~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/sched/sched.h

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.75 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 
  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>
 12 
 13 #include "cpupri.h"
 14 #include "cpudeadline.h"
 15 #include "cpuacct.h"
 16 
 17 struct rq;
 18 struct cpuidle_state;
 19 
 20 /* task_struct::on_rq states: */
 21 #define TASK_ON_RQ_QUEUED       1
 22 #define TASK_ON_RQ_MIGRATING    2
 23 
 24 extern __read_mostly int scheduler_running;
 25 
 26 extern unsigned long calc_load_update;
 27 extern atomic_long_t calc_load_tasks;
 28 
 29 extern void calc_global_load_tick(struct rq *this_rq);
 30 extern long calc_load_fold_active(struct rq *this_rq);
 31 
 32 #ifdef CONFIG_SMP
 33 extern void update_cpu_load_active(struct rq *this_rq);
 34 #else
 35 static inline void update_cpu_load_active(struct rq *this_rq) { }
 36 #endif
 37 
 38 /*
 39  * Helpers for converting nanosecond timing to jiffy resolution
 40  */
 41 #define NS_TO_JIFFIES(TIME)     ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
 42 
 43 /*
 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.
 49  *
 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
 53  * increased costs.
 54  */
 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)
 59 #else
 60 # define SCHED_LOAD_RESOLUTION  0
 61 # define scale_load(w)          (w)
 62 # define scale_load_down(w)     (w)
 63 #endif
 64 
 65 #define SCHED_LOAD_SHIFT        (10 + SCHED_LOAD_RESOLUTION)
 66 #define SCHED_LOAD_SCALE        (1L << SCHED_LOAD_SHIFT)
 67 
 68 #define NICE_0_LOAD             SCHED_LOAD_SCALE
 69 #define NICE_0_SHIFT            SCHED_LOAD_SHIFT
 70 
 71 /*
 72  * Single value that decides SCHED_DEADLINE internal math precision.
 73  * 10 -> just above 1us
 74  * 9  -> just above 0.5us
 75  */
 76 #define DL_SCALE (10)
 77 
 78 /*
 79  * These are the 'tuning knobs' of the scheduler:
 80  */
 81 
 82 /*
 83  * single value that denotes runtime == period, ie unlimited time.
 84  */
 85 #define RUNTIME_INF     ((u64)~0ULL)
 86 
 87 static inline int fair_policy(int policy)
 88 {
 89         return policy == SCHED_NORMAL || policy == SCHED_BATCH;
 90 }
 91 
 92 static inline int rt_policy(int policy)
 93 {
 94         return policy == SCHED_FIFO || policy == SCHED_RR;
 95 }
 96 
 97 static inline int dl_policy(int policy)
 98 {
 99         return policy == SCHED_DEADLINE;
100 }
101 
102 static inline int task_has_rt_policy(struct task_struct *p)
103 {
104         return rt_policy(p->policy);
105 }
106 
107 static inline int task_has_dl_policy(struct task_struct *p)
108 {
109         return dl_policy(p->policy);
110 }
111 
112 static inline bool dl_time_before(u64 a, u64 b)
113 {
114         return (s64)(a - b) < 0;
115 }
116 
117 /*
118  * Tells if entity @a should preempt entity @b.
119  */
120 static inline bool
121 dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
122 {
123         return dl_time_before(a->deadline, b->deadline);
124 }
125 
126 /*
127  * This is the priority-queue data structure of the RT scheduling class:
128  */
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];
132 };
133 
134 struct rt_bandwidth {
135         /* nests inside the rq lock: */
136         raw_spinlock_t          rt_runtime_lock;
137         ktime_t                 rt_period;
138         u64                     rt_runtime;
139         struct hrtimer          rt_period_timer;
140         unsigned int            rt_period_active;
141 };
142 
143 void __dl_clear_params(struct task_struct *p);
144 
145 /*
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.
150  *
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.
156  *
157  * With respect to SMP, the bandwidth is given on a per-CPU basis,
158  * meaning that:
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
167  * control.
168  */
169 struct dl_bandwidth {
170         raw_spinlock_t dl_runtime_lock;
171         u64 dl_runtime;
172         u64 dl_period;
173 };
174 
175 static inline int dl_bandwidth_enabled(void)
176 {
177         return sysctl_sched_rt_runtime >= 0;
178 }
179 
180 extern struct dl_bw *dl_bw_of(int i);
181 
182 struct dl_bw {
183         raw_spinlock_t lock;
184         u64 bw, total_bw;
185 };
186 
187 static inline
188 void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw)
189 {
190         dl_b->total_bw -= tsk_bw;
191 }
192 
193 static inline
194 void __dl_add(struct dl_bw *dl_b, u64 tsk_bw)
195 {
196         dl_b->total_bw += tsk_bw;
197 }
198 
199 static inline
200 bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw)
201 {
202         return dl_b->bw != -1 &&
203                dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw;
204 }
205 
206 extern struct mutex sched_domains_mutex;
207 
208 #ifdef CONFIG_CGROUP_SCHED
209 
210 #include <linux/cgroup.h>
211 
212 struct cfs_rq;
213 struct rt_rq;
214 
215 extern struct list_head task_groups;
216 
217 struct cfs_bandwidth {
218 #ifdef CONFIG_CFS_BANDWIDTH
219         raw_spinlock_t lock;
220         ktime_t period;
221         u64 quota, runtime;
222         s64 hierarchical_quota;
223         u64 runtime_expires;
224 
225         int idle, period_active;
226         struct hrtimer period_timer, slack_timer;
227         struct list_head throttled_cfs_rq;
228 
229         /* statistics */
230         int nr_periods, nr_throttled;
231         u64 throttled_time;
232 #endif
233 };
234 
235 /* task group related information */
236 struct task_group {
237         struct cgroup_subsys_state css;
238 
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;
245 
246 #ifdef  CONFIG_SMP
247         atomic_long_t load_avg;
248         atomic_t runnable_avg;
249 #endif
250 #endif
251 
252 #ifdef CONFIG_RT_GROUP_SCHED
253         struct sched_rt_entity **rt_se;
254         struct rt_rq **rt_rq;
255 
256         struct rt_bandwidth rt_bandwidth;
257 #endif
258 
259         struct rcu_head rcu;
260         struct list_head list;
261 
262         struct task_group *parent;
263         struct list_head siblings;
264         struct list_head children;
265 
266 #ifdef CONFIG_SCHED_AUTOGROUP
267         struct autogroup *autogroup;
268 #endif
269 
270         struct cfs_bandwidth cfs_bandwidth;
271 };
272 
273 #ifdef CONFIG_FAIR_GROUP_SCHED
274 #define ROOT_TASK_GROUP_LOAD    NICE_0_LOAD
275 
276 /*
277  * A weight of 0 or 1 can cause arithmetics problems.
278  * A weight of a cfs_rq is the sum of weights of which entities
279  * are queued on this cfs_rq, so a weight of a entity should not be
280  * too large, so as the shares value of a task group.
281  * (The default weight is 1024 - so there's no practical
282  *  limitation from this.)
283  */
284 #define MIN_SHARES      (1UL <<  1)
285 #define MAX_SHARES      (1UL << 18)
286 #endif
287 
288 typedef int (*tg_visitor)(struct task_group *, void *);
289 
290 extern int walk_tg_tree_from(struct task_group *from,
291                              tg_visitor down, tg_visitor up, void *data);
292 
293 /*
294  * Iterate the full tree, calling @down when first entering a node and @up when
295  * leaving it for the final time.
296  *
297  * Caller must hold rcu_lock or sufficient equivalent.
298  */
299 static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
300 {
301         return walk_tg_tree_from(&root_task_group, down, up, data);
302 }
303 
304 extern int tg_nop(struct task_group *tg, void *data);
305 
306 extern void free_fair_sched_group(struct task_group *tg);
307 extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent);
308 extern void unregister_fair_sched_group(struct task_group *tg, int cpu);
309 extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
310                         struct sched_entity *se, int cpu,
311                         struct sched_entity *parent);
312 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
313 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
314 
315 extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
316 extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
317 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
318 
319 extern void free_rt_sched_group(struct task_group *tg);
320 extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent);
321 extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
322                 struct sched_rt_entity *rt_se, int cpu,
323                 struct sched_rt_entity *parent);
324 
325 extern struct task_group *sched_create_group(struct task_group *parent);
326 extern void sched_online_group(struct task_group *tg,
327                                struct task_group *parent);
328 extern void sched_destroy_group(struct task_group *tg);
329 extern void sched_offline_group(struct task_group *tg);
330 
331 extern void sched_move_task(struct task_struct *tsk);
332 
333 #ifdef CONFIG_FAIR_GROUP_SCHED
334 extern int sched_group_set_shares(struct task_group *tg, unsigned long shares);
335 #endif
336 
337 #else /* CONFIG_CGROUP_SCHED */
338 
339 struct cfs_bandwidth { };
340 
341 #endif  /* CONFIG_CGROUP_SCHED */
342 
343 /* CFS-related fields in a runqueue */
344 struct cfs_rq {
345         struct load_weight load;
346         unsigned int nr_running, h_nr_running;
347 
348         u64 exec_clock;
349         u64 min_vruntime;
350 #ifndef CONFIG_64BIT
351         u64 min_vruntime_copy;
352 #endif
353 
354         struct rb_root tasks_timeline;
355         struct rb_node *rb_leftmost;
356 
357         /*
358          * 'curr' points to currently running entity on this cfs_rq.
359          * It is set to NULL otherwise (i.e when none are currently running).
360          */
361         struct sched_entity *curr, *next, *last, *skip;
362 
363 #ifdef  CONFIG_SCHED_DEBUG
364         unsigned int nr_spread_over;
365 #endif
366 
367 #ifdef CONFIG_SMP
368         /*
369          * CFS Load tracking
370          * Under CFS, load is tracked on a per-entity basis and aggregated up.
371          * This allows for the description of both thread and group usage (in
372          * the FAIR_GROUP_SCHED case).
373          * runnable_load_avg is the sum of the load_avg_contrib of the
374          * sched_entities on the rq.
375          * blocked_load_avg is similar to runnable_load_avg except that its
376          * the blocked sched_entities on the rq.
377          * utilization_load_avg is the sum of the average running time of the
378          * sched_entities on the rq.
379          */
380         unsigned long runnable_load_avg, blocked_load_avg, utilization_load_avg;
381         atomic64_t decay_counter;
382         u64 last_decay;
383         atomic_long_t removed_load;
384 
385 #ifdef CONFIG_FAIR_GROUP_SCHED
386         /* Required to track per-cpu representation of a task_group */
387         u32 tg_runnable_contrib;
388         unsigned long tg_load_contrib;
389 
390         /*
391          *   h_load = weight * f(tg)
392          *
393          * Where f(tg) is the recursive weight fraction assigned to
394          * this group.
395          */
396         unsigned long h_load;
397         u64 last_h_load_update;
398         struct sched_entity *h_load_next;
399 #endif /* CONFIG_FAIR_GROUP_SCHED */
400 #endif /* CONFIG_SMP */
401 
402 #ifdef CONFIG_FAIR_GROUP_SCHED
403         struct rq *rq;  /* cpu runqueue to which this cfs_rq is attached */
404 
405         /*
406          * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
407          * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
408          * (like users, containers etc.)
409          *
410          * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
411          * list is used during load balance.
412          */
413         int on_list;
414         struct list_head leaf_cfs_rq_list;
415         struct task_group *tg;  /* group that "owns" this runqueue */
416 
417 #ifdef CONFIG_CFS_BANDWIDTH
418         int runtime_enabled;
419         u64 runtime_expires;
420         s64 runtime_remaining;
421 
422         u64 throttled_clock, throttled_clock_task;
423         u64 throttled_clock_task_time;
424         int throttled, throttle_count;
425         struct list_head throttled_list;
426 #endif /* CONFIG_CFS_BANDWIDTH */
427 #endif /* CONFIG_FAIR_GROUP_SCHED */
428 };
429 
430 static inline int rt_bandwidth_enabled(void)
431 {
432         return sysctl_sched_rt_runtime >= 0;
433 }
434 
435 /* RT IPI pull logic requires IRQ_WORK */
436 #ifdef CONFIG_IRQ_WORK
437 # define HAVE_RT_PUSH_IPI
438 #endif
439 
440 /* Real-Time classes' related field in a runqueue: */
441 struct rt_rq {
442         struct rt_prio_array active;
443         unsigned int rt_nr_running;
444 #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
445         struct {
446                 int curr; /* highest queued rt task prio */
447 #ifdef CONFIG_SMP
448                 int next; /* next highest */
449 #endif
450         } highest_prio;
451 #endif
452 #ifdef CONFIG_SMP
453         unsigned long rt_nr_migratory;
454         unsigned long rt_nr_total;
455         int overloaded;
456         struct plist_head pushable_tasks;
457 #ifdef HAVE_RT_PUSH_IPI
458         int push_flags;
459         int push_cpu;
460         struct irq_work push_work;
461         raw_spinlock_t push_lock;
462 #endif
463 #endif /* CONFIG_SMP */
464         int rt_queued;
465 
466         int rt_throttled;
467         u64 rt_time;
468         u64 rt_runtime;
469         /* Nests inside the rq lock: */
470         raw_spinlock_t rt_runtime_lock;
471 
472 #ifdef CONFIG_RT_GROUP_SCHED
473         unsigned long rt_nr_boosted;
474 
475         struct rq *rq;
476         struct task_group *tg;
477 #endif
478 };
479 
480 /* Deadline class' related fields in a runqueue */
481 struct dl_rq {
482         /* runqueue is an rbtree, ordered by deadline */
483         struct rb_root rb_root;
484         struct rb_node *rb_leftmost;
485 
486         unsigned long dl_nr_running;
487 
488 #ifdef CONFIG_SMP
489         /*
490          * Deadline values of the currently executing and the
491          * earliest ready task on this rq. Caching these facilitates
492          * the decision wether or not a ready but not running task
493          * should migrate somewhere else.
494          */
495         struct {
496                 u64 curr;
497                 u64 next;
498         } earliest_dl;
499 
500         unsigned long dl_nr_migratory;
501         int overloaded;
502 
503         /*
504          * Tasks on this rq that can be pushed away. They are kept in
505          * an rb-tree, ordered by tasks' deadlines, with caching
506          * of the leftmost (earliest deadline) element.
507          */
508         struct rb_root pushable_dl_tasks_root;
509         struct rb_node *pushable_dl_tasks_leftmost;
510 #else
511         struct dl_bw dl_bw;
512 #endif
513 };
514 
515 #ifdef CONFIG_SMP
516 
517 /*
518  * We add the notion of a root-domain which will be used to define per-domain
519  * variables. Each exclusive cpuset essentially defines an island domain by
520  * fully partitioning the member cpus from any other cpuset. Whenever a new
521  * exclusive cpuset is created, we also create and attach a new root-domain
522  * object.
523  *
524  */
525 struct root_domain {
526         atomic_t refcount;
527         atomic_t rto_count;
528         struct rcu_head rcu;
529         cpumask_var_t span;
530         cpumask_var_t online;
531 
532         /* Indicate more than one runnable task for any CPU */
533         bool overload;
534 
535         /*
536          * The bit corresponding to a CPU gets set here if such CPU has more
537          * than one runnable -deadline task (as it is below for RT tasks).
538          */
539         cpumask_var_t dlo_mask;
540         atomic_t dlo_count;
541         struct dl_bw dl_bw;
542         struct cpudl cpudl;
543 
544         /*
545          * The "RT overload" flag: it gets set if a CPU has more than
546          * one runnable RT task.
547          */
548         cpumask_var_t rto_mask;
549         struct cpupri cpupri;
550 };
551 
552 extern struct root_domain def_root_domain;
553 
554 #endif /* CONFIG_SMP */
555 
556 /*
557  * This is the main, per-CPU runqueue data structure.
558  *
559  * Locking rule: those places that want to lock multiple runqueues
560  * (such as the load balancing or the thread migration code), lock
561  * acquire operations must be ordered by ascending &runqueue.
562  */
563 struct rq {
564         /* runqueue lock: */
565         raw_spinlock_t lock;
566 
567         /*
568          * nr_running and cpu_load should be in the same cacheline because
569          * remote CPUs use both these fields when doing load calculation.
570          */
571         unsigned int nr_running;
572 #ifdef CONFIG_NUMA_BALANCING
573         unsigned int nr_numa_running;
574         unsigned int nr_preferred_running;
575 #endif
576         #define CPU_LOAD_IDX_MAX 5
577         unsigned long cpu_load[CPU_LOAD_IDX_MAX];
578         unsigned long last_load_update_tick;
579 #ifdef CONFIG_NO_HZ_COMMON
580         u64 nohz_stamp;
581         unsigned long nohz_flags;
582 #endif
583 #ifdef CONFIG_NO_HZ_FULL
584         unsigned long last_sched_tick;
585 #endif
586         /* capture load from *all* tasks on this cpu: */
587         struct load_weight load;
588         unsigned long nr_load_updates;
589         u64 nr_switches;
590 
591         struct cfs_rq cfs;
592         struct rt_rq rt;
593         struct dl_rq dl;
594 
595 #ifdef CONFIG_FAIR_GROUP_SCHED
596         /* list of leaf cfs_rq on this cpu: */
597         struct list_head leaf_cfs_rq_list;
598 
599         struct sched_avg avg;
600 #endif /* CONFIG_FAIR_GROUP_SCHED */
601 
602         /*
603          * This is part of a global counter where only the total sum
604          * over all CPUs matters. A task can increase this counter on
605          * one CPU and if it got migrated afterwards it may decrease
606          * it on another CPU. Always updated under the runqueue lock:
607          */
608         unsigned long nr_uninterruptible;
609 
610         struct task_struct *curr, *idle, *stop;
611         unsigned long next_balance;
612         struct mm_struct *prev_mm;
613 
614         unsigned int clock_skip_update;
615         u64 clock;
616         u64 clock_task;
617 
618         atomic_t nr_iowait;
619 
620 #ifdef CONFIG_SMP
621         struct root_domain *rd;
622         struct sched_domain *sd;
623 
624         unsigned long cpu_capacity;
625         unsigned long cpu_capacity_orig;
626 
627         struct callback_head *balance_callback;
628 
629         unsigned char idle_balance;
630         /* For active balancing */
631         int active_balance;
632         int push_cpu;
633         struct cpu_stop_work active_balance_work;
634         /* cpu of this runqueue: */
635         int cpu;
636         int online;
637 
638         struct list_head cfs_tasks;
639 
640         u64 rt_avg;
641         u64 age_stamp;
642         u64 idle_stamp;
643         u64 avg_idle;
644 
645         /* This is used to determine avg_idle's max value */
646         u64 max_idle_balance_cost;
647 #endif
648 
649 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
650         u64 prev_irq_time;
651 #endif
652 #ifdef CONFIG_PARAVIRT
653         u64 prev_steal_time;
654 #endif
655 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
656         u64 prev_steal_time_rq;
657 #endif
658 
659         /* calc_load related fields */
660         unsigned long calc_load_update;
661         long calc_load_active;
662 
663 #ifdef CONFIG_SCHED_HRTICK
664 #ifdef CONFIG_SMP
665         int hrtick_csd_pending;
666         struct call_single_data hrtick_csd;
667 #endif
668         struct hrtimer hrtick_timer;
669 #endif
670 
671 #ifdef CONFIG_SCHEDSTATS
672         /* latency stats */
673         struct sched_info rq_sched_info;
674         unsigned long long rq_cpu_time;
675         /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
676 
677         /* sys_sched_yield() stats */
678         unsigned int yld_count;
679 
680         /* schedule() stats */
681         unsigned int sched_count;
682         unsigned int sched_goidle;
683 
684         /* try_to_wake_up() stats */
685         unsigned int ttwu_count;
686         unsigned int ttwu_local;
687 #endif
688 
689 #ifdef CONFIG_SMP
690         struct llist_head wake_list;
691 #endif
692 
693 #ifdef CONFIG_CPU_IDLE
694         /* Must be inspected within a rcu lock section */
695         struct cpuidle_state *idle_state;
696 #endif
697 };
698 
699 static inline int cpu_of(struct rq *rq)
700 {
701 #ifdef CONFIG_SMP
702         return rq->cpu;
703 #else
704         return 0;
705 #endif
706 }
707 
708 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
709 
710 #define cpu_rq(cpu)             (&per_cpu(runqueues, (cpu)))
711 #define this_rq()               this_cpu_ptr(&runqueues)
712 #define task_rq(p)              cpu_rq(task_cpu(p))
713 #define cpu_curr(cpu)           (cpu_rq(cpu)->curr)
714 #define raw_rq()                raw_cpu_ptr(&runqueues)
715 
716 static inline u64 __rq_clock_broken(struct rq *rq)
717 {
718         return READ_ONCE(rq->clock);
719 }
720 
721 static inline u64 rq_clock(struct rq *rq)
722 {
723         lockdep_assert_held(&rq->lock);
724         return rq->clock;
725 }
726 
727 static inline u64 rq_clock_task(struct rq *rq)
728 {
729         lockdep_assert_held(&rq->lock);
730         return rq->clock_task;
731 }
732 
733 #define RQCF_REQ_SKIP   0x01
734 #define RQCF_ACT_SKIP   0x02
735 
736 static inline void rq_clock_skip_update(struct rq *rq, bool skip)
737 {
738         lockdep_assert_held(&rq->lock);
739         if (skip)
740                 rq->clock_skip_update |= RQCF_REQ_SKIP;
741         else
742                 rq->clock_skip_update &= ~RQCF_REQ_SKIP;
743 }
744 
745 #ifdef CONFIG_NUMA
746 enum numa_topology_type {
747         NUMA_DIRECT,
748         NUMA_GLUELESS_MESH,
749         NUMA_BACKPLANE,
750 };
751 extern enum numa_topology_type sched_numa_topology_type;
752 extern int sched_max_numa_distance;
753 extern bool find_numa_distance(int distance);
754 #endif
755 
756 #ifdef CONFIG_NUMA_BALANCING
757 /* The regions in numa_faults array from task_struct */
758 enum numa_faults_stats {
759         NUMA_MEM = 0,
760         NUMA_CPU,
761         NUMA_MEMBUF,
762         NUMA_CPUBUF
763 };
764 extern void sched_setnuma(struct task_struct *p, int node);
765 extern int migrate_task_to(struct task_struct *p, int cpu);
766 extern int migrate_swap(struct task_struct *, struct task_struct *);
767 #endif /* CONFIG_NUMA_BALANCING */
768 
769 #ifdef CONFIG_SMP
770 
771 static inline void
772 queue_balance_callback(struct rq *rq,
773                        struct callback_head *head,
774                        void (*func)(struct rq *rq))
775 {
776         lockdep_assert_held(&rq->lock);
777 
778         if (unlikely(head->next))
779                 return;
780 
781         head->func = (void (*)(struct callback_head *))func;
782         head->next = rq->balance_callback;
783         rq->balance_callback = head;
784 }
785 
786 extern void sched_ttwu_pending(void);
787 
788 #define rcu_dereference_check_sched_domain(p) \
789         rcu_dereference_check((p), \
790                               lockdep_is_held(&sched_domains_mutex))
791 
792 /*
793  * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
794  * See detach_destroy_domains: synchronize_sched for details.
795  *
796  * The domain tree of any CPU may only be accessed from within
797  * preempt-disabled sections.
798  */
799 #define for_each_domain(cpu, __sd) \
800         for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \
801                         __sd; __sd = __sd->parent)
802 
803 #define for_each_lower_domain(sd) for (; sd; sd = sd->child)
804 
805 /**
806  * highest_flag_domain - Return highest sched_domain containing flag.
807  * @cpu:        The cpu whose highest level of sched domain is to
808  *              be returned.
809  * @flag:       The flag to check for the highest sched_domain
810  *              for the given cpu.
811  *
812  * Returns the highest sched_domain of a cpu which contains the given flag.
813  */
814 static inline struct sched_domain *highest_flag_domain(int cpu, int flag)
815 {
816         struct sched_domain *sd, *hsd = NULL;
817 
818         for_each_domain(cpu, sd) {
819                 if (!(sd->flags & flag))
820                         break;
821                 hsd = sd;
822         }
823 
824         return hsd;
825 }
826 
827 static inline struct sched_domain *lowest_flag_domain(int cpu, int flag)
828 {
829         struct sched_domain *sd;
830 
831         for_each_domain(cpu, sd) {
832                 if (sd->flags & flag)
833                         break;
834         }
835 
836         return sd;
837 }
838 
839 DECLARE_PER_CPU(struct sched_domain *, sd_llc);
840 DECLARE_PER_CPU(int, sd_llc_size);
841 DECLARE_PER_CPU(int, sd_llc_id);
842 DECLARE_PER_CPU(struct sched_domain *, sd_numa);
843 DECLARE_PER_CPU(struct sched_domain *, sd_busy);
844 DECLARE_PER_CPU(struct sched_domain *, sd_asym);
845 
846 struct sched_group_capacity {
847         atomic_t ref;
848         /*
849          * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity
850          * for a single CPU.
851          */
852         unsigned int capacity;
853         unsigned long next_update;
854         int imbalance; /* XXX unrelated to capacity but shared group state */
855         /*
856          * Number of busy cpus in this group.
857          */
858         atomic_t nr_busy_cpus;
859 
860         unsigned long cpumask[0]; /* iteration mask */
861 };
862 
863 struct sched_group {
864         struct sched_group *next;       /* Must be a circular list */
865         atomic_t ref;
866 
867         unsigned int group_weight;
868         struct sched_group_capacity *sgc;
869 
870         /*
871          * The CPUs this group covers.
872          *
873          * NOTE: this field is variable length. (Allocated dynamically
874          * by attaching extra space to the end of the structure,
875          * depending on how many CPUs the kernel has booted up with)
876          */
877         unsigned long cpumask[0];
878 };
879 
880 static inline struct cpumask *sched_group_cpus(struct sched_group *sg)
881 {
882         return to_cpumask(sg->cpumask);
883 }
884 
885 /*
886  * cpumask masking which cpus in the group are allowed to iterate up the domain
887  * tree.
888  */
889 static inline struct cpumask *sched_group_mask(struct sched_group *sg)
890 {
891         return to_cpumask(sg->sgc->cpumask);
892 }
893 
894 /**
895  * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
896  * @group: The group whose first cpu is to be returned.
897  */
898 static inline unsigned int group_first_cpu(struct sched_group *group)
899 {
900         return cpumask_first(sched_group_cpus(group));
901 }
902 
903 extern int group_balance_cpu(struct sched_group *sg);
904 
905 #else
906 
907 static inline void sched_ttwu_pending(void) { }
908 
909 #endif /* CONFIG_SMP */
910 
911 #include "stats.h"
912 #include "auto_group.h"
913 
914 #ifdef CONFIG_CGROUP_SCHED
915 
916 /*
917  * Return the group to which this tasks belongs.
918  *
919  * We cannot use task_css() and friends because the cgroup subsystem
920  * changes that value before the cgroup_subsys::attach() method is called,
921  * therefore we cannot pin it and might observe the wrong value.
922  *
923  * The same is true for autogroup's p->signal->autogroup->tg, the autogroup
924  * core changes this before calling sched_move_task().
925  *
926  * Instead we use a 'copy' which is updated from sched_move_task() while
927  * holding both task_struct::pi_lock and rq::lock.
928  */
929 static inline struct task_group *task_group(struct task_struct *p)
930 {
931         return p->sched_task_group;
932 }
933 
934 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
935 static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
936 {
937 #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED)
938         struct task_group *tg = task_group(p);
939 #endif
940 
941 #ifdef CONFIG_FAIR_GROUP_SCHED
942         p->se.cfs_rq = tg->cfs_rq[cpu];
943         p->se.parent = tg->se[cpu];
944 #endif
945 
946 #ifdef CONFIG_RT_GROUP_SCHED
947         p->rt.rt_rq  = tg->rt_rq[cpu];
948         p->rt.parent = tg->rt_se[cpu];
949 #endif
950 }
951 
952 #else /* CONFIG_CGROUP_SCHED */
953 
954 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
955 static inline struct task_group *task_group(struct task_struct *p)
956 {
957         return NULL;
958 }
959 
960 #endif /* CONFIG_CGROUP_SCHED */
961 
962 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
963 {
964         set_task_rq(p, cpu);
965 #ifdef CONFIG_SMP
966         /*
967          * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
968          * successfuly executed on another CPU. We must ensure that updates of
969          * per-task data have been completed by this moment.
970          */
971         smp_wmb();
972         task_thread_info(p)->cpu = cpu;
973         p->wake_cpu = cpu;
974 #endif
975 }
976 
977 /*
978  * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
979  */
980 #ifdef CONFIG_SCHED_DEBUG
981 # include <linux/static_key.h>
982 # define const_debug __read_mostly
983 #else
984 # define const_debug const
985 #endif
986 
987 extern const_debug unsigned int sysctl_sched_features;
988 
989 #define SCHED_FEAT(name, enabled)       \
990         __SCHED_FEAT_##name ,
991 
992 enum {
993 #include "features.h"
994         __SCHED_FEAT_NR,
995 };
996 
997 #undef SCHED_FEAT
998 
999 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
1000 #define SCHED_FEAT(name, enabled)                                       \
1001 static __always_inline bool static_branch_##name(struct static_key *key) \
1002 {                                                                       \
1003         return static_key_##enabled(key);                               \
1004 }
1005 
1006 #include "features.h"
1007 
1008 #undef SCHED_FEAT
1009 
1010 extern struct static_key sched_feat_keys[__SCHED_FEAT_NR];
1011 #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x]))
1012 #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */
1013 #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
1014 #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */
1015 
1016 #ifdef CONFIG_NUMA_BALANCING
1017 #define sched_feat_numa(x) sched_feat(x)
1018 #ifdef CONFIG_SCHED_DEBUG
1019 #define numabalancing_enabled sched_feat_numa(NUMA)
1020 #else
1021 extern bool numabalancing_enabled;
1022 #endif /* CONFIG_SCHED_DEBUG */
1023 #else
1024 #define sched_feat_numa(x) (0)
1025 #define numabalancing_enabled (0)
1026 #endif /* CONFIG_NUMA_BALANCING */
1027 
1028 static inline u64 global_rt_period(void)
1029 {
1030         return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
1031 }
1032 
1033 static inline u64 global_rt_runtime(void)
1034 {
1035         if (sysctl_sched_rt_runtime < 0)
1036                 return RUNTIME_INF;
1037 
1038         return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
1039 }
1040 
1041 static inline int task_current(struct rq *rq, struct task_struct *p)
1042 {
1043         return rq->curr == p;
1044 }
1045 
1046 static inline int task_running(struct rq *rq, struct task_struct *p)
1047 {
1048 #ifdef CONFIG_SMP
1049         return p->on_cpu;
1050 #else
1051         return task_current(rq, p);
1052 #endif
1053 }
1054 
1055 static inline int task_on_rq_queued(struct task_struct *p)
1056 {
1057         return p->on_rq == TASK_ON_RQ_QUEUED;
1058 }
1059 
1060 static inline int task_on_rq_migrating(struct task_struct *p)
1061 {
1062         return p->on_rq == TASK_ON_RQ_MIGRATING;
1063 }
1064 
1065 #ifndef prepare_arch_switch
1066 # define prepare_arch_switch(next)      do { } while (0)
1067 #endif
1068 #ifndef finish_arch_switch
1069 # define finish_arch_switch(prev)       do { } while (0)
1070 #endif
1071 #ifndef finish_arch_post_lock_switch
1072 # define finish_arch_post_lock_switch() do { } while (0)
1073 #endif
1074 
1075 static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
1076 {
1077 #ifdef CONFIG_SMP
1078         /*
1079          * We can optimise this out completely for !SMP, because the
1080          * SMP rebalancing from interrupt is the only thing that cares
1081          * here.
1082          */
1083         next->on_cpu = 1;
1084 #endif
1085 }
1086 
1087 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
1088 {
1089 #ifdef CONFIG_SMP
1090         /*
1091          * After ->on_cpu is cleared, the task can be moved to a different CPU.
1092          * We must ensure this doesn't happen until the switch is completely
1093          * finished.
1094          *
1095          * Pairs with the control dependency and rmb in try_to_wake_up().
1096          */
1097         smp_store_release(&prev->on_cpu, 0);
1098 #endif
1099 #ifdef CONFIG_DEBUG_SPINLOCK
1100         /* this is a valid case when another task releases the spinlock */
1101         rq->lock.owner = current;
1102 #endif
1103         /*
1104          * If we are tracking spinlock dependencies then we have to
1105          * fix up the runqueue lock - which gets 'carried over' from
1106          * prev into current:
1107          */
1108         spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
1109 
1110         raw_spin_unlock_irq(&rq->lock);
1111 }
1112 
1113 /*
1114  * wake flags
1115  */
1116 #define WF_SYNC         0x01            /* waker goes to sleep after wakeup */
1117 #define WF_FORK         0x02            /* child wakeup after fork */
1118 #define WF_MIGRATED     0x4             /* internal use, task got migrated */
1119 
1120 /*
1121  * To aid in avoiding the subversion of "niceness" due to uneven distribution
1122  * of tasks with abnormal "nice" values across CPUs the contribution that
1123  * each task makes to its run queue's load is weighted according to its
1124  * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
1125  * scaled version of the new time slice allocation that they receive on time
1126  * slice expiry etc.
1127  */
1128 
1129 #define WEIGHT_IDLEPRIO                3
1130 #define WMULT_IDLEPRIO         1431655765
1131 
1132 /*
1133  * Nice levels are multiplicative, with a gentle 10% change for every
1134  * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1135  * nice 1, it will get ~10% less CPU time than another CPU-bound task
1136  * that remained on nice 0.
1137  *
1138  * The "10% effect" is relative and cumulative: from _any_ nice level,
1139  * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
1140  * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1141  * If a task goes up by ~10% and another task goes down by ~10% then
1142  * the relative distance between them is ~25%.)
1143  */
1144 static const int prio_to_weight[40] = {
1145  /* -20 */     88761,     71755,     56483,     46273,     36291,
1146  /* -15 */     29154,     23254,     18705,     14949,     11916,
1147  /* -10 */      9548,      7620,      6100,      4904,      3906,
1148  /*  -5 */      3121,      2501,      1991,      1586,      1277,
1149  /*   0 */      1024,       820,       655,       526,       423,
1150  /*   5 */       335,       272,       215,       172,       137,
1151  /*  10 */       110,        87,        70,        56,        45,
1152  /*  15 */        36,        29,        23,        18,        15,
1153 };
1154 
1155 /*
1156  * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1157  *
1158  * In cases where the weight does not change often, we can use the
1159  * precalculated inverse to speed up arithmetics by turning divisions
1160  * into multiplications:
1161  */
1162 static const u32 prio_to_wmult[40] = {
1163  /* -20 */     48388,     59856,     76040,     92818,    118348,
1164  /* -15 */    147320,    184698,    229616,    287308,    360437,
1165  /* -10 */    449829,    563644,    704093,    875809,   1099582,
1166  /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
1167  /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
1168  /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
1169  /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
1170  /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
1171 };
1172 
1173 #define ENQUEUE_WAKEUP          1
1174 #define ENQUEUE_HEAD            2
1175 #ifdef CONFIG_SMP
1176 #define ENQUEUE_WAKING          4       /* sched_class::task_waking was called */
1177 #else
1178 #define ENQUEUE_WAKING          0
1179 #endif
1180 #define ENQUEUE_REPLENISH       8
1181 
1182 #define DEQUEUE_SLEEP           1
1183 
1184 #define RETRY_TASK              ((void *)-1UL)
1185 
1186 struct sched_class {
1187         const struct sched_class *next;
1188 
1189         void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags);
1190         void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags);
1191         void (*yield_task) (struct rq *rq);
1192         bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt);
1193 
1194         void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
1195 
1196         /*
1197          * It is the responsibility of the pick_next_task() method that will
1198          * return the next task to call put_prev_task() on the @prev task or
1199          * something equivalent.
1200          *
1201          * May return RETRY_TASK when it finds a higher prio class has runnable
1202          * tasks.
1203          */
1204         struct task_struct * (*pick_next_task) (struct rq *rq,
1205                                                 struct task_struct *prev);
1206         void (*put_prev_task) (struct rq *rq, struct task_struct *p);
1207 
1208 #ifdef CONFIG_SMP
1209         int  (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags);
1210         void (*migrate_task_rq)(struct task_struct *p, int next_cpu);
1211 
1212         void (*task_waking) (struct task_struct *task);
1213         void (*task_woken) (struct rq *this_rq, struct task_struct *task);
1214 
1215         void (*set_cpus_allowed)(struct task_struct *p,
1216                                  const struct cpumask *newmask);
1217 
1218         void (*rq_online)(struct rq *rq);
1219         void (*rq_offline)(struct rq *rq);
1220 #endif
1221 
1222         void (*set_curr_task) (struct rq *rq);
1223         void (*task_tick) (struct rq *rq, struct task_struct *p, int queued);
1224         void (*task_fork) (struct task_struct *p);
1225         void (*task_dead) (struct task_struct *p);
1226 
1227         /*
1228          * The switched_from() call is allowed to drop rq->lock, therefore we
1229          * cannot assume the switched_from/switched_to pair is serliazed by
1230          * rq->lock. They are however serialized by p->pi_lock.
1231          */
1232         void (*switched_from) (struct rq *this_rq, struct task_struct *task);
1233         void (*switched_to) (struct rq *this_rq, struct task_struct *task);
1234         void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
1235                              int oldprio);
1236 
1237         unsigned int (*get_rr_interval) (struct rq *rq,
1238                                          struct task_struct *task);
1239 
1240         void (*update_curr) (struct rq *rq);
1241 
1242 #ifdef CONFIG_FAIR_GROUP_SCHED
1243         void (*task_move_group) (struct task_struct *p, int on_rq);
1244 #endif
1245 };
1246 
1247 static inline void put_prev_task(struct rq *rq, struct task_struct *prev)
1248 {
1249         prev->sched_class->put_prev_task(rq, prev);
1250 }
1251 
1252 #define sched_class_highest (&stop_sched_class)
1253 #define for_each_class(class) \
1254    for (class = sched_class_highest; class; class = class->next)
1255 
1256 extern const struct sched_class stop_sched_class;
1257 extern const struct sched_class dl_sched_class;
1258 extern const struct sched_class rt_sched_class;
1259 extern const struct sched_class fair_sched_class;
1260 extern const struct sched_class idle_sched_class;
1261 
1262 
1263 #ifdef CONFIG_SMP
1264 
1265 extern void update_group_capacity(struct sched_domain *sd, int cpu);
1266 
1267 extern void trigger_load_balance(struct rq *rq);
1268 
1269 extern void idle_enter_fair(struct rq *this_rq);
1270 extern void idle_exit_fair(struct rq *this_rq);
1271 
1272 #else
1273 
1274 static inline void idle_enter_fair(struct rq *rq) { }
1275 static inline void idle_exit_fair(struct rq *rq) { }
1276 
1277 #endif
1278 
1279 #ifdef CONFIG_CPU_IDLE
1280 static inline void idle_set_state(struct rq *rq,
1281                                   struct cpuidle_state *idle_state)
1282 {
1283         rq->idle_state = idle_state;
1284 }
1285 
1286 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1287 {
1288         WARN_ON(!rcu_read_lock_held());
1289         return rq->idle_state;
1290 }
1291 #else
1292 static inline void idle_set_state(struct rq *rq,
1293                                   struct cpuidle_state *idle_state)
1294 {
1295 }
1296 
1297 static inline struct cpuidle_state *idle_get_state(struct rq *rq)
1298 {
1299         return NULL;
1300 }
1301 #endif
1302 
1303 extern void sysrq_sched_debug_show(void);
1304 extern void sched_init_granularity(void);
1305 extern void update_max_interval(void);
1306 
1307 extern void init_sched_dl_class(void);
1308 extern void init_sched_rt_class(void);
1309 extern void init_sched_fair_class(void);
1310 
1311 extern void resched_curr(struct rq *rq);
1312 extern void resched_cpu(int cpu);
1313 
1314 extern struct rt_bandwidth def_rt_bandwidth;
1315 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
1316 
1317 extern struct dl_bandwidth def_dl_bandwidth;
1318 extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
1319 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
1320 
1321 unsigned long to_ratio(u64 period, u64 runtime);
1322 
1323 extern void init_task_runnable_average(struct task_struct *p);
1324 
1325 static inline void add_nr_running(struct rq *rq, unsigned count)
1326 {
1327         unsigned prev_nr = rq->nr_running;
1328 
1329         rq->nr_running = prev_nr + count;
1330 
1331         if (prev_nr < 2 && rq->nr_running >= 2) {
1332 #ifdef CONFIG_SMP
1333                 if (!rq->rd->overload)
1334                         rq->rd->overload = true;
1335 #endif
1336 
1337 #ifdef CONFIG_NO_HZ_FULL
1338                 if (tick_nohz_full_cpu(rq->cpu)) {
1339                         /*
1340                          * Tick is needed if more than one task runs on a CPU.
1341                          * Send the target an IPI to kick it out of nohz mode.
1342                          *
1343                          * We assume that IPI implies full memory barrier and the
1344                          * new value of rq->nr_running is visible on reception
1345                          * from the target.
1346                          */
1347                         tick_nohz_full_kick_cpu(rq->cpu);
1348                 }
1349 #endif
1350         }
1351 }
1352 
1353 static inline void sub_nr_running(struct rq *rq, unsigned count)
1354 {
1355         rq->nr_running -= count;
1356 }
1357 
1358 static inline void rq_last_tick_reset(struct rq *rq)
1359 {
1360 #ifdef CONFIG_NO_HZ_FULL
1361         rq->last_sched_tick = jiffies;
1362 #endif
1363 }
1364 
1365 extern void update_rq_clock(struct rq *rq);
1366 
1367 extern void activate_task(struct rq *rq, struct task_struct *p, int flags);
1368 extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags);
1369 
1370 extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
1371 
1372 extern const_debug unsigned int sysctl_sched_time_avg;
1373 extern const_debug unsigned int sysctl_sched_nr_migrate;
1374 extern const_debug unsigned int sysctl_sched_migration_cost;
1375 
1376 static inline u64 sched_avg_period(void)
1377 {
1378         return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1379 }
1380 
1381 #ifdef CONFIG_SCHED_HRTICK
1382 
1383 /*
1384  * Use hrtick when:
1385  *  - enabled by features
1386  *  - hrtimer is actually high res
1387  */
1388 static inline int hrtick_enabled(struct rq *rq)
1389 {
1390         if (!sched_feat(HRTICK))
1391                 return 0;
1392         if (!cpu_active(cpu_of(rq)))
1393                 return 0;
1394         return hrtimer_is_hres_active(&rq->hrtick_timer);
1395 }
1396 
1397 void hrtick_start(struct rq *rq, u64 delay);
1398 
1399 #else
1400 
1401 static inline int hrtick_enabled(struct rq *rq)
1402 {
1403         return 0;
1404 }
1405 
1406 #endif /* CONFIG_SCHED_HRTICK */
1407 
1408 #ifdef CONFIG_SMP
1409 extern void sched_avg_update(struct rq *rq);
1410 
1411 #ifndef arch_scale_freq_capacity
1412 static __always_inline
1413 unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu)
1414 {
1415         return SCHED_CAPACITY_SCALE;
1416 }
1417 #endif
1418 
1419 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1420 {
1421         rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
1422         sched_avg_update(rq);
1423 }
1424 #else
1425 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
1426 static inline void sched_avg_update(struct rq *rq) { }
1427 #endif
1428 
1429 /*
1430  * __task_rq_lock - lock the rq @p resides on.
1431  */
1432 static inline struct rq *__task_rq_lock(struct task_struct *p)
1433         __acquires(rq->lock)
1434 {
1435         struct rq *rq;
1436 
1437         lockdep_assert_held(&p->pi_lock);
1438 
1439         for (;;) {
1440                 rq = task_rq(p);
1441                 raw_spin_lock(&rq->lock);
1442                 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1443                         lockdep_pin_lock(&rq->lock);
1444                         return rq;
1445                 }
1446                 raw_spin_unlock(&rq->lock);
1447 
1448                 while (unlikely(task_on_rq_migrating(p)))
1449                         cpu_relax();
1450         }
1451 }
1452 
1453 /*
1454  * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
1455  */
1456 static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1457         __acquires(p->pi_lock)
1458         __acquires(rq->lock)
1459 {
1460         struct rq *rq;
1461 
1462         for (;;) {
1463                 raw_spin_lock_irqsave(&p->pi_lock, *flags);
1464                 rq = task_rq(p);
1465                 raw_spin_lock(&rq->lock);
1466                 /*
1467                  *      move_queued_task()              task_rq_lock()
1468                  *
1469                  *      ACQUIRE (rq->lock)
1470                  *      [S] ->on_rq = MIGRATING         [L] rq = task_rq()
1471                  *      WMB (__set_task_cpu())          ACQUIRE (rq->lock);
1472                  *      [S] ->cpu = new_cpu             [L] task_rq()
1473                  *                                      [L] ->on_rq
1474                  *      RELEASE (rq->lock)
1475                  *
1476                  * If we observe the old cpu in task_rq_lock, the acquire of
1477                  * the old rq->lock will fully serialize against the stores.
1478                  *
1479                  * If we observe the new cpu in task_rq_lock, the acquire will
1480                  * pair with the WMB to ensure we must then also see migrating.
1481                  */
1482                 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
1483                         lockdep_pin_lock(&rq->lock);
1484                         return rq;
1485                 }
1486                 raw_spin_unlock(&rq->lock);
1487                 raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1488 
1489                 while (unlikely(task_on_rq_migrating(p)))
1490                         cpu_relax();
1491         }
1492 }
1493 
1494 static inline void __task_rq_unlock(struct rq *rq)
1495         __releases(rq->lock)
1496 {
1497         lockdep_unpin_lock(&rq->lock);
1498         raw_spin_unlock(&rq->lock);
1499 }
1500 
1501 static inline void
1502 task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
1503         __releases(rq->lock)
1504         __releases(p->pi_lock)
1505 {
1506         lockdep_unpin_lock(&rq->lock);
1507         raw_spin_unlock(&rq->lock);
1508         raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
1509 }
1510 
1511 #ifdef CONFIG_SMP
1512 #ifdef CONFIG_PREEMPT
1513 
1514 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2);
1515 
1516 /*
1517  * fair double_lock_balance: Safely acquires both rq->locks in a fair
1518  * way at the expense of forcing extra atomic operations in all
1519  * invocations.  This assures that the double_lock is acquired using the
1520  * same underlying policy as the spinlock_t on this architecture, which
1521  * reduces latency compared to the unfair variant below.  However, it
1522  * also adds more overhead and therefore may reduce throughput.
1523  */
1524 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1525         __releases(this_rq->lock)
1526         __acquires(busiest->lock)
1527         __acquires(this_rq->lock)
1528 {
1529         raw_spin_unlock(&this_rq->lock);
1530         double_rq_lock(this_rq, busiest);
1531 
1532         return 1;
1533 }
1534 
1535 #else
1536 /*
1537  * Unfair double_lock_balance: Optimizes throughput at the expense of
1538  * latency by eliminating extra atomic operations when the locks are
1539  * already in proper order on entry.  This favors lower cpu-ids and will
1540  * grant the double lock to lower cpus over higher ids under contention,
1541  * regardless of entry order into the function.
1542  */
1543 static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1544         __releases(this_rq->lock)
1545         __acquires(busiest->lock)
1546         __acquires(this_rq->lock)
1547 {
1548         int ret = 0;
1549 
1550         if (unlikely(!raw_spin_trylock(&busiest->lock))) {
1551                 if (busiest < this_rq) {
1552                         raw_spin_unlock(&this_rq->lock);
1553                         raw_spin_lock(&busiest->lock);
1554                         raw_spin_lock_nested(&this_rq->lock,
1555                                               SINGLE_DEPTH_NESTING);
1556                         ret = 1;
1557                 } else
1558                         raw_spin_lock_nested(&busiest->lock,
1559                                               SINGLE_DEPTH_NESTING);
1560         }
1561         return ret;
1562 }
1563 
1564 #endif /* CONFIG_PREEMPT */
1565 
1566 /*
1567  * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1568  */
1569 static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1570 {
1571         if (unlikely(!irqs_disabled())) {
1572                 /* printk() doesn't work good under rq->lock */
1573                 raw_spin_unlock(&this_rq->lock);
1574                 BUG_ON(1);
1575         }
1576 
1577         return _double_lock_balance(this_rq, busiest);
1578 }
1579 
1580 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1581         __releases(busiest->lock)
1582 {
1583         raw_spin_unlock(&busiest->lock);
1584         lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1585 }
1586 
1587 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
1588 {
1589         if (l1 > l2)
1590                 swap(l1, l2);
1591 
1592         spin_lock(l1);
1593         spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1594 }
1595 
1596 static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2)
1597 {
1598         if (l1 > l2)
1599                 swap(l1, l2);
1600 
1601         spin_lock_irq(l1);
1602         spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1603 }
1604 
1605 static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2)
1606 {
1607         if (l1 > l2)
1608                 swap(l1, l2);
1609 
1610         raw_spin_lock(l1);
1611         raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
1612 }
1613 
1614 /*
1615  * double_rq_lock - safely lock two runqueues
1616  *
1617  * Note this does not disable interrupts like task_rq_lock,
1618  * you need to do so manually before calling.
1619  */
1620 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1621         __acquires(rq1->lock)
1622         __acquires(rq2->lock)
1623 {
1624         BUG_ON(!irqs_disabled());
1625         if (rq1 == rq2) {
1626                 raw_spin_lock(&rq1->lock);
1627                 __acquire(rq2->lock);   /* Fake it out ;) */
1628         } else {
1629                 if (rq1 < rq2) {
1630                         raw_spin_lock(&rq1->lock);
1631                         raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1632                 } else {
1633                         raw_spin_lock(&rq2->lock);
1634                         raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1635                 }
1636         }
1637 }
1638 
1639 /*
1640  * double_rq_unlock - safely unlock two runqueues
1641  *
1642  * Note this does not restore interrupts like task_rq_unlock,
1643  * you need to do so manually after calling.
1644  */
1645 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1646         __releases(rq1->lock)
1647         __releases(rq2->lock)
1648 {
1649         raw_spin_unlock(&rq1->lock);
1650         if (rq1 != rq2)
1651                 raw_spin_unlock(&rq2->lock);
1652         else
1653                 __release(rq2->lock);
1654 }
1655 
1656 #else /* CONFIG_SMP */
1657 
1658 /*
1659  * double_rq_lock - safely lock two runqueues
1660  *
1661  * Note this does not disable interrupts like task_rq_lock,
1662  * you need to do so manually before calling.
1663  */
1664 static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
1665         __acquires(rq1->lock)
1666         __acquires(rq2->lock)
1667 {
1668         BUG_ON(!irqs_disabled());
1669         BUG_ON(rq1 != rq2);
1670         raw_spin_lock(&rq1->lock);
1671         __acquire(rq2->lock);   /* Fake it out ;) */
1672 }
1673 
1674 /*
1675  * double_rq_unlock - safely unlock two runqueues
1676  *
1677  * Note this does not restore interrupts like task_rq_unlock,
1678  * you need to do so manually after calling.
1679  */
1680 static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1681         __releases(rq1->lock)
1682         __releases(rq2->lock)
1683 {
1684         BUG_ON(rq1 != rq2);
1685         raw_spin_unlock(&rq1->lock);
1686         __release(rq2->lock);
1687 }
1688 
1689 #endif
1690 
1691 extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq);
1692 extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq);
1693 
1694 #ifdef  CONFIG_SCHED_DEBUG
1695 extern void print_cfs_stats(struct seq_file *m, int cpu);
1696 extern void print_rt_stats(struct seq_file *m, int cpu);
1697 extern void print_dl_stats(struct seq_file *m, int cpu);
1698 extern void
1699 print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq);
1700 
1701 #ifdef CONFIG_NUMA_BALANCING
1702 extern void
1703 show_numa_stats(struct task_struct *p, struct seq_file *m);
1704 extern void
1705 print_numa_stats(struct seq_file *m, int node, unsigned long tsf,
1706         unsigned long tpf, unsigned long gsf, unsigned long gpf);
1707 #endif /* CONFIG_NUMA_BALANCING */
1708 #endif /* CONFIG_SCHED_DEBUG */
1709 
1710 extern void init_cfs_rq(struct cfs_rq *cfs_rq);
1711 extern void init_rt_rq(struct rt_rq *rt_rq);
1712 extern void init_dl_rq(struct dl_rq *dl_rq);
1713 
1714 extern void cfs_bandwidth_usage_inc(void);
1715 extern void cfs_bandwidth_usage_dec(void);
1716 
1717 #ifdef CONFIG_NO_HZ_COMMON
1718 enum rq_nohz_flag_bits {
1719         NOHZ_TICK_STOPPED,
1720         NOHZ_BALANCE_KICK,
1721 };
1722 
1723 #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1724 #endif
1725 
1726 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
1727 
1728 DECLARE_PER_CPU(u64, cpu_hardirq_time);
1729 DECLARE_PER_CPU(u64, cpu_softirq_time);
1730 
1731 #ifndef CONFIG_64BIT
1732 DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1733 
1734 static inline void irq_time_write_begin(void)
1735 {
1736         __this_cpu_inc(irq_time_seq.sequence);
1737         smp_wmb();
1738 }
1739 
1740 static inline void irq_time_write_end(void)
1741 {
1742         smp_wmb();
1743         __this_cpu_inc(irq_time_seq.sequence);
1744 }
1745 
1746 static inline u64 irq_time_read(int cpu)
1747 {
1748         u64 irq_time;
1749         unsigned seq;
1750 
1751         do {
1752                 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1753                 irq_time = per_cpu(cpu_softirq_time, cpu) +
1754                            per_cpu(cpu_hardirq_time, cpu);
1755         } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1756 
1757         return irq_time;
1758 }
1759 #else /* CONFIG_64BIT */
1760 static inline void irq_time_write_begin(void)
1761 {
1762 }
1763 
1764 static inline void irq_time_write_end(void)
1765 {
1766 }
1767 
1768 static inline u64 irq_time_read(int cpu)
1769 {
1770         return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1771 }
1772 #endif /* CONFIG_64BIT */
1773 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1774 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp