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" |
60fed789 | 13 | #include "cpuacct.h" |
029632fb | 14 | |
45ceebf7 PG |
15 | struct rq; |
16 | ||
029632fb PZ |
17 | extern __read_mostly int scheduler_running; |
18 | ||
45ceebf7 PG |
19 | extern unsigned long calc_load_update; |
20 | extern atomic_long_t calc_load_tasks; | |
21 | ||
22 | extern long calc_load_fold_active(struct rq *this_rq); | |
23 | extern void update_cpu_load_active(struct rq *this_rq); | |
24 | ||
029632fb PZ |
25 | /* |
26 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
27 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
28 | * and back. | |
29 | */ | |
30 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
31 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
32 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
33 | ||
34 | /* | |
35 | * 'User priority' is the nice value converted to something we | |
36 | * can work with better when scaling various scheduler parameters, | |
37 | * it's a [ 0 ... 39 ] range. | |
38 | */ | |
39 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
40 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
41 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
42 | ||
43 | /* | |
44 | * Helpers for converting nanosecond timing to jiffy resolution | |
45 | */ | |
46 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
47 | ||
cc1f4b1f LZ |
48 | /* |
49 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
50 | * The extra resolution improves shares distribution and load balancing of | |
51 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
52 | * hierarchies, especially on larger systems. This is not a user-visible change | |
53 | * and does not change the user-interface for setting shares/weights. | |
54 | * | |
55 | * We increase resolution only if we have enough bits to allow this increased | |
56 | * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution | |
57 | * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the | |
58 | * increased costs. | |
59 | */ | |
60 | #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */ | |
61 | # define SCHED_LOAD_RESOLUTION 10 | |
62 | # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION) | |
63 | # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION) | |
64 | #else | |
65 | # define SCHED_LOAD_RESOLUTION 0 | |
66 | # define scale_load(w) (w) | |
67 | # define scale_load_down(w) (w) | |
68 | #endif | |
69 | ||
70 | #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION) | |
71 | #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT) | |
72 | ||
029632fb PZ |
73 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
74 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
75 | ||
332ac17e DF |
76 | /* |
77 | * Single value that decides SCHED_DEADLINE internal math precision. | |
78 | * 10 -> just above 1us | |
79 | * 9 -> just above 0.5us | |
80 | */ | |
81 | #define DL_SCALE (10) | |
82 | ||
029632fb PZ |
83 | /* |
84 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 85 | */ |
029632fb PZ |
86 | |
87 | /* | |
88 | * single value that denotes runtime == period, ie unlimited time. | |
89 | */ | |
90 | #define RUNTIME_INF ((u64)~0ULL) | |
91 | ||
d50dde5a DF |
92 | static inline int fair_policy(int policy) |
93 | { | |
94 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
95 | } | |
96 | ||
029632fb PZ |
97 | static inline int rt_policy(int policy) |
98 | { | |
d50dde5a | 99 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
100 | } |
101 | ||
aab03e05 DF |
102 | static inline int dl_policy(int policy) |
103 | { | |
104 | return policy == SCHED_DEADLINE; | |
105 | } | |
106 | ||
029632fb PZ |
107 | static inline int task_has_rt_policy(struct task_struct *p) |
108 | { | |
109 | return rt_policy(p->policy); | |
110 | } | |
111 | ||
aab03e05 DF |
112 | static inline int task_has_dl_policy(struct task_struct *p) |
113 | { | |
114 | return dl_policy(p->policy); | |
115 | } | |
116 | ||
332ac17e | 117 | static inline bool dl_time_before(u64 a, u64 b) |
2d3d891d DF |
118 | { |
119 | return (s64)(a - b) < 0; | |
120 | } | |
121 | ||
122 | /* | |
123 | * Tells if entity @a should preempt entity @b. | |
124 | */ | |
332ac17e DF |
125 | static inline bool |
126 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
127 | { |
128 | return dl_time_before(a->deadline, b->deadline); | |
129 | } | |
130 | ||
029632fb PZ |
131 | /* |
132 | * This is the priority-queue data structure of the RT scheduling class: | |
133 | */ | |
134 | struct rt_prio_array { | |
135 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
136 | struct list_head queue[MAX_RT_PRIO]; | |
137 | }; | |
138 | ||
139 | struct rt_bandwidth { | |
140 | /* nests inside the rq lock: */ | |
141 | raw_spinlock_t rt_runtime_lock; | |
142 | ktime_t rt_period; | |
143 | u64 rt_runtime; | |
144 | struct hrtimer rt_period_timer; | |
145 | }; | |
332ac17e DF |
146 | /* |
147 | * To keep the bandwidth of -deadline tasks and groups under control | |
148 | * we need some place where: | |
149 | * - store the maximum -deadline bandwidth of the system (the group); | |
150 | * - cache the fraction of that bandwidth that is currently allocated. | |
151 | * | |
152 | * This is all done in the data structure below. It is similar to the | |
153 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
154 | * that, since here we are only interested in admission control, we | |
155 | * do not decrease any runtime while the group "executes", neither we | |
156 | * need a timer to replenish it. | |
157 | * | |
158 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
159 | * meaning that: | |
160 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
161 | * - dl_total_bw array contains, in the i-eth element, the currently | |
162 | * allocated bandwidth on the i-eth CPU. | |
163 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
164 | * consume bandwidth on the CPU they're running on. | |
165 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
166 | * that will be shown the next time the proc or cgroup controls will | |
167 | * be red. It on its turn can be changed by writing on its own | |
168 | * control. | |
169 | */ | |
170 | struct dl_bandwidth { | |
171 | raw_spinlock_t dl_runtime_lock; | |
172 | u64 dl_runtime; | |
173 | u64 dl_period; | |
174 | }; | |
175 | ||
176 | static inline int dl_bandwidth_enabled(void) | |
177 | { | |
178 | return sysctl_sched_dl_runtime >= 0; | |
179 | } | |
180 | ||
181 | extern struct dl_bw *dl_bw_of(int i); | |
182 | ||
183 | struct dl_bw { | |
184 | raw_spinlock_t lock; | |
185 | u64 bw, total_bw; | |
186 | }; | |
187 | ||
188 | static inline u64 global_dl_period(void); | |
189 | static inline u64 global_dl_runtime(void); | |
029632fb PZ |
190 | |
191 | extern struct mutex sched_domains_mutex; | |
192 | ||
193 | #ifdef CONFIG_CGROUP_SCHED | |
194 | ||
195 | #include <linux/cgroup.h> | |
196 | ||
197 | struct cfs_rq; | |
198 | struct rt_rq; | |
199 | ||
35cf4e50 | 200 | extern struct list_head task_groups; |
029632fb PZ |
201 | |
202 | struct cfs_bandwidth { | |
203 | #ifdef CONFIG_CFS_BANDWIDTH | |
204 | raw_spinlock_t lock; | |
205 | ktime_t period; | |
206 | u64 quota, runtime; | |
207 | s64 hierarchal_quota; | |
208 | u64 runtime_expires; | |
209 | ||
210 | int idle, timer_active; | |
211 | struct hrtimer period_timer, slack_timer; | |
212 | struct list_head throttled_cfs_rq; | |
213 | ||
214 | /* statistics */ | |
215 | int nr_periods, nr_throttled; | |
216 | u64 throttled_time; | |
217 | #endif | |
218 | }; | |
219 | ||
220 | /* task group related information */ | |
221 | struct task_group { | |
222 | struct cgroup_subsys_state css; | |
223 | ||
224 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
225 | /* schedulable entities of this group on each cpu */ | |
226 | struct sched_entity **se; | |
227 | /* runqueue "owned" by this group on each cpu */ | |
228 | struct cfs_rq **cfs_rq; | |
229 | unsigned long shares; | |
230 | ||
fa6bddeb | 231 | #ifdef CONFIG_SMP |
bf5b986e | 232 | atomic_long_t load_avg; |
bb17f655 | 233 | atomic_t runnable_avg; |
029632fb | 234 | #endif |
fa6bddeb | 235 | #endif |
029632fb PZ |
236 | |
237 | #ifdef CONFIG_RT_GROUP_SCHED | |
238 | struct sched_rt_entity **rt_se; | |
239 | struct rt_rq **rt_rq; | |
240 | ||
241 | struct rt_bandwidth rt_bandwidth; | |
242 | #endif | |
243 | ||
244 | struct rcu_head rcu; | |
245 | struct list_head list; | |
246 | ||
247 | struct task_group *parent; | |
248 | struct list_head siblings; | |
249 | struct list_head children; | |
250 | ||
251 | #ifdef CONFIG_SCHED_AUTOGROUP | |
252 | struct autogroup *autogroup; | |
253 | #endif | |
254 | ||
255 | struct cfs_bandwidth cfs_bandwidth; | |
256 | }; | |
257 | ||
258 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
259 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
260 | ||
261 | /* | |
262 | * A weight of 0 or 1 can cause arithmetics problems. | |
263 | * A weight of a cfs_rq is the sum of weights of which entities | |
264 | * are queued on this cfs_rq, so a weight of a entity should not be | |
265 | * too large, so as the shares value of a task group. | |
266 | * (The default weight is 1024 - so there's no practical | |
267 | * limitation from this.) | |
268 | */ | |
269 | #define MIN_SHARES (1UL << 1) | |
270 | #define MAX_SHARES (1UL << 18) | |
271 | #endif | |
272 | ||
029632fb PZ |
273 | typedef int (*tg_visitor)(struct task_group *, void *); |
274 | ||
275 | extern int walk_tg_tree_from(struct task_group *from, | |
276 | tg_visitor down, tg_visitor up, void *data); | |
277 | ||
278 | /* | |
279 | * Iterate the full tree, calling @down when first entering a node and @up when | |
280 | * leaving it for the final time. | |
281 | * | |
282 | * Caller must hold rcu_lock or sufficient equivalent. | |
283 | */ | |
284 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
285 | { | |
286 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
287 | } | |
288 | ||
289 | extern int tg_nop(struct task_group *tg, void *data); | |
290 | ||
291 | extern void free_fair_sched_group(struct task_group *tg); | |
292 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
293 | extern void unregister_fair_sched_group(struct task_group *tg, int cpu); | |
294 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, | |
295 | struct sched_entity *se, int cpu, | |
296 | struct sched_entity *parent); | |
297 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
298 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
299 | ||
300 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
301 | extern void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
302 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); | |
303 | ||
304 | extern void free_rt_sched_group(struct task_group *tg); | |
305 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
306 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
307 | struct sched_rt_entity *rt_se, int cpu, | |
308 | struct sched_rt_entity *parent); | |
309 | ||
25cc7da7 LZ |
310 | extern struct task_group *sched_create_group(struct task_group *parent); |
311 | extern void sched_online_group(struct task_group *tg, | |
312 | struct task_group *parent); | |
313 | extern void sched_destroy_group(struct task_group *tg); | |
314 | extern void sched_offline_group(struct task_group *tg); | |
315 | ||
316 | extern void sched_move_task(struct task_struct *tsk); | |
317 | ||
318 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
319 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
320 | #endif | |
321 | ||
029632fb PZ |
322 | #else /* CONFIG_CGROUP_SCHED */ |
323 | ||
324 | struct cfs_bandwidth { }; | |
325 | ||
326 | #endif /* CONFIG_CGROUP_SCHED */ | |
327 | ||
328 | /* CFS-related fields in a runqueue */ | |
329 | struct cfs_rq { | |
330 | struct load_weight load; | |
c82513e5 | 331 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
332 | |
333 | u64 exec_clock; | |
334 | u64 min_vruntime; | |
335 | #ifndef CONFIG_64BIT | |
336 | u64 min_vruntime_copy; | |
337 | #endif | |
338 | ||
339 | struct rb_root tasks_timeline; | |
340 | struct rb_node *rb_leftmost; | |
341 | ||
029632fb PZ |
342 | /* |
343 | * 'curr' points to currently running entity on this cfs_rq. | |
344 | * It is set to NULL otherwise (i.e when none are currently running). | |
345 | */ | |
346 | struct sched_entity *curr, *next, *last, *skip; | |
347 | ||
348 | #ifdef CONFIG_SCHED_DEBUG | |
349 | unsigned int nr_spread_over; | |
350 | #endif | |
351 | ||
2dac754e PT |
352 | #ifdef CONFIG_SMP |
353 | /* | |
354 | * CFS Load tracking | |
355 | * Under CFS, load is tracked on a per-entity basis and aggregated up. | |
356 | * This allows for the description of both thread and group usage (in | |
357 | * the FAIR_GROUP_SCHED case). | |
358 | */ | |
72a4cf20 | 359 | unsigned long runnable_load_avg, blocked_load_avg; |
2509940f | 360 | atomic64_t decay_counter; |
9ee474f5 | 361 | u64 last_decay; |
2509940f | 362 | atomic_long_t removed_load; |
141965c7 | 363 | |
c566e8e9 | 364 | #ifdef CONFIG_FAIR_GROUP_SCHED |
141965c7 | 365 | /* Required to track per-cpu representation of a task_group */ |
bb17f655 | 366 | u32 tg_runnable_contrib; |
bf5b986e | 367 | unsigned long tg_load_contrib; |
82958366 PT |
368 | |
369 | /* | |
370 | * h_load = weight * f(tg) | |
371 | * | |
372 | * Where f(tg) is the recursive weight fraction assigned to | |
373 | * this group. | |
374 | */ | |
375 | unsigned long h_load; | |
68520796 VD |
376 | u64 last_h_load_update; |
377 | struct sched_entity *h_load_next; | |
378 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
379 | #endif /* CONFIG_SMP */ |
380 | ||
029632fb PZ |
381 | #ifdef CONFIG_FAIR_GROUP_SCHED |
382 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
383 | ||
384 | /* | |
385 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
386 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
387 | * (like users, containers etc.) | |
388 | * | |
389 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
390 | * list is used during load balance. | |
391 | */ | |
392 | int on_list; | |
393 | struct list_head leaf_cfs_rq_list; | |
394 | struct task_group *tg; /* group that "owns" this runqueue */ | |
395 | ||
029632fb PZ |
396 | #ifdef CONFIG_CFS_BANDWIDTH |
397 | int runtime_enabled; | |
398 | u64 runtime_expires; | |
399 | s64 runtime_remaining; | |
400 | ||
f1b17280 PT |
401 | u64 throttled_clock, throttled_clock_task; |
402 | u64 throttled_clock_task_time; | |
029632fb PZ |
403 | int throttled, throttle_count; |
404 | struct list_head throttled_list; | |
405 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
406 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
407 | }; | |
408 | ||
409 | static inline int rt_bandwidth_enabled(void) | |
410 | { | |
411 | return sysctl_sched_rt_runtime >= 0; | |
412 | } | |
413 | ||
414 | /* Real-Time classes' related field in a runqueue: */ | |
415 | struct rt_rq { | |
416 | struct rt_prio_array active; | |
c82513e5 | 417 | unsigned int rt_nr_running; |
029632fb PZ |
418 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
419 | struct { | |
420 | int curr; /* highest queued rt task prio */ | |
421 | #ifdef CONFIG_SMP | |
422 | int next; /* next highest */ | |
423 | #endif | |
424 | } highest_prio; | |
425 | #endif | |
426 | #ifdef CONFIG_SMP | |
427 | unsigned long rt_nr_migratory; | |
428 | unsigned long rt_nr_total; | |
429 | int overloaded; | |
430 | struct plist_head pushable_tasks; | |
431 | #endif | |
432 | int rt_throttled; | |
433 | u64 rt_time; | |
434 | u64 rt_runtime; | |
435 | /* Nests inside the rq lock: */ | |
436 | raw_spinlock_t rt_runtime_lock; | |
437 | ||
438 | #ifdef CONFIG_RT_GROUP_SCHED | |
439 | unsigned long rt_nr_boosted; | |
440 | ||
441 | struct rq *rq; | |
029632fb PZ |
442 | struct task_group *tg; |
443 | #endif | |
444 | }; | |
445 | ||
aab03e05 DF |
446 | /* Deadline class' related fields in a runqueue */ |
447 | struct dl_rq { | |
448 | /* runqueue is an rbtree, ordered by deadline */ | |
449 | struct rb_root rb_root; | |
450 | struct rb_node *rb_leftmost; | |
451 | ||
452 | unsigned long dl_nr_running; | |
1baca4ce JL |
453 | |
454 | #ifdef CONFIG_SMP | |
455 | /* | |
456 | * Deadline values of the currently executing and the | |
457 | * earliest ready task on this rq. Caching these facilitates | |
458 | * the decision wether or not a ready but not running task | |
459 | * should migrate somewhere else. | |
460 | */ | |
461 | struct { | |
462 | u64 curr; | |
463 | u64 next; | |
464 | } earliest_dl; | |
465 | ||
466 | unsigned long dl_nr_migratory; | |
467 | unsigned long dl_nr_total; | |
468 | int overloaded; | |
469 | ||
470 | /* | |
471 | * Tasks on this rq that can be pushed away. They are kept in | |
472 | * an rb-tree, ordered by tasks' deadlines, with caching | |
473 | * of the leftmost (earliest deadline) element. | |
474 | */ | |
475 | struct rb_root pushable_dl_tasks_root; | |
476 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
477 | #else |
478 | struct dl_bw dl_bw; | |
1baca4ce | 479 | #endif |
aab03e05 DF |
480 | }; |
481 | ||
029632fb PZ |
482 | #ifdef CONFIG_SMP |
483 | ||
484 | /* | |
485 | * We add the notion of a root-domain which will be used to define per-domain | |
486 | * variables. Each exclusive cpuset essentially defines an island domain by | |
487 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
488 | * exclusive cpuset is created, we also create and attach a new root-domain | |
489 | * object. | |
490 | * | |
491 | */ | |
492 | struct root_domain { | |
493 | atomic_t refcount; | |
494 | atomic_t rto_count; | |
495 | struct rcu_head rcu; | |
496 | cpumask_var_t span; | |
497 | cpumask_var_t online; | |
498 | ||
1baca4ce JL |
499 | /* |
500 | * The bit corresponding to a CPU gets set here if such CPU has more | |
501 | * than one runnable -deadline task (as it is below for RT tasks). | |
502 | */ | |
503 | cpumask_var_t dlo_mask; | |
504 | atomic_t dlo_count; | |
332ac17e | 505 | struct dl_bw dl_bw; |
1baca4ce | 506 | |
029632fb PZ |
507 | /* |
508 | * The "RT overload" flag: it gets set if a CPU has more than | |
509 | * one runnable RT task. | |
510 | */ | |
511 | cpumask_var_t rto_mask; | |
512 | struct cpupri cpupri; | |
513 | }; | |
514 | ||
515 | extern struct root_domain def_root_domain; | |
516 | ||
517 | #endif /* CONFIG_SMP */ | |
518 | ||
519 | /* | |
520 | * This is the main, per-CPU runqueue data structure. | |
521 | * | |
522 | * Locking rule: those places that want to lock multiple runqueues | |
523 | * (such as the load balancing or the thread migration code), lock | |
524 | * acquire operations must be ordered by ascending &runqueue. | |
525 | */ | |
526 | struct rq { | |
527 | /* runqueue lock: */ | |
528 | raw_spinlock_t lock; | |
529 | ||
530 | /* | |
531 | * nr_running and cpu_load should be in the same cacheline because | |
532 | * remote CPUs use both these fields when doing load calculation. | |
533 | */ | |
c82513e5 | 534 | unsigned int nr_running; |
0ec8aa00 PZ |
535 | #ifdef CONFIG_NUMA_BALANCING |
536 | unsigned int nr_numa_running; | |
537 | unsigned int nr_preferred_running; | |
538 | #endif | |
029632fb PZ |
539 | #define CPU_LOAD_IDX_MAX 5 |
540 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
541 | unsigned long last_load_update_tick; | |
3451d024 | 542 | #ifdef CONFIG_NO_HZ_COMMON |
029632fb | 543 | u64 nohz_stamp; |
1c792db7 | 544 | unsigned long nohz_flags; |
265f22a9 FW |
545 | #endif |
546 | #ifdef CONFIG_NO_HZ_FULL | |
547 | unsigned long last_sched_tick; | |
029632fb PZ |
548 | #endif |
549 | int skip_clock_update; | |
550 | ||
551 | /* capture load from *all* tasks on this cpu: */ | |
552 | struct load_weight load; | |
553 | unsigned long nr_load_updates; | |
554 | u64 nr_switches; | |
555 | ||
556 | struct cfs_rq cfs; | |
557 | struct rt_rq rt; | |
aab03e05 | 558 | struct dl_rq dl; |
029632fb PZ |
559 | |
560 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
561 | /* list of leaf cfs_rq on this cpu: */ | |
562 | struct list_head leaf_cfs_rq_list; | |
a35b6466 PZ |
563 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
564 | ||
029632fb PZ |
565 | #ifdef CONFIG_RT_GROUP_SCHED |
566 | struct list_head leaf_rt_rq_list; | |
567 | #endif | |
568 | ||
569 | /* | |
570 | * This is part of a global counter where only the total sum | |
571 | * over all CPUs matters. A task can increase this counter on | |
572 | * one CPU and if it got migrated afterwards it may decrease | |
573 | * it on another CPU. Always updated under the runqueue lock: | |
574 | */ | |
575 | unsigned long nr_uninterruptible; | |
576 | ||
577 | struct task_struct *curr, *idle, *stop; | |
578 | unsigned long next_balance; | |
579 | struct mm_struct *prev_mm; | |
580 | ||
581 | u64 clock; | |
582 | u64 clock_task; | |
583 | ||
584 | atomic_t nr_iowait; | |
585 | ||
586 | #ifdef CONFIG_SMP | |
587 | struct root_domain *rd; | |
588 | struct sched_domain *sd; | |
589 | ||
590 | unsigned long cpu_power; | |
591 | ||
592 | unsigned char idle_balance; | |
593 | /* For active balancing */ | |
594 | int post_schedule; | |
595 | int active_balance; | |
596 | int push_cpu; | |
597 | struct cpu_stop_work active_balance_work; | |
598 | /* cpu of this runqueue: */ | |
599 | int cpu; | |
600 | int online; | |
601 | ||
367456c7 PZ |
602 | struct list_head cfs_tasks; |
603 | ||
029632fb PZ |
604 | u64 rt_avg; |
605 | u64 age_stamp; | |
606 | u64 idle_stamp; | |
607 | u64 avg_idle; | |
9bd721c5 JL |
608 | |
609 | /* This is used to determine avg_idle's max value */ | |
610 | u64 max_idle_balance_cost; | |
029632fb PZ |
611 | #endif |
612 | ||
613 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
614 | u64 prev_irq_time; | |
615 | #endif | |
616 | #ifdef CONFIG_PARAVIRT | |
617 | u64 prev_steal_time; | |
618 | #endif | |
619 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
620 | u64 prev_steal_time_rq; | |
621 | #endif | |
622 | ||
623 | /* calc_load related fields */ | |
624 | unsigned long calc_load_update; | |
625 | long calc_load_active; | |
626 | ||
627 | #ifdef CONFIG_SCHED_HRTICK | |
628 | #ifdef CONFIG_SMP | |
629 | int hrtick_csd_pending; | |
630 | struct call_single_data hrtick_csd; | |
631 | #endif | |
632 | struct hrtimer hrtick_timer; | |
633 | #endif | |
634 | ||
635 | #ifdef CONFIG_SCHEDSTATS | |
636 | /* latency stats */ | |
637 | struct sched_info rq_sched_info; | |
638 | unsigned long long rq_cpu_time; | |
639 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
640 | ||
641 | /* sys_sched_yield() stats */ | |
642 | unsigned int yld_count; | |
643 | ||
644 | /* schedule() stats */ | |
029632fb PZ |
645 | unsigned int sched_count; |
646 | unsigned int sched_goidle; | |
647 | ||
648 | /* try_to_wake_up() stats */ | |
649 | unsigned int ttwu_count; | |
650 | unsigned int ttwu_local; | |
651 | #endif | |
652 | ||
653 | #ifdef CONFIG_SMP | |
654 | struct llist_head wake_list; | |
655 | #endif | |
18bf2805 BS |
656 | |
657 | struct sched_avg avg; | |
029632fb PZ |
658 | }; |
659 | ||
660 | static inline int cpu_of(struct rq *rq) | |
661 | { | |
662 | #ifdef CONFIG_SMP | |
663 | return rq->cpu; | |
664 | #else | |
665 | return 0; | |
666 | #endif | |
667 | } | |
668 | ||
669 | DECLARE_PER_CPU(struct rq, runqueues); | |
670 | ||
518cd623 PZ |
671 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
672 | #define this_rq() (&__get_cpu_var(runqueues)) | |
673 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
674 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
675 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) | |
676 | ||
78becc27 FW |
677 | static inline u64 rq_clock(struct rq *rq) |
678 | { | |
679 | return rq->clock; | |
680 | } | |
681 | ||
682 | static inline u64 rq_clock_task(struct rq *rq) | |
683 | { | |
684 | return rq->clock_task; | |
685 | } | |
686 | ||
f809ca9a | 687 | #ifdef CONFIG_NUMA_BALANCING |
0ec8aa00 | 688 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 689 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 690 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
691 | #endif /* CONFIG_NUMA_BALANCING */ |
692 | ||
518cd623 PZ |
693 | #ifdef CONFIG_SMP |
694 | ||
029632fb PZ |
695 | #define rcu_dereference_check_sched_domain(p) \ |
696 | rcu_dereference_check((p), \ | |
697 | lockdep_is_held(&sched_domains_mutex)) | |
698 | ||
699 | /* | |
700 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
701 | * See detach_destroy_domains: synchronize_sched for details. | |
702 | * | |
703 | * The domain tree of any CPU may only be accessed from within | |
704 | * preempt-disabled sections. | |
705 | */ | |
706 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
707 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
708 | __sd; __sd = __sd->parent) | |
029632fb | 709 | |
77e81365 SS |
710 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
711 | ||
518cd623 PZ |
712 | /** |
713 | * highest_flag_domain - Return highest sched_domain containing flag. | |
714 | * @cpu: The cpu whose highest level of sched domain is to | |
715 | * be returned. | |
716 | * @flag: The flag to check for the highest sched_domain | |
717 | * for the given cpu. | |
718 | * | |
719 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
720 | */ | |
721 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
722 | { | |
723 | struct sched_domain *sd, *hsd = NULL; | |
724 | ||
725 | for_each_domain(cpu, sd) { | |
726 | if (!(sd->flags & flag)) | |
727 | break; | |
728 | hsd = sd; | |
729 | } | |
730 | ||
731 | return hsd; | |
732 | } | |
733 | ||
fb13c7ee MG |
734 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
735 | { | |
736 | struct sched_domain *sd; | |
737 | ||
738 | for_each_domain(cpu, sd) { | |
739 | if (sd->flags & flag) | |
740 | break; | |
741 | } | |
742 | ||
743 | return sd; | |
744 | } | |
745 | ||
518cd623 | 746 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 747 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 748 | DECLARE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 749 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
750 | DECLARE_PER_CPU(struct sched_domain *, sd_busy); |
751 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 | 752 | |
5e6521ea LZ |
753 | struct sched_group_power { |
754 | atomic_t ref; | |
755 | /* | |
756 | * CPU power of this group, SCHED_LOAD_SCALE being max power for a | |
757 | * single CPU. | |
758 | */ | |
759 | unsigned int power, power_orig; | |
760 | unsigned long next_update; | |
6263322c | 761 | int imbalance; /* XXX unrelated to power but shared group state */ |
5e6521ea LZ |
762 | /* |
763 | * Number of busy cpus in this group. | |
764 | */ | |
765 | atomic_t nr_busy_cpus; | |
766 | ||
767 | unsigned long cpumask[0]; /* iteration mask */ | |
768 | }; | |
769 | ||
770 | struct sched_group { | |
771 | struct sched_group *next; /* Must be a circular list */ | |
772 | atomic_t ref; | |
773 | ||
774 | unsigned int group_weight; | |
775 | struct sched_group_power *sgp; | |
776 | ||
777 | /* | |
778 | * The CPUs this group covers. | |
779 | * | |
780 | * NOTE: this field is variable length. (Allocated dynamically | |
781 | * by attaching extra space to the end of the structure, | |
782 | * depending on how many CPUs the kernel has booted up with) | |
783 | */ | |
784 | unsigned long cpumask[0]; | |
785 | }; | |
786 | ||
787 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) | |
788 | { | |
789 | return to_cpumask(sg->cpumask); | |
790 | } | |
791 | ||
792 | /* | |
793 | * cpumask masking which cpus in the group are allowed to iterate up the domain | |
794 | * tree. | |
795 | */ | |
796 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) | |
797 | { | |
798 | return to_cpumask(sg->sgp->cpumask); | |
799 | } | |
800 | ||
801 | /** | |
802 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
803 | * @group: The group whose first cpu is to be returned. | |
804 | */ | |
805 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
806 | { | |
807 | return cpumask_first(sched_group_cpus(group)); | |
808 | } | |
809 | ||
c1174876 PZ |
810 | extern int group_balance_cpu(struct sched_group *sg); |
811 | ||
518cd623 | 812 | #endif /* CONFIG_SMP */ |
029632fb | 813 | |
391e43da PZ |
814 | #include "stats.h" |
815 | #include "auto_group.h" | |
029632fb PZ |
816 | |
817 | #ifdef CONFIG_CGROUP_SCHED | |
818 | ||
819 | /* | |
820 | * Return the group to which this tasks belongs. | |
821 | * | |
8af01f56 TH |
822 | * We cannot use task_css() and friends because the cgroup subsystem |
823 | * changes that value before the cgroup_subsys::attach() method is called, | |
824 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
825 | * |
826 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
827 | * core changes this before calling sched_move_task(). | |
828 | * | |
829 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
830 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
831 | */ |
832 | static inline struct task_group *task_group(struct task_struct *p) | |
833 | { | |
8323f26c | 834 | return p->sched_task_group; |
029632fb PZ |
835 | } |
836 | ||
837 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
838 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
839 | { | |
840 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
841 | struct task_group *tg = task_group(p); | |
842 | #endif | |
843 | ||
844 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
845 | p->se.cfs_rq = tg->cfs_rq[cpu]; | |
846 | p->se.parent = tg->se[cpu]; | |
847 | #endif | |
848 | ||
849 | #ifdef CONFIG_RT_GROUP_SCHED | |
850 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
851 | p->rt.parent = tg->rt_se[cpu]; | |
852 | #endif | |
853 | } | |
854 | ||
855 | #else /* CONFIG_CGROUP_SCHED */ | |
856 | ||
857 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
858 | static inline struct task_group *task_group(struct task_struct *p) | |
859 | { | |
860 | return NULL; | |
861 | } | |
862 | ||
863 | #endif /* CONFIG_CGROUP_SCHED */ | |
864 | ||
865 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
866 | { | |
867 | set_task_rq(p, cpu); | |
868 | #ifdef CONFIG_SMP | |
869 | /* | |
870 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
871 | * successfuly executed on another CPU. We must ensure that updates of | |
872 | * per-task data have been completed by this moment. | |
873 | */ | |
874 | smp_wmb(); | |
875 | task_thread_info(p)->cpu = cpu; | |
ac66f547 | 876 | p->wake_cpu = cpu; |
029632fb PZ |
877 | #endif |
878 | } | |
879 | ||
880 | /* | |
881 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
882 | */ | |
883 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 884 | # include <linux/static_key.h> |
029632fb PZ |
885 | # define const_debug __read_mostly |
886 | #else | |
887 | # define const_debug const | |
888 | #endif | |
889 | ||
890 | extern const_debug unsigned int sysctl_sched_features; | |
891 | ||
892 | #define SCHED_FEAT(name, enabled) \ | |
893 | __SCHED_FEAT_##name , | |
894 | ||
895 | enum { | |
391e43da | 896 | #include "features.h" |
f8b6d1cc | 897 | __SCHED_FEAT_NR, |
029632fb PZ |
898 | }; |
899 | ||
900 | #undef SCHED_FEAT | |
901 | ||
f8b6d1cc | 902 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
c5905afb | 903 | static __always_inline bool static_branch__true(struct static_key *key) |
f8b6d1cc | 904 | { |
c5905afb | 905 | return static_key_true(key); /* Not out of line branch. */ |
f8b6d1cc PZ |
906 | } |
907 | ||
c5905afb | 908 | static __always_inline bool static_branch__false(struct static_key *key) |
f8b6d1cc | 909 | { |
c5905afb | 910 | return static_key_false(key); /* Out of line branch. */ |
f8b6d1cc PZ |
911 | } |
912 | ||
913 | #define SCHED_FEAT(name, enabled) \ | |
c5905afb | 914 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc PZ |
915 | { \ |
916 | return static_branch__##enabled(key); \ | |
917 | } | |
918 | ||
919 | #include "features.h" | |
920 | ||
921 | #undef SCHED_FEAT | |
922 | ||
c5905afb | 923 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
924 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
925 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 926 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 927 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 928 | |
cbee9f88 PZ |
929 | #ifdef CONFIG_NUMA_BALANCING |
930 | #define sched_feat_numa(x) sched_feat(x) | |
3105b86a MG |
931 | #ifdef CONFIG_SCHED_DEBUG |
932 | #define numabalancing_enabled sched_feat_numa(NUMA) | |
933 | #else | |
934 | extern bool numabalancing_enabled; | |
935 | #endif /* CONFIG_SCHED_DEBUG */ | |
cbee9f88 PZ |
936 | #else |
937 | #define sched_feat_numa(x) (0) | |
3105b86a MG |
938 | #define numabalancing_enabled (0) |
939 | #endif /* CONFIG_NUMA_BALANCING */ | |
cbee9f88 | 940 | |
029632fb PZ |
941 | static inline u64 global_rt_period(void) |
942 | { | |
943 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
944 | } | |
945 | ||
946 | static inline u64 global_rt_runtime(void) | |
947 | { | |
948 | if (sysctl_sched_rt_runtime < 0) | |
949 | return RUNTIME_INF; | |
950 | ||
951 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
952 | } | |
953 | ||
332ac17e DF |
954 | static inline u64 global_dl_period(void) |
955 | { | |
956 | return (u64)sysctl_sched_dl_period * NSEC_PER_USEC; | |
957 | } | |
958 | ||
959 | static inline u64 global_dl_runtime(void) | |
960 | { | |
961 | if (sysctl_sched_dl_runtime < 0) | |
962 | return RUNTIME_INF; | |
029632fb | 963 | |
332ac17e DF |
964 | return (u64)sysctl_sched_dl_runtime * NSEC_PER_USEC; |
965 | } | |
029632fb PZ |
966 | |
967 | static inline int task_current(struct rq *rq, struct task_struct *p) | |
968 | { | |
969 | return rq->curr == p; | |
970 | } | |
971 | ||
972 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
973 | { | |
974 | #ifdef CONFIG_SMP | |
975 | return p->on_cpu; | |
976 | #else | |
977 | return task_current(rq, p); | |
978 | #endif | |
979 | } | |
980 | ||
981 | ||
982 | #ifndef prepare_arch_switch | |
983 | # define prepare_arch_switch(next) do { } while (0) | |
984 | #endif | |
985 | #ifndef finish_arch_switch | |
986 | # define finish_arch_switch(prev) do { } while (0) | |
987 | #endif | |
01f23e16 CM |
988 | #ifndef finish_arch_post_lock_switch |
989 | # define finish_arch_post_lock_switch() do { } while (0) | |
990 | #endif | |
029632fb PZ |
991 | |
992 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
993 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
994 | { | |
995 | #ifdef CONFIG_SMP | |
996 | /* | |
997 | * We can optimise this out completely for !SMP, because the | |
998 | * SMP rebalancing from interrupt is the only thing that cares | |
999 | * here. | |
1000 | */ | |
1001 | next->on_cpu = 1; | |
1002 | #endif | |
1003 | } | |
1004 | ||
1005 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1006 | { | |
1007 | #ifdef CONFIG_SMP | |
1008 | /* | |
1009 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1010 | * We must ensure this doesn't happen until the switch is completely | |
1011 | * finished. | |
1012 | */ | |
1013 | smp_wmb(); | |
1014 | prev->on_cpu = 0; | |
1015 | #endif | |
1016 | #ifdef CONFIG_DEBUG_SPINLOCK | |
1017 | /* this is a valid case when another task releases the spinlock */ | |
1018 | rq->lock.owner = current; | |
1019 | #endif | |
1020 | /* | |
1021 | * If we are tracking spinlock dependencies then we have to | |
1022 | * fix up the runqueue lock - which gets 'carried over' from | |
1023 | * prev into current: | |
1024 | */ | |
1025 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1026 | ||
1027 | raw_spin_unlock_irq(&rq->lock); | |
1028 | } | |
1029 | ||
1030 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1031 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) | |
1032 | { | |
1033 | #ifdef CONFIG_SMP | |
1034 | /* | |
1035 | * We can optimise this out completely for !SMP, because the | |
1036 | * SMP rebalancing from interrupt is the only thing that cares | |
1037 | * here. | |
1038 | */ | |
1039 | next->on_cpu = 1; | |
1040 | #endif | |
029632fb | 1041 | raw_spin_unlock(&rq->lock); |
029632fb PZ |
1042 | } |
1043 | ||
1044 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1045 | { | |
1046 | #ifdef CONFIG_SMP | |
1047 | /* | |
1048 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1049 | * We must ensure this doesn't happen until the switch is completely | |
1050 | * finished. | |
1051 | */ | |
1052 | smp_wmb(); | |
1053 | prev->on_cpu = 0; | |
1054 | #endif | |
029632fb | 1055 | local_irq_enable(); |
029632fb PZ |
1056 | } |
1057 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1058 | ||
b13095f0 LZ |
1059 | /* |
1060 | * wake flags | |
1061 | */ | |
1062 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1063 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1064 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1065 | ||
029632fb PZ |
1066 | /* |
1067 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1068 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1069 | * each task makes to its run queue's load is weighted according to its | |
1070 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1071 | * scaled version of the new time slice allocation that they receive on time | |
1072 | * slice expiry etc. | |
1073 | */ | |
1074 | ||
1075 | #define WEIGHT_IDLEPRIO 3 | |
1076 | #define WMULT_IDLEPRIO 1431655765 | |
1077 | ||
1078 | /* | |
1079 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1080 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1081 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1082 | * that remained on nice 0. | |
1083 | * | |
1084 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1085 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
1086 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
1087 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1088 | * the relative distance between them is ~25%.) | |
1089 | */ | |
1090 | static const int prio_to_weight[40] = { | |
1091 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
1092 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1093 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1094 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1095 | /* 0 */ 1024, 820, 655, 526, 423, | |
1096 | /* 5 */ 335, 272, 215, 172, 137, | |
1097 | /* 10 */ 110, 87, 70, 56, 45, | |
1098 | /* 15 */ 36, 29, 23, 18, 15, | |
1099 | }; | |
1100 | ||
1101 | /* | |
1102 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1103 | * | |
1104 | * In cases where the weight does not change often, we can use the | |
1105 | * precalculated inverse to speed up arithmetics by turning divisions | |
1106 | * into multiplications: | |
1107 | */ | |
1108 | static const u32 prio_to_wmult[40] = { | |
1109 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
1110 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1111 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1112 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1113 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1114 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1115 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1116 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
1117 | }; | |
1118 | ||
c82ba9fa LZ |
1119 | #define ENQUEUE_WAKEUP 1 |
1120 | #define ENQUEUE_HEAD 2 | |
1121 | #ifdef CONFIG_SMP | |
1122 | #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */ | |
1123 | #else | |
1124 | #define ENQUEUE_WAKING 0 | |
1125 | #endif | |
aab03e05 | 1126 | #define ENQUEUE_REPLENISH 8 |
c82ba9fa LZ |
1127 | |
1128 | #define DEQUEUE_SLEEP 1 | |
1129 | ||
1130 | struct sched_class { | |
1131 | const struct sched_class *next; | |
1132 | ||
1133 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1134 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1135 | void (*yield_task) (struct rq *rq); | |
1136 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1137 | ||
1138 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1139 | ||
1140 | struct task_struct * (*pick_next_task) (struct rq *rq); | |
1141 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); | |
1142 | ||
1143 | #ifdef CONFIG_SMP | |
ac66f547 | 1144 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
c82ba9fa LZ |
1145 | void (*migrate_task_rq)(struct task_struct *p, int next_cpu); |
1146 | ||
1147 | void (*pre_schedule) (struct rq *this_rq, struct task_struct *task); | |
1148 | void (*post_schedule) (struct rq *this_rq); | |
1149 | void (*task_waking) (struct task_struct *task); | |
1150 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); | |
1151 | ||
1152 | void (*set_cpus_allowed)(struct task_struct *p, | |
1153 | const struct cpumask *newmask); | |
1154 | ||
1155 | void (*rq_online)(struct rq *rq); | |
1156 | void (*rq_offline)(struct rq *rq); | |
1157 | #endif | |
1158 | ||
1159 | void (*set_curr_task) (struct rq *rq); | |
1160 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1161 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1162 | void (*task_dead) (struct task_struct *p); |
c82ba9fa LZ |
1163 | |
1164 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); | |
1165 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1166 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1167 | int oldprio); | |
1168 | ||
1169 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1170 | struct task_struct *task); | |
1171 | ||
1172 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1173 | void (*task_move_group) (struct task_struct *p, int on_rq); | |
1174 | #endif | |
1175 | }; | |
029632fb PZ |
1176 | |
1177 | #define sched_class_highest (&stop_sched_class) | |
1178 | #define for_each_class(class) \ | |
1179 | for (class = sched_class_highest; class; class = class->next) | |
1180 | ||
1181 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1182 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1183 | extern const struct sched_class rt_sched_class; |
1184 | extern const struct sched_class fair_sched_class; | |
1185 | extern const struct sched_class idle_sched_class; | |
1186 | ||
1187 | ||
1188 | #ifdef CONFIG_SMP | |
1189 | ||
b719203b LZ |
1190 | extern void update_group_power(struct sched_domain *sd, int cpu); |
1191 | ||
029632fb PZ |
1192 | extern void trigger_load_balance(struct rq *rq, int cpu); |
1193 | extern void idle_balance(int this_cpu, struct rq *this_rq); | |
1194 | ||
642dbc39 VG |
1195 | extern void idle_enter_fair(struct rq *this_rq); |
1196 | extern void idle_exit_fair(struct rq *this_rq); | |
642dbc39 | 1197 | |
029632fb PZ |
1198 | #else /* CONFIG_SMP */ |
1199 | ||
1200 | static inline void idle_balance(int cpu, struct rq *rq) | |
1201 | { | |
1202 | } | |
1203 | ||
1204 | #endif | |
1205 | ||
1206 | extern void sysrq_sched_debug_show(void); | |
1207 | extern void sched_init_granularity(void); | |
1208 | extern void update_max_interval(void); | |
1baca4ce JL |
1209 | |
1210 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1211 | extern void init_sched_rt_class(void); |
1212 | extern void init_sched_fair_class(void); | |
332ac17e | 1213 | extern void init_sched_dl_class(void); |
029632fb PZ |
1214 | |
1215 | extern void resched_task(struct task_struct *p); | |
1216 | extern void resched_cpu(int cpu); | |
1217 | ||
1218 | extern struct rt_bandwidth def_rt_bandwidth; | |
1219 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1220 | ||
332ac17e DF |
1221 | extern struct dl_bandwidth def_dl_bandwidth; |
1222 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 DF |
1223 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
1224 | ||
332ac17e DF |
1225 | unsigned long to_ratio(u64 period, u64 runtime); |
1226 | ||
556061b0 | 1227 | extern void update_idle_cpu_load(struct rq *this_rq); |
029632fb | 1228 | |
a75cdaa9 AS |
1229 | extern void init_task_runnable_average(struct task_struct *p); |
1230 | ||
73fbec60 FW |
1231 | #ifdef CONFIG_PARAVIRT |
1232 | static inline u64 steal_ticks(u64 steal) | |
1233 | { | |
1234 | if (unlikely(steal > NSEC_PER_SEC)) | |
1235 | return div_u64(steal, TICK_NSEC); | |
1236 | ||
1237 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); | |
1238 | } | |
1239 | #endif | |
1240 | ||
029632fb PZ |
1241 | static inline void inc_nr_running(struct rq *rq) |
1242 | { | |
1243 | rq->nr_running++; | |
9f3660c2 FW |
1244 | |
1245 | #ifdef CONFIG_NO_HZ_FULL | |
1246 | if (rq->nr_running == 2) { | |
1247 | if (tick_nohz_full_cpu(rq->cpu)) { | |
1248 | /* Order rq->nr_running write against the IPI */ | |
1249 | smp_wmb(); | |
1250 | smp_send_reschedule(rq->cpu); | |
1251 | } | |
1252 | } | |
1253 | #endif | |
029632fb PZ |
1254 | } |
1255 | ||
1256 | static inline void dec_nr_running(struct rq *rq) | |
1257 | { | |
1258 | rq->nr_running--; | |
1259 | } | |
1260 | ||
265f22a9 FW |
1261 | static inline void rq_last_tick_reset(struct rq *rq) |
1262 | { | |
1263 | #ifdef CONFIG_NO_HZ_FULL | |
1264 | rq->last_sched_tick = jiffies; | |
1265 | #endif | |
1266 | } | |
1267 | ||
029632fb PZ |
1268 | extern void update_rq_clock(struct rq *rq); |
1269 | ||
1270 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1271 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1272 | ||
1273 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1274 | ||
1275 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1276 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1277 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1278 | ||
1279 | static inline u64 sched_avg_period(void) | |
1280 | { | |
1281 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1282 | } | |
1283 | ||
029632fb PZ |
1284 | #ifdef CONFIG_SCHED_HRTICK |
1285 | ||
1286 | /* | |
1287 | * Use hrtick when: | |
1288 | * - enabled by features | |
1289 | * - hrtimer is actually high res | |
1290 | */ | |
1291 | static inline int hrtick_enabled(struct rq *rq) | |
1292 | { | |
1293 | if (!sched_feat(HRTICK)) | |
1294 | return 0; | |
1295 | if (!cpu_active(cpu_of(rq))) | |
1296 | return 0; | |
1297 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1298 | } | |
1299 | ||
1300 | void hrtick_start(struct rq *rq, u64 delay); | |
1301 | ||
b39e66ea MG |
1302 | #else |
1303 | ||
1304 | static inline int hrtick_enabled(struct rq *rq) | |
1305 | { | |
1306 | return 0; | |
1307 | } | |
1308 | ||
029632fb PZ |
1309 | #endif /* CONFIG_SCHED_HRTICK */ |
1310 | ||
1311 | #ifdef CONFIG_SMP | |
1312 | extern void sched_avg_update(struct rq *rq); | |
1313 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1314 | { | |
1315 | rq->rt_avg += rt_delta; | |
1316 | sched_avg_update(rq); | |
1317 | } | |
1318 | #else | |
1319 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1320 | static inline void sched_avg_update(struct rq *rq) { } | |
1321 | #endif | |
1322 | ||
1323 | extern void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period); | |
1324 | ||
1325 | #ifdef CONFIG_SMP | |
1326 | #ifdef CONFIG_PREEMPT | |
1327 | ||
1328 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1329 | ||
1330 | /* | |
1331 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1332 | * way at the expense of forcing extra atomic operations in all | |
1333 | * invocations. This assures that the double_lock is acquired using the | |
1334 | * same underlying policy as the spinlock_t on this architecture, which | |
1335 | * reduces latency compared to the unfair variant below. However, it | |
1336 | * also adds more overhead and therefore may reduce throughput. | |
1337 | */ | |
1338 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1339 | __releases(this_rq->lock) | |
1340 | __acquires(busiest->lock) | |
1341 | __acquires(this_rq->lock) | |
1342 | { | |
1343 | raw_spin_unlock(&this_rq->lock); | |
1344 | double_rq_lock(this_rq, busiest); | |
1345 | ||
1346 | return 1; | |
1347 | } | |
1348 | ||
1349 | #else | |
1350 | /* | |
1351 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1352 | * latency by eliminating extra atomic operations when the locks are | |
1353 | * already in proper order on entry. This favors lower cpu-ids and will | |
1354 | * grant the double lock to lower cpus over higher ids under contention, | |
1355 | * regardless of entry order into the function. | |
1356 | */ | |
1357 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1358 | __releases(this_rq->lock) | |
1359 | __acquires(busiest->lock) | |
1360 | __acquires(this_rq->lock) | |
1361 | { | |
1362 | int ret = 0; | |
1363 | ||
1364 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1365 | if (busiest < this_rq) { | |
1366 | raw_spin_unlock(&this_rq->lock); | |
1367 | raw_spin_lock(&busiest->lock); | |
1368 | raw_spin_lock_nested(&this_rq->lock, | |
1369 | SINGLE_DEPTH_NESTING); | |
1370 | ret = 1; | |
1371 | } else | |
1372 | raw_spin_lock_nested(&busiest->lock, | |
1373 | SINGLE_DEPTH_NESTING); | |
1374 | } | |
1375 | return ret; | |
1376 | } | |
1377 | ||
1378 | #endif /* CONFIG_PREEMPT */ | |
1379 | ||
1380 | /* | |
1381 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1382 | */ | |
1383 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1384 | { | |
1385 | if (unlikely(!irqs_disabled())) { | |
1386 | /* printk() doesn't work good under rq->lock */ | |
1387 | raw_spin_unlock(&this_rq->lock); | |
1388 | BUG_ON(1); | |
1389 | } | |
1390 | ||
1391 | return _double_lock_balance(this_rq, busiest); | |
1392 | } | |
1393 | ||
1394 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1395 | __releases(busiest->lock) | |
1396 | { | |
1397 | raw_spin_unlock(&busiest->lock); | |
1398 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1399 | } | |
1400 | ||
74602315 PZ |
1401 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1402 | { | |
1403 | if (l1 > l2) | |
1404 | swap(l1, l2); | |
1405 | ||
1406 | spin_lock(l1); | |
1407 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1408 | } | |
1409 | ||
1410 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) | |
1411 | { | |
1412 | if (l1 > l2) | |
1413 | swap(l1, l2); | |
1414 | ||
1415 | raw_spin_lock(l1); | |
1416 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1417 | } | |
1418 | ||
029632fb PZ |
1419 | /* |
1420 | * double_rq_lock - safely lock two runqueues | |
1421 | * | |
1422 | * Note this does not disable interrupts like task_rq_lock, | |
1423 | * you need to do so manually before calling. | |
1424 | */ | |
1425 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1426 | __acquires(rq1->lock) | |
1427 | __acquires(rq2->lock) | |
1428 | { | |
1429 | BUG_ON(!irqs_disabled()); | |
1430 | if (rq1 == rq2) { | |
1431 | raw_spin_lock(&rq1->lock); | |
1432 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1433 | } else { | |
1434 | if (rq1 < rq2) { | |
1435 | raw_spin_lock(&rq1->lock); | |
1436 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1437 | } else { | |
1438 | raw_spin_lock(&rq2->lock); | |
1439 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1440 | } | |
1441 | } | |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * double_rq_unlock - safely unlock two runqueues | |
1446 | * | |
1447 | * Note this does not restore interrupts like task_rq_unlock, | |
1448 | * you need to do so manually after calling. | |
1449 | */ | |
1450 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1451 | __releases(rq1->lock) | |
1452 | __releases(rq2->lock) | |
1453 | { | |
1454 | raw_spin_unlock(&rq1->lock); | |
1455 | if (rq1 != rq2) | |
1456 | raw_spin_unlock(&rq2->lock); | |
1457 | else | |
1458 | __release(rq2->lock); | |
1459 | } | |
1460 | ||
1461 | #else /* CONFIG_SMP */ | |
1462 | ||
1463 | /* | |
1464 | * double_rq_lock - safely lock two runqueues | |
1465 | * | |
1466 | * Note this does not disable interrupts like task_rq_lock, | |
1467 | * you need to do so manually before calling. | |
1468 | */ | |
1469 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1470 | __acquires(rq1->lock) | |
1471 | __acquires(rq2->lock) | |
1472 | { | |
1473 | BUG_ON(!irqs_disabled()); | |
1474 | BUG_ON(rq1 != rq2); | |
1475 | raw_spin_lock(&rq1->lock); | |
1476 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * double_rq_unlock - safely unlock two runqueues | |
1481 | * | |
1482 | * Note this does not restore interrupts like task_rq_unlock, | |
1483 | * you need to do so manually after calling. | |
1484 | */ | |
1485 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1486 | __releases(rq1->lock) | |
1487 | __releases(rq2->lock) | |
1488 | { | |
1489 | BUG_ON(rq1 != rq2); | |
1490 | raw_spin_unlock(&rq1->lock); | |
1491 | __release(rq2->lock); | |
1492 | } | |
1493 | ||
1494 | #endif | |
1495 | ||
1496 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1497 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
1498 | extern void print_cfs_stats(struct seq_file *m, int cpu); | |
1499 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
1500 | ||
1501 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
1502 | extern void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq); | |
aab03e05 | 1503 | extern void init_dl_rq(struct dl_rq *dl_rq, struct rq *rq); |
029632fb | 1504 | |
1ee14e6c BS |
1505 | extern void cfs_bandwidth_usage_inc(void); |
1506 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1507 | |
3451d024 | 1508 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1509 | enum rq_nohz_flag_bits { |
1510 | NOHZ_TICK_STOPPED, | |
1511 | NOHZ_BALANCE_KICK, | |
1512 | }; | |
1513 | ||
1514 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
1515 | #endif | |
73fbec60 FW |
1516 | |
1517 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
1518 | ||
1519 | DECLARE_PER_CPU(u64, cpu_hardirq_time); | |
1520 | DECLARE_PER_CPU(u64, cpu_softirq_time); | |
1521 | ||
1522 | #ifndef CONFIG_64BIT | |
1523 | DECLARE_PER_CPU(seqcount_t, irq_time_seq); | |
1524 | ||
1525 | static inline void irq_time_write_begin(void) | |
1526 | { | |
1527 | __this_cpu_inc(irq_time_seq.sequence); | |
1528 | smp_wmb(); | |
1529 | } | |
1530 | ||
1531 | static inline void irq_time_write_end(void) | |
1532 | { | |
1533 | smp_wmb(); | |
1534 | __this_cpu_inc(irq_time_seq.sequence); | |
1535 | } | |
1536 | ||
1537 | static inline u64 irq_time_read(int cpu) | |
1538 | { | |
1539 | u64 irq_time; | |
1540 | unsigned seq; | |
1541 | ||
1542 | do { | |
1543 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1544 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1545 | per_cpu(cpu_hardirq_time, cpu); | |
1546 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1547 | ||
1548 | return irq_time; | |
1549 | } | |
1550 | #else /* CONFIG_64BIT */ | |
1551 | static inline void irq_time_write_begin(void) | |
1552 | { | |
1553 | } | |
1554 | ||
1555 | static inline void irq_time_write_end(void) | |
1556 | { | |
1557 | } | |
1558 | ||
1559 | static inline u64 irq_time_read(int cpu) | |
1560 | { | |
1561 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1562 | } | |
1563 | #endif /* CONFIG_64BIT */ | |
1564 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |