEQUEUE_SLEEP - task is no longer runnable
2073 * ENQUEUE_WAKEUP - task just became runnable
2075 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
2076 * are in a known state which allows modification. Such pairs
2077 * should preserve as much state as possible.
2079 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
2082 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
2083 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
2084 * ENQUEUE_MIGRATED - the task was migrated during wakeup
2088 #define DEQUEUE_SLEEP 0x01
2089 #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */
2090 #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */
2091 #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */
2093 #define ENQUEUE_WAKEUP 0x01
2094 #define ENQUEUE_RESTORE 0x02
2095 #define ENQUEUE_MOVE 0x04
2096 #define ENQUEUE_NOCLOCK 0x08
2098 #define ENQUEUE_HEAD 0x10
2099 #define ENQUEUE_REPLENISH 0x20
2101 #define ENQUEUE_MIGRATED 0x40
2103 #define ENQUEUE_MIGRATED 0x00
2106 #define RETRY_TASK ((void *)-1UL)
2108 struct sched_class
{
2110 #ifdef CONFIG_UCLAMP_TASK
2114 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
2115 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
2116 void (*yield_task
) (struct rq
*rq
);
2117 bool (*yield_to_task
)(struct rq
*rq
, struct task_struct
*p
);
2119 void (*check_preempt_curr
)(struct rq
*rq
, struct task_struct
*p
, int flags
);
2121 struct task_struct
*(*pick_next_task
)(struct rq
*rq
);
2123 void (*put_prev_task
)(struct rq
*rq
, struct task_struct
*p
);
2124 void (*set_next_task
)(struct rq
*rq
, struct task_struct
*p
, bool first
);
2127 int (*balance
)(struct rq
*rq
, struct task_struct
*prev
, struct rq_flags
*rf
);
2128 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int flags
);
2130 struct task_struct
* (*pick_task
)(struct rq
*rq
);
2132 void (*migrate_task_rq
)(struct task_struct
*p
, int new_cpu
);
2134 void (*task_woken
)(struct rq
*this_rq
, struct task_struct
*task
);
2136 void (*set_cpus_allowed
)(struct task_struct
*p
,
2137 const struct cpumask
*newmask
,
2140 void (*rq_online
)(struct rq
*rq
);
2141 void (*rq_offline
)(struct rq
*rq
);
2143 struct rq
*(*find_lock_rq
)(struct task_struct
*p
, struct rq
*rq
);
2146 void (*task_tick
)(struct rq
*rq
, struct task_struct
*p
, int queued
);
2147 void (*task_fork
)(struct task_struct
*p
);
2148 void (*task_dead
)(struct task_struct
*p
);
2151 * The switched_from() call is allowed to drop rq->lock, therefore we
2152 * cannot assume the switched_from/switched_to pair is serialized by
2153 * rq->lock. They are however serialized by p->pi_lock.
2155 void (*switched_from
)(struct rq
*this_rq
, struct task_struct
*task
);
2156 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
2157 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
2160 unsigned int (*get_rr_interval
)(struct rq
*rq
,
2161 struct task_struct
*task
);
2163 void (*update_curr
)(struct rq
*rq
);
2165 #define TASK_SET_GROUP 0
2166 #define TASK_MOVE_GROUP 1
2168 #ifdef CONFIG_FAIR_GROUP_SCHED
2169 void (*task_change_group
)(struct task_struct
*p
, int type
);
2173 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
2175 WARN_ON_ONCE(rq
->curr
!= prev
);
2176 prev
->sched_class
->put_prev_task(rq
, prev
);
2179 static inline void set_next_task(struct rq
*rq
, struct task_struct
*next
)
2181 next
->sched_class
->set_next_task(rq
, next
, false);
2186 * Helper to define a sched_class instance; each one is placed in a separate
2187 * section which is ordered by the linker script:
2189 * include/asm-generic/vmlinux.lds.h
2191 * Also enforce alignment on the instance, not the type, to guarantee layout.
2193 #define DEFINE_SCHED_CLASS(name) \
2194 const struct sched_class name##_sched_class \
2195 __aligned(__alignof__(struct sched_class)) \
2196 __section("__" #name "_sched_class")
2198 /* Defined in include/asm-generic/vmlinux.lds.h */
2199 extern struct sched_class __begin_sched_classes
[];
2200 extern struct sched_class __end_sched_classes
[];
2202 #define sched_class_highest (__end_sched_classes - 1)
2203 #define sched_class_lowest (__begin_sched_classes - 1)
2205 #define for_class_range(class, _from, _to) \
2206 for (class = (_from); class != (_to); class--)
2208 #define for_each_class(class) \
2209 for_class_range(class, sched_class_highest, sched_class_lowest)
2211 extern const struct sched_class stop_sched_class
;
2212 extern const struct sched_class dl_sched_class
;
2213 extern const struct sched_class rt_sched_class
;
2214 extern const struct sched_class fair_sched_class
;
2215 extern const struct sched_class idle_sched_class
;
2217 static inline bool sched_stop_runnable(struct rq
*rq
)
2219 return rq
->stop
&& task_on_rq_queued(rq
->stop
);
2222 static inline bool sched_dl_runnable(struct rq
*rq
)
2224 return rq
->dl
.dl_nr_running
> 0;
2227 static inline bool sched_rt_runnable(struct rq
*rq
)
2229 return rq
->rt
.rt_queued
> 0;
2232 static inline bool sched_fair_runnable(struct rq
*rq
)
2234 return rq
->cfs
.nr_running
> 0;
2237 extern struct task_struct
*pick_next_task_fair(struct rq
*rq
, struct task_struct
*prev
, struct rq_flags
*rf
);
2238 extern struct task_struct
*pick_next_task_idle(struct rq
*rq
);
2240 #define SCA_CHECK 0x01
2241 #define SCA_MIGRATE_DISABLE 0x02
2242 #define SCA_MIGRATE_ENABLE 0x04
2243 #define SCA_USER 0x08
2247 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
2249 extern void trigger_load_balance(struct rq
*rq
);
2251 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
, u32 flags
);
2253 static inline struct task_struct
*get_push_task(struct rq
*rq
)
2255 struct task_struct
*p
= rq
->curr
;
2257 lockdep_assert_rq_held(rq
);
2262 if (p
->nr_cpus_allowed
== 1)
2265 if (p
->migration_disabled
)
2268 rq
->push_busy
= true;
2269 return get_task_struct(p
);
2272 extern int push_cpu_stop(void *arg
);
2276 #ifdef CONFIG_CPU_IDLE
2277 static inline void idle_set_state(struct rq
*rq
,
2278 struct cpuidle_state
*idle_state
)
2280 rq
->idle_state
= idle_state
;
2283 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
2285 SCHED_WARN_ON(!rcu_read_lock_held());
2287 return rq
->idle_state
;
2290 static inline void idle_set_state(struct rq
*rq
,
2291 struct cpuidle_state
*idle_state
)
2295 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
2301 extern void schedule_idle(void);
2303 extern void sysrq_sched_debug_show(void);
2304 extern void sched_init_granularity(void);
2305 extern void update_max_interval(void);
2307 extern void init_sched_dl_class(void);
2308 extern void init_sched_rt_class(void);
2309 extern void init_sched_fair_class(void);
2311 extern void reweight_task(struct task_struct
*p
, int prio
);
2313 extern void resched_curr(struct rq
*rq
);
2314 extern void resched_cpu(int cpu
);
2316 extern struct rt_bandwidth def_rt_bandwidth
;
2317 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
2319 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
2320 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
2321 extern void init_dl_inactive_task_timer(struct sched_dl_entity
*dl_se
);
2324 #define BW_UNIT (1 << BW_SHIFT)
2325 #define RATIO_SHIFT 8
2326 #define MAX_BW_BITS (64 - BW_SHIFT)
2327 #define MAX_BW ((1ULL << MAX_BW_BITS) - 1)
2328 unsigned long to_ratio(u64 period
, u64 runtime
);
2330 extern void init_entity_runnable_average(struct sched_entity
*se
);
2331 extern void post_init_entity_util_avg(struct task_struct
*p
);
2333 #ifdef CONFIG_NO_HZ_FULL
2334 extern bool sched_can_stop_tick(struct rq
*rq
);
2335 extern int __init
sched_tick_offload_init(void);
2338 * Tick may be needed by tasks in the runqueue depending on their policy and
2339 * requirements. If tick is needed, lets send the target an IPI to kick it out of
2340 * nohz mode if necessary.
2342 static inline void sched_update_tick_dependency(struct rq
*rq
)
2344 int cpu
= cpu_of(rq
);
2346 if (!tick_nohz_full_cpu(cpu
))
2349 if (sched_can_stop_tick(rq
))
2350 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
2352 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
2355 static inline int sched_tick_offload_init(void) { return 0; }
2356 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
2359 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
2361 unsigned prev_nr
= rq
->nr_running
;
2363 rq
->nr_running
= prev_nr
+ count
;
2364 if (trace_sched_update_nr_running_tp_enabled()) {
2365 call_trace_sched_update_nr_running(rq
, count
);
2369 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
2370 if (!READ_ONCE(rq
->rd
->overload
))
2371 WRITE_ONCE(rq
->rd
->overload
, 1);
2375 sched_update_tick_dependency(rq
);
2378 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
2380 rq
->nr_running
-= count
;
2381 if (trace_sched_update_nr_running_tp_enabled()) {
2382 call_trace_sched_update_nr_running(rq
, -count
);
2385 /* Check if we still need preemption */
2386 sched_update_tick_dependency(rq
);
2389 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
2390 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
2392 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
2394 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
2395 extern const_debug
unsigned int sysctl_sched_migration_cost
;
2397 #ifdef CONFIG_SCHED_DEBUG
2398 extern unsigned int sysctl_sched_latency
;
2399 extern unsigned int sysctl_sched_min_granularity
;
2400 extern unsigned int sysctl_sched_idle_min_granularity
;
2401 extern unsigned int sysctl_sched_wakeup_granularity
;
2402 extern int sysctl_resched_latency_warn_ms
;
2403 extern int sysctl_resched_latency_warn_once
;
2405 extern unsigned int sysctl_sched_tunable_scaling
;
2407 extern unsigned int sysctl_numa_balancing_scan_delay
;
2408 extern unsigned int sysctl_numa_balancing_scan_period_min
;
2409 extern unsigned int sysctl_numa_balancing_scan_period_max
;
2410 extern unsigned int sysctl_numa_balancing_scan_size
;
2413 #ifdef CONFIG_SCHED_HRTICK
2417 * - enabled by features
2418 * - hrtimer is actually high res
2420 static inline int hrtick_enabled(struct rq
*rq
)
2422 if (!cpu_active(cpu_of(rq
)))
2424 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
2427 static inline int hrtick_enabled_fair(struct rq
*rq
)
2429 if (!sched_feat(HRTICK
))
2431 return hrtick_enabled(rq
);
2434 static inline int hrtick_enabled_dl(struct rq
*rq
)
2436 if (!sched_feat(HRTICK_DL
))
2438 return hrtick_enabled(rq
);
2441 void hrtick_start(struct rq
*rq
, u64 delay
);
2445 static inline int hrtick_enabled_fair(struct rq
*rq
)
2450 static inline int hrtick_enabled_dl(struct rq
*rq
)
2455 static inline int hrtick_enabled(struct rq
*rq
)
2460 #endif /* CONFIG_SCHED_HRTICK */
2462 #ifndef arch_scale_freq_tick
2463 static __always_inline
2464 void arch_scale_freq_tick(void)
2469 #ifndef arch_scale_freq_capacity
2471 * arch_scale_freq_capacity - get the frequency scale factor of a given CPU.
2472 * @cpu: the CPU in question.
2474 * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e.
2477 * ------ * SCHED_CAPACITY_SCALE
2480 static __always_inline
2481 unsigned long arch_scale_freq_capacity(int cpu
)
2483 return SCHED_CAPACITY_SCALE
;
2490 static inline bool rq_order_less(struct rq
*rq1
, struct rq
*rq2
)
2492 #ifdef CONFIG_SCHED_CORE
2494 * In order to not have {0,2},{1,3} turn into into an AB-BA,
2495 * order by core-id first and cpu-id second.
2499 * double_rq_lock(0,3); will take core-0, core-1 lock
2500 * double_rq_lock(1,2); will take core-1, core-0 lock
2502 * when only cpu-id is considered.
2504 if (rq1
->core
->cpu
< rq2
->core
->cpu
)
2506 if (rq1
->core
->cpu
> rq2
->core
->cpu
)
2510 * __sched_core_flip() relies on SMT having cpu-id lock order.
2513 return rq1
->cpu
< rq2
->cpu
;
2516 extern void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
2518 #ifdef CONFIG_PREEMPTION
2521 * fair double_lock_balance: Safely acquires both rq->locks in a fair
2522 * way at the expense of forcing extra atomic operations in all
2523 * invocations. This assures that the double_lock is acquired using the
2524 * same underlying policy as the spinlock_t on this architecture, which
2525 * reduces latency compared to the unfair variant below. However, it
2526 * also adds more overhead and therefore may reduce throughput.
2528 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
2529 __releases(this_rq
->lock
)
2530 __acquires(busiest
->lock
)
2531 __acquires(this_rq
->lock
)
2533 raw_spin_rq_unlock(this_rq
);
2534 double_rq_lock(this_rq
, busiest
);
2541 * Unfair double_lock_balance: Optimizes throughput at the expense of
2542 * latency by eliminating extra atomic operations when the locks are
2543 * already in proper order on entry. This favors lower CPU-ids and will
2544 * grant the double lock to lower CPUs over higher ids under contention,
2545 * regardless of entry order into the function.
2547 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
2548 __releases(this_rq
->lock
)
2549 __acquires(busiest
->lock
)
2550 __acquires(this_rq
->lock
)
2552 if (__rq_lockp(this_rq
) == __rq_lockp(busiest
))
2555 if (likely(raw_spin_rq_trylock(busiest
)))
2558 if (rq_order_less(this_rq
, busiest
)) {
2559 raw_spin_rq_lock_nested(busiest
, SINGLE_DEPTH_NESTING
);
2563 raw_spin_rq_unlock(this_rq
);
2564 double_rq_lock(this_rq
, busiest
);
2569 #endif /* CONFIG_PREEMPTION */
2572 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2574 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
2576 lockdep_assert_irqs_disabled();
2578 return _double_lock_balance(this_rq
, busiest
);
2581 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
2582 __releases(busiest
->lock
)
2584 if (__rq_lockp(this_rq
) != __rq_lockp(busiest
))
2585 raw_spin_rq_unlock(busiest
);
2586 lock_set_subclass(&__rq_lockp(this_rq
)->dep_map
, 0, _RET_IP_
);
2589 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
2595 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
2598 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
2604 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
2607 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
2613 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
2617 * double_rq_unlock - safely unlock two runqueues
2619 * Note this does not restore interrupts like task_rq_unlock,
2620 * you need to do so manually after calling.
2622 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
2623 __releases(rq1
->lock
)
2624 __releases(rq2
->lock
)
2626 if (__rq_lockp(rq1
) != __rq_lockp(rq2
))
2627 raw_spin_rq_unlock(rq2
);
2629 __release(rq2
->lock
);
2630 raw_spin_rq_unlock(rq1
);
2633 extern void set_rq_online (struct rq
*rq
);
2634 extern void set_rq_offline(struct rq
*rq
);
2635 extern bool sched_smp_initialized
;
2637 #else /* CONFIG_SMP */
2640 * double_rq_lock - safely lock two runqueues
2642 * Note this does not disable interrupts like task_rq_lock,
2643 * you need to do so manually before calling.
2645 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
2646 __acquires(rq1
->lock
)
2647 __acquires(rq2
->lock
)
2649 BUG_ON(!irqs_disabled());
2651 raw_spin_rq_lock(rq1
);
2652 __acquire(rq2
->lock
); /* Fake it out ;) */
2656 * double_rq_unlock - safely unlock two runqueues
2658 * Note this does not restore interrupts like task_rq_unlock,
2659 * you need to do so manually after calling.
2661 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
2662 __releases(rq1
->lock
)
2663 __releases(rq2
->lock
)
2666 raw_spin_rq_unlock(rq1
);
2667 __release(rq2
->lock
);
2672 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
2673 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
2675 #ifdef CONFIG_SCHED_DEBUG
2676 extern bool sched_debug_verbose
;
2678 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
2679 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
2680 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
2681 extern void print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
2682 extern void print_rt_rq(struct seq_file
*m
, int cpu
, struct rt_rq
*rt_rq
);
2683 extern void print_dl_rq(struct seq_file
*m
, int cpu
, struct dl_rq
*dl_rq
);
2685 extern void resched_latency_warn(int cpu
, u64 latency
);
2686 #ifdef CONFIG_NUMA_BALANCING
2688 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
2690 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
2691 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
2692 #endif /* CONFIG_NUMA_BALANCING */
2694 static inline void resched_latency_warn(int cpu
, u64 latency
) {}
2695 #endif /* CONFIG_SCHED_DEBUG */
2697 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
2698 extern void init_rt_rq(struct rt_rq
*rt_rq
);
2699 extern void init_dl_rq(struct dl_rq
*dl_rq
);
2701 extern void cfs_bandwidth_usage_inc(void);
2702 extern void cfs_bandwidth_usage_dec(void);
2704 #ifdef CONFIG_NO_HZ_COMMON
2705 #define NOHZ_BALANCE_KICK_BIT 0
2706 #define NOHZ_STATS_KICK_BIT 1
2707 #define NOHZ_NEWILB_KICK_BIT 2
2708 #define NOHZ_NEXT_KICK_BIT 3
2710 /* Run rebalance_domains() */
2711 #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT)
2712 /* Update blocked load */
2713 #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT)
2714 /* Update blocked load when entering idle */
2715 #define NOHZ_NEWILB_KICK BIT(NOHZ_NEWILB_KICK_BIT)
2716 /* Update nohz.next_balance */
2717 #define NOHZ_NEXT_KICK BIT(NOHZ_NEXT_KICK_BIT)
2719 #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK | NOHZ_NEXT_KICK)
2721 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
2723 extern void nohz_balance_exit_idle(struct rq
*rq
);
2725 static inline void nohz_balance_exit_idle(struct rq
*rq
) { }
2728 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
2729 extern void nohz_run_idle_balance(int cpu
);
2731 static inline void nohz_run_idle_balance(int cpu
) { }
2734 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
2739 struct u64_stats_sync sync
;
2742 DECLARE_PER_CPU(struct irqtime
, cpu_irqtime
);
2745 * Returns the irqtime minus the softirq time computed by ksoftirqd.
2746 * Otherwise ksoftirqd's sum_exec_runtime is subtracted its own runtime
2747 * and never move forward.
2749 static inline u64
irq_time_read(int cpu
)
2751 struct irqtime
*irqtime
= &per_cpu(cpu_irqtime
, cpu
);
2756 seq
= __u64_stats_fetch_begin(&irqtime
->sync
);
2757 total
= irqtime
->total
;
2758 } while (__u64_stats_fetch_retry(&irqtime
->sync
, seq
));
2762 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
2764 #ifdef CONFIG_CPU_FREQ
2765 DECLARE_PER_CPU(struct update_util_data __rcu
*, cpufreq_update_util_data
);
2768 * cpufreq_update_util - Take a note about CPU utilization changes.
2769 * @rq: Runqueue to carry out the update for.
2770 * @flags: Update reason flags.
2772 * This function is called by the scheduler on the CPU whose utilization is
2775 * It can only be called from RCU-sched read-side critical sections.
2777 * The way cpufreq is currently arranged requires it to evaluate the CPU
2778 * performance state (frequency/voltage) on a regular basis to prevent it from
2779 * being stuck in a completely inadequate performance level for too long.
2780 * That is not guaranteed to happen if the updates are only triggered from CFS
2781 * and DL, though, because they may not be coming in if only RT tasks are
2782 * active all the time (or there are RT tasks only).
2784 * As a workaround for that issue, this function is called periodically by the
2785 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,
2786 * but that really is a band-aid. Going forward it should be replaced with
2787 * solutions targeted more specifically at RT tasks.
2789 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
2791 struct update_util_data
*data
;
2793 data
= rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data
,
2796 data
->func(data
, rq_clock(rq
), flags
);
2799 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
2800 #endif /* CONFIG_CPU_FREQ */
2802 #ifdef arch_scale_freq_capacity
2803 # ifndef arch_scale_freq_invariant
2804 # define arch_scale_freq_invariant() true
2807 # define arch_scale_freq_invariant() false
2811 static inline unsigned long capacity_orig_of(int cpu
)
2813 return cpu_rq(cpu
)->cpu_capacity_orig
;
2817 * enum cpu_util_type - CPU utilization type
2818 * @FREQUENCY_UTIL: Utilization used to select frequency
2819 * @ENERGY_UTIL: Utilization used during energy calculation
2821 * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time
2822 * need to be aggregated differently depending on the usage made of them. This
2823 * enum is used within effective_cpu_util() to differentiate the types of
2824 * utilization expected by the callers, and adjust the aggregation accordingly.
2826 enum cpu_util_type
{
2831 unsigned long effective_cpu_util(int cpu
, unsigned long util_cfs
,
2832 unsigned long max
, enum cpu_util_type type
,
2833 struct task_struct
*p
);
2835 static inline unsigned long cpu_bw_dl(struct rq
*rq
)
2837 return (rq
->dl
.running_bw
* SCHED_CAPACITY_SCALE
) >> BW_SHIFT
;
2840 static inline unsigned long cpu_util_dl(struct rq
*rq
)
2842 return READ_ONCE(rq
->avg_dl
.util_avg
);
2846 * cpu_util_cfs() - Estimates the amount of CPU capacity used by CFS tasks.
2847 * @cpu: the CPU to get the utilization for.
2849 * The unit of the return value must be the same as the one of CPU capacity
2850 * so that CPU utilization can be compared with CPU capacity.
2852 * CPU utilization is the sum of running time of runnable tasks plus the
2853 * recent utilization of currently non-runnable tasks on that CPU.
2854 * It represents the amount of CPU capacity currently used by CFS tasks in
2855 * the range [0..max CPU capacity] with max CPU capacity being the CPU
2856 * capacity at f_max.
2858 * The estimated CPU utilization is defined as the maximum between CPU
2859 * utilization and sum of the estimated utilization of the currently
2860 * runnable tasks on that CPU. It preserves a utilization "snapshot" of
2861 * previously-executed tasks, which helps better deduce how busy a CPU will
2862 * be when a long-sleeping task wakes up. The contribution to CPU utilization
2863 * of such a task would be significantly decayed at this point of time.
2865 * CPU utilization can be higher than the current CPU capacity
2866 * (f_curr/f_max * max CPU capacity) or even the max CPU capacity because
2867 * of rounding errors as well as task migrations or wakeups of new tasks.
2868 * CPU utilization has to be capped to fit into the [0..max CPU capacity]
2869 * range. Otherwise a group of CPUs (CPU0 util = 121% + CPU1 util = 80%)
2870 * could be seen as over-utilized even though CPU1 has 20% of spare CPU
2871 * capacity. CPU utilization is allowed to overshoot current CPU capacity
2872 * though since this is useful for predicting the CPU capacity required
2873 * after task migrations (scheduler-driven DVFS).
2875 * Return: (Estimated) utilization for the specified CPU.
2877 static inline unsigned long cpu_util_cfs(int cpu
)
2879 struct cfs_rq
*cfs_rq
;
2882 cfs_rq
= &cpu_rq(cpu
)->cfs
;
2883 util
= READ_ONCE(cfs_rq
->avg
.util_avg
);
2885 if (sched_feat(UTIL_EST
)) {
2886 util
= max_t(unsigned long, util
,
2887 READ_ONCE(cfs_rq
->avg
.util_est
.enqueued
));
2890 return min(util
, capacity_orig_of(cpu
));
2893 static inline unsigned long cpu_util_rt(struct rq
*rq
)
2895 return READ_ONCE(rq
->avg_rt
.util_avg
);
2899 #ifdef CONFIG_UCLAMP_TASK
2900 unsigned long uclamp_eff_value(struct task_struct
*p
, enum uclamp_id clamp_id
);
2903 * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values.
2904 * @rq: The rq to clamp against. Must not be NULL.
2905 * @util: The util value to clamp.
2906 * @p: The task to clamp against. Can be NULL if you want to clamp
2909 * Clamps the passed @util to the max(@rq, @p) effective uclamp values.
2911 * If sched_uclamp_used static key is disabled, then just return the util
2912 * without any clamping since uclamp aggregation at the rq level in the fast
2913 * path is disabled, rendering this operation a NOP.
2915 * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It
2916 * will return the correct effective uclamp value of the task even if the
2917 * static key is disabled.
2919 static __always_inline
2920 unsigned long uclamp_rq_util_with(struct rq
*rq
, unsigned long util
,
2921 struct task_struct
*p
)
2923 unsigned long min_util
= 0;
2924 unsigned long max_util
= 0;
2926 if (!static_branch_likely(&sched_uclamp_used
))
2930 min_util
= uclamp_eff_value(p
, UCLAMP_MIN
);
2931 max_util
= uclamp_eff_value(p
, UCLAMP_MAX
);
2934 * Ignore last runnable task's max clamp, as this task will
2935 * reset it. Similarly, no need to read the rq's min clamp.
2937 if (rq
->uclamp_flags
& UCLAMP_FLAG_IDLE
)
2941 min_util
= max_t(unsigned long, min_util
, READ_ONCE(rq
->uclamp
[UCLAMP_MIN
].value
));
2942 max_util
= max_t(unsigned long, max_util
, READ_ONCE(rq
->uclamp
[UCLAMP_MAX
].value
));
2945 * Since CPU's {min,max}_util clamps are MAX aggregated considering
2946 * RUNNABLE tasks with _different_ clamps, we can end up with an
2947 * inversion. Fix it now when the clamps are applied.
2949 if (unlikely(min_util
>= max_util
))
2952 return clamp(util
, min_util
, max_util
);
2955 /* Is the rq being capped/throttled by uclamp_max? */
2956 static inline bool uclamp_rq_is_capped(struct rq
*rq
)
2958 unsigned long rq_util
;
2959 unsigned long max_util
;
2961 if (!static_branch_likely(&sched_uclamp_used
))
2964 rq_util
= cpu_util_cfs(cpu_of(rq
)) + cpu_util_rt(rq
);
2965 max_util
= READ_ONCE(rq
->uclamp
[UCLAMP_MAX
].value
);
2967 return max_util
!= SCHED_CAPACITY_SCALE
&& rq_util
>= max_util
;
2971 * When uclamp is compiled in, the aggregation at rq level is 'turned off'
2972 * by default in the fast path and only gets turned on once userspace performs
2973 * an operation that requires it.
2975 * Returns true if userspace opted-in to use uclamp and aggregation at rq level
2978 static inline bool uclamp_is_used(void)
2980 return static_branch_likely(&sched_uclamp_used
);
2982 #else /* CONFIG_UCLAMP_TASK */
2984 unsigned long uclamp_rq_util_with(struct rq
*rq
, unsigned long util
,
2985 struct task_struct
*p
)
2990 static inline bool uclamp_rq_is_capped(struct rq
*rq
) { return false; }
2992 static inline bool uclamp_is_used(void)
2996 #endif /* CONFIG_UCLAMP_TASK */
2998 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
2999 static inline unsigned long cpu_util_irq(struct rq
*rq
)
3001 return rq
->avg_irq
.util_avg
;
3005 unsigned long scale_irq_capacity(unsigned long util
, unsigned long irq
, unsigned long max
)
3007 util
*= (max
- irq
);
3014 static inline unsigned long cpu_util_irq(struct rq
*rq
)
3020 unsigned long scale_irq_capacity(unsigned long util
, unsigned long irq
, unsigned long max
)
3026 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
3028 #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus)))
3030 DECLARE_STATIC_KEY_FALSE(sched_energy_present
);
3032 static inline bool sched_energy_enabled(void)
3034 return static_branch_unlikely(&sched_energy_present
);
3037 #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */
3039 #define perf_domain_span(pd) NULL
3040 static inline bool sched_energy_enabled(void) { return false; }
3042 #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */
3044 #ifdef CONFIG_MEMBARRIER
3046 * The scheduler provides memory barriers required by membarrier between:
3047 * - prior user-space memory accesses and store to rq->membarrier_state,
3048 * - store to rq->membarrier_state and following user-space memory accesses.
3049 * In the same way it provides those guarantees around store to rq->curr.
3051 static inline void membarrier_switch_mm(struct rq
*rq
,
3052 struct mm_struct
*prev_mm
,
3053 struct mm_struct
*next_mm
)
3055 int membarrier_state
;
3057 if (prev_mm
== next_mm
)
3060 membarrier_state
= atomic_read(&next_mm
->membarrier_state
);
3061 if (READ_ONCE(rq
->membarrier_state
) == membarrier_state
)
3064 WRITE_ONCE(rq
->membarrier_state
, membarrier_state
);
3067 static inline void membarrier_switch_mm(struct rq
*rq
,
3068 struct mm_struct
*prev_mm
,
3069 struct mm_struct
*next_mm
)
3075 static inline bool is_per_cpu_kthread(struct task_struct
*p
)
3077 if (!(p
->flags
& PF_KTHREAD
))
3080 if (p
->nr_cpus_allowed
!= 1)
3087 extern void swake_up_all_locked(struct swait_queue_head
*q
);
3088 extern void __prepare_to_swait(struct swait_queue_head
*q
, struct swait_queue
*wait
);
3090 #ifdef CONFIG_PREEMPT_DYNAMIC
3091 extern int preempt_dynamic_mode
;
3092 extern int sched_dynamic_mode(const char *str
);
3093 extern void sched_dynamic_update(int mode
);
3096 #endif /* _KERNEL_SCHED_SCHED_H */