2 #include <linux/sched.h>
3 #include <linux/sched/sysctl.h>
4 #include <linux/sched/rt.h>
5 #include <linux/mutex.h>
6 #include <linux/spinlock.h>
7 #include <linux/stop_machine.h>
11 extern __read_mostly
int scheduler_running
;
14 * Convert user-nice values [ -20 ... 0 ... 19 ]
15 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
18 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
19 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
20 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
23 * 'User priority' is the nice value converted to something we
24 * can work with better when scaling various scheduler parameters,
25 * it's a [ 0 ... 39 ] range.
27 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
28 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
29 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
32 * Helpers for converting nanosecond timing to jiffy resolution
34 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
37 * Increase resolution of nice-level calculations for 64-bit architectures.
38 * The extra resolution improves shares distribution and load balancing of
39 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
40 * hierarchies, especially on larger systems. This is not a user-visible change
41 * and does not change the user-interface for setting shares/weights.
43 * We increase resolution only if we have enough bits to allow this increased
44 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
45 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
48 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
49 # define SCHED_LOAD_RESOLUTION 10
50 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
51 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
53 # define SCHED_LOAD_RESOLUTION 0
54 # define scale_load(w) (w)
55 # define scale_load_down(w) (w)
58 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
59 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
61 #define NICE_0_LOAD SCHED_LOAD_SCALE
62 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
65 * These are the 'tuning knobs' of the scheduler:
69 * single value that denotes runtime == period, ie unlimited time.
71 #define RUNTIME_INF ((u64)~0ULL)
73 static inline int rt_policy(int policy
)
75 if (policy
== SCHED_FIFO
|| policy
== SCHED_RR
)
80 static inline int task_has_rt_policy(struct task_struct
*p
)
82 return rt_policy(p
->policy
);
86 * This is the priority-queue data structure of the RT scheduling class:
88 struct rt_prio_array
{
89 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
90 struct list_head queue
[MAX_RT_PRIO
];
94 /* nests inside the rq lock: */
95 raw_spinlock_t rt_runtime_lock
;
98 struct hrtimer rt_period_timer
;
101 extern struct mutex sched_domains_mutex
;
103 #ifdef CONFIG_CGROUP_SCHED
105 #include <linux/cgroup.h>
110 extern struct list_head task_groups
;
112 struct cfs_bandwidth
{
113 #ifdef CONFIG_CFS_BANDWIDTH
117 s64 hierarchal_quota
;
120 int idle
, timer_active
;
121 struct hrtimer period_timer
, slack_timer
;
122 struct list_head throttled_cfs_rq
;
125 int nr_periods
, nr_throttled
;
130 /* task group related information */
132 struct cgroup_subsys_state css
;
134 #ifdef CONFIG_FAIR_GROUP_SCHED
135 /* schedulable entities of this group on each cpu */
136 struct sched_entity
**se
;
137 /* runqueue "owned" by this group on each cpu */
138 struct cfs_rq
**cfs_rq
;
139 unsigned long shares
;
141 atomic_t load_weight
;
143 atomic_t runnable_avg
;
146 #ifdef CONFIG_RT_GROUP_SCHED
147 struct sched_rt_entity
**rt_se
;
148 struct rt_rq
**rt_rq
;
150 struct rt_bandwidth rt_bandwidth
;
154 struct list_head list
;
156 struct task_group
*parent
;
157 struct list_head siblings
;
158 struct list_head children
;
160 #ifdef CONFIG_SCHED_AUTOGROUP
161 struct autogroup
*autogroup
;
164 struct cfs_bandwidth cfs_bandwidth
;
167 #ifdef CONFIG_FAIR_GROUP_SCHED
168 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
171 * A weight of 0 or 1 can cause arithmetics problems.
172 * A weight of a cfs_rq is the sum of weights of which entities
173 * are queued on this cfs_rq, so a weight of a entity should not be
174 * too large, so as the shares value of a task group.
175 * (The default weight is 1024 - so there's no practical
176 * limitation from this.)
178 #define MIN_SHARES (1UL << 1)
179 #define MAX_SHARES (1UL << 18)
182 typedef int (*tg_visitor
)(struct task_group
*, void *);
184 extern int walk_tg_tree_from(struct task_group
*from
,
185 tg_visitor down
, tg_visitor up
, void *data
);
188 * Iterate the full tree, calling @down when first entering a node and @up when
189 * leaving it for the final time.
191 * Caller must hold rcu_lock or sufficient equivalent.
193 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
195 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
198 extern int tg_nop(struct task_group
*tg
, void *data
);
200 extern void free_fair_sched_group(struct task_group
*tg
);
201 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
202 extern void unregister_fair_sched_group(struct task_group
*tg
, int cpu
);
203 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
204 struct sched_entity
*se
, int cpu
,
205 struct sched_entity
*parent
);
206 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
207 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
209 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
210 extern void __start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
211 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
213 extern void free_rt_sched_group(struct task_group
*tg
);
214 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
215 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
216 struct sched_rt_entity
*rt_se
, int cpu
,
217 struct sched_rt_entity
*parent
);
219 extern struct task_group
*sched_create_group(struct task_group
*parent
);
220 extern void sched_online_group(struct task_group
*tg
,
221 struct task_group
*parent
);
222 extern void sched_destroy_group(struct task_group
*tg
);
223 extern void sched_offline_group(struct task_group
*tg
);
225 extern void sched_move_task(struct task_struct
*tsk
);
227 #ifdef CONFIG_FAIR_GROUP_SCHED
228 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
231 #else /* CONFIG_CGROUP_SCHED */
233 struct cfs_bandwidth
{ };
235 #endif /* CONFIG_CGROUP_SCHED */
237 /* CFS-related fields in a runqueue */
239 struct load_weight load
;
240 unsigned int nr_running
, h_nr_running
;
245 u64 min_vruntime_copy
;
248 struct rb_root tasks_timeline
;
249 struct rb_node
*rb_leftmost
;
252 * 'curr' points to currently running entity on this cfs_rq.
253 * It is set to NULL otherwise (i.e when none are currently running).
255 struct sched_entity
*curr
, *next
, *last
, *skip
;
257 #ifdef CONFIG_SCHED_DEBUG
258 unsigned int nr_spread_over
;
263 * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be
264 * removed when useful for applications beyond shares distribution (e.g.
267 #ifdef CONFIG_FAIR_GROUP_SCHED
270 * Under CFS, load is tracked on a per-entity basis and aggregated up.
271 * This allows for the description of both thread and group usage (in
272 * the FAIR_GROUP_SCHED case).
274 u64 runnable_load_avg
, blocked_load_avg
;
275 atomic64_t decay_counter
, removed_load
;
277 #endif /* CONFIG_FAIR_GROUP_SCHED */
278 /* These always depend on CONFIG_FAIR_GROUP_SCHED */
279 #ifdef CONFIG_FAIR_GROUP_SCHED
280 u32 tg_runnable_contrib
;
282 #endif /* CONFIG_FAIR_GROUP_SCHED */
285 * h_load = weight * f(tg)
287 * Where f(tg) is the recursive weight fraction assigned to
290 unsigned long h_load
;
291 #endif /* CONFIG_SMP */
293 #ifdef CONFIG_FAIR_GROUP_SCHED
294 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
297 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
298 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
299 * (like users, containers etc.)
301 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
302 * list is used during load balance.
305 struct list_head leaf_cfs_rq_list
;
306 struct task_group
*tg
; /* group that "owns" this runqueue */
308 #ifdef CONFIG_CFS_BANDWIDTH
311 s64 runtime_remaining
;
313 u64 throttled_clock
, throttled_clock_task
;
314 u64 throttled_clock_task_time
;
315 int throttled
, throttle_count
;
316 struct list_head throttled_list
;
317 #endif /* CONFIG_CFS_BANDWIDTH */
318 #endif /* CONFIG_FAIR_GROUP_SCHED */
321 static inline int rt_bandwidth_enabled(void)
323 return sysctl_sched_rt_runtime
>= 0;
326 /* Real-Time classes' related field in a runqueue: */
328 struct rt_prio_array active
;
329 unsigned int rt_nr_running
;
330 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
332 int curr
; /* highest queued rt task prio */
334 int next
; /* next highest */
339 unsigned long rt_nr_migratory
;
340 unsigned long rt_nr_total
;
342 struct plist_head pushable_tasks
;
347 /* Nests inside the rq lock: */
348 raw_spinlock_t rt_runtime_lock
;
350 #ifdef CONFIG_RT_GROUP_SCHED
351 unsigned long rt_nr_boosted
;
354 struct list_head leaf_rt_rq_list
;
355 struct task_group
*tg
;
362 * We add the notion of a root-domain which will be used to define per-domain
363 * variables. Each exclusive cpuset essentially defines an island domain by
364 * fully partitioning the member cpus from any other cpuset. Whenever a new
365 * exclusive cpuset is created, we also create and attach a new root-domain
374 cpumask_var_t online
;
377 * The "RT overload" flag: it gets set if a CPU has more than
378 * one runnable RT task.
380 cpumask_var_t rto_mask
;
381 struct cpupri cpupri
;
384 extern struct root_domain def_root_domain
;
386 #endif /* CONFIG_SMP */
389 * This is the main, per-CPU runqueue data structure.
391 * Locking rule: those places that want to lock multiple runqueues
392 * (such as the load balancing or the thread migration code), lock
393 * acquire operations must be ordered by ascending &runqueue.
400 * nr_running and cpu_load should be in the same cacheline because
401 * remote CPUs use both these fields when doing load calculation.
403 unsigned int nr_running
;
404 #define CPU_LOAD_IDX_MAX 5
405 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
406 unsigned long last_load_update_tick
;
407 #ifdef CONFIG_NO_HZ_COMMON
409 unsigned long nohz_flags
;
411 int skip_clock_update
;
413 /* capture load from *all* tasks on this cpu: */
414 struct load_weight load
;
415 unsigned long nr_load_updates
;
421 #ifdef CONFIG_FAIR_GROUP_SCHED
422 /* list of leaf cfs_rq on this cpu: */
423 struct list_head leaf_cfs_rq_list
;
425 unsigned long h_load_throttle
;
426 #endif /* CONFIG_SMP */
427 #endif /* CONFIG_FAIR_GROUP_SCHED */
429 #ifdef CONFIG_RT_GROUP_SCHED
430 struct list_head leaf_rt_rq_list
;
434 * This is part of a global counter where only the total sum
435 * over all CPUs matters. A task can increase this counter on
436 * one CPU and if it got migrated afterwards it may decrease
437 * it on another CPU. Always updated under the runqueue lock:
439 unsigned long nr_uninterruptible
;
441 struct task_struct
*curr
, *idle
, *stop
;
442 unsigned long next_balance
;
443 struct mm_struct
*prev_mm
;
451 struct root_domain
*rd
;
452 struct sched_domain
*sd
;
454 unsigned long cpu_power
;
456 unsigned char idle_balance
;
457 /* For active balancing */
461 struct cpu_stop_work active_balance_work
;
462 /* cpu of this runqueue: */
466 struct list_head cfs_tasks
;
474 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
477 #ifdef CONFIG_PARAVIRT
480 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
481 u64 prev_steal_time_rq
;
484 /* calc_load related fields */
485 unsigned long calc_load_update
;
486 long calc_load_active
;
488 #ifdef CONFIG_SCHED_HRTICK
490 int hrtick_csd_pending
;
491 struct call_single_data hrtick_csd
;
493 struct hrtimer hrtick_timer
;
496 #ifdef CONFIG_SCHEDSTATS
498 struct sched_info rq_sched_info
;
499 unsigned long long rq_cpu_time
;
500 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
502 /* sys_sched_yield() stats */
503 unsigned int yld_count
;
505 /* schedule() stats */
506 unsigned int sched_count
;
507 unsigned int sched_goidle
;
509 /* try_to_wake_up() stats */
510 unsigned int ttwu_count
;
511 unsigned int ttwu_local
;
515 struct llist_head wake_list
;
518 struct sched_avg avg
;
521 static inline int cpu_of(struct rq
*rq
)
530 DECLARE_PER_CPU(struct rq
, runqueues
);
532 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
533 #define this_rq() (&__get_cpu_var(runqueues))
534 #define task_rq(p) cpu_rq(task_cpu(p))
535 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
536 #define raw_rq() (&__raw_get_cpu_var(runqueues))
540 #define rcu_dereference_check_sched_domain(p) \
541 rcu_dereference_check((p), \
542 lockdep_is_held(&sched_domains_mutex))
545 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
546 * See detach_destroy_domains: synchronize_sched for details.
548 * The domain tree of any CPU may only be accessed from within
549 * preempt-disabled sections.
551 #define for_each_domain(cpu, __sd) \
552 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
553 __sd; __sd = __sd->parent)
555 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
558 * highest_flag_domain - Return highest sched_domain containing flag.
559 * @cpu: The cpu whose highest level of sched domain is to
561 * @flag: The flag to check for the highest sched_domain
564 * Returns the highest sched_domain of a cpu which contains the given flag.
566 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
568 struct sched_domain
*sd
, *hsd
= NULL
;
570 for_each_domain(cpu
, sd
) {
571 if (!(sd
->flags
& flag
))
579 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
580 DECLARE_PER_CPU(int, sd_llc_id
);
582 struct sched_group_power
{
585 * CPU power of this group, SCHED_LOAD_SCALE being max power for a
588 unsigned int power
, power_orig
;
589 unsigned long next_update
;
591 * Number of busy cpus in this group.
593 atomic_t nr_busy_cpus
;
595 unsigned long cpumask
[0]; /* iteration mask */
599 struct sched_group
*next
; /* Must be a circular list */
602 unsigned int group_weight
;
603 struct sched_group_power
*sgp
;
606 * The CPUs this group covers.
608 * NOTE: this field is variable length. (Allocated dynamically
609 * by attaching extra space to the end of the structure,
610 * depending on how many CPUs the kernel has booted up with)
612 unsigned long cpumask
[0];
615 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
617 return to_cpumask(sg
->cpumask
);
621 * cpumask masking which cpus in the group are allowed to iterate up the domain
624 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
626 return to_cpumask(sg
->sgp
->cpumask
);
630 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
631 * @group: The group whose first cpu is to be returned.
633 static inline unsigned int group_first_cpu(struct sched_group
*group
)
635 return cpumask_first(sched_group_cpus(group
));
638 extern int group_balance_cpu(struct sched_group
*sg
);
640 #endif /* CONFIG_SMP */
643 #include "auto_group.h"
645 #ifdef CONFIG_CGROUP_SCHED
648 * Return the group to which this tasks belongs.
650 * We cannot use task_subsys_state() and friends because the cgroup
651 * subsystem changes that value before the cgroup_subsys::attach() method
652 * is called, therefore we cannot pin it and might observe the wrong value.
654 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
655 * core changes this before calling sched_move_task().
657 * Instead we use a 'copy' which is updated from sched_move_task() while
658 * holding both task_struct::pi_lock and rq::lock.
660 static inline struct task_group
*task_group(struct task_struct
*p
)
662 return p
->sched_task_group
;
665 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
666 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
668 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
669 struct task_group
*tg
= task_group(p
);
672 #ifdef CONFIG_FAIR_GROUP_SCHED
673 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
674 p
->se
.parent
= tg
->se
[cpu
];
677 #ifdef CONFIG_RT_GROUP_SCHED
678 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
679 p
->rt
.parent
= tg
->rt_se
[cpu
];
683 #else /* CONFIG_CGROUP_SCHED */
685 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
686 static inline struct task_group
*task_group(struct task_struct
*p
)
691 #endif /* CONFIG_CGROUP_SCHED */
693 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
698 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
699 * successfuly executed on another CPU. We must ensure that updates of
700 * per-task data have been completed by this moment.
703 task_thread_info(p
)->cpu
= cpu
;
708 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
710 #ifdef CONFIG_SCHED_DEBUG
711 # include <linux/static_key.h>
712 # define const_debug __read_mostly
714 # define const_debug const
717 extern const_debug
unsigned int sysctl_sched_features
;
719 #define SCHED_FEAT(name, enabled) \
720 __SCHED_FEAT_##name ,
723 #include "features.h"
729 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
730 static __always_inline
bool static_branch__true(struct static_key
*key
)
732 return static_key_true(key
); /* Not out of line branch. */
735 static __always_inline
bool static_branch__false(struct static_key
*key
)
737 return static_key_false(key
); /* Out of line branch. */
740 #define SCHED_FEAT(name, enabled) \
741 static __always_inline bool static_branch_##name(struct static_key *key) \
743 return static_branch__##enabled(key); \
746 #include "features.h"
750 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
751 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
752 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
753 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
754 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
756 #ifdef CONFIG_NUMA_BALANCING
757 #define sched_feat_numa(x) sched_feat(x)
758 #ifdef CONFIG_SCHED_DEBUG
759 #define numabalancing_enabled sched_feat_numa(NUMA)
761 extern bool numabalancing_enabled
;
762 #endif /* CONFIG_SCHED_DEBUG */
764 #define sched_feat_numa(x) (0)
765 #define numabalancing_enabled (0)
766 #endif /* CONFIG_NUMA_BALANCING */
768 static inline u64
global_rt_period(void)
770 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
773 static inline u64
global_rt_runtime(void)
775 if (sysctl_sched_rt_runtime
< 0)
778 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
783 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
785 return rq
->curr
== p
;
788 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
793 return task_current(rq
, p
);
798 #ifndef prepare_arch_switch
799 # define prepare_arch_switch(next) do { } while (0)
801 #ifndef finish_arch_switch
802 # define finish_arch_switch(prev) do { } while (0)
804 #ifndef finish_arch_post_lock_switch
805 # define finish_arch_post_lock_switch() do { } while (0)
808 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
809 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
813 * We can optimise this out completely for !SMP, because the
814 * SMP rebalancing from interrupt is the only thing that cares
821 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
825 * After ->on_cpu is cleared, the task can be moved to a different CPU.
826 * We must ensure this doesn't happen until the switch is completely
832 #ifdef CONFIG_DEBUG_SPINLOCK
833 /* this is a valid case when another task releases the spinlock */
834 rq
->lock
.owner
= current
;
837 * If we are tracking spinlock dependencies then we have to
838 * fix up the runqueue lock - which gets 'carried over' from
841 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
843 raw_spin_unlock_irq(&rq
->lock
);
846 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
847 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
851 * We can optimise this out completely for !SMP, because the
852 * SMP rebalancing from interrupt is the only thing that cares
857 raw_spin_unlock(&rq
->lock
);
860 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
864 * After ->on_cpu is cleared, the task can be moved to a different CPU.
865 * We must ensure this doesn't happen until the switch is completely
873 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
878 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
879 #define WF_FORK 0x02 /* child wakeup after fork */
880 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
882 static inline void update_load_add(struct load_weight
*lw
, unsigned long inc
)
888 static inline void update_load_sub(struct load_weight
*lw
, unsigned long dec
)
894 static inline void update_load_set(struct load_weight
*lw
, unsigned long w
)
901 * To aid in avoiding the subversion of "niceness" due to uneven distribution
902 * of tasks with abnormal "nice" values across CPUs the contribution that
903 * each task makes to its run queue's load is weighted according to its
904 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
905 * scaled version of the new time slice allocation that they receive on time
909 #define WEIGHT_IDLEPRIO 3
910 #define WMULT_IDLEPRIO 1431655765
913 * Nice levels are multiplicative, with a gentle 10% change for every
914 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
915 * nice 1, it will get ~10% less CPU time than another CPU-bound task
916 * that remained on nice 0.
918 * The "10% effect" is relative and cumulative: from _any_ nice level,
919 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
920 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
921 * If a task goes up by ~10% and another task goes down by ~10% then
922 * the relative distance between them is ~25%.)
924 static const int prio_to_weight
[40] = {
925 /* -20 */ 88761, 71755, 56483, 46273, 36291,
926 /* -15 */ 29154, 23254, 18705, 14949, 11916,
927 /* -10 */ 9548, 7620, 6100, 4904, 3906,
928 /* -5 */ 3121, 2501, 1991, 1586, 1277,
929 /* 0 */ 1024, 820, 655, 526, 423,
930 /* 5 */ 335, 272, 215, 172, 137,
931 /* 10 */ 110, 87, 70, 56, 45,
932 /* 15 */ 36, 29, 23, 18, 15,
936 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
938 * In cases where the weight does not change often, we can use the
939 * precalculated inverse to speed up arithmetics by turning divisions
940 * into multiplications:
942 static const u32 prio_to_wmult
[40] = {
943 /* -20 */ 48388, 59856, 76040, 92818, 118348,
944 /* -15 */ 147320, 184698, 229616, 287308, 360437,
945 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
946 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
947 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
948 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
949 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
950 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
953 /* Time spent by the tasks of the cpu accounting group executing in ... */
954 enum cpuacct_stat_index
{
955 CPUACCT_STAT_USER
, /* ... user mode */
956 CPUACCT_STAT_SYSTEM
, /* ... kernel mode */
961 #define ENQUEUE_WAKEUP 1
962 #define ENQUEUE_HEAD 2
964 #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */
966 #define ENQUEUE_WAKING 0
969 #define DEQUEUE_SLEEP 1
972 const struct sched_class
*next
;
974 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
975 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
976 void (*yield_task
) (struct rq
*rq
);
977 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
979 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
981 struct task_struct
* (*pick_next_task
) (struct rq
*rq
);
982 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
985 int (*select_task_rq
)(struct task_struct
*p
, int sd_flag
, int flags
);
986 void (*migrate_task_rq
)(struct task_struct
*p
, int next_cpu
);
988 void (*pre_schedule
) (struct rq
*this_rq
, struct task_struct
*task
);
989 void (*post_schedule
) (struct rq
*this_rq
);
990 void (*task_waking
) (struct task_struct
*task
);
991 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
993 void (*set_cpus_allowed
)(struct task_struct
*p
,
994 const struct cpumask
*newmask
);
996 void (*rq_online
)(struct rq
*rq
);
997 void (*rq_offline
)(struct rq
*rq
);
1000 void (*set_curr_task
) (struct rq
*rq
);
1001 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1002 void (*task_fork
) (struct task_struct
*p
);
1004 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1005 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1006 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1009 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1010 struct task_struct
*task
);
1012 #ifdef CONFIG_FAIR_GROUP_SCHED
1013 void (*task_move_group
) (struct task_struct
*p
, int on_rq
);
1017 #define sched_class_highest (&stop_sched_class)
1018 #define for_each_class(class) \
1019 for (class = sched_class_highest; class; class = class->next)
1021 extern const struct sched_class stop_sched_class
;
1022 extern const struct sched_class rt_sched_class
;
1023 extern const struct sched_class fair_sched_class
;
1024 extern const struct sched_class idle_sched_class
;
1029 extern void update_group_power(struct sched_domain
*sd
, int cpu
);
1031 extern void trigger_load_balance(struct rq
*rq
, int cpu
);
1032 extern void idle_balance(int this_cpu
, struct rq
*this_rq
);
1034 #else /* CONFIG_SMP */
1036 static inline void idle_balance(int cpu
, struct rq
*rq
)
1042 extern void sysrq_sched_debug_show(void);
1043 extern void sched_init_granularity(void);
1044 extern void update_max_interval(void);
1045 extern int update_runtime(struct notifier_block
*nfb
, unsigned long action
, void *hcpu
);
1046 extern void init_sched_rt_class(void);
1047 extern void init_sched_fair_class(void);
1049 extern void resched_task(struct task_struct
*p
);
1050 extern void resched_cpu(int cpu
);
1052 extern struct rt_bandwidth def_rt_bandwidth
;
1053 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1055 extern void update_idle_cpu_load(struct rq
*this_rq
);
1057 #ifdef CONFIG_CGROUP_CPUACCT
1058 #include <linux/cgroup.h>
1059 /* track cpu usage of a group of tasks and its child groups */
1061 struct cgroup_subsys_state css
;
1062 /* cpuusage holds pointer to a u64-type object on every cpu */
1063 u64 __percpu
*cpuusage
;
1064 struct kernel_cpustat __percpu
*cpustat
;
1067 extern struct cgroup_subsys cpuacct_subsys
;
1068 extern struct cpuacct root_cpuacct
;
1070 /* return cpu accounting group corresponding to this container */
1071 static inline struct cpuacct
*cgroup_ca(struct cgroup
*cgrp
)
1073 return container_of(cgroup_subsys_state(cgrp
, cpuacct_subsys_id
),
1074 struct cpuacct
, css
);
1077 /* return cpu accounting group to which this task belongs */
1078 static inline struct cpuacct
*task_ca(struct task_struct
*tsk
)
1080 return container_of(task_subsys_state(tsk
, cpuacct_subsys_id
),
1081 struct cpuacct
, css
);
1084 static inline struct cpuacct
*parent_ca(struct cpuacct
*ca
)
1086 if (!ca
|| !ca
->css
.cgroup
->parent
)
1088 return cgroup_ca(ca
->css
.cgroup
->parent
);
1091 extern void cpuacct_charge(struct task_struct
*tsk
, u64 cputime
);
1093 static inline void cpuacct_charge(struct task_struct
*tsk
, u64 cputime
) {}
1096 #ifdef CONFIG_PARAVIRT
1097 static inline u64
steal_ticks(u64 steal
)
1099 if (unlikely(steal
> NSEC_PER_SEC
))
1100 return div_u64(steal
, TICK_NSEC
);
1102 return __iter_div_u64_rem(steal
, TICK_NSEC
, &steal
);
1106 static inline void inc_nr_running(struct rq
*rq
)
1111 static inline void dec_nr_running(struct rq
*rq
)
1116 extern void update_rq_clock(struct rq
*rq
);
1118 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1119 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1121 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1123 extern const_debug
unsigned int sysctl_sched_time_avg
;
1124 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1125 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1127 static inline u64
sched_avg_period(void)
1129 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1132 #ifdef CONFIG_SCHED_HRTICK
1136 * - enabled by features
1137 * - hrtimer is actually high res
1139 static inline int hrtick_enabled(struct rq
*rq
)
1141 if (!sched_feat(HRTICK
))
1143 if (!cpu_active(cpu_of(rq
)))
1145 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1148 void hrtick_start(struct rq
*rq
, u64 delay
);
1152 static inline int hrtick_enabled(struct rq
*rq
)
1157 #endif /* CONFIG_SCHED_HRTICK */
1160 extern void sched_avg_update(struct rq
*rq
);
1161 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1163 rq
->rt_avg
+= rt_delta
;
1164 sched_avg_update(rq
);
1167 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1168 static inline void sched_avg_update(struct rq
*rq
) { }
1171 extern void start_bandwidth_timer(struct hrtimer
*period_timer
, ktime_t period
);
1174 #ifdef CONFIG_PREEMPT
1176 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1179 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1180 * way at the expense of forcing extra atomic operations in all
1181 * invocations. This assures that the double_lock is acquired using the
1182 * same underlying policy as the spinlock_t on this architecture, which
1183 * reduces latency compared to the unfair variant below. However, it
1184 * also adds more overhead and therefore may reduce throughput.
1186 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1187 __releases(this_rq
->lock
)
1188 __acquires(busiest
->lock
)
1189 __acquires(this_rq
->lock
)
1191 raw_spin_unlock(&this_rq
->lock
);
1192 double_rq_lock(this_rq
, busiest
);
1199 * Unfair double_lock_balance: Optimizes throughput at the expense of
1200 * latency by eliminating extra atomic operations when the locks are
1201 * already in proper order on entry. This favors lower cpu-ids and will
1202 * grant the double lock to lower cpus over higher ids under contention,
1203 * regardless of entry order into the function.
1205 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1206 __releases(this_rq
->lock
)
1207 __acquires(busiest
->lock
)
1208 __acquires(this_rq
->lock
)
1212 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1213 if (busiest
< this_rq
) {
1214 raw_spin_unlock(&this_rq
->lock
);
1215 raw_spin_lock(&busiest
->lock
);
1216 raw_spin_lock_nested(&this_rq
->lock
,
1217 SINGLE_DEPTH_NESTING
);
1220 raw_spin_lock_nested(&busiest
->lock
,
1221 SINGLE_DEPTH_NESTING
);
1226 #endif /* CONFIG_PREEMPT */
1229 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1231 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1233 if (unlikely(!irqs_disabled())) {
1234 /* printk() doesn't work good under rq->lock */
1235 raw_spin_unlock(&this_rq
->lock
);
1239 return _double_lock_balance(this_rq
, busiest
);
1242 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1243 __releases(busiest
->lock
)
1245 raw_spin_unlock(&busiest
->lock
);
1246 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1250 * double_rq_lock - safely lock two runqueues
1252 * Note this does not disable interrupts like task_rq_lock,
1253 * you need to do so manually before calling.
1255 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1256 __acquires(rq1
->lock
)
1257 __acquires(rq2
->lock
)
1259 BUG_ON(!irqs_disabled());
1261 raw_spin_lock(&rq1
->lock
);
1262 __acquire(rq2
->lock
); /* Fake it out ;) */
1265 raw_spin_lock(&rq1
->lock
);
1266 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1268 raw_spin_lock(&rq2
->lock
);
1269 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1275 * double_rq_unlock - safely unlock two runqueues
1277 * Note this does not restore interrupts like task_rq_unlock,
1278 * you need to do so manually after calling.
1280 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1281 __releases(rq1
->lock
)
1282 __releases(rq2
->lock
)
1284 raw_spin_unlock(&rq1
->lock
);
1286 raw_spin_unlock(&rq2
->lock
);
1288 __release(rq2
->lock
);
1291 #else /* CONFIG_SMP */
1294 * double_rq_lock - safely lock two runqueues
1296 * Note this does not disable interrupts like task_rq_lock,
1297 * you need to do so manually before calling.
1299 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1300 __acquires(rq1
->lock
)
1301 __acquires(rq2
->lock
)
1303 BUG_ON(!irqs_disabled());
1305 raw_spin_lock(&rq1
->lock
);
1306 __acquire(rq2
->lock
); /* Fake it out ;) */
1310 * double_rq_unlock - safely unlock two runqueues
1312 * Note this does not restore interrupts like task_rq_unlock,
1313 * you need to do so manually after calling.
1315 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1316 __releases(rq1
->lock
)
1317 __releases(rq2
->lock
)
1320 raw_spin_unlock(&rq1
->lock
);
1321 __release(rq2
->lock
);
1326 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1327 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1328 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1329 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1331 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1332 extern void init_rt_rq(struct rt_rq
*rt_rq
, struct rq
*rq
);
1334 extern void account_cfs_bandwidth_used(int enabled
, int was_enabled
);
1336 #ifdef CONFIG_NO_HZ_COMMON
1337 enum rq_nohz_flag_bits
{
1343 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1346 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1348 DECLARE_PER_CPU(u64
, cpu_hardirq_time
);
1349 DECLARE_PER_CPU(u64
, cpu_softirq_time
);
1351 #ifndef CONFIG_64BIT
1352 DECLARE_PER_CPU(seqcount_t
, irq_time_seq
);
1354 static inline void irq_time_write_begin(void)
1356 __this_cpu_inc(irq_time_seq
.sequence
);
1360 static inline void irq_time_write_end(void)
1363 __this_cpu_inc(irq_time_seq
.sequence
);
1366 static inline u64
irq_time_read(int cpu
)
1372 seq
= read_seqcount_begin(&per_cpu(irq_time_seq
, cpu
));
1373 irq_time
= per_cpu(cpu_softirq_time
, cpu
) +
1374 per_cpu(cpu_hardirq_time
, cpu
);
1375 } while (read_seqcount_retry(&per_cpu(irq_time_seq
, cpu
), seq
));
1379 #else /* CONFIG_64BIT */
1380 static inline void irq_time_write_begin(void)
1384 static inline void irq_time_write_end(void)
1388 static inline u64
irq_time_read(int cpu
)
1390 return per_cpu(cpu_softirq_time
, cpu
) + per_cpu(cpu_hardirq_time
, cpu
);
1392 #endif /* CONFIG_64BIT */
1393 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */