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/mutex.h>
7 #include <linux/spinlock.h>
8 #include <linux/stop_machine.h>
9 #include <linux/irq_work.h>
10 #include <linux/tick.h>
11 #include <linux/slab.h>
14 #include "cpudeadline.h"
20 /* task_struct::on_rq states: */
21 #define TASK_ON_RQ_QUEUED 1
22 #define TASK_ON_RQ_MIGRATING 2
24 extern __read_mostly
int scheduler_running
;
26 extern unsigned long calc_load_update
;
27 extern atomic_long_t calc_load_tasks
;
29 extern void calc_global_load_tick(struct rq
*this_rq
);
30 extern long calc_load_fold_active(struct rq
*this_rq
);
33 extern void update_cpu_load_active(struct rq
*this_rq
);
35 static inline void update_cpu_load_active(struct rq
*this_rq
) { }
39 * Helpers for converting nanosecond timing to jiffy resolution
41 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
44 * Increase resolution of nice-level calculations for 64-bit architectures.
45 * The extra resolution improves shares distribution and load balancing of
46 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
47 * hierarchies, especially on larger systems. This is not a user-visible change
48 * and does not change the user-interface for setting shares/weights.
50 * We increase resolution only if we have enough bits to allow this increased
51 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
52 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
55 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
56 # define SCHED_LOAD_RESOLUTION 10
57 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
58 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
60 # define SCHED_LOAD_RESOLUTION 0
61 # define scale_load(w) (w)
62 # define scale_load_down(w) (w)
65 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
66 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
68 #define NICE_0_LOAD SCHED_LOAD_SCALE
69 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
72 * Single value that decides SCHED_DEADLINE internal math precision.
73 * 10 -> just above 1us
74 * 9 -> just above 0.5us
79 * These are the 'tuning knobs' of the scheduler:
83 * single value that denotes runtime == period, ie unlimited time.
85 #define RUNTIME_INF ((u64)~0ULL)
87 static inline int idle_policy(int policy
)
89 return policy
== SCHED_IDLE
;
91 static inline int fair_policy(int policy
)
93 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
96 static inline int rt_policy(int policy
)
98 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
101 static inline int dl_policy(int policy
)
103 return policy
== SCHED_DEADLINE
;
105 static inline bool valid_policy(int policy
)
107 return idle_policy(policy
) || fair_policy(policy
) ||
108 rt_policy(policy
) || dl_policy(policy
);
111 static inline int task_has_rt_policy(struct task_struct
*p
)
113 return rt_policy(p
->policy
);
116 static inline int task_has_dl_policy(struct task_struct
*p
)
118 return dl_policy(p
->policy
);
122 * Tells if entity @a should preempt entity @b.
125 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
127 return dl_time_before(a
->deadline
, b
->deadline
);
131 * This is the priority-queue data structure of the RT scheduling class:
133 struct rt_prio_array
{
134 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
135 struct list_head queue
[MAX_RT_PRIO
];
138 struct rt_bandwidth
{
139 /* nests inside the rq lock: */
140 raw_spinlock_t rt_runtime_lock
;
143 struct hrtimer rt_period_timer
;
144 unsigned int rt_period_active
;
147 void __dl_clear_params(struct task_struct
*p
);
150 * To keep the bandwidth of -deadline tasks and groups under control
151 * we need some place where:
152 * - store the maximum -deadline bandwidth of the system (the group);
153 * - cache the fraction of that bandwidth that is currently allocated.
155 * This is all done in the data structure below. It is similar to the
156 * one used for RT-throttling (rt_bandwidth), with the main difference
157 * that, since here we are only interested in admission control, we
158 * do not decrease any runtime while the group "executes", neither we
159 * need a timer to replenish it.
161 * With respect to SMP, the bandwidth is given on a per-CPU basis,
163 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
164 * - dl_total_bw array contains, in the i-eth element, the currently
165 * allocated bandwidth on the i-eth CPU.
166 * Moreover, groups consume bandwidth on each CPU, while tasks only
167 * consume bandwidth on the CPU they're running on.
168 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
169 * that will be shown the next time the proc or cgroup controls will
170 * be red. It on its turn can be changed by writing on its own
173 struct dl_bandwidth
{
174 raw_spinlock_t dl_runtime_lock
;
179 static inline int dl_bandwidth_enabled(void)
181 return sysctl_sched_rt_runtime
>= 0;
184 extern struct dl_bw
*dl_bw_of(int i
);
192 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
194 dl_b
->total_bw
-= tsk_bw
;
198 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
200 dl_b
->total_bw
+= tsk_bw
;
204 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
206 return dl_b
->bw
!= -1 &&
207 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
210 extern struct mutex sched_domains_mutex
;
212 #ifdef CONFIG_CGROUP_SCHED
214 #include <linux/cgroup.h>
219 extern struct list_head task_groups
;
221 struct cfs_bandwidth
{
222 #ifdef CONFIG_CFS_BANDWIDTH
226 s64 hierarchical_quota
;
229 int idle
, period_active
;
230 struct hrtimer period_timer
, slack_timer
;
231 struct list_head throttled_cfs_rq
;
234 int nr_periods
, nr_throttled
;
239 /* task group related information */
241 struct cgroup_subsys_state css
;
243 #ifdef CONFIG_FAIR_GROUP_SCHED
244 /* schedulable entities of this group on each cpu */
245 struct sched_entity
**se
;
246 /* runqueue "owned" by this group on each cpu */
247 struct cfs_rq
**cfs_rq
;
248 unsigned long shares
;
251 atomic_long_t load_avg
;
255 #ifdef CONFIG_RT_GROUP_SCHED
256 struct sched_rt_entity
**rt_se
;
257 struct rt_rq
**rt_rq
;
259 struct rt_bandwidth rt_bandwidth
;
263 struct list_head list
;
265 struct task_group
*parent
;
266 struct list_head siblings
;
267 struct list_head children
;
269 #ifdef CONFIG_SCHED_AUTOGROUP
270 struct autogroup
*autogroup
;
273 struct cfs_bandwidth cfs_bandwidth
;
276 #ifdef CONFIG_FAIR_GROUP_SCHED
277 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
280 * A weight of 0 or 1 can cause arithmetics problems.
281 * A weight of a cfs_rq is the sum of weights of which entities
282 * are queued on this cfs_rq, so a weight of a entity should not be
283 * too large, so as the shares value of a task group.
284 * (The default weight is 1024 - so there's no practical
285 * limitation from this.)
287 #define MIN_SHARES (1UL << 1)
288 #define MAX_SHARES (1UL << 18)
291 typedef int (*tg_visitor
)(struct task_group
*, void *);
293 extern int walk_tg_tree_from(struct task_group
*from
,
294 tg_visitor down
, tg_visitor up
, void *data
);
297 * Iterate the full tree, calling @down when first entering a node and @up when
298 * leaving it for the final time.
300 * Caller must hold rcu_lock or sufficient equivalent.
302 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
304 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
307 extern int tg_nop(struct task_group
*tg
, void *data
);
309 extern void free_fair_sched_group(struct task_group
*tg
);
310 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
311 extern void unregister_fair_sched_group(struct task_group
*tg
, int cpu
);
312 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
313 struct sched_entity
*se
, int cpu
,
314 struct sched_entity
*parent
);
315 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
316 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
318 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
319 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
320 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
322 extern void free_rt_sched_group(struct task_group
*tg
);
323 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
324 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
325 struct sched_rt_entity
*rt_se
, int cpu
,
326 struct sched_rt_entity
*parent
);
328 extern struct task_group
*sched_create_group(struct task_group
*parent
);
329 extern void sched_online_group(struct task_group
*tg
,
330 struct task_group
*parent
);
331 extern void sched_destroy_group(struct task_group
*tg
);
332 extern void sched_offline_group(struct task_group
*tg
);
334 extern void sched_move_task(struct task_struct
*tsk
);
336 #ifdef CONFIG_FAIR_GROUP_SCHED
337 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
340 extern void set_task_rq_fair(struct sched_entity
*se
,
341 struct cfs_rq
*prev
, struct cfs_rq
*next
);
342 #else /* !CONFIG_SMP */
343 static inline void set_task_rq_fair(struct sched_entity
*se
,
344 struct cfs_rq
*prev
, struct cfs_rq
*next
) { }
345 #endif /* CONFIG_SMP */
346 #endif /* CONFIG_FAIR_GROUP_SCHED */
348 #else /* CONFIG_CGROUP_SCHED */
350 struct cfs_bandwidth
{ };
352 #endif /* CONFIG_CGROUP_SCHED */
354 /* CFS-related fields in a runqueue */
356 struct load_weight load
;
357 unsigned int nr_running
, h_nr_running
;
362 u64 min_vruntime_copy
;
365 struct rb_root tasks_timeline
;
366 struct rb_node
*rb_leftmost
;
369 * 'curr' points to currently running entity on this cfs_rq.
370 * It is set to NULL otherwise (i.e when none are currently running).
372 struct sched_entity
*curr
, *next
, *last
, *skip
;
374 #ifdef CONFIG_SCHED_DEBUG
375 unsigned int nr_spread_over
;
382 struct sched_avg avg
;
383 u64 runnable_load_sum
;
384 unsigned long runnable_load_avg
;
385 #ifdef CONFIG_FAIR_GROUP_SCHED
386 unsigned long tg_load_avg_contrib
;
388 atomic_long_t removed_load_avg
, removed_util_avg
;
390 u64 load_last_update_time_copy
;
393 #ifdef CONFIG_FAIR_GROUP_SCHED
395 * h_load = weight * f(tg)
397 * Where f(tg) is the recursive weight fraction assigned to
400 unsigned long h_load
;
401 u64 last_h_load_update
;
402 struct sched_entity
*h_load_next
;
403 #endif /* CONFIG_FAIR_GROUP_SCHED */
404 #endif /* CONFIG_SMP */
406 #ifdef CONFIG_FAIR_GROUP_SCHED
407 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
410 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
411 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
412 * (like users, containers etc.)
414 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
415 * list is used during load balance.
418 struct list_head leaf_cfs_rq_list
;
419 struct task_group
*tg
; /* group that "owns" this runqueue */
421 #ifdef CONFIG_CFS_BANDWIDTH
424 s64 runtime_remaining
;
426 u64 throttled_clock
, throttled_clock_task
;
427 u64 throttled_clock_task_time
;
428 int throttled
, throttle_count
;
429 struct list_head throttled_list
;
430 #endif /* CONFIG_CFS_BANDWIDTH */
431 #endif /* CONFIG_FAIR_GROUP_SCHED */
434 static inline int rt_bandwidth_enabled(void)
436 return sysctl_sched_rt_runtime
>= 0;
439 /* RT IPI pull logic requires IRQ_WORK */
440 #ifdef CONFIG_IRQ_WORK
441 # define HAVE_RT_PUSH_IPI
444 /* Real-Time classes' related field in a runqueue: */
446 struct rt_prio_array active
;
447 unsigned int rt_nr_running
;
448 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
450 int curr
; /* highest queued rt task prio */
452 int next
; /* next highest */
457 unsigned long rt_nr_migratory
;
458 unsigned long rt_nr_total
;
460 struct plist_head pushable_tasks
;
461 #ifdef HAVE_RT_PUSH_IPI
464 struct irq_work push_work
;
465 raw_spinlock_t push_lock
;
467 #endif /* CONFIG_SMP */
473 /* Nests inside the rq lock: */
474 raw_spinlock_t rt_runtime_lock
;
476 #ifdef CONFIG_RT_GROUP_SCHED
477 unsigned long rt_nr_boosted
;
480 struct task_group
*tg
;
484 /* Deadline class' related fields in a runqueue */
486 /* runqueue is an rbtree, ordered by deadline */
487 struct rb_root rb_root
;
488 struct rb_node
*rb_leftmost
;
490 unsigned long dl_nr_running
;
494 * Deadline values of the currently executing and the
495 * earliest ready task on this rq. Caching these facilitates
496 * the decision wether or not a ready but not running task
497 * should migrate somewhere else.
504 unsigned long dl_nr_migratory
;
508 * Tasks on this rq that can be pushed away. They are kept in
509 * an rb-tree, ordered by tasks' deadlines, with caching
510 * of the leftmost (earliest deadline) element.
512 struct rb_root pushable_dl_tasks_root
;
513 struct rb_node
*pushable_dl_tasks_leftmost
;
522 * We add the notion of a root-domain which will be used to define per-domain
523 * variables. Each exclusive cpuset essentially defines an island domain by
524 * fully partitioning the member cpus from any other cpuset. Whenever a new
525 * exclusive cpuset is created, we also create and attach a new root-domain
534 cpumask_var_t online
;
536 /* Indicate more than one runnable task for any CPU */
540 * The bit corresponding to a CPU gets set here if such CPU has more
541 * than one runnable -deadline task (as it is below for RT tasks).
543 cpumask_var_t dlo_mask
;
549 * The "RT overload" flag: it gets set if a CPU has more than
550 * one runnable RT task.
552 cpumask_var_t rto_mask
;
553 struct cpupri cpupri
;
556 extern struct root_domain def_root_domain
;
558 #endif /* CONFIG_SMP */
561 * This is the main, per-CPU runqueue data structure.
563 * Locking rule: those places that want to lock multiple runqueues
564 * (such as the load balancing or the thread migration code), lock
565 * acquire operations must be ordered by ascending &runqueue.
572 * nr_running and cpu_load should be in the same cacheline because
573 * remote CPUs use both these fields when doing load calculation.
575 unsigned int nr_running
;
576 #ifdef CONFIG_NUMA_BALANCING
577 unsigned int nr_numa_running
;
578 unsigned int nr_preferred_running
;
580 #define CPU_LOAD_IDX_MAX 5
581 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
582 unsigned long last_load_update_tick
;
583 #ifdef CONFIG_NO_HZ_COMMON
585 unsigned long nohz_flags
;
587 #ifdef CONFIG_NO_HZ_FULL
588 unsigned long last_sched_tick
;
590 /* capture load from *all* tasks on this cpu: */
591 struct load_weight load
;
592 unsigned long nr_load_updates
;
599 #ifdef CONFIG_FAIR_GROUP_SCHED
600 /* list of leaf cfs_rq on this cpu: */
601 struct list_head leaf_cfs_rq_list
;
602 #endif /* CONFIG_FAIR_GROUP_SCHED */
605 * This is part of a global counter where only the total sum
606 * over all CPUs matters. A task can increase this counter on
607 * one CPU and if it got migrated afterwards it may decrease
608 * it on another CPU. Always updated under the runqueue lock:
610 unsigned long nr_uninterruptible
;
612 struct task_struct
*curr
, *idle
, *stop
;
613 unsigned long next_balance
;
614 struct mm_struct
*prev_mm
;
616 unsigned int clock_skip_update
;
623 struct root_domain
*rd
;
624 struct sched_domain
*sd
;
626 unsigned long cpu_capacity
;
627 unsigned long cpu_capacity_orig
;
629 struct callback_head
*balance_callback
;
631 unsigned char idle_balance
;
632 /* For active balancing */
635 struct cpu_stop_work active_balance_work
;
636 /* cpu of this runqueue: */
640 struct list_head cfs_tasks
;
647 /* This is used to determine avg_idle's max value */
648 u64 max_idle_balance_cost
;
651 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
654 #ifdef CONFIG_PARAVIRT
657 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
658 u64 prev_steal_time_rq
;
661 /* calc_load related fields */
662 unsigned long calc_load_update
;
663 long calc_load_active
;
665 #ifdef CONFIG_SCHED_HRTICK
667 int hrtick_csd_pending
;
668 struct call_single_data hrtick_csd
;
670 struct hrtimer hrtick_timer
;
673 #ifdef CONFIG_SCHEDSTATS
675 struct sched_info rq_sched_info
;
676 unsigned long long rq_cpu_time
;
677 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
679 /* sys_sched_yield() stats */
680 unsigned int yld_count
;
682 /* schedule() stats */
683 unsigned int sched_count
;
684 unsigned int sched_goidle
;
686 /* try_to_wake_up() stats */
687 unsigned int ttwu_count
;
688 unsigned int ttwu_local
;
692 struct llist_head wake_list
;
695 #ifdef CONFIG_CPU_IDLE
696 /* Must be inspected within a rcu lock section */
697 struct cpuidle_state
*idle_state
;
701 static inline int cpu_of(struct rq
*rq
)
710 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
712 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
713 #define this_rq() this_cpu_ptr(&runqueues)
714 #define task_rq(p) cpu_rq(task_cpu(p))
715 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
716 #define raw_rq() raw_cpu_ptr(&runqueues)
718 static inline u64
__rq_clock_broken(struct rq
*rq
)
720 return READ_ONCE(rq
->clock
);
723 static inline u64
rq_clock(struct rq
*rq
)
725 lockdep_assert_held(&rq
->lock
);
729 static inline u64
rq_clock_task(struct rq
*rq
)
731 lockdep_assert_held(&rq
->lock
);
732 return rq
->clock_task
;
735 #define RQCF_REQ_SKIP 0x01
736 #define RQCF_ACT_SKIP 0x02
738 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
740 lockdep_assert_held(&rq
->lock
);
742 rq
->clock_skip_update
|= RQCF_REQ_SKIP
;
744 rq
->clock_skip_update
&= ~RQCF_REQ_SKIP
;
748 enum numa_topology_type
{
753 extern enum numa_topology_type sched_numa_topology_type
;
754 extern int sched_max_numa_distance
;
755 extern bool find_numa_distance(int distance
);
758 #ifdef CONFIG_NUMA_BALANCING
759 /* The regions in numa_faults array from task_struct */
760 enum numa_faults_stats
{
766 extern void sched_setnuma(struct task_struct
*p
, int node
);
767 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
768 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
769 #endif /* CONFIG_NUMA_BALANCING */
774 queue_balance_callback(struct rq
*rq
,
775 struct callback_head
*head
,
776 void (*func
)(struct rq
*rq
))
778 lockdep_assert_held(&rq
->lock
);
780 if (unlikely(head
->next
))
783 head
->func
= (void (*)(struct callback_head
*))func
;
784 head
->next
= rq
->balance_callback
;
785 rq
->balance_callback
= head
;
788 extern void sched_ttwu_pending(void);
790 #define rcu_dereference_check_sched_domain(p) \
791 rcu_dereference_check((p), \
792 lockdep_is_held(&sched_domains_mutex))
795 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
796 * See detach_destroy_domains: synchronize_sched for details.
798 * The domain tree of any CPU may only be accessed from within
799 * preempt-disabled sections.
801 #define for_each_domain(cpu, __sd) \
802 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
803 __sd; __sd = __sd->parent)
805 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
808 * highest_flag_domain - Return highest sched_domain containing flag.
809 * @cpu: The cpu whose highest level of sched domain is to
811 * @flag: The flag to check for the highest sched_domain
814 * Returns the highest sched_domain of a cpu which contains the given flag.
816 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
818 struct sched_domain
*sd
, *hsd
= NULL
;
820 for_each_domain(cpu
, sd
) {
821 if (!(sd
->flags
& flag
))
829 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
831 struct sched_domain
*sd
;
833 for_each_domain(cpu
, sd
) {
834 if (sd
->flags
& flag
)
841 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
842 DECLARE_PER_CPU(int, sd_llc_size
);
843 DECLARE_PER_CPU(int, sd_llc_id
);
844 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
845 DECLARE_PER_CPU(struct sched_domain
*, sd_busy
);
846 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
848 struct sched_group_capacity
{
851 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
854 unsigned int capacity
;
855 unsigned long next_update
;
856 int imbalance
; /* XXX unrelated to capacity but shared group state */
858 * Number of busy cpus in this group.
860 atomic_t nr_busy_cpus
;
862 unsigned long cpumask
[0]; /* iteration mask */
866 struct sched_group
*next
; /* Must be a circular list */
869 unsigned int group_weight
;
870 struct sched_group_capacity
*sgc
;
873 * The CPUs this group covers.
875 * NOTE: this field is variable length. (Allocated dynamically
876 * by attaching extra space to the end of the structure,
877 * depending on how many CPUs the kernel has booted up with)
879 unsigned long cpumask
[0];
882 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
884 return to_cpumask(sg
->cpumask
);
888 * cpumask masking which cpus in the group are allowed to iterate up the domain
891 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
893 return to_cpumask(sg
->sgc
->cpumask
);
897 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
898 * @group: The group whose first cpu is to be returned.
900 static inline unsigned int group_first_cpu(struct sched_group
*group
)
902 return cpumask_first(sched_group_cpus(group
));
905 extern int group_balance_cpu(struct sched_group
*sg
);
909 static inline void sched_ttwu_pending(void) { }
911 #endif /* CONFIG_SMP */
914 #include "auto_group.h"
916 #ifdef CONFIG_CGROUP_SCHED
919 * Return the group to which this tasks belongs.
921 * We cannot use task_css() and friends because the cgroup subsystem
922 * changes that value before the cgroup_subsys::attach() method is called,
923 * therefore we cannot pin it and might observe the wrong value.
925 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
926 * core changes this before calling sched_move_task().
928 * Instead we use a 'copy' which is updated from sched_move_task() while
929 * holding both task_struct::pi_lock and rq::lock.
931 static inline struct task_group
*task_group(struct task_struct
*p
)
933 return p
->sched_task_group
;
936 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
937 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
939 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
940 struct task_group
*tg
= task_group(p
);
943 #ifdef CONFIG_FAIR_GROUP_SCHED
944 set_task_rq_fair(&p
->se
, p
->se
.cfs_rq
, tg
->cfs_rq
[cpu
]);
945 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
946 p
->se
.parent
= tg
->se
[cpu
];
949 #ifdef CONFIG_RT_GROUP_SCHED
950 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
951 p
->rt
.parent
= tg
->rt_se
[cpu
];
955 #else /* CONFIG_CGROUP_SCHED */
957 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
958 static inline struct task_group
*task_group(struct task_struct
*p
)
963 #endif /* CONFIG_CGROUP_SCHED */
965 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
970 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
971 * successfuly executed on another CPU. We must ensure that updates of
972 * per-task data have been completed by this moment.
975 task_thread_info(p
)->cpu
= cpu
;
981 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
983 #ifdef CONFIG_SCHED_DEBUG
984 # include <linux/static_key.h>
985 # define const_debug __read_mostly
987 # define const_debug const
990 extern const_debug
unsigned int sysctl_sched_features
;
992 #define SCHED_FEAT(name, enabled) \
993 __SCHED_FEAT_##name ,
996 #include "features.h"
1002 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1003 #define SCHED_FEAT(name, enabled) \
1004 static __always_inline bool static_branch_##name(struct static_key *key) \
1006 return static_key_##enabled(key); \
1009 #include "features.h"
1013 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1014 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1015 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1016 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1017 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1019 extern struct static_key_false sched_numa_balancing
;
1021 static inline u64
global_rt_period(void)
1023 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1026 static inline u64
global_rt_runtime(void)
1028 if (sysctl_sched_rt_runtime
< 0)
1031 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1034 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1036 return rq
->curr
== p
;
1039 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1044 return task_current(rq
, p
);
1048 static inline int task_on_rq_queued(struct task_struct
*p
)
1050 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1053 static inline int task_on_rq_migrating(struct task_struct
*p
)
1055 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1058 #ifndef prepare_arch_switch
1059 # define prepare_arch_switch(next) do { } while (0)
1061 #ifndef finish_arch_post_lock_switch
1062 # define finish_arch_post_lock_switch() do { } while (0)
1065 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1069 * We can optimise this out completely for !SMP, because the
1070 * SMP rebalancing from interrupt is the only thing that cares
1077 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1081 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1082 * We must ensure this doesn't happen until the switch is completely
1085 * In particular, the load of prev->state in finish_task_switch() must
1086 * happen before this.
1088 * Pairs with the control dependency and rmb in try_to_wake_up().
1090 smp_store_release(&prev
->on_cpu
, 0);
1092 #ifdef CONFIG_DEBUG_SPINLOCK
1093 /* this is a valid case when another task releases the spinlock */
1094 rq
->lock
.owner
= current
;
1097 * If we are tracking spinlock dependencies then we have to
1098 * fix up the runqueue lock - which gets 'carried over' from
1099 * prev into current:
1101 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1103 raw_spin_unlock_irq(&rq
->lock
);
1109 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1110 #define WF_FORK 0x02 /* child wakeup after fork */
1111 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1114 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1115 * of tasks with abnormal "nice" values across CPUs the contribution that
1116 * each task makes to its run queue's load is weighted according to its
1117 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1118 * scaled version of the new time slice allocation that they receive on time
1122 #define WEIGHT_IDLEPRIO 3
1123 #define WMULT_IDLEPRIO 1431655765
1125 extern const int sched_prio_to_weight
[40];
1126 extern const u32 sched_prio_to_wmult
[40];
1128 #define ENQUEUE_WAKEUP 0x01
1129 #define ENQUEUE_HEAD 0x02
1131 #define ENQUEUE_WAKING 0x04 /* sched_class::task_waking was called */
1133 #define ENQUEUE_WAKING 0x00
1135 #define ENQUEUE_REPLENISH 0x08
1136 #define ENQUEUE_RESTORE 0x10
1138 #define DEQUEUE_SLEEP 0x01
1139 #define DEQUEUE_SAVE 0x02
1141 #define RETRY_TASK ((void *)-1UL)
1143 struct sched_class
{
1144 const struct sched_class
*next
;
1146 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1147 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1148 void (*yield_task
) (struct rq
*rq
);
1149 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1151 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1154 * It is the responsibility of the pick_next_task() method that will
1155 * return the next task to call put_prev_task() on the @prev task or
1156 * something equivalent.
1158 * May return RETRY_TASK when it finds a higher prio class has runnable
1161 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1162 struct task_struct
*prev
);
1163 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1166 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1167 void (*migrate_task_rq
)(struct task_struct
*p
);
1169 void (*task_waking
) (struct task_struct
*task
);
1170 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1172 void (*set_cpus_allowed
)(struct task_struct
*p
,
1173 const struct cpumask
*newmask
);
1175 void (*rq_online
)(struct rq
*rq
);
1176 void (*rq_offline
)(struct rq
*rq
);
1179 void (*set_curr_task
) (struct rq
*rq
);
1180 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1181 void (*task_fork
) (struct task_struct
*p
);
1182 void (*task_dead
) (struct task_struct
*p
);
1185 * The switched_from() call is allowed to drop rq->lock, therefore we
1186 * cannot assume the switched_from/switched_to pair is serliazed by
1187 * rq->lock. They are however serialized by p->pi_lock.
1189 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1190 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1191 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1194 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1195 struct task_struct
*task
);
1197 void (*update_curr
) (struct rq
*rq
);
1199 #ifdef CONFIG_FAIR_GROUP_SCHED
1200 void (*task_move_group
) (struct task_struct
*p
);
1204 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1206 prev
->sched_class
->put_prev_task(rq
, prev
);
1209 #define sched_class_highest (&stop_sched_class)
1210 #define for_each_class(class) \
1211 for (class = sched_class_highest; class; class = class->next)
1213 extern const struct sched_class stop_sched_class
;
1214 extern const struct sched_class dl_sched_class
;
1215 extern const struct sched_class rt_sched_class
;
1216 extern const struct sched_class fair_sched_class
;
1217 extern const struct sched_class idle_sched_class
;
1222 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1224 extern void trigger_load_balance(struct rq
*rq
);
1226 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1230 #ifdef CONFIG_CPU_IDLE
1231 static inline void idle_set_state(struct rq
*rq
,
1232 struct cpuidle_state
*idle_state
)
1234 rq
->idle_state
= idle_state
;
1237 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1239 WARN_ON(!rcu_read_lock_held());
1240 return rq
->idle_state
;
1243 static inline void idle_set_state(struct rq
*rq
,
1244 struct cpuidle_state
*idle_state
)
1248 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1254 extern void sysrq_sched_debug_show(void);
1255 extern void sched_init_granularity(void);
1256 extern void update_max_interval(void);
1258 extern void init_sched_dl_class(void);
1259 extern void init_sched_rt_class(void);
1260 extern void init_sched_fair_class(void);
1262 extern void resched_curr(struct rq
*rq
);
1263 extern void resched_cpu(int cpu
);
1265 extern struct rt_bandwidth def_rt_bandwidth
;
1266 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1268 extern struct dl_bandwidth def_dl_bandwidth
;
1269 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1270 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1272 unsigned long to_ratio(u64 period
, u64 runtime
);
1274 extern void init_entity_runnable_average(struct sched_entity
*se
);
1276 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1278 unsigned prev_nr
= rq
->nr_running
;
1280 rq
->nr_running
= prev_nr
+ count
;
1282 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1284 if (!rq
->rd
->overload
)
1285 rq
->rd
->overload
= true;
1288 #ifdef CONFIG_NO_HZ_FULL
1289 if (tick_nohz_full_cpu(rq
->cpu
)) {
1291 * Tick is needed if more than one task runs on a CPU.
1292 * Send the target an IPI to kick it out of nohz mode.
1294 * We assume that IPI implies full memory barrier and the
1295 * new value of rq->nr_running is visible on reception
1298 tick_nohz_full_kick_cpu(rq
->cpu
);
1304 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1306 rq
->nr_running
-= count
;
1309 static inline void rq_last_tick_reset(struct rq
*rq
)
1311 #ifdef CONFIG_NO_HZ_FULL
1312 rq
->last_sched_tick
= jiffies
;
1316 extern void update_rq_clock(struct rq
*rq
);
1318 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1319 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1321 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1323 extern const_debug
unsigned int sysctl_sched_time_avg
;
1324 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1325 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1327 static inline u64
sched_avg_period(void)
1329 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1332 #ifdef CONFIG_SCHED_HRTICK
1336 * - enabled by features
1337 * - hrtimer is actually high res
1339 static inline int hrtick_enabled(struct rq
*rq
)
1341 if (!sched_feat(HRTICK
))
1343 if (!cpu_active(cpu_of(rq
)))
1345 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1348 void hrtick_start(struct rq
*rq
, u64 delay
);
1352 static inline int hrtick_enabled(struct rq
*rq
)
1357 #endif /* CONFIG_SCHED_HRTICK */
1360 extern void sched_avg_update(struct rq
*rq
);
1362 #ifndef arch_scale_freq_capacity
1363 static __always_inline
1364 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1366 return SCHED_CAPACITY_SCALE
;
1370 #ifndef arch_scale_cpu_capacity
1371 static __always_inline
1372 unsigned long arch_scale_cpu_capacity(struct sched_domain
*sd
, int cpu
)
1374 if (sd
&& (sd
->flags
& SD_SHARE_CPUCAPACITY
) && (sd
->span_weight
> 1))
1375 return sd
->smt_gain
/ sd
->span_weight
;
1377 return SCHED_CAPACITY_SCALE
;
1381 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1383 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1384 sched_avg_update(rq
);
1387 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1388 static inline void sched_avg_update(struct rq
*rq
) { }
1392 * __task_rq_lock - lock the rq @p resides on.
1394 static inline struct rq
*__task_rq_lock(struct task_struct
*p
)
1395 __acquires(rq
->lock
)
1399 lockdep_assert_held(&p
->pi_lock
);
1403 raw_spin_lock(&rq
->lock
);
1404 if (likely(rq
== task_rq(p
) && !task_on_rq_migrating(p
))) {
1405 lockdep_pin_lock(&rq
->lock
);
1408 raw_spin_unlock(&rq
->lock
);
1410 while (unlikely(task_on_rq_migrating(p
)))
1416 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1418 static inline struct rq
*task_rq_lock(struct task_struct
*p
, unsigned long *flags
)
1419 __acquires(p
->pi_lock
)
1420 __acquires(rq
->lock
)
1425 raw_spin_lock_irqsave(&p
->pi_lock
, *flags
);
1427 raw_spin_lock(&rq
->lock
);
1429 * move_queued_task() task_rq_lock()
1431 * ACQUIRE (rq->lock)
1432 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1433 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1434 * [S] ->cpu = new_cpu [L] task_rq()
1436 * RELEASE (rq->lock)
1438 * If we observe the old cpu in task_rq_lock, the acquire of
1439 * the old rq->lock will fully serialize against the stores.
1441 * If we observe the new cpu in task_rq_lock, the acquire will
1442 * pair with the WMB to ensure we must then also see migrating.
1444 if (likely(rq
== task_rq(p
) && !task_on_rq_migrating(p
))) {
1445 lockdep_pin_lock(&rq
->lock
);
1448 raw_spin_unlock(&rq
->lock
);
1449 raw_spin_unlock_irqrestore(&p
->pi_lock
, *flags
);
1451 while (unlikely(task_on_rq_migrating(p
)))
1456 static inline void __task_rq_unlock(struct rq
*rq
)
1457 __releases(rq
->lock
)
1459 lockdep_unpin_lock(&rq
->lock
);
1460 raw_spin_unlock(&rq
->lock
);
1464 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, unsigned long *flags
)
1465 __releases(rq
->lock
)
1466 __releases(p
->pi_lock
)
1468 lockdep_unpin_lock(&rq
->lock
);
1469 raw_spin_unlock(&rq
->lock
);
1470 raw_spin_unlock_irqrestore(&p
->pi_lock
, *flags
);
1474 #ifdef CONFIG_PREEMPT
1476 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1479 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1480 * way at the expense of forcing extra atomic operations in all
1481 * invocations. This assures that the double_lock is acquired using the
1482 * same underlying policy as the spinlock_t on this architecture, which
1483 * reduces latency compared to the unfair variant below. However, it
1484 * also adds more overhead and therefore may reduce throughput.
1486 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1487 __releases(this_rq
->lock
)
1488 __acquires(busiest
->lock
)
1489 __acquires(this_rq
->lock
)
1491 raw_spin_unlock(&this_rq
->lock
);
1492 double_rq_lock(this_rq
, busiest
);
1499 * Unfair double_lock_balance: Optimizes throughput at the expense of
1500 * latency by eliminating extra atomic operations when the locks are
1501 * already in proper order on entry. This favors lower cpu-ids and will
1502 * grant the double lock to lower cpus over higher ids under contention,
1503 * regardless of entry order into the function.
1505 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1506 __releases(this_rq
->lock
)
1507 __acquires(busiest
->lock
)
1508 __acquires(this_rq
->lock
)
1512 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1513 if (busiest
< this_rq
) {
1514 raw_spin_unlock(&this_rq
->lock
);
1515 raw_spin_lock(&busiest
->lock
);
1516 raw_spin_lock_nested(&this_rq
->lock
,
1517 SINGLE_DEPTH_NESTING
);
1520 raw_spin_lock_nested(&busiest
->lock
,
1521 SINGLE_DEPTH_NESTING
);
1526 #endif /* CONFIG_PREEMPT */
1529 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1531 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1533 if (unlikely(!irqs_disabled())) {
1534 /* printk() doesn't work good under rq->lock */
1535 raw_spin_unlock(&this_rq
->lock
);
1539 return _double_lock_balance(this_rq
, busiest
);
1542 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1543 __releases(busiest
->lock
)
1545 raw_spin_unlock(&busiest
->lock
);
1546 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1549 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1555 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1558 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1564 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1567 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1573 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1577 * double_rq_lock - safely lock two runqueues
1579 * Note this does not disable interrupts like task_rq_lock,
1580 * you need to do so manually before calling.
1582 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1583 __acquires(rq1
->lock
)
1584 __acquires(rq2
->lock
)
1586 BUG_ON(!irqs_disabled());
1588 raw_spin_lock(&rq1
->lock
);
1589 __acquire(rq2
->lock
); /* Fake it out ;) */
1592 raw_spin_lock(&rq1
->lock
);
1593 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1595 raw_spin_lock(&rq2
->lock
);
1596 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1602 * double_rq_unlock - safely unlock two runqueues
1604 * Note this does not restore interrupts like task_rq_unlock,
1605 * you need to do so manually after calling.
1607 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1608 __releases(rq1
->lock
)
1609 __releases(rq2
->lock
)
1611 raw_spin_unlock(&rq1
->lock
);
1613 raw_spin_unlock(&rq2
->lock
);
1615 __release(rq2
->lock
);
1618 #else /* CONFIG_SMP */
1621 * double_rq_lock - safely lock two runqueues
1623 * Note this does not disable interrupts like task_rq_lock,
1624 * you need to do so manually before calling.
1626 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1627 __acquires(rq1
->lock
)
1628 __acquires(rq2
->lock
)
1630 BUG_ON(!irqs_disabled());
1632 raw_spin_lock(&rq1
->lock
);
1633 __acquire(rq2
->lock
); /* Fake it out ;) */
1637 * double_rq_unlock - safely unlock two runqueues
1639 * Note this does not restore interrupts like task_rq_unlock,
1640 * you need to do so manually after calling.
1642 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1643 __releases(rq1
->lock
)
1644 __releases(rq2
->lock
)
1647 raw_spin_unlock(&rq1
->lock
);
1648 __release(rq2
->lock
);
1653 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1654 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1656 #ifdef CONFIG_SCHED_DEBUG
1657 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1658 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1659 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1661 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1663 #ifdef CONFIG_NUMA_BALANCING
1665 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1667 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1668 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1669 #endif /* CONFIG_NUMA_BALANCING */
1670 #endif /* CONFIG_SCHED_DEBUG */
1672 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1673 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1674 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1676 extern void cfs_bandwidth_usage_inc(void);
1677 extern void cfs_bandwidth_usage_dec(void);
1679 #ifdef CONFIG_NO_HZ_COMMON
1680 enum rq_nohz_flag_bits
{
1685 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1688 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1690 DECLARE_PER_CPU(u64
, cpu_hardirq_time
);
1691 DECLARE_PER_CPU(u64
, cpu_softirq_time
);
1693 #ifndef CONFIG_64BIT
1694 DECLARE_PER_CPU(seqcount_t
, irq_time_seq
);
1696 static inline void irq_time_write_begin(void)
1698 __this_cpu_inc(irq_time_seq
.sequence
);
1702 static inline void irq_time_write_end(void)
1705 __this_cpu_inc(irq_time_seq
.sequence
);
1708 static inline u64
irq_time_read(int cpu
)
1714 seq
= read_seqcount_begin(&per_cpu(irq_time_seq
, cpu
));
1715 irq_time
= per_cpu(cpu_softirq_time
, cpu
) +
1716 per_cpu(cpu_hardirq_time
, cpu
);
1717 } while (read_seqcount_retry(&per_cpu(irq_time_seq
, cpu
), seq
));
1721 #else /* CONFIG_64BIT */
1722 static inline void irq_time_write_begin(void)
1726 static inline void irq_time_write_end(void)
1730 static inline u64
irq_time_read(int cpu
)
1732 return per_cpu(cpu_softirq_time
, cpu
) + per_cpu(cpu_hardirq_time
, cpu
);
1734 #endif /* CONFIG_64BIT */
1735 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */