Commit | Line | Data |
---|---|---|
029632fb PZ |
1 | |
2 | #include <linux/sched.h> | |
3 | #include <linux/mutex.h> | |
4 | #include <linux/spinlock.h> | |
5 | #include <linux/stop_machine.h> | |
6 | ||
391e43da | 7 | #include "cpupri.h" |
029632fb PZ |
8 | |
9 | extern __read_mostly int scheduler_running; | |
10 | ||
11 | /* | |
12 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
13 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
14 | * and back. | |
15 | */ | |
16 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
17 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
18 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
19 | ||
20 | /* | |
21 | * 'User priority' is the nice value converted to something we | |
22 | * can work with better when scaling various scheduler parameters, | |
23 | * it's a [ 0 ... 39 ] range. | |
24 | */ | |
25 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
26 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
27 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
28 | ||
29 | /* | |
30 | * Helpers for converting nanosecond timing to jiffy resolution | |
31 | */ | |
32 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
33 | ||
34 | #define NICE_0_LOAD SCHED_LOAD_SCALE | |
35 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
36 | ||
37 | /* | |
38 | * These are the 'tuning knobs' of the scheduler: | |
39 | * | |
40 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). | |
41 | * Timeslices get refilled after they expire. | |
42 | */ | |
43 | #define DEF_TIMESLICE (100 * HZ / 1000) | |
44 | ||
45 | /* | |
46 | * single value that denotes runtime == period, ie unlimited time. | |
47 | */ | |
48 | #define RUNTIME_INF ((u64)~0ULL) | |
49 | ||
50 | static inline int rt_policy(int policy) | |
51 | { | |
52 | if (policy == SCHED_FIFO || policy == SCHED_RR) | |
53 | return 1; | |
54 | return 0; | |
55 | } | |
56 | ||
57 | static inline int task_has_rt_policy(struct task_struct *p) | |
58 | { | |
59 | return rt_policy(p->policy); | |
60 | } | |
61 | ||
62 | /* | |
63 | * This is the priority-queue data structure of the RT scheduling class: | |
64 | */ | |
65 | struct rt_prio_array { | |
66 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
67 | struct list_head queue[MAX_RT_PRIO]; | |
68 | }; | |
69 | ||
70 | struct rt_bandwidth { | |
71 | /* nests inside the rq lock: */ | |
72 | raw_spinlock_t rt_runtime_lock; | |
73 | ktime_t rt_period; | |
74 | u64 rt_runtime; | |
75 | struct hrtimer rt_period_timer; | |
76 | }; | |
77 | ||
78 | extern struct mutex sched_domains_mutex; | |
79 | ||
80 | #ifdef CONFIG_CGROUP_SCHED | |
81 | ||
82 | #include <linux/cgroup.h> | |
83 | ||
84 | struct cfs_rq; | |
85 | struct rt_rq; | |
86 | ||
87 | static LIST_HEAD(task_groups); | |
88 | ||
89 | struct cfs_bandwidth { | |
90 | #ifdef CONFIG_CFS_BANDWIDTH | |
91 | raw_spinlock_t lock; | |
92 | ktime_t period; | |
93 | u64 quota, runtime; | |
94 | s64 hierarchal_quota; | |
95 | u64 runtime_expires; | |
96 | ||
97 | int idle, timer_active; | |
98 | struct hrtimer period_timer, slack_timer; | |
99 | struct list_head throttled_cfs_rq; | |
100 | ||
101 | /* statistics */ | |
102 | int nr_periods, nr_throttled; | |
103 | u64 throttled_time; | |
104 | #endif | |
105 | }; | |
106 | ||
107 | /* task group related information */ | |
108 | struct task_group { | |
109 | struct cgroup_subsys_state css; | |
110 | ||
111 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
112 | /* schedulable entities of this group on each cpu */ | |
113 | struct sched_entity **se; | |
114 | /* runqueue "owned" by this group on each cpu */ | |
115 | struct cfs_rq **cfs_rq; | |
116 | unsigned long shares; | |
117 | ||
118 | atomic_t load_weight; | |
119 | #endif | |
120 | ||
121 | #ifdef CONFIG_RT_GROUP_SCHED | |
122 | struct sched_rt_entity **rt_se; | |
123 | struct rt_rq **rt_rq; | |
124 | ||
125 | struct rt_bandwidth rt_bandwidth; | |
126 | #endif | |
127 | ||
128 | struct rcu_head rcu; | |
129 | struct list_head list; | |
130 | ||
131 | struct task_group *parent; | |
132 | struct list_head siblings; | |
133 | struct list_head children; | |
134 | ||
135 | #ifdef CONFIG_SCHED_AUTOGROUP | |
136 | struct autogroup *autogroup; | |
137 | #endif | |
138 | ||
139 | struct cfs_bandwidth cfs_bandwidth; | |
140 | }; | |
141 | ||
142 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
143 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
144 | ||
145 | /* | |
146 | * A weight of 0 or 1 can cause arithmetics problems. | |
147 | * A weight of a cfs_rq is the sum of weights of which entities | |
148 | * are queued on this cfs_rq, so a weight of a entity should not be | |
149 | * too large, so as the shares value of a task group. | |
150 | * (The default weight is 1024 - so there's no practical | |
151 | * limitation from this.) | |
152 | */ | |
153 | #define MIN_SHARES (1UL << 1) | |
154 | #define MAX_SHARES (1UL << 18) | |
155 | #endif | |
156 | ||
157 | /* Default task group. | |
158 | * Every task in system belong to this group at bootup. | |
159 | */ | |
160 | extern struct task_group root_task_group; | |
161 | ||
162 | typedef int (*tg_visitor)(struct task_group *, void *); | |
163 | ||
164 | extern int walk_tg_tree_from(struct task_group *from, | |
165 | tg_visitor down, tg_visitor up, void *data); | |
166 | ||
167 | /* | |
168 | * Iterate the full tree, calling @down when first entering a node and @up when | |
169 | * leaving it for the final time. | |
170 | * | |
171 | * Caller must hold rcu_lock or sufficient equivalent. | |
172 | */ | |
173 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
174 | { | |
175 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
176 | } | |
177 | ||
178 | extern int tg_nop(struct task_group *tg, void *data); | |
179 | ||
180 | extern void free_fair_sched_group(struct task_group *tg); | |
181 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
182 | extern void unregister_fair_sched_group(struct task_group *tg, int cpu); | |
183 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, | |
184 | struct sched_entity *se, int cpu, | |
185 | struct sched_entity *parent); | |
186 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
187 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
188 | ||
189 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
190 | extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
191 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); | |
192 | ||
193 | extern void free_rt_sched_group(struct task_group *tg); | |
194 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
195 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
196 | struct sched_rt_entity *rt_se, int cpu, | |
197 | struct sched_rt_entity *parent); | |
198 | ||
199 | #else /* CONFIG_CGROUP_SCHED */ | |
200 | ||
201 | struct cfs_bandwidth { }; | |
202 | ||
203 | #endif /* CONFIG_CGROUP_SCHED */ | |
204 | ||
205 | /* CFS-related fields in a runqueue */ | |
206 | struct cfs_rq { | |
207 | struct load_weight load; | |
208 | unsigned long nr_running, h_nr_running; | |
209 | ||
210 | u64 exec_clock; | |
211 | u64 min_vruntime; | |
212 | #ifndef CONFIG_64BIT | |
213 | u64 min_vruntime_copy; | |
214 | #endif | |
215 | ||
216 | struct rb_root tasks_timeline; | |
217 | struct rb_node *rb_leftmost; | |
218 | ||
219 | struct list_head tasks; | |
220 | struct list_head *balance_iterator; | |
221 | ||
222 | /* | |
223 | * 'curr' points to currently running entity on this cfs_rq. | |
224 | * It is set to NULL otherwise (i.e when none are currently running). | |
225 | */ | |
226 | struct sched_entity *curr, *next, *last, *skip; | |
227 | ||
228 | #ifdef CONFIG_SCHED_DEBUG | |
229 | unsigned int nr_spread_over; | |
230 | #endif | |
231 | ||
232 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
233 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
234 | ||
235 | /* | |
236 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
237 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
238 | * (like users, containers etc.) | |
239 | * | |
240 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
241 | * list is used during load balance. | |
242 | */ | |
243 | int on_list; | |
244 | struct list_head leaf_cfs_rq_list; | |
245 | struct task_group *tg; /* group that "owns" this runqueue */ | |
246 | ||
247 | #ifdef CONFIG_SMP | |
248 | /* | |
249 | * the part of load.weight contributed by tasks | |
250 | */ | |
251 | unsigned long task_weight; | |
252 | ||
253 | /* | |
254 | * h_load = weight * f(tg) | |
255 | * | |
256 | * Where f(tg) is the recursive weight fraction assigned to | |
257 | * this group. | |
258 | */ | |
259 | unsigned long h_load; | |
260 | ||
261 | /* | |
262 | * Maintaining per-cpu shares distribution for group scheduling | |
263 | * | |
264 | * load_stamp is the last time we updated the load average | |
265 | * load_last is the last time we updated the load average and saw load | |
266 | * load_unacc_exec_time is currently unaccounted execution time | |
267 | */ | |
268 | u64 load_avg; | |
269 | u64 load_period; | |
270 | u64 load_stamp, load_last, load_unacc_exec_time; | |
271 | ||
272 | unsigned long load_contribution; | |
273 | #endif /* CONFIG_SMP */ | |
274 | #ifdef CONFIG_CFS_BANDWIDTH | |
275 | int runtime_enabled; | |
276 | u64 runtime_expires; | |
277 | s64 runtime_remaining; | |
278 | ||
279 | u64 throttled_timestamp; | |
280 | int throttled, throttle_count; | |
281 | struct list_head throttled_list; | |
282 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
283 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
284 | }; | |
285 | ||
286 | static inline int rt_bandwidth_enabled(void) | |
287 | { | |
288 | return sysctl_sched_rt_runtime >= 0; | |
289 | } | |
290 | ||
291 | /* Real-Time classes' related field in a runqueue: */ | |
292 | struct rt_rq { | |
293 | struct rt_prio_array active; | |
294 | unsigned long rt_nr_running; | |
295 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED | |
296 | struct { | |
297 | int curr; /* highest queued rt task prio */ | |
298 | #ifdef CONFIG_SMP | |
299 | int next; /* next highest */ | |
300 | #endif | |
301 | } highest_prio; | |
302 | #endif | |
303 | #ifdef CONFIG_SMP | |
304 | unsigned long rt_nr_migratory; | |
305 | unsigned long rt_nr_total; | |
306 | int overloaded; | |
307 | struct plist_head pushable_tasks; | |
308 | #endif | |
309 | int rt_throttled; | |
310 | u64 rt_time; | |
311 | u64 rt_runtime; | |
312 | /* Nests inside the rq lock: */ | |
313 | raw_spinlock_t rt_runtime_lock; | |
314 | ||
315 | #ifdef CONFIG_RT_GROUP_SCHED | |
316 | unsigned long rt_nr_boosted; | |
317 | ||
318 | struct rq *rq; | |
319 | struct list_head leaf_rt_rq_list; | |
320 | struct task_group *tg; | |
321 | #endif | |
322 | }; | |
323 | ||
324 | #ifdef CONFIG_SMP | |
325 | ||
326 | /* | |
327 | * We add the notion of a root-domain which will be used to define per-domain | |
328 | * variables. Each exclusive cpuset essentially defines an island domain by | |
329 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
330 | * exclusive cpuset is created, we also create and attach a new root-domain | |
331 | * object. | |
332 | * | |
333 | */ | |
334 | struct root_domain { | |
335 | atomic_t refcount; | |
336 | atomic_t rto_count; | |
337 | struct rcu_head rcu; | |
338 | cpumask_var_t span; | |
339 | cpumask_var_t online; | |
340 | ||
341 | /* | |
342 | * The "RT overload" flag: it gets set if a CPU has more than | |
343 | * one runnable RT task. | |
344 | */ | |
345 | cpumask_var_t rto_mask; | |
346 | struct cpupri cpupri; | |
347 | }; | |
348 | ||
349 | extern struct root_domain def_root_domain; | |
350 | ||
351 | #endif /* CONFIG_SMP */ | |
352 | ||
353 | /* | |
354 | * This is the main, per-CPU runqueue data structure. | |
355 | * | |
356 | * Locking rule: those places that want to lock multiple runqueues | |
357 | * (such as the load balancing or the thread migration code), lock | |
358 | * acquire operations must be ordered by ascending &runqueue. | |
359 | */ | |
360 | struct rq { | |
361 | /* runqueue lock: */ | |
362 | raw_spinlock_t lock; | |
363 | ||
364 | /* | |
365 | * nr_running and cpu_load should be in the same cacheline because | |
366 | * remote CPUs use both these fields when doing load calculation. | |
367 | */ | |
368 | unsigned long nr_running; | |
369 | #define CPU_LOAD_IDX_MAX 5 | |
370 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
371 | unsigned long last_load_update_tick; | |
372 | #ifdef CONFIG_NO_HZ | |
373 | u64 nohz_stamp; | |
374 | unsigned char nohz_balance_kick; | |
375 | #endif | |
376 | int skip_clock_update; | |
377 | ||
378 | /* capture load from *all* tasks on this cpu: */ | |
379 | struct load_weight load; | |
380 | unsigned long nr_load_updates; | |
381 | u64 nr_switches; | |
382 | ||
383 | struct cfs_rq cfs; | |
384 | struct rt_rq rt; | |
385 | ||
386 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
387 | /* list of leaf cfs_rq on this cpu: */ | |
388 | struct list_head leaf_cfs_rq_list; | |
389 | #endif | |
390 | #ifdef CONFIG_RT_GROUP_SCHED | |
391 | struct list_head leaf_rt_rq_list; | |
392 | #endif | |
393 | ||
394 | /* | |
395 | * This is part of a global counter where only the total sum | |
396 | * over all CPUs matters. A task can increase this counter on | |
397 | * one CPU and if it got migrated afterwards it may decrease | |
398 | * it on another CPU. Always updated under the runqueue lock: | |
399 | */ | |
400 | unsigned long nr_uninterruptible; | |
401 | ||
402 | struct task_struct *curr, *idle, *stop; | |
403 | unsigned long next_balance; | |
404 | struct mm_struct *prev_mm; | |
405 | ||
406 | u64 clock; | |
407 | u64 clock_task; | |
408 | ||
409 | atomic_t nr_iowait; | |
410 | ||
411 | #ifdef CONFIG_SMP | |
412 | struct root_domain *rd; | |
413 | struct sched_domain *sd; | |
414 | ||
415 | unsigned long cpu_power; | |
416 | ||
417 | unsigned char idle_balance; | |
418 | /* For active balancing */ | |
419 | int post_schedule; | |
420 | int active_balance; | |
421 | int push_cpu; | |
422 | struct cpu_stop_work active_balance_work; | |
423 | /* cpu of this runqueue: */ | |
424 | int cpu; | |
425 | int online; | |
426 | ||
427 | u64 rt_avg; | |
428 | u64 age_stamp; | |
429 | u64 idle_stamp; | |
430 | u64 avg_idle; | |
431 | #endif | |
432 | ||
433 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
434 | u64 prev_irq_time; | |
435 | #endif | |
436 | #ifdef CONFIG_PARAVIRT | |
437 | u64 prev_steal_time; | |
438 | #endif | |
439 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
440 | u64 prev_steal_time_rq; | |
441 | #endif | |
442 | ||
443 | /* calc_load related fields */ | |
444 | unsigned long calc_load_update; | |
445 | long calc_load_active; | |
446 | ||
447 | #ifdef CONFIG_SCHED_HRTICK | |
448 | #ifdef CONFIG_SMP | |
449 | int hrtick_csd_pending; | |
450 | struct call_single_data hrtick_csd; | |
451 | #endif | |
452 | struct hrtimer hrtick_timer; | |
453 | #endif | |
454 | ||
455 | #ifdef CONFIG_SCHEDSTATS | |
456 | /* latency stats */ | |
457 | struct sched_info rq_sched_info; | |
458 | unsigned long long rq_cpu_time; | |
459 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
460 | ||
461 | /* sys_sched_yield() stats */ | |
462 | unsigned int yld_count; | |
463 | ||
464 | /* schedule() stats */ | |
465 | unsigned int sched_switch; | |
466 | unsigned int sched_count; | |
467 | unsigned int sched_goidle; | |
468 | ||
469 | /* try_to_wake_up() stats */ | |
470 | unsigned int ttwu_count; | |
471 | unsigned int ttwu_local; | |
472 | #endif | |
473 | ||
474 | #ifdef CONFIG_SMP | |
475 | struct llist_head wake_list; | |
476 | #endif | |
477 | }; | |
478 | ||
479 | static inline int cpu_of(struct rq *rq) | |
480 | { | |
481 | #ifdef CONFIG_SMP | |
482 | return rq->cpu; | |
483 | #else | |
484 | return 0; | |
485 | #endif | |
486 | } | |
487 | ||
488 | DECLARE_PER_CPU(struct rq, runqueues); | |
489 | ||
490 | #define rcu_dereference_check_sched_domain(p) \ | |
491 | rcu_dereference_check((p), \ | |
492 | lockdep_is_held(&sched_domains_mutex)) | |
493 | ||
494 | /* | |
495 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
496 | * See detach_destroy_domains: synchronize_sched for details. | |
497 | * | |
498 | * The domain tree of any CPU may only be accessed from within | |
499 | * preempt-disabled sections. | |
500 | */ | |
501 | #define for_each_domain(cpu, __sd) \ | |
502 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
503 | ||
504 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
505 | #define this_rq() (&__get_cpu_var(runqueues)) | |
506 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
507 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
508 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) | |
509 | ||
391e43da PZ |
510 | #include "stats.h" |
511 | #include "auto_group.h" | |
029632fb PZ |
512 | |
513 | #ifdef CONFIG_CGROUP_SCHED | |
514 | ||
515 | /* | |
516 | * Return the group to which this tasks belongs. | |
517 | * | |
518 | * We use task_subsys_state_check() and extend the RCU verification with | |
519 | * pi->lock and rq->lock because cpu_cgroup_attach() holds those locks for each | |
520 | * task it moves into the cgroup. Therefore by holding either of those locks, | |
521 | * we pin the task to the current cgroup. | |
522 | */ | |
523 | static inline struct task_group *task_group(struct task_struct *p) | |
524 | { | |
525 | struct task_group *tg; | |
526 | struct cgroup_subsys_state *css; | |
527 | ||
528 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
529 | lockdep_is_held(&p->pi_lock) || | |
530 | lockdep_is_held(&task_rq(p)->lock)); | |
531 | tg = container_of(css, struct task_group, css); | |
532 | ||
533 | return autogroup_task_group(p, tg); | |
534 | } | |
535 | ||
536 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
537 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
538 | { | |
539 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
540 | struct task_group *tg = task_group(p); | |
541 | #endif | |
542 | ||
543 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
544 | p->se.cfs_rq = tg->cfs_rq[cpu]; | |
545 | p->se.parent = tg->se[cpu]; | |
546 | #endif | |
547 | ||
548 | #ifdef CONFIG_RT_GROUP_SCHED | |
549 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
550 | p->rt.parent = tg->rt_se[cpu]; | |
551 | #endif | |
552 | } | |
553 | ||
554 | #else /* CONFIG_CGROUP_SCHED */ | |
555 | ||
556 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
557 | static inline struct task_group *task_group(struct task_struct *p) | |
558 | { | |
559 | return NULL; | |
560 | } | |
561 | ||
562 | #endif /* CONFIG_CGROUP_SCHED */ | |
563 | ||
564 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
565 | { | |
566 | set_task_rq(p, cpu); | |
567 | #ifdef CONFIG_SMP | |
568 | /* | |
569 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
570 | * successfuly executed on another CPU. We must ensure that updates of | |
571 | * per-task data have been completed by this moment. | |
572 | */ | |
573 | smp_wmb(); | |
574 | task_thread_info(p)->cpu = cpu; | |
575 | #endif | |
576 | } | |
577 | ||
578 | /* | |
579 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
580 | */ | |
581 | #ifdef CONFIG_SCHED_DEBUG | |
582 | # define const_debug __read_mostly | |
583 | #else | |
584 | # define const_debug const | |
585 | #endif | |
586 | ||
587 | extern const_debug unsigned int sysctl_sched_features; | |
588 | ||
589 | #define SCHED_FEAT(name, enabled) \ | |
590 | __SCHED_FEAT_##name , | |
591 | ||
592 | enum { | |
391e43da | 593 | #include "features.h" |
029632fb PZ |
594 | }; |
595 | ||
596 | #undef SCHED_FEAT | |
597 | ||
598 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
599 | ||
600 | static inline u64 global_rt_period(void) | |
601 | { | |
602 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
603 | } | |
604 | ||
605 | static inline u64 global_rt_runtime(void) | |
606 | { | |
607 | if (sysctl_sched_rt_runtime < 0) | |
608 | return RUNTIME_INF; | |
609 | ||
610 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
611 | } | |
612 | ||
613 | ||
614 | ||
615 | static inline int task_current(struct rq *rq, struct task_struct *p) | |
616 | { | |
617 | return rq->curr == p; | |
618 | } | |
619 | ||
620 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
621 | { | |
622 | #ifdef CONFIG_SMP | |
623 | return p->on_cpu; | |
624 | #else | |
625 | return task_current(rq, p); | |
626 | #endif | |
627 | } | |
628 | ||
629 | ||
630 | #ifndef prepare_arch_switch | |
631 | # define prepare_arch_switch(next) do { } while (0) | |
632 | #endif | |
633 | #ifndef finish_arch_switch | |
634 | # define finish_arch_switch(prev) do { } while (0) | |
635 | #endif | |
636 | ||
637 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
638 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
639 | { | |
640 | #ifdef CONFIG_SMP | |
641 | /* | |
642 | * We can optimise this out completely for !SMP, because the | |
643 | * SMP rebalancing from interrupt is the only thing that cares | |
644 | * here. | |
645 | */ | |
646 | next->on_cpu = 1; | |
647 | #endif | |
648 | } | |
649 | ||
650 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
651 | { | |
652 | #ifdef CONFIG_SMP | |
653 | /* | |
654 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
655 | * We must ensure this doesn't happen until the switch is completely | |
656 | * finished. | |
657 | */ | |
658 | smp_wmb(); | |
659 | prev->on_cpu = 0; | |
660 | #endif | |
661 | #ifdef CONFIG_DEBUG_SPINLOCK | |
662 | /* this is a valid case when another task releases the spinlock */ | |
663 | rq->lock.owner = current; | |
664 | #endif | |
665 | /* | |
666 | * If we are tracking spinlock dependencies then we have to | |
667 | * fix up the runqueue lock - which gets 'carried over' from | |
668 | * prev into current: | |
669 | */ | |
670 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
671 | ||
672 | raw_spin_unlock_irq(&rq->lock); | |
673 | } | |
674 | ||
675 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
676 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
677 | { | |
678 | #ifdef CONFIG_SMP | |
679 | /* | |
680 | * We can optimise this out completely for !SMP, because the | |
681 | * SMP rebalancing from interrupt is the only thing that cares | |
682 | * here. | |
683 | */ | |
684 | next->on_cpu = 1; | |
685 | #endif | |
686 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
687 | raw_spin_unlock_irq(&rq->lock); | |
688 | #else | |
689 | raw_spin_unlock(&rq->lock); | |
690 | #endif | |
691 | } | |
692 | ||
693 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
694 | { | |
695 | #ifdef CONFIG_SMP | |
696 | /* | |
697 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
698 | * We must ensure this doesn't happen until the switch is completely | |
699 | * finished. | |
700 | */ | |
701 | smp_wmb(); | |
702 | prev->on_cpu = 0; | |
703 | #endif | |
704 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
705 | local_irq_enable(); | |
706 | #endif | |
707 | } | |
708 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
709 | ||
710 | ||
711 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) | |
712 | { | |
713 | lw->weight += inc; | |
714 | lw->inv_weight = 0; | |
715 | } | |
716 | ||
717 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) | |
718 | { | |
719 | lw->weight -= dec; | |
720 | lw->inv_weight = 0; | |
721 | } | |
722 | ||
723 | static inline void update_load_set(struct load_weight *lw, unsigned long w) | |
724 | { | |
725 | lw->weight = w; | |
726 | lw->inv_weight = 0; | |
727 | } | |
728 | ||
729 | /* | |
730 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
731 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
732 | * each task makes to its run queue's load is weighted according to its | |
733 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
734 | * scaled version of the new time slice allocation that they receive on time | |
735 | * slice expiry etc. | |
736 | */ | |
737 | ||
738 | #define WEIGHT_IDLEPRIO 3 | |
739 | #define WMULT_IDLEPRIO 1431655765 | |
740 | ||
741 | /* | |
742 | * Nice levels are multiplicative, with a gentle 10% change for every | |
743 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
744 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
745 | * that remained on nice 0. | |
746 | * | |
747 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
748 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
749 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
750 | * If a task goes up by ~10% and another task goes down by ~10% then | |
751 | * the relative distance between them is ~25%.) | |
752 | */ | |
753 | static const int prio_to_weight[40] = { | |
754 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
755 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
756 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
757 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
758 | /* 0 */ 1024, 820, 655, 526, 423, | |
759 | /* 5 */ 335, 272, 215, 172, 137, | |
760 | /* 10 */ 110, 87, 70, 56, 45, | |
761 | /* 15 */ 36, 29, 23, 18, 15, | |
762 | }; | |
763 | ||
764 | /* | |
765 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
766 | * | |
767 | * In cases where the weight does not change often, we can use the | |
768 | * precalculated inverse to speed up arithmetics by turning divisions | |
769 | * into multiplications: | |
770 | */ | |
771 | static const u32 prio_to_wmult[40] = { | |
772 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
773 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
774 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
775 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
776 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
777 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
778 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
779 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
780 | }; | |
781 | ||
782 | /* Time spent by the tasks of the cpu accounting group executing in ... */ | |
783 | enum cpuacct_stat_index { | |
784 | CPUACCT_STAT_USER, /* ... user mode */ | |
785 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
786 | ||
787 | CPUACCT_STAT_NSTATS, | |
788 | }; | |
789 | ||
790 | ||
791 | #define sched_class_highest (&stop_sched_class) | |
792 | #define for_each_class(class) \ | |
793 | for (class = sched_class_highest; class; class = class->next) | |
794 | ||
795 | extern const struct sched_class stop_sched_class; | |
796 | extern const struct sched_class rt_sched_class; | |
797 | extern const struct sched_class fair_sched_class; | |
798 | extern const struct sched_class idle_sched_class; | |
799 | ||
800 | ||
801 | #ifdef CONFIG_SMP | |
802 | ||
803 | extern void trigger_load_balance(struct rq *rq, int cpu); | |
804 | extern void idle_balance(int this_cpu, struct rq *this_rq); | |
805 | ||
806 | #else /* CONFIG_SMP */ | |
807 | ||
808 | static inline void idle_balance(int cpu, struct rq *rq) | |
809 | { | |
810 | } | |
811 | ||
812 | #endif | |
813 | ||
814 | extern void sysrq_sched_debug_show(void); | |
815 | extern void sched_init_granularity(void); | |
816 | extern void update_max_interval(void); | |
817 | extern void update_group_power(struct sched_domain *sd, int cpu); | |
818 | extern int update_runtime(struct notifier_block *nfb, unsigned long action, void *hcpu); | |
819 | extern void init_sched_rt_class(void); | |
820 | extern void init_sched_fair_class(void); | |
821 | ||
822 | extern void resched_task(struct task_struct *p); | |
823 | extern void resched_cpu(int cpu); | |
824 | ||
825 | extern struct rt_bandwidth def_rt_bandwidth; | |
826 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
827 | ||
828 | extern void update_cpu_load(struct rq *this_rq); | |
829 | ||
830 | #ifdef CONFIG_CGROUP_CPUACCT | |
831 | extern void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
832 | extern void cpuacct_update_stats(struct task_struct *tsk, | |
833 | enum cpuacct_stat_index idx, cputime_t val); | |
834 | #else | |
835 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
836 | static inline void cpuacct_update_stats(struct task_struct *tsk, | |
837 | enum cpuacct_stat_index idx, cputime_t val) {} | |
838 | #endif | |
839 | ||
840 | static inline void inc_nr_running(struct rq *rq) | |
841 | { | |
842 | rq->nr_running++; | |
843 | } | |
844 | ||
845 | static inline void dec_nr_running(struct rq *rq) | |
846 | { | |
847 | rq->nr_running--; | |
848 | } | |
849 | ||
850 | extern void update_rq_clock(struct rq *rq); | |
851 | ||
852 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
853 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
854 | ||
855 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
856 | ||
857 | extern const_debug unsigned int sysctl_sched_time_avg; | |
858 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
859 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
860 | ||
861 | static inline u64 sched_avg_period(void) | |
862 | { | |
863 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
864 | } | |
865 | ||
866 | void calc_load_account_idle(struct rq *this_rq); | |
867 | ||
868 | #ifdef CONFIG_SCHED_HRTICK | |
869 | ||
870 | /* | |
871 | * Use hrtick when: | |
872 | * - enabled by features | |
873 | * - hrtimer is actually high res | |
874 | */ | |
875 | static inline int hrtick_enabled(struct rq *rq) | |
876 | { | |
877 | if (!sched_feat(HRTICK)) | |
878 | return 0; | |
879 | if (!cpu_active(cpu_of(rq))) | |
880 | return 0; | |
881 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
882 | } | |
883 | ||
884 | void hrtick_start(struct rq *rq, u64 delay); | |
885 | ||
886 | #endif /* CONFIG_SCHED_HRTICK */ | |
887 | ||
888 | #ifdef CONFIG_SMP | |
889 | extern void sched_avg_update(struct rq *rq); | |
890 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
891 | { | |
892 | rq->rt_avg += rt_delta; | |
893 | sched_avg_update(rq); | |
894 | } | |
895 | #else | |
896 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
897 | static inline void sched_avg_update(struct rq *rq) { } | |
898 | #endif | |
899 | ||
900 | extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period); | |
901 | ||
902 | #ifdef CONFIG_SMP | |
903 | #ifdef CONFIG_PREEMPT | |
904 | ||
905 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
906 | ||
907 | /* | |
908 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
909 | * way at the expense of forcing extra atomic operations in all | |
910 | * invocations. This assures that the double_lock is acquired using the | |
911 | * same underlying policy as the spinlock_t on this architecture, which | |
912 | * reduces latency compared to the unfair variant below. However, it | |
913 | * also adds more overhead and therefore may reduce throughput. | |
914 | */ | |
915 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
916 | __releases(this_rq->lock) | |
917 | __acquires(busiest->lock) | |
918 | __acquires(this_rq->lock) | |
919 | { | |
920 | raw_spin_unlock(&this_rq->lock); | |
921 | double_rq_lock(this_rq, busiest); | |
922 | ||
923 | return 1; | |
924 | } | |
925 | ||
926 | #else | |
927 | /* | |
928 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
929 | * latency by eliminating extra atomic operations when the locks are | |
930 | * already in proper order on entry. This favors lower cpu-ids and will | |
931 | * grant the double lock to lower cpus over higher ids under contention, | |
932 | * regardless of entry order into the function. | |
933 | */ | |
934 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
935 | __releases(this_rq->lock) | |
936 | __acquires(busiest->lock) | |
937 | __acquires(this_rq->lock) | |
938 | { | |
939 | int ret = 0; | |
940 | ||
941 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
942 | if (busiest < this_rq) { | |
943 | raw_spin_unlock(&this_rq->lock); | |
944 | raw_spin_lock(&busiest->lock); | |
945 | raw_spin_lock_nested(&this_rq->lock, | |
946 | SINGLE_DEPTH_NESTING); | |
947 | ret = 1; | |
948 | } else | |
949 | raw_spin_lock_nested(&busiest->lock, | |
950 | SINGLE_DEPTH_NESTING); | |
951 | } | |
952 | return ret; | |
953 | } | |
954 | ||
955 | #endif /* CONFIG_PREEMPT */ | |
956 | ||
957 | /* | |
958 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
959 | */ | |
960 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
961 | { | |
962 | if (unlikely(!irqs_disabled())) { | |
963 | /* printk() doesn't work good under rq->lock */ | |
964 | raw_spin_unlock(&this_rq->lock); | |
965 | BUG_ON(1); | |
966 | } | |
967 | ||
968 | return _double_lock_balance(this_rq, busiest); | |
969 | } | |
970 | ||
971 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
972 | __releases(busiest->lock) | |
973 | { | |
974 | raw_spin_unlock(&busiest->lock); | |
975 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
976 | } | |
977 | ||
978 | /* | |
979 | * double_rq_lock - safely lock two runqueues | |
980 | * | |
981 | * Note this does not disable interrupts like task_rq_lock, | |
982 | * you need to do so manually before calling. | |
983 | */ | |
984 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
985 | __acquires(rq1->lock) | |
986 | __acquires(rq2->lock) | |
987 | { | |
988 | BUG_ON(!irqs_disabled()); | |
989 | if (rq1 == rq2) { | |
990 | raw_spin_lock(&rq1->lock); | |
991 | __acquire(rq2->lock); /* Fake it out ;) */ | |
992 | } else { | |
993 | if (rq1 < rq2) { | |
994 | raw_spin_lock(&rq1->lock); | |
995 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
996 | } else { | |
997 | raw_spin_lock(&rq2->lock); | |
998 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
999 | } | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * double_rq_unlock - safely unlock two runqueues | |
1005 | * | |
1006 | * Note this does not restore interrupts like task_rq_unlock, | |
1007 | * you need to do so manually after calling. | |
1008 | */ | |
1009 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1010 | __releases(rq1->lock) | |
1011 | __releases(rq2->lock) | |
1012 | { | |
1013 | raw_spin_unlock(&rq1->lock); | |
1014 | if (rq1 != rq2) | |
1015 | raw_spin_unlock(&rq2->lock); | |
1016 | else | |
1017 | __release(rq2->lock); | |
1018 | } | |
1019 | ||
1020 | #else /* CONFIG_SMP */ | |
1021 | ||
1022 | /* | |
1023 | * double_rq_lock - safely lock two runqueues | |
1024 | * | |
1025 | * Note this does not disable interrupts like task_rq_lock, | |
1026 | * you need to do so manually before calling. | |
1027 | */ | |
1028 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1029 | __acquires(rq1->lock) | |
1030 | __acquires(rq2->lock) | |
1031 | { | |
1032 | BUG_ON(!irqs_disabled()); | |
1033 | BUG_ON(rq1 != rq2); | |
1034 | raw_spin_lock(&rq1->lock); | |
1035 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * double_rq_unlock - safely unlock two runqueues | |
1040 | * | |
1041 | * Note this does not restore interrupts like task_rq_unlock, | |
1042 | * you need to do so manually after calling. | |
1043 | */ | |
1044 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1045 | __releases(rq1->lock) | |
1046 | __releases(rq2->lock) | |
1047 | { | |
1048 | BUG_ON(rq1 != rq2); | |
1049 | raw_spin_unlock(&rq1->lock); | |
1050 | __release(rq2->lock); | |
1051 | } | |
1052 | ||
1053 | #endif | |
1054 | ||
1055 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1056 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
1057 | extern void print_cfs_stats(struct seq_file *m, int cpu); | |
1058 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
1059 | ||
1060 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
1061 | extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq); | |
1062 | extern void unthrottle_offline_cfs_rqs(struct rq *rq); | |
1063 | ||
1064 | extern void account_cfs_bandwidth_used(int enabled, int was_enabled); |