]>
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> |
3866e845 | 6 | #include <linux/binfmts.h> |
029632fb PZ |
7 | #include <linux/mutex.h> |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/stop_machine.h> | |
b6366f04 | 10 | #include <linux/irq_work.h> |
9f3660c2 | 11 | #include <linux/tick.h> |
f809ca9a | 12 | #include <linux/slab.h> |
029632fb | 13 | |
391e43da | 14 | #include "cpupri.h" |
6bfd6d72 | 15 | #include "cpudeadline.h" |
60fed789 | 16 | #include "cpuacct.h" |
029632fb | 17 | |
45ceebf7 | 18 | struct rq; |
442bf3aa | 19 | struct cpuidle_state; |
45ceebf7 | 20 | |
da0c1e65 KT |
21 | /* task_struct::on_rq states: */ |
22 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 23 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 24 | |
029632fb PZ |
25 | extern __read_mostly int scheduler_running; |
26 | ||
45ceebf7 PG |
27 | extern unsigned long calc_load_update; |
28 | extern atomic_long_t calc_load_tasks; | |
29 | ||
3289bdb4 | 30 | extern void calc_global_load_tick(struct rq *this_rq); |
d60585c5 | 31 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
3289bdb4 PZ |
32 | |
33 | #ifdef CONFIG_SMP | |
cee1afce | 34 | extern void cpu_load_update_active(struct rq *this_rq); |
3289bdb4 | 35 | #else |
cee1afce | 36 | static inline void cpu_load_update_active(struct rq *this_rq) { } |
3289bdb4 | 37 | #endif |
45ceebf7 | 38 | |
029632fb PZ |
39 | /* |
40 | * Helpers for converting nanosecond timing to jiffy resolution | |
41 | */ | |
42 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
43 | ||
cc1f4b1f LZ |
44 | /* |
45 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
46 | * The extra resolution improves shares distribution and load balancing of | |
47 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
48 | * hierarchies, especially on larger systems. This is not a user-visible change | |
49 | * and does not change the user-interface for setting shares/weights. | |
50 | * | |
51 | * We increase resolution only if we have enough bits to allow this increased | |
2159197d PZ |
52 | * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are |
53 | * pretty high and the returns do not justify the increased costs. | |
54 | * | |
55 | * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to | |
56 | * increase coverage and consistency always enable it on 64bit platforms. | |
cc1f4b1f | 57 | */ |
2159197d | 58 | #ifdef CONFIG_64BIT |
172895e6 | 59 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
6ecdd749 YD |
60 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
61 | # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) | |
cc1f4b1f | 62 | #else |
172895e6 | 63 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
cc1f4b1f LZ |
64 | # define scale_load(w) (w) |
65 | # define scale_load_down(w) (w) | |
66 | #endif | |
67 | ||
6ecdd749 | 68 | /* |
172895e6 YD |
69 | * Task weight (visible to users) and its load (invisible to users) have |
70 | * independent resolution, but they should be well calibrated. We use | |
71 | * scale_load() and scale_load_down(w) to convert between them. The | |
72 | * following must be true: | |
73 | * | |
74 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD | |
75 | * | |
6ecdd749 | 76 | */ |
172895e6 | 77 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
029632fb | 78 | |
332ac17e DF |
79 | /* |
80 | * Single value that decides SCHED_DEADLINE internal math precision. | |
81 | * 10 -> just above 1us | |
82 | * 9 -> just above 0.5us | |
83 | */ | |
84 | #define DL_SCALE (10) | |
85 | ||
029632fb PZ |
86 | /* |
87 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 88 | */ |
029632fb PZ |
89 | |
90 | /* | |
91 | * single value that denotes runtime == period, ie unlimited time. | |
92 | */ | |
93 | #define RUNTIME_INF ((u64)~0ULL) | |
94 | ||
20f9cd2a HA |
95 | static inline int idle_policy(int policy) |
96 | { | |
97 | return policy == SCHED_IDLE; | |
98 | } | |
d50dde5a DF |
99 | static inline int fair_policy(int policy) |
100 | { | |
101 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
102 | } | |
103 | ||
029632fb PZ |
104 | static inline int rt_policy(int policy) |
105 | { | |
d50dde5a | 106 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
107 | } |
108 | ||
aab03e05 DF |
109 | static inline int dl_policy(int policy) |
110 | { | |
111 | return policy == SCHED_DEADLINE; | |
112 | } | |
20f9cd2a HA |
113 | static inline bool valid_policy(int policy) |
114 | { | |
115 | return idle_policy(policy) || fair_policy(policy) || | |
116 | rt_policy(policy) || dl_policy(policy); | |
117 | } | |
aab03e05 | 118 | |
029632fb PZ |
119 | static inline int task_has_rt_policy(struct task_struct *p) |
120 | { | |
121 | return rt_policy(p->policy); | |
122 | } | |
123 | ||
aab03e05 DF |
124 | static inline int task_has_dl_policy(struct task_struct *p) |
125 | { | |
126 | return dl_policy(p->policy); | |
127 | } | |
128 | ||
2d3d891d DF |
129 | /* |
130 | * Tells if entity @a should preempt entity @b. | |
131 | */ | |
332ac17e DF |
132 | static inline bool |
133 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
134 | { |
135 | return dl_time_before(a->deadline, b->deadline); | |
136 | } | |
137 | ||
029632fb PZ |
138 | /* |
139 | * This is the priority-queue data structure of the RT scheduling class: | |
140 | */ | |
141 | struct rt_prio_array { | |
142 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
143 | struct list_head queue[MAX_RT_PRIO]; | |
144 | }; | |
145 | ||
146 | struct rt_bandwidth { | |
147 | /* nests inside the rq lock: */ | |
148 | raw_spinlock_t rt_runtime_lock; | |
149 | ktime_t rt_period; | |
150 | u64 rt_runtime; | |
151 | struct hrtimer rt_period_timer; | |
4cfafd30 | 152 | unsigned int rt_period_active; |
029632fb | 153 | }; |
a5e7be3b JL |
154 | |
155 | void __dl_clear_params(struct task_struct *p); | |
156 | ||
332ac17e DF |
157 | /* |
158 | * To keep the bandwidth of -deadline tasks and groups under control | |
159 | * we need some place where: | |
160 | * - store the maximum -deadline bandwidth of the system (the group); | |
161 | * - cache the fraction of that bandwidth that is currently allocated. | |
162 | * | |
163 | * This is all done in the data structure below. It is similar to the | |
164 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
165 | * that, since here we are only interested in admission control, we | |
166 | * do not decrease any runtime while the group "executes", neither we | |
167 | * need a timer to replenish it. | |
168 | * | |
169 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
170 | * meaning that: | |
171 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
172 | * - dl_total_bw array contains, in the i-eth element, the currently | |
173 | * allocated bandwidth on the i-eth CPU. | |
174 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
175 | * consume bandwidth on the CPU they're running on. | |
176 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
177 | * that will be shown the next time the proc or cgroup controls will | |
178 | * be red. It on its turn can be changed by writing on its own | |
179 | * control. | |
180 | */ | |
181 | struct dl_bandwidth { | |
182 | raw_spinlock_t dl_runtime_lock; | |
183 | u64 dl_runtime; | |
184 | u64 dl_period; | |
185 | }; | |
186 | ||
187 | static inline int dl_bandwidth_enabled(void) | |
188 | { | |
1724813d | 189 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
190 | } |
191 | ||
192 | extern struct dl_bw *dl_bw_of(int i); | |
193 | ||
194 | struct dl_bw { | |
195 | raw_spinlock_t lock; | |
196 | u64 bw, total_bw; | |
197 | }; | |
198 | ||
7f51412a JL |
199 | static inline |
200 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
201 | { | |
202 | dl_b->total_bw -= tsk_bw; | |
203 | } | |
204 | ||
205 | static inline | |
206 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
207 | { | |
208 | dl_b->total_bw += tsk_bw; | |
209 | } | |
210 | ||
211 | static inline | |
212 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
213 | { | |
214 | return dl_b->bw != -1 && | |
215 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
216 | } | |
217 | ||
029632fb PZ |
218 | extern struct mutex sched_domains_mutex; |
219 | ||
220 | #ifdef CONFIG_CGROUP_SCHED | |
221 | ||
222 | #include <linux/cgroup.h> | |
223 | ||
224 | struct cfs_rq; | |
225 | struct rt_rq; | |
226 | ||
35cf4e50 | 227 | extern struct list_head task_groups; |
029632fb PZ |
228 | |
229 | struct cfs_bandwidth { | |
230 | #ifdef CONFIG_CFS_BANDWIDTH | |
231 | raw_spinlock_t lock; | |
232 | ktime_t period; | |
233 | u64 quota, runtime; | |
9c58c79a | 234 | s64 hierarchical_quota; |
029632fb PZ |
235 | u64 runtime_expires; |
236 | ||
4cfafd30 | 237 | int idle, period_active; |
029632fb PZ |
238 | struct hrtimer period_timer, slack_timer; |
239 | struct list_head throttled_cfs_rq; | |
240 | ||
241 | /* statistics */ | |
242 | int nr_periods, nr_throttled; | |
243 | u64 throttled_time; | |
244 | #endif | |
245 | }; | |
246 | ||
247 | /* task group related information */ | |
248 | struct task_group { | |
249 | struct cgroup_subsys_state css; | |
250 | ||
251 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
252 | /* schedulable entities of this group on each cpu */ | |
253 | struct sched_entity **se; | |
254 | /* runqueue "owned" by this group on each cpu */ | |
255 | struct cfs_rq **cfs_rq; | |
256 | unsigned long shares; | |
257 | ||
fa6bddeb | 258 | #ifdef CONFIG_SMP |
b0367629 WL |
259 | /* |
260 | * load_avg can be heavily contended at clock tick time, so put | |
261 | * it in its own cacheline separated from the fields above which | |
262 | * will also be accessed at each tick. | |
263 | */ | |
264 | atomic_long_t load_avg ____cacheline_aligned; | |
029632fb | 265 | #endif |
fa6bddeb | 266 | #endif |
029632fb PZ |
267 | |
268 | #ifdef CONFIG_RT_GROUP_SCHED | |
269 | struct sched_rt_entity **rt_se; | |
270 | struct rt_rq **rt_rq; | |
271 | ||
272 | struct rt_bandwidth rt_bandwidth; | |
273 | #endif | |
274 | ||
275 | struct rcu_head rcu; | |
276 | struct list_head list; | |
277 | ||
278 | struct task_group *parent; | |
279 | struct list_head siblings; | |
280 | struct list_head children; | |
281 | ||
282 | #ifdef CONFIG_SCHED_AUTOGROUP | |
283 | struct autogroup *autogroup; | |
284 | #endif | |
285 | ||
286 | struct cfs_bandwidth cfs_bandwidth; | |
287 | }; | |
288 | ||
289 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
290 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
291 | ||
292 | /* | |
293 | * A weight of 0 or 1 can cause arithmetics problems. | |
294 | * A weight of a cfs_rq is the sum of weights of which entities | |
295 | * are queued on this cfs_rq, so a weight of a entity should not be | |
296 | * too large, so as the shares value of a task group. | |
297 | * (The default weight is 1024 - so there's no practical | |
298 | * limitation from this.) | |
299 | */ | |
300 | #define MIN_SHARES (1UL << 1) | |
301 | #define MAX_SHARES (1UL << 18) | |
302 | #endif | |
303 | ||
029632fb PZ |
304 | typedef int (*tg_visitor)(struct task_group *, void *); |
305 | ||
306 | extern int walk_tg_tree_from(struct task_group *from, | |
307 | tg_visitor down, tg_visitor up, void *data); | |
308 | ||
309 | /* | |
310 | * Iterate the full tree, calling @down when first entering a node and @up when | |
311 | * leaving it for the final time. | |
312 | * | |
313 | * Caller must hold rcu_lock or sufficient equivalent. | |
314 | */ | |
315 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
316 | { | |
317 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
318 | } | |
319 | ||
320 | extern int tg_nop(struct task_group *tg, void *data); | |
321 | ||
322 | extern void free_fair_sched_group(struct task_group *tg); | |
323 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
8663e24d | 324 | extern void online_fair_sched_group(struct task_group *tg); |
6fe1f348 | 325 | extern void unregister_fair_sched_group(struct task_group *tg); |
029632fb PZ |
326 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
327 | struct sched_entity *se, int cpu, | |
328 | struct sched_entity *parent); | |
329 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
029632fb PZ |
330 | |
331 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 332 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
333 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
334 | ||
335 | extern void free_rt_sched_group(struct task_group *tg); | |
336 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
337 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
338 | struct sched_rt_entity *rt_se, int cpu, | |
339 | struct sched_rt_entity *parent); | |
340 | ||
25cc7da7 LZ |
341 | extern struct task_group *sched_create_group(struct task_group *parent); |
342 | extern void sched_online_group(struct task_group *tg, | |
343 | struct task_group *parent); | |
344 | extern void sched_destroy_group(struct task_group *tg); | |
345 | extern void sched_offline_group(struct task_group *tg); | |
346 | ||
347 | extern void sched_move_task(struct task_struct *tsk); | |
348 | ||
349 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
350 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
ad936d86 BP |
351 | |
352 | #ifdef CONFIG_SMP | |
353 | extern void set_task_rq_fair(struct sched_entity *se, | |
354 | struct cfs_rq *prev, struct cfs_rq *next); | |
355 | #else /* !CONFIG_SMP */ | |
356 | static inline void set_task_rq_fair(struct sched_entity *se, | |
357 | struct cfs_rq *prev, struct cfs_rq *next) { } | |
358 | #endif /* CONFIG_SMP */ | |
359 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
25cc7da7 | 360 | |
029632fb PZ |
361 | #else /* CONFIG_CGROUP_SCHED */ |
362 | ||
363 | struct cfs_bandwidth { }; | |
364 | ||
365 | #endif /* CONFIG_CGROUP_SCHED */ | |
366 | ||
367 | /* CFS-related fields in a runqueue */ | |
368 | struct cfs_rq { | |
369 | struct load_weight load; | |
c82513e5 | 370 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
371 | |
372 | u64 exec_clock; | |
373 | u64 min_vruntime; | |
374 | #ifndef CONFIG_64BIT | |
375 | u64 min_vruntime_copy; | |
376 | #endif | |
377 | ||
378 | struct rb_root tasks_timeline; | |
379 | struct rb_node *rb_leftmost; | |
380 | ||
029632fb PZ |
381 | /* |
382 | * 'curr' points to currently running entity on this cfs_rq. | |
383 | * It is set to NULL otherwise (i.e when none are currently running). | |
384 | */ | |
385 | struct sched_entity *curr, *next, *last, *skip; | |
386 | ||
387 | #ifdef CONFIG_SCHED_DEBUG | |
388 | unsigned int nr_spread_over; | |
389 | #endif | |
390 | ||
2dac754e PT |
391 | #ifdef CONFIG_SMP |
392 | /* | |
9d89c257 | 393 | * CFS load tracking |
2dac754e | 394 | */ |
9d89c257 | 395 | struct sched_avg avg; |
13962234 YD |
396 | u64 runnable_load_sum; |
397 | unsigned long runnable_load_avg; | |
c566e8e9 | 398 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9d89c257 YD |
399 | unsigned long tg_load_avg_contrib; |
400 | #endif | |
401 | atomic_long_t removed_load_avg, removed_util_avg; | |
402 | #ifndef CONFIG_64BIT | |
403 | u64 load_last_update_time_copy; | |
404 | #endif | |
82958366 | 405 | |
9d89c257 | 406 | #ifdef CONFIG_FAIR_GROUP_SCHED |
82958366 PT |
407 | /* |
408 | * h_load = weight * f(tg) | |
409 | * | |
410 | * Where f(tg) is the recursive weight fraction assigned to | |
411 | * this group. | |
412 | */ | |
413 | unsigned long h_load; | |
68520796 VD |
414 | u64 last_h_load_update; |
415 | struct sched_entity *h_load_next; | |
416 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
417 | #endif /* CONFIG_SMP */ |
418 | ||
029632fb PZ |
419 | #ifdef CONFIG_FAIR_GROUP_SCHED |
420 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
421 | ||
422 | /* | |
423 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
424 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
425 | * (like users, containers etc.) | |
426 | * | |
427 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
428 | * list is used during load balance. | |
429 | */ | |
430 | int on_list; | |
431 | struct list_head leaf_cfs_rq_list; | |
432 | struct task_group *tg; /* group that "owns" this runqueue */ | |
433 | ||
029632fb PZ |
434 | #ifdef CONFIG_CFS_BANDWIDTH |
435 | int runtime_enabled; | |
436 | u64 runtime_expires; | |
437 | s64 runtime_remaining; | |
438 | ||
f1b17280 PT |
439 | u64 throttled_clock, throttled_clock_task; |
440 | u64 throttled_clock_task_time; | |
55e16d30 | 441 | int throttled, throttle_count; |
029632fb PZ |
442 | struct list_head throttled_list; |
443 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
444 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
445 | }; | |
446 | ||
447 | static inline int rt_bandwidth_enabled(void) | |
448 | { | |
449 | return sysctl_sched_rt_runtime >= 0; | |
450 | } | |
451 | ||
b6366f04 SR |
452 | /* RT IPI pull logic requires IRQ_WORK */ |
453 | #ifdef CONFIG_IRQ_WORK | |
454 | # define HAVE_RT_PUSH_IPI | |
455 | #endif | |
456 | ||
029632fb PZ |
457 | /* Real-Time classes' related field in a runqueue: */ |
458 | struct rt_rq { | |
459 | struct rt_prio_array active; | |
c82513e5 | 460 | unsigned int rt_nr_running; |
01d36d0a | 461 | unsigned int rr_nr_running; |
029632fb PZ |
462 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
463 | struct { | |
464 | int curr; /* highest queued rt task prio */ | |
465 | #ifdef CONFIG_SMP | |
466 | int next; /* next highest */ | |
467 | #endif | |
468 | } highest_prio; | |
469 | #endif | |
470 | #ifdef CONFIG_SMP | |
471 | unsigned long rt_nr_migratory; | |
472 | unsigned long rt_nr_total; | |
473 | int overloaded; | |
474 | struct plist_head pushable_tasks; | |
b6366f04 SR |
475 | #ifdef HAVE_RT_PUSH_IPI |
476 | int push_flags; | |
477 | int push_cpu; | |
478 | struct irq_work push_work; | |
479 | raw_spinlock_t push_lock; | |
029632fb | 480 | #endif |
b6366f04 | 481 | #endif /* CONFIG_SMP */ |
f4ebcbc0 KT |
482 | int rt_queued; |
483 | ||
029632fb PZ |
484 | int rt_throttled; |
485 | u64 rt_time; | |
486 | u64 rt_runtime; | |
487 | /* Nests inside the rq lock: */ | |
488 | raw_spinlock_t rt_runtime_lock; | |
489 | ||
490 | #ifdef CONFIG_RT_GROUP_SCHED | |
491 | unsigned long rt_nr_boosted; | |
492 | ||
493 | struct rq *rq; | |
029632fb PZ |
494 | struct task_group *tg; |
495 | #endif | |
496 | }; | |
497 | ||
aab03e05 DF |
498 | /* Deadline class' related fields in a runqueue */ |
499 | struct dl_rq { | |
500 | /* runqueue is an rbtree, ordered by deadline */ | |
501 | struct rb_root rb_root; | |
502 | struct rb_node *rb_leftmost; | |
503 | ||
504 | unsigned long dl_nr_running; | |
1baca4ce JL |
505 | |
506 | #ifdef CONFIG_SMP | |
507 | /* | |
508 | * Deadline values of the currently executing and the | |
509 | * earliest ready task on this rq. Caching these facilitates | |
510 | * the decision wether or not a ready but not running task | |
511 | * should migrate somewhere else. | |
512 | */ | |
513 | struct { | |
514 | u64 curr; | |
515 | u64 next; | |
516 | } earliest_dl; | |
517 | ||
518 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
519 | int overloaded; |
520 | ||
521 | /* | |
522 | * Tasks on this rq that can be pushed away. They are kept in | |
523 | * an rb-tree, ordered by tasks' deadlines, with caching | |
524 | * of the leftmost (earliest deadline) element. | |
525 | */ | |
526 | struct rb_root pushable_dl_tasks_root; | |
527 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
528 | #else |
529 | struct dl_bw dl_bw; | |
1baca4ce | 530 | #endif |
aab03e05 DF |
531 | }; |
532 | ||
029632fb PZ |
533 | #ifdef CONFIG_SMP |
534 | ||
535 | /* | |
536 | * We add the notion of a root-domain which will be used to define per-domain | |
537 | * variables. Each exclusive cpuset essentially defines an island domain by | |
538 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
539 | * exclusive cpuset is created, we also create and attach a new root-domain | |
540 | * object. | |
541 | * | |
542 | */ | |
543 | struct root_domain { | |
544 | atomic_t refcount; | |
545 | atomic_t rto_count; | |
546 | struct rcu_head rcu; | |
547 | cpumask_var_t span; | |
548 | cpumask_var_t online; | |
549 | ||
4486edd1 TC |
550 | /* Indicate more than one runnable task for any CPU */ |
551 | bool overload; | |
552 | ||
1baca4ce JL |
553 | /* |
554 | * The bit corresponding to a CPU gets set here if such CPU has more | |
555 | * than one runnable -deadline task (as it is below for RT tasks). | |
556 | */ | |
557 | cpumask_var_t dlo_mask; | |
558 | atomic_t dlo_count; | |
332ac17e | 559 | struct dl_bw dl_bw; |
6bfd6d72 | 560 | struct cpudl cpudl; |
1baca4ce | 561 | |
029632fb PZ |
562 | /* |
563 | * The "RT overload" flag: it gets set if a CPU has more than | |
564 | * one runnable RT task. | |
565 | */ | |
566 | cpumask_var_t rto_mask; | |
567 | struct cpupri cpupri; | |
cd92bfd3 DE |
568 | |
569 | unsigned long max_cpu_capacity; | |
029632fb PZ |
570 | }; |
571 | ||
572 | extern struct root_domain def_root_domain; | |
573 | ||
574 | #endif /* CONFIG_SMP */ | |
575 | ||
576 | /* | |
577 | * This is the main, per-CPU runqueue data structure. | |
578 | * | |
579 | * Locking rule: those places that want to lock multiple runqueues | |
580 | * (such as the load balancing or the thread migration code), lock | |
581 | * acquire operations must be ordered by ascending &runqueue. | |
582 | */ | |
583 | struct rq { | |
584 | /* runqueue lock: */ | |
585 | raw_spinlock_t lock; | |
586 | ||
587 | /* | |
588 | * nr_running and cpu_load should be in the same cacheline because | |
589 | * remote CPUs use both these fields when doing load calculation. | |
590 | */ | |
c82513e5 | 591 | unsigned int nr_running; |
0ec8aa00 PZ |
592 | #ifdef CONFIG_NUMA_BALANCING |
593 | unsigned int nr_numa_running; | |
594 | unsigned int nr_preferred_running; | |
595 | #endif | |
029632fb PZ |
596 | #define CPU_LOAD_IDX_MAX 5 |
597 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
3451d024 | 598 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 FW |
599 | #ifdef CONFIG_SMP |
600 | unsigned long last_load_update_tick; | |
601 | #endif /* CONFIG_SMP */ | |
1c792db7 | 602 | unsigned long nohz_flags; |
9fd81dd5 | 603 | #endif /* CONFIG_NO_HZ_COMMON */ |
265f22a9 FW |
604 | #ifdef CONFIG_NO_HZ_FULL |
605 | unsigned long last_sched_tick; | |
029632fb | 606 | #endif |
029632fb PZ |
607 | /* capture load from *all* tasks on this cpu: */ |
608 | struct load_weight load; | |
609 | unsigned long nr_load_updates; | |
610 | u64 nr_switches; | |
611 | ||
612 | struct cfs_rq cfs; | |
613 | struct rt_rq rt; | |
aab03e05 | 614 | struct dl_rq dl; |
029632fb PZ |
615 | |
616 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
617 | /* list of leaf cfs_rq on this cpu: */ | |
618 | struct list_head leaf_cfs_rq_list; | |
a35b6466 PZ |
619 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
620 | ||
029632fb PZ |
621 | /* |
622 | * This is part of a global counter where only the total sum | |
623 | * over all CPUs matters. A task can increase this counter on | |
624 | * one CPU and if it got migrated afterwards it may decrease | |
625 | * it on another CPU. Always updated under the runqueue lock: | |
626 | */ | |
627 | unsigned long nr_uninterruptible; | |
628 | ||
629 | struct task_struct *curr, *idle, *stop; | |
630 | unsigned long next_balance; | |
631 | struct mm_struct *prev_mm; | |
632 | ||
9edfbfed | 633 | unsigned int clock_skip_update; |
029632fb PZ |
634 | u64 clock; |
635 | u64 clock_task; | |
636 | ||
637 | atomic_t nr_iowait; | |
638 | ||
639 | #ifdef CONFIG_SMP | |
640 | struct root_domain *rd; | |
641 | struct sched_domain *sd; | |
642 | ||
ced549fa | 643 | unsigned long cpu_capacity; |
ca6d75e6 | 644 | unsigned long cpu_capacity_orig; |
029632fb | 645 | |
e3fca9e7 PZ |
646 | struct callback_head *balance_callback; |
647 | ||
029632fb PZ |
648 | unsigned char idle_balance; |
649 | /* For active balancing */ | |
029632fb PZ |
650 | int active_balance; |
651 | int push_cpu; | |
652 | struct cpu_stop_work active_balance_work; | |
653 | /* cpu of this runqueue: */ | |
654 | int cpu; | |
655 | int online; | |
656 | ||
367456c7 PZ |
657 | struct list_head cfs_tasks; |
658 | ||
029632fb PZ |
659 | u64 rt_avg; |
660 | u64 age_stamp; | |
661 | u64 idle_stamp; | |
662 | u64 avg_idle; | |
9bd721c5 JL |
663 | |
664 | /* This is used to determine avg_idle's max value */ | |
665 | u64 max_idle_balance_cost; | |
029632fb PZ |
666 | #endif |
667 | ||
668 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
669 | u64 prev_irq_time; | |
670 | #endif | |
671 | #ifdef CONFIG_PARAVIRT | |
672 | u64 prev_steal_time; | |
673 | #endif | |
674 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
675 | u64 prev_steal_time_rq; | |
676 | #endif | |
677 | ||
678 | /* calc_load related fields */ | |
679 | unsigned long calc_load_update; | |
680 | long calc_load_active; | |
681 | ||
682 | #ifdef CONFIG_SCHED_HRTICK | |
683 | #ifdef CONFIG_SMP | |
684 | int hrtick_csd_pending; | |
685 | struct call_single_data hrtick_csd; | |
686 | #endif | |
687 | struct hrtimer hrtick_timer; | |
688 | #endif | |
689 | ||
690 | #ifdef CONFIG_SCHEDSTATS | |
691 | /* latency stats */ | |
692 | struct sched_info rq_sched_info; | |
693 | unsigned long long rq_cpu_time; | |
694 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
695 | ||
696 | /* sys_sched_yield() stats */ | |
697 | unsigned int yld_count; | |
698 | ||
699 | /* schedule() stats */ | |
029632fb PZ |
700 | unsigned int sched_count; |
701 | unsigned int sched_goidle; | |
702 | ||
703 | /* try_to_wake_up() stats */ | |
704 | unsigned int ttwu_count; | |
705 | unsigned int ttwu_local; | |
706 | #endif | |
707 | ||
708 | #ifdef CONFIG_SMP | |
709 | struct llist_head wake_list; | |
710 | #endif | |
442bf3aa DL |
711 | |
712 | #ifdef CONFIG_CPU_IDLE | |
713 | /* Must be inspected within a rcu lock section */ | |
714 | struct cpuidle_state *idle_state; | |
715 | #endif | |
029632fb PZ |
716 | }; |
717 | ||
718 | static inline int cpu_of(struct rq *rq) | |
719 | { | |
720 | #ifdef CONFIG_SMP | |
721 | return rq->cpu; | |
722 | #else | |
723 | return 0; | |
724 | #endif | |
725 | } | |
726 | ||
1b568f0a PZ |
727 | |
728 | #ifdef CONFIG_SCHED_SMT | |
729 | ||
730 | extern struct static_key_false sched_smt_present; | |
731 | ||
732 | extern void __update_idle_core(struct rq *rq); | |
733 | ||
734 | static inline void update_idle_core(struct rq *rq) | |
735 | { | |
736 | if (static_branch_unlikely(&sched_smt_present)) | |
737 | __update_idle_core(rq); | |
738 | } | |
739 | ||
740 | #else | |
741 | static inline void update_idle_core(struct rq *rq) { } | |
742 | #endif | |
743 | ||
8b06c55b | 744 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 745 | |
518cd623 | 746 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 747 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
748 | #define task_rq(p) cpu_rq(task_cpu(p)) |
749 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 750 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 751 | |
cebde6d6 PZ |
752 | static inline u64 __rq_clock_broken(struct rq *rq) |
753 | { | |
316c1608 | 754 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
755 | } |
756 | ||
78becc27 FW |
757 | static inline u64 rq_clock(struct rq *rq) |
758 | { | |
cebde6d6 | 759 | lockdep_assert_held(&rq->lock); |
78becc27 FW |
760 | return rq->clock; |
761 | } | |
762 | ||
763 | static inline u64 rq_clock_task(struct rq *rq) | |
764 | { | |
cebde6d6 | 765 | lockdep_assert_held(&rq->lock); |
78becc27 FW |
766 | return rq->clock_task; |
767 | } | |
768 | ||
9edfbfed PZ |
769 | #define RQCF_REQ_SKIP 0x01 |
770 | #define RQCF_ACT_SKIP 0x02 | |
771 | ||
772 | static inline void rq_clock_skip_update(struct rq *rq, bool skip) | |
773 | { | |
774 | lockdep_assert_held(&rq->lock); | |
775 | if (skip) | |
776 | rq->clock_skip_update |= RQCF_REQ_SKIP; | |
777 | else | |
778 | rq->clock_skip_update &= ~RQCF_REQ_SKIP; | |
779 | } | |
780 | ||
9942f79b | 781 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
782 | enum numa_topology_type { |
783 | NUMA_DIRECT, | |
784 | NUMA_GLUELESS_MESH, | |
785 | NUMA_BACKPLANE, | |
786 | }; | |
787 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
788 | extern int sched_max_numa_distance; |
789 | extern bool find_numa_distance(int distance); | |
790 | #endif | |
791 | ||
f809ca9a | 792 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
793 | /* The regions in numa_faults array from task_struct */ |
794 | enum numa_faults_stats { | |
795 | NUMA_MEM = 0, | |
796 | NUMA_CPU, | |
797 | NUMA_MEMBUF, | |
798 | NUMA_CPUBUF | |
799 | }; | |
0ec8aa00 | 800 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 801 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 802 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
803 | #endif /* CONFIG_NUMA_BALANCING */ |
804 | ||
518cd623 PZ |
805 | #ifdef CONFIG_SMP |
806 | ||
e3fca9e7 PZ |
807 | static inline void |
808 | queue_balance_callback(struct rq *rq, | |
809 | struct callback_head *head, | |
810 | void (*func)(struct rq *rq)) | |
811 | { | |
812 | lockdep_assert_held(&rq->lock); | |
813 | ||
814 | if (unlikely(head->next)) | |
815 | return; | |
816 | ||
817 | head->func = (void (*)(struct callback_head *))func; | |
818 | head->next = rq->balance_callback; | |
819 | rq->balance_callback = head; | |
820 | } | |
821 | ||
e3baac47 PZ |
822 | extern void sched_ttwu_pending(void); |
823 | ||
029632fb PZ |
824 | #define rcu_dereference_check_sched_domain(p) \ |
825 | rcu_dereference_check((p), \ | |
826 | lockdep_is_held(&sched_domains_mutex)) | |
827 | ||
828 | /* | |
829 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
830 | * See detach_destroy_domains: synchronize_sched for details. | |
831 | * | |
832 | * The domain tree of any CPU may only be accessed from within | |
833 | * preempt-disabled sections. | |
834 | */ | |
835 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
836 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
837 | __sd; __sd = __sd->parent) | |
029632fb | 838 | |
77e81365 SS |
839 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
840 | ||
518cd623 PZ |
841 | /** |
842 | * highest_flag_domain - Return highest sched_domain containing flag. | |
843 | * @cpu: The cpu whose highest level of sched domain is to | |
844 | * be returned. | |
845 | * @flag: The flag to check for the highest sched_domain | |
846 | * for the given cpu. | |
847 | * | |
848 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
849 | */ | |
850 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
851 | { | |
852 | struct sched_domain *sd, *hsd = NULL; | |
853 | ||
854 | for_each_domain(cpu, sd) { | |
855 | if (!(sd->flags & flag)) | |
856 | break; | |
857 | hsd = sd; | |
858 | } | |
859 | ||
860 | return hsd; | |
861 | } | |
862 | ||
fb13c7ee MG |
863 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
864 | { | |
865 | struct sched_domain *sd; | |
866 | ||
867 | for_each_domain(cpu, sd) { | |
868 | if (sd->flags & flag) | |
869 | break; | |
870 | } | |
871 | ||
872 | return sd; | |
873 | } | |
874 | ||
518cd623 | 875 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 876 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 877 | DECLARE_PER_CPU(int, sd_llc_id); |
0e369d75 | 878 | DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
fb13c7ee | 879 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 | 880 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); |
518cd623 | 881 | |
63b2ca30 | 882 | struct sched_group_capacity { |
5e6521ea LZ |
883 | atomic_t ref; |
884 | /* | |
172895e6 | 885 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
63b2ca30 | 886 | * for a single CPU. |
5e6521ea | 887 | */ |
dc7ff76e | 888 | unsigned int capacity; |
5e6521ea | 889 | unsigned long next_update; |
63b2ca30 | 890 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea LZ |
891 | |
892 | unsigned long cpumask[0]; /* iteration mask */ | |
893 | }; | |
894 | ||
895 | struct sched_group { | |
896 | struct sched_group *next; /* Must be a circular list */ | |
897 | atomic_t ref; | |
898 | ||
899 | unsigned int group_weight; | |
63b2ca30 | 900 | struct sched_group_capacity *sgc; |
5e6521ea LZ |
901 | |
902 | /* | |
903 | * The CPUs this group covers. | |
904 | * | |
905 | * NOTE: this field is variable length. (Allocated dynamically | |
906 | * by attaching extra space to the end of the structure, | |
907 | * depending on how many CPUs the kernel has booted up with) | |
908 | */ | |
909 | unsigned long cpumask[0]; | |
910 | }; | |
911 | ||
912 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) | |
913 | { | |
914 | return to_cpumask(sg->cpumask); | |
915 | } | |
916 | ||
917 | /* | |
918 | * cpumask masking which cpus in the group are allowed to iterate up the domain | |
919 | * tree. | |
920 | */ | |
921 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) | |
922 | { | |
63b2ca30 | 923 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
924 | } |
925 | ||
926 | /** | |
927 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
928 | * @group: The group whose first cpu is to be returned. | |
929 | */ | |
930 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
931 | { | |
932 | return cpumask_first(sched_group_cpus(group)); | |
933 | } | |
934 | ||
c1174876 PZ |
935 | extern int group_balance_cpu(struct sched_group *sg); |
936 | ||
3866e845 SRRH |
937 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
938 | void register_sched_domain_sysctl(void); | |
939 | void unregister_sched_domain_sysctl(void); | |
940 | #else | |
941 | static inline void register_sched_domain_sysctl(void) | |
942 | { | |
943 | } | |
944 | static inline void unregister_sched_domain_sysctl(void) | |
945 | { | |
946 | } | |
947 | #endif | |
948 | ||
e3baac47 PZ |
949 | #else |
950 | ||
951 | static inline void sched_ttwu_pending(void) { } | |
952 | ||
518cd623 | 953 | #endif /* CONFIG_SMP */ |
029632fb | 954 | |
391e43da PZ |
955 | #include "stats.h" |
956 | #include "auto_group.h" | |
029632fb PZ |
957 | |
958 | #ifdef CONFIG_CGROUP_SCHED | |
959 | ||
960 | /* | |
961 | * Return the group to which this tasks belongs. | |
962 | * | |
8af01f56 TH |
963 | * We cannot use task_css() and friends because the cgroup subsystem |
964 | * changes that value before the cgroup_subsys::attach() method is called, | |
965 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
966 | * |
967 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
968 | * core changes this before calling sched_move_task(). | |
969 | * | |
970 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
971 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
972 | */ |
973 | static inline struct task_group *task_group(struct task_struct *p) | |
974 | { | |
8323f26c | 975 | return p->sched_task_group; |
029632fb PZ |
976 | } |
977 | ||
978 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
979 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
980 | { | |
981 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
982 | struct task_group *tg = task_group(p); | |
983 | #endif | |
984 | ||
985 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
ad936d86 | 986 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
029632fb PZ |
987 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
988 | p->se.parent = tg->se[cpu]; | |
989 | #endif | |
990 | ||
991 | #ifdef CONFIG_RT_GROUP_SCHED | |
992 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
993 | p->rt.parent = tg->rt_se[cpu]; | |
994 | #endif | |
995 | } | |
996 | ||
997 | #else /* CONFIG_CGROUP_SCHED */ | |
998 | ||
999 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
1000 | static inline struct task_group *task_group(struct task_struct *p) | |
1001 | { | |
1002 | return NULL; | |
1003 | } | |
1004 | ||
1005 | #endif /* CONFIG_CGROUP_SCHED */ | |
1006 | ||
1007 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1008 | { | |
1009 | set_task_rq(p, cpu); | |
1010 | #ifdef CONFIG_SMP | |
1011 | /* | |
1012 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1013 | * successfuly executed on another CPU. We must ensure that updates of | |
1014 | * per-task data have been completed by this moment. | |
1015 | */ | |
1016 | smp_wmb(); | |
1017 | task_thread_info(p)->cpu = cpu; | |
ac66f547 | 1018 | p->wake_cpu = cpu; |
029632fb PZ |
1019 | #endif |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
1024 | */ | |
1025 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 1026 | # include <linux/static_key.h> |
029632fb PZ |
1027 | # define const_debug __read_mostly |
1028 | #else | |
1029 | # define const_debug const | |
1030 | #endif | |
1031 | ||
1032 | extern const_debug unsigned int sysctl_sched_features; | |
1033 | ||
1034 | #define SCHED_FEAT(name, enabled) \ | |
1035 | __SCHED_FEAT_##name , | |
1036 | ||
1037 | enum { | |
391e43da | 1038 | #include "features.h" |
f8b6d1cc | 1039 | __SCHED_FEAT_NR, |
029632fb PZ |
1040 | }; |
1041 | ||
1042 | #undef SCHED_FEAT | |
1043 | ||
f8b6d1cc | 1044 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
f8b6d1cc | 1045 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1046 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1047 | { \ |
6e76ea8a | 1048 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1049 | } |
1050 | ||
1051 | #include "features.h" | |
1052 | ||
1053 | #undef SCHED_FEAT | |
1054 | ||
c5905afb | 1055 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
1056 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
1057 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 1058 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 1059 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 1060 | |
2a595721 | 1061 | extern struct static_key_false sched_numa_balancing; |
cb251765 | 1062 | extern struct static_key_false sched_schedstats; |
cbee9f88 | 1063 | |
029632fb PZ |
1064 | static inline u64 global_rt_period(void) |
1065 | { | |
1066 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1067 | } | |
1068 | ||
1069 | static inline u64 global_rt_runtime(void) | |
1070 | { | |
1071 | if (sysctl_sched_rt_runtime < 0) | |
1072 | return RUNTIME_INF; | |
1073 | ||
1074 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1075 | } | |
1076 | ||
029632fb PZ |
1077 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1078 | { | |
1079 | return rq->curr == p; | |
1080 | } | |
1081 | ||
1082 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1083 | { | |
1084 | #ifdef CONFIG_SMP | |
1085 | return p->on_cpu; | |
1086 | #else | |
1087 | return task_current(rq, p); | |
1088 | #endif | |
1089 | } | |
1090 | ||
da0c1e65 KT |
1091 | static inline int task_on_rq_queued(struct task_struct *p) |
1092 | { | |
1093 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1094 | } | |
029632fb | 1095 | |
cca26e80 KT |
1096 | static inline int task_on_rq_migrating(struct task_struct *p) |
1097 | { | |
1098 | return p->on_rq == TASK_ON_RQ_MIGRATING; | |
1099 | } | |
1100 | ||
029632fb PZ |
1101 | #ifndef prepare_arch_switch |
1102 | # define prepare_arch_switch(next) do { } while (0) | |
1103 | #endif | |
01f23e16 CM |
1104 | #ifndef finish_arch_post_lock_switch |
1105 | # define finish_arch_post_lock_switch() do { } while (0) | |
1106 | #endif | |
029632fb | 1107 | |
029632fb PZ |
1108 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
1109 | { | |
1110 | #ifdef CONFIG_SMP | |
1111 | /* | |
1112 | * We can optimise this out completely for !SMP, because the | |
1113 | * SMP rebalancing from interrupt is the only thing that cares | |
1114 | * here. | |
1115 | */ | |
1116 | next->on_cpu = 1; | |
1117 | #endif | |
1118 | } | |
1119 | ||
1120 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1121 | { | |
1122 | #ifdef CONFIG_SMP | |
1123 | /* | |
1124 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1125 | * We must ensure this doesn't happen until the switch is completely | |
1126 | * finished. | |
95913d97 | 1127 | * |
b75a2253 PZ |
1128 | * In particular, the load of prev->state in finish_task_switch() must |
1129 | * happen before this. | |
1130 | * | |
1f03e8d2 | 1131 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). |
029632fb | 1132 | */ |
95913d97 | 1133 | smp_store_release(&prev->on_cpu, 0); |
029632fb PZ |
1134 | #endif |
1135 | #ifdef CONFIG_DEBUG_SPINLOCK | |
1136 | /* this is a valid case when another task releases the spinlock */ | |
1137 | rq->lock.owner = current; | |
1138 | #endif | |
1139 | /* | |
1140 | * If we are tracking spinlock dependencies then we have to | |
1141 | * fix up the runqueue lock - which gets 'carried over' from | |
1142 | * prev into current: | |
1143 | */ | |
1144 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1145 | ||
1146 | raw_spin_unlock_irq(&rq->lock); | |
1147 | } | |
1148 | ||
b13095f0 LZ |
1149 | /* |
1150 | * wake flags | |
1151 | */ | |
1152 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1153 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1154 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1155 | ||
029632fb PZ |
1156 | /* |
1157 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1158 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1159 | * each task makes to its run queue's load is weighted according to its | |
1160 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1161 | * scaled version of the new time slice allocation that they receive on time | |
1162 | * slice expiry etc. | |
1163 | */ | |
1164 | ||
1165 | #define WEIGHT_IDLEPRIO 3 | |
1166 | #define WMULT_IDLEPRIO 1431655765 | |
1167 | ||
ed82b8a1 AK |
1168 | extern const int sched_prio_to_weight[40]; |
1169 | extern const u32 sched_prio_to_wmult[40]; | |
029632fb | 1170 | |
ff77e468 PZ |
1171 | /* |
1172 | * {de,en}queue flags: | |
1173 | * | |
1174 | * DEQUEUE_SLEEP - task is no longer runnable | |
1175 | * ENQUEUE_WAKEUP - task just became runnable | |
1176 | * | |
1177 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks | |
1178 | * are in a known state which allows modification. Such pairs | |
1179 | * should preserve as much state as possible. | |
1180 | * | |
1181 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location | |
1182 | * in the runqueue. | |
1183 | * | |
1184 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) | |
1185 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) | |
59efa0ba | 1186 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
ff77e468 PZ |
1187 | * |
1188 | */ | |
1189 | ||
1190 | #define DEQUEUE_SLEEP 0x01 | |
1191 | #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */ | |
1192 | #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */ | |
1193 | ||
1de64443 | 1194 | #define ENQUEUE_WAKEUP 0x01 |
ff77e468 PZ |
1195 | #define ENQUEUE_RESTORE 0x02 |
1196 | #define ENQUEUE_MOVE 0x04 | |
1197 | ||
1198 | #define ENQUEUE_HEAD 0x08 | |
1199 | #define ENQUEUE_REPLENISH 0x10 | |
c82ba9fa | 1200 | #ifdef CONFIG_SMP |
59efa0ba | 1201 | #define ENQUEUE_MIGRATED 0x20 |
c82ba9fa | 1202 | #else |
59efa0ba | 1203 | #define ENQUEUE_MIGRATED 0x00 |
c82ba9fa | 1204 | #endif |
c82ba9fa | 1205 | |
37e117c0 PZ |
1206 | #define RETRY_TASK ((void *)-1UL) |
1207 | ||
c82ba9fa LZ |
1208 | struct sched_class { |
1209 | const struct sched_class *next; | |
1210 | ||
1211 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1212 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1213 | void (*yield_task) (struct rq *rq); | |
1214 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1215 | ||
1216 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1217 | ||
606dba2e PZ |
1218 | /* |
1219 | * It is the responsibility of the pick_next_task() method that will | |
1220 | * return the next task to call put_prev_task() on the @prev task or | |
1221 | * something equivalent. | |
37e117c0 PZ |
1222 | * |
1223 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1224 | * tasks. | |
606dba2e PZ |
1225 | */ |
1226 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
e7904a28 PZ |
1227 | struct task_struct *prev, |
1228 | struct pin_cookie cookie); | |
c82ba9fa LZ |
1229 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1230 | ||
1231 | #ifdef CONFIG_SMP | |
ac66f547 | 1232 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
5a4fd036 | 1233 | void (*migrate_task_rq)(struct task_struct *p); |
c82ba9fa | 1234 | |
c82ba9fa LZ |
1235 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); |
1236 | ||
1237 | void (*set_cpus_allowed)(struct task_struct *p, | |
1238 | const struct cpumask *newmask); | |
1239 | ||
1240 | void (*rq_online)(struct rq *rq); | |
1241 | void (*rq_offline)(struct rq *rq); | |
1242 | #endif | |
1243 | ||
1244 | void (*set_curr_task) (struct rq *rq); | |
1245 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1246 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1247 | void (*task_dead) (struct task_struct *p); |
c82ba9fa | 1248 | |
67dfa1b7 KT |
1249 | /* |
1250 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1251 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1252 | * rq->lock. They are however serialized by p->pi_lock. | |
1253 | */ | |
c82ba9fa LZ |
1254 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); |
1255 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1256 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1257 | int oldprio); | |
1258 | ||
1259 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1260 | struct task_struct *task); | |
1261 | ||
6e998916 SG |
1262 | void (*update_curr) (struct rq *rq); |
1263 | ||
ea86cb4b VG |
1264 | #define TASK_SET_GROUP 0 |
1265 | #define TASK_MOVE_GROUP 1 | |
1266 | ||
c82ba9fa | 1267 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b | 1268 | void (*task_change_group) (struct task_struct *p, int type); |
c82ba9fa LZ |
1269 | #endif |
1270 | }; | |
029632fb | 1271 | |
3f1d2a31 PZ |
1272 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1273 | { | |
1274 | prev->sched_class->put_prev_task(rq, prev); | |
1275 | } | |
1276 | ||
b2bf6c31 PZ |
1277 | static inline void set_curr_task(struct rq *rq, struct task_struct *curr) |
1278 | { | |
1279 | curr->sched_class->set_curr_task(rq); | |
1280 | } | |
1281 | ||
029632fb PZ |
1282 | #define sched_class_highest (&stop_sched_class) |
1283 | #define for_each_class(class) \ | |
1284 | for (class = sched_class_highest; class; class = class->next) | |
1285 | ||
1286 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1287 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1288 | extern const struct sched_class rt_sched_class; |
1289 | extern const struct sched_class fair_sched_class; | |
1290 | extern const struct sched_class idle_sched_class; | |
1291 | ||
1292 | ||
1293 | #ifdef CONFIG_SMP | |
1294 | ||
63b2ca30 | 1295 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1296 | |
7caff66f | 1297 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1298 | |
c5b28038 PZ |
1299 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
1300 | ||
029632fb PZ |
1301 | #endif |
1302 | ||
442bf3aa DL |
1303 | #ifdef CONFIG_CPU_IDLE |
1304 | static inline void idle_set_state(struct rq *rq, | |
1305 | struct cpuidle_state *idle_state) | |
1306 | { | |
1307 | rq->idle_state = idle_state; | |
1308 | } | |
1309 | ||
1310 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1311 | { | |
1312 | WARN_ON(!rcu_read_lock_held()); | |
1313 | return rq->idle_state; | |
1314 | } | |
1315 | #else | |
1316 | static inline void idle_set_state(struct rq *rq, | |
1317 | struct cpuidle_state *idle_state) | |
1318 | { | |
1319 | } | |
1320 | ||
1321 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1322 | { | |
1323 | return NULL; | |
1324 | } | |
1325 | #endif | |
1326 | ||
029632fb PZ |
1327 | extern void sysrq_sched_debug_show(void); |
1328 | extern void sched_init_granularity(void); | |
1329 | extern void update_max_interval(void); | |
1baca4ce JL |
1330 | |
1331 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1332 | extern void init_sched_rt_class(void); |
1333 | extern void init_sched_fair_class(void); | |
1334 | ||
8875125e | 1335 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1336 | extern void resched_cpu(int cpu); |
1337 | ||
1338 | extern struct rt_bandwidth def_rt_bandwidth; | |
1339 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1340 | ||
332ac17e DF |
1341 | extern struct dl_bandwidth def_dl_bandwidth; |
1342 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 DF |
1343 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
1344 | ||
332ac17e DF |
1345 | unsigned long to_ratio(u64 period, u64 runtime); |
1346 | ||
540247fb | 1347 | extern void init_entity_runnable_average(struct sched_entity *se); |
2b8c41da | 1348 | extern void post_init_entity_util_avg(struct sched_entity *se); |
a75cdaa9 | 1349 | |
76d92ac3 FW |
1350 | #ifdef CONFIG_NO_HZ_FULL |
1351 | extern bool sched_can_stop_tick(struct rq *rq); | |
1352 | ||
1353 | /* | |
1354 | * Tick may be needed by tasks in the runqueue depending on their policy and | |
1355 | * requirements. If tick is needed, lets send the target an IPI to kick it out of | |
1356 | * nohz mode if necessary. | |
1357 | */ | |
1358 | static inline void sched_update_tick_dependency(struct rq *rq) | |
1359 | { | |
1360 | int cpu; | |
1361 | ||
1362 | if (!tick_nohz_full_enabled()) | |
1363 | return; | |
1364 | ||
1365 | cpu = cpu_of(rq); | |
1366 | ||
1367 | if (!tick_nohz_full_cpu(cpu)) | |
1368 | return; | |
1369 | ||
1370 | if (sched_can_stop_tick(rq)) | |
1371 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1372 | else | |
1373 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1374 | } | |
1375 | #else | |
1376 | static inline void sched_update_tick_dependency(struct rq *rq) { } | |
1377 | #endif | |
1378 | ||
72465447 | 1379 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1380 | { |
72465447 KT |
1381 | unsigned prev_nr = rq->nr_running; |
1382 | ||
1383 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1384 | |
72465447 | 1385 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1386 | #ifdef CONFIG_SMP |
1387 | if (!rq->rd->overload) | |
1388 | rq->rd->overload = true; | |
1389 | #endif | |
4486edd1 | 1390 | } |
76d92ac3 FW |
1391 | |
1392 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1393 | } |
1394 | ||
72465447 | 1395 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1396 | { |
72465447 | 1397 | rq->nr_running -= count; |
76d92ac3 FW |
1398 | /* Check if we still need preemption */ |
1399 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1400 | } |
1401 | ||
265f22a9 FW |
1402 | static inline void rq_last_tick_reset(struct rq *rq) |
1403 | { | |
1404 | #ifdef CONFIG_NO_HZ_FULL | |
1405 | rq->last_sched_tick = jiffies; | |
1406 | #endif | |
1407 | } | |
1408 | ||
029632fb PZ |
1409 | extern void update_rq_clock(struct rq *rq); |
1410 | ||
1411 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1412 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1413 | ||
1414 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1415 | ||
1416 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1417 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1418 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1419 | ||
1420 | static inline u64 sched_avg_period(void) | |
1421 | { | |
1422 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1423 | } | |
1424 | ||
029632fb PZ |
1425 | #ifdef CONFIG_SCHED_HRTICK |
1426 | ||
1427 | /* | |
1428 | * Use hrtick when: | |
1429 | * - enabled by features | |
1430 | * - hrtimer is actually high res | |
1431 | */ | |
1432 | static inline int hrtick_enabled(struct rq *rq) | |
1433 | { | |
1434 | if (!sched_feat(HRTICK)) | |
1435 | return 0; | |
1436 | if (!cpu_active(cpu_of(rq))) | |
1437 | return 0; | |
1438 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1439 | } | |
1440 | ||
1441 | void hrtick_start(struct rq *rq, u64 delay); | |
1442 | ||
b39e66ea MG |
1443 | #else |
1444 | ||
1445 | static inline int hrtick_enabled(struct rq *rq) | |
1446 | { | |
1447 | return 0; | |
1448 | } | |
1449 | ||
029632fb PZ |
1450 | #endif /* CONFIG_SCHED_HRTICK */ |
1451 | ||
1452 | #ifdef CONFIG_SMP | |
1453 | extern void sched_avg_update(struct rq *rq); | |
dfbca41f PZ |
1454 | |
1455 | #ifndef arch_scale_freq_capacity | |
1456 | static __always_inline | |
1457 | unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) | |
1458 | { | |
1459 | return SCHED_CAPACITY_SCALE; | |
1460 | } | |
1461 | #endif | |
b5b4860d | 1462 | |
8cd5601c MR |
1463 | #ifndef arch_scale_cpu_capacity |
1464 | static __always_inline | |
1465 | unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) | |
1466 | { | |
e3279a2e | 1467 | if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) |
8cd5601c MR |
1468 | return sd->smt_gain / sd->span_weight; |
1469 | ||
1470 | return SCHED_CAPACITY_SCALE; | |
1471 | } | |
1472 | #endif | |
1473 | ||
029632fb PZ |
1474 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) |
1475 | { | |
b5b4860d | 1476 | rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); |
029632fb PZ |
1477 | sched_avg_update(rq); |
1478 | } | |
1479 | #else | |
1480 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1481 | static inline void sched_avg_update(struct rq *rq) { } | |
1482 | #endif | |
1483 | ||
eb580751 PZ |
1484 | struct rq_flags { |
1485 | unsigned long flags; | |
e7904a28 | 1486 | struct pin_cookie cookie; |
eb580751 PZ |
1487 | }; |
1488 | ||
1489 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) | |
3e71a462 | 1490 | __acquires(rq->lock); |
eb580751 | 1491 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3960c8c0 | 1492 | __acquires(p->pi_lock) |
3e71a462 | 1493 | __acquires(rq->lock); |
3960c8c0 | 1494 | |
eb580751 | 1495 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) |
3960c8c0 PZ |
1496 | __releases(rq->lock) |
1497 | { | |
e7904a28 | 1498 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
3960c8c0 PZ |
1499 | raw_spin_unlock(&rq->lock); |
1500 | } | |
1501 | ||
1502 | static inline void | |
eb580751 | 1503 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) |
3960c8c0 PZ |
1504 | __releases(rq->lock) |
1505 | __releases(p->pi_lock) | |
1506 | { | |
e7904a28 | 1507 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
3960c8c0 | 1508 | raw_spin_unlock(&rq->lock); |
eb580751 | 1509 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3960c8c0 PZ |
1510 | } |
1511 | ||
029632fb PZ |
1512 | #ifdef CONFIG_SMP |
1513 | #ifdef CONFIG_PREEMPT | |
1514 | ||
1515 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1516 | ||
1517 | /* | |
1518 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1519 | * way at the expense of forcing extra atomic operations in all | |
1520 | * invocations. This assures that the double_lock is acquired using the | |
1521 | * same underlying policy as the spinlock_t on this architecture, which | |
1522 | * reduces latency compared to the unfair variant below. However, it | |
1523 | * also adds more overhead and therefore may reduce throughput. | |
1524 | */ | |
1525 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1526 | __releases(this_rq->lock) | |
1527 | __acquires(busiest->lock) | |
1528 | __acquires(this_rq->lock) | |
1529 | { | |
1530 | raw_spin_unlock(&this_rq->lock); | |
1531 | double_rq_lock(this_rq, busiest); | |
1532 | ||
1533 | return 1; | |
1534 | } | |
1535 | ||
1536 | #else | |
1537 | /* | |
1538 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1539 | * latency by eliminating extra atomic operations when the locks are | |
1540 | * already in proper order on entry. This favors lower cpu-ids and will | |
1541 | * grant the double lock to lower cpus over higher ids under contention, | |
1542 | * regardless of entry order into the function. | |
1543 | */ | |
1544 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1545 | __releases(this_rq->lock) | |
1546 | __acquires(busiest->lock) | |
1547 | __acquires(this_rq->lock) | |
1548 | { | |
1549 | int ret = 0; | |
1550 | ||
1551 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1552 | if (busiest < this_rq) { | |
1553 | raw_spin_unlock(&this_rq->lock); | |
1554 | raw_spin_lock(&busiest->lock); | |
1555 | raw_spin_lock_nested(&this_rq->lock, | |
1556 | SINGLE_DEPTH_NESTING); | |
1557 | ret = 1; | |
1558 | } else | |
1559 | raw_spin_lock_nested(&busiest->lock, | |
1560 | SINGLE_DEPTH_NESTING); | |
1561 | } | |
1562 | return ret; | |
1563 | } | |
1564 | ||
1565 | #endif /* CONFIG_PREEMPT */ | |
1566 | ||
1567 | /* | |
1568 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1569 | */ | |
1570 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1571 | { | |
1572 | if (unlikely(!irqs_disabled())) { | |
1573 | /* printk() doesn't work good under rq->lock */ | |
1574 | raw_spin_unlock(&this_rq->lock); | |
1575 | BUG_ON(1); | |
1576 | } | |
1577 | ||
1578 | return _double_lock_balance(this_rq, busiest); | |
1579 | } | |
1580 | ||
1581 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1582 | __releases(busiest->lock) | |
1583 | { | |
1584 | raw_spin_unlock(&busiest->lock); | |
1585 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1586 | } | |
1587 | ||
74602315 PZ |
1588 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1589 | { | |
1590 | if (l1 > l2) | |
1591 | swap(l1, l2); | |
1592 | ||
1593 | spin_lock(l1); | |
1594 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1595 | } | |
1596 | ||
60e69eed MG |
1597 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1598 | { | |
1599 | if (l1 > l2) | |
1600 | swap(l1, l2); | |
1601 | ||
1602 | spin_lock_irq(l1); | |
1603 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1604 | } | |
1605 | ||
74602315 PZ |
1606 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1607 | { | |
1608 | if (l1 > l2) | |
1609 | swap(l1, l2); | |
1610 | ||
1611 | raw_spin_lock(l1); | |
1612 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1613 | } | |
1614 | ||
029632fb PZ |
1615 | /* |
1616 | * double_rq_lock - safely lock two runqueues | |
1617 | * | |
1618 | * Note this does not disable interrupts like task_rq_lock, | |
1619 | * you need to do so manually before calling. | |
1620 | */ | |
1621 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1622 | __acquires(rq1->lock) | |
1623 | __acquires(rq2->lock) | |
1624 | { | |
1625 | BUG_ON(!irqs_disabled()); | |
1626 | if (rq1 == rq2) { | |
1627 | raw_spin_lock(&rq1->lock); | |
1628 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1629 | } else { | |
1630 | if (rq1 < rq2) { | |
1631 | raw_spin_lock(&rq1->lock); | |
1632 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1633 | } else { | |
1634 | raw_spin_lock(&rq2->lock); | |
1635 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1636 | } | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | /* | |
1641 | * double_rq_unlock - safely unlock two runqueues | |
1642 | * | |
1643 | * Note this does not restore interrupts like task_rq_unlock, | |
1644 | * you need to do so manually after calling. | |
1645 | */ | |
1646 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1647 | __releases(rq1->lock) | |
1648 | __releases(rq2->lock) | |
1649 | { | |
1650 | raw_spin_unlock(&rq1->lock); | |
1651 | if (rq1 != rq2) | |
1652 | raw_spin_unlock(&rq2->lock); | |
1653 | else | |
1654 | __release(rq2->lock); | |
1655 | } | |
1656 | ||
1657 | #else /* CONFIG_SMP */ | |
1658 | ||
1659 | /* | |
1660 | * double_rq_lock - safely lock two runqueues | |
1661 | * | |
1662 | * Note this does not disable interrupts like task_rq_lock, | |
1663 | * you need to do so manually before calling. | |
1664 | */ | |
1665 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1666 | __acquires(rq1->lock) | |
1667 | __acquires(rq2->lock) | |
1668 | { | |
1669 | BUG_ON(!irqs_disabled()); | |
1670 | BUG_ON(rq1 != rq2); | |
1671 | raw_spin_lock(&rq1->lock); | |
1672 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1673 | } | |
1674 | ||
1675 | /* | |
1676 | * double_rq_unlock - safely unlock two runqueues | |
1677 | * | |
1678 | * Note this does not restore interrupts like task_rq_unlock, | |
1679 | * you need to do so manually after calling. | |
1680 | */ | |
1681 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1682 | __releases(rq1->lock) | |
1683 | __releases(rq2->lock) | |
1684 | { | |
1685 | BUG_ON(rq1 != rq2); | |
1686 | raw_spin_unlock(&rq1->lock); | |
1687 | __release(rq2->lock); | |
1688 | } | |
1689 | ||
1690 | #endif | |
1691 | ||
1692 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1693 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
1694 | |
1695 | #ifdef CONFIG_SCHED_DEBUG | |
029632fb PZ |
1696 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
1697 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 1698 | extern void print_dl_stats(struct seq_file *m, int cpu); |
6b55c965 SD |
1699 | extern void |
1700 | print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); | |
397f2378 SD |
1701 | |
1702 | #ifdef CONFIG_NUMA_BALANCING | |
1703 | extern void | |
1704 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
1705 | extern void | |
1706 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
1707 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
1708 | #endif /* CONFIG_NUMA_BALANCING */ | |
1709 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
1710 | |
1711 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
1712 | extern void init_rt_rq(struct rt_rq *rt_rq); |
1713 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 1714 | |
1ee14e6c BS |
1715 | extern void cfs_bandwidth_usage_inc(void); |
1716 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1717 | |
3451d024 | 1718 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1719 | enum rq_nohz_flag_bits { |
1720 | NOHZ_TICK_STOPPED, | |
1721 | NOHZ_BALANCE_KICK, | |
1722 | }; | |
1723 | ||
1724 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
20a5c8cc TG |
1725 | |
1726 | extern void nohz_balance_exit_idle(unsigned int cpu); | |
1727 | #else | |
1728 | static inline void nohz_balance_exit_idle(unsigned int cpu) { } | |
1c792db7 | 1729 | #endif |
73fbec60 FW |
1730 | |
1731 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
1732 | ||
1733 | DECLARE_PER_CPU(u64, cpu_hardirq_time); | |
1734 | DECLARE_PER_CPU(u64, cpu_softirq_time); | |
1735 | ||
1736 | #ifndef CONFIG_64BIT | |
1737 | DECLARE_PER_CPU(seqcount_t, irq_time_seq); | |
1738 | ||
1739 | static inline void irq_time_write_begin(void) | |
1740 | { | |
1741 | __this_cpu_inc(irq_time_seq.sequence); | |
1742 | smp_wmb(); | |
1743 | } | |
1744 | ||
1745 | static inline void irq_time_write_end(void) | |
1746 | { | |
1747 | smp_wmb(); | |
1748 | __this_cpu_inc(irq_time_seq.sequence); | |
1749 | } | |
1750 | ||
1751 | static inline u64 irq_time_read(int cpu) | |
1752 | { | |
1753 | u64 irq_time; | |
1754 | unsigned seq; | |
1755 | ||
1756 | do { | |
1757 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1758 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1759 | per_cpu(cpu_hardirq_time, cpu); | |
1760 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1761 | ||
1762 | return irq_time; | |
1763 | } | |
1764 | #else /* CONFIG_64BIT */ | |
1765 | static inline void irq_time_write_begin(void) | |
1766 | { | |
1767 | } | |
1768 | ||
1769 | static inline void irq_time_write_end(void) | |
1770 | { | |
1771 | } | |
1772 | ||
1773 | static inline u64 irq_time_read(int cpu) | |
1774 | { | |
1775 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1776 | } | |
1777 | #endif /* CONFIG_64BIT */ | |
1778 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
adaf9fcd RW |
1779 | |
1780 | #ifdef CONFIG_CPU_FREQ | |
1781 | DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); | |
1782 | ||
1783 | /** | |
1784 | * cpufreq_update_util - Take a note about CPU utilization changes. | |
1785 | * @time: Current time. | |
1786 | * @util: Current utilization. | |
1787 | * @max: Utilization ceiling. | |
1788 | * | |
1789 | * This function is called by the scheduler on every invocation of | |
1790 | * update_load_avg() on the CPU whose utilization is being updated. | |
1791 | * | |
1792 | * It can only be called from RCU-sched read-side critical sections. | |
1793 | */ | |
1794 | static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) | |
1795 | { | |
1796 | struct update_util_data *data; | |
1797 | ||
1798 | data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data)); | |
1799 | if (data) | |
1800 | data->func(data, time, util, max); | |
1801 | } | |
1802 | ||
1803 | /** | |
1804 | * cpufreq_trigger_update - Trigger CPU performance state evaluation if needed. | |
1805 | * @time: Current time. | |
1806 | * | |
1807 | * The way cpufreq is currently arranged requires it to evaluate the CPU | |
1808 | * performance state (frequency/voltage) on a regular basis to prevent it from | |
1809 | * being stuck in a completely inadequate performance level for too long. | |
1810 | * That is not guaranteed to happen if the updates are only triggered from CFS, | |
1811 | * though, because they may not be coming in if RT or deadline tasks are active | |
1812 | * all the time (or there are RT and DL tasks only). | |
1813 | * | |
1814 | * As a workaround for that issue, this function is called by the RT and DL | |
1815 | * sched classes to trigger extra cpufreq updates to prevent it from stalling, | |
1816 | * but that really is a band-aid. Going forward it should be replaced with | |
1817 | * solutions targeted more specifically at RT and DL tasks. | |
1818 | */ | |
1819 | static inline void cpufreq_trigger_update(u64 time) | |
1820 | { | |
1821 | cpufreq_update_util(time, ULONG_MAX, 0); | |
1822 | } | |
1823 | #else | |
1824 | static inline void cpufreq_update_util(u64 time, unsigned long util, unsigned long max) {} | |
1825 | static inline void cpufreq_trigger_update(u64 time) {} | |
1826 | #endif /* CONFIG_CPU_FREQ */ | |
be53f58f | 1827 | |
9bdcb44e RW |
1828 | #ifdef arch_scale_freq_capacity |
1829 | #ifndef arch_scale_freq_invariant | |
1830 | #define arch_scale_freq_invariant() (true) | |
1831 | #endif | |
1832 | #else /* arch_scale_freq_capacity */ | |
1833 | #define arch_scale_freq_invariant() (false) | |
1834 | #endif |