2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/sched/deadline.h>
6 #include <linux/binfmts.h>
7 #include <linux/mutex.h>
8 #include <linux/spinlock.h>
9 #include <linux/stop_machine.h>
10 #include <linux/irq_work.h>
11 #include <linux/tick.h>
12 #include <linux/slab.h>
15 #include "cpudeadline.h"
21 /* task_struct::on_rq states: */
22 #define TASK_ON_RQ_QUEUED 1
23 #define TASK_ON_RQ_MIGRATING 2
25 extern __read_mostly
int scheduler_running
;
27 extern unsigned long calc_load_update
;
28 extern atomic_long_t calc_load_tasks
;
30 extern void calc_global_load_tick(struct rq
*this_rq
);
31 extern long calc_load_fold_active(struct rq
*this_rq
, long adjust
);
34 extern void cpu_load_update_active(struct rq
*this_rq
);
36 static inline void cpu_load_update_active(struct rq
*this_rq
) { }
40 * Helpers for converting nanosecond timing to jiffy resolution
42 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
45 * Increase resolution of nice-level calculations for 64-bit architectures.
46 * The extra resolution improves shares distribution and load balancing of
47 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
48 * hierarchies, especially on larger systems. This is not a user-visible change
49 * and does not change the user-interface for setting shares/weights.
51 * We increase resolution only if we have enough bits to allow this increased
52 * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are
53 * pretty high and the returns do not justify the increased costs.
55 * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to
56 * increase coverage and consistency always enable it on 64bit platforms.
59 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
60 # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT)
61 # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT)
63 # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT)
64 # define scale_load(w) (w)
65 # define scale_load_down(w) (w)
69 * Task weight (visible to users) and its load (invisible to users) have
70 * independent resolution, but they should be well calibrated. We use
71 * scale_load() and scale_load_down(w) to convert between them. The
72 * following must be true:
74 * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD
77 #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT)
80 * Single value that decides SCHED_DEADLINE internal math precision.
81 * 10 -> just above 1us
82 * 9 -> just above 0.5us
87 * These are the 'tuning knobs' of the scheduler:
91 * single value that denotes runtime == period, ie unlimited time.
93 #define RUNTIME_INF ((u64)~0ULL)
95 static inline int idle_policy(int policy
)
97 return policy
== SCHED_IDLE
;
99 static inline int fair_policy(int policy
)
101 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
104 static inline int rt_policy(int policy
)
106 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
109 static inline int dl_policy(int policy
)
111 return policy
== SCHED_DEADLINE
;
113 static inline bool valid_policy(int policy
)
115 return idle_policy(policy
) || fair_policy(policy
) ||
116 rt_policy(policy
) || dl_policy(policy
);
119 static inline int task_has_rt_policy(struct task_struct
*p
)
121 return rt_policy(p
->policy
);
124 static inline int task_has_dl_policy(struct task_struct
*p
)
126 return dl_policy(p
->policy
);
130 * Tells if entity @a should preempt entity @b.
133 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
135 return dl_time_before(a
->deadline
, b
->deadline
);
139 * This is the priority-queue data structure of the RT scheduling class:
141 struct rt_prio_array
{
142 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
143 struct list_head queue
[MAX_RT_PRIO
];
146 struct rt_bandwidth
{
147 /* nests inside the rq lock: */
148 raw_spinlock_t rt_runtime_lock
;
151 struct hrtimer rt_period_timer
;
152 unsigned int rt_period_active
;
155 void __dl_clear_params(struct task_struct
*p
);
158 * To keep the bandwidth of -deadline tasks and groups under control
159 * we need some place where:
160 * - store the maximum -deadline bandwidth of the system (the group);
161 * - cache the fraction of that bandwidth that is currently allocated.
163 * This is all done in the data structure below. It is similar to the
164 * one used for RT-throttling (rt_bandwidth), with the main difference
165 * that, since here we are only interested in admission control, we
166 * do not decrease any runtime while the group "executes", neither we
167 * need a timer to replenish it.
169 * With respect to SMP, the bandwidth is given on a per-CPU basis,
171 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
172 * - dl_total_bw array contains, in the i-eth element, the currently
173 * allocated bandwidth on the i-eth CPU.
174 * Moreover, groups consume bandwidth on each CPU, while tasks only
175 * consume bandwidth on the CPU they're running on.
176 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
177 * that will be shown the next time the proc or cgroup controls will
178 * be red. It on its turn can be changed by writing on its own
181 struct dl_bandwidth
{
182 raw_spinlock_t dl_runtime_lock
;
187 static inline int dl_bandwidth_enabled(void)
189 return sysctl_sched_rt_runtime
>= 0;
192 extern struct dl_bw
*dl_bw_of(int i
);
200 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
202 dl_b
->total_bw
-= tsk_bw
;
206 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
208 dl_b
->total_bw
+= tsk_bw
;
212 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
214 return dl_b
->bw
!= -1 &&
215 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
218 extern struct mutex sched_domains_mutex
;
220 #ifdef CONFIG_CGROUP_SCHED
222 #include <linux/cgroup.h>
227 extern struct list_head task_groups
;
229 struct cfs_bandwidth
{
230 #ifdef CONFIG_CFS_BANDWIDTH
234 s64 hierarchical_quota
;
237 int idle
, period_active
;
238 struct hrtimer period_timer
, slack_timer
;
239 struct list_head throttled_cfs_rq
;
242 int nr_periods
, nr_throttled
;
247 /* task group related information */
249 struct cgroup_subsys_state css
;
251 #ifdef CONFIG_FAIR_GROUP_SCHED
252 /* schedulable entities of this group on each cpu */
253 struct sched_entity
**se
;
254 /* runqueue "owned" by this group on each cpu */
255 struct cfs_rq
**cfs_rq
;
256 unsigned long shares
;
260 * load_avg can be heavily contended at clock tick time, so put
261 * it in its own cacheline separated from the fields above which
262 * will also be accessed at each tick.
264 atomic_long_t load_avg ____cacheline_aligned
;
268 #ifdef CONFIG_RT_GROUP_SCHED
269 struct sched_rt_entity
**rt_se
;
270 struct rt_rq
**rt_rq
;
272 struct rt_bandwidth rt_bandwidth
;
276 struct list_head list
;
278 struct task_group
*parent
;
279 struct list_head siblings
;
280 struct list_head children
;
282 #ifdef CONFIG_SCHED_AUTOGROUP
283 struct autogroup
*autogroup
;
286 struct cfs_bandwidth cfs_bandwidth
;
289 #ifdef CONFIG_FAIR_GROUP_SCHED
290 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
293 * A weight of 0 or 1 can cause arithmetics problems.
294 * A weight of a cfs_rq is the sum of weights of which entities
295 * are queued on this cfs_rq, so a weight of a entity should not be
296 * too large, so as the shares value of a task group.
297 * (The default weight is 1024 - so there's no practical
298 * limitation from this.)
300 #define MIN_SHARES (1UL << 1)
301 #define MAX_SHARES (1UL << 18)
304 typedef int (*tg_visitor
)(struct task_group
*, void *);
306 extern int walk_tg_tree_from(struct task_group
*from
,
307 tg_visitor down
, tg_visitor up
, void *data
);
310 * Iterate the full tree, calling @down when first entering a node and @up when
311 * leaving it for the final time.
313 * Caller must hold rcu_lock or sufficient equivalent.
315 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
317 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
320 extern int tg_nop(struct task_group
*tg
, void *data
);
322 extern void free_fair_sched_group(struct task_group
*tg
);
323 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
324 extern void online_fair_sched_group(struct task_group
*tg
);
325 extern void unregister_fair_sched_group(struct task_group
*tg
);
326 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
327 struct sched_entity
*se
, int cpu
,
328 struct sched_entity
*parent
);
329 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
331 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
332 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
333 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
335 extern void free_rt_sched_group(struct task_group
*tg
);
336 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
337 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
338 struct sched_rt_entity
*rt_se
, int cpu
,
339 struct sched_rt_entity
*parent
);
341 extern struct task_group
*sched_create_group(struct task_group
*parent
);
342 extern void sched_online_group(struct task_group
*tg
,
343 struct task_group
*parent
);
344 extern void sched_destroy_group(struct task_group
*tg
);
345 extern void sched_offline_group(struct task_group
*tg
);
347 extern void sched_move_task(struct task_struct
*tsk
);
349 #ifdef CONFIG_FAIR_GROUP_SCHED
350 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
353 extern void set_task_rq_fair(struct sched_entity
*se
,
354 struct cfs_rq
*prev
, struct cfs_rq
*next
);
355 #else /* !CONFIG_SMP */
356 static inline void set_task_rq_fair(struct sched_entity
*se
,
357 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
358 #endif /* CONFIG_SMP */
359 #endif /* CONFIG_FAIR_GROUP_SCHED */
361 #else /* CONFIG_CGROUP_SCHED */
363 struct cfs_bandwidth
{ };
365 #endif /* CONFIG_CGROUP_SCHED */
367 /* CFS-related fields in a runqueue */
369 struct load_weight load
;
370 unsigned int nr_running
, h_nr_running
;
375 u64 min_vruntime_copy
;
378 struct rb_root tasks_timeline
;
379 struct rb_node
*rb_leftmost
;
382 * 'curr' points to currently running entity on this cfs_rq.
383 * It is set to NULL otherwise (i.e when none are currently running).
385 struct sched_entity
*curr
, *next
, *last
, *skip
;
387 #ifdef CONFIG_SCHED_DEBUG
388 unsigned int nr_spread_over
;
395 struct sched_avg avg
;
396 u64 runnable_load_sum
;
397 unsigned long runnable_load_avg
;
398 #ifdef CONFIG_FAIR_GROUP_SCHED
399 unsigned long tg_load_avg_contrib
;
401 atomic_long_t removed_load_avg
, removed_util_avg
;
403 u64 load_last_update_time_copy
;
406 #ifdef CONFIG_FAIR_GROUP_SCHED
408 * h_load = weight * f(tg)
410 * Where f(tg) is the recursive weight fraction assigned to
413 unsigned long h_load
;
414 u64 last_h_load_update
;
415 struct sched_entity
*h_load_next
;
416 #endif /* CONFIG_FAIR_GROUP_SCHED */
417 #endif /* CONFIG_SMP */
419 #ifdef CONFIG_FAIR_GROUP_SCHED
420 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
423 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
424 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
425 * (like users, containers etc.)
427 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
428 * list is used during load balance.
431 struct list_head leaf_cfs_rq_list
;
432 struct task_group
*tg
; /* group that "owns" this runqueue */
434 #ifdef CONFIG_CFS_BANDWIDTH
437 s64 runtime_remaining
;
439 u64 throttled_clock
, throttled_clock_task
;
440 u64 throttled_clock_task_time
;
441 int throttled
, throttle_count
;
442 struct list_head throttled_list
;
443 #endif /* CONFIG_CFS_BANDWIDTH */
444 #endif /* CONFIG_FAIR_GROUP_SCHED */
447 static inline int rt_bandwidth_enabled(void)
449 return sysctl_sched_rt_runtime
>= 0;
452 /* RT IPI pull logic requires IRQ_WORK */
453 #ifdef CONFIG_IRQ_WORK
454 # define HAVE_RT_PUSH_IPI
457 /* Real-Time classes' related field in a runqueue: */
459 struct rt_prio_array active
;
460 unsigned int rt_nr_running
;
461 unsigned int rr_nr_running
;
462 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
464 int curr
; /* highest queued rt task prio */
466 int next
; /* next highest */
471 unsigned long rt_nr_migratory
;
472 unsigned long rt_nr_total
;
474 struct plist_head pushable_tasks
;
475 #ifdef HAVE_RT_PUSH_IPI
478 struct irq_work push_work
;
479 raw_spinlock_t push_lock
;
481 #endif /* CONFIG_SMP */
487 /* Nests inside the rq lock: */
488 raw_spinlock_t rt_runtime_lock
;
490 #ifdef CONFIG_RT_GROUP_SCHED
491 unsigned long rt_nr_boosted
;
494 struct task_group
*tg
;
498 /* Deadline class' related fields in a runqueue */
500 /* runqueue is an rbtree, ordered by deadline */
501 struct rb_root rb_root
;
502 struct rb_node
*rb_leftmost
;
504 unsigned long dl_nr_running
;
508 * Deadline values of the currently executing and the
509 * earliest ready task on this rq. Caching these facilitates
510 * the decision wether or not a ready but not running task
511 * should migrate somewhere else.
518 unsigned long dl_nr_migratory
;
522 * Tasks on this rq that can be pushed away. They are kept in
523 * an rb-tree, ordered by tasks' deadlines, with caching
524 * of the leftmost (earliest deadline) element.
526 struct rb_root pushable_dl_tasks_root
;
527 struct rb_node
*pushable_dl_tasks_leftmost
;
536 * We add the notion of a root-domain which will be used to define per-domain
537 * variables. Each exclusive cpuset essentially defines an island domain by
538 * fully partitioning the member cpus from any other cpuset. Whenever a new
539 * exclusive cpuset is created, we also create and attach a new root-domain
548 cpumask_var_t online
;
550 /* Indicate more than one runnable task for any CPU */
554 * The bit corresponding to a CPU gets set here if such CPU has more
555 * than one runnable -deadline task (as it is below for RT tasks).
557 cpumask_var_t dlo_mask
;
563 * The "RT overload" flag: it gets set if a CPU has more than
564 * one runnable RT task.
566 cpumask_var_t rto_mask
;
567 struct cpupri cpupri
;
569 unsigned long max_cpu_capacity
;
572 extern struct root_domain def_root_domain
;
574 #endif /* CONFIG_SMP */
577 * This is the main, per-CPU runqueue data structure.
579 * Locking rule: those places that want to lock multiple runqueues
580 * (such as the load balancing or the thread migration code), lock
581 * acquire operations must be ordered by ascending &runqueue.
588 * nr_running and cpu_load should be in the same cacheline because
589 * remote CPUs use both these fields when doing load calculation.
591 unsigned int nr_running
;
592 #ifdef CONFIG_NUMA_BALANCING
593 unsigned int nr_numa_running
;
594 unsigned int nr_preferred_running
;
596 #define CPU_LOAD_IDX_MAX 5
597 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
598 #ifdef CONFIG_NO_HZ_COMMON
600 unsigned long last_load_update_tick
;
601 #endif /* CONFIG_SMP */
602 unsigned long nohz_flags
;
603 #endif /* CONFIG_NO_HZ_COMMON */
604 #ifdef CONFIG_NO_HZ_FULL
605 unsigned long last_sched_tick
;
607 /* capture load from *all* tasks on this cpu: */
608 struct load_weight load
;
609 unsigned long nr_load_updates
;
616 #ifdef CONFIG_FAIR_GROUP_SCHED
617 /* list of leaf cfs_rq on this cpu: */
618 struct list_head leaf_cfs_rq_list
;
619 #endif /* CONFIG_FAIR_GROUP_SCHED */
622 * This is part of a global counter where only the total sum
623 * over all CPUs matters. A task can increase this counter on
624 * one CPU and if it got migrated afterwards it may decrease
625 * it on another CPU. Always updated under the runqueue lock:
627 unsigned long nr_uninterruptible
;
629 struct task_struct
*curr
, *idle
, *stop
;
630 unsigned long next_balance
;
631 struct mm_struct
*prev_mm
;
633 unsigned int clock_skip_update
;
640 struct root_domain
*rd
;
641 struct sched_domain
*sd
;
643 unsigned long cpu_capacity
;
644 unsigned long cpu_capacity_orig
;
646 struct callback_head
*balance_callback
;
648 unsigned char idle_balance
;
649 /* For active balancing */
652 struct cpu_stop_work active_balance_work
;
653 /* cpu of this runqueue: */
657 struct list_head cfs_tasks
;
664 /* This is used to determine avg_idle's max value */
665 u64 max_idle_balance_cost
;
668 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
671 #ifdef CONFIG_PARAVIRT
674 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
675 u64 prev_steal_time_rq
;
678 /* calc_load related fields */
679 unsigned long calc_load_update
;
680 long calc_load_active
;
682 #ifdef CONFIG_SCHED_HRTICK
684 int hrtick_csd_pending
;
685 struct call_single_data hrtick_csd
;
687 struct hrtimer hrtick_timer
;
690 #ifdef CONFIG_SCHEDSTATS
692 struct sched_info rq_sched_info
;
693 unsigned long long rq_cpu_time
;
694 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
696 /* sys_sched_yield() stats */
697 unsigned int yld_count
;
699 /* schedule() stats */
700 unsigned int sched_count
;
701 unsigned int sched_goidle
;
703 /* try_to_wake_up() stats */
704 unsigned int ttwu_count
;
705 unsigned int ttwu_local
;
709 struct llist_head wake_list
;
712 #ifdef CONFIG_CPU_IDLE
713 /* Must be inspected within a rcu lock section */
714 struct cpuidle_state
*idle_state
;
718 static inline int cpu_of(struct rq
*rq
)
727 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
729 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
730 #define this_rq() this_cpu_ptr(&runqueues)
731 #define task_rq(p) cpu_rq(task_cpu(p))
732 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
733 #define raw_rq() raw_cpu_ptr(&runqueues)
735 static inline u64
__rq_clock_broken(struct rq
*rq
)
737 return READ_ONCE(rq
->clock
);
740 static inline u64
rq_clock(struct rq
*rq
)
742 lockdep_assert_held(&rq
->lock
);
746 static inline u64
rq_clock_task(struct rq
*rq
)
748 lockdep_assert_held(&rq
->lock
);
749 return rq
->clock_task
;
752 #define RQCF_REQ_SKIP 0x01
753 #define RQCF_ACT_SKIP 0x02
755 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
757 lockdep_assert_held(&rq
->lock
);
759 rq
->clock_skip_update
|= RQCF_REQ_SKIP
;
761 rq
->clock_skip_update
&= ~RQCF_REQ_SKIP
;
765 enum numa_topology_type
{
770 extern enum numa_topology_type sched_numa_topology_type
;
771 extern int sched_max_numa_distance
;
772 extern bool find_numa_distance(int distance
);
775 #ifdef CONFIG_NUMA_BALANCING
776 /* The regions in numa_faults array from task_struct */
777 enum numa_faults_stats
{
783 extern void sched_setnuma(struct task_struct
*p
, int node
);
784 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
785 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
786 #endif /* CONFIG_NUMA_BALANCING */
791 queue_balance_callback(struct rq
*rq
,
792 struct callback_head
*head
,
793 void (*func
)(struct rq
*rq
))
795 lockdep_assert_held(&rq
->lock
);
797 if (unlikely(head
->next
))
800 head
->func
= (void (*)(struct callback_head
*))func
;
801 head
->next
= rq
->balance_callback
;
802 rq
->balance_callback
= head
;
805 extern void sched_ttwu_pending(void);
807 #define rcu_dereference_check_sched_domain(p) \
808 rcu_dereference_check((p), \
809 lockdep_is_held(&sched_domains_mutex))
812 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
813 * See detach_destroy_domains: synchronize_sched for details.
815 * The domain tree of any CPU may only be accessed from within
816 * preempt-disabled sections.
818 #define for_each_domain(cpu, __sd) \
819 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
820 __sd; __sd = __sd->parent)
822 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
825 * highest_flag_domain - Return highest sched_domain containing flag.
826 * @cpu: The cpu whose highest level of sched domain is to
828 * @flag: The flag to check for the highest sched_domain
831 * Returns the highest sched_domain of a cpu which contains the given flag.
833 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
835 struct sched_domain
*sd
, *hsd
= NULL
;
837 for_each_domain(cpu
, sd
) {
838 if (!(sd
->flags
& flag
))
846 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
848 struct sched_domain
*sd
;
850 for_each_domain(cpu
, sd
) {
851 if (sd
->flags
& flag
)
858 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
859 DECLARE_PER_CPU(int, sd_llc_size
);
860 DECLARE_PER_CPU(int, sd_llc_id
);
861 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
862 DECLARE_PER_CPU(struct sched_domain
*, sd_busy
);
863 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
865 struct sched_group_capacity
{
868 * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity
871 unsigned int capacity
;
872 unsigned long next_update
;
873 int imbalance
; /* XXX unrelated to capacity but shared group state */
875 * Number of busy cpus in this group.
877 atomic_t nr_busy_cpus
;
879 unsigned long cpumask
[0]; /* iteration mask */
883 struct sched_group
*next
; /* Must be a circular list */
886 unsigned int group_weight
;
887 struct sched_group_capacity
*sgc
;
890 * The CPUs this group covers.
892 * NOTE: this field is variable length. (Allocated dynamically
893 * by attaching extra space to the end of the structure,
894 * depending on how many CPUs the kernel has booted up with)
896 unsigned long cpumask
[0];
899 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
901 return to_cpumask(sg
->cpumask
);
905 * cpumask masking which cpus in the group are allowed to iterate up the domain
908 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
910 return to_cpumask(sg
->sgc
->cpumask
);
914 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
915 * @group: The group whose first cpu is to be returned.
917 static inline unsigned int group_first_cpu(struct sched_group
*group
)
919 return cpumask_first(sched_group_cpus(group
));
922 extern int group_balance_cpu(struct sched_group
*sg
);
924 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
925 void register_sched_domain_sysctl(void);
926 void unregister_sched_domain_sysctl(void);
928 static inline void register_sched_domain_sysctl(void)
931 static inline void unregister_sched_domain_sysctl(void)
938 static inline void sched_ttwu_pending(void) { }
940 #endif /* CONFIG_SMP */
943 #include "auto_group.h"
945 #ifdef CONFIG_CGROUP_SCHED
948 * Return the group to which this tasks belongs.
950 * We cannot use task_css() and friends because the cgroup subsystem
951 * changes that value before the cgroup_subsys::attach() method is called,
952 * therefore we cannot pin it and might observe the wrong value.
954 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
955 * core changes this before calling sched_move_task().
957 * Instead we use a 'copy' which is updated from sched_move_task() while
958 * holding both task_struct::pi_lock and rq::lock.
960 static inline struct task_group
*task_group(struct task_struct
*p
)
962 return p
->sched_task_group
;
965 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
966 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
968 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
969 struct task_group
*tg
= task_group(p
);
972 #ifdef CONFIG_FAIR_GROUP_SCHED
973 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
974 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
975 p
->se
.parent
= tg
->se
[cpu
];
978 #ifdef CONFIG_RT_GROUP_SCHED
979 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
980 p
->rt
.parent
= tg
->rt_se
[cpu
];
984 #else /* CONFIG_CGROUP_SCHED */
986 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
987 static inline struct task_group
*task_group(struct task_struct
*p
)
992 #endif /* CONFIG_CGROUP_SCHED */
994 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
999 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1000 * successfuly executed on another CPU. We must ensure that updates of
1001 * per-task data have been completed by this moment.
1004 task_thread_info(p
)->cpu
= cpu
;
1010 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
1012 #ifdef CONFIG_SCHED_DEBUG
1013 # include <linux/static_key.h>
1014 # define const_debug __read_mostly
1016 # define const_debug const
1019 extern const_debug
unsigned int sysctl_sched_features
;
1021 #define SCHED_FEAT(name, enabled) \
1022 __SCHED_FEAT_##name ,
1025 #include "features.h"
1031 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1032 #define SCHED_FEAT(name, enabled) \
1033 static __always_inline bool static_branch_##name(struct static_key *key) \
1035 return static_key_##enabled(key); \
1038 #include "features.h"
1042 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1043 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1044 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1045 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1046 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1048 extern struct static_key_false sched_numa_balancing
;
1049 extern struct static_key_false sched_schedstats
;
1051 static inline u64
global_rt_period(void)
1053 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1056 static inline u64
global_rt_runtime(void)
1058 if (sysctl_sched_rt_runtime
< 0)
1061 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1064 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1066 return rq
->curr
== p
;
1069 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1074 return task_current(rq
, p
);
1078 static inline int task_on_rq_queued(struct task_struct
*p
)
1080 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1083 static inline int task_on_rq_migrating(struct task_struct
*p
)
1085 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1088 #ifndef prepare_arch_switch
1089 # define prepare_arch_switch(next) do { } while (0)
1091 #ifndef finish_arch_post_lock_switch
1092 # define finish_arch_post_lock_switch() do { } while (0)
1095 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1099 * We can optimise this out completely for !SMP, because the
1100 * SMP rebalancing from interrupt is the only thing that cares
1107 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1111 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1112 * We must ensure this doesn't happen until the switch is completely
1115 * In particular, the load of prev->state in finish_task_switch() must
1116 * happen before this.
1118 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
1120 smp_store_release(&prev
->on_cpu
, 0);
1122 #ifdef CONFIG_DEBUG_SPINLOCK
1123 /* this is a valid case when another task releases the spinlock */
1124 rq
->lock
.owner
= current
;
1127 * If we are tracking spinlock dependencies then we have to
1128 * fix up the runqueue lock - which gets 'carried over' from
1129 * prev into current:
1131 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1133 raw_spin_unlock_irq(&rq
->lock
);
1139 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1140 #define WF_FORK 0x02 /* child wakeup after fork */
1141 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1144 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1145 * of tasks with abnormal "nice" values across CPUs the contribution that
1146 * each task makes to its run queue's load is weighted according to its
1147 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1148 * scaled version of the new time slice allocation that they receive on time
1152 #define WEIGHT_IDLEPRIO 3
1153 #define WMULT_IDLEPRIO 1431655765
1155 extern const int sched_prio_to_weight
[40];
1156 extern const u32 sched_prio_to_wmult
[40];
1159 * {de,en}queue flags:
1161 * DEQUEUE_SLEEP - task is no longer runnable
1162 * ENQUEUE_WAKEUP - task just became runnable
1164 * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks
1165 * are in a known state which allows modification. Such pairs
1166 * should preserve as much state as possible.
1168 * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location
1171 * ENQUEUE_HEAD - place at front of runqueue (tail if not specified)
1172 * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline)
1173 * ENQUEUE_MIGRATED - the task was migrated during wakeup
1177 #define DEQUEUE_SLEEP 0x01
1178 #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */
1179 #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */
1181 #define ENQUEUE_WAKEUP 0x01
1182 #define ENQUEUE_RESTORE 0x02
1183 #define ENQUEUE_MOVE 0x04
1185 #define ENQUEUE_HEAD 0x08
1186 #define ENQUEUE_REPLENISH 0x10
1188 #define ENQUEUE_MIGRATED 0x20
1190 #define ENQUEUE_MIGRATED 0x00
1193 #define RETRY_TASK ((void *)-1UL)
1195 struct sched_class
{
1196 const struct sched_class
*next
;
1198 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1199 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1200 void (*yield_task
) (struct rq
*rq
);
1201 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1203 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1206 * It is the responsibility of the pick_next_task() method that will
1207 * return the next task to call put_prev_task() on the @prev task or
1208 * something equivalent.
1210 * May return RETRY_TASK when it finds a higher prio class has runnable
1213 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1214 struct task_struct
*prev
,
1215 struct pin_cookie cookie
);
1216 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1219 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1220 void (*migrate_task_rq
)(struct task_struct
*p
);
1222 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1224 void (*set_cpus_allowed
)(struct task_struct
*p
,
1225 const struct cpumask
*newmask
);
1227 void (*rq_online
)(struct rq
*rq
);
1228 void (*rq_offline
)(struct rq
*rq
);
1231 void (*set_curr_task
) (struct rq
*rq
);
1232 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1233 void (*task_fork
) (struct task_struct
*p
);
1234 void (*task_dead
) (struct task_struct
*p
);
1237 * The switched_from() call is allowed to drop rq->lock, therefore we
1238 * cannot assume the switched_from/switched_to pair is serliazed by
1239 * rq->lock. They are however serialized by p->pi_lock.
1241 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1242 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1243 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1246 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1247 struct task_struct
*task
);
1249 void (*update_curr
) (struct rq
*rq
);
1251 #define TASK_SET_GROUP 0
1252 #define TASK_MOVE_GROUP 1
1254 #ifdef CONFIG_FAIR_GROUP_SCHED
1255 void (*task_change_group
) (struct task_struct
*p
, int type
);
1259 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1261 prev
->sched_class
->put_prev_task(rq
, prev
);
1264 #define sched_class_highest (&stop_sched_class)
1265 #define for_each_class(class) \
1266 for (class = sched_class_highest; class; class = class->next)
1268 extern const struct sched_class stop_sched_class
;
1269 extern const struct sched_class dl_sched_class
;
1270 extern const struct sched_class rt_sched_class
;
1271 extern const struct sched_class fair_sched_class
;
1272 extern const struct sched_class idle_sched_class
;
1277 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1279 extern void trigger_load_balance(struct rq
*rq
);
1281 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1285 #ifdef CONFIG_CPU_IDLE
1286 static inline void idle_set_state(struct rq
*rq
,
1287 struct cpuidle_state
*idle_state
)
1289 rq
->idle_state
= idle_state
;
1292 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1294 WARN_ON(!rcu_read_lock_held());
1295 return rq
->idle_state
;
1298 static inline void idle_set_state(struct rq
*rq
,
1299 struct cpuidle_state
*idle_state
)
1303 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1309 extern void sysrq_sched_debug_show(void);
1310 extern void sched_init_granularity(void);
1311 extern void update_max_interval(void);
1313 extern void init_sched_dl_class(void);
1314 extern void init_sched_rt_class(void);
1315 extern void init_sched_fair_class(void);
1317 extern void resched_curr(struct rq
*rq
);
1318 extern void resched_cpu(int cpu
);
1320 extern struct rt_bandwidth def_rt_bandwidth
;
1321 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1323 extern struct dl_bandwidth def_dl_bandwidth
;
1324 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1325 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1327 unsigned long to_ratio(u64 period
, u64 runtime
);
1329 extern void init_entity_runnable_average(struct sched_entity
*se
);
1330 extern void post_init_entity_util_avg(struct sched_entity
*se
);
1332 #ifdef CONFIG_NO_HZ_FULL
1333 extern bool sched_can_stop_tick(struct rq
*rq
);
1336 * Tick may be needed by tasks in the runqueue depending on their policy and
1337 * requirements. If tick is needed, lets send the target an IPI to kick it out of
1338 * nohz mode if necessary.
1340 static inline void sched_update_tick_dependency(struct rq
*rq
)
1344 if (!tick_nohz_full_enabled())
1349 if (!tick_nohz_full_cpu(cpu
))
1352 if (sched_can_stop_tick(rq
))
1353 tick_nohz_dep_clear_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1355 tick_nohz_dep_set_cpu(cpu
, TICK_DEP_BIT_SCHED
);
1358 static inline void sched_update_tick_dependency(struct rq
*rq
) { }
1361 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1363 unsigned prev_nr
= rq
->nr_running
;
1365 rq
->nr_running
= prev_nr
+ count
;
1367 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1369 if (!rq
->rd
->overload
)
1370 rq
->rd
->overload
= true;
1374 sched_update_tick_dependency(rq
);
1377 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1379 rq
->nr_running
-= count
;
1380 /* Check if we still need preemption */
1381 sched_update_tick_dependency(rq
);
1384 static inline void rq_last_tick_reset(struct rq
*rq
)
1386 #ifdef CONFIG_NO_HZ_FULL
1387 rq
->last_sched_tick
= jiffies
;
1391 extern void update_rq_clock(struct rq
*rq
);
1393 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1394 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1396 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1398 extern const_debug
unsigned int sysctl_sched_time_avg
;
1399 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1400 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1402 static inline u64
sched_avg_period(void)
1404 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1407 #ifdef CONFIG_SCHED_HRTICK
1411 * - enabled by features
1412 * - hrtimer is actually high res
1414 static inline int hrtick_enabled(struct rq
*rq
)
1416 if (!sched_feat(HRTICK
))
1418 if (!cpu_active(cpu_of(rq
)))
1420 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1423 void hrtick_start(struct rq
*rq
, u64 delay
);
1427 static inline int hrtick_enabled(struct rq
*rq
)
1432 #endif /* CONFIG_SCHED_HRTICK */
1435 extern void sched_avg_update(struct rq
*rq
);
1437 #ifndef arch_scale_freq_capacity
1438 static __always_inline
1439 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1441 return SCHED_CAPACITY_SCALE
;
1445 #ifndef arch_scale_cpu_capacity
1446 static __always_inline
1447 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1449 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1450 return sd
->smt_gain
/ sd
->span_weight
;
1452 return SCHED_CAPACITY_SCALE
;
1456 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1458 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1459 sched_avg_update(rq
);
1462 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1463 static inline void sched_avg_update(struct rq
*rq
) { }
1467 unsigned long flags
;
1468 struct pin_cookie cookie
;
1471 struct rq
*__task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1472 __acquires(rq
->lock
);
1473 struct rq
*task_rq_lock(struct task_struct
*p
, struct rq_flags
*rf
)
1474 __acquires(p
->pi_lock
)
1475 __acquires(rq
->lock
);
1477 static inline void __task_rq_unlock(struct rq
*rq
, struct rq_flags
*rf
)
1478 __releases(rq
->lock
)
1480 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
1481 raw_spin_unlock(&rq
->lock
);
1485 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, struct rq_flags
*rf
)
1486 __releases(rq
->lock
)
1487 __releases(p
->pi_lock
)
1489 lockdep_unpin_lock(&rq
->lock
, rf
->cookie
);
1490 raw_spin_unlock(&rq
->lock
);
1491 raw_spin_unlock_irqrestore(&p
->pi_lock
, rf
->flags
);
1495 #ifdef CONFIG_PREEMPT
1497 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1500 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1501 * way at the expense of forcing extra atomic operations in all
1502 * invocations. This assures that the double_lock is acquired using the
1503 * same underlying policy as the spinlock_t on this architecture, which
1504 * reduces latency compared to the unfair variant below. However, it
1505 * also adds more overhead and therefore may reduce throughput.
1507 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1508 __releases(this_rq
->lock
)
1509 __acquires(busiest
->lock
)
1510 __acquires(this_rq
->lock
)
1512 raw_spin_unlock(&this_rq
->lock
);
1513 double_rq_lock(this_rq
, busiest
);
1520 * Unfair double_lock_balance: Optimizes throughput at the expense of
1521 * latency by eliminating extra atomic operations when the locks are
1522 * already in proper order on entry. This favors lower cpu-ids and will
1523 * grant the double lock to lower cpus over higher ids under contention,
1524 * regardless of entry order into the function.
1526 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1527 __releases(this_rq
->lock
)
1528 __acquires(busiest
->lock
)
1529 __acquires(this_rq
->lock
)
1533 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1534 if (busiest
< this_rq
) {
1535 raw_spin_unlock(&this_rq
->lock
);
1536 raw_spin_lock(&busiest
->lock
);
1537 raw_spin_lock_nested(&this_rq
->lock
,
1538 SINGLE_DEPTH_NESTING
);
1541 raw_spin_lock_nested(&busiest
->lock
,
1542 SINGLE_DEPTH_NESTING
);
1547 #endif /* CONFIG_PREEMPT */
1550 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1552 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1554 if (unlikely(!irqs_disabled())) {
1555 /* printk() doesn't work good under rq->lock */
1556 raw_spin_unlock(&this_rq
->lock
);
1560 return _double_lock_balance(this_rq
, busiest
);
1563 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1564 __releases(busiest
->lock
)
1566 raw_spin_unlock(&busiest
->lock
);
1567 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1570 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1576 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1579 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1585 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1588 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1594 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1598 * double_rq_lock - safely lock two runqueues
1600 * Note this does not disable interrupts like task_rq_lock,
1601 * you need to do so manually before calling.
1603 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1604 __acquires(rq1
->lock
)
1605 __acquires(rq2
->lock
)
1607 BUG_ON(!irqs_disabled());
1609 raw_spin_lock(&rq1
->lock
);
1610 __acquire(rq2
->lock
); /* Fake it out ;) */
1613 raw_spin_lock(&rq1
->lock
);
1614 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1616 raw_spin_lock(&rq2
->lock
);
1617 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1623 * double_rq_unlock - safely unlock two runqueues
1625 * Note this does not restore interrupts like task_rq_unlock,
1626 * you need to do so manually after calling.
1628 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1629 __releases(rq1
->lock
)
1630 __releases(rq2
->lock
)
1632 raw_spin_unlock(&rq1
->lock
);
1634 raw_spin_unlock(&rq2
->lock
);
1636 __release(rq2
->lock
);
1639 #else /* CONFIG_SMP */
1642 * double_rq_lock - safely lock two runqueues
1644 * Note this does not disable interrupts like task_rq_lock,
1645 * you need to do so manually before calling.
1647 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1648 __acquires(rq1
->lock
)
1649 __acquires(rq2
->lock
)
1651 BUG_ON(!irqs_disabled());
1653 raw_spin_lock(&rq1
->lock
);
1654 __acquire(rq2
->lock
); /* Fake it out ;) */
1658 * double_rq_unlock - safely unlock two runqueues
1660 * Note this does not restore interrupts like task_rq_unlock,
1661 * you need to do so manually after calling.
1663 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1664 __releases(rq1
->lock
)
1665 __releases(rq2
->lock
)
1668 raw_spin_unlock(&rq1
->lock
);
1669 __release(rq2
->lock
);
1674 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1675 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1677 #ifdef CONFIG_SCHED_DEBUG
1678 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1679 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1680 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1682 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1684 #ifdef CONFIG_NUMA_BALANCING
1686 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1688 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1689 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1690 #endif /* CONFIG_NUMA_BALANCING */
1691 #endif /* CONFIG_SCHED_DEBUG */
1693 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1694 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1695 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1697 extern void cfs_bandwidth_usage_inc(void);
1698 extern void cfs_bandwidth_usage_dec(void);
1700 #ifdef CONFIG_NO_HZ_COMMON
1701 enum rq_nohz_flag_bits
{
1706 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1708 extern void nohz_balance_exit_idle(unsigned int cpu
);
1710 static inline void nohz_balance_exit_idle(unsigned int cpu
) { }
1713 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1715 DECLARE_PER_CPU(u64
, cpu_hardirq_time
);
1716 DECLARE_PER_CPU(u64
, cpu_softirq_time
);
1718 #ifndef CONFIG_64BIT
1719 DECLARE_PER_CPU(seqcount_t
, irq_time_seq
);
1721 static inline void irq_time_write_begin(void)
1723 __this_cpu_inc(irq_time_seq
.sequence
);
1727 static inline void irq_time_write_end(void)
1730 __this_cpu_inc(irq_time_seq
.sequence
);
1733 static inline u64
irq_time_read(int cpu
)
1739 seq
= read_seqcount_begin(&per_cpu(irq_time_seq
, cpu
));
1740 irq_time
= per_cpu(cpu_softirq_time
, cpu
) +
1741 per_cpu(cpu_hardirq_time
, cpu
);
1742 } while (read_seqcount_retry(&per_cpu(irq_time_seq
, cpu
), seq
));
1746 #else /* CONFIG_64BIT */
1747 static inline void irq_time_write_begin(void)
1751 static inline void irq_time_write_end(void)
1755 static inline u64
irq_time_read(int cpu
)
1757 return per_cpu(cpu_softirq_time
, cpu
) + per_cpu(cpu_hardirq_time
, cpu
);
1759 #endif /* CONFIG_64BIT */
1760 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1762 #ifdef CONFIG_CPU_FREQ
1763 DECLARE_PER_CPU(struct update_util_data
*, cpufreq_update_util_data
);
1766 * cpufreq_update_util - Take a note about CPU utilization changes.
1767 * @rq: Runqueue to carry out the update for.
1768 * @flags: Update reason flags.
1770 * This function is called by the scheduler on the CPU whose utilization is
1773 * It can only be called from RCU-sched read-side critical sections.
1775 * The way cpufreq is currently arranged requires it to evaluate the CPU
1776 * performance state (frequency/voltage) on a regular basis to prevent it from
1777 * being stuck in a completely inadequate performance level for too long.
1778 * That is not guaranteed to happen if the updates are only triggered from CFS,
1779 * though, because they may not be coming in if RT or deadline tasks are active
1780 * all the time (or there are RT and DL tasks only).
1782 * As a workaround for that issue, this function is called by the RT and DL
1783 * sched classes to trigger extra cpufreq updates to prevent it from stalling,
1784 * but that really is a band-aid. Going forward it should be replaced with
1785 * solutions targeted more specifically at RT and DL tasks.
1787 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
)
1789 struct update_util_data
*data
;
1791 data
= rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data
));
1793 data
->func(data
, rq_clock(rq
), flags
);
1796 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
)
1798 if (cpu_of(rq
) == smp_processor_id())
1799 cpufreq_update_util(rq
, flags
);
1802 static inline void cpufreq_update_util(struct rq
*rq
, unsigned int flags
) {}
1803 static inline void cpufreq_update_this_cpu(struct rq
*rq
, unsigned int flags
) {}
1804 #endif /* CONFIG_CPU_FREQ */
1806 #ifdef arch_scale_freq_capacity
1807 #ifndef arch_scale_freq_invariant
1808 #define arch_scale_freq_invariant() (true)
1810 #else /* arch_scale_freq_capacity */
1811 #define arch_scale_freq_invariant() (false)