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 fair_policy(int policy
)
89 return policy
== SCHED_NORMAL
|| policy
== SCHED_BATCH
;
92 static inline int rt_policy(int policy
)
94 return policy
== SCHED_FIFO
|| policy
== SCHED_RR
;
97 static inline int dl_policy(int policy
)
99 return policy
== SCHED_DEADLINE
;
102 static inline int task_has_rt_policy(struct task_struct
*p
)
104 return rt_policy(p
->policy
);
107 static inline int task_has_dl_policy(struct task_struct
*p
)
109 return dl_policy(p
->policy
);
112 static inline bool dl_time_before(u64 a
, u64 b
)
114 return (s64
)(a
- b
) < 0;
118 * Tells if entity @a should preempt entity @b.
121 dl_entity_preempt(struct sched_dl_entity
*a
, struct sched_dl_entity
*b
)
123 return dl_time_before(a
->deadline
, b
->deadline
);
127 * This is the priority-queue data structure of the RT scheduling class:
129 struct rt_prio_array
{
130 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
131 struct list_head queue
[MAX_RT_PRIO
];
134 struct rt_bandwidth
{
135 /* nests inside the rq lock: */
136 raw_spinlock_t rt_runtime_lock
;
139 struct hrtimer rt_period_timer
;
140 unsigned int rt_period_active
;
143 void __dl_clear_params(struct task_struct
*p
);
146 * To keep the bandwidth of -deadline tasks and groups under control
147 * we need some place where:
148 * - store the maximum -deadline bandwidth of the system (the group);
149 * - cache the fraction of that bandwidth that is currently allocated.
151 * This is all done in the data structure below. It is similar to the
152 * one used for RT-throttling (rt_bandwidth), with the main difference
153 * that, since here we are only interested in admission control, we
154 * do not decrease any runtime while the group "executes", neither we
155 * need a timer to replenish it.
157 * With respect to SMP, the bandwidth is given on a per-CPU basis,
159 * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
160 * - dl_total_bw array contains, in the i-eth element, the currently
161 * allocated bandwidth on the i-eth CPU.
162 * Moreover, groups consume bandwidth on each CPU, while tasks only
163 * consume bandwidth on the CPU they're running on.
164 * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
165 * that will be shown the next time the proc or cgroup controls will
166 * be red. It on its turn can be changed by writing on its own
169 struct dl_bandwidth
{
170 raw_spinlock_t dl_runtime_lock
;
175 static inline int dl_bandwidth_enabled(void)
177 return sysctl_sched_rt_runtime
>= 0;
180 extern struct dl_bw
*dl_bw_of(int i
);
188 void __dl_clear(struct dl_bw
*dl_b
, u64 tsk_bw
)
190 dl_b
->total_bw
-= tsk_bw
;
194 void __dl_add(struct dl_bw
*dl_b
, u64 tsk_bw
)
196 dl_b
->total_bw
+= tsk_bw
;
200 bool __dl_overflow(struct dl_bw
*dl_b
, int cpus
, u64 old_bw
, u64 new_bw
)
202 return dl_b
->bw
!= -1 &&
203 dl_b
->bw
* cpus
< dl_b
->total_bw
- old_bw
+ new_bw
;
206 extern struct mutex sched_domains_mutex
;
208 #ifdef CONFIG_CGROUP_SCHED
210 #include <linux/cgroup.h>
215 extern struct list_head task_groups
;
217 struct cfs_bandwidth
{
218 #ifdef CONFIG_CFS_BANDWIDTH
222 s64 hierarchical_quota
;
225 int idle
, period_active
;
226 struct hrtimer period_timer
, slack_timer
;
227 struct list_head throttled_cfs_rq
;
230 int nr_periods
, nr_throttled
;
235 /* task group related information */
237 struct cgroup_subsys_state css
;
239 #ifdef CONFIG_FAIR_GROUP_SCHED
240 /* schedulable entities of this group on each cpu */
241 struct sched_entity
**se
;
242 /* runqueue "owned" by this group on each cpu */
243 struct cfs_rq
**cfs_rq
;
244 unsigned long shares
;
247 atomic_long_t load_avg
;
251 #ifdef CONFIG_RT_GROUP_SCHED
252 struct sched_rt_entity
**rt_se
;
253 struct rt_rq
**rt_rq
;
255 struct rt_bandwidth rt_bandwidth
;
259 struct list_head list
;
261 struct task_group
*parent
;
262 struct list_head siblings
;
263 struct list_head children
;
265 #ifdef CONFIG_SCHED_AUTOGROUP
266 struct autogroup
*autogroup
;
269 struct cfs_bandwidth cfs_bandwidth
;
272 #ifdef CONFIG_FAIR_GROUP_SCHED
273 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
276 * A weight of 0 or 1 can cause arithmetics problems.
277 * A weight of a cfs_rq is the sum of weights of which entities
278 * are queued on this cfs_rq, so a weight of a entity should not be
279 * too large, so as the shares value of a task group.
280 * (The default weight is 1024 - so there's no practical
281 * limitation from this.)
283 #define MIN_SHARES (1UL << 1)
284 #define MAX_SHARES (1UL << 18)
287 typedef int (*tg_visitor
)(struct task_group
*, void *);
289 extern int walk_tg_tree_from(struct task_group
*from
,
290 tg_visitor down
, tg_visitor up
, void *data
);
293 * Iterate the full tree, calling @down when first entering a node and @up when
294 * leaving it for the final time.
296 * Caller must hold rcu_lock or sufficient equivalent.
298 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
300 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
303 extern int tg_nop(struct task_group
*tg
, void *data
);
305 extern void free_fair_sched_group(struct task_group
*tg
);
306 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
307 extern void unregister_fair_sched_group(struct task_group
*tg
, int cpu
);
308 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
309 struct sched_entity
*se
, int cpu
,
310 struct sched_entity
*parent
);
311 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
312 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
314 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
315 extern void start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
316 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
318 extern void free_rt_sched_group(struct task_group
*tg
);
319 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
320 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
321 struct sched_rt_entity
*rt_se
, int cpu
,
322 struct sched_rt_entity
*parent
);
324 extern struct task_group
*sched_create_group(struct task_group
*parent
);
325 extern void sched_online_group(struct task_group
*tg
,
326 struct task_group
*parent
);
327 extern void sched_destroy_group(struct task_group
*tg
);
328 extern void sched_offline_group(struct task_group
*tg
);
330 extern void sched_move_task(struct task_struct
*tsk
);
332 #ifdef CONFIG_FAIR_GROUP_SCHED
333 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
336 #else /* CONFIG_CGROUP_SCHED */
338 struct cfs_bandwidth
{ };
340 #endif /* CONFIG_CGROUP_SCHED */
342 /* CFS-related fields in a runqueue */
344 struct load_weight load
;
345 unsigned int nr_running
, h_nr_running
;
350 u64 min_vruntime_copy
;
353 struct rb_root tasks_timeline
;
354 struct rb_node
*rb_leftmost
;
357 * 'curr' points to currently running entity on this cfs_rq.
358 * It is set to NULL otherwise (i.e when none are currently running).
360 struct sched_entity
*curr
, *next
, *last
, *skip
;
362 #ifdef CONFIG_SCHED_DEBUG
363 unsigned int nr_spread_over
;
370 struct sched_avg avg
;
371 u64 runnable_load_sum
;
372 unsigned long runnable_load_avg
;
373 #ifdef CONFIG_FAIR_GROUP_SCHED
374 unsigned long tg_load_avg_contrib
;
376 atomic_long_t removed_load_avg
, removed_util_avg
;
378 u64 load_last_update_time_copy
;
381 #ifdef CONFIG_FAIR_GROUP_SCHED
383 * h_load = weight * f(tg)
385 * Where f(tg) is the recursive weight fraction assigned to
388 unsigned long h_load
;
389 u64 last_h_load_update
;
390 struct sched_entity
*h_load_next
;
391 #endif /* CONFIG_FAIR_GROUP_SCHED */
392 #endif /* CONFIG_SMP */
394 #ifdef CONFIG_FAIR_GROUP_SCHED
395 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
398 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
399 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
400 * (like users, containers etc.)
402 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
403 * list is used during load balance.
406 struct list_head leaf_cfs_rq_list
;
407 struct task_group
*tg
; /* group that "owns" this runqueue */
409 #ifdef CONFIG_CFS_BANDWIDTH
412 s64 runtime_remaining
;
414 u64 throttled_clock
, throttled_clock_task
;
415 u64 throttled_clock_task_time
;
416 int throttled
, throttle_count
;
417 struct list_head throttled_list
;
418 #endif /* CONFIG_CFS_BANDWIDTH */
419 #endif /* CONFIG_FAIR_GROUP_SCHED */
422 static inline int rt_bandwidth_enabled(void)
424 return sysctl_sched_rt_runtime
>= 0;
427 /* RT IPI pull logic requires IRQ_WORK */
428 #ifdef CONFIG_IRQ_WORK
429 # define HAVE_RT_PUSH_IPI
432 /* Real-Time classes' related field in a runqueue: */
434 struct rt_prio_array active
;
435 unsigned int rt_nr_running
;
436 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
438 int curr
; /* highest queued rt task prio */
440 int next
; /* next highest */
445 unsigned long rt_nr_migratory
;
446 unsigned long rt_nr_total
;
448 struct plist_head pushable_tasks
;
449 #ifdef HAVE_RT_PUSH_IPI
452 struct irq_work push_work
;
453 raw_spinlock_t push_lock
;
455 #endif /* CONFIG_SMP */
461 /* Nests inside the rq lock: */
462 raw_spinlock_t rt_runtime_lock
;
464 #ifdef CONFIG_RT_GROUP_SCHED
465 unsigned long rt_nr_boosted
;
468 struct task_group
*tg
;
472 /* Deadline class' related fields in a runqueue */
474 /* runqueue is an rbtree, ordered by deadline */
475 struct rb_root rb_root
;
476 struct rb_node
*rb_leftmost
;
478 unsigned long dl_nr_running
;
482 * Deadline values of the currently executing and the
483 * earliest ready task on this rq. Caching these facilitates
484 * the decision wether or not a ready but not running task
485 * should migrate somewhere else.
492 unsigned long dl_nr_migratory
;
496 * Tasks on this rq that can be pushed away. They are kept in
497 * an rb-tree, ordered by tasks' deadlines, with caching
498 * of the leftmost (earliest deadline) element.
500 struct rb_root pushable_dl_tasks_root
;
501 struct rb_node
*pushable_dl_tasks_leftmost
;
510 * We add the notion of a root-domain which will be used to define per-domain
511 * variables. Each exclusive cpuset essentially defines an island domain by
512 * fully partitioning the member cpus from any other cpuset. Whenever a new
513 * exclusive cpuset is created, we also create and attach a new root-domain
522 cpumask_var_t online
;
524 /* Indicate more than one runnable task for any CPU */
528 * The bit corresponding to a CPU gets set here if such CPU has more
529 * than one runnable -deadline task (as it is below for RT tasks).
531 cpumask_var_t dlo_mask
;
537 * The "RT overload" flag: it gets set if a CPU has more than
538 * one runnable RT task.
540 cpumask_var_t rto_mask
;
541 struct cpupri cpupri
;
544 extern struct root_domain def_root_domain
;
546 #endif /* CONFIG_SMP */
549 * This is the main, per-CPU runqueue data structure.
551 * Locking rule: those places that want to lock multiple runqueues
552 * (such as the load balancing or the thread migration code), lock
553 * acquire operations must be ordered by ascending &runqueue.
560 * nr_running and cpu_load should be in the same cacheline because
561 * remote CPUs use both these fields when doing load calculation.
563 unsigned int nr_running
;
564 #ifdef CONFIG_NUMA_BALANCING
565 unsigned int nr_numa_running
;
566 unsigned int nr_preferred_running
;
568 #define CPU_LOAD_IDX_MAX 5
569 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
570 unsigned long last_load_update_tick
;
571 #ifdef CONFIG_NO_HZ_COMMON
573 unsigned long nohz_flags
;
575 #ifdef CONFIG_NO_HZ_FULL
576 unsigned long last_sched_tick
;
578 /* capture load from *all* tasks on this cpu: */
579 struct load_weight load
;
580 unsigned long nr_load_updates
;
587 #ifdef CONFIG_FAIR_GROUP_SCHED
588 /* list of leaf cfs_rq on this cpu: */
589 struct list_head leaf_cfs_rq_list
;
590 #endif /* CONFIG_FAIR_GROUP_SCHED */
593 * This is part of a global counter where only the total sum
594 * over all CPUs matters. A task can increase this counter on
595 * one CPU and if it got migrated afterwards it may decrease
596 * it on another CPU. Always updated under the runqueue lock:
598 unsigned long nr_uninterruptible
;
600 struct task_struct
*curr
, *idle
, *stop
;
601 unsigned long next_balance
;
602 struct mm_struct
*prev_mm
;
604 unsigned int clock_skip_update
;
611 struct root_domain
*rd
;
612 struct sched_domain
*sd
;
614 unsigned long cpu_capacity
;
615 unsigned long cpu_capacity_orig
;
617 struct callback_head
*balance_callback
;
619 unsigned char idle_balance
;
620 /* For active balancing */
623 struct cpu_stop_work active_balance_work
;
624 /* cpu of this runqueue: */
628 struct list_head cfs_tasks
;
635 /* This is used to determine avg_idle's max value */
636 u64 max_idle_balance_cost
;
639 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
642 #ifdef CONFIG_PARAVIRT
645 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
646 u64 prev_steal_time_rq
;
649 /* calc_load related fields */
650 unsigned long calc_load_update
;
651 long calc_load_active
;
653 #ifdef CONFIG_SCHED_HRTICK
655 int hrtick_csd_pending
;
656 struct call_single_data hrtick_csd
;
658 struct hrtimer hrtick_timer
;
661 #ifdef CONFIG_SCHEDSTATS
663 struct sched_info rq_sched_info
;
664 unsigned long long rq_cpu_time
;
665 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
667 /* sys_sched_yield() stats */
668 unsigned int yld_count
;
670 /* schedule() stats */
671 unsigned int sched_count
;
672 unsigned int sched_goidle
;
674 /* try_to_wake_up() stats */
675 unsigned int ttwu_count
;
676 unsigned int ttwu_local
;
680 struct llist_head wake_list
;
683 #ifdef CONFIG_CPU_IDLE
684 /* Must be inspected within a rcu lock section */
685 struct cpuidle_state
*idle_state
;
689 static inline int cpu_of(struct rq
*rq
)
698 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq
, runqueues
);
700 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
701 #define this_rq() this_cpu_ptr(&runqueues)
702 #define task_rq(p) cpu_rq(task_cpu(p))
703 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
704 #define raw_rq() raw_cpu_ptr(&runqueues)
706 static inline u64
__rq_clock_broken(struct rq
*rq
)
708 return READ_ONCE(rq
->clock
);
711 static inline u64
rq_clock(struct rq
*rq
)
713 lockdep_assert_held(&rq
->lock
);
717 static inline u64
rq_clock_task(struct rq
*rq
)
719 lockdep_assert_held(&rq
->lock
);
720 return rq
->clock_task
;
723 #define RQCF_REQ_SKIP 0x01
724 #define RQCF_ACT_SKIP 0x02
726 static inline void rq_clock_skip_update(struct rq
*rq
, bool skip
)
728 lockdep_assert_held(&rq
->lock
);
730 rq
->clock_skip_update
|= RQCF_REQ_SKIP
;
732 rq
->clock_skip_update
&= ~RQCF_REQ_SKIP
;
736 enum numa_topology_type
{
741 extern enum numa_topology_type sched_numa_topology_type
;
742 extern int sched_max_numa_distance
;
743 extern bool find_numa_distance(int distance
);
746 #ifdef CONFIG_NUMA_BALANCING
747 /* The regions in numa_faults array from task_struct */
748 enum numa_faults_stats
{
754 extern void sched_setnuma(struct task_struct
*p
, int node
);
755 extern int migrate_task_to(struct task_struct
*p
, int cpu
);
756 extern int migrate_swap(struct task_struct
*, struct task_struct
*);
757 #endif /* CONFIG_NUMA_BALANCING */
762 queue_balance_callback(struct rq
*rq
,
763 struct callback_head
*head
,
764 void (*func
)(struct rq
*rq
))
766 lockdep_assert_held(&rq
->lock
);
768 if (unlikely(head
->next
))
771 head
->func
= (void (*)(struct callback_head
*))func
;
772 head
->next
= rq
->balance_callback
;
773 rq
->balance_callback
= head
;
776 extern void sched_ttwu_pending(void);
778 #define rcu_dereference_check_sched_domain(p) \
779 rcu_dereference_check((p), \
780 lockdep_is_held(&sched_domains_mutex))
783 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
784 * See detach_destroy_domains: synchronize_sched for details.
786 * The domain tree of any CPU may only be accessed from within
787 * preempt-disabled sections.
789 #define for_each_domain(cpu, __sd) \
790 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
791 __sd; __sd = __sd->parent)
793 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
796 * highest_flag_domain - Return highest sched_domain containing flag.
797 * @cpu: The cpu whose highest level of sched domain is to
799 * @flag: The flag to check for the highest sched_domain
802 * Returns the highest sched_domain of a cpu which contains the given flag.
804 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
806 struct sched_domain
*sd
, *hsd
= NULL
;
808 for_each_domain(cpu
, sd
) {
809 if (!(sd
->flags
& flag
))
817 static inline struct sched_domain
*lowest_flag_domain(int cpu
, int flag
)
819 struct sched_domain
*sd
;
821 for_each_domain(cpu
, sd
) {
822 if (sd
->flags
& flag
)
829 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
830 DECLARE_PER_CPU(int, sd_llc_size
);
831 DECLARE_PER_CPU(int, sd_llc_id
);
832 DECLARE_PER_CPU(struct sched_domain
*, sd_numa
);
833 DECLARE_PER_CPU(struct sched_domain
*, sd_busy
);
834 DECLARE_PER_CPU(struct sched_domain
*, sd_asym
);
836 struct sched_group_capacity
{
839 * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
842 unsigned int capacity
;
843 unsigned long next_update
;
844 int imbalance
; /* XXX unrelated to capacity but shared group state */
846 * Number of busy cpus in this group.
848 atomic_t nr_busy_cpus
;
850 unsigned long cpumask
[0]; /* iteration mask */
854 struct sched_group
*next
; /* Must be a circular list */
857 unsigned int group_weight
;
858 struct sched_group_capacity
*sgc
;
861 * The CPUs this group covers.
863 * NOTE: this field is variable length. (Allocated dynamically
864 * by attaching extra space to the end of the structure,
865 * depending on how many CPUs the kernel has booted up with)
867 unsigned long cpumask
[0];
870 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
872 return to_cpumask(sg
->cpumask
);
876 * cpumask masking which cpus in the group are allowed to iterate up the domain
879 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
881 return to_cpumask(sg
->sgc
->cpumask
);
885 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
886 * @group: The group whose first cpu is to be returned.
888 static inline unsigned int group_first_cpu(struct sched_group
*group
)
890 return cpumask_first(sched_group_cpus(group
));
893 extern int group_balance_cpu(struct sched_group
*sg
);
897 static inline void sched_ttwu_pending(void) { }
899 #endif /* CONFIG_SMP */
902 #include "auto_group.h"
904 #ifdef CONFIG_CGROUP_SCHED
907 * Return the group to which this tasks belongs.
909 * We cannot use task_css() and friends because the cgroup subsystem
910 * changes that value before the cgroup_subsys::attach() method is called,
911 * therefore we cannot pin it and might observe the wrong value.
913 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
914 * core changes this before calling sched_move_task().
916 * Instead we use a 'copy' which is updated from sched_move_task() while
917 * holding both task_struct::pi_lock and rq::lock.
919 static inline struct task_group
*task_group(struct task_struct
*p
)
921 return p
->sched_task_group
;
924 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
925 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
927 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
928 struct task_group
*tg
= task_group(p
);
931 #ifdef CONFIG_FAIR_GROUP_SCHED
932 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
933 p
->se
.parent
= tg
->se
[cpu
];
936 #ifdef CONFIG_RT_GROUP_SCHED
937 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
938 p
->rt
.parent
= tg
->rt_se
[cpu
];
942 #else /* CONFIG_CGROUP_SCHED */
944 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
945 static inline struct task_group
*task_group(struct task_struct
*p
)
950 #endif /* CONFIG_CGROUP_SCHED */
952 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
957 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
958 * successfuly executed on another CPU. We must ensure that updates of
959 * per-task data have been completed by this moment.
962 task_thread_info(p
)->cpu
= cpu
;
968 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
970 #ifdef CONFIG_SCHED_DEBUG
971 # include <linux/static_key.h>
972 # define const_debug __read_mostly
974 # define const_debug const
977 extern const_debug
unsigned int sysctl_sched_features
;
979 #define SCHED_FEAT(name, enabled) \
980 __SCHED_FEAT_##name ,
983 #include "features.h"
989 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
990 #define SCHED_FEAT(name, enabled) \
991 static __always_inline bool static_branch_##name(struct static_key *key) \
993 return static_key_##enabled(key); \
996 #include "features.h"
1000 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
1001 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1002 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1003 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1004 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1006 #ifdef CONFIG_NUMA_BALANCING
1007 #define sched_feat_numa(x) sched_feat(x)
1008 #ifdef CONFIG_SCHED_DEBUG
1009 #define numabalancing_enabled sched_feat_numa(NUMA)
1011 extern bool numabalancing_enabled
;
1012 #endif /* CONFIG_SCHED_DEBUG */
1014 #define sched_feat_numa(x) (0)
1015 #define numabalancing_enabled (0)
1016 #endif /* CONFIG_NUMA_BALANCING */
1018 static inline u64
global_rt_period(void)
1020 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
1023 static inline u64
global_rt_runtime(void)
1025 if (sysctl_sched_rt_runtime
< 0)
1028 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
1031 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
1033 return rq
->curr
== p
;
1036 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
1041 return task_current(rq
, p
);
1045 static inline int task_on_rq_queued(struct task_struct
*p
)
1047 return p
->on_rq
== TASK_ON_RQ_QUEUED
;
1050 static inline int task_on_rq_migrating(struct task_struct
*p
)
1052 return p
->on_rq
== TASK_ON_RQ_MIGRATING
;
1055 #ifndef prepare_arch_switch
1056 # define prepare_arch_switch(next) do { } while (0)
1058 #ifndef finish_arch_post_lock_switch
1059 # define finish_arch_post_lock_switch() do { } while (0)
1062 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
1066 * We can optimise this out completely for !SMP, because the
1067 * SMP rebalancing from interrupt is the only thing that cares
1074 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
1078 * After ->on_cpu is cleared, the task can be moved to a different CPU.
1079 * We must ensure this doesn't happen until the switch is completely
1082 * Pairs with the control dependency and rmb in try_to_wake_up().
1084 smp_store_release(&prev
->on_cpu
, 0);
1086 #ifdef CONFIG_DEBUG_SPINLOCK
1087 /* this is a valid case when another task releases the spinlock */
1088 rq
->lock
.owner
= current
;
1091 * If we are tracking spinlock dependencies then we have to
1092 * fix up the runqueue lock - which gets 'carried over' from
1093 * prev into current:
1095 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
1097 raw_spin_unlock_irq(&rq
->lock
);
1103 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
1104 #define WF_FORK 0x02 /* child wakeup after fork */
1105 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
1108 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1109 * of tasks with abnormal "nice" values across CPUs the contribution that
1110 * each task makes to its run queue's load is weighted according to its
1111 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1112 * scaled version of the new time slice allocation that they receive on time
1116 #define WEIGHT_IDLEPRIO 3
1117 #define WMULT_IDLEPRIO 1431655765
1120 * Nice levels are multiplicative, with a gentle 10% change for every
1121 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1122 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1123 * that remained on nice 0.
1125 * The "10% effect" is relative and cumulative: from _any_ nice level,
1126 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1127 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1128 * If a task goes up by ~10% and another task goes down by ~10% then
1129 * the relative distance between them is ~25%.)
1131 static const int prio_to_weight
[40] = {
1132 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1133 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1134 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1135 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1136 /* 0 */ 1024, 820, 655, 526, 423,
1137 /* 5 */ 335, 272, 215, 172, 137,
1138 /* 10 */ 110, 87, 70, 56, 45,
1139 /* 15 */ 36, 29, 23, 18, 15,
1143 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1145 * In cases where the weight does not change often, we can use the
1146 * precalculated inverse to speed up arithmetics by turning divisions
1147 * into multiplications:
1149 static const u32 prio_to_wmult
[40] = {
1150 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1151 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1152 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1153 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1154 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1155 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1156 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1157 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1160 #define ENQUEUE_WAKEUP 1
1161 #define ENQUEUE_HEAD 2
1163 #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */
1165 #define ENQUEUE_WAKING 0
1167 #define ENQUEUE_REPLENISH 8
1169 #define DEQUEUE_SLEEP 1
1171 #define RETRY_TASK ((void *)-1UL)
1173 struct sched_class
{
1174 const struct sched_class
*next
;
1176 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1177 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1178 void (*yield_task
) (struct rq
*rq
);
1179 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1181 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1184 * It is the responsibility of the pick_next_task() method that will
1185 * return the next task to call put_prev_task() on the @prev task or
1186 * something equivalent.
1188 * May return RETRY_TASK when it finds a higher prio class has runnable
1191 struct task_struct
* (*pick_next_task
) (struct rq
*rq
,
1192 struct task_struct
*prev
);
1193 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1196 int (*select_task_rq
)(struct task_struct
*p
, int task_cpu
, int sd_flag
, int flags
);
1197 void (*migrate_task_rq
)(struct task_struct
*p
, int next_cpu
);
1199 void (*task_waking
) (struct task_struct
*task
);
1200 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1202 void (*set_cpus_allowed
)(struct task_struct
*p
,
1203 const struct cpumask
*newmask
);
1205 void (*rq_online
)(struct rq
*rq
);
1206 void (*rq_offline
)(struct rq
*rq
);
1209 void (*set_curr_task
) (struct rq
*rq
);
1210 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1211 void (*task_fork
) (struct task_struct
*p
);
1212 void (*task_dead
) (struct task_struct
*p
);
1215 * The switched_from() call is allowed to drop rq->lock, therefore we
1216 * cannot assume the switched_from/switched_to pair is serliazed by
1217 * rq->lock. They are however serialized by p->pi_lock.
1219 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1220 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1221 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1224 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1225 struct task_struct
*task
);
1227 void (*update_curr
) (struct rq
*rq
);
1229 #ifdef CONFIG_FAIR_GROUP_SCHED
1230 void (*task_move_group
) (struct task_struct
*p
, int on_rq
);
1234 static inline void put_prev_task(struct rq
*rq
, struct task_struct
*prev
)
1236 prev
->sched_class
->put_prev_task(rq
, prev
);
1239 #define sched_class_highest (&stop_sched_class)
1240 #define for_each_class(class) \
1241 for (class = sched_class_highest; class; class = class->next)
1243 extern const struct sched_class stop_sched_class
;
1244 extern const struct sched_class dl_sched_class
;
1245 extern const struct sched_class rt_sched_class
;
1246 extern const struct sched_class fair_sched_class
;
1247 extern const struct sched_class idle_sched_class
;
1252 extern void update_group_capacity(struct sched_domain
*sd
, int cpu
);
1254 extern void trigger_load_balance(struct rq
*rq
);
1256 extern void idle_enter_fair(struct rq
*this_rq
);
1257 extern void idle_exit_fair(struct rq
*this_rq
);
1259 extern void set_cpus_allowed_common(struct task_struct
*p
, const struct cpumask
*new_mask
);
1263 static inline void idle_enter_fair(struct rq
*rq
) { }
1264 static inline void idle_exit_fair(struct rq
*rq
) { }
1268 #ifdef CONFIG_CPU_IDLE
1269 static inline void idle_set_state(struct rq
*rq
,
1270 struct cpuidle_state
*idle_state
)
1272 rq
->idle_state
= idle_state
;
1275 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1277 WARN_ON(!rcu_read_lock_held());
1278 return rq
->idle_state
;
1281 static inline void idle_set_state(struct rq
*rq
,
1282 struct cpuidle_state
*idle_state
)
1286 static inline struct cpuidle_state
*idle_get_state(struct rq
*rq
)
1292 extern void sysrq_sched_debug_show(void);
1293 extern void sched_init_granularity(void);
1294 extern void update_max_interval(void);
1296 extern void init_sched_dl_class(void);
1297 extern void init_sched_rt_class(void);
1298 extern void init_sched_fair_class(void);
1300 extern void resched_curr(struct rq
*rq
);
1301 extern void resched_cpu(int cpu
);
1303 extern struct rt_bandwidth def_rt_bandwidth
;
1304 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1306 extern struct dl_bandwidth def_dl_bandwidth
;
1307 extern void init_dl_bandwidth(struct dl_bandwidth
*dl_b
, u64 period
, u64 runtime
);
1308 extern void init_dl_task_timer(struct sched_dl_entity
*dl_se
);
1310 unsigned long to_ratio(u64 period
, u64 runtime
);
1312 extern void init_entity_runnable_average(struct sched_entity
*se
);
1314 static inline void add_nr_running(struct rq
*rq
, unsigned count
)
1316 unsigned prev_nr
= rq
->nr_running
;
1318 rq
->nr_running
= prev_nr
+ count
;
1320 if (prev_nr
< 2 && rq
->nr_running
>= 2) {
1322 if (!rq
->rd
->overload
)
1323 rq
->rd
->overload
= true;
1326 #ifdef CONFIG_NO_HZ_FULL
1327 if (tick_nohz_full_cpu(rq
->cpu
)) {
1329 * Tick is needed if more than one task runs on a CPU.
1330 * Send the target an IPI to kick it out of nohz mode.
1332 * We assume that IPI implies full memory barrier and the
1333 * new value of rq->nr_running is visible on reception
1336 tick_nohz_full_kick_cpu(rq
->cpu
);
1342 static inline void sub_nr_running(struct rq
*rq
, unsigned count
)
1344 rq
->nr_running
-= count
;
1347 static inline void rq_last_tick_reset(struct rq
*rq
)
1349 #ifdef CONFIG_NO_HZ_FULL
1350 rq
->last_sched_tick
= jiffies
;
1354 extern void update_rq_clock(struct rq
*rq
);
1356 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1357 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1359 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1361 extern const_debug
unsigned int sysctl_sched_time_avg
;
1362 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1363 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1365 static inline u64
sched_avg_period(void)
1367 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1370 #ifdef CONFIG_SCHED_HRTICK
1374 * - enabled by features
1375 * - hrtimer is actually high res
1377 static inline int hrtick_enabled(struct rq
*rq
)
1379 if (!sched_feat(HRTICK
))
1381 if (!cpu_active(cpu_of(rq
)))
1383 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1386 void hrtick_start(struct rq
*rq
, u64 delay
);
1390 static inline int hrtick_enabled(struct rq
*rq
)
1395 #endif /* CONFIG_SCHED_HRTICK */
1398 extern void sched_avg_update(struct rq
*rq
);
1400 #ifndef arch_scale_freq_capacity
1401 static __always_inline
1402 unsigned long arch_scale_freq_capacity(struct sched_domain
*sd
, int cpu
)
1404 return SCHED_CAPACITY_SCALE
;
1408 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1410 rq
->rt_avg
+= rt_delta
* arch_scale_freq_capacity(NULL
, cpu_of(rq
));
1411 sched_avg_update(rq
);
1414 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1415 static inline void sched_avg_update(struct rq
*rq
) { }
1419 * __task_rq_lock - lock the rq @p resides on.
1421 static inline struct rq
*__task_rq_lock(struct task_struct
*p
)
1422 __acquires(rq
->lock
)
1426 lockdep_assert_held(&p
->pi_lock
);
1430 raw_spin_lock(&rq
->lock
);
1431 if (likely(rq
== task_rq(p
) && !task_on_rq_migrating(p
))) {
1432 lockdep_pin_lock(&rq
->lock
);
1435 raw_spin_unlock(&rq
->lock
);
1437 while (unlikely(task_on_rq_migrating(p
)))
1443 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1445 static inline struct rq
*task_rq_lock(struct task_struct
*p
, unsigned long *flags
)
1446 __acquires(p
->pi_lock
)
1447 __acquires(rq
->lock
)
1452 raw_spin_lock_irqsave(&p
->pi_lock
, *flags
);
1454 raw_spin_lock(&rq
->lock
);
1456 * move_queued_task() task_rq_lock()
1458 * ACQUIRE (rq->lock)
1459 * [S] ->on_rq = MIGRATING [L] rq = task_rq()
1460 * WMB (__set_task_cpu()) ACQUIRE (rq->lock);
1461 * [S] ->cpu = new_cpu [L] task_rq()
1463 * RELEASE (rq->lock)
1465 * If we observe the old cpu in task_rq_lock, the acquire of
1466 * the old rq->lock will fully serialize against the stores.
1468 * If we observe the new cpu in task_rq_lock, the acquire will
1469 * pair with the WMB to ensure we must then also see migrating.
1471 if (likely(rq
== task_rq(p
) && !task_on_rq_migrating(p
))) {
1472 lockdep_pin_lock(&rq
->lock
);
1475 raw_spin_unlock(&rq
->lock
);
1476 raw_spin_unlock_irqrestore(&p
->pi_lock
, *flags
);
1478 while (unlikely(task_on_rq_migrating(p
)))
1483 static inline void __task_rq_unlock(struct rq
*rq
)
1484 __releases(rq
->lock
)
1486 lockdep_unpin_lock(&rq
->lock
);
1487 raw_spin_unlock(&rq
->lock
);
1491 task_rq_unlock(struct rq
*rq
, struct task_struct
*p
, unsigned long *flags
)
1492 __releases(rq
->lock
)
1493 __releases(p
->pi_lock
)
1495 lockdep_unpin_lock(&rq
->lock
);
1496 raw_spin_unlock(&rq
->lock
);
1497 raw_spin_unlock_irqrestore(&p
->pi_lock
, *flags
);
1501 #ifdef CONFIG_PREEMPT
1503 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1506 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1507 * way at the expense of forcing extra atomic operations in all
1508 * invocations. This assures that the double_lock is acquired using the
1509 * same underlying policy as the spinlock_t on this architecture, which
1510 * reduces latency compared to the unfair variant below. However, it
1511 * also adds more overhead and therefore may reduce throughput.
1513 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1514 __releases(this_rq
->lock
)
1515 __acquires(busiest
->lock
)
1516 __acquires(this_rq
->lock
)
1518 raw_spin_unlock(&this_rq
->lock
);
1519 double_rq_lock(this_rq
, busiest
);
1526 * Unfair double_lock_balance: Optimizes throughput at the expense of
1527 * latency by eliminating extra atomic operations when the locks are
1528 * already in proper order on entry. This favors lower cpu-ids and will
1529 * grant the double lock to lower cpus over higher ids under contention,
1530 * regardless of entry order into the function.
1532 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1533 __releases(this_rq
->lock
)
1534 __acquires(busiest
->lock
)
1535 __acquires(this_rq
->lock
)
1539 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1540 if (busiest
< this_rq
) {
1541 raw_spin_unlock(&this_rq
->lock
);
1542 raw_spin_lock(&busiest
->lock
);
1543 raw_spin_lock_nested(&this_rq
->lock
,
1544 SINGLE_DEPTH_NESTING
);
1547 raw_spin_lock_nested(&busiest
->lock
,
1548 SINGLE_DEPTH_NESTING
);
1553 #endif /* CONFIG_PREEMPT */
1556 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1558 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1560 if (unlikely(!irqs_disabled())) {
1561 /* printk() doesn't work good under rq->lock */
1562 raw_spin_unlock(&this_rq
->lock
);
1566 return _double_lock_balance(this_rq
, busiest
);
1569 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1570 __releases(busiest
->lock
)
1572 raw_spin_unlock(&busiest
->lock
);
1573 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1576 static inline void double_lock(spinlock_t
*l1
, spinlock_t
*l2
)
1582 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1585 static inline void double_lock_irq(spinlock_t
*l1
, spinlock_t
*l2
)
1591 spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1594 static inline void double_raw_lock(raw_spinlock_t
*l1
, raw_spinlock_t
*l2
)
1600 raw_spin_lock_nested(l2
, SINGLE_DEPTH_NESTING
);
1604 * double_rq_lock - safely lock two runqueues
1606 * Note this does not disable interrupts like task_rq_lock,
1607 * you need to do so manually before calling.
1609 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1610 __acquires(rq1
->lock
)
1611 __acquires(rq2
->lock
)
1613 BUG_ON(!irqs_disabled());
1615 raw_spin_lock(&rq1
->lock
);
1616 __acquire(rq2
->lock
); /* Fake it out ;) */
1619 raw_spin_lock(&rq1
->lock
);
1620 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1622 raw_spin_lock(&rq2
->lock
);
1623 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1629 * double_rq_unlock - safely unlock two runqueues
1631 * Note this does not restore interrupts like task_rq_unlock,
1632 * you need to do so manually after calling.
1634 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1635 __releases(rq1
->lock
)
1636 __releases(rq2
->lock
)
1638 raw_spin_unlock(&rq1
->lock
);
1640 raw_spin_unlock(&rq2
->lock
);
1642 __release(rq2
->lock
);
1645 #else /* CONFIG_SMP */
1648 * double_rq_lock - safely lock two runqueues
1650 * Note this does not disable interrupts like task_rq_lock,
1651 * you need to do so manually before calling.
1653 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1654 __acquires(rq1
->lock
)
1655 __acquires(rq2
->lock
)
1657 BUG_ON(!irqs_disabled());
1659 raw_spin_lock(&rq1
->lock
);
1660 __acquire(rq2
->lock
); /* Fake it out ;) */
1664 * double_rq_unlock - safely unlock two runqueues
1666 * Note this does not restore interrupts like task_rq_unlock,
1667 * you need to do so manually after calling.
1669 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1670 __releases(rq1
->lock
)
1671 __releases(rq2
->lock
)
1674 raw_spin_unlock(&rq1
->lock
);
1675 __release(rq2
->lock
);
1680 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1681 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1683 #ifdef CONFIG_SCHED_DEBUG
1684 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1685 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1686 extern void print_dl_stats(struct seq_file
*m
, int cpu
);
1688 print_cfs_rq(struct seq_file
*m
, int cpu
, struct cfs_rq
*cfs_rq
);
1690 #ifdef CONFIG_NUMA_BALANCING
1692 show_numa_stats(struct task_struct
*p
, struct seq_file
*m
);
1694 print_numa_stats(struct seq_file
*m
, int node
, unsigned long tsf
,
1695 unsigned long tpf
, unsigned long gsf
, unsigned long gpf
);
1696 #endif /* CONFIG_NUMA_BALANCING */
1697 #endif /* CONFIG_SCHED_DEBUG */
1699 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1700 extern void init_rt_rq(struct rt_rq
*rt_rq
);
1701 extern void init_dl_rq(struct dl_rq
*dl_rq
);
1703 extern void cfs_bandwidth_usage_inc(void);
1704 extern void cfs_bandwidth_usage_dec(void);
1706 #ifdef CONFIG_NO_HZ_COMMON
1707 enum rq_nohz_flag_bits
{
1712 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1715 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1717 DECLARE_PER_CPU(u64
, cpu_hardirq_time
);
1718 DECLARE_PER_CPU(u64
, cpu_softirq_time
);
1720 #ifndef CONFIG_64BIT
1721 DECLARE_PER_CPU(seqcount_t
, irq_time_seq
);
1723 static inline void irq_time_write_begin(void)
1725 __this_cpu_inc(irq_time_seq
.sequence
);
1729 static inline void irq_time_write_end(void)
1732 __this_cpu_inc(irq_time_seq
.sequence
);
1735 static inline u64
irq_time_read(int cpu
)
1741 seq
= read_seqcount_begin(&per_cpu(irq_time_seq
, cpu
));
1742 irq_time
= per_cpu(cpu_softirq_time
, cpu
) +
1743 per_cpu(cpu_hardirq_time
, cpu
);
1744 } while (read_seqcount_retry(&per_cpu(irq_time_seq
, cpu
), seq
));
1748 #else /* CONFIG_64BIT */
1749 static inline void irq_time_write_begin(void)
1753 static inline void irq_time_write_end(void)
1757 static inline u64
irq_time_read(int cpu
)
1759 return per_cpu(cpu_softirq_time
, cpu
) + per_cpu(cpu_hardirq_time
, cpu
);
1761 #endif /* CONFIG_64BIT */
1762 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */