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>
8 #include <linux/tick.h>
13 extern __read_mostly
int scheduler_running
;
16 * Convert user-nice values [ -20 ... 0 ... 19 ]
17 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
20 #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
21 #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
22 #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
24 extern unsigned long get_cpu_load(int cpu
);
26 * 'User priority' is the nice value converted to something we
27 * can work with better when scaling various scheduler parameters,
28 * it's a [ 0 ... 39 ] range.
30 #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
31 #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
32 #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
35 * Helpers for converting nanosecond timing to jiffy resolution
37 #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
40 * Increase resolution of nice-level calculations for 64-bit architectures.
41 * The extra resolution improves shares distribution and load balancing of
42 * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup
43 * hierarchies, especially on larger systems. This is not a user-visible change
44 * and does not change the user-interface for setting shares/weights.
46 * We increase resolution only if we have enough bits to allow this increased
47 * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution
48 * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the
51 #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */
52 # define SCHED_LOAD_RESOLUTION 10
53 # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION)
54 # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION)
56 # define SCHED_LOAD_RESOLUTION 0
57 # define scale_load(w) (w)
58 # define scale_load_down(w) (w)
61 #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION)
62 #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT)
64 #define NICE_0_LOAD SCHED_LOAD_SCALE
65 #define NICE_0_SHIFT SCHED_LOAD_SHIFT
68 * These are the 'tuning knobs' of the scheduler:
72 * single value that denotes runtime == period, ie unlimited time.
74 #define RUNTIME_INF ((u64)~0ULL)
76 static inline int rt_policy(int policy
)
78 if (policy
== SCHED_FIFO
|| policy
== SCHED_RR
)
83 static inline int task_has_rt_policy(struct task_struct
*p
)
85 return rt_policy(p
->policy
);
89 * This is the priority-queue data structure of the RT scheduling class:
91 struct rt_prio_array
{
92 DECLARE_BITMAP(bitmap
, MAX_RT_PRIO
+1); /* include 1 bit for delimiter */
93 struct list_head queue
[MAX_RT_PRIO
];
97 /* nests inside the rq lock: */
98 raw_spinlock_t rt_runtime_lock
;
101 struct hrtimer rt_period_timer
;
104 extern struct mutex sched_domains_mutex
;
106 #ifdef CONFIG_CGROUP_SCHED
108 #include <linux/cgroup.h>
113 extern struct list_head task_groups
;
115 struct cfs_bandwidth
{
116 #ifdef CONFIG_CFS_BANDWIDTH
120 s64 hierarchal_quota
;
123 int idle
, timer_active
;
124 struct hrtimer period_timer
, slack_timer
;
125 struct list_head throttled_cfs_rq
;
128 int nr_periods
, nr_throttled
;
133 /* task group related information */
135 struct cgroup_subsys_state css
;
137 #ifdef CONFIG_FAIR_GROUP_SCHED
138 /* schedulable entities of this group on each cpu */
139 struct sched_entity
**se
;
140 /* runqueue "owned" by this group on each cpu */
141 struct cfs_rq
**cfs_rq
;
142 unsigned long shares
;
145 atomic_long_t load_avg
;
146 atomic_t runnable_avg
, usage_avg
;
150 #ifdef CONFIG_RT_GROUP_SCHED
151 struct sched_rt_entity
**rt_se
;
152 struct rt_rq
**rt_rq
;
154 struct rt_bandwidth rt_bandwidth
;
158 struct list_head list
;
160 struct task_group
*parent
;
161 struct list_head siblings
;
162 struct list_head children
;
164 #ifdef CONFIG_SCHED_AUTOGROUP
165 struct autogroup
*autogroup
;
168 struct cfs_bandwidth cfs_bandwidth
;
171 #ifdef CONFIG_FAIR_GROUP_SCHED
172 #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
175 * A weight of 0 or 1 can cause arithmetics problems.
176 * A weight of a cfs_rq is the sum of weights of which entities
177 * are queued on this cfs_rq, so a weight of a entity should not be
178 * too large, so as the shares value of a task group.
179 * (The default weight is 1024 - so there's no practical
180 * limitation from this.)
182 #define MIN_SHARES (1UL << 1)
183 #define MAX_SHARES (1UL << 18)
186 typedef int (*tg_visitor
)(struct task_group
*, void *);
188 extern int walk_tg_tree_from(struct task_group
*from
,
189 tg_visitor down
, tg_visitor up
, void *data
);
192 * Iterate the full tree, calling @down when first entering a node and @up when
193 * leaving it for the final time.
195 * Caller must hold rcu_lock or sufficient equivalent.
197 static inline int walk_tg_tree(tg_visitor down
, tg_visitor up
, void *data
)
199 return walk_tg_tree_from(&root_task_group
, down
, up
, data
);
202 extern int tg_nop(struct task_group
*tg
, void *data
);
204 extern void free_fair_sched_group(struct task_group
*tg
);
205 extern int alloc_fair_sched_group(struct task_group
*tg
, struct task_group
*parent
);
206 extern void unregister_fair_sched_group(struct task_group
*tg
, int cpu
);
207 extern void init_tg_cfs_entry(struct task_group
*tg
, struct cfs_rq
*cfs_rq
,
208 struct sched_entity
*se
, int cpu
,
209 struct sched_entity
*parent
);
210 extern void init_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
211 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
213 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth
*cfs_b
);
214 extern void __start_cfs_bandwidth(struct cfs_bandwidth
*cfs_b
);
215 extern void unthrottle_cfs_rq(struct cfs_rq
*cfs_rq
);
217 extern void free_rt_sched_group(struct task_group
*tg
);
218 extern int alloc_rt_sched_group(struct task_group
*tg
, struct task_group
*parent
);
219 extern void init_tg_rt_entry(struct task_group
*tg
, struct rt_rq
*rt_rq
,
220 struct sched_rt_entity
*rt_se
, int cpu
,
221 struct sched_rt_entity
*parent
);
223 extern struct task_group
*sched_create_group(struct task_group
*parent
);
224 extern void sched_online_group(struct task_group
*tg
,
225 struct task_group
*parent
);
226 extern void sched_destroy_group(struct task_group
*tg
);
227 extern void sched_offline_group(struct task_group
*tg
);
229 extern void sched_move_task(struct task_struct
*tsk
);
231 #ifdef CONFIG_FAIR_GROUP_SCHED
232 extern int sched_group_set_shares(struct task_group
*tg
, unsigned long shares
);
235 #else /* CONFIG_CGROUP_SCHED */
237 struct cfs_bandwidth
{ };
239 #endif /* CONFIG_CGROUP_SCHED */
241 /* CFS-related fields in a runqueue */
243 struct load_weight load
;
244 unsigned int nr_running
, h_nr_running
;
249 u64 min_vruntime_copy
;
252 struct rb_root tasks_timeline
;
253 struct rb_node
*rb_leftmost
;
256 * 'curr' points to currently running entity on this cfs_rq.
257 * It is set to NULL otherwise (i.e when none are currently running).
259 struct sched_entity
*curr
, *next
, *last
, *skip
;
261 #ifdef CONFIG_SCHED_DEBUG
262 unsigned int nr_spread_over
;
268 * Under CFS, load is tracked on a per-entity basis and aggregated up.
269 * This allows for the description of both thread and group usage (in
270 * the FAIR_GROUP_SCHED case).
272 unsigned long runnable_load_avg
, blocked_load_avg
;
273 atomic64_t decay_counter
;
275 atomic_long_t removed_load
;
277 #ifdef CONFIG_FAIR_GROUP_SCHED
278 /* Required to track per-cpu representation of a task_group */
279 u32 tg_runnable_contrib
, tg_usage_contrib
;
280 unsigned long tg_load_contrib
;
281 #endif /* CONFIG_FAIR_GROUP_SCHED */
283 struct sched_avg avg
;
286 * h_load = weight * f(tg)
288 * Where f(tg) is the recursive weight fraction assigned to
291 unsigned long h_load
;
292 #endif /* CONFIG_SMP */
294 #ifdef CONFIG_FAIR_GROUP_SCHED
295 struct rq
*rq
; /* cpu runqueue to which this cfs_rq is attached */
298 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
299 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
300 * (like users, containers etc.)
302 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
303 * list is used during load balance.
306 struct list_head leaf_cfs_rq_list
;
307 struct task_group
*tg
; /* group that "owns" this runqueue */
309 #ifdef CONFIG_CFS_BANDWIDTH
312 s64 runtime_remaining
;
314 u64 throttled_clock
, throttled_clock_task
;
315 u64 throttled_clock_task_time
;
316 int throttled
, throttle_count
;
317 struct list_head throttled_list
;
318 #endif /* CONFIG_CFS_BANDWIDTH */
319 #endif /* CONFIG_FAIR_GROUP_SCHED */
322 static inline int rt_bandwidth_enabled(void)
324 return sysctl_sched_rt_runtime
>= 0;
327 /* Real-Time classes' related field in a runqueue: */
329 struct rt_prio_array active
;
330 unsigned int rt_nr_running
;
331 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
333 int curr
; /* highest queued rt task prio */
335 int next
; /* next highest */
340 unsigned long rt_nr_migratory
;
341 unsigned long rt_nr_total
;
343 struct plist_head pushable_tasks
;
346 int rt_disable_borrow
;
349 /* Nests inside the rq lock: */
350 raw_spinlock_t rt_runtime_lock
;
352 #ifdef CONFIG_RT_GROUP_SCHED
353 unsigned long rt_nr_boosted
;
356 struct list_head leaf_rt_rq_list
;
357 struct task_group
*tg
;
364 * We add the notion of a root-domain which will be used to define per-domain
365 * variables. Each exclusive cpuset essentially defines an island domain by
366 * fully partitioning the member cpus from any other cpuset. Whenever a new
367 * exclusive cpuset is created, we also create and attach a new root-domain
376 cpumask_var_t online
;
379 * The "RT overload" flag: it gets set if a CPU has more than
380 * one runnable RT task.
382 cpumask_var_t rto_mask
;
383 struct cpupri cpupri
;
386 extern struct root_domain def_root_domain
;
388 #endif /* CONFIG_SMP */
391 * This is the main, per-CPU runqueue data structure.
393 * Locking rule: those places that want to lock multiple runqueues
394 * (such as the load balancing or the thread migration code), lock
395 * acquire operations must be ordered by ascending &runqueue.
402 * nr_running and cpu_load should be in the same cacheline because
403 * remote CPUs use both these fields when doing load calculation.
405 unsigned int nr_running
;
406 #define CPU_LOAD_IDX_MAX 5
407 unsigned long cpu_load
[CPU_LOAD_IDX_MAX
];
408 unsigned long last_load_update_tick
;
409 #ifdef CONFIG_NO_HZ_COMMON
411 unsigned long nohz_flags
;
413 #ifdef CONFIG_NO_HZ_FULL
414 unsigned long last_sched_tick
;
416 int skip_clock_update
;
418 /* capture load from *all* tasks on this cpu: */
419 struct load_weight load
;
420 unsigned long nr_load_updates
;
426 #ifdef CONFIG_FAIR_GROUP_SCHED
427 /* list of leaf cfs_rq on this cpu: */
428 struct list_head leaf_cfs_rq_list
;
430 unsigned long h_load_throttle
;
431 #endif /* CONFIG_SMP */
432 #endif /* CONFIG_FAIR_GROUP_SCHED */
434 #ifdef CONFIG_RT_GROUP_SCHED
435 struct list_head leaf_rt_rq_list
;
439 * This is part of a global counter where only the total sum
440 * over all CPUs matters. A task can increase this counter on
441 * one CPU and if it got migrated afterwards it may decrease
442 * it on another CPU. Always updated under the runqueue lock:
444 unsigned long nr_uninterruptible
;
446 struct task_struct
*curr
, *idle
, *stop
;
447 unsigned long next_balance
;
448 struct mm_struct
*prev_mm
;
456 struct root_domain
*rd
;
457 struct sched_domain
*sd
;
459 unsigned long cpu_power
;
461 unsigned char idle_balance
;
462 /* For active balancing */
466 struct cpu_stop_work active_balance_work
;
467 #ifdef CONFIG_SCHED_HMP
468 struct task_struct
*migrate_task
;
470 /* cpu of this runqueue: */
474 struct list_head cfs_tasks
;
482 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
485 #ifdef CONFIG_PARAVIRT
488 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
489 u64 prev_steal_time_rq
;
492 /* calc_load related fields */
493 unsigned long calc_load_update
;
494 long calc_load_active
;
496 #ifdef CONFIG_SCHED_HRTICK
498 int hrtick_csd_pending
;
499 struct call_single_data hrtick_csd
;
501 struct hrtimer hrtick_timer
;
504 #ifdef CONFIG_SCHEDSTATS
506 struct sched_info rq_sched_info
;
507 unsigned long long rq_cpu_time
;
508 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
510 /* sys_sched_yield() stats */
511 unsigned int yld_count
;
513 /* schedule() stats */
514 unsigned int sched_count
;
515 unsigned int sched_goidle
;
517 /* try_to_wake_up() stats */
518 unsigned int ttwu_count
;
519 unsigned int ttwu_local
;
523 struct llist_head wake_list
;
526 struct sched_avg avg
;
529 static inline int cpu_of(struct rq
*rq
)
538 DECLARE_PER_CPU(struct rq
, runqueues
);
540 #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
541 #define this_rq() (&__get_cpu_var(runqueues))
542 #define task_rq(p) cpu_rq(task_cpu(p))
543 #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
544 #define raw_rq() (&__raw_get_cpu_var(runqueues))
548 #define rcu_dereference_check_sched_domain(p) \
549 rcu_dereference_check((p), \
550 lockdep_is_held(&sched_domains_mutex))
553 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
554 * See detach_destroy_domains: synchronize_sched for details.
556 * The domain tree of any CPU may only be accessed from within
557 * preempt-disabled sections.
559 #define for_each_domain(cpu, __sd) \
560 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
561 __sd; __sd = __sd->parent)
563 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
566 * highest_flag_domain - Return highest sched_domain containing flag.
567 * @cpu: The cpu whose highest level of sched domain is to
569 * @flag: The flag to check for the highest sched_domain
572 * Returns the highest sched_domain of a cpu which contains the given flag.
574 static inline struct sched_domain
*highest_flag_domain(int cpu
, int flag
)
576 struct sched_domain
*sd
, *hsd
= NULL
;
578 for_each_domain(cpu
, sd
) {
579 if (!(sd
->flags
& flag
))
587 DECLARE_PER_CPU(struct sched_domain
*, sd_llc
);
588 DECLARE_PER_CPU(int, sd_llc_id
);
590 struct sched_group_power
{
593 * CPU power of this group, SCHED_LOAD_SCALE being max power for a
596 unsigned int power
, power_orig
;
597 unsigned long next_update
;
599 * Number of busy cpus in this group.
601 atomic_t nr_busy_cpus
;
603 unsigned long cpumask
[0]; /* iteration mask */
607 struct sched_group
*next
; /* Must be a circular list */
610 unsigned int group_weight
;
611 struct sched_group_power
*sgp
;
614 * The CPUs this group covers.
616 * NOTE: this field is variable length. (Allocated dynamically
617 * by attaching extra space to the end of the structure,
618 * depending on how many CPUs the kernel has booted up with)
620 unsigned long cpumask
[0];
623 static inline struct cpumask
*sched_group_cpus(struct sched_group
*sg
)
625 return to_cpumask(sg
->cpumask
);
629 * cpumask masking which cpus in the group are allowed to iterate up the domain
632 static inline struct cpumask
*sched_group_mask(struct sched_group
*sg
)
634 return to_cpumask(sg
->sgp
->cpumask
);
638 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
639 * @group: The group whose first cpu is to be returned.
641 static inline unsigned int group_first_cpu(struct sched_group
*group
)
643 return cpumask_first(sched_group_cpus(group
));
646 extern int group_balance_cpu(struct sched_group
*sg
);
648 #ifdef CONFIG_SCHED_HMP
649 static LIST_HEAD(hmp_domains
);
650 DECLARE_PER_CPU(struct hmp_domain
*, hmp_cpu_domain
);
651 #define hmp_cpu_domain(cpu) (per_cpu(hmp_cpu_domain, (cpu)))
652 #endif /* CONFIG_SCHED_HMP */
654 #endif /* CONFIG_SMP */
657 #include "auto_group.h"
659 #ifdef CONFIG_CGROUP_SCHED
662 * Return the group to which this tasks belongs.
664 * We cannot use task_subsys_state() and friends because the cgroup
665 * subsystem changes that value before the cgroup_subsys::attach() method
666 * is called, therefore we cannot pin it and might observe the wrong value.
668 * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
669 * core changes this before calling sched_move_task().
671 * Instead we use a 'copy' which is updated from sched_move_task() while
672 * holding both task_struct::pi_lock and rq::lock.
674 static inline struct task_group
*task_group(struct task_struct
*p
)
676 return p
->sched_task_group
;
679 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
680 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
)
682 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
683 struct task_group
*tg
= task_group(p
);
686 #ifdef CONFIG_FAIR_GROUP_SCHED
687 p
->se
.cfs_rq
= tg
->cfs_rq
[cpu
];
688 p
->se
.parent
= tg
->se
[cpu
];
691 #ifdef CONFIG_RT_GROUP_SCHED
692 p
->rt
.rt_rq
= tg
->rt_rq
[cpu
];
693 p
->rt
.parent
= tg
->rt_se
[cpu
];
697 #else /* CONFIG_CGROUP_SCHED */
699 static inline void set_task_rq(struct task_struct
*p
, unsigned int cpu
) { }
700 static inline struct task_group
*task_group(struct task_struct
*p
)
705 #endif /* CONFIG_CGROUP_SCHED */
707 static inline void __set_task_cpu(struct task_struct
*p
, unsigned int cpu
)
712 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
713 * successfuly executed on another CPU. We must ensure that updates of
714 * per-task data have been completed by this moment.
717 task_thread_info(p
)->cpu
= cpu
;
719 #ifdef CONFIG_FAIR_GROUP_SCHED
720 BUG_ON(p
->se
.cfs_rq
->rq
->cpu
!= cpu
);
727 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
729 #ifdef CONFIG_SCHED_DEBUG
730 # include <linux/static_key.h>
731 # define const_debug __read_mostly
733 # define const_debug const
736 extern const_debug
unsigned int sysctl_sched_features
;
738 #define SCHED_FEAT(name, enabled) \
739 __SCHED_FEAT_##name ,
742 #include "features.h"
748 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
749 static __always_inline
bool static_branch__true(struct static_key
*key
)
751 return static_key_true(key
); /* Not out of line branch. */
754 static __always_inline
bool static_branch__false(struct static_key
*key
)
756 return static_key_false(key
); /* Out of line branch. */
759 #define SCHED_FEAT(name, enabled) \
760 static __always_inline bool static_branch_##name(struct static_key *key) \
762 return static_branch__##enabled(key); \
765 #include "features.h"
769 extern struct static_key sched_feat_keys
[__SCHED_FEAT_NR
];
770 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
771 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
772 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
773 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
775 #ifdef CONFIG_NUMA_BALANCING
776 #define sched_feat_numa(x) sched_feat(x)
777 #ifdef CONFIG_SCHED_DEBUG
778 #define numabalancing_enabled sched_feat_numa(NUMA)
780 extern bool numabalancing_enabled
;
781 #endif /* CONFIG_SCHED_DEBUG */
783 #define sched_feat_numa(x) (0)
784 #define numabalancing_enabled (0)
785 #endif /* CONFIG_NUMA_BALANCING */
787 static inline u64
global_rt_period(void)
789 return (u64
)sysctl_sched_rt_period
* NSEC_PER_USEC
;
792 static inline u64
global_rt_runtime(void)
794 if (sysctl_sched_rt_runtime
< 0)
797 return (u64
)sysctl_sched_rt_runtime
* NSEC_PER_USEC
;
802 static inline int task_current(struct rq
*rq
, struct task_struct
*p
)
804 return rq
->curr
== p
;
807 static inline int task_running(struct rq
*rq
, struct task_struct
*p
)
812 return task_current(rq
, p
);
817 #ifndef prepare_arch_switch
818 # define prepare_arch_switch(next) do { } while (0)
820 #ifndef finish_arch_switch
821 # define finish_arch_switch(prev) do { } while (0)
823 #ifndef finish_arch_post_lock_switch
824 # define finish_arch_post_lock_switch() do { } while (0)
827 #ifdef CONFIG_MT_RT_SCHED
828 extern void mt_check_rt_policy(struct rq
*this_rq
);
829 extern int push_need_released_rt_task(struct rq
*rq
, struct task_struct
*p
);
830 extern int pull_rt_task(struct rq
*this_rq
);
831 extern int mt_post_schedule(struct rq
*rq
);
834 #ifdef CONFIG_MT_RT_SCHED_LOG
835 #ifdef CONFIG_MT_RT_SCHED_DEBUG
836 #define mt_rt_printf(x...) \
838 char strings[128]=""; \
839 snprintf(strings, 128, x); \
840 printk(KERN_NOTICE x); \
841 trace_sched_rt_log(strings); \
844 #define mt_rt_printf(x...) \
846 char strings[128]=""; \
847 snprintf(strings, 128, x); \
848 trace_sched_rt_log(strings); \
852 #define mt_rt_printf do {} while (0)
855 #ifndef __ARCH_WANT_UNLOCKED_CTXSW
856 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
860 * We can optimise this out completely for !SMP, because the
861 * SMP rebalancing from interrupt is the only thing that cares
868 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
872 * After ->on_cpu is cleared, the task can be moved to a different CPU.
873 * We must ensure this doesn't happen until the switch is completely
879 #ifdef CONFIG_DEBUG_SPINLOCK
880 /* this is a valid case when another task releases the spinlock */
881 rq
->lock
.owner
= current
;
884 * If we are tracking spinlock dependencies then we have to
885 * fix up the runqueue lock - which gets 'carried over' from
888 spin_acquire(&rq
->lock
.dep_map
, 0, 0, _THIS_IP_
);
889 #ifdef CONFIG_MT_RT_SCHED
890 if(test_tsk_need_released(prev
)){
891 clear_tsk_need_released(prev
);
892 push_need_released_rt_task(rq
, prev
);
895 raw_spin_unlock_irq(&rq
->lock
);
898 #else /* __ARCH_WANT_UNLOCKED_CTXSW */
899 static inline void prepare_lock_switch(struct rq
*rq
, struct task_struct
*next
)
903 * We can optimise this out completely for !SMP, because the
904 * SMP rebalancing from interrupt is the only thing that cares
909 raw_spin_unlock(&rq
->lock
);
912 static inline void finish_lock_switch(struct rq
*rq
, struct task_struct
*prev
)
916 * After ->on_cpu is cleared, the task can be moved to a different CPU.
917 * We must ensure this doesn't happen until the switch is completely
925 #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
930 #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
931 #define WF_FORK 0x02 /* child wakeup after fork */
932 #define WF_MIGRATED 0x4 /* internal use, task got migrated */
934 static inline void update_load_add(struct load_weight
*lw
, unsigned long inc
)
940 static inline void update_load_sub(struct load_weight
*lw
, unsigned long dec
)
946 static inline void update_load_set(struct load_weight
*lw
, unsigned long w
)
953 * To aid in avoiding the subversion of "niceness" due to uneven distribution
954 * of tasks with abnormal "nice" values across CPUs the contribution that
955 * each task makes to its run queue's load is weighted according to its
956 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
957 * scaled version of the new time slice allocation that they receive on time
961 #define WEIGHT_IDLEPRIO 3
962 #define WMULT_IDLEPRIO 1431655765
965 * Nice levels are multiplicative, with a gentle 10% change for every
966 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
967 * nice 1, it will get ~10% less CPU time than another CPU-bound task
968 * that remained on nice 0.
970 * The "10% effect" is relative and cumulative: from _any_ nice level,
971 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
972 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
973 * If a task goes up by ~10% and another task goes down by ~10% then
974 * the relative distance between them is ~25%.)
976 static const int prio_to_weight
[40] = {
977 /* -20 */ 88761, 71755, 56483, 46273, 36291,
978 /* -15 */ 29154, 23254, 18705, 14949, 11916,
979 /* -10 */ 9548, 7620, 6100, 4904, 3906,
980 /* -5 */ 3121, 2501, 1991, 1586, 1277,
981 /* 0 */ 1024, 820, 655, 526, 423,
982 /* 5 */ 335, 272, 215, 172, 137,
983 /* 10 */ 110, 87, 70, 56, 45,
984 /* 15 */ 36, 29, 23, 18, 15,
988 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
990 * In cases where the weight does not change often, we can use the
991 * precalculated inverse to speed up arithmetics by turning divisions
992 * into multiplications:
994 static const u32 prio_to_wmult
[40] = {
995 /* -20 */ 48388, 59856, 76040, 92818, 118348,
996 /* -15 */ 147320, 184698, 229616, 287308, 360437,
997 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
998 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
999 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1000 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1001 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1002 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1005 #define ENQUEUE_WAKEUP 1
1006 #define ENQUEUE_HEAD 2
1008 #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */
1010 #define ENQUEUE_WAKING 0
1013 #define DEQUEUE_SLEEP 1
1015 struct sched_class
{
1016 const struct sched_class
*next
;
1018 void (*enqueue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1019 void (*dequeue_task
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1020 void (*yield_task
) (struct rq
*rq
);
1021 bool (*yield_to_task
) (struct rq
*rq
, struct task_struct
*p
, bool preempt
);
1023 void (*check_preempt_curr
) (struct rq
*rq
, struct task_struct
*p
, int flags
);
1025 struct task_struct
* (*pick_next_task
) (struct rq
*rq
);
1026 void (*put_prev_task
) (struct rq
*rq
, struct task_struct
*p
);
1029 int (*select_task_rq
)(struct task_struct
*p
, int sd_flag
, int flags
);
1030 void (*migrate_task_rq
)(struct task_struct
*p
, int next_cpu
);
1032 void (*pre_schedule
) (struct rq
*this_rq
, struct task_struct
*task
);
1033 void (*post_schedule
) (struct rq
*this_rq
);
1034 void (*task_waking
) (struct task_struct
*task
);
1035 void (*task_woken
) (struct rq
*this_rq
, struct task_struct
*task
);
1037 void (*set_cpus_allowed
)(struct task_struct
*p
,
1038 const struct cpumask
*newmask
);
1040 void (*rq_online
)(struct rq
*rq
);
1041 void (*rq_offline
)(struct rq
*rq
);
1044 void (*set_curr_task
) (struct rq
*rq
);
1045 void (*task_tick
) (struct rq
*rq
, struct task_struct
*p
, int queued
);
1046 void (*task_fork
) (struct task_struct
*p
);
1048 void (*switched_from
) (struct rq
*this_rq
, struct task_struct
*task
);
1049 void (*switched_to
) (struct rq
*this_rq
, struct task_struct
*task
);
1050 void (*prio_changed
) (struct rq
*this_rq
, struct task_struct
*task
,
1053 unsigned int (*get_rr_interval
) (struct rq
*rq
,
1054 struct task_struct
*task
);
1056 #ifdef CONFIG_FAIR_GROUP_SCHED
1057 void (*task_move_group
) (struct task_struct
*p
, int on_rq
);
1061 #define sched_class_highest (&stop_sched_class)
1062 #define for_each_class(class) \
1063 for (class = sched_class_highest; class; class = class->next)
1065 extern const struct sched_class stop_sched_class
;
1066 extern const struct sched_class rt_sched_class
;
1067 extern const struct sched_class fair_sched_class
;
1068 extern const struct sched_class idle_sched_class
;
1073 extern void update_group_power(struct sched_domain
*sd
, int cpu
);
1075 extern void trigger_load_balance(struct rq
*rq
, int cpu
);
1076 extern void idle_balance(int this_cpu
, struct rq
*this_rq
);
1078 extern void idle_enter_fair(struct rq
*this_rq
);
1079 extern void idle_exit_fair(struct rq
*this_rq
);
1082 # ifdef CONFIG_MTK_SCHED_CMP_TGS
1083 extern int group_leader_is_empty(struct task_struct
*p
);
1084 # endif /* CONFIG_MTK_SCHED_CMP_TGS */
1086 # ifdef CONFIG_MTK_SCHED_CMP
1087 extern void get_cluster_cpus(struct cpumask
*cpus
, int cluster_id
,
1088 bool exclusive_offline
);
1089 extern int get_cluster_id(unsigned int cpu
);
1090 # endif /* CONFIG_MTK_SCHED_CMP */
1092 #else /* CONFIG_SMP */
1094 static inline void idle_balance(int cpu
, struct rq
*rq
)
1100 extern void sysrq_sched_debug_show(void);
1101 extern void sched_init_granularity(void);
1102 #if defined (CONFIG_MTK_SCHED_CMP_PACK_SMALL_TASK) || defined (CONFIG_HMP_PACK_SMALL_TASK)
1103 extern void update_packing_domain(int cpu
);
1104 #endif /* CONFIG_HMP_PACK_SMALL_TASK */
1105 extern void update_max_interval(void);
1106 extern int update_runtime(struct notifier_block
*nfb
, unsigned long action
, void *hcpu
);
1107 extern void init_sched_rt_class(void);
1108 extern void init_sched_fair_class(void);
1110 extern void resched_task(struct task_struct
*p
);
1111 extern void resched_cpu(int cpu
);
1113 extern struct rt_bandwidth def_rt_bandwidth
;
1114 extern void init_rt_bandwidth(struct rt_bandwidth
*rt_b
, u64 period
, u64 runtime
);
1116 extern void update_idle_cpu_load(struct rq
*this_rq
);
1118 extern void init_task_runnable_average(struct task_struct
*p
);
1120 #ifdef CONFIG_PARAVIRT
1121 static inline u64
steal_ticks(u64 steal
)
1123 if (unlikely(steal
> NSEC_PER_SEC
))
1124 return div_u64(steal
, TICK_NSEC
);
1126 return __iter_div_u64_rem(steal
, TICK_NSEC
, &steal
);
1130 static inline void inc_nr_running(struct rq
*rq
)
1132 #ifdef CONFIG_MTK_SCHED_RQAVG_KS
1133 sched_update_nr_prod(cpu_of(rq
), rq
->nr_running
, true);
1134 #endif /* CONFIG_MTK_SCHED_RQAVG_KS */
1137 #ifdef CONFIG_NO_HZ_FULL
1138 if (rq
->nr_running
== 2) {
1139 if (tick_nohz_full_cpu(rq
->cpu
)) {
1140 /* Order rq->nr_running write against the IPI */
1142 smp_send_reschedule(rq
->cpu
);
1148 static inline void dec_nr_running(struct rq
*rq
)
1150 #ifdef CONFIG_MTK_SCHED_RQAVG_KS
1151 sched_update_nr_prod(cpu_of(rq
), rq
->nr_running
, false);
1152 #endif /* CONFIG_MTK_SCHED_RQAVG_KS */
1156 static inline void rq_last_tick_reset(struct rq
*rq
)
1158 #ifdef CONFIG_NO_HZ_FULL
1159 rq
->last_sched_tick
= jiffies
;
1163 extern void update_rq_clock(struct rq
*rq
);
1165 extern void activate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1166 extern void deactivate_task(struct rq
*rq
, struct task_struct
*p
, int flags
);
1168 extern void check_preempt_curr(struct rq
*rq
, struct task_struct
*p
, int flags
);
1170 extern const_debug
unsigned int sysctl_sched_time_avg
;
1171 extern const_debug
unsigned int sysctl_sched_nr_migrate
;
1172 extern const_debug
unsigned int sysctl_sched_migration_cost
;
1174 static inline u64
sched_avg_period(void)
1176 return (u64
)sysctl_sched_time_avg
* NSEC_PER_MSEC
/ 2;
1179 #ifdef CONFIG_SCHED_HRTICK
1183 * - enabled by features
1184 * - hrtimer is actually high res
1186 static inline int hrtick_enabled(struct rq
*rq
)
1188 if (!sched_feat(HRTICK
))
1190 if (!cpu_active(cpu_of(rq
)))
1192 return hrtimer_is_hres_active(&rq
->hrtick_timer
);
1195 void hrtick_start(struct rq
*rq
, u64 delay
);
1199 static inline int hrtick_enabled(struct rq
*rq
)
1204 #endif /* CONFIG_SCHED_HRTICK */
1207 extern void sched_avg_update(struct rq
*rq
);
1208 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
)
1210 rq
->rt_avg
+= rt_delta
;
1211 sched_avg_update(rq
);
1214 static inline void sched_rt_avg_update(struct rq
*rq
, u64 rt_delta
) { }
1215 static inline void sched_avg_update(struct rq
*rq
) { }
1218 extern void start_bandwidth_timer(struct hrtimer
*period_timer
, ktime_t period
);
1221 #ifdef CONFIG_PREEMPT
1223 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
);
1226 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1227 * way at the expense of forcing extra atomic operations in all
1228 * invocations. This assures that the double_lock is acquired using the
1229 * same underlying policy as the spinlock_t on this architecture, which
1230 * reduces latency compared to the unfair variant below. However, it
1231 * also adds more overhead and therefore may reduce throughput.
1233 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1234 __releases(this_rq
->lock
)
1235 __acquires(busiest
->lock
)
1236 __acquires(this_rq
->lock
)
1238 raw_spin_unlock(&this_rq
->lock
);
1239 double_rq_lock(this_rq
, busiest
);
1246 * Unfair double_lock_balance: Optimizes throughput at the expense of
1247 * latency by eliminating extra atomic operations when the locks are
1248 * already in proper order on entry. This favors lower cpu-ids and will
1249 * grant the double lock to lower cpus over higher ids under contention,
1250 * regardless of entry order into the function.
1252 static inline int _double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1253 __releases(this_rq
->lock
)
1254 __acquires(busiest
->lock
)
1255 __acquires(this_rq
->lock
)
1259 if (unlikely(!raw_spin_trylock(&busiest
->lock
))) {
1260 if (busiest
< this_rq
) {
1261 raw_spin_unlock(&this_rq
->lock
);
1262 raw_spin_lock(&busiest
->lock
);
1263 raw_spin_lock_nested(&this_rq
->lock
,
1264 SINGLE_DEPTH_NESTING
);
1267 raw_spin_lock_nested(&busiest
->lock
,
1268 SINGLE_DEPTH_NESTING
);
1273 #endif /* CONFIG_PREEMPT */
1276 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1278 static inline int double_lock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1280 if (unlikely(!irqs_disabled())) {
1281 /* printk() doesn't work good under rq->lock */
1282 raw_spin_unlock(&this_rq
->lock
);
1286 return _double_lock_balance(this_rq
, busiest
);
1289 static inline void double_unlock_balance(struct rq
*this_rq
, struct rq
*busiest
)
1290 __releases(busiest
->lock
)
1292 raw_spin_unlock(&busiest
->lock
);
1293 lock_set_subclass(&this_rq
->lock
.dep_map
, 0, _RET_IP_
);
1297 * double_rq_lock - safely lock two runqueues
1299 * Note this does not disable interrupts like task_rq_lock,
1300 * you need to do so manually before calling.
1302 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1303 __acquires(rq1
->lock
)
1304 __acquires(rq2
->lock
)
1306 BUG_ON(!irqs_disabled());
1308 raw_spin_lock(&rq1
->lock
);
1309 __acquire(rq2
->lock
); /* Fake it out ;) */
1312 raw_spin_lock(&rq1
->lock
);
1313 raw_spin_lock_nested(&rq2
->lock
, SINGLE_DEPTH_NESTING
);
1315 raw_spin_lock(&rq2
->lock
);
1316 raw_spin_lock_nested(&rq1
->lock
, SINGLE_DEPTH_NESTING
);
1322 * double_rq_unlock - safely unlock two runqueues
1324 * Note this does not restore interrupts like task_rq_unlock,
1325 * you need to do so manually after calling.
1327 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1328 __releases(rq1
->lock
)
1329 __releases(rq2
->lock
)
1331 raw_spin_unlock(&rq1
->lock
);
1333 raw_spin_unlock(&rq2
->lock
);
1335 __release(rq2
->lock
);
1338 #else /* CONFIG_SMP */
1341 * double_rq_lock - safely lock two runqueues
1343 * Note this does not disable interrupts like task_rq_lock,
1344 * you need to do so manually before calling.
1346 static inline void double_rq_lock(struct rq
*rq1
, struct rq
*rq2
)
1347 __acquires(rq1
->lock
)
1348 __acquires(rq2
->lock
)
1350 BUG_ON(!irqs_disabled());
1352 raw_spin_lock(&rq1
->lock
);
1353 __acquire(rq2
->lock
); /* Fake it out ;) */
1357 * double_rq_unlock - safely unlock two runqueues
1359 * Note this does not restore interrupts like task_rq_unlock,
1360 * you need to do so manually after calling.
1362 static inline void double_rq_unlock(struct rq
*rq1
, struct rq
*rq2
)
1363 __releases(rq1
->lock
)
1364 __releases(rq2
->lock
)
1367 raw_spin_unlock(&rq1
->lock
);
1368 __release(rq2
->lock
);
1373 extern struct sched_entity
*__pick_first_entity(struct cfs_rq
*cfs_rq
);
1374 extern struct sched_entity
*__pick_last_entity(struct cfs_rq
*cfs_rq
);
1375 extern void print_cfs_stats(struct seq_file
*m
, int cpu
);
1376 extern void print_rt_stats(struct seq_file
*m
, int cpu
);
1378 extern void init_cfs_rq(struct cfs_rq
*cfs_rq
);
1379 extern void init_rt_rq(struct rt_rq
*rt_rq
, struct rq
*rq
);
1380 extern void unthrottle_offline_rt_rqs(struct rq
*rq
);
1382 extern void cfs_bandwidth_usage_inc(void);
1383 extern void cfs_bandwidth_usage_dec(void);
1385 #ifdef CONFIG_NO_HZ_COMMON
1386 enum rq_nohz_flag_bits
{
1391 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1394 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1396 DECLARE_PER_CPU(u64
, cpu_hardirq_time
);
1397 DECLARE_PER_CPU(u64
, cpu_softirq_time
);
1399 #ifndef CONFIG_64BIT
1400 DECLARE_PER_CPU(seqcount_t
, irq_time_seq
);
1402 static inline void irq_time_write_begin(void)
1404 __this_cpu_inc(irq_time_seq
.sequence
);
1408 static inline void irq_time_write_end(void)
1411 __this_cpu_inc(irq_time_seq
.sequence
);
1414 static inline u64
irq_time_read(int cpu
)
1420 seq
= read_seqcount_begin(&per_cpu(irq_time_seq
, cpu
));
1421 irq_time
= per_cpu(cpu_softirq_time
, cpu
) +
1422 per_cpu(cpu_hardirq_time
, cpu
);
1423 } while (read_seqcount_retry(&per_cpu(irq_time_seq
, cpu
), seq
));
1427 #else /* CONFIG_64BIT */
1428 static inline void irq_time_write_begin(void)
1432 static inline void irq_time_write_end(void)
1436 static inline u64
irq_time_read(int cpu
)
1438 return per_cpu(cpu_softirq_time
, cpu
) + per_cpu(cpu_hardirq_time
, cpu
);
1440 #endif /* CONFIG_64BIT */
1441 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1444 static inline int rq_cpu(const struct rq
*rq
) { return rq
->cpu
; }
1446 static inline int rq_cpu(const struct rq
*rq
) { return 0; }