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> | |
b6366f04 | 9 | #include <linux/irq_work.h> |
9f3660c2 | 10 | #include <linux/tick.h> |
f809ca9a | 11 | #include <linux/slab.h> |
029632fb | 12 | |
391e43da | 13 | #include "cpupri.h" |
6bfd6d72 | 14 | #include "cpudeadline.h" |
60fed789 | 15 | #include "cpuacct.h" |
029632fb | 16 | |
45ceebf7 | 17 | struct rq; |
442bf3aa | 18 | struct cpuidle_state; |
45ceebf7 | 19 | |
da0c1e65 KT |
20 | /* task_struct::on_rq states: */ |
21 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 22 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 23 | |
029632fb PZ |
24 | extern __read_mostly int scheduler_running; |
25 | ||
45ceebf7 PG |
26 | extern unsigned long calc_load_update; |
27 | extern atomic_long_t calc_load_tasks; | |
28 | ||
3289bdb4 | 29 | extern void calc_global_load_tick(struct rq *this_rq); |
45ceebf7 | 30 | extern long calc_load_fold_active(struct rq *this_rq); |
3289bdb4 PZ |
31 | |
32 | #ifdef CONFIG_SMP | |
45ceebf7 | 33 | extern void update_cpu_load_active(struct rq *this_rq); |
3289bdb4 PZ |
34 | #else |
35 | static inline void update_cpu_load_active(struct rq *this_rq) { } | |
36 | #endif | |
45ceebf7 | 37 | |
029632fb PZ |
38 | /* |
39 | * Helpers for converting nanosecond timing to jiffy resolution | |
40 | */ | |
41 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
42 | ||
cc1f4b1f LZ |
43 | /* |
44 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
45 | * The extra resolution improves shares distribution and load balancing of | |
46 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
47 | * hierarchies, especially on larger systems. This is not a user-visible change | |
48 | * and does not change the user-interface for setting shares/weights. | |
49 | * | |
50 | * We increase resolution only if we have enough bits to allow this increased | |
51 | * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution | |
52 | * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the | |
53 | * increased costs. | |
54 | */ | |
55 | #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */ | |
56 | # define SCHED_LOAD_RESOLUTION 10 | |
57 | # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION) | |
58 | # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION) | |
59 | #else | |
60 | # define SCHED_LOAD_RESOLUTION 0 | |
61 | # define scale_load(w) (w) | |
62 | # define scale_load_down(w) (w) | |
63 | #endif | |
64 | ||
65 | #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION) | |
66 | #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT) | |
67 | ||
029632fb PZ |
68 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
69 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
70 | ||
332ac17e DF |
71 | /* |
72 | * Single value that decides SCHED_DEADLINE internal math precision. | |
73 | * 10 -> just above 1us | |
74 | * 9 -> just above 0.5us | |
75 | */ | |
76 | #define DL_SCALE (10) | |
77 | ||
029632fb PZ |
78 | /* |
79 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 80 | */ |
029632fb PZ |
81 | |
82 | /* | |
83 | * single value that denotes runtime == period, ie unlimited time. | |
84 | */ | |
85 | #define RUNTIME_INF ((u64)~0ULL) | |
86 | ||
d50dde5a DF |
87 | static inline int fair_policy(int policy) |
88 | { | |
89 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
90 | } | |
91 | ||
029632fb PZ |
92 | static inline int rt_policy(int policy) |
93 | { | |
d50dde5a | 94 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
95 | } |
96 | ||
aab03e05 DF |
97 | static inline int dl_policy(int policy) |
98 | { | |
99 | return policy == SCHED_DEADLINE; | |
100 | } | |
101 | ||
029632fb PZ |
102 | static inline int task_has_rt_policy(struct task_struct *p) |
103 | { | |
104 | return rt_policy(p->policy); | |
105 | } | |
106 | ||
aab03e05 DF |
107 | static inline int task_has_dl_policy(struct task_struct *p) |
108 | { | |
109 | return dl_policy(p->policy); | |
110 | } | |
111 | ||
332ac17e | 112 | static inline bool dl_time_before(u64 a, u64 b) |
2d3d891d DF |
113 | { |
114 | return (s64)(a - b) < 0; | |
115 | } | |
116 | ||
117 | /* | |
118 | * Tells if entity @a should preempt entity @b. | |
119 | */ | |
332ac17e DF |
120 | static inline bool |
121 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
122 | { |
123 | return dl_time_before(a->deadline, b->deadline); | |
124 | } | |
125 | ||
029632fb PZ |
126 | /* |
127 | * This is the priority-queue data structure of the RT scheduling class: | |
128 | */ | |
129 | struct rt_prio_array { | |
130 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
131 | struct list_head queue[MAX_RT_PRIO]; | |
132 | }; | |
133 | ||
134 | struct rt_bandwidth { | |
135 | /* nests inside the rq lock: */ | |
136 | raw_spinlock_t rt_runtime_lock; | |
137 | ktime_t rt_period; | |
138 | u64 rt_runtime; | |
139 | struct hrtimer rt_period_timer; | |
4cfafd30 | 140 | unsigned int rt_period_active; |
029632fb | 141 | }; |
a5e7be3b JL |
142 | |
143 | void __dl_clear_params(struct task_struct *p); | |
144 | ||
332ac17e DF |
145 | /* |
146 | * To keep the bandwidth of -deadline tasks and groups under control | |
147 | * we need some place where: | |
148 | * - store the maximum -deadline bandwidth of the system (the group); | |
149 | * - cache the fraction of that bandwidth that is currently allocated. | |
150 | * | |
151 | * This is all done in the data structure below. It is similar to the | |
152 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
153 | * that, since here we are only interested in admission control, we | |
154 | * do not decrease any runtime while the group "executes", neither we | |
155 | * need a timer to replenish it. | |
156 | * | |
157 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
158 | * meaning that: | |
159 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
160 | * - dl_total_bw array contains, in the i-eth element, the currently | |
161 | * allocated bandwidth on the i-eth CPU. | |
162 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
163 | * consume bandwidth on the CPU they're running on. | |
164 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
165 | * that will be shown the next time the proc or cgroup controls will | |
166 | * be red. It on its turn can be changed by writing on its own | |
167 | * control. | |
168 | */ | |
169 | struct dl_bandwidth { | |
170 | raw_spinlock_t dl_runtime_lock; | |
171 | u64 dl_runtime; | |
172 | u64 dl_period; | |
173 | }; | |
174 | ||
175 | static inline int dl_bandwidth_enabled(void) | |
176 | { | |
1724813d | 177 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
178 | } |
179 | ||
180 | extern struct dl_bw *dl_bw_of(int i); | |
181 | ||
182 | struct dl_bw { | |
183 | raw_spinlock_t lock; | |
184 | u64 bw, total_bw; | |
185 | }; | |
186 | ||
7f51412a JL |
187 | static inline |
188 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
189 | { | |
190 | dl_b->total_bw -= tsk_bw; | |
191 | } | |
192 | ||
193 | static inline | |
194 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
195 | { | |
196 | dl_b->total_bw += tsk_bw; | |
197 | } | |
198 | ||
199 | static inline | |
200 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
201 | { | |
202 | return dl_b->bw != -1 && | |
203 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
204 | } | |
205 | ||
029632fb PZ |
206 | extern struct mutex sched_domains_mutex; |
207 | ||
208 | #ifdef CONFIG_CGROUP_SCHED | |
209 | ||
210 | #include <linux/cgroup.h> | |
211 | ||
212 | struct cfs_rq; | |
213 | struct rt_rq; | |
214 | ||
35cf4e50 | 215 | extern struct list_head task_groups; |
029632fb PZ |
216 | |
217 | struct cfs_bandwidth { | |
218 | #ifdef CONFIG_CFS_BANDWIDTH | |
219 | raw_spinlock_t lock; | |
220 | ktime_t period; | |
221 | u64 quota, runtime; | |
9c58c79a | 222 | s64 hierarchical_quota; |
029632fb PZ |
223 | u64 runtime_expires; |
224 | ||
4cfafd30 | 225 | int idle, period_active; |
029632fb PZ |
226 | struct hrtimer period_timer, slack_timer; |
227 | struct list_head throttled_cfs_rq; | |
228 | ||
229 | /* statistics */ | |
230 | int nr_periods, nr_throttled; | |
231 | u64 throttled_time; | |
232 | #endif | |
233 | }; | |
234 | ||
235 | /* task group related information */ | |
236 | struct task_group { | |
237 | struct cgroup_subsys_state css; | |
238 | ||
239 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
240 | /* schedulable entities of this group on each cpu */ | |
241 | struct sched_entity **se; | |
242 | /* runqueue "owned" by this group on each cpu */ | |
243 | struct cfs_rq **cfs_rq; | |
244 | unsigned long shares; | |
245 | ||
fa6bddeb | 246 | #ifdef CONFIG_SMP |
bf5b986e | 247 | atomic_long_t load_avg; |
bb17f655 | 248 | atomic_t runnable_avg; |
029632fb | 249 | #endif |
fa6bddeb | 250 | #endif |
029632fb PZ |
251 | |
252 | #ifdef CONFIG_RT_GROUP_SCHED | |
253 | struct sched_rt_entity **rt_se; | |
254 | struct rt_rq **rt_rq; | |
255 | ||
256 | struct rt_bandwidth rt_bandwidth; | |
257 | #endif | |
258 | ||
259 | struct rcu_head rcu; | |
260 | struct list_head list; | |
261 | ||
262 | struct task_group *parent; | |
263 | struct list_head siblings; | |
264 | struct list_head children; | |
265 | ||
266 | #ifdef CONFIG_SCHED_AUTOGROUP | |
267 | struct autogroup *autogroup; | |
268 | #endif | |
269 | ||
270 | struct cfs_bandwidth cfs_bandwidth; | |
271 | }; | |
272 | ||
273 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
274 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
275 | ||
276 | /* | |
277 | * A weight of 0 or 1 can cause arithmetics problems. | |
278 | * A weight of a cfs_rq is the sum of weights of which entities | |
279 | * are queued on this cfs_rq, so a weight of a entity should not be | |
280 | * too large, so as the shares value of a task group. | |
281 | * (The default weight is 1024 - so there's no practical | |
282 | * limitation from this.) | |
283 | */ | |
284 | #define MIN_SHARES (1UL << 1) | |
285 | #define MAX_SHARES (1UL << 18) | |
286 | #endif | |
287 | ||
029632fb PZ |
288 | typedef int (*tg_visitor)(struct task_group *, void *); |
289 | ||
290 | extern int walk_tg_tree_from(struct task_group *from, | |
291 | tg_visitor down, tg_visitor up, void *data); | |
292 | ||
293 | /* | |
294 | * Iterate the full tree, calling @down when first entering a node and @up when | |
295 | * leaving it for the final time. | |
296 | * | |
297 | * Caller must hold rcu_lock or sufficient equivalent. | |
298 | */ | |
299 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
300 | { | |
301 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
302 | } | |
303 | ||
304 | extern int tg_nop(struct task_group *tg, void *data); | |
305 | ||
306 | extern void free_fair_sched_group(struct task_group *tg); | |
307 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
308 | extern void unregister_fair_sched_group(struct task_group *tg, int cpu); | |
309 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, | |
310 | struct sched_entity *se, int cpu, | |
311 | struct sched_entity *parent); | |
312 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
313 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
314 | ||
315 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 316 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
317 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
318 | ||
319 | extern void free_rt_sched_group(struct task_group *tg); | |
320 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
321 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
322 | struct sched_rt_entity *rt_se, int cpu, | |
323 | struct sched_rt_entity *parent); | |
324 | ||
25cc7da7 LZ |
325 | extern struct task_group *sched_create_group(struct task_group *parent); |
326 | extern void sched_online_group(struct task_group *tg, | |
327 | struct task_group *parent); | |
328 | extern void sched_destroy_group(struct task_group *tg); | |
329 | extern void sched_offline_group(struct task_group *tg); | |
330 | ||
331 | extern void sched_move_task(struct task_struct *tsk); | |
332 | ||
333 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
334 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
335 | #endif | |
336 | ||
029632fb PZ |
337 | #else /* CONFIG_CGROUP_SCHED */ |
338 | ||
339 | struct cfs_bandwidth { }; | |
340 | ||
341 | #endif /* CONFIG_CGROUP_SCHED */ | |
342 | ||
343 | /* CFS-related fields in a runqueue */ | |
344 | struct cfs_rq { | |
345 | struct load_weight load; | |
c82513e5 | 346 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
347 | |
348 | u64 exec_clock; | |
349 | u64 min_vruntime; | |
350 | #ifndef CONFIG_64BIT | |
351 | u64 min_vruntime_copy; | |
352 | #endif | |
353 | ||
354 | struct rb_root tasks_timeline; | |
355 | struct rb_node *rb_leftmost; | |
356 | ||
029632fb PZ |
357 | /* |
358 | * 'curr' points to currently running entity on this cfs_rq. | |
359 | * It is set to NULL otherwise (i.e when none are currently running). | |
360 | */ | |
361 | struct sched_entity *curr, *next, *last, *skip; | |
362 | ||
363 | #ifdef CONFIG_SCHED_DEBUG | |
364 | unsigned int nr_spread_over; | |
365 | #endif | |
366 | ||
2dac754e PT |
367 | #ifdef CONFIG_SMP |
368 | /* | |
369 | * CFS Load tracking | |
370 | * Under CFS, load is tracked on a per-entity basis and aggregated up. | |
371 | * This allows for the description of both thread and group usage (in | |
372 | * the FAIR_GROUP_SCHED case). | |
36ee28e4 VG |
373 | * runnable_load_avg is the sum of the load_avg_contrib of the |
374 | * sched_entities on the rq. | |
375 | * blocked_load_avg is similar to runnable_load_avg except that its | |
376 | * the blocked sched_entities on the rq. | |
377 | * utilization_load_avg is the sum of the average running time of the | |
378 | * sched_entities on the rq. | |
2dac754e | 379 | */ |
36ee28e4 | 380 | unsigned long runnable_load_avg, blocked_load_avg, utilization_load_avg; |
2509940f | 381 | atomic64_t decay_counter; |
9ee474f5 | 382 | u64 last_decay; |
2509940f | 383 | atomic_long_t removed_load; |
141965c7 | 384 | |
c566e8e9 | 385 | #ifdef CONFIG_FAIR_GROUP_SCHED |
141965c7 | 386 | /* Required to track per-cpu representation of a task_group */ |
bb17f655 | 387 | u32 tg_runnable_contrib; |
bf5b986e | 388 | unsigned long tg_load_contrib; |
82958366 PT |
389 | |
390 | /* | |
391 | * h_load = weight * f(tg) | |
392 | * | |
393 | * Where f(tg) is the recursive weight fraction assigned to | |
394 | * this group. | |
395 | */ | |
396 | unsigned long h_load; | |
68520796 VD |
397 | u64 last_h_load_update; |
398 | struct sched_entity *h_load_next; | |
399 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
400 | #endif /* CONFIG_SMP */ |
401 | ||
029632fb PZ |
402 | #ifdef CONFIG_FAIR_GROUP_SCHED |
403 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
404 | ||
405 | /* | |
406 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
407 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
408 | * (like users, containers etc.) | |
409 | * | |
410 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
411 | * list is used during load balance. | |
412 | */ | |
413 | int on_list; | |
414 | struct list_head leaf_cfs_rq_list; | |
415 | struct task_group *tg; /* group that "owns" this runqueue */ | |
416 | ||
029632fb PZ |
417 | #ifdef CONFIG_CFS_BANDWIDTH |
418 | int runtime_enabled; | |
419 | u64 runtime_expires; | |
420 | s64 runtime_remaining; | |
421 | ||
f1b17280 PT |
422 | u64 throttled_clock, throttled_clock_task; |
423 | u64 throttled_clock_task_time; | |
029632fb PZ |
424 | int throttled, throttle_count; |
425 | struct list_head throttled_list; | |
426 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
427 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
428 | }; | |
429 | ||
430 | static inline int rt_bandwidth_enabled(void) | |
431 | { | |
432 | return sysctl_sched_rt_runtime >= 0; | |
433 | } | |
434 | ||
b6366f04 SR |
435 | /* RT IPI pull logic requires IRQ_WORK */ |
436 | #ifdef CONFIG_IRQ_WORK | |
437 | # define HAVE_RT_PUSH_IPI | |
438 | #endif | |
439 | ||
029632fb PZ |
440 | /* Real-Time classes' related field in a runqueue: */ |
441 | struct rt_rq { | |
442 | struct rt_prio_array active; | |
c82513e5 | 443 | unsigned int rt_nr_running; |
029632fb PZ |
444 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
445 | struct { | |
446 | int curr; /* highest queued rt task prio */ | |
447 | #ifdef CONFIG_SMP | |
448 | int next; /* next highest */ | |
449 | #endif | |
450 | } highest_prio; | |
451 | #endif | |
452 | #ifdef CONFIG_SMP | |
453 | unsigned long rt_nr_migratory; | |
454 | unsigned long rt_nr_total; | |
455 | int overloaded; | |
456 | struct plist_head pushable_tasks; | |
b6366f04 SR |
457 | #ifdef HAVE_RT_PUSH_IPI |
458 | int push_flags; | |
459 | int push_cpu; | |
460 | struct irq_work push_work; | |
461 | raw_spinlock_t push_lock; | |
029632fb | 462 | #endif |
b6366f04 | 463 | #endif /* CONFIG_SMP */ |
f4ebcbc0 KT |
464 | int rt_queued; |
465 | ||
029632fb PZ |
466 | int rt_throttled; |
467 | u64 rt_time; | |
468 | u64 rt_runtime; | |
469 | /* Nests inside the rq lock: */ | |
470 | raw_spinlock_t rt_runtime_lock; | |
471 | ||
472 | #ifdef CONFIG_RT_GROUP_SCHED | |
473 | unsigned long rt_nr_boosted; | |
474 | ||
475 | struct rq *rq; | |
029632fb PZ |
476 | struct task_group *tg; |
477 | #endif | |
478 | }; | |
479 | ||
aab03e05 DF |
480 | /* Deadline class' related fields in a runqueue */ |
481 | struct dl_rq { | |
482 | /* runqueue is an rbtree, ordered by deadline */ | |
483 | struct rb_root rb_root; | |
484 | struct rb_node *rb_leftmost; | |
485 | ||
486 | unsigned long dl_nr_running; | |
1baca4ce JL |
487 | |
488 | #ifdef CONFIG_SMP | |
489 | /* | |
490 | * Deadline values of the currently executing and the | |
491 | * earliest ready task on this rq. Caching these facilitates | |
492 | * the decision wether or not a ready but not running task | |
493 | * should migrate somewhere else. | |
494 | */ | |
495 | struct { | |
496 | u64 curr; | |
497 | u64 next; | |
498 | } earliest_dl; | |
499 | ||
500 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
501 | int overloaded; |
502 | ||
503 | /* | |
504 | * Tasks on this rq that can be pushed away. They are kept in | |
505 | * an rb-tree, ordered by tasks' deadlines, with caching | |
506 | * of the leftmost (earliest deadline) element. | |
507 | */ | |
508 | struct rb_root pushable_dl_tasks_root; | |
509 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
510 | #else |
511 | struct dl_bw dl_bw; | |
1baca4ce | 512 | #endif |
aab03e05 DF |
513 | }; |
514 | ||
029632fb PZ |
515 | #ifdef CONFIG_SMP |
516 | ||
517 | /* | |
518 | * We add the notion of a root-domain which will be used to define per-domain | |
519 | * variables. Each exclusive cpuset essentially defines an island domain by | |
520 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
521 | * exclusive cpuset is created, we also create and attach a new root-domain | |
522 | * object. | |
523 | * | |
524 | */ | |
525 | struct root_domain { | |
526 | atomic_t refcount; | |
527 | atomic_t rto_count; | |
528 | struct rcu_head rcu; | |
529 | cpumask_var_t span; | |
530 | cpumask_var_t online; | |
531 | ||
4486edd1 TC |
532 | /* Indicate more than one runnable task for any CPU */ |
533 | bool overload; | |
534 | ||
1baca4ce JL |
535 | /* |
536 | * The bit corresponding to a CPU gets set here if such CPU has more | |
537 | * than one runnable -deadline task (as it is below for RT tasks). | |
538 | */ | |
539 | cpumask_var_t dlo_mask; | |
540 | atomic_t dlo_count; | |
332ac17e | 541 | struct dl_bw dl_bw; |
6bfd6d72 | 542 | struct cpudl cpudl; |
1baca4ce | 543 | |
029632fb PZ |
544 | /* |
545 | * The "RT overload" flag: it gets set if a CPU has more than | |
546 | * one runnable RT task. | |
547 | */ | |
548 | cpumask_var_t rto_mask; | |
549 | struct cpupri cpupri; | |
550 | }; | |
551 | ||
552 | extern struct root_domain def_root_domain; | |
553 | ||
554 | #endif /* CONFIG_SMP */ | |
555 | ||
556 | /* | |
557 | * This is the main, per-CPU runqueue data structure. | |
558 | * | |
559 | * Locking rule: those places that want to lock multiple runqueues | |
560 | * (such as the load balancing or the thread migration code), lock | |
561 | * acquire operations must be ordered by ascending &runqueue. | |
562 | */ | |
563 | struct rq { | |
564 | /* runqueue lock: */ | |
565 | raw_spinlock_t lock; | |
566 | ||
567 | /* | |
568 | * nr_running and cpu_load should be in the same cacheline because | |
569 | * remote CPUs use both these fields when doing load calculation. | |
570 | */ | |
c82513e5 | 571 | unsigned int nr_running; |
0ec8aa00 PZ |
572 | #ifdef CONFIG_NUMA_BALANCING |
573 | unsigned int nr_numa_running; | |
574 | unsigned int nr_preferred_running; | |
575 | #endif | |
029632fb PZ |
576 | #define CPU_LOAD_IDX_MAX 5 |
577 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
578 | unsigned long last_load_update_tick; | |
3451d024 | 579 | #ifdef CONFIG_NO_HZ_COMMON |
029632fb | 580 | u64 nohz_stamp; |
1c792db7 | 581 | unsigned long nohz_flags; |
265f22a9 FW |
582 | #endif |
583 | #ifdef CONFIG_NO_HZ_FULL | |
584 | unsigned long last_sched_tick; | |
029632fb | 585 | #endif |
029632fb PZ |
586 | /* capture load from *all* tasks on this cpu: */ |
587 | struct load_weight load; | |
588 | unsigned long nr_load_updates; | |
589 | u64 nr_switches; | |
590 | ||
591 | struct cfs_rq cfs; | |
592 | struct rt_rq rt; | |
aab03e05 | 593 | struct dl_rq dl; |
029632fb PZ |
594 | |
595 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
596 | /* list of leaf cfs_rq on this cpu: */ | |
597 | struct list_head leaf_cfs_rq_list; | |
f5f9739d DE |
598 | |
599 | struct sched_avg avg; | |
a35b6466 PZ |
600 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
601 | ||
029632fb PZ |
602 | /* |
603 | * This is part of a global counter where only the total sum | |
604 | * over all CPUs matters. A task can increase this counter on | |
605 | * one CPU and if it got migrated afterwards it may decrease | |
606 | * it on another CPU. Always updated under the runqueue lock: | |
607 | */ | |
608 | unsigned long nr_uninterruptible; | |
609 | ||
610 | struct task_struct *curr, *idle, *stop; | |
611 | unsigned long next_balance; | |
612 | struct mm_struct *prev_mm; | |
613 | ||
9edfbfed | 614 | unsigned int clock_skip_update; |
029632fb PZ |
615 | u64 clock; |
616 | u64 clock_task; | |
617 | ||
618 | atomic_t nr_iowait; | |
619 | ||
620 | #ifdef CONFIG_SMP | |
621 | struct root_domain *rd; | |
622 | struct sched_domain *sd; | |
623 | ||
ced549fa | 624 | unsigned long cpu_capacity; |
ca6d75e6 | 625 | unsigned long cpu_capacity_orig; |
029632fb | 626 | |
e3fca9e7 PZ |
627 | struct callback_head *balance_callback; |
628 | ||
029632fb PZ |
629 | unsigned char idle_balance; |
630 | /* For active balancing */ | |
029632fb PZ |
631 | int active_balance; |
632 | int push_cpu; | |
633 | struct cpu_stop_work active_balance_work; | |
634 | /* cpu of this runqueue: */ | |
635 | int cpu; | |
636 | int online; | |
637 | ||
367456c7 PZ |
638 | struct list_head cfs_tasks; |
639 | ||
029632fb PZ |
640 | u64 rt_avg; |
641 | u64 age_stamp; | |
642 | u64 idle_stamp; | |
643 | u64 avg_idle; | |
9bd721c5 JL |
644 | |
645 | /* This is used to determine avg_idle's max value */ | |
646 | u64 max_idle_balance_cost; | |
029632fb PZ |
647 | #endif |
648 | ||
649 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
650 | u64 prev_irq_time; | |
651 | #endif | |
652 | #ifdef CONFIG_PARAVIRT | |
653 | u64 prev_steal_time; | |
654 | #endif | |
655 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
656 | u64 prev_steal_time_rq; | |
657 | #endif | |
658 | ||
659 | /* calc_load related fields */ | |
660 | unsigned long calc_load_update; | |
661 | long calc_load_active; | |
662 | ||
663 | #ifdef CONFIG_SCHED_HRTICK | |
664 | #ifdef CONFIG_SMP | |
665 | int hrtick_csd_pending; | |
666 | struct call_single_data hrtick_csd; | |
667 | #endif | |
668 | struct hrtimer hrtick_timer; | |
669 | #endif | |
670 | ||
671 | #ifdef CONFIG_SCHEDSTATS | |
672 | /* latency stats */ | |
673 | struct sched_info rq_sched_info; | |
674 | unsigned long long rq_cpu_time; | |
675 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
676 | ||
677 | /* sys_sched_yield() stats */ | |
678 | unsigned int yld_count; | |
679 | ||
680 | /* schedule() stats */ | |
029632fb PZ |
681 | unsigned int sched_count; |
682 | unsigned int sched_goidle; | |
683 | ||
684 | /* try_to_wake_up() stats */ | |
685 | unsigned int ttwu_count; | |
686 | unsigned int ttwu_local; | |
687 | #endif | |
688 | ||
689 | #ifdef CONFIG_SMP | |
690 | struct llist_head wake_list; | |
691 | #endif | |
442bf3aa DL |
692 | |
693 | #ifdef CONFIG_CPU_IDLE | |
694 | /* Must be inspected within a rcu lock section */ | |
695 | struct cpuidle_state *idle_state; | |
696 | #endif | |
029632fb PZ |
697 | }; |
698 | ||
699 | static inline int cpu_of(struct rq *rq) | |
700 | { | |
701 | #ifdef CONFIG_SMP | |
702 | return rq->cpu; | |
703 | #else | |
704 | return 0; | |
705 | #endif | |
706 | } | |
707 | ||
8b06c55b | 708 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 709 | |
518cd623 | 710 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 711 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
712 | #define task_rq(p) cpu_rq(task_cpu(p)) |
713 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 714 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 715 | |
cebde6d6 PZ |
716 | static inline u64 __rq_clock_broken(struct rq *rq) |
717 | { | |
316c1608 | 718 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
719 | } |
720 | ||
78becc27 FW |
721 | static inline u64 rq_clock(struct rq *rq) |
722 | { | |
cebde6d6 | 723 | lockdep_assert_held(&rq->lock); |
78becc27 FW |
724 | return rq->clock; |
725 | } | |
726 | ||
727 | static inline u64 rq_clock_task(struct rq *rq) | |
728 | { | |
cebde6d6 | 729 | lockdep_assert_held(&rq->lock); |
78becc27 FW |
730 | return rq->clock_task; |
731 | } | |
732 | ||
9edfbfed PZ |
733 | #define RQCF_REQ_SKIP 0x01 |
734 | #define RQCF_ACT_SKIP 0x02 | |
735 | ||
736 | static inline void rq_clock_skip_update(struct rq *rq, bool skip) | |
737 | { | |
738 | lockdep_assert_held(&rq->lock); | |
739 | if (skip) | |
740 | rq->clock_skip_update |= RQCF_REQ_SKIP; | |
741 | else | |
742 | rq->clock_skip_update &= ~RQCF_REQ_SKIP; | |
743 | } | |
744 | ||
9942f79b | 745 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
746 | enum numa_topology_type { |
747 | NUMA_DIRECT, | |
748 | NUMA_GLUELESS_MESH, | |
749 | NUMA_BACKPLANE, | |
750 | }; | |
751 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
752 | extern int sched_max_numa_distance; |
753 | extern bool find_numa_distance(int distance); | |
754 | #endif | |
755 | ||
f809ca9a | 756 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
757 | /* The regions in numa_faults array from task_struct */ |
758 | enum numa_faults_stats { | |
759 | NUMA_MEM = 0, | |
760 | NUMA_CPU, | |
761 | NUMA_MEMBUF, | |
762 | NUMA_CPUBUF | |
763 | }; | |
0ec8aa00 | 764 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 765 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 766 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
767 | #endif /* CONFIG_NUMA_BALANCING */ |
768 | ||
518cd623 PZ |
769 | #ifdef CONFIG_SMP |
770 | ||
e3fca9e7 PZ |
771 | static inline void |
772 | queue_balance_callback(struct rq *rq, | |
773 | struct callback_head *head, | |
774 | void (*func)(struct rq *rq)) | |
775 | { | |
776 | lockdep_assert_held(&rq->lock); | |
777 | ||
778 | if (unlikely(head->next)) | |
779 | return; | |
780 | ||
781 | head->func = (void (*)(struct callback_head *))func; | |
782 | head->next = rq->balance_callback; | |
783 | rq->balance_callback = head; | |
784 | } | |
785 | ||
e3baac47 PZ |
786 | extern void sched_ttwu_pending(void); |
787 | ||
029632fb PZ |
788 | #define rcu_dereference_check_sched_domain(p) \ |
789 | rcu_dereference_check((p), \ | |
790 | lockdep_is_held(&sched_domains_mutex)) | |
791 | ||
792 | /* | |
793 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
794 | * See detach_destroy_domains: synchronize_sched for details. | |
795 | * | |
796 | * The domain tree of any CPU may only be accessed from within | |
797 | * preempt-disabled sections. | |
798 | */ | |
799 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
800 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
801 | __sd; __sd = __sd->parent) | |
029632fb | 802 | |
77e81365 SS |
803 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
804 | ||
518cd623 PZ |
805 | /** |
806 | * highest_flag_domain - Return highest sched_domain containing flag. | |
807 | * @cpu: The cpu whose highest level of sched domain is to | |
808 | * be returned. | |
809 | * @flag: The flag to check for the highest sched_domain | |
810 | * for the given cpu. | |
811 | * | |
812 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
813 | */ | |
814 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
815 | { | |
816 | struct sched_domain *sd, *hsd = NULL; | |
817 | ||
818 | for_each_domain(cpu, sd) { | |
819 | if (!(sd->flags & flag)) | |
820 | break; | |
821 | hsd = sd; | |
822 | } | |
823 | ||
824 | return hsd; | |
825 | } | |
826 | ||
fb13c7ee MG |
827 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
828 | { | |
829 | struct sched_domain *sd; | |
830 | ||
831 | for_each_domain(cpu, sd) { | |
832 | if (sd->flags & flag) | |
833 | break; | |
834 | } | |
835 | ||
836 | return sd; | |
837 | } | |
838 | ||
518cd623 | 839 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 840 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 841 | DECLARE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 842 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
843 | DECLARE_PER_CPU(struct sched_domain *, sd_busy); |
844 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 | 845 | |
63b2ca30 | 846 | struct sched_group_capacity { |
5e6521ea LZ |
847 | atomic_t ref; |
848 | /* | |
63b2ca30 NP |
849 | * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity |
850 | * for a single CPU. | |
5e6521ea | 851 | */ |
dc7ff76e | 852 | unsigned int capacity; |
5e6521ea | 853 | unsigned long next_update; |
63b2ca30 | 854 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea LZ |
855 | /* |
856 | * Number of busy cpus in this group. | |
857 | */ | |
858 | atomic_t nr_busy_cpus; | |
859 | ||
860 | unsigned long cpumask[0]; /* iteration mask */ | |
861 | }; | |
862 | ||
863 | struct sched_group { | |
864 | struct sched_group *next; /* Must be a circular list */ | |
865 | atomic_t ref; | |
866 | ||
867 | unsigned int group_weight; | |
63b2ca30 | 868 | struct sched_group_capacity *sgc; |
5e6521ea LZ |
869 | |
870 | /* | |
871 | * The CPUs this group covers. | |
872 | * | |
873 | * NOTE: this field is variable length. (Allocated dynamically | |
874 | * by attaching extra space to the end of the structure, | |
875 | * depending on how many CPUs the kernel has booted up with) | |
876 | */ | |
877 | unsigned long cpumask[0]; | |
878 | }; | |
879 | ||
880 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) | |
881 | { | |
882 | return to_cpumask(sg->cpumask); | |
883 | } | |
884 | ||
885 | /* | |
886 | * cpumask masking which cpus in the group are allowed to iterate up the domain | |
887 | * tree. | |
888 | */ | |
889 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) | |
890 | { | |
63b2ca30 | 891 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
892 | } |
893 | ||
894 | /** | |
895 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
896 | * @group: The group whose first cpu is to be returned. | |
897 | */ | |
898 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
899 | { | |
900 | return cpumask_first(sched_group_cpus(group)); | |
901 | } | |
902 | ||
c1174876 PZ |
903 | extern int group_balance_cpu(struct sched_group *sg); |
904 | ||
e3baac47 PZ |
905 | #else |
906 | ||
907 | static inline void sched_ttwu_pending(void) { } | |
908 | ||
518cd623 | 909 | #endif /* CONFIG_SMP */ |
029632fb | 910 | |
391e43da PZ |
911 | #include "stats.h" |
912 | #include "auto_group.h" | |
029632fb PZ |
913 | |
914 | #ifdef CONFIG_CGROUP_SCHED | |
915 | ||
916 | /* | |
917 | * Return the group to which this tasks belongs. | |
918 | * | |
8af01f56 TH |
919 | * We cannot use task_css() and friends because the cgroup subsystem |
920 | * changes that value before the cgroup_subsys::attach() method is called, | |
921 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
922 | * |
923 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
924 | * core changes this before calling sched_move_task(). | |
925 | * | |
926 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
927 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
928 | */ |
929 | static inline struct task_group *task_group(struct task_struct *p) | |
930 | { | |
8323f26c | 931 | return p->sched_task_group; |
029632fb PZ |
932 | } |
933 | ||
934 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
935 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
936 | { | |
937 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
938 | struct task_group *tg = task_group(p); | |
939 | #endif | |
940 | ||
941 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
942 | p->se.cfs_rq = tg->cfs_rq[cpu]; | |
943 | p->se.parent = tg->se[cpu]; | |
944 | #endif | |
945 | ||
946 | #ifdef CONFIG_RT_GROUP_SCHED | |
947 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
948 | p->rt.parent = tg->rt_se[cpu]; | |
949 | #endif | |
950 | } | |
951 | ||
952 | #else /* CONFIG_CGROUP_SCHED */ | |
953 | ||
954 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
955 | static inline struct task_group *task_group(struct task_struct *p) | |
956 | { | |
957 | return NULL; | |
958 | } | |
959 | ||
960 | #endif /* CONFIG_CGROUP_SCHED */ | |
961 | ||
962 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
963 | { | |
964 | set_task_rq(p, cpu); | |
965 | #ifdef CONFIG_SMP | |
966 | /* | |
967 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
968 | * successfuly executed on another CPU. We must ensure that updates of | |
969 | * per-task data have been completed by this moment. | |
970 | */ | |
971 | smp_wmb(); | |
972 | task_thread_info(p)->cpu = cpu; | |
ac66f547 | 973 | p->wake_cpu = cpu; |
029632fb PZ |
974 | #endif |
975 | } | |
976 | ||
977 | /* | |
978 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
979 | */ | |
980 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 981 | # include <linux/static_key.h> |
029632fb PZ |
982 | # define const_debug __read_mostly |
983 | #else | |
984 | # define const_debug const | |
985 | #endif | |
986 | ||
987 | extern const_debug unsigned int sysctl_sched_features; | |
988 | ||
989 | #define SCHED_FEAT(name, enabled) \ | |
990 | __SCHED_FEAT_##name , | |
991 | ||
992 | enum { | |
391e43da | 993 | #include "features.h" |
f8b6d1cc | 994 | __SCHED_FEAT_NR, |
029632fb PZ |
995 | }; |
996 | ||
997 | #undef SCHED_FEAT | |
998 | ||
f8b6d1cc | 999 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
f8b6d1cc | 1000 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1001 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1002 | { \ |
6e76ea8a | 1003 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1004 | } |
1005 | ||
1006 | #include "features.h" | |
1007 | ||
1008 | #undef SCHED_FEAT | |
1009 | ||
c5905afb | 1010 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
1011 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
1012 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 1013 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 1014 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 1015 | |
cbee9f88 PZ |
1016 | #ifdef CONFIG_NUMA_BALANCING |
1017 | #define sched_feat_numa(x) sched_feat(x) | |
3105b86a MG |
1018 | #ifdef CONFIG_SCHED_DEBUG |
1019 | #define numabalancing_enabled sched_feat_numa(NUMA) | |
1020 | #else | |
1021 | extern bool numabalancing_enabled; | |
1022 | #endif /* CONFIG_SCHED_DEBUG */ | |
cbee9f88 PZ |
1023 | #else |
1024 | #define sched_feat_numa(x) (0) | |
3105b86a MG |
1025 | #define numabalancing_enabled (0) |
1026 | #endif /* CONFIG_NUMA_BALANCING */ | |
cbee9f88 | 1027 | |
029632fb PZ |
1028 | static inline u64 global_rt_period(void) |
1029 | { | |
1030 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1031 | } | |
1032 | ||
1033 | static inline u64 global_rt_runtime(void) | |
1034 | { | |
1035 | if (sysctl_sched_rt_runtime < 0) | |
1036 | return RUNTIME_INF; | |
1037 | ||
1038 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1039 | } | |
1040 | ||
029632fb PZ |
1041 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1042 | { | |
1043 | return rq->curr == p; | |
1044 | } | |
1045 | ||
1046 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1047 | { | |
1048 | #ifdef CONFIG_SMP | |
1049 | return p->on_cpu; | |
1050 | #else | |
1051 | return task_current(rq, p); | |
1052 | #endif | |
1053 | } | |
1054 | ||
da0c1e65 KT |
1055 | static inline int task_on_rq_queued(struct task_struct *p) |
1056 | { | |
1057 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1058 | } | |
029632fb | 1059 | |
cca26e80 KT |
1060 | static inline int task_on_rq_migrating(struct task_struct *p) |
1061 | { | |
1062 | return p->on_rq == TASK_ON_RQ_MIGRATING; | |
1063 | } | |
1064 | ||
029632fb PZ |
1065 | #ifndef prepare_arch_switch |
1066 | # define prepare_arch_switch(next) do { } while (0) | |
1067 | #endif | |
1068 | #ifndef finish_arch_switch | |
1069 | # define finish_arch_switch(prev) do { } while (0) | |
1070 | #endif | |
01f23e16 CM |
1071 | #ifndef finish_arch_post_lock_switch |
1072 | # define finish_arch_post_lock_switch() do { } while (0) | |
1073 | #endif | |
029632fb | 1074 | |
029632fb PZ |
1075 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
1076 | { | |
1077 | #ifdef CONFIG_SMP | |
1078 | /* | |
1079 | * We can optimise this out completely for !SMP, because the | |
1080 | * SMP rebalancing from interrupt is the only thing that cares | |
1081 | * here. | |
1082 | */ | |
1083 | next->on_cpu = 1; | |
1084 | #endif | |
1085 | } | |
1086 | ||
1087 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1088 | { | |
1089 | #ifdef CONFIG_SMP | |
1090 | /* | |
1091 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1092 | * We must ensure this doesn't happen until the switch is completely | |
1093 | * finished. | |
1094 | */ | |
1095 | smp_wmb(); | |
1096 | prev->on_cpu = 0; | |
1097 | #endif | |
1098 | #ifdef CONFIG_DEBUG_SPINLOCK | |
1099 | /* this is a valid case when another task releases the spinlock */ | |
1100 | rq->lock.owner = current; | |
1101 | #endif | |
1102 | /* | |
1103 | * If we are tracking spinlock dependencies then we have to | |
1104 | * fix up the runqueue lock - which gets 'carried over' from | |
1105 | * prev into current: | |
1106 | */ | |
1107 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1108 | ||
1109 | raw_spin_unlock_irq(&rq->lock); | |
1110 | } | |
1111 | ||
b13095f0 LZ |
1112 | /* |
1113 | * wake flags | |
1114 | */ | |
1115 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1116 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1117 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1118 | ||
029632fb PZ |
1119 | /* |
1120 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1121 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1122 | * each task makes to its run queue's load is weighted according to its | |
1123 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1124 | * scaled version of the new time slice allocation that they receive on time | |
1125 | * slice expiry etc. | |
1126 | */ | |
1127 | ||
1128 | #define WEIGHT_IDLEPRIO 3 | |
1129 | #define WMULT_IDLEPRIO 1431655765 | |
1130 | ||
1131 | /* | |
1132 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1133 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1134 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1135 | * that remained on nice 0. | |
1136 | * | |
1137 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1138 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
1139 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
1140 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1141 | * the relative distance between them is ~25%.) | |
1142 | */ | |
1143 | static const int prio_to_weight[40] = { | |
1144 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
1145 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1146 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1147 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1148 | /* 0 */ 1024, 820, 655, 526, 423, | |
1149 | /* 5 */ 335, 272, 215, 172, 137, | |
1150 | /* 10 */ 110, 87, 70, 56, 45, | |
1151 | /* 15 */ 36, 29, 23, 18, 15, | |
1152 | }; | |
1153 | ||
1154 | /* | |
1155 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1156 | * | |
1157 | * In cases where the weight does not change often, we can use the | |
1158 | * precalculated inverse to speed up arithmetics by turning divisions | |
1159 | * into multiplications: | |
1160 | */ | |
1161 | static const u32 prio_to_wmult[40] = { | |
1162 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
1163 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1164 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1165 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1166 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1167 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1168 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1169 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
1170 | }; | |
1171 | ||
c82ba9fa LZ |
1172 | #define ENQUEUE_WAKEUP 1 |
1173 | #define ENQUEUE_HEAD 2 | |
1174 | #ifdef CONFIG_SMP | |
1175 | #define ENQUEUE_WAKING 4 /* sched_class::task_waking was called */ | |
1176 | #else | |
1177 | #define ENQUEUE_WAKING 0 | |
1178 | #endif | |
aab03e05 | 1179 | #define ENQUEUE_REPLENISH 8 |
c82ba9fa LZ |
1180 | |
1181 | #define DEQUEUE_SLEEP 1 | |
1182 | ||
37e117c0 PZ |
1183 | #define RETRY_TASK ((void *)-1UL) |
1184 | ||
c82ba9fa LZ |
1185 | struct sched_class { |
1186 | const struct sched_class *next; | |
1187 | ||
1188 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1189 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1190 | void (*yield_task) (struct rq *rq); | |
1191 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1192 | ||
1193 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1194 | ||
606dba2e PZ |
1195 | /* |
1196 | * It is the responsibility of the pick_next_task() method that will | |
1197 | * return the next task to call put_prev_task() on the @prev task or | |
1198 | * something equivalent. | |
37e117c0 PZ |
1199 | * |
1200 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1201 | * tasks. | |
606dba2e PZ |
1202 | */ |
1203 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
1204 | struct task_struct *prev); | |
c82ba9fa LZ |
1205 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1206 | ||
1207 | #ifdef CONFIG_SMP | |
ac66f547 | 1208 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
c82ba9fa LZ |
1209 | void (*migrate_task_rq)(struct task_struct *p, int next_cpu); |
1210 | ||
c82ba9fa LZ |
1211 | void (*task_waking) (struct task_struct *task); |
1212 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); | |
1213 | ||
1214 | void (*set_cpus_allowed)(struct task_struct *p, | |
1215 | const struct cpumask *newmask); | |
1216 | ||
1217 | void (*rq_online)(struct rq *rq); | |
1218 | void (*rq_offline)(struct rq *rq); | |
1219 | #endif | |
1220 | ||
1221 | void (*set_curr_task) (struct rq *rq); | |
1222 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1223 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1224 | void (*task_dead) (struct task_struct *p); |
c82ba9fa | 1225 | |
67dfa1b7 KT |
1226 | /* |
1227 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1228 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1229 | * rq->lock. They are however serialized by p->pi_lock. | |
1230 | */ | |
c82ba9fa LZ |
1231 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); |
1232 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1233 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1234 | int oldprio); | |
1235 | ||
1236 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1237 | struct task_struct *task); | |
1238 | ||
6e998916 SG |
1239 | void (*update_curr) (struct rq *rq); |
1240 | ||
c82ba9fa LZ |
1241 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1242 | void (*task_move_group) (struct task_struct *p, int on_rq); | |
1243 | #endif | |
1244 | }; | |
029632fb | 1245 | |
3f1d2a31 PZ |
1246 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1247 | { | |
1248 | prev->sched_class->put_prev_task(rq, prev); | |
1249 | } | |
1250 | ||
029632fb PZ |
1251 | #define sched_class_highest (&stop_sched_class) |
1252 | #define for_each_class(class) \ | |
1253 | for (class = sched_class_highest; class; class = class->next) | |
1254 | ||
1255 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1256 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1257 | extern const struct sched_class rt_sched_class; |
1258 | extern const struct sched_class fair_sched_class; | |
1259 | extern const struct sched_class idle_sched_class; | |
1260 | ||
1261 | ||
1262 | #ifdef CONFIG_SMP | |
1263 | ||
63b2ca30 | 1264 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1265 | |
7caff66f | 1266 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1267 | |
642dbc39 VG |
1268 | extern void idle_enter_fair(struct rq *this_rq); |
1269 | extern void idle_exit_fair(struct rq *this_rq); | |
642dbc39 | 1270 | |
dc877341 PZ |
1271 | #else |
1272 | ||
1273 | static inline void idle_enter_fair(struct rq *rq) { } | |
1274 | static inline void idle_exit_fair(struct rq *rq) { } | |
1275 | ||
029632fb PZ |
1276 | #endif |
1277 | ||
442bf3aa DL |
1278 | #ifdef CONFIG_CPU_IDLE |
1279 | static inline void idle_set_state(struct rq *rq, | |
1280 | struct cpuidle_state *idle_state) | |
1281 | { | |
1282 | rq->idle_state = idle_state; | |
1283 | } | |
1284 | ||
1285 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1286 | { | |
1287 | WARN_ON(!rcu_read_lock_held()); | |
1288 | return rq->idle_state; | |
1289 | } | |
1290 | #else | |
1291 | static inline void idle_set_state(struct rq *rq, | |
1292 | struct cpuidle_state *idle_state) | |
1293 | { | |
1294 | } | |
1295 | ||
1296 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1297 | { | |
1298 | return NULL; | |
1299 | } | |
1300 | #endif | |
1301 | ||
029632fb PZ |
1302 | extern void sysrq_sched_debug_show(void); |
1303 | extern void sched_init_granularity(void); | |
1304 | extern void update_max_interval(void); | |
1baca4ce JL |
1305 | |
1306 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1307 | extern void init_sched_rt_class(void); |
1308 | extern void init_sched_fair_class(void); | |
1309 | ||
8875125e | 1310 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1311 | extern void resched_cpu(int cpu); |
1312 | ||
1313 | extern struct rt_bandwidth def_rt_bandwidth; | |
1314 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1315 | ||
332ac17e DF |
1316 | extern struct dl_bandwidth def_dl_bandwidth; |
1317 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 DF |
1318 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
1319 | ||
332ac17e DF |
1320 | unsigned long to_ratio(u64 period, u64 runtime); |
1321 | ||
a75cdaa9 AS |
1322 | extern void init_task_runnable_average(struct task_struct *p); |
1323 | ||
72465447 | 1324 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1325 | { |
72465447 KT |
1326 | unsigned prev_nr = rq->nr_running; |
1327 | ||
1328 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1329 | |
72465447 | 1330 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1331 | #ifdef CONFIG_SMP |
1332 | if (!rq->rd->overload) | |
1333 | rq->rd->overload = true; | |
1334 | #endif | |
1335 | ||
1336 | #ifdef CONFIG_NO_HZ_FULL | |
9f3660c2 | 1337 | if (tick_nohz_full_cpu(rq->cpu)) { |
3882ec64 FW |
1338 | /* |
1339 | * Tick is needed if more than one task runs on a CPU. | |
1340 | * Send the target an IPI to kick it out of nohz mode. | |
1341 | * | |
1342 | * We assume that IPI implies full memory barrier and the | |
1343 | * new value of rq->nr_running is visible on reception | |
1344 | * from the target. | |
1345 | */ | |
fd2ac4f4 | 1346 | tick_nohz_full_kick_cpu(rq->cpu); |
9f3660c2 | 1347 | } |
9f3660c2 | 1348 | #endif |
4486edd1 | 1349 | } |
029632fb PZ |
1350 | } |
1351 | ||
72465447 | 1352 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1353 | { |
72465447 | 1354 | rq->nr_running -= count; |
029632fb PZ |
1355 | } |
1356 | ||
265f22a9 FW |
1357 | static inline void rq_last_tick_reset(struct rq *rq) |
1358 | { | |
1359 | #ifdef CONFIG_NO_HZ_FULL | |
1360 | rq->last_sched_tick = jiffies; | |
1361 | #endif | |
1362 | } | |
1363 | ||
029632fb PZ |
1364 | extern void update_rq_clock(struct rq *rq); |
1365 | ||
1366 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1367 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1368 | ||
1369 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1370 | ||
1371 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1372 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1373 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1374 | ||
1375 | static inline u64 sched_avg_period(void) | |
1376 | { | |
1377 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1378 | } | |
1379 | ||
029632fb PZ |
1380 | #ifdef CONFIG_SCHED_HRTICK |
1381 | ||
1382 | /* | |
1383 | * Use hrtick when: | |
1384 | * - enabled by features | |
1385 | * - hrtimer is actually high res | |
1386 | */ | |
1387 | static inline int hrtick_enabled(struct rq *rq) | |
1388 | { | |
1389 | if (!sched_feat(HRTICK)) | |
1390 | return 0; | |
1391 | if (!cpu_active(cpu_of(rq))) | |
1392 | return 0; | |
1393 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1394 | } | |
1395 | ||
1396 | void hrtick_start(struct rq *rq, u64 delay); | |
1397 | ||
b39e66ea MG |
1398 | #else |
1399 | ||
1400 | static inline int hrtick_enabled(struct rq *rq) | |
1401 | { | |
1402 | return 0; | |
1403 | } | |
1404 | ||
029632fb PZ |
1405 | #endif /* CONFIG_SCHED_HRTICK */ |
1406 | ||
1407 | #ifdef CONFIG_SMP | |
1408 | extern void sched_avg_update(struct rq *rq); | |
dfbca41f PZ |
1409 | |
1410 | #ifndef arch_scale_freq_capacity | |
1411 | static __always_inline | |
1412 | unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) | |
1413 | { | |
1414 | return SCHED_CAPACITY_SCALE; | |
1415 | } | |
1416 | #endif | |
b5b4860d | 1417 | |
029632fb PZ |
1418 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) |
1419 | { | |
b5b4860d | 1420 | rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); |
029632fb PZ |
1421 | sched_avg_update(rq); |
1422 | } | |
1423 | #else | |
1424 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1425 | static inline void sched_avg_update(struct rq *rq) { } | |
1426 | #endif | |
1427 | ||
3960c8c0 PZ |
1428 | /* |
1429 | * __task_rq_lock - lock the rq @p resides on. | |
1430 | */ | |
1431 | static inline struct rq *__task_rq_lock(struct task_struct *p) | |
1432 | __acquires(rq->lock) | |
1433 | { | |
1434 | struct rq *rq; | |
1435 | ||
1436 | lockdep_assert_held(&p->pi_lock); | |
1437 | ||
1438 | for (;;) { | |
1439 | rq = task_rq(p); | |
1440 | raw_spin_lock(&rq->lock); | |
cbce1a68 PZ |
1441 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
1442 | lockdep_pin_lock(&rq->lock); | |
3960c8c0 | 1443 | return rq; |
cbce1a68 | 1444 | } |
3960c8c0 PZ |
1445 | raw_spin_unlock(&rq->lock); |
1446 | ||
1447 | while (unlikely(task_on_rq_migrating(p))) | |
1448 | cpu_relax(); | |
1449 | } | |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
1454 | */ | |
1455 | static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) | |
1456 | __acquires(p->pi_lock) | |
1457 | __acquires(rq->lock) | |
1458 | { | |
1459 | struct rq *rq; | |
1460 | ||
1461 | for (;;) { | |
1462 | raw_spin_lock_irqsave(&p->pi_lock, *flags); | |
1463 | rq = task_rq(p); | |
1464 | raw_spin_lock(&rq->lock); | |
1465 | /* | |
1466 | * move_queued_task() task_rq_lock() | |
1467 | * | |
1468 | * ACQUIRE (rq->lock) | |
1469 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
1470 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
1471 | * [S] ->cpu = new_cpu [L] task_rq() | |
1472 | * [L] ->on_rq | |
1473 | * RELEASE (rq->lock) | |
1474 | * | |
1475 | * If we observe the old cpu in task_rq_lock, the acquire of | |
1476 | * the old rq->lock will fully serialize against the stores. | |
1477 | * | |
1478 | * If we observe the new cpu in task_rq_lock, the acquire will | |
1479 | * pair with the WMB to ensure we must then also see migrating. | |
1480 | */ | |
cbce1a68 PZ |
1481 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
1482 | lockdep_pin_lock(&rq->lock); | |
3960c8c0 | 1483 | return rq; |
cbce1a68 | 1484 | } |
3960c8c0 PZ |
1485 | raw_spin_unlock(&rq->lock); |
1486 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1487 | ||
1488 | while (unlikely(task_on_rq_migrating(p))) | |
1489 | cpu_relax(); | |
1490 | } | |
1491 | } | |
1492 | ||
1493 | static inline void __task_rq_unlock(struct rq *rq) | |
1494 | __releases(rq->lock) | |
1495 | { | |
cbce1a68 | 1496 | lockdep_unpin_lock(&rq->lock); |
3960c8c0 PZ |
1497 | raw_spin_unlock(&rq->lock); |
1498 | } | |
1499 | ||
1500 | static inline void | |
1501 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1502 | __releases(rq->lock) | |
1503 | __releases(p->pi_lock) | |
1504 | { | |
cbce1a68 | 1505 | lockdep_unpin_lock(&rq->lock); |
3960c8c0 PZ |
1506 | raw_spin_unlock(&rq->lock); |
1507 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1508 | } | |
1509 | ||
029632fb PZ |
1510 | #ifdef CONFIG_SMP |
1511 | #ifdef CONFIG_PREEMPT | |
1512 | ||
1513 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1514 | ||
1515 | /* | |
1516 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1517 | * way at the expense of forcing extra atomic operations in all | |
1518 | * invocations. This assures that the double_lock is acquired using the | |
1519 | * same underlying policy as the spinlock_t on this architecture, which | |
1520 | * reduces latency compared to the unfair variant below. However, it | |
1521 | * also adds more overhead and therefore may reduce throughput. | |
1522 | */ | |
1523 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1524 | __releases(this_rq->lock) | |
1525 | __acquires(busiest->lock) | |
1526 | __acquires(this_rq->lock) | |
1527 | { | |
1528 | raw_spin_unlock(&this_rq->lock); | |
1529 | double_rq_lock(this_rq, busiest); | |
1530 | ||
1531 | return 1; | |
1532 | } | |
1533 | ||
1534 | #else | |
1535 | /* | |
1536 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1537 | * latency by eliminating extra atomic operations when the locks are | |
1538 | * already in proper order on entry. This favors lower cpu-ids and will | |
1539 | * grant the double lock to lower cpus over higher ids under contention, | |
1540 | * regardless of entry order into the function. | |
1541 | */ | |
1542 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1543 | __releases(this_rq->lock) | |
1544 | __acquires(busiest->lock) | |
1545 | __acquires(this_rq->lock) | |
1546 | { | |
1547 | int ret = 0; | |
1548 | ||
1549 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1550 | if (busiest < this_rq) { | |
1551 | raw_spin_unlock(&this_rq->lock); | |
1552 | raw_spin_lock(&busiest->lock); | |
1553 | raw_spin_lock_nested(&this_rq->lock, | |
1554 | SINGLE_DEPTH_NESTING); | |
1555 | ret = 1; | |
1556 | } else | |
1557 | raw_spin_lock_nested(&busiest->lock, | |
1558 | SINGLE_DEPTH_NESTING); | |
1559 | } | |
1560 | return ret; | |
1561 | } | |
1562 | ||
1563 | #endif /* CONFIG_PREEMPT */ | |
1564 | ||
1565 | /* | |
1566 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1567 | */ | |
1568 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1569 | { | |
1570 | if (unlikely(!irqs_disabled())) { | |
1571 | /* printk() doesn't work good under rq->lock */ | |
1572 | raw_spin_unlock(&this_rq->lock); | |
1573 | BUG_ON(1); | |
1574 | } | |
1575 | ||
1576 | return _double_lock_balance(this_rq, busiest); | |
1577 | } | |
1578 | ||
1579 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1580 | __releases(busiest->lock) | |
1581 | { | |
1582 | raw_spin_unlock(&busiest->lock); | |
1583 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1584 | } | |
1585 | ||
74602315 PZ |
1586 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1587 | { | |
1588 | if (l1 > l2) | |
1589 | swap(l1, l2); | |
1590 | ||
1591 | spin_lock(l1); | |
1592 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1593 | } | |
1594 | ||
60e69eed MG |
1595 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1596 | { | |
1597 | if (l1 > l2) | |
1598 | swap(l1, l2); | |
1599 | ||
1600 | spin_lock_irq(l1); | |
1601 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1602 | } | |
1603 | ||
74602315 PZ |
1604 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1605 | { | |
1606 | if (l1 > l2) | |
1607 | swap(l1, l2); | |
1608 | ||
1609 | raw_spin_lock(l1); | |
1610 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1611 | } | |
1612 | ||
029632fb PZ |
1613 | /* |
1614 | * double_rq_lock - safely lock two runqueues | |
1615 | * | |
1616 | * Note this does not disable interrupts like task_rq_lock, | |
1617 | * you need to do so manually before calling. | |
1618 | */ | |
1619 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1620 | __acquires(rq1->lock) | |
1621 | __acquires(rq2->lock) | |
1622 | { | |
1623 | BUG_ON(!irqs_disabled()); | |
1624 | if (rq1 == rq2) { | |
1625 | raw_spin_lock(&rq1->lock); | |
1626 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1627 | } else { | |
1628 | if (rq1 < rq2) { | |
1629 | raw_spin_lock(&rq1->lock); | |
1630 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1631 | } else { | |
1632 | raw_spin_lock(&rq2->lock); | |
1633 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1634 | } | |
1635 | } | |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * double_rq_unlock - safely unlock two runqueues | |
1640 | * | |
1641 | * Note this does not restore interrupts like task_rq_unlock, | |
1642 | * you need to do so manually after calling. | |
1643 | */ | |
1644 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1645 | __releases(rq1->lock) | |
1646 | __releases(rq2->lock) | |
1647 | { | |
1648 | raw_spin_unlock(&rq1->lock); | |
1649 | if (rq1 != rq2) | |
1650 | raw_spin_unlock(&rq2->lock); | |
1651 | else | |
1652 | __release(rq2->lock); | |
1653 | } | |
1654 | ||
1655 | #else /* CONFIG_SMP */ | |
1656 | ||
1657 | /* | |
1658 | * double_rq_lock - safely lock two runqueues | |
1659 | * | |
1660 | * Note this does not disable interrupts like task_rq_lock, | |
1661 | * you need to do so manually before calling. | |
1662 | */ | |
1663 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1664 | __acquires(rq1->lock) | |
1665 | __acquires(rq2->lock) | |
1666 | { | |
1667 | BUG_ON(!irqs_disabled()); | |
1668 | BUG_ON(rq1 != rq2); | |
1669 | raw_spin_lock(&rq1->lock); | |
1670 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * double_rq_unlock - safely unlock two runqueues | |
1675 | * | |
1676 | * Note this does not restore interrupts like task_rq_unlock, | |
1677 | * you need to do so manually after calling. | |
1678 | */ | |
1679 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1680 | __releases(rq1->lock) | |
1681 | __releases(rq2->lock) | |
1682 | { | |
1683 | BUG_ON(rq1 != rq2); | |
1684 | raw_spin_unlock(&rq1->lock); | |
1685 | __release(rq2->lock); | |
1686 | } | |
1687 | ||
1688 | #endif | |
1689 | ||
1690 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1691 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
1692 | |
1693 | #ifdef CONFIG_SCHED_DEBUG | |
029632fb PZ |
1694 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
1695 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 1696 | extern void print_dl_stats(struct seq_file *m, int cpu); |
6b55c965 SD |
1697 | extern void |
1698 | print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); | |
397f2378 SD |
1699 | |
1700 | #ifdef CONFIG_NUMA_BALANCING | |
1701 | extern void | |
1702 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
1703 | extern void | |
1704 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
1705 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
1706 | #endif /* CONFIG_NUMA_BALANCING */ | |
1707 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
1708 | |
1709 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
1710 | extern void init_rt_rq(struct rt_rq *rt_rq); |
1711 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 1712 | |
1ee14e6c BS |
1713 | extern void cfs_bandwidth_usage_inc(void); |
1714 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1715 | |
3451d024 | 1716 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1717 | enum rq_nohz_flag_bits { |
1718 | NOHZ_TICK_STOPPED, | |
1719 | NOHZ_BALANCE_KICK, | |
1720 | }; | |
1721 | ||
1722 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
1723 | #endif | |
73fbec60 FW |
1724 | |
1725 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
1726 | ||
1727 | DECLARE_PER_CPU(u64, cpu_hardirq_time); | |
1728 | DECLARE_PER_CPU(u64, cpu_softirq_time); | |
1729 | ||
1730 | #ifndef CONFIG_64BIT | |
1731 | DECLARE_PER_CPU(seqcount_t, irq_time_seq); | |
1732 | ||
1733 | static inline void irq_time_write_begin(void) | |
1734 | { | |
1735 | __this_cpu_inc(irq_time_seq.sequence); | |
1736 | smp_wmb(); | |
1737 | } | |
1738 | ||
1739 | static inline void irq_time_write_end(void) | |
1740 | { | |
1741 | smp_wmb(); | |
1742 | __this_cpu_inc(irq_time_seq.sequence); | |
1743 | } | |
1744 | ||
1745 | static inline u64 irq_time_read(int cpu) | |
1746 | { | |
1747 | u64 irq_time; | |
1748 | unsigned seq; | |
1749 | ||
1750 | do { | |
1751 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1752 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1753 | per_cpu(cpu_hardirq_time, cpu); | |
1754 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1755 | ||
1756 | return irq_time; | |
1757 | } | |
1758 | #else /* CONFIG_64BIT */ | |
1759 | static inline void irq_time_write_begin(void) | |
1760 | { | |
1761 | } | |
1762 | ||
1763 | static inline void irq_time_write_end(void) | |
1764 | { | |
1765 | } | |
1766 | ||
1767 | static inline u64 irq_time_read(int cpu) | |
1768 | { | |
1769 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1770 | } | |
1771 | #endif /* CONFIG_64BIT */ | |
1772 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |