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 | ||
da0c1e65 KT |
18 | /* task_struct::on_rq states: */ |
19 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 20 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 21 | |
029632fb PZ |
22 | extern __read_mostly int scheduler_running; |
23 | ||
45ceebf7 PG |
24 | extern unsigned long calc_load_update; |
25 | extern atomic_long_t calc_load_tasks; | |
26 | ||
27 | extern long calc_load_fold_active(struct rq *this_rq); | |
28 | extern void update_cpu_load_active(struct rq *this_rq); | |
29 | ||
029632fb PZ |
30 | /* |
31 | * Helpers for converting nanosecond timing to jiffy resolution | |
32 | */ | |
33 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
34 | ||
cc1f4b1f LZ |
35 | /* |
36 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
37 | * The extra resolution improves shares distribution and load balancing of | |
38 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
39 | * hierarchies, especially on larger systems. This is not a user-visible change | |
40 | * and does not change the user-interface for setting shares/weights. | |
41 | * | |
42 | * We increase resolution only if we have enough bits to allow this increased | |
43 | * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution | |
44 | * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the | |
45 | * increased costs. | |
46 | */ | |
47 | #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */ | |
48 | # define SCHED_LOAD_RESOLUTION 10 | |
49 | # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION) | |
50 | # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION) | |
51 | #else | |
52 | # define SCHED_LOAD_RESOLUTION 0 | |
53 | # define scale_load(w) (w) | |
54 | # define scale_load_down(w) (w) | |
55 | #endif | |
56 | ||
57 | #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION) | |
58 | #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT) | |
59 | ||
029632fb PZ |
60 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
61 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
62 | ||
332ac17e DF |
63 | /* |
64 | * Single value that decides SCHED_DEADLINE internal math precision. | |
65 | * 10 -> just above 1us | |
66 | * 9 -> just above 0.5us | |
67 | */ | |
68 | #define DL_SCALE (10) | |
69 | ||
029632fb PZ |
70 | /* |
71 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 72 | */ |
029632fb PZ |
73 | |
74 | /* | |
75 | * single value that denotes runtime == period, ie unlimited time. | |
76 | */ | |
77 | #define RUNTIME_INF ((u64)~0ULL) | |
78 | ||
d50dde5a DF |
79 | static inline int fair_policy(int policy) |
80 | { | |
81 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
82 | } | |
83 | ||
029632fb PZ |
84 | static inline int rt_policy(int policy) |
85 | { | |
d50dde5a | 86 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
87 | } |
88 | ||
aab03e05 DF |
89 | static inline int dl_policy(int policy) |
90 | { | |
91 | return policy == SCHED_DEADLINE; | |
92 | } | |
93 | ||
029632fb PZ |
94 | static inline int task_has_rt_policy(struct task_struct *p) |
95 | { | |
96 | return rt_policy(p->policy); | |
97 | } | |
98 | ||
aab03e05 DF |
99 | static inline int task_has_dl_policy(struct task_struct *p) |
100 | { | |
101 | return dl_policy(p->policy); | |
102 | } | |
103 | ||
332ac17e | 104 | static inline bool dl_time_before(u64 a, u64 b) |
2d3d891d DF |
105 | { |
106 | return (s64)(a - b) < 0; | |
107 | } | |
108 | ||
109 | /* | |
110 | * Tells if entity @a should preempt entity @b. | |
111 | */ | |
332ac17e DF |
112 | static inline bool |
113 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
114 | { |
115 | return dl_time_before(a->deadline, b->deadline); | |
116 | } | |
117 | ||
029632fb PZ |
118 | /* |
119 | * This is the priority-queue data structure of the RT scheduling class: | |
120 | */ | |
121 | struct rt_prio_array { | |
122 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
123 | struct list_head queue[MAX_RT_PRIO]; | |
124 | }; | |
125 | ||
126 | struct rt_bandwidth { | |
127 | /* nests inside the rq lock: */ | |
128 | raw_spinlock_t rt_runtime_lock; | |
129 | ktime_t rt_period; | |
130 | u64 rt_runtime; | |
131 | struct hrtimer rt_period_timer; | |
132 | }; | |
332ac17e DF |
133 | /* |
134 | * To keep the bandwidth of -deadline tasks and groups under control | |
135 | * we need some place where: | |
136 | * - store the maximum -deadline bandwidth of the system (the group); | |
137 | * - cache the fraction of that bandwidth that is currently allocated. | |
138 | * | |
139 | * This is all done in the data structure below. It is similar to the | |
140 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
141 | * that, since here we are only interested in admission control, we | |
142 | * do not decrease any runtime while the group "executes", neither we | |
143 | * need a timer to replenish it. | |
144 | * | |
145 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
146 | * meaning that: | |
147 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
148 | * - dl_total_bw array contains, in the i-eth element, the currently | |
149 | * allocated bandwidth on the i-eth CPU. | |
150 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
151 | * consume bandwidth on the CPU they're running on. | |
152 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
153 | * that will be shown the next time the proc or cgroup controls will | |
154 | * be red. It on its turn can be changed by writing on its own | |
155 | * control. | |
156 | */ | |
157 | struct dl_bandwidth { | |
158 | raw_spinlock_t dl_runtime_lock; | |
159 | u64 dl_runtime; | |
160 | u64 dl_period; | |
161 | }; | |
162 | ||
163 | static inline int dl_bandwidth_enabled(void) | |
164 | { | |
1724813d | 165 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
166 | } |
167 | ||
168 | extern struct dl_bw *dl_bw_of(int i); | |
169 | ||
170 | struct dl_bw { | |
171 | raw_spinlock_t lock; | |
172 | u64 bw, total_bw; | |
173 | }; | |
174 | ||
029632fb PZ |
175 | extern struct mutex sched_domains_mutex; |
176 | ||
177 | #ifdef CONFIG_CGROUP_SCHED | |
178 | ||
179 | #include <linux/cgroup.h> | |
180 | ||
181 | struct cfs_rq; | |
182 | struct rt_rq; | |
183 | ||
35cf4e50 | 184 | extern struct list_head task_groups; |
029632fb PZ |
185 | |
186 | struct cfs_bandwidth { | |
187 | #ifdef CONFIG_CFS_BANDWIDTH | |
188 | raw_spinlock_t lock; | |
189 | ktime_t period; | |
190 | u64 quota, runtime; | |
191 | s64 hierarchal_quota; | |
192 | u64 runtime_expires; | |
193 | ||
194 | int idle, timer_active; | |
195 | struct hrtimer period_timer, slack_timer; | |
196 | struct list_head throttled_cfs_rq; | |
197 | ||
198 | /* statistics */ | |
199 | int nr_periods, nr_throttled; | |
200 | u64 throttled_time; | |
201 | #endif | |
202 | }; | |
203 | ||
204 | /* task group related information */ | |
205 | struct task_group { | |
206 | struct cgroup_subsys_state css; | |
207 | ||
208 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
209 | /* schedulable entities of this group on each cpu */ | |
210 | struct sched_entity **se; | |
211 | /* runqueue "owned" by this group on each cpu */ | |
212 | struct cfs_rq **cfs_rq; | |
213 | unsigned long shares; | |
214 | ||
fa6bddeb | 215 | #ifdef CONFIG_SMP |
bf5b986e | 216 | atomic_long_t load_avg; |
bb17f655 | 217 | atomic_t runnable_avg; |
029632fb | 218 | #endif |
fa6bddeb | 219 | #endif |
029632fb PZ |
220 | |
221 | #ifdef CONFIG_RT_GROUP_SCHED | |
222 | struct sched_rt_entity **rt_se; | |
223 | struct rt_rq **rt_rq; | |
224 | ||
225 | struct rt_bandwidth rt_bandwidth; | |
226 | #endif | |
227 | ||
228 | struct rcu_head rcu; | |
229 | struct list_head list; | |
230 | ||
231 | struct task_group *parent; | |
232 | struct list_head siblings; | |
233 | struct list_head children; | |
234 | ||
235 | #ifdef CONFIG_SCHED_AUTOGROUP | |
236 | struct autogroup *autogroup; | |
237 | #endif | |
238 | ||
239 | struct cfs_bandwidth cfs_bandwidth; | |
240 | }; | |
241 | ||
242 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
243 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
244 | ||
245 | /* | |
246 | * A weight of 0 or 1 can cause arithmetics problems. | |
247 | * A weight of a cfs_rq is the sum of weights of which entities | |
248 | * are queued on this cfs_rq, so a weight of a entity should not be | |
249 | * too large, so as the shares value of a task group. | |
250 | * (The default weight is 1024 - so there's no practical | |
251 | * limitation from this.) | |
252 | */ | |
253 | #define MIN_SHARES (1UL << 1) | |
254 | #define MAX_SHARES (1UL << 18) | |
255 | #endif | |
256 | ||
029632fb PZ |
257 | typedef int (*tg_visitor)(struct task_group *, void *); |
258 | ||
259 | extern int walk_tg_tree_from(struct task_group *from, | |
260 | tg_visitor down, tg_visitor up, void *data); | |
261 | ||
262 | /* | |
263 | * Iterate the full tree, calling @down when first entering a node and @up when | |
264 | * leaving it for the final time. | |
265 | * | |
266 | * Caller must hold rcu_lock or sufficient equivalent. | |
267 | */ | |
268 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
269 | { | |
270 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
271 | } | |
272 | ||
273 | extern int tg_nop(struct task_group *tg, void *data); | |
274 | ||
275 | extern void free_fair_sched_group(struct task_group *tg); | |
276 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
277 | extern void unregister_fair_sched_group(struct task_group *tg, int cpu); | |
278 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, | |
279 | struct sched_entity *se, int cpu, | |
280 | struct sched_entity *parent); | |
281 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
282 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
283 | ||
284 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
09dc4ab0 | 285 | extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, bool force); |
029632fb PZ |
286 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
287 | ||
288 | extern void free_rt_sched_group(struct task_group *tg); | |
289 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
290 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
291 | struct sched_rt_entity *rt_se, int cpu, | |
292 | struct sched_rt_entity *parent); | |
293 | ||
25cc7da7 LZ |
294 | extern struct task_group *sched_create_group(struct task_group *parent); |
295 | extern void sched_online_group(struct task_group *tg, | |
296 | struct task_group *parent); | |
297 | extern void sched_destroy_group(struct task_group *tg); | |
298 | extern void sched_offline_group(struct task_group *tg); | |
299 | ||
300 | extern void sched_move_task(struct task_struct *tsk); | |
301 | ||
302 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
303 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
304 | #endif | |
305 | ||
029632fb PZ |
306 | #else /* CONFIG_CGROUP_SCHED */ |
307 | ||
308 | struct cfs_bandwidth { }; | |
309 | ||
310 | #endif /* CONFIG_CGROUP_SCHED */ | |
311 | ||
312 | /* CFS-related fields in a runqueue */ | |
313 | struct cfs_rq { | |
314 | struct load_weight load; | |
c82513e5 | 315 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
316 | |
317 | u64 exec_clock; | |
318 | u64 min_vruntime; | |
319 | #ifndef CONFIG_64BIT | |
320 | u64 min_vruntime_copy; | |
321 | #endif | |
322 | ||
323 | struct rb_root tasks_timeline; | |
324 | struct rb_node *rb_leftmost; | |
325 | ||
029632fb PZ |
326 | /* |
327 | * 'curr' points to currently running entity on this cfs_rq. | |
328 | * It is set to NULL otherwise (i.e when none are currently running). | |
329 | */ | |
330 | struct sched_entity *curr, *next, *last, *skip; | |
331 | ||
332 | #ifdef CONFIG_SCHED_DEBUG | |
333 | unsigned int nr_spread_over; | |
334 | #endif | |
335 | ||
2dac754e PT |
336 | #ifdef CONFIG_SMP |
337 | /* | |
338 | * CFS Load tracking | |
339 | * Under CFS, load is tracked on a per-entity basis and aggregated up. | |
340 | * This allows for the description of both thread and group usage (in | |
341 | * the FAIR_GROUP_SCHED case). | |
342 | */ | |
72a4cf20 | 343 | unsigned long runnable_load_avg, blocked_load_avg; |
2509940f | 344 | atomic64_t decay_counter; |
9ee474f5 | 345 | u64 last_decay; |
2509940f | 346 | atomic_long_t removed_load; |
141965c7 | 347 | |
c566e8e9 | 348 | #ifdef CONFIG_FAIR_GROUP_SCHED |
141965c7 | 349 | /* Required to track per-cpu representation of a task_group */ |
bb17f655 | 350 | u32 tg_runnable_contrib; |
bf5b986e | 351 | unsigned long tg_load_contrib; |
82958366 PT |
352 | |
353 | /* | |
354 | * h_load = weight * f(tg) | |
355 | * | |
356 | * Where f(tg) is the recursive weight fraction assigned to | |
357 | * this group. | |
358 | */ | |
359 | unsigned long h_load; | |
68520796 VD |
360 | u64 last_h_load_update; |
361 | struct sched_entity *h_load_next; | |
362 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
363 | #endif /* CONFIG_SMP */ |
364 | ||
029632fb PZ |
365 | #ifdef CONFIG_FAIR_GROUP_SCHED |
366 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
367 | ||
368 | /* | |
369 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
370 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
371 | * (like users, containers etc.) | |
372 | * | |
373 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
374 | * list is used during load balance. | |
375 | */ | |
376 | int on_list; | |
377 | struct list_head leaf_cfs_rq_list; | |
378 | struct task_group *tg; /* group that "owns" this runqueue */ | |
379 | ||
029632fb PZ |
380 | #ifdef CONFIG_CFS_BANDWIDTH |
381 | int runtime_enabled; | |
382 | u64 runtime_expires; | |
383 | s64 runtime_remaining; | |
384 | ||
f1b17280 PT |
385 | u64 throttled_clock, throttled_clock_task; |
386 | u64 throttled_clock_task_time; | |
029632fb PZ |
387 | int throttled, throttle_count; |
388 | struct list_head throttled_list; | |
389 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
390 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
391 | }; | |
392 | ||
393 | static inline int rt_bandwidth_enabled(void) | |
394 | { | |
395 | return sysctl_sched_rt_runtime >= 0; | |
396 | } | |
397 | ||
398 | /* Real-Time classes' related field in a runqueue: */ | |
399 | struct rt_rq { | |
400 | struct rt_prio_array active; | |
c82513e5 | 401 | unsigned int rt_nr_running; |
029632fb PZ |
402 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
403 | struct { | |
404 | int curr; /* highest queued rt task prio */ | |
405 | #ifdef CONFIG_SMP | |
406 | int next; /* next highest */ | |
407 | #endif | |
408 | } highest_prio; | |
409 | #endif | |
410 | #ifdef CONFIG_SMP | |
411 | unsigned long rt_nr_migratory; | |
412 | unsigned long rt_nr_total; | |
413 | int overloaded; | |
414 | struct plist_head pushable_tasks; | |
415 | #endif | |
f4ebcbc0 KT |
416 | int rt_queued; |
417 | ||
029632fb PZ |
418 | int rt_throttled; |
419 | u64 rt_time; | |
420 | u64 rt_runtime; | |
421 | /* Nests inside the rq lock: */ | |
422 | raw_spinlock_t rt_runtime_lock; | |
423 | ||
424 | #ifdef CONFIG_RT_GROUP_SCHED | |
425 | unsigned long rt_nr_boosted; | |
426 | ||
427 | struct rq *rq; | |
029632fb PZ |
428 | struct task_group *tg; |
429 | #endif | |
430 | }; | |
431 | ||
aab03e05 DF |
432 | /* Deadline class' related fields in a runqueue */ |
433 | struct dl_rq { | |
434 | /* runqueue is an rbtree, ordered by deadline */ | |
435 | struct rb_root rb_root; | |
436 | struct rb_node *rb_leftmost; | |
437 | ||
438 | unsigned long dl_nr_running; | |
1baca4ce JL |
439 | |
440 | #ifdef CONFIG_SMP | |
441 | /* | |
442 | * Deadline values of the currently executing and the | |
443 | * earliest ready task on this rq. Caching these facilitates | |
444 | * the decision wether or not a ready but not running task | |
445 | * should migrate somewhere else. | |
446 | */ | |
447 | struct { | |
448 | u64 curr; | |
449 | u64 next; | |
450 | } earliest_dl; | |
451 | ||
452 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
453 | int overloaded; |
454 | ||
455 | /* | |
456 | * Tasks on this rq that can be pushed away. They are kept in | |
457 | * an rb-tree, ordered by tasks' deadlines, with caching | |
458 | * of the leftmost (earliest deadline) element. | |
459 | */ | |
460 | struct rb_root pushable_dl_tasks_root; | |
461 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
462 | #else |
463 | struct dl_bw dl_bw; | |
1baca4ce | 464 | #endif |
aab03e05 DF |
465 | }; |
466 | ||
029632fb PZ |
467 | #ifdef CONFIG_SMP |
468 | ||
469 | /* | |
470 | * We add the notion of a root-domain which will be used to define per-domain | |
471 | * variables. Each exclusive cpuset essentially defines an island domain by | |
472 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
473 | * exclusive cpuset is created, we also create and attach a new root-domain | |
474 | * object. | |
475 | * | |
476 | */ | |
477 | struct root_domain { | |
478 | atomic_t refcount; | |
479 | atomic_t rto_count; | |
480 | struct rcu_head rcu; | |
481 | cpumask_var_t span; | |
482 | cpumask_var_t online; | |
483 | ||
4486edd1 TC |
484 | /* Indicate more than one runnable task for any CPU */ |
485 | bool overload; | |
486 | ||
1baca4ce JL |
487 | /* |
488 | * The bit corresponding to a CPU gets set here if such CPU has more | |
489 | * than one runnable -deadline task (as it is below for RT tasks). | |
490 | */ | |
491 | cpumask_var_t dlo_mask; | |
492 | atomic_t dlo_count; | |
332ac17e | 493 | struct dl_bw dl_bw; |
6bfd6d72 | 494 | struct cpudl cpudl; |
1baca4ce | 495 | |
029632fb PZ |
496 | /* |
497 | * The "RT overload" flag: it gets set if a CPU has more than | |
498 | * one runnable RT task. | |
499 | */ | |
500 | cpumask_var_t rto_mask; | |
501 | struct cpupri cpupri; | |
502 | }; | |
503 | ||
504 | extern struct root_domain def_root_domain; | |
505 | ||
506 | #endif /* CONFIG_SMP */ | |
507 | ||
508 | /* | |
509 | * This is the main, per-CPU runqueue data structure. | |
510 | * | |
511 | * Locking rule: those places that want to lock multiple runqueues | |
512 | * (such as the load balancing or the thread migration code), lock | |
513 | * acquire operations must be ordered by ascending &runqueue. | |
514 | */ | |
515 | struct rq { | |
516 | /* runqueue lock: */ | |
517 | raw_spinlock_t lock; | |
518 | ||
519 | /* | |
520 | * nr_running and cpu_load should be in the same cacheline because | |
521 | * remote CPUs use both these fields when doing load calculation. | |
522 | */ | |
c82513e5 | 523 | unsigned int nr_running; |
0ec8aa00 PZ |
524 | #ifdef CONFIG_NUMA_BALANCING |
525 | unsigned int nr_numa_running; | |
526 | unsigned int nr_preferred_running; | |
527 | #endif | |
029632fb PZ |
528 | #define CPU_LOAD_IDX_MAX 5 |
529 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
530 | unsigned long last_load_update_tick; | |
3451d024 | 531 | #ifdef CONFIG_NO_HZ_COMMON |
029632fb | 532 | u64 nohz_stamp; |
1c792db7 | 533 | unsigned long nohz_flags; |
265f22a9 FW |
534 | #endif |
535 | #ifdef CONFIG_NO_HZ_FULL | |
536 | unsigned long last_sched_tick; | |
029632fb PZ |
537 | #endif |
538 | int skip_clock_update; | |
539 | ||
540 | /* capture load from *all* tasks on this cpu: */ | |
541 | struct load_weight load; | |
542 | unsigned long nr_load_updates; | |
543 | u64 nr_switches; | |
544 | ||
545 | struct cfs_rq cfs; | |
546 | struct rt_rq rt; | |
aab03e05 | 547 | struct dl_rq dl; |
029632fb PZ |
548 | |
549 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
550 | /* list of leaf cfs_rq on this cpu: */ | |
551 | struct list_head leaf_cfs_rq_list; | |
f5f9739d DE |
552 | |
553 | struct sched_avg avg; | |
a35b6466 PZ |
554 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
555 | ||
029632fb PZ |
556 | /* |
557 | * This is part of a global counter where only the total sum | |
558 | * over all CPUs matters. A task can increase this counter on | |
559 | * one CPU and if it got migrated afterwards it may decrease | |
560 | * it on another CPU. Always updated under the runqueue lock: | |
561 | */ | |
562 | unsigned long nr_uninterruptible; | |
563 | ||
564 | struct task_struct *curr, *idle, *stop; | |
565 | unsigned long next_balance; | |
566 | struct mm_struct *prev_mm; | |
567 | ||
568 | u64 clock; | |
569 | u64 clock_task; | |
570 | ||
571 | atomic_t nr_iowait; | |
572 | ||
573 | #ifdef CONFIG_SMP | |
574 | struct root_domain *rd; | |
575 | struct sched_domain *sd; | |
576 | ||
ced549fa | 577 | unsigned long cpu_capacity; |
029632fb PZ |
578 | |
579 | unsigned char idle_balance; | |
580 | /* For active balancing */ | |
581 | int post_schedule; | |
582 | int active_balance; | |
583 | int push_cpu; | |
584 | struct cpu_stop_work active_balance_work; | |
585 | /* cpu of this runqueue: */ | |
586 | int cpu; | |
587 | int online; | |
588 | ||
367456c7 PZ |
589 | struct list_head cfs_tasks; |
590 | ||
029632fb PZ |
591 | u64 rt_avg; |
592 | u64 age_stamp; | |
593 | u64 idle_stamp; | |
594 | u64 avg_idle; | |
9bd721c5 JL |
595 | |
596 | /* This is used to determine avg_idle's max value */ | |
597 | u64 max_idle_balance_cost; | |
029632fb PZ |
598 | #endif |
599 | ||
600 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
601 | u64 prev_irq_time; | |
602 | #endif | |
603 | #ifdef CONFIG_PARAVIRT | |
604 | u64 prev_steal_time; | |
605 | #endif | |
606 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
607 | u64 prev_steal_time_rq; | |
608 | #endif | |
609 | ||
610 | /* calc_load related fields */ | |
611 | unsigned long calc_load_update; | |
612 | long calc_load_active; | |
613 | ||
614 | #ifdef CONFIG_SCHED_HRTICK | |
615 | #ifdef CONFIG_SMP | |
616 | int hrtick_csd_pending; | |
617 | struct call_single_data hrtick_csd; | |
618 | #endif | |
619 | struct hrtimer hrtick_timer; | |
620 | #endif | |
621 | ||
622 | #ifdef CONFIG_SCHEDSTATS | |
623 | /* latency stats */ | |
624 | struct sched_info rq_sched_info; | |
625 | unsigned long long rq_cpu_time; | |
626 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
627 | ||
628 | /* sys_sched_yield() stats */ | |
629 | unsigned int yld_count; | |
630 | ||
631 | /* schedule() stats */ | |
029632fb PZ |
632 | unsigned int sched_count; |
633 | unsigned int sched_goidle; | |
634 | ||
635 | /* try_to_wake_up() stats */ | |
636 | unsigned int ttwu_count; | |
637 | unsigned int ttwu_local; | |
638 | #endif | |
639 | ||
640 | #ifdef CONFIG_SMP | |
641 | struct llist_head wake_list; | |
642 | #endif | |
643 | }; | |
644 | ||
645 | static inline int cpu_of(struct rq *rq) | |
646 | { | |
647 | #ifdef CONFIG_SMP | |
648 | return rq->cpu; | |
649 | #else | |
650 | return 0; | |
651 | #endif | |
652 | } | |
653 | ||
8b06c55b | 654 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 655 | |
518cd623 PZ |
656 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
657 | #define this_rq() (&__get_cpu_var(runqueues)) | |
658 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
659 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
660 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) | |
661 | ||
78becc27 FW |
662 | static inline u64 rq_clock(struct rq *rq) |
663 | { | |
664 | return rq->clock; | |
665 | } | |
666 | ||
667 | static inline u64 rq_clock_task(struct rq *rq) | |
668 | { | |
669 | return rq->clock_task; | |
670 | } | |
671 | ||
f809ca9a | 672 | #ifdef CONFIG_NUMA_BALANCING |
0ec8aa00 | 673 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 674 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 675 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
676 | #endif /* CONFIG_NUMA_BALANCING */ |
677 | ||
518cd623 PZ |
678 | #ifdef CONFIG_SMP |
679 | ||
e3baac47 PZ |
680 | extern void sched_ttwu_pending(void); |
681 | ||
029632fb PZ |
682 | #define rcu_dereference_check_sched_domain(p) \ |
683 | rcu_dereference_check((p), \ | |
684 | lockdep_is_held(&sched_domains_mutex)) | |
685 | ||
686 | /* | |
687 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
688 | * See detach_destroy_domains: synchronize_sched for details. | |
689 | * | |
690 | * The domain tree of any CPU may only be accessed from within | |
691 | * preempt-disabled sections. | |
692 | */ | |
693 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
694 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
695 | __sd; __sd = __sd->parent) | |
029632fb | 696 | |
77e81365 SS |
697 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
698 | ||
518cd623 PZ |
699 | /** |
700 | * highest_flag_domain - Return highest sched_domain containing flag. | |
701 | * @cpu: The cpu whose highest level of sched domain is to | |
702 | * be returned. | |
703 | * @flag: The flag to check for the highest sched_domain | |
704 | * for the given cpu. | |
705 | * | |
706 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
707 | */ | |
708 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
709 | { | |
710 | struct sched_domain *sd, *hsd = NULL; | |
711 | ||
712 | for_each_domain(cpu, sd) { | |
713 | if (!(sd->flags & flag)) | |
714 | break; | |
715 | hsd = sd; | |
716 | } | |
717 | ||
718 | return hsd; | |
719 | } | |
720 | ||
fb13c7ee MG |
721 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
722 | { | |
723 | struct sched_domain *sd; | |
724 | ||
725 | for_each_domain(cpu, sd) { | |
726 | if (sd->flags & flag) | |
727 | break; | |
728 | } | |
729 | ||
730 | return sd; | |
731 | } | |
732 | ||
518cd623 | 733 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 734 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 735 | DECLARE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 736 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
737 | DECLARE_PER_CPU(struct sched_domain *, sd_busy); |
738 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 | 739 | |
63b2ca30 | 740 | struct sched_group_capacity { |
5e6521ea LZ |
741 | atomic_t ref; |
742 | /* | |
63b2ca30 NP |
743 | * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity |
744 | * for a single CPU. | |
5e6521ea | 745 | */ |
63b2ca30 | 746 | unsigned int capacity, capacity_orig; |
5e6521ea | 747 | unsigned long next_update; |
63b2ca30 | 748 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea LZ |
749 | /* |
750 | * Number of busy cpus in this group. | |
751 | */ | |
752 | atomic_t nr_busy_cpus; | |
753 | ||
754 | unsigned long cpumask[0]; /* iteration mask */ | |
755 | }; | |
756 | ||
757 | struct sched_group { | |
758 | struct sched_group *next; /* Must be a circular list */ | |
759 | atomic_t ref; | |
760 | ||
761 | unsigned int group_weight; | |
63b2ca30 | 762 | struct sched_group_capacity *sgc; |
5e6521ea LZ |
763 | |
764 | /* | |
765 | * The CPUs this group covers. | |
766 | * | |
767 | * NOTE: this field is variable length. (Allocated dynamically | |
768 | * by attaching extra space to the end of the structure, | |
769 | * depending on how many CPUs the kernel has booted up with) | |
770 | */ | |
771 | unsigned long cpumask[0]; | |
772 | }; | |
773 | ||
774 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) | |
775 | { | |
776 | return to_cpumask(sg->cpumask); | |
777 | } | |
778 | ||
779 | /* | |
780 | * cpumask masking which cpus in the group are allowed to iterate up the domain | |
781 | * tree. | |
782 | */ | |
783 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) | |
784 | { | |
63b2ca30 | 785 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
786 | } |
787 | ||
788 | /** | |
789 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
790 | * @group: The group whose first cpu is to be returned. | |
791 | */ | |
792 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
793 | { | |
794 | return cpumask_first(sched_group_cpus(group)); | |
795 | } | |
796 | ||
c1174876 PZ |
797 | extern int group_balance_cpu(struct sched_group *sg); |
798 | ||
e3baac47 PZ |
799 | #else |
800 | ||
801 | static inline void sched_ttwu_pending(void) { } | |
802 | ||
518cd623 | 803 | #endif /* CONFIG_SMP */ |
029632fb | 804 | |
391e43da PZ |
805 | #include "stats.h" |
806 | #include "auto_group.h" | |
029632fb PZ |
807 | |
808 | #ifdef CONFIG_CGROUP_SCHED | |
809 | ||
810 | /* | |
811 | * Return the group to which this tasks belongs. | |
812 | * | |
8af01f56 TH |
813 | * We cannot use task_css() and friends because the cgroup subsystem |
814 | * changes that value before the cgroup_subsys::attach() method is called, | |
815 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
816 | * |
817 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
818 | * core changes this before calling sched_move_task(). | |
819 | * | |
820 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
821 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
822 | */ |
823 | static inline struct task_group *task_group(struct task_struct *p) | |
824 | { | |
8323f26c | 825 | return p->sched_task_group; |
029632fb PZ |
826 | } |
827 | ||
828 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
829 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
830 | { | |
831 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
832 | struct task_group *tg = task_group(p); | |
833 | #endif | |
834 | ||
835 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
836 | p->se.cfs_rq = tg->cfs_rq[cpu]; | |
837 | p->se.parent = tg->se[cpu]; | |
838 | #endif | |
839 | ||
840 | #ifdef CONFIG_RT_GROUP_SCHED | |
841 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
842 | p->rt.parent = tg->rt_se[cpu]; | |
843 | #endif | |
844 | } | |
845 | ||
846 | #else /* CONFIG_CGROUP_SCHED */ | |
847 | ||
848 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
849 | static inline struct task_group *task_group(struct task_struct *p) | |
850 | { | |
851 | return NULL; | |
852 | } | |
853 | ||
854 | #endif /* CONFIG_CGROUP_SCHED */ | |
855 | ||
856 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
857 | { | |
858 | set_task_rq(p, cpu); | |
859 | #ifdef CONFIG_SMP | |
860 | /* | |
861 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
862 | * successfuly executed on another CPU. We must ensure that updates of | |
863 | * per-task data have been completed by this moment. | |
864 | */ | |
865 | smp_wmb(); | |
866 | task_thread_info(p)->cpu = cpu; | |
ac66f547 | 867 | p->wake_cpu = cpu; |
029632fb PZ |
868 | #endif |
869 | } | |
870 | ||
871 | /* | |
872 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
873 | */ | |
874 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 875 | # include <linux/static_key.h> |
029632fb PZ |
876 | # define const_debug __read_mostly |
877 | #else | |
878 | # define const_debug const | |
879 | #endif | |
880 | ||
881 | extern const_debug unsigned int sysctl_sched_features; | |
882 | ||
883 | #define SCHED_FEAT(name, enabled) \ | |
884 | __SCHED_FEAT_##name , | |
885 | ||
886 | enum { | |
391e43da | 887 | #include "features.h" |
f8b6d1cc | 888 | __SCHED_FEAT_NR, |
029632fb PZ |
889 | }; |
890 | ||
891 | #undef SCHED_FEAT | |
892 | ||
f8b6d1cc | 893 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
f8b6d1cc | 894 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 895 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 896 | { \ |
6e76ea8a | 897 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
898 | } |
899 | ||
900 | #include "features.h" | |
901 | ||
902 | #undef SCHED_FEAT | |
903 | ||
c5905afb | 904 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
905 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
906 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 907 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 908 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 909 | |
cbee9f88 PZ |
910 | #ifdef CONFIG_NUMA_BALANCING |
911 | #define sched_feat_numa(x) sched_feat(x) | |
3105b86a MG |
912 | #ifdef CONFIG_SCHED_DEBUG |
913 | #define numabalancing_enabled sched_feat_numa(NUMA) | |
914 | #else | |
915 | extern bool numabalancing_enabled; | |
916 | #endif /* CONFIG_SCHED_DEBUG */ | |
cbee9f88 PZ |
917 | #else |
918 | #define sched_feat_numa(x) (0) | |
3105b86a MG |
919 | #define numabalancing_enabled (0) |
920 | #endif /* CONFIG_NUMA_BALANCING */ | |
cbee9f88 | 921 | |
029632fb PZ |
922 | static inline u64 global_rt_period(void) |
923 | { | |
924 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
925 | } | |
926 | ||
927 | static inline u64 global_rt_runtime(void) | |
928 | { | |
929 | if (sysctl_sched_rt_runtime < 0) | |
930 | return RUNTIME_INF; | |
931 | ||
932 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
933 | } | |
934 | ||
029632fb PZ |
935 | static inline int task_current(struct rq *rq, struct task_struct *p) |
936 | { | |
937 | return rq->curr == p; | |
938 | } | |
939 | ||
940 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
941 | { | |
942 | #ifdef CONFIG_SMP | |
943 | return p->on_cpu; | |
944 | #else | |
945 | return task_current(rq, p); | |
946 | #endif | |
947 | } | |
948 | ||
da0c1e65 KT |
949 | static inline int task_on_rq_queued(struct task_struct *p) |
950 | { | |
951 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
952 | } | |
029632fb | 953 | |
cca26e80 KT |
954 | static inline int task_on_rq_migrating(struct task_struct *p) |
955 | { | |
956 | return p->on_rq == TASK_ON_RQ_MIGRATING; | |
957 | } | |
958 | ||
029632fb PZ |
959 | #ifndef prepare_arch_switch |
960 | # define prepare_arch_switch(next) do { } while (0) | |
961 | #endif | |
962 | #ifndef finish_arch_switch | |
963 | # define finish_arch_switch(prev) do { } while (0) | |
964 | #endif | |
01f23e16 CM |
965 | #ifndef finish_arch_post_lock_switch |
966 | # define finish_arch_post_lock_switch() do { } while (0) | |
967 | #endif | |
029632fb PZ |
968 | |
969 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
970 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
971 | { | |
972 | #ifdef CONFIG_SMP | |
973 | /* | |
974 | * We can optimise this out completely for !SMP, because the | |
975 | * SMP rebalancing from interrupt is the only thing that cares | |
976 | * here. | |
977 | */ | |
978 | next->on_cpu = 1; | |
979 | #endif | |
980 | } | |
981 | ||
982 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
983 | { | |
984 | #ifdef CONFIG_SMP | |
985 | /* | |
986 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
987 | * We must ensure this doesn't happen until the switch is completely | |
988 | * finished. | |
989 | */ | |
990 | smp_wmb(); | |
991 | prev->on_cpu = 0; | |
992 | #endif | |
993 | #ifdef CONFIG_DEBUG_SPINLOCK | |
994 | /* this is a valid case when another task releases the spinlock */ | |
995 | rq->lock.owner = current; | |
996 | #endif | |
997 | /* | |
998 | * If we are tracking spinlock dependencies then we have to | |
999 | * fix up the runqueue lock - which gets 'carried over' from | |
1000 | * prev into current: | |
1001 | */ | |
1002 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1003 | ||
1004 | raw_spin_unlock_irq(&rq->lock); | |
1005 | } | |
1006 | ||
1007 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1008 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
1009 | { | |
1010 | #ifdef CONFIG_SMP | |
1011 | /* | |
1012 | * We can optimise this out completely for !SMP, because the | |
1013 | * SMP rebalancing from interrupt is the only thing that cares | |
1014 | * here. | |
1015 | */ | |
1016 | next->on_cpu = 1; | |
1017 | #endif | |
029632fb | 1018 | raw_spin_unlock(&rq->lock); |
029632fb PZ |
1019 | } |
1020 | ||
1021 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1022 | { | |
1023 | #ifdef CONFIG_SMP | |
1024 | /* | |
1025 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1026 | * We must ensure this doesn't happen until the switch is completely | |
1027 | * finished. | |
1028 | */ | |
1029 | smp_wmb(); | |
1030 | prev->on_cpu = 0; | |
1031 | #endif | |
029632fb | 1032 | local_irq_enable(); |
029632fb PZ |
1033 | } |
1034 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1035 | ||
b13095f0 LZ |
1036 | /* |
1037 | * wake flags | |
1038 | */ | |
1039 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1040 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1041 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1042 | ||
029632fb PZ |
1043 | /* |
1044 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1045 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1046 | * each task makes to its run queue's load is weighted according to its | |
1047 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1048 | * scaled version of the new time slice allocation that they receive on time | |
1049 | * slice expiry etc. | |
1050 | */ | |
1051 | ||
1052 | #define WEIGHT_IDLEPRIO 3 | |
1053 | #define WMULT_IDLEPRIO 1431655765 | |
1054 | ||
1055 | /* | |
1056 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1057 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1058 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1059 | * that remained on nice 0. | |
1060 | * | |
1061 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1062 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
1063 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
1064 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1065 | * the relative distance between them is ~25%.) | |
1066 | */ | |
1067 | static const int prio_to_weight[40] = { | |
1068 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
1069 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1070 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1071 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1072 | /* 0 */ 1024, 820, 655, 526, 423, | |
1073 | /* 5 */ 335, 272, 215, 172, 137, | |
1074 | /* 10 */ 110, 87, 70, 56, 45, | |
1075 | /* 15 */ 36, 29, 23, 18, 15, | |
1076 | }; | |
1077 | ||
1078 | /* | |
1079 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1080 | * | |
1081 | * In cases where the weight does not change often, we can use the | |
1082 | * precalculated inverse to speed up arithmetics by turning divisions | |
1083 | * into multiplications: | |
1084 | */ | |
1085 | static const u32 prio_to_wmult[40] = { | |
1086 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
1087 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1088 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1089 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1090 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1091 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1092 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1093 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
1094 | }; | |
1095 | ||
c82ba9fa LZ |
1096 | #define ENQUEUE_WAKEUP 1 |
1097 | #define ENQUEUE_HEAD 2 | |
1098 | #ifdef CONFIG_SMP | |
1099 | #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */ | |
1100 | #else | |
1101 | #define ENQUEUE_WAKING 0 | |
1102 | #endif | |
aab03e05 | 1103 | #define ENQUEUE_REPLENISH 8 |
c82ba9fa LZ |
1104 | |
1105 | #define DEQUEUE_SLEEP 1 | |
1106 | ||
37e117c0 PZ |
1107 | #define RETRY_TASK ((void *)-1UL) |
1108 | ||
c82ba9fa LZ |
1109 | struct sched_class { |
1110 | const struct sched_class *next; | |
1111 | ||
1112 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1113 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1114 | void (*yield_task) (struct rq *rq); | |
1115 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1116 | ||
1117 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1118 | ||
606dba2e PZ |
1119 | /* |
1120 | * It is the responsibility of the pick_next_task() method that will | |
1121 | * return the next task to call put_prev_task() on the @prev task or | |
1122 | * something equivalent. | |
37e117c0 PZ |
1123 | * |
1124 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1125 | * tasks. | |
606dba2e PZ |
1126 | */ |
1127 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
1128 | struct task_struct *prev); | |
c82ba9fa LZ |
1129 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1130 | ||
1131 | #ifdef CONFIG_SMP | |
ac66f547 | 1132 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
c82ba9fa LZ |
1133 | void (*migrate_task_rq)(struct task_struct *p, int next_cpu); |
1134 | ||
c82ba9fa LZ |
1135 | void (*post_schedule) (struct rq *this_rq); |
1136 | void (*task_waking) (struct task_struct *task); | |
1137 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); | |
1138 | ||
1139 | void (*set_cpus_allowed)(struct task_struct *p, | |
1140 | const struct cpumask *newmask); | |
1141 | ||
1142 | void (*rq_online)(struct rq *rq); | |
1143 | void (*rq_offline)(struct rq *rq); | |
1144 | #endif | |
1145 | ||
1146 | void (*set_curr_task) (struct rq *rq); | |
1147 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1148 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1149 | void (*task_dead) (struct task_struct *p); |
c82ba9fa LZ |
1150 | |
1151 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); | |
1152 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1153 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1154 | int oldprio); | |
1155 | ||
1156 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1157 | struct task_struct *task); | |
1158 | ||
1159 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1160 | void (*task_move_group) (struct task_struct *p, int on_rq); | |
1161 | #endif | |
1162 | }; | |
029632fb | 1163 | |
3f1d2a31 PZ |
1164 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1165 | { | |
1166 | prev->sched_class->put_prev_task(rq, prev); | |
1167 | } | |
1168 | ||
029632fb PZ |
1169 | #define sched_class_highest (&stop_sched_class) |
1170 | #define for_each_class(class) \ | |
1171 | for (class = sched_class_highest; class; class = class->next) | |
1172 | ||
1173 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1174 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1175 | extern const struct sched_class rt_sched_class; |
1176 | extern const struct sched_class fair_sched_class; | |
1177 | extern const struct sched_class idle_sched_class; | |
1178 | ||
1179 | ||
1180 | #ifdef CONFIG_SMP | |
1181 | ||
63b2ca30 | 1182 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1183 | |
7caff66f | 1184 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1185 | |
642dbc39 VG |
1186 | extern void idle_enter_fair(struct rq *this_rq); |
1187 | extern void idle_exit_fair(struct rq *this_rq); | |
642dbc39 | 1188 | |
dc877341 PZ |
1189 | #else |
1190 | ||
1191 | static inline void idle_enter_fair(struct rq *rq) { } | |
1192 | static inline void idle_exit_fair(struct rq *rq) { } | |
1193 | ||
029632fb PZ |
1194 | #endif |
1195 | ||
1196 | extern void sysrq_sched_debug_show(void); | |
1197 | extern void sched_init_granularity(void); | |
1198 | extern void update_max_interval(void); | |
1baca4ce JL |
1199 | |
1200 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1201 | extern void init_sched_rt_class(void); |
1202 | extern void init_sched_fair_class(void); | |
332ac17e | 1203 | extern void init_sched_dl_class(void); |
029632fb | 1204 | |
8875125e | 1205 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1206 | extern void resched_cpu(int cpu); |
1207 | ||
1208 | extern struct rt_bandwidth def_rt_bandwidth; | |
1209 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1210 | ||
332ac17e DF |
1211 | extern struct dl_bandwidth def_dl_bandwidth; |
1212 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 DF |
1213 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
1214 | ||
332ac17e DF |
1215 | unsigned long to_ratio(u64 period, u64 runtime); |
1216 | ||
556061b0 | 1217 | extern void update_idle_cpu_load(struct rq *this_rq); |
029632fb | 1218 | |
a75cdaa9 AS |
1219 | extern void init_task_runnable_average(struct task_struct *p); |
1220 | ||
72465447 | 1221 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1222 | { |
72465447 KT |
1223 | unsigned prev_nr = rq->nr_running; |
1224 | ||
1225 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1226 | |
72465447 | 1227 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1228 | #ifdef CONFIG_SMP |
1229 | if (!rq->rd->overload) | |
1230 | rq->rd->overload = true; | |
1231 | #endif | |
1232 | ||
1233 | #ifdef CONFIG_NO_HZ_FULL | |
9f3660c2 | 1234 | if (tick_nohz_full_cpu(rq->cpu)) { |
3882ec64 FW |
1235 | /* |
1236 | * Tick is needed if more than one task runs on a CPU. | |
1237 | * Send the target an IPI to kick it out of nohz mode. | |
1238 | * | |
1239 | * We assume that IPI implies full memory barrier and the | |
1240 | * new value of rq->nr_running is visible on reception | |
1241 | * from the target. | |
1242 | */ | |
fd2ac4f4 | 1243 | tick_nohz_full_kick_cpu(rq->cpu); |
9f3660c2 | 1244 | } |
9f3660c2 | 1245 | #endif |
4486edd1 | 1246 | } |
029632fb PZ |
1247 | } |
1248 | ||
72465447 | 1249 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1250 | { |
72465447 | 1251 | rq->nr_running -= count; |
029632fb PZ |
1252 | } |
1253 | ||
265f22a9 FW |
1254 | static inline void rq_last_tick_reset(struct rq *rq) |
1255 | { | |
1256 | #ifdef CONFIG_NO_HZ_FULL | |
1257 | rq->last_sched_tick = jiffies; | |
1258 | #endif | |
1259 | } | |
1260 | ||
029632fb PZ |
1261 | extern void update_rq_clock(struct rq *rq); |
1262 | ||
1263 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1264 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1265 | ||
1266 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1267 | ||
1268 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1269 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1270 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1271 | ||
1272 | static inline u64 sched_avg_period(void) | |
1273 | { | |
1274 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1275 | } | |
1276 | ||
029632fb PZ |
1277 | #ifdef CONFIG_SCHED_HRTICK |
1278 | ||
1279 | /* | |
1280 | * Use hrtick when: | |
1281 | * - enabled by features | |
1282 | * - hrtimer is actually high res | |
1283 | */ | |
1284 | static inline int hrtick_enabled(struct rq *rq) | |
1285 | { | |
1286 | if (!sched_feat(HRTICK)) | |
1287 | return 0; | |
1288 | if (!cpu_active(cpu_of(rq))) | |
1289 | return 0; | |
1290 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1291 | } | |
1292 | ||
1293 | void hrtick_start(struct rq *rq, u64 delay); | |
1294 | ||
b39e66ea MG |
1295 | #else |
1296 | ||
1297 | static inline int hrtick_enabled(struct rq *rq) | |
1298 | { | |
1299 | return 0; | |
1300 | } | |
1301 | ||
029632fb PZ |
1302 | #endif /* CONFIG_SCHED_HRTICK */ |
1303 | ||
1304 | #ifdef CONFIG_SMP | |
1305 | extern void sched_avg_update(struct rq *rq); | |
1306 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1307 | { | |
1308 | rq->rt_avg += rt_delta; | |
1309 | sched_avg_update(rq); | |
1310 | } | |
1311 | #else | |
1312 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1313 | static inline void sched_avg_update(struct rq *rq) { } | |
1314 | #endif | |
1315 | ||
1316 | extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period); | |
1317 | ||
1318 | #ifdef CONFIG_SMP | |
1319 | #ifdef CONFIG_PREEMPT | |
1320 | ||
1321 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1322 | ||
1323 | /* | |
1324 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1325 | * way at the expense of forcing extra atomic operations in all | |
1326 | * invocations. This assures that the double_lock is acquired using the | |
1327 | * same underlying policy as the spinlock_t on this architecture, which | |
1328 | * reduces latency compared to the unfair variant below. However, it | |
1329 | * also adds more overhead and therefore may reduce throughput. | |
1330 | */ | |
1331 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1332 | __releases(this_rq->lock) | |
1333 | __acquires(busiest->lock) | |
1334 | __acquires(this_rq->lock) | |
1335 | { | |
1336 | raw_spin_unlock(&this_rq->lock); | |
1337 | double_rq_lock(this_rq, busiest); | |
1338 | ||
1339 | return 1; | |
1340 | } | |
1341 | ||
1342 | #else | |
1343 | /* | |
1344 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1345 | * latency by eliminating extra atomic operations when the locks are | |
1346 | * already in proper order on entry. This favors lower cpu-ids and will | |
1347 | * grant the double lock to lower cpus over higher ids under contention, | |
1348 | * regardless of entry order into the function. | |
1349 | */ | |
1350 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1351 | __releases(this_rq->lock) | |
1352 | __acquires(busiest->lock) | |
1353 | __acquires(this_rq->lock) | |
1354 | { | |
1355 | int ret = 0; | |
1356 | ||
1357 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1358 | if (busiest < this_rq) { | |
1359 | raw_spin_unlock(&this_rq->lock); | |
1360 | raw_spin_lock(&busiest->lock); | |
1361 | raw_spin_lock_nested(&this_rq->lock, | |
1362 | SINGLE_DEPTH_NESTING); | |
1363 | ret = 1; | |
1364 | } else | |
1365 | raw_spin_lock_nested(&busiest->lock, | |
1366 | SINGLE_DEPTH_NESTING); | |
1367 | } | |
1368 | return ret; | |
1369 | } | |
1370 | ||
1371 | #endif /* CONFIG_PREEMPT */ | |
1372 | ||
1373 | /* | |
1374 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1375 | */ | |
1376 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1377 | { | |
1378 | if (unlikely(!irqs_disabled())) { | |
1379 | /* printk() doesn't work good under rq->lock */ | |
1380 | raw_spin_unlock(&this_rq->lock); | |
1381 | BUG_ON(1); | |
1382 | } | |
1383 | ||
1384 | return _double_lock_balance(this_rq, busiest); | |
1385 | } | |
1386 | ||
1387 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1388 | __releases(busiest->lock) | |
1389 | { | |
1390 | raw_spin_unlock(&busiest->lock); | |
1391 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1392 | } | |
1393 | ||
74602315 PZ |
1394 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1395 | { | |
1396 | if (l1 > l2) | |
1397 | swap(l1, l2); | |
1398 | ||
1399 | spin_lock(l1); | |
1400 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1401 | } | |
1402 | ||
60e69eed MG |
1403 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1404 | { | |
1405 | if (l1 > l2) | |
1406 | swap(l1, l2); | |
1407 | ||
1408 | spin_lock_irq(l1); | |
1409 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1410 | } | |
1411 | ||
74602315 PZ |
1412 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1413 | { | |
1414 | if (l1 > l2) | |
1415 | swap(l1, l2); | |
1416 | ||
1417 | raw_spin_lock(l1); | |
1418 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1419 | } | |
1420 | ||
029632fb PZ |
1421 | /* |
1422 | * double_rq_lock - safely lock two runqueues | |
1423 | * | |
1424 | * Note this does not disable interrupts like task_rq_lock, | |
1425 | * you need to do so manually before calling. | |
1426 | */ | |
1427 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1428 | __acquires(rq1->lock) | |
1429 | __acquires(rq2->lock) | |
1430 | { | |
1431 | BUG_ON(!irqs_disabled()); | |
1432 | if (rq1 == rq2) { | |
1433 | raw_spin_lock(&rq1->lock); | |
1434 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1435 | } else { | |
1436 | if (rq1 < rq2) { | |
1437 | raw_spin_lock(&rq1->lock); | |
1438 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1439 | } else { | |
1440 | raw_spin_lock(&rq2->lock); | |
1441 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1442 | } | |
1443 | } | |
1444 | } | |
1445 | ||
1446 | /* | |
1447 | * double_rq_unlock - safely unlock two runqueues | |
1448 | * | |
1449 | * Note this does not restore interrupts like task_rq_unlock, | |
1450 | * you need to do so manually after calling. | |
1451 | */ | |
1452 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1453 | __releases(rq1->lock) | |
1454 | __releases(rq2->lock) | |
1455 | { | |
1456 | raw_spin_unlock(&rq1->lock); | |
1457 | if (rq1 != rq2) | |
1458 | raw_spin_unlock(&rq2->lock); | |
1459 | else | |
1460 | __release(rq2->lock); | |
1461 | } | |
1462 | ||
1463 | #else /* CONFIG_SMP */ | |
1464 | ||
1465 | /* | |
1466 | * double_rq_lock - safely lock two runqueues | |
1467 | * | |
1468 | * Note this does not disable interrupts like task_rq_lock, | |
1469 | * you need to do so manually before calling. | |
1470 | */ | |
1471 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1472 | __acquires(rq1->lock) | |
1473 | __acquires(rq2->lock) | |
1474 | { | |
1475 | BUG_ON(!irqs_disabled()); | |
1476 | BUG_ON(rq1 != rq2); | |
1477 | raw_spin_lock(&rq1->lock); | |
1478 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1479 | } | |
1480 | ||
1481 | /* | |
1482 | * double_rq_unlock - safely unlock two runqueues | |
1483 | * | |
1484 | * Note this does not restore interrupts like task_rq_unlock, | |
1485 | * you need to do so manually after calling. | |
1486 | */ | |
1487 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1488 | __releases(rq1->lock) | |
1489 | __releases(rq2->lock) | |
1490 | { | |
1491 | BUG_ON(rq1 != rq2); | |
1492 | raw_spin_unlock(&rq1->lock); | |
1493 | __release(rq2->lock); | |
1494 | } | |
1495 | ||
1496 | #endif | |
1497 | ||
1498 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1499 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
1500 | extern void print_cfs_stats(struct seq_file *m, int cpu); | |
1501 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
1502 | ||
1503 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
1504 | extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq); | |
aab03e05 | 1505 | extern void init_dl_rq(struct dl_rq *dl_rq, struct rq *rq); |
029632fb | 1506 | |
1ee14e6c BS |
1507 | extern void cfs_bandwidth_usage_inc(void); |
1508 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1509 | |
3451d024 | 1510 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1511 | enum rq_nohz_flag_bits { |
1512 | NOHZ_TICK_STOPPED, | |
1513 | NOHZ_BALANCE_KICK, | |
1514 | }; | |
1515 | ||
1516 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
1517 | #endif | |
73fbec60 FW |
1518 | |
1519 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
1520 | ||
1521 | DECLARE_PER_CPU(u64, cpu_hardirq_time); | |
1522 | DECLARE_PER_CPU(u64, cpu_softirq_time); | |
1523 | ||
1524 | #ifndef CONFIG_64BIT | |
1525 | DECLARE_PER_CPU(seqcount_t, irq_time_seq); | |
1526 | ||
1527 | static inline void irq_time_write_begin(void) | |
1528 | { | |
1529 | __this_cpu_inc(irq_time_seq.sequence); | |
1530 | smp_wmb(); | |
1531 | } | |
1532 | ||
1533 | static inline void irq_time_write_end(void) | |
1534 | { | |
1535 | smp_wmb(); | |
1536 | __this_cpu_inc(irq_time_seq.sequence); | |
1537 | } | |
1538 | ||
1539 | static inline u64 irq_time_read(int cpu) | |
1540 | { | |
1541 | u64 irq_time; | |
1542 | unsigned seq; | |
1543 | ||
1544 | do { | |
1545 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1546 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1547 | per_cpu(cpu_hardirq_time, cpu); | |
1548 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1549 | ||
1550 | return irq_time; | |
1551 | } | |
1552 | #else /* CONFIG_64BIT */ | |
1553 | static inline void irq_time_write_begin(void) | |
1554 | { | |
1555 | } | |
1556 | ||
1557 | static inline void irq_time_write_end(void) | |
1558 | { | |
1559 | } | |
1560 | ||
1561 | static inline u64 irq_time_read(int cpu) | |
1562 | { | |
1563 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1564 | } | |
1565 | #endif /* CONFIG_64BIT */ | |
1566 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |