Commit | Line | Data |
---|---|---|
029632fb PZ |
1 | |
2 | #include <linux/sched.h> | |
cf4aebc2 | 3 | #include <linux/sched/sysctl.h> |
8bd75c77 | 4 | #include <linux/sched/rt.h> |
aab03e05 | 5 | #include <linux/sched/deadline.h> |
029632fb PZ |
6 | #include <linux/mutex.h> |
7 | #include <linux/spinlock.h> | |
8 | #include <linux/stop_machine.h> | |
9f3660c2 | 9 | #include <linux/tick.h> |
f809ca9a | 10 | #include <linux/slab.h> |
029632fb | 11 | |
391e43da | 12 | #include "cpupri.h" |
6bfd6d72 | 13 | #include "cpudeadline.h" |
60fed789 | 14 | #include "cpuacct.h" |
029632fb | 15 | |
45ceebf7 PG |
16 | struct rq; |
17 | ||
029632fb PZ |
18 | extern __read_mostly int scheduler_running; |
19 | ||
45ceebf7 PG |
20 | extern unsigned long calc_load_update; |
21 | extern atomic_long_t calc_load_tasks; | |
22 | ||
23 | extern long calc_load_fold_active(struct rq *this_rq); | |
24 | extern void update_cpu_load_active(struct rq *this_rq); | |
25 | ||
029632fb PZ |
26 | /* |
27 | * Helpers for converting nanosecond timing to jiffy resolution | |
28 | */ | |
29 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
30 | ||
cc1f4b1f LZ |
31 | /* |
32 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
33 | * The extra resolution improves shares distribution and load balancing of | |
34 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
35 | * hierarchies, especially on larger systems. This is not a user-visible change | |
36 | * and does not change the user-interface for setting shares/weights. | |
37 | * | |
38 | * We increase resolution only if we have enough bits to allow this increased | |
39 | * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution | |
40 | * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the | |
41 | * increased costs. | |
42 | */ | |
43 | #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */ | |
44 | # define SCHED_LOAD_RESOLUTION 10 | |
45 | # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION) | |
46 | # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION) | |
47 | #else | |
48 | # define SCHED_LOAD_RESOLUTION 0 | |
49 | # define scale_load(w) (w) | |
50 | # define scale_load_down(w) (w) | |
51 | #endif | |
52 | ||
53 | #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION) | |
54 | #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT) | |
55 | ||
029632fb PZ |
56 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
57 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
58 | ||
332ac17e DF |
59 | /* |
60 | * Single value that decides SCHED_DEADLINE internal math precision. | |
61 | * 10 -> just above 1us | |
62 | * 9 -> just above 0.5us | |
63 | */ | |
64 | #define DL_SCALE (10) | |
65 | ||
029632fb PZ |
66 | /* |
67 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 68 | */ |
029632fb PZ |
69 | |
70 | /* | |
71 | * single value that denotes runtime == period, ie unlimited time. | |
72 | */ | |
73 | #define RUNTIME_INF ((u64)~0ULL) | |
74 | ||
d50dde5a DF |
75 | static inline int fair_policy(int policy) |
76 | { | |
77 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
78 | } | |
79 | ||
029632fb PZ |
80 | static inline int rt_policy(int policy) |
81 | { | |
d50dde5a | 82 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
83 | } |
84 | ||
aab03e05 DF |
85 | static inline int dl_policy(int policy) |
86 | { | |
87 | return policy == SCHED_DEADLINE; | |
88 | } | |
89 | ||
029632fb PZ |
90 | static inline int task_has_rt_policy(struct task_struct *p) |
91 | { | |
92 | return rt_policy(p->policy); | |
93 | } | |
94 | ||
aab03e05 DF |
95 | static inline int task_has_dl_policy(struct task_struct *p) |
96 | { | |
97 | return dl_policy(p->policy); | |
98 | } | |
99 | ||
332ac17e | 100 | static inline bool dl_time_before(u64 a, u64 b) |
2d3d891d DF |
101 | { |
102 | return (s64)(a - b) < 0; | |
103 | } | |
104 | ||
105 | /* | |
106 | * Tells if entity @a should preempt entity @b. | |
107 | */ | |
332ac17e DF |
108 | static inline bool |
109 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
110 | { |
111 | return dl_time_before(a->deadline, b->deadline); | |
112 | } | |
113 | ||
029632fb PZ |
114 | /* |
115 | * This is the priority-queue data structure of the RT scheduling class: | |
116 | */ | |
117 | struct rt_prio_array { | |
118 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
119 | struct list_head queue[MAX_RT_PRIO]; | |
120 | }; | |
121 | ||
122 | struct rt_bandwidth { | |
123 | /* nests inside the rq lock: */ | |
124 | raw_spinlock_t rt_runtime_lock; | |
125 | ktime_t rt_period; | |
126 | u64 rt_runtime; | |
127 | struct hrtimer rt_period_timer; | |
128 | }; | |
332ac17e DF |
129 | /* |
130 | * To keep the bandwidth of -deadline tasks and groups under control | |
131 | * we need some place where: | |
132 | * - store the maximum -deadline bandwidth of the system (the group); | |
133 | * - cache the fraction of that bandwidth that is currently allocated. | |
134 | * | |
135 | * This is all done in the data structure below. It is similar to the | |
136 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
137 | * that, since here we are only interested in admission control, we | |
138 | * do not decrease any runtime while the group "executes", neither we | |
139 | * need a timer to replenish it. | |
140 | * | |
141 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
142 | * meaning that: | |
143 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
144 | * - dl_total_bw array contains, in the i-eth element, the currently | |
145 | * allocated bandwidth on the i-eth CPU. | |
146 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
147 | * consume bandwidth on the CPU they're running on. | |
148 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
149 | * that will be shown the next time the proc or cgroup controls will | |
150 | * be red. It on its turn can be changed by writing on its own | |
151 | * control. | |
152 | */ | |
153 | struct dl_bandwidth { | |
154 | raw_spinlock_t dl_runtime_lock; | |
155 | u64 dl_runtime; | |
156 | u64 dl_period; | |
157 | }; | |
158 | ||
159 | static inline int dl_bandwidth_enabled(void) | |
160 | { | |
1724813d | 161 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
162 | } |
163 | ||
164 | extern struct dl_bw *dl_bw_of(int i); | |
165 | ||
166 | struct dl_bw { | |
167 | raw_spinlock_t lock; | |
168 | u64 bw, total_bw; | |
169 | }; | |
170 | ||
029632fb PZ |
171 | extern struct mutex sched_domains_mutex; |
172 | ||
173 | #ifdef CONFIG_CGROUP_SCHED | |
174 | ||
175 | #include <linux/cgroup.h> | |
176 | ||
177 | struct cfs_rq; | |
178 | struct rt_rq; | |
179 | ||
35cf4e50 | 180 | extern struct list_head task_groups; |
029632fb PZ |
181 | |
182 | struct cfs_bandwidth { | |
183 | #ifdef CONFIG_CFS_BANDWIDTH | |
184 | raw_spinlock_t lock; | |
185 | ktime_t period; | |
186 | u64 quota, runtime; | |
187 | s64 hierarchal_quota; | |
188 | u64 runtime_expires; | |
189 | ||
190 | int idle, timer_active; | |
191 | struct hrtimer period_timer, slack_timer; | |
192 | struct list_head throttled_cfs_rq; | |
193 | ||
194 | /* statistics */ | |
195 | int nr_periods, nr_throttled; | |
196 | u64 throttled_time; | |
197 | #endif | |
198 | }; | |
199 | ||
200 | /* task group related information */ | |
201 | struct task_group { | |
202 | struct cgroup_subsys_state css; | |
203 | ||
204 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
205 | /* schedulable entities of this group on each cpu */ | |
206 | struct sched_entity **se; | |
207 | /* runqueue "owned" by this group on each cpu */ | |
208 | struct cfs_rq **cfs_rq; | |
209 | unsigned long shares; | |
210 | ||
fa6bddeb | 211 | #ifdef CONFIG_SMP |
bf5b986e | 212 | atomic_long_t load_avg; |
bb17f655 | 213 | atomic_t runnable_avg; |
029632fb | 214 | #endif |
fa6bddeb | 215 | #endif |
029632fb PZ |
216 | |
217 | #ifdef CONFIG_RT_GROUP_SCHED | |
218 | struct sched_rt_entity **rt_se; | |
219 | struct rt_rq **rt_rq; | |
220 | ||
221 | struct rt_bandwidth rt_bandwidth; | |
222 | #endif | |
223 | ||
224 | struct rcu_head rcu; | |
225 | struct list_head list; | |
226 | ||
227 | struct task_group *parent; | |
228 | struct list_head siblings; | |
229 | struct list_head children; | |
230 | ||
231 | #ifdef CONFIG_SCHED_AUTOGROUP | |
232 | struct autogroup *autogroup; | |
233 | #endif | |
234 | ||
235 | struct cfs_bandwidth cfs_bandwidth; | |
236 | }; | |
237 | ||
238 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
239 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
240 | ||
241 | /* | |
242 | * A weight of 0 or 1 can cause arithmetics problems. | |
243 | * A weight of a cfs_rq is the sum of weights of which entities | |
244 | * are queued on this cfs_rq, so a weight of a entity should not be | |
245 | * too large, so as the shares value of a task group. | |
246 | * (The default weight is 1024 - so there's no practical | |
247 | * limitation from this.) | |
248 | */ | |
249 | #define MIN_SHARES (1UL << 1) | |
250 | #define MAX_SHARES (1UL << 18) | |
251 | #endif | |
252 | ||
029632fb PZ |
253 | typedef int (*tg_visitor)(struct task_group *, void *); |
254 | ||
255 | extern int walk_tg_tree_from(struct task_group *from, | |
256 | tg_visitor down, tg_visitor up, void *data); | |
257 | ||
258 | /* | |
259 | * Iterate the full tree, calling @down when first entering a node and @up when | |
260 | * leaving it for the final time. | |
261 | * | |
262 | * Caller must hold rcu_lock or sufficient equivalent. | |
263 | */ | |
264 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
265 | { | |
266 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
267 | } | |
268 | ||
269 | extern int tg_nop(struct task_group *tg, void *data); | |
270 | ||
271 | extern void free_fair_sched_group(struct task_group *tg); | |
272 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
273 | extern void unregister_fair_sched_group(struct task_group *tg, int cpu); | |
274 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, | |
275 | struct sched_entity *se, int cpu, | |
276 | struct sched_entity *parent); | |
277 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
278 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
279 | ||
280 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
09dc4ab0 | 281 | extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force); |
029632fb PZ |
282 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
283 | ||
284 | extern void free_rt_sched_group(struct task_group *tg); | |
285 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
286 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
287 | struct sched_rt_entity *rt_se, int cpu, | |
288 | struct sched_rt_entity *parent); | |
289 | ||
25cc7da7 LZ |
290 | extern struct task_group *sched_create_group(struct task_group *parent); |
291 | extern void sched_online_group(struct task_group *tg, | |
292 | struct task_group *parent); | |
293 | extern void sched_destroy_group(struct task_group *tg); | |
294 | extern void sched_offline_group(struct task_group *tg); | |
295 | ||
296 | extern void sched_move_task(struct task_struct *tsk); | |
297 | ||
298 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
299 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
300 | #endif | |
301 | ||
029632fb PZ |
302 | #else /* CONFIG_CGROUP_SCHED */ |
303 | ||
304 | struct cfs_bandwidth { }; | |
305 | ||
306 | #endif /* CONFIG_CGROUP_SCHED */ | |
307 | ||
308 | /* CFS-related fields in a runqueue */ | |
309 | struct cfs_rq { | |
310 | struct load_weight load; | |
c82513e5 | 311 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
312 | |
313 | u64 exec_clock; | |
314 | u64 min_vruntime; | |
315 | #ifndef CONFIG_64BIT | |
316 | u64 min_vruntime_copy; | |
317 | #endif | |
318 | ||
319 | struct rb_root tasks_timeline; | |
320 | struct rb_node *rb_leftmost; | |
321 | ||
029632fb PZ |
322 | /* |
323 | * 'curr' points to currently running entity on this cfs_rq. | |
324 | * It is set to NULL otherwise (i.e when none are currently running). | |
325 | */ | |
326 | struct sched_entity *curr, *next, *last, *skip; | |
327 | ||
328 | #ifdef CONFIG_SCHED_DEBUG | |
329 | unsigned int nr_spread_over; | |
330 | #endif | |
331 | ||
2dac754e PT |
332 | #ifdef CONFIG_SMP |
333 | /* | |
334 | * CFS Load tracking | |
335 | * Under CFS, load is tracked on a per-entity basis and aggregated up. | |
336 | * This allows for the description of both thread and group usage (in | |
337 | * the FAIR_GROUP_SCHED case). | |
338 | */ | |
72a4cf20 | 339 | unsigned long runnable_load_avg, blocked_load_avg; |
2509940f | 340 | atomic64_t decay_counter; |
9ee474f5 | 341 | u64 last_decay; |
2509940f | 342 | atomic_long_t removed_load; |
141965c7 | 343 | |
c566e8e9 | 344 | #ifdef CONFIG_FAIR_GROUP_SCHED |
141965c7 | 345 | /* Required to track per-cpu representation of a task_group */ |
bb17f655 | 346 | u32 tg_runnable_contrib; |
bf5b986e | 347 | unsigned long tg_load_contrib; |
82958366 PT |
348 | |
349 | /* | |
350 | * h_load = weight * f(tg) | |
351 | * | |
352 | * Where f(tg) is the recursive weight fraction assigned to | |
353 | * this group. | |
354 | */ | |
355 | unsigned long h_load; | |
68520796 VD |
356 | u64 last_h_load_update; |
357 | struct sched_entity *h_load_next; | |
358 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
359 | #endif /* CONFIG_SMP */ |
360 | ||
029632fb PZ |
361 | #ifdef CONFIG_FAIR_GROUP_SCHED |
362 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
363 | ||
364 | /* | |
365 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
366 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
367 | * (like users, containers etc.) | |
368 | * | |
369 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
370 | * list is used during load balance. | |
371 | */ | |
372 | int on_list; | |
373 | struct list_head leaf_cfs_rq_list; | |
374 | struct task_group *tg; /* group that "owns" this runqueue */ | |
375 | ||
029632fb PZ |
376 | #ifdef CONFIG_CFS_BANDWIDTH |
377 | int runtime_enabled; | |
378 | u64 runtime_expires; | |
379 | s64 runtime_remaining; | |
380 | ||
f1b17280 PT |
381 | u64 throttled_clock, throttled_clock_task; |
382 | u64 throttled_clock_task_time; | |
029632fb PZ |
383 | int throttled, throttle_count; |
384 | struct list_head throttled_list; | |
385 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
386 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
387 | }; | |
388 | ||
389 | static inline int rt_bandwidth_enabled(void) | |
390 | { | |
391 | return sysctl_sched_rt_runtime >= 0; | |
392 | } | |
393 | ||
394 | /* Real-Time classes' related field in a runqueue: */ | |
395 | struct rt_rq { | |
396 | struct rt_prio_array active; | |
c82513e5 | 397 | unsigned int rt_nr_running; |
029632fb PZ |
398 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
399 | struct { | |
400 | int curr; /* highest queued rt task prio */ | |
401 | #ifdef CONFIG_SMP | |
402 | int next; /* next highest */ | |
403 | #endif | |
404 | } highest_prio; | |
405 | #endif | |
406 | #ifdef CONFIG_SMP | |
407 | unsigned long rt_nr_migratory; | |
408 | unsigned long rt_nr_total; | |
409 | int overloaded; | |
410 | struct plist_head pushable_tasks; | |
411 | #endif | |
f4ebcbc0 KT |
412 | int rt_queued; |
413 | ||
029632fb PZ |
414 | int rt_throttled; |
415 | u64 rt_time; | |
416 | u64 rt_runtime; | |
417 | /* Nests inside the rq lock: */ | |
418 | raw_spinlock_t rt_runtime_lock; | |
419 | ||
420 | #ifdef CONFIG_RT_GROUP_SCHED | |
421 | unsigned long rt_nr_boosted; | |
422 | ||
423 | struct rq *rq; | |
029632fb PZ |
424 | struct task_group *tg; |
425 | #endif | |
426 | }; | |
427 | ||
aab03e05 DF |
428 | /* Deadline class' related fields in a runqueue */ |
429 | struct dl_rq { | |
430 | /* runqueue is an rbtree, ordered by deadline */ | |
431 | struct rb_root rb_root; | |
432 | struct rb_node *rb_leftmost; | |
433 | ||
434 | unsigned long dl_nr_running; | |
1baca4ce JL |
435 | |
436 | #ifdef CONFIG_SMP | |
437 | /* | |
438 | * Deadline values of the currently executing and the | |
439 | * earliest ready task on this rq. Caching these facilitates | |
440 | * the decision wether or not a ready but not running task | |
441 | * should migrate somewhere else. | |
442 | */ | |
443 | struct { | |
444 | u64 curr; | |
445 | u64 next; | |
446 | } earliest_dl; | |
447 | ||
448 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
449 | int overloaded; |
450 | ||
451 | /* | |
452 | * Tasks on this rq that can be pushed away. They are kept in | |
453 | * an rb-tree, ordered by tasks' deadlines, with caching | |
454 | * of the leftmost (earliest deadline) element. | |
455 | */ | |
456 | struct rb_root pushable_dl_tasks_root; | |
457 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
458 | #else |
459 | struct dl_bw dl_bw; | |
1baca4ce | 460 | #endif |
aab03e05 DF |
461 | }; |
462 | ||
029632fb PZ |
463 | #ifdef CONFIG_SMP |
464 | ||
465 | /* | |
466 | * We add the notion of a root-domain which will be used to define per-domain | |
467 | * variables. Each exclusive cpuset essentially defines an island domain by | |
468 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
469 | * exclusive cpuset is created, we also create and attach a new root-domain | |
470 | * object. | |
471 | * | |
472 | */ | |
473 | struct root_domain { | |
474 | atomic_t refcount; | |
475 | atomic_t rto_count; | |
476 | struct rcu_head rcu; | |
477 | cpumask_var_t span; | |
478 | cpumask_var_t online; | |
479 | ||
4486edd1 TC |
480 | /* Indicate more than one runnable task for any CPU */ |
481 | bool overload; | |
482 | ||
1baca4ce JL |
483 | /* |
484 | * The bit corresponding to a CPU gets set here if such CPU has more | |
485 | * than one runnable -deadline task (as it is below for RT tasks). | |
486 | */ | |
487 | cpumask_var_t dlo_mask; | |
488 | atomic_t dlo_count; | |
332ac17e | 489 | struct dl_bw dl_bw; |
6bfd6d72 | 490 | struct cpudl cpudl; |
1baca4ce | 491 | |
029632fb PZ |
492 | /* |
493 | * The "RT overload" flag: it gets set if a CPU has more than | |
494 | * one runnable RT task. | |
495 | */ | |
496 | cpumask_var_t rto_mask; | |
497 | struct cpupri cpupri; | |
498 | }; | |
499 | ||
500 | extern struct root_domain def_root_domain; | |
501 | ||
502 | #endif /* CONFIG_SMP */ | |
503 | ||
504 | /* | |
505 | * This is the main, per-CPU runqueue data structure. | |
506 | * | |
507 | * Locking rule: those places that want to lock multiple runqueues | |
508 | * (such as the load balancing or the thread migration code), lock | |
509 | * acquire operations must be ordered by ascending &runqueue. | |
510 | */ | |
511 | struct rq { | |
512 | /* runqueue lock: */ | |
513 | raw_spinlock_t lock; | |
514 | ||
515 | /* | |
516 | * nr_running and cpu_load should be in the same cacheline because | |
517 | * remote CPUs use both these fields when doing load calculation. | |
518 | */ | |
c82513e5 | 519 | unsigned int nr_running; |
0ec8aa00 PZ |
520 | #ifdef CONFIG_NUMA_BALANCING |
521 | unsigned int nr_numa_running; | |
522 | unsigned int nr_preferred_running; | |
523 | #endif | |
029632fb PZ |
524 | #define CPU_LOAD_IDX_MAX 5 |
525 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
526 | unsigned long last_load_update_tick; | |
3451d024 | 527 | #ifdef CONFIG_NO_HZ_COMMON |
029632fb | 528 | u64 nohz_stamp; |
1c792db7 | 529 | unsigned long nohz_flags; |
265f22a9 FW |
530 | #endif |
531 | #ifdef CONFIG_NO_HZ_FULL | |
532 | unsigned long last_sched_tick; | |
029632fb PZ |
533 | #endif |
534 | int skip_clock_update; | |
535 | ||
536 | /* capture load from *all* tasks on this cpu: */ | |
537 | struct load_weight load; | |
538 | unsigned long nr_load_updates; | |
539 | u64 nr_switches; | |
540 | ||
541 | struct cfs_rq cfs; | |
542 | struct rt_rq rt; | |
aab03e05 | 543 | struct dl_rq dl; |
029632fb PZ |
544 | |
545 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
546 | /* list of leaf cfs_rq on this cpu: */ | |
547 | struct list_head leaf_cfs_rq_list; | |
f5f9739d DE |
548 | |
549 | struct sched_avg avg; | |
a35b6466 PZ |
550 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
551 | ||
029632fb PZ |
552 | /* |
553 | * This is part of a global counter where only the total sum | |
554 | * over all CPUs matters. A task can increase this counter on | |
555 | * one CPU and if it got migrated afterwards it may decrease | |
556 | * it on another CPU. Always updated under the runqueue lock: | |
557 | */ | |
558 | unsigned long nr_uninterruptible; | |
559 | ||
560 | struct task_struct *curr, *idle, *stop; | |
561 | unsigned long next_balance; | |
562 | struct mm_struct *prev_mm; | |
563 | ||
564 | u64 clock; | |
565 | u64 clock_task; | |
566 | ||
567 | atomic_t nr_iowait; | |
568 | ||
569 | #ifdef CONFIG_SMP | |
570 | struct root_domain *rd; | |
571 | struct sched_domain *sd; | |
572 | ||
ced549fa | 573 | unsigned long cpu_capacity; |
029632fb PZ |
574 | |
575 | unsigned char idle_balance; | |
576 | /* For active balancing */ | |
577 | int post_schedule; | |
578 | int active_balance; | |
579 | int push_cpu; | |
580 | struct cpu_stop_work active_balance_work; | |
581 | /* cpu of this runqueue: */ | |
582 | int cpu; | |
583 | int online; | |
584 | ||
367456c7 PZ |
585 | struct list_head cfs_tasks; |
586 | ||
029632fb PZ |
587 | u64 rt_avg; |
588 | u64 age_stamp; | |
589 | u64 idle_stamp; | |
590 | u64 avg_idle; | |
9bd721c5 JL |
591 | |
592 | /* This is used to determine avg_idle's max value */ | |
593 | u64 max_idle_balance_cost; | |
029632fb PZ |
594 | #endif |
595 | ||
596 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
597 | u64 prev_irq_time; | |
598 | #endif | |
599 | #ifdef CONFIG_PARAVIRT | |
600 | u64 prev_steal_time; | |
601 | #endif | |
602 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
603 | u64 prev_steal_time_rq; | |
604 | #endif | |
605 | ||
606 | /* calc_load related fields */ | |
607 | unsigned long calc_load_update; | |
608 | long calc_load_active; | |
609 | ||
610 | #ifdef CONFIG_SCHED_HRTICK | |
611 | #ifdef CONFIG_SMP | |
612 | int hrtick_csd_pending; | |
613 | struct call_single_data hrtick_csd; | |
614 | #endif | |
615 | struct hrtimer hrtick_timer; | |
616 | #endif | |
617 | ||
618 | #ifdef CONFIG_SCHEDSTATS | |
619 | /* latency stats */ | |
620 | struct sched_info rq_sched_info; | |
621 | unsigned long long rq_cpu_time; | |
622 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
623 | ||
624 | /* sys_sched_yield() stats */ | |
625 | unsigned int yld_count; | |
626 | ||
627 | /* schedule() stats */ | |
029632fb PZ |
628 | unsigned int sched_count; |
629 | unsigned int sched_goidle; | |
630 | ||
631 | /* try_to_wake_up() stats */ | |
632 | unsigned int ttwu_count; | |
633 | unsigned int ttwu_local; | |
634 | #endif | |
635 | ||
636 | #ifdef CONFIG_SMP | |
637 | struct llist_head wake_list; | |
638 | #endif | |
639 | }; | |
640 | ||
641 | static inline int cpu_of(struct rq *rq) | |
642 | { | |
643 | #ifdef CONFIG_SMP | |
644 | return rq->cpu; | |
645 | #else | |
646 | return 0; | |
647 | #endif | |
648 | } | |
649 | ||
650 | DECLARE_PER_CPU(struct rq, runqueues); | |
651 | ||
518cd623 PZ |
652 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
653 | #define this_rq() (&__get_cpu_var(runqueues)) | |
654 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
655 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
656 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) | |
657 | ||
78becc27 FW |
658 | static inline u64 rq_clock(struct rq *rq) |
659 | { | |
660 | return rq->clock; | |
661 | } | |
662 | ||
663 | static inline u64 rq_clock_task(struct rq *rq) | |
664 | { | |
665 | return rq->clock_task; | |
666 | } | |
667 | ||
f809ca9a | 668 | #ifdef CONFIG_NUMA_BALANCING |
0ec8aa00 | 669 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 670 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 671 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
672 | #endif /* CONFIG_NUMA_BALANCING */ |
673 | ||
518cd623 PZ |
674 | #ifdef CONFIG_SMP |
675 | ||
e3baac47 PZ |
676 | extern void sched_ttwu_pending(void); |
677 | ||
029632fb PZ |
678 | #define rcu_dereference_check_sched_domain(p) \ |
679 | rcu_dereference_check((p), \ | |
680 | lockdep_is_held(&sched_domains_mutex)) | |
681 | ||
682 | /* | |
683 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
684 | * See detach_destroy_domains: synchronize_sched for details. | |
685 | * | |
686 | * The domain tree of any CPU may only be accessed from within | |
687 | * preempt-disabled sections. | |
688 | */ | |
689 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
690 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
691 | __sd; __sd = __sd->parent) | |
029632fb | 692 | |
77e81365 SS |
693 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
694 | ||
518cd623 PZ |
695 | /** |
696 | * highest_flag_domain - Return highest sched_domain containing flag. | |
697 | * @cpu: The cpu whose highest level of sched domain is to | |
698 | * be returned. | |
699 | * @flag: The flag to check for the highest sched_domain | |
700 | * for the given cpu. | |
701 | * | |
702 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
703 | */ | |
704 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
705 | { | |
706 | struct sched_domain *sd, *hsd = NULL; | |
707 | ||
708 | for_each_domain(cpu, sd) { | |
709 | if (!(sd->flags & flag)) | |
710 | break; | |
711 | hsd = sd; | |
712 | } | |
713 | ||
714 | return hsd; | |
715 | } | |
716 | ||
fb13c7ee MG |
717 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
718 | { | |
719 | struct sched_domain *sd; | |
720 | ||
721 | for_each_domain(cpu, sd) { | |
722 | if (sd->flags & flag) | |
723 | break; | |
724 | } | |
725 | ||
726 | return sd; | |
727 | } | |
728 | ||
518cd623 | 729 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 730 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 731 | DECLARE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 732 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
733 | DECLARE_PER_CPU(struct sched_domain *, sd_busy); |
734 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 | 735 | |
63b2ca30 | 736 | struct sched_group_capacity { |
5e6521ea LZ |
737 | atomic_t ref; |
738 | /* | |
63b2ca30 NP |
739 | * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity |
740 | * for a single CPU. | |
5e6521ea | 741 | */ |
63b2ca30 | 742 | unsigned int capacity, capacity_orig; |
5e6521ea | 743 | unsigned long next_update; |
63b2ca30 | 744 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea LZ |
745 | /* |
746 | * Number of busy cpus in this group. | |
747 | */ | |
748 | atomic_t nr_busy_cpus; | |
749 | ||
750 | unsigned long cpumask[0]; /* iteration mask */ | |
751 | }; | |
752 | ||
753 | struct sched_group { | |
754 | struct sched_group *next; /* Must be a circular list */ | |
755 | atomic_t ref; | |
756 | ||
757 | unsigned int group_weight; | |
63b2ca30 | 758 | struct sched_group_capacity *sgc; |
5e6521ea LZ |
759 | |
760 | /* | |
761 | * The CPUs this group covers. | |
762 | * | |
763 | * NOTE: this field is variable length. (Allocated dynamically | |
764 | * by attaching extra space to the end of the structure, | |
765 | * depending on how many CPUs the kernel has booted up with) | |
766 | */ | |
767 | unsigned long cpumask[0]; | |
768 | }; | |
769 | ||
770 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) | |
771 | { | |
772 | return to_cpumask(sg->cpumask); | |
773 | } | |
774 | ||
775 | /* | |
776 | * cpumask masking which cpus in the group are allowed to iterate up the domain | |
777 | * tree. | |
778 | */ | |
779 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) | |
780 | { | |
63b2ca30 | 781 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
782 | } |
783 | ||
784 | /** | |
785 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
786 | * @group: The group whose first cpu is to be returned. | |
787 | */ | |
788 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
789 | { | |
790 | return cpumask_first(sched_group_cpus(group)); | |
791 | } | |
792 | ||
c1174876 PZ |
793 | extern int group_balance_cpu(struct sched_group *sg); |
794 | ||
e3baac47 PZ |
795 | #else |
796 | ||
797 | static inline void sched_ttwu_pending(void) { } | |
798 | ||
518cd623 | 799 | #endif /* CONFIG_SMP */ |
029632fb | 800 | |
391e43da PZ |
801 | #include "stats.h" |
802 | #include "auto_group.h" | |
029632fb PZ |
803 | |
804 | #ifdef CONFIG_CGROUP_SCHED | |
805 | ||
806 | /* | |
807 | * Return the group to which this tasks belongs. | |
808 | * | |
8af01f56 TH |
809 | * We cannot use task_css() and friends because the cgroup subsystem |
810 | * changes that value before the cgroup_subsys::attach() method is called, | |
811 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
812 | * |
813 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
814 | * core changes this before calling sched_move_task(). | |
815 | * | |
816 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
817 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
818 | */ |
819 | static inline struct task_group *task_group(struct task_struct *p) | |
820 | { | |
8323f26c | 821 | return p->sched_task_group; |
029632fb PZ |
822 | } |
823 | ||
824 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
825 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
826 | { | |
827 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
828 | struct task_group *tg = task_group(p); | |
829 | #endif | |
830 | ||
831 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
832 | p->se.cfs_rq = tg->cfs_rq[cpu]; | |
833 | p->se.parent = tg->se[cpu]; | |
834 | #endif | |
835 | ||
836 | #ifdef CONFIG_RT_GROUP_SCHED | |
837 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
838 | p->rt.parent = tg->rt_se[cpu]; | |
839 | #endif | |
840 | } | |
841 | ||
842 | #else /* CONFIG_CGROUP_SCHED */ | |
843 | ||
844 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
845 | static inline struct task_group *task_group(struct task_struct *p) | |
846 | { | |
847 | return NULL; | |
848 | } | |
849 | ||
850 | #endif /* CONFIG_CGROUP_SCHED */ | |
851 | ||
852 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
853 | { | |
854 | set_task_rq(p, cpu); | |
855 | #ifdef CONFIG_SMP | |
856 | /* | |
857 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
858 | * successfuly executed on another CPU. We must ensure that updates of | |
859 | * per-task data have been completed by this moment. | |
860 | */ | |
861 | smp_wmb(); | |
862 | task_thread_info(p)->cpu = cpu; | |
ac66f547 | 863 | p->wake_cpu = cpu; |
029632fb PZ |
864 | #endif |
865 | } | |
866 | ||
867 | /* | |
868 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
869 | */ | |
870 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 871 | # include <linux/static_key.h> |
029632fb PZ |
872 | # define const_debug __read_mostly |
873 | #else | |
874 | # define const_debug const | |
875 | #endif | |
876 | ||
877 | extern const_debug unsigned int sysctl_sched_features; | |
878 | ||
879 | #define SCHED_FEAT(name, enabled) \ | |
880 | __SCHED_FEAT_##name , | |
881 | ||
882 | enum { | |
391e43da | 883 | #include "features.h" |
f8b6d1cc | 884 | __SCHED_FEAT_NR, |
029632fb PZ |
885 | }; |
886 | ||
887 | #undef SCHED_FEAT | |
888 | ||
f8b6d1cc | 889 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
c5905afb | 890 | static __always_inline bool static_branch__true(struct static_key *key) |
f8b6d1cc | 891 | { |
c5905afb | 892 | return static_key_true(key); /* Not out of line branch. */ |
f8b6d1cc PZ |
893 | } |
894 | ||
c5905afb | 895 | static __always_inline bool static_branch__false(struct static_key *key) |
f8b6d1cc | 896 | { |
c5905afb | 897 | return static_key_false(key); /* Out of line branch. */ |
f8b6d1cc PZ |
898 | } |
899 | ||
900 | #define SCHED_FEAT(name, enabled) \ | |
c5905afb | 901 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc PZ |
902 | { \ |
903 | return static_branch__##enabled(key); \ | |
904 | } | |
905 | ||
906 | #include "features.h" | |
907 | ||
908 | #undef SCHED_FEAT | |
909 | ||
c5905afb | 910 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
911 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
912 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 913 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 914 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 915 | |
cbee9f88 PZ |
916 | #ifdef CONFIG_NUMA_BALANCING |
917 | #define sched_feat_numa(x) sched_feat(x) | |
3105b86a MG |
918 | #ifdef CONFIG_SCHED_DEBUG |
919 | #define numabalancing_enabled sched_feat_numa(NUMA) | |
920 | #else | |
921 | extern bool numabalancing_enabled; | |
922 | #endif /* CONFIG_SCHED_DEBUG */ | |
cbee9f88 PZ |
923 | #else |
924 | #define sched_feat_numa(x) (0) | |
3105b86a MG |
925 | #define numabalancing_enabled (0) |
926 | #endif /* CONFIG_NUMA_BALANCING */ | |
cbee9f88 | 927 | |
029632fb PZ |
928 | static inline u64 global_rt_period(void) |
929 | { | |
930 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
931 | } | |
932 | ||
933 | static inline u64 global_rt_runtime(void) | |
934 | { | |
935 | if (sysctl_sched_rt_runtime < 0) | |
936 | return RUNTIME_INF; | |
937 | ||
938 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
939 | } | |
940 | ||
029632fb PZ |
941 | static inline int task_current(struct rq *rq, struct task_struct *p) |
942 | { | |
943 | return rq->curr == p; | |
944 | } | |
945 | ||
946 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
947 | { | |
948 | #ifdef CONFIG_SMP | |
949 | return p->on_cpu; | |
950 | #else | |
951 | return task_current(rq, p); | |
952 | #endif | |
953 | } | |
954 | ||
955 | ||
956 | #ifndef prepare_arch_switch | |
957 | # define prepare_arch_switch(next) do { } while (0) | |
958 | #endif | |
959 | #ifndef finish_arch_switch | |
960 | # define finish_arch_switch(prev) do { } while (0) | |
961 | #endif | |
01f23e16 CM |
962 | #ifndef finish_arch_post_lock_switch |
963 | # define finish_arch_post_lock_switch() do { } while (0) | |
964 | #endif | |
029632fb PZ |
965 | |
966 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
967 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
968 | { | |
969 | #ifdef CONFIG_SMP | |
970 | /* | |
971 | * We can optimise this out completely for !SMP, because the | |
972 | * SMP rebalancing from interrupt is the only thing that cares | |
973 | * here. | |
974 | */ | |
975 | next->on_cpu = 1; | |
976 | #endif | |
977 | } | |
978 | ||
979 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
980 | { | |
981 | #ifdef CONFIG_SMP | |
982 | /* | |
983 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
984 | * We must ensure this doesn't happen until the switch is completely | |
985 | * finished. | |
986 | */ | |
987 | smp_wmb(); | |
988 | prev->on_cpu = 0; | |
989 | #endif | |
990 | #ifdef CONFIG_DEBUG_SPINLOCK | |
991 | /* this is a valid case when another task releases the spinlock */ | |
992 | rq->lock.owner = current; | |
993 | #endif | |
994 | /* | |
995 | * If we are tracking spinlock dependencies then we have to | |
996 | * fix up the runqueue lock - which gets 'carried over' from | |
997 | * prev into current: | |
998 | */ | |
999 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1000 | ||
1001 | raw_spin_unlock_irq(&rq->lock); | |
1002 | } | |
1003 | ||
1004 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1005 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
1006 | { | |
1007 | #ifdef CONFIG_SMP | |
1008 | /* | |
1009 | * We can optimise this out completely for !SMP, because the | |
1010 | * SMP rebalancing from interrupt is the only thing that cares | |
1011 | * here. | |
1012 | */ | |
1013 | next->on_cpu = 1; | |
1014 | #endif | |
029632fb | 1015 | raw_spin_unlock(&rq->lock); |
029632fb PZ |
1016 | } |
1017 | ||
1018 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1019 | { | |
1020 | #ifdef CONFIG_SMP | |
1021 | /* | |
1022 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1023 | * We must ensure this doesn't happen until the switch is completely | |
1024 | * finished. | |
1025 | */ | |
1026 | smp_wmb(); | |
1027 | prev->on_cpu = 0; | |
1028 | #endif | |
029632fb | 1029 | local_irq_enable(); |
029632fb PZ |
1030 | } |
1031 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1032 | ||
b13095f0 LZ |
1033 | /* |
1034 | * wake flags | |
1035 | */ | |
1036 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1037 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1038 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1039 | ||
029632fb PZ |
1040 | /* |
1041 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1042 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1043 | * each task makes to its run queue's load is weighted according to its | |
1044 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1045 | * scaled version of the new time slice allocation that they receive on time | |
1046 | * slice expiry etc. | |
1047 | */ | |
1048 | ||
1049 | #define WEIGHT_IDLEPRIO 3 | |
1050 | #define WMULT_IDLEPRIO 1431655765 | |
1051 | ||
1052 | /* | |
1053 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1054 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1055 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1056 | * that remained on nice 0. | |
1057 | * | |
1058 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1059 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
1060 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
1061 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1062 | * the relative distance between them is ~25%.) | |
1063 | */ | |
1064 | static const int prio_to_weight[40] = { | |
1065 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
1066 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1067 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1068 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1069 | /* 0 */ 1024, 820, 655, 526, 423, | |
1070 | /* 5 */ 335, 272, 215, 172, 137, | |
1071 | /* 10 */ 110, 87, 70, 56, 45, | |
1072 | /* 15 */ 36, 29, 23, 18, 15, | |
1073 | }; | |
1074 | ||
1075 | /* | |
1076 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1077 | * | |
1078 | * In cases where the weight does not change often, we can use the | |
1079 | * precalculated inverse to speed up arithmetics by turning divisions | |
1080 | * into multiplications: | |
1081 | */ | |
1082 | static const u32 prio_to_wmult[40] = { | |
1083 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
1084 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1085 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1086 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1087 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1088 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1089 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1090 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
1091 | }; | |
1092 | ||
c82ba9fa LZ |
1093 | #define ENQUEUE_WAKEUP 1 |
1094 | #define ENQUEUE_HEAD 2 | |
1095 | #ifdef CONFIG_SMP | |
1096 | #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */ | |
1097 | #else | |
1098 | #define ENQUEUE_WAKING 0 | |
1099 | #endif | |
aab03e05 | 1100 | #define ENQUEUE_REPLENISH 8 |
c82ba9fa LZ |
1101 | |
1102 | #define DEQUEUE_SLEEP 1 | |
1103 | ||
37e117c0 PZ |
1104 | #define RETRY_TASK ((void *)-1UL) |
1105 | ||
c82ba9fa LZ |
1106 | struct sched_class { |
1107 | const struct sched_class *next; | |
1108 | ||
1109 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1110 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1111 | void (*yield_task) (struct rq *rq); | |
1112 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1113 | ||
1114 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1115 | ||
606dba2e PZ |
1116 | /* |
1117 | * It is the responsibility of the pick_next_task() method that will | |
1118 | * return the next task to call put_prev_task() on the @prev task or | |
1119 | * something equivalent. | |
37e117c0 PZ |
1120 | * |
1121 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1122 | * tasks. | |
606dba2e PZ |
1123 | */ |
1124 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
1125 | struct task_struct *prev); | |
c82ba9fa LZ |
1126 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1127 | ||
1128 | #ifdef CONFIG_SMP | |
ac66f547 | 1129 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
c82ba9fa LZ |
1130 | void (*migrate_task_rq)(struct task_struct *p, int next_cpu); |
1131 | ||
c82ba9fa LZ |
1132 | void (*post_schedule) (struct rq *this_rq); |
1133 | void (*task_waking) (struct task_struct *task); | |
1134 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); | |
1135 | ||
1136 | void (*set_cpus_allowed)(struct task_struct *p, | |
1137 | const struct cpumask *newmask); | |
1138 | ||
1139 | void (*rq_online)(struct rq *rq); | |
1140 | void (*rq_offline)(struct rq *rq); | |
1141 | #endif | |
1142 | ||
1143 | void (*set_curr_task) (struct rq *rq); | |
1144 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1145 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1146 | void (*task_dead) (struct task_struct *p); |
c82ba9fa LZ |
1147 | |
1148 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); | |
1149 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1150 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1151 | int oldprio); | |
1152 | ||
1153 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1154 | struct task_struct *task); | |
1155 | ||
1156 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1157 | void (*task_move_group) (struct task_struct *p, int on_rq); | |
1158 | #endif | |
1159 | }; | |
029632fb | 1160 | |
3f1d2a31 PZ |
1161 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1162 | { | |
1163 | prev->sched_class->put_prev_task(rq, prev); | |
1164 | } | |
1165 | ||
029632fb PZ |
1166 | #define sched_class_highest (&stop_sched_class) |
1167 | #define for_each_class(class) \ | |
1168 | for (class = sched_class_highest; class; class = class->next) | |
1169 | ||
1170 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1171 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1172 | extern const struct sched_class rt_sched_class; |
1173 | extern const struct sched_class fair_sched_class; | |
1174 | extern const struct sched_class idle_sched_class; | |
1175 | ||
1176 | ||
1177 | #ifdef CONFIG_SMP | |
1178 | ||
63b2ca30 | 1179 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1180 | |
7caff66f | 1181 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1182 | |
642dbc39 VG |
1183 | extern void idle_enter_fair(struct rq *this_rq); |
1184 | extern void idle_exit_fair(struct rq *this_rq); | |
642dbc39 | 1185 | |
dc877341 PZ |
1186 | #else |
1187 | ||
1188 | static inline void idle_enter_fair(struct rq *rq) { } | |
1189 | static inline void idle_exit_fair(struct rq *rq) { } | |
1190 | ||
029632fb PZ |
1191 | #endif |
1192 | ||
1193 | extern void sysrq_sched_debug_show(void); | |
1194 | extern void sched_init_granularity(void); | |
1195 | extern void update_max_interval(void); | |
1baca4ce JL |
1196 | |
1197 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1198 | extern void init_sched_rt_class(void); |
1199 | extern void init_sched_fair_class(void); | |
332ac17e | 1200 | extern void init_sched_dl_class(void); |
029632fb | 1201 | |
8875125e | 1202 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1203 | extern void resched_cpu(int cpu); |
1204 | ||
1205 | extern struct rt_bandwidth def_rt_bandwidth; | |
1206 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1207 | ||
332ac17e DF |
1208 | extern struct dl_bandwidth def_dl_bandwidth; |
1209 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 DF |
1210 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
1211 | ||
332ac17e DF |
1212 | unsigned long to_ratio(u64 period, u64 runtime); |
1213 | ||
556061b0 | 1214 | extern void update_idle_cpu_load(struct rq *this_rq); |
029632fb | 1215 | |
a75cdaa9 AS |
1216 | extern void init_task_runnable_average(struct task_struct *p); |
1217 | ||
72465447 | 1218 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1219 | { |
72465447 KT |
1220 | unsigned prev_nr = rq->nr_running; |
1221 | ||
1222 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1223 | |
72465447 | 1224 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1225 | #ifdef CONFIG_SMP |
1226 | if (!rq->rd->overload) | |
1227 | rq->rd->overload = true; | |
1228 | #endif | |
1229 | ||
1230 | #ifdef CONFIG_NO_HZ_FULL | |
9f3660c2 | 1231 | if (tick_nohz_full_cpu(rq->cpu)) { |
3882ec64 FW |
1232 | /* |
1233 | * Tick is needed if more than one task runs on a CPU. | |
1234 | * Send the target an IPI to kick it out of nohz mode. | |
1235 | * | |
1236 | * We assume that IPI implies full memory barrier and the | |
1237 | * new value of rq->nr_running is visible on reception | |
1238 | * from the target. | |
1239 | */ | |
fd2ac4f4 | 1240 | tick_nohz_full_kick_cpu(rq->cpu); |
9f3660c2 | 1241 | } |
9f3660c2 | 1242 | #endif |
4486edd1 | 1243 | } |
029632fb PZ |
1244 | } |
1245 | ||
72465447 | 1246 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1247 | { |
72465447 | 1248 | rq->nr_running -= count; |
029632fb PZ |
1249 | } |
1250 | ||
265f22a9 FW |
1251 | static inline void rq_last_tick_reset(struct rq *rq) |
1252 | { | |
1253 | #ifdef CONFIG_NO_HZ_FULL | |
1254 | rq->last_sched_tick = jiffies; | |
1255 | #endif | |
1256 | } | |
1257 | ||
029632fb PZ |
1258 | extern void update_rq_clock(struct rq *rq); |
1259 | ||
1260 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1261 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1262 | ||
1263 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1264 | ||
1265 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1266 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1267 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1268 | ||
1269 | static inline u64 sched_avg_period(void) | |
1270 | { | |
1271 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1272 | } | |
1273 | ||
029632fb PZ |
1274 | #ifdef CONFIG_SCHED_HRTICK |
1275 | ||
1276 | /* | |
1277 | * Use hrtick when: | |
1278 | * - enabled by features | |
1279 | * - hrtimer is actually high res | |
1280 | */ | |
1281 | static inline int hrtick_enabled(struct rq *rq) | |
1282 | { | |
1283 | if (!sched_feat(HRTICK)) | |
1284 | return 0; | |
1285 | if (!cpu_active(cpu_of(rq))) | |
1286 | return 0; | |
1287 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1288 | } | |
1289 | ||
1290 | void hrtick_start(struct rq *rq, u64 delay); | |
1291 | ||
b39e66ea MG |
1292 | #else |
1293 | ||
1294 | static inline int hrtick_enabled(struct rq *rq) | |
1295 | { | |
1296 | return 0; | |
1297 | } | |
1298 | ||
029632fb PZ |
1299 | #endif /* CONFIG_SCHED_HRTICK */ |
1300 | ||
1301 | #ifdef CONFIG_SMP | |
1302 | extern void sched_avg_update(struct rq *rq); | |
1303 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1304 | { | |
1305 | rq->rt_avg += rt_delta; | |
1306 | sched_avg_update(rq); | |
1307 | } | |
1308 | #else | |
1309 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1310 | static inline void sched_avg_update(struct rq *rq) { } | |
1311 | #endif | |
1312 | ||
1313 | extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period); | |
1314 | ||
1315 | #ifdef CONFIG_SMP | |
1316 | #ifdef CONFIG_PREEMPT | |
1317 | ||
1318 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1319 | ||
1320 | /* | |
1321 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1322 | * way at the expense of forcing extra atomic operations in all | |
1323 | * invocations. This assures that the double_lock is acquired using the | |
1324 | * same underlying policy as the spinlock_t on this architecture, which | |
1325 | * reduces latency compared to the unfair variant below. However, it | |
1326 | * also adds more overhead and therefore may reduce throughput. | |
1327 | */ | |
1328 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1329 | __releases(this_rq->lock) | |
1330 | __acquires(busiest->lock) | |
1331 | __acquires(this_rq->lock) | |
1332 | { | |
1333 | raw_spin_unlock(&this_rq->lock); | |
1334 | double_rq_lock(this_rq, busiest); | |
1335 | ||
1336 | return 1; | |
1337 | } | |
1338 | ||
1339 | #else | |
1340 | /* | |
1341 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1342 | * latency by eliminating extra atomic operations when the locks are | |
1343 | * already in proper order on entry. This favors lower cpu-ids and will | |
1344 | * grant the double lock to lower cpus over higher ids under contention, | |
1345 | * regardless of entry order into the function. | |
1346 | */ | |
1347 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1348 | __releases(this_rq->lock) | |
1349 | __acquires(busiest->lock) | |
1350 | __acquires(this_rq->lock) | |
1351 | { | |
1352 | int ret = 0; | |
1353 | ||
1354 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1355 | if (busiest < this_rq) { | |
1356 | raw_spin_unlock(&this_rq->lock); | |
1357 | raw_spin_lock(&busiest->lock); | |
1358 | raw_spin_lock_nested(&this_rq->lock, | |
1359 | SINGLE_DEPTH_NESTING); | |
1360 | ret = 1; | |
1361 | } else | |
1362 | raw_spin_lock_nested(&busiest->lock, | |
1363 | SINGLE_DEPTH_NESTING); | |
1364 | } | |
1365 | return ret; | |
1366 | } | |
1367 | ||
1368 | #endif /* CONFIG_PREEMPT */ | |
1369 | ||
1370 | /* | |
1371 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1372 | */ | |
1373 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1374 | { | |
1375 | if (unlikely(!irqs_disabled())) { | |
1376 | /* printk() doesn't work good under rq->lock */ | |
1377 | raw_spin_unlock(&this_rq->lock); | |
1378 | BUG_ON(1); | |
1379 | } | |
1380 | ||
1381 | return _double_lock_balance(this_rq, busiest); | |
1382 | } | |
1383 | ||
1384 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1385 | __releases(busiest->lock) | |
1386 | { | |
1387 | raw_spin_unlock(&busiest->lock); | |
1388 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1389 | } | |
1390 | ||
74602315 PZ |
1391 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1392 | { | |
1393 | if (l1 > l2) | |
1394 | swap(l1, l2); | |
1395 | ||
1396 | spin_lock(l1); | |
1397 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1398 | } | |
1399 | ||
60e69eed MG |
1400 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1401 | { | |
1402 | if (l1 > l2) | |
1403 | swap(l1, l2); | |
1404 | ||
1405 | spin_lock_irq(l1); | |
1406 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1407 | } | |
1408 | ||
74602315 PZ |
1409 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1410 | { | |
1411 | if (l1 > l2) | |
1412 | swap(l1, l2); | |
1413 | ||
1414 | raw_spin_lock(l1); | |
1415 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1416 | } | |
1417 | ||
029632fb PZ |
1418 | /* |
1419 | * double_rq_lock - safely lock two runqueues | |
1420 | * | |
1421 | * Note this does not disable interrupts like task_rq_lock, | |
1422 | * you need to do so manually before calling. | |
1423 | */ | |
1424 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1425 | __acquires(rq1->lock) | |
1426 | __acquires(rq2->lock) | |
1427 | { | |
1428 | BUG_ON(!irqs_disabled()); | |
1429 | if (rq1 == rq2) { | |
1430 | raw_spin_lock(&rq1->lock); | |
1431 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1432 | } else { | |
1433 | if (rq1 < rq2) { | |
1434 | raw_spin_lock(&rq1->lock); | |
1435 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1436 | } else { | |
1437 | raw_spin_lock(&rq2->lock); | |
1438 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1439 | } | |
1440 | } | |
1441 | } | |
1442 | ||
1443 | /* | |
1444 | * double_rq_unlock - safely unlock two runqueues | |
1445 | * | |
1446 | * Note this does not restore interrupts like task_rq_unlock, | |
1447 | * you need to do so manually after calling. | |
1448 | */ | |
1449 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1450 | __releases(rq1->lock) | |
1451 | __releases(rq2->lock) | |
1452 | { | |
1453 | raw_spin_unlock(&rq1->lock); | |
1454 | if (rq1 != rq2) | |
1455 | raw_spin_unlock(&rq2->lock); | |
1456 | else | |
1457 | __release(rq2->lock); | |
1458 | } | |
1459 | ||
1460 | #else /* CONFIG_SMP */ | |
1461 | ||
1462 | /* | |
1463 | * double_rq_lock - safely lock two runqueues | |
1464 | * | |
1465 | * Note this does not disable interrupts like task_rq_lock, | |
1466 | * you need to do so manually before calling. | |
1467 | */ | |
1468 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1469 | __acquires(rq1->lock) | |
1470 | __acquires(rq2->lock) | |
1471 | { | |
1472 | BUG_ON(!irqs_disabled()); | |
1473 | BUG_ON(rq1 != rq2); | |
1474 | raw_spin_lock(&rq1->lock); | |
1475 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1476 | } | |
1477 | ||
1478 | /* | |
1479 | * double_rq_unlock - safely unlock two runqueues | |
1480 | * | |
1481 | * Note this does not restore interrupts like task_rq_unlock, | |
1482 | * you need to do so manually after calling. | |
1483 | */ | |
1484 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1485 | __releases(rq1->lock) | |
1486 | __releases(rq2->lock) | |
1487 | { | |
1488 | BUG_ON(rq1 != rq2); | |
1489 | raw_spin_unlock(&rq1->lock); | |
1490 | __release(rq2->lock); | |
1491 | } | |
1492 | ||
1493 | #endif | |
1494 | ||
1495 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1496 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
1497 | extern void print_cfs_stats(struct seq_file *m, int cpu); | |
1498 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
1499 | ||
1500 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
1501 | extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq); | |
aab03e05 | 1502 | extern void init_dl_rq(struct dl_rq *dl_rq, struct rq *rq); |
029632fb | 1503 | |
1ee14e6c BS |
1504 | extern void cfs_bandwidth_usage_inc(void); |
1505 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1506 | |
3451d024 | 1507 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1508 | enum rq_nohz_flag_bits { |
1509 | NOHZ_TICK_STOPPED, | |
1510 | NOHZ_BALANCE_KICK, | |
1511 | }; | |
1512 | ||
1513 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
1514 | #endif | |
73fbec60 FW |
1515 | |
1516 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
1517 | ||
1518 | DECLARE_PER_CPU(u64, cpu_hardirq_time); | |
1519 | DECLARE_PER_CPU(u64, cpu_softirq_time); | |
1520 | ||
1521 | #ifndef CONFIG_64BIT | |
1522 | DECLARE_PER_CPU(seqcount_t, irq_time_seq); | |
1523 | ||
1524 | static inline void irq_time_write_begin(void) | |
1525 | { | |
1526 | __this_cpu_inc(irq_time_seq.sequence); | |
1527 | smp_wmb(); | |
1528 | } | |
1529 | ||
1530 | static inline void irq_time_write_end(void) | |
1531 | { | |
1532 | smp_wmb(); | |
1533 | __this_cpu_inc(irq_time_seq.sequence); | |
1534 | } | |
1535 | ||
1536 | static inline u64 irq_time_read(int cpu) | |
1537 | { | |
1538 | u64 irq_time; | |
1539 | unsigned seq; | |
1540 | ||
1541 | do { | |
1542 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1543 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1544 | per_cpu(cpu_hardirq_time, cpu); | |
1545 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1546 | ||
1547 | return irq_time; | |
1548 | } | |
1549 | #else /* CONFIG_64BIT */ | |
1550 | static inline void irq_time_write_begin(void) | |
1551 | { | |
1552 | } | |
1553 | ||
1554 | static inline void irq_time_write_end(void) | |
1555 | { | |
1556 | } | |
1557 | ||
1558 | static inline u64 irq_time_read(int cpu) | |
1559 | { | |
1560 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1561 | } | |
1562 | #endif /* CONFIG_64BIT */ | |
1563 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |