]>
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> |
19d23dbf | 5 | #include <linux/u64_stats_sync.h> |
aab03e05 | 6 | #include <linux/sched/deadline.h> |
a499a5a1 | 7 | #include <linux/kernel_stat.h> |
3866e845 | 8 | #include <linux/binfmts.h> |
029632fb PZ |
9 | #include <linux/mutex.h> |
10 | #include <linux/spinlock.h> | |
11 | #include <linux/stop_machine.h> | |
b6366f04 | 12 | #include <linux/irq_work.h> |
9f3660c2 | 13 | #include <linux/tick.h> |
f809ca9a | 14 | #include <linux/slab.h> |
029632fb | 15 | |
391e43da | 16 | #include "cpupri.h" |
6bfd6d72 | 17 | #include "cpudeadline.h" |
60fed789 | 18 | #include "cpuacct.h" |
029632fb | 19 | |
9148a3a1 PZ |
20 | #ifdef CONFIG_SCHED_DEBUG |
21 | #define SCHED_WARN_ON(x) WARN_ONCE(x, #x) | |
22 | #else | |
23 | #define SCHED_WARN_ON(x) ((void)(x)) | |
24 | #endif | |
25 | ||
45ceebf7 | 26 | struct rq; |
442bf3aa | 27 | struct cpuidle_state; |
45ceebf7 | 28 | |
da0c1e65 KT |
29 | /* task_struct::on_rq states: */ |
30 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 31 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 32 | |
029632fb PZ |
33 | extern __read_mostly int scheduler_running; |
34 | ||
45ceebf7 PG |
35 | extern unsigned long calc_load_update; |
36 | extern atomic_long_t calc_load_tasks; | |
37 | ||
3289bdb4 | 38 | extern void calc_global_load_tick(struct rq *this_rq); |
d60585c5 | 39 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
3289bdb4 PZ |
40 | |
41 | #ifdef CONFIG_SMP | |
cee1afce | 42 | extern void cpu_load_update_active(struct rq *this_rq); |
3289bdb4 | 43 | #else |
cee1afce | 44 | static inline void cpu_load_update_active(struct rq *this_rq) { } |
3289bdb4 | 45 | #endif |
45ceebf7 | 46 | |
029632fb PZ |
47 | /* |
48 | * Helpers for converting nanosecond timing to jiffy resolution | |
49 | */ | |
50 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
51 | ||
cc1f4b1f LZ |
52 | /* |
53 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
54 | * The extra resolution improves shares distribution and load balancing of | |
55 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
56 | * hierarchies, especially on larger systems. This is not a user-visible change | |
57 | * and does not change the user-interface for setting shares/weights. | |
58 | * | |
59 | * We increase resolution only if we have enough bits to allow this increased | |
2159197d PZ |
60 | * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are |
61 | * pretty high and the returns do not justify the increased costs. | |
62 | * | |
63 | * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to | |
64 | * increase coverage and consistency always enable it on 64bit platforms. | |
cc1f4b1f | 65 | */ |
2159197d | 66 | #ifdef CONFIG_64BIT |
172895e6 | 67 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
6ecdd749 YD |
68 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
69 | # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) | |
cc1f4b1f | 70 | #else |
172895e6 | 71 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
cc1f4b1f LZ |
72 | # define scale_load(w) (w) |
73 | # define scale_load_down(w) (w) | |
74 | #endif | |
75 | ||
6ecdd749 | 76 | /* |
172895e6 YD |
77 | * Task weight (visible to users) and its load (invisible to users) have |
78 | * independent resolution, but they should be well calibrated. We use | |
79 | * scale_load() and scale_load_down(w) to convert between them. The | |
80 | * following must be true: | |
81 | * | |
82 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD | |
83 | * | |
6ecdd749 | 84 | */ |
172895e6 | 85 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
029632fb | 86 | |
332ac17e DF |
87 | /* |
88 | * Single value that decides SCHED_DEADLINE internal math precision. | |
89 | * 10 -> just above 1us | |
90 | * 9 -> just above 0.5us | |
91 | */ | |
92 | #define DL_SCALE (10) | |
93 | ||
029632fb PZ |
94 | /* |
95 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 96 | */ |
029632fb PZ |
97 | |
98 | /* | |
99 | * single value that denotes runtime == period, ie unlimited time. | |
100 | */ | |
101 | #define RUNTIME_INF ((u64)~0ULL) | |
102 | ||
20f9cd2a HA |
103 | static inline int idle_policy(int policy) |
104 | { | |
105 | return policy == SCHED_IDLE; | |
106 | } | |
d50dde5a DF |
107 | static inline int fair_policy(int policy) |
108 | { | |
109 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
110 | } | |
111 | ||
029632fb PZ |
112 | static inline int rt_policy(int policy) |
113 | { | |
d50dde5a | 114 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
115 | } |
116 | ||
aab03e05 DF |
117 | static inline int dl_policy(int policy) |
118 | { | |
119 | return policy == SCHED_DEADLINE; | |
120 | } | |
20f9cd2a HA |
121 | static inline bool valid_policy(int policy) |
122 | { | |
123 | return idle_policy(policy) || fair_policy(policy) || | |
124 | rt_policy(policy) || dl_policy(policy); | |
125 | } | |
aab03e05 | 126 | |
029632fb PZ |
127 | static inline int task_has_rt_policy(struct task_struct *p) |
128 | { | |
129 | return rt_policy(p->policy); | |
130 | } | |
131 | ||
aab03e05 DF |
132 | static inline int task_has_dl_policy(struct task_struct *p) |
133 | { | |
134 | return dl_policy(p->policy); | |
135 | } | |
136 | ||
2d3d891d DF |
137 | /* |
138 | * Tells if entity @a should preempt entity @b. | |
139 | */ | |
332ac17e DF |
140 | static inline bool |
141 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
142 | { |
143 | return dl_time_before(a->deadline, b->deadline); | |
144 | } | |
145 | ||
029632fb PZ |
146 | /* |
147 | * This is the priority-queue data structure of the RT scheduling class: | |
148 | */ | |
149 | struct rt_prio_array { | |
150 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
151 | struct list_head queue[MAX_RT_PRIO]; | |
152 | }; | |
153 | ||
154 | struct rt_bandwidth { | |
155 | /* nests inside the rq lock: */ | |
156 | raw_spinlock_t rt_runtime_lock; | |
157 | ktime_t rt_period; | |
158 | u64 rt_runtime; | |
159 | struct hrtimer rt_period_timer; | |
4cfafd30 | 160 | unsigned int rt_period_active; |
029632fb | 161 | }; |
a5e7be3b JL |
162 | |
163 | void __dl_clear_params(struct task_struct *p); | |
164 | ||
332ac17e DF |
165 | /* |
166 | * To keep the bandwidth of -deadline tasks and groups under control | |
167 | * we need some place where: | |
168 | * - store the maximum -deadline bandwidth of the system (the group); | |
169 | * - cache the fraction of that bandwidth that is currently allocated. | |
170 | * | |
171 | * This is all done in the data structure below. It is similar to the | |
172 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
173 | * that, since here we are only interested in admission control, we | |
174 | * do not decrease any runtime while the group "executes", neither we | |
175 | * need a timer to replenish it. | |
176 | * | |
177 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
178 | * meaning that: | |
179 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
180 | * - dl_total_bw array contains, in the i-eth element, the currently | |
181 | * allocated bandwidth on the i-eth CPU. | |
182 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
183 | * consume bandwidth on the CPU they're running on. | |
184 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
185 | * that will be shown the next time the proc or cgroup controls will | |
186 | * be red. It on its turn can be changed by writing on its own | |
187 | * control. | |
188 | */ | |
189 | struct dl_bandwidth { | |
190 | raw_spinlock_t dl_runtime_lock; | |
191 | u64 dl_runtime; | |
192 | u64 dl_period; | |
193 | }; | |
194 | ||
195 | static inline int dl_bandwidth_enabled(void) | |
196 | { | |
1724813d | 197 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
198 | } |
199 | ||
200 | extern struct dl_bw *dl_bw_of(int i); | |
201 | ||
202 | struct dl_bw { | |
203 | raw_spinlock_t lock; | |
204 | u64 bw, total_bw; | |
205 | }; | |
206 | ||
7f51412a JL |
207 | static inline |
208 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
209 | { | |
210 | dl_b->total_bw -= tsk_bw; | |
211 | } | |
212 | ||
213 | static inline | |
214 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
215 | { | |
216 | dl_b->total_bw += tsk_bw; | |
217 | } | |
218 | ||
219 | static inline | |
220 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
221 | { | |
222 | return dl_b->bw != -1 && | |
223 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
224 | } | |
225 | ||
f2cb1360 | 226 | extern void init_dl_bw(struct dl_bw *dl_b); |
029632fb PZ |
227 | |
228 | #ifdef CONFIG_CGROUP_SCHED | |
229 | ||
230 | #include <linux/cgroup.h> | |
231 | ||
232 | struct cfs_rq; | |
233 | struct rt_rq; | |
234 | ||
35cf4e50 | 235 | extern struct list_head task_groups; |
029632fb PZ |
236 | |
237 | struct cfs_bandwidth { | |
238 | #ifdef CONFIG_CFS_BANDWIDTH | |
239 | raw_spinlock_t lock; | |
240 | ktime_t period; | |
241 | u64 quota, runtime; | |
9c58c79a | 242 | s64 hierarchical_quota; |
029632fb PZ |
243 | u64 runtime_expires; |
244 | ||
4cfafd30 | 245 | int idle, period_active; |
029632fb PZ |
246 | struct hrtimer period_timer, slack_timer; |
247 | struct list_head throttled_cfs_rq; | |
248 | ||
249 | /* statistics */ | |
250 | int nr_periods, nr_throttled; | |
251 | u64 throttled_time; | |
252 | #endif | |
253 | }; | |
254 | ||
255 | /* task group related information */ | |
256 | struct task_group { | |
257 | struct cgroup_subsys_state css; | |
258 | ||
259 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
260 | /* schedulable entities of this group on each cpu */ | |
261 | struct sched_entity **se; | |
262 | /* runqueue "owned" by this group on each cpu */ | |
263 | struct cfs_rq **cfs_rq; | |
264 | unsigned long shares; | |
265 | ||
fa6bddeb | 266 | #ifdef CONFIG_SMP |
b0367629 WL |
267 | /* |
268 | * load_avg can be heavily contended at clock tick time, so put | |
269 | * it in its own cacheline separated from the fields above which | |
270 | * will also be accessed at each tick. | |
271 | */ | |
272 | atomic_long_t load_avg ____cacheline_aligned; | |
029632fb | 273 | #endif |
fa6bddeb | 274 | #endif |
029632fb PZ |
275 | |
276 | #ifdef CONFIG_RT_GROUP_SCHED | |
277 | struct sched_rt_entity **rt_se; | |
278 | struct rt_rq **rt_rq; | |
279 | ||
280 | struct rt_bandwidth rt_bandwidth; | |
281 | #endif | |
282 | ||
283 | struct rcu_head rcu; | |
284 | struct list_head list; | |
285 | ||
286 | struct task_group *parent; | |
287 | struct list_head siblings; | |
288 | struct list_head children; | |
289 | ||
290 | #ifdef CONFIG_SCHED_AUTOGROUP | |
291 | struct autogroup *autogroup; | |
292 | #endif | |
293 | ||
294 | struct cfs_bandwidth cfs_bandwidth; | |
295 | }; | |
296 | ||
297 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
298 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
299 | ||
300 | /* | |
301 | * A weight of 0 or 1 can cause arithmetics problems. | |
302 | * A weight of a cfs_rq is the sum of weights of which entities | |
303 | * are queued on this cfs_rq, so a weight of a entity should not be | |
304 | * too large, so as the shares value of a task group. | |
305 | * (The default weight is 1024 - so there's no practical | |
306 | * limitation from this.) | |
307 | */ | |
308 | #define MIN_SHARES (1UL << 1) | |
309 | #define MAX_SHARES (1UL << 18) | |
310 | #endif | |
311 | ||
029632fb PZ |
312 | typedef int (*tg_visitor)(struct task_group *, void *); |
313 | ||
314 | extern int walk_tg_tree_from(struct task_group *from, | |
315 | tg_visitor down, tg_visitor up, void *data); | |
316 | ||
317 | /* | |
318 | * Iterate the full tree, calling @down when first entering a node and @up when | |
319 | * leaving it for the final time. | |
320 | * | |
321 | * Caller must hold rcu_lock or sufficient equivalent. | |
322 | */ | |
323 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
324 | { | |
325 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
326 | } | |
327 | ||
328 | extern int tg_nop(struct task_group *tg, void *data); | |
329 | ||
330 | extern void free_fair_sched_group(struct task_group *tg); | |
331 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
8663e24d | 332 | extern void online_fair_sched_group(struct task_group *tg); |
6fe1f348 | 333 | extern void unregister_fair_sched_group(struct task_group *tg); |
029632fb PZ |
334 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
335 | struct sched_entity *se, int cpu, | |
336 | struct sched_entity *parent); | |
337 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
029632fb PZ |
338 | |
339 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 340 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
341 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
342 | ||
343 | extern void free_rt_sched_group(struct task_group *tg); | |
344 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
345 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
346 | struct sched_rt_entity *rt_se, int cpu, | |
347 | struct sched_rt_entity *parent); | |
348 | ||
25cc7da7 LZ |
349 | extern struct task_group *sched_create_group(struct task_group *parent); |
350 | extern void sched_online_group(struct task_group *tg, | |
351 | struct task_group *parent); | |
352 | extern void sched_destroy_group(struct task_group *tg); | |
353 | extern void sched_offline_group(struct task_group *tg); | |
354 | ||
355 | extern void sched_move_task(struct task_struct *tsk); | |
356 | ||
357 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
358 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
ad936d86 BP |
359 | |
360 | #ifdef CONFIG_SMP | |
361 | extern void set_task_rq_fair(struct sched_entity *se, | |
362 | struct cfs_rq *prev, struct cfs_rq *next); | |
363 | #else /* !CONFIG_SMP */ | |
364 | static inline void set_task_rq_fair(struct sched_entity *se, | |
365 | struct cfs_rq *prev, struct cfs_rq *next) { } | |
366 | #endif /* CONFIG_SMP */ | |
367 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
25cc7da7 | 368 | |
029632fb PZ |
369 | #else /* CONFIG_CGROUP_SCHED */ |
370 | ||
371 | struct cfs_bandwidth { }; | |
372 | ||
373 | #endif /* CONFIG_CGROUP_SCHED */ | |
374 | ||
375 | /* CFS-related fields in a runqueue */ | |
376 | struct cfs_rq { | |
377 | struct load_weight load; | |
c82513e5 | 378 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
379 | |
380 | u64 exec_clock; | |
381 | u64 min_vruntime; | |
382 | #ifndef CONFIG_64BIT | |
383 | u64 min_vruntime_copy; | |
384 | #endif | |
385 | ||
386 | struct rb_root tasks_timeline; | |
387 | struct rb_node *rb_leftmost; | |
388 | ||
029632fb PZ |
389 | /* |
390 | * 'curr' points to currently running entity on this cfs_rq. | |
391 | * It is set to NULL otherwise (i.e when none are currently running). | |
392 | */ | |
393 | struct sched_entity *curr, *next, *last, *skip; | |
394 | ||
395 | #ifdef CONFIG_SCHED_DEBUG | |
396 | unsigned int nr_spread_over; | |
397 | #endif | |
398 | ||
2dac754e PT |
399 | #ifdef CONFIG_SMP |
400 | /* | |
9d89c257 | 401 | * CFS load tracking |
2dac754e | 402 | */ |
9d89c257 | 403 | struct sched_avg avg; |
13962234 YD |
404 | u64 runnable_load_sum; |
405 | unsigned long runnable_load_avg; | |
c566e8e9 | 406 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9d89c257 | 407 | unsigned long tg_load_avg_contrib; |
09a43ace | 408 | unsigned long propagate_avg; |
9d89c257 YD |
409 | #endif |
410 | atomic_long_t removed_load_avg, removed_util_avg; | |
411 | #ifndef CONFIG_64BIT | |
412 | u64 load_last_update_time_copy; | |
413 | #endif | |
82958366 | 414 | |
9d89c257 | 415 | #ifdef CONFIG_FAIR_GROUP_SCHED |
82958366 PT |
416 | /* |
417 | * h_load = weight * f(tg) | |
418 | * | |
419 | * Where f(tg) is the recursive weight fraction assigned to | |
420 | * this group. | |
421 | */ | |
422 | unsigned long h_load; | |
68520796 VD |
423 | u64 last_h_load_update; |
424 | struct sched_entity *h_load_next; | |
425 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
426 | #endif /* CONFIG_SMP */ |
427 | ||
029632fb PZ |
428 | #ifdef CONFIG_FAIR_GROUP_SCHED |
429 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
430 | ||
431 | /* | |
432 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
433 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
434 | * (like users, containers etc.) | |
435 | * | |
436 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
437 | * list is used during load balance. | |
438 | */ | |
439 | int on_list; | |
440 | struct list_head leaf_cfs_rq_list; | |
441 | struct task_group *tg; /* group that "owns" this runqueue */ | |
442 | ||
029632fb PZ |
443 | #ifdef CONFIG_CFS_BANDWIDTH |
444 | int runtime_enabled; | |
445 | u64 runtime_expires; | |
446 | s64 runtime_remaining; | |
447 | ||
f1b17280 PT |
448 | u64 throttled_clock, throttled_clock_task; |
449 | u64 throttled_clock_task_time; | |
55e16d30 | 450 | int throttled, throttle_count; |
029632fb PZ |
451 | struct list_head throttled_list; |
452 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
453 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
454 | }; | |
455 | ||
456 | static inline int rt_bandwidth_enabled(void) | |
457 | { | |
458 | return sysctl_sched_rt_runtime >= 0; | |
459 | } | |
460 | ||
b6366f04 SR |
461 | /* RT IPI pull logic requires IRQ_WORK */ |
462 | #ifdef CONFIG_IRQ_WORK | |
463 | # define HAVE_RT_PUSH_IPI | |
464 | #endif | |
465 | ||
029632fb PZ |
466 | /* Real-Time classes' related field in a runqueue: */ |
467 | struct rt_rq { | |
468 | struct rt_prio_array active; | |
c82513e5 | 469 | unsigned int rt_nr_running; |
01d36d0a | 470 | unsigned int rr_nr_running; |
029632fb PZ |
471 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
472 | struct { | |
473 | int curr; /* highest queued rt task prio */ | |
474 | #ifdef CONFIG_SMP | |
475 | int next; /* next highest */ | |
476 | #endif | |
477 | } highest_prio; | |
478 | #endif | |
479 | #ifdef CONFIG_SMP | |
480 | unsigned long rt_nr_migratory; | |
481 | unsigned long rt_nr_total; | |
482 | int overloaded; | |
483 | struct plist_head pushable_tasks; | |
b6366f04 SR |
484 | #ifdef HAVE_RT_PUSH_IPI |
485 | int push_flags; | |
486 | int push_cpu; | |
487 | struct irq_work push_work; | |
488 | raw_spinlock_t push_lock; | |
029632fb | 489 | #endif |
b6366f04 | 490 | #endif /* CONFIG_SMP */ |
f4ebcbc0 KT |
491 | int rt_queued; |
492 | ||
029632fb PZ |
493 | int rt_throttled; |
494 | u64 rt_time; | |
495 | u64 rt_runtime; | |
496 | /* Nests inside the rq lock: */ | |
497 | raw_spinlock_t rt_runtime_lock; | |
498 | ||
499 | #ifdef CONFIG_RT_GROUP_SCHED | |
500 | unsigned long rt_nr_boosted; | |
501 | ||
502 | struct rq *rq; | |
029632fb PZ |
503 | struct task_group *tg; |
504 | #endif | |
505 | }; | |
506 | ||
aab03e05 DF |
507 | /* Deadline class' related fields in a runqueue */ |
508 | struct dl_rq { | |
509 | /* runqueue is an rbtree, ordered by deadline */ | |
510 | struct rb_root rb_root; | |
511 | struct rb_node *rb_leftmost; | |
512 | ||
513 | unsigned long dl_nr_running; | |
1baca4ce JL |
514 | |
515 | #ifdef CONFIG_SMP | |
516 | /* | |
517 | * Deadline values of the currently executing and the | |
518 | * earliest ready task on this rq. Caching these facilitates | |
519 | * the decision wether or not a ready but not running task | |
520 | * should migrate somewhere else. | |
521 | */ | |
522 | struct { | |
523 | u64 curr; | |
524 | u64 next; | |
525 | } earliest_dl; | |
526 | ||
527 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
528 | int overloaded; |
529 | ||
530 | /* | |
531 | * Tasks on this rq that can be pushed away. They are kept in | |
532 | * an rb-tree, ordered by tasks' deadlines, with caching | |
533 | * of the leftmost (earliest deadline) element. | |
534 | */ | |
535 | struct rb_root pushable_dl_tasks_root; | |
536 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
537 | #else |
538 | struct dl_bw dl_bw; | |
1baca4ce | 539 | #endif |
aab03e05 DF |
540 | }; |
541 | ||
029632fb PZ |
542 | #ifdef CONFIG_SMP |
543 | ||
afe06efd TC |
544 | static inline bool sched_asym_prefer(int a, int b) |
545 | { | |
546 | return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); | |
547 | } | |
548 | ||
029632fb PZ |
549 | /* |
550 | * We add the notion of a root-domain which will be used to define per-domain | |
551 | * variables. Each exclusive cpuset essentially defines an island domain by | |
552 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
553 | * exclusive cpuset is created, we also create and attach a new root-domain | |
554 | * object. | |
555 | * | |
556 | */ | |
557 | struct root_domain { | |
558 | atomic_t refcount; | |
559 | atomic_t rto_count; | |
560 | struct rcu_head rcu; | |
561 | cpumask_var_t span; | |
562 | cpumask_var_t online; | |
563 | ||
4486edd1 TC |
564 | /* Indicate more than one runnable task for any CPU */ |
565 | bool overload; | |
566 | ||
1baca4ce JL |
567 | /* |
568 | * The bit corresponding to a CPU gets set here if such CPU has more | |
569 | * than one runnable -deadline task (as it is below for RT tasks). | |
570 | */ | |
571 | cpumask_var_t dlo_mask; | |
572 | atomic_t dlo_count; | |
332ac17e | 573 | struct dl_bw dl_bw; |
6bfd6d72 | 574 | struct cpudl cpudl; |
1baca4ce | 575 | |
029632fb PZ |
576 | /* |
577 | * The "RT overload" flag: it gets set if a CPU has more than | |
578 | * one runnable RT task. | |
579 | */ | |
580 | cpumask_var_t rto_mask; | |
581 | struct cpupri cpupri; | |
cd92bfd3 DE |
582 | |
583 | unsigned long max_cpu_capacity; | |
029632fb PZ |
584 | }; |
585 | ||
586 | extern struct root_domain def_root_domain; | |
f2cb1360 IM |
587 | extern struct mutex sched_domains_mutex; |
588 | extern cpumask_var_t fallback_doms; | |
589 | extern cpumask_var_t sched_domains_tmpmask; | |
590 | ||
591 | extern void init_defrootdomain(void); | |
592 | extern int init_sched_domains(const struct cpumask *cpu_map); | |
593 | extern void rq_attach_root(struct rq *rq, struct root_domain *rd); | |
029632fb PZ |
594 | |
595 | #endif /* CONFIG_SMP */ | |
596 | ||
597 | /* | |
598 | * This is the main, per-CPU runqueue data structure. | |
599 | * | |
600 | * Locking rule: those places that want to lock multiple runqueues | |
601 | * (such as the load balancing or the thread migration code), lock | |
602 | * acquire operations must be ordered by ascending &runqueue. | |
603 | */ | |
604 | struct rq { | |
605 | /* runqueue lock: */ | |
606 | raw_spinlock_t lock; | |
607 | ||
608 | /* | |
609 | * nr_running and cpu_load should be in the same cacheline because | |
610 | * remote CPUs use both these fields when doing load calculation. | |
611 | */ | |
c82513e5 | 612 | unsigned int nr_running; |
0ec8aa00 PZ |
613 | #ifdef CONFIG_NUMA_BALANCING |
614 | unsigned int nr_numa_running; | |
615 | unsigned int nr_preferred_running; | |
616 | #endif | |
029632fb PZ |
617 | #define CPU_LOAD_IDX_MAX 5 |
618 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
3451d024 | 619 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 FW |
620 | #ifdef CONFIG_SMP |
621 | unsigned long last_load_update_tick; | |
622 | #endif /* CONFIG_SMP */ | |
1c792db7 | 623 | unsigned long nohz_flags; |
9fd81dd5 | 624 | #endif /* CONFIG_NO_HZ_COMMON */ |
265f22a9 FW |
625 | #ifdef CONFIG_NO_HZ_FULL |
626 | unsigned long last_sched_tick; | |
029632fb | 627 | #endif |
029632fb PZ |
628 | /* capture load from *all* tasks on this cpu: */ |
629 | struct load_weight load; | |
630 | unsigned long nr_load_updates; | |
631 | u64 nr_switches; | |
632 | ||
633 | struct cfs_rq cfs; | |
634 | struct rt_rq rt; | |
aab03e05 | 635 | struct dl_rq dl; |
029632fb PZ |
636 | |
637 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
638 | /* list of leaf cfs_rq on this cpu: */ | |
639 | struct list_head leaf_cfs_rq_list; | |
9c2791f9 | 640 | struct list_head *tmp_alone_branch; |
a35b6466 PZ |
641 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
642 | ||
029632fb PZ |
643 | /* |
644 | * This is part of a global counter where only the total sum | |
645 | * over all CPUs matters. A task can increase this counter on | |
646 | * one CPU and if it got migrated afterwards it may decrease | |
647 | * it on another CPU. Always updated under the runqueue lock: | |
648 | */ | |
649 | unsigned long nr_uninterruptible; | |
650 | ||
651 | struct task_struct *curr, *idle, *stop; | |
652 | unsigned long next_balance; | |
653 | struct mm_struct *prev_mm; | |
654 | ||
cb42c9a3 | 655 | unsigned int clock_update_flags; |
029632fb PZ |
656 | u64 clock; |
657 | u64 clock_task; | |
658 | ||
659 | atomic_t nr_iowait; | |
660 | ||
661 | #ifdef CONFIG_SMP | |
662 | struct root_domain *rd; | |
663 | struct sched_domain *sd; | |
664 | ||
ced549fa | 665 | unsigned long cpu_capacity; |
ca6d75e6 | 666 | unsigned long cpu_capacity_orig; |
029632fb | 667 | |
e3fca9e7 PZ |
668 | struct callback_head *balance_callback; |
669 | ||
029632fb PZ |
670 | unsigned char idle_balance; |
671 | /* For active balancing */ | |
029632fb PZ |
672 | int active_balance; |
673 | int push_cpu; | |
674 | struct cpu_stop_work active_balance_work; | |
675 | /* cpu of this runqueue: */ | |
676 | int cpu; | |
677 | int online; | |
678 | ||
367456c7 PZ |
679 | struct list_head cfs_tasks; |
680 | ||
029632fb PZ |
681 | u64 rt_avg; |
682 | u64 age_stamp; | |
683 | u64 idle_stamp; | |
684 | u64 avg_idle; | |
9bd721c5 JL |
685 | |
686 | /* This is used to determine avg_idle's max value */ | |
687 | u64 max_idle_balance_cost; | |
029632fb PZ |
688 | #endif |
689 | ||
690 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
691 | u64 prev_irq_time; | |
692 | #endif | |
693 | #ifdef CONFIG_PARAVIRT | |
694 | u64 prev_steal_time; | |
695 | #endif | |
696 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
697 | u64 prev_steal_time_rq; | |
698 | #endif | |
699 | ||
700 | /* calc_load related fields */ | |
701 | unsigned long calc_load_update; | |
702 | long calc_load_active; | |
703 | ||
704 | #ifdef CONFIG_SCHED_HRTICK | |
705 | #ifdef CONFIG_SMP | |
706 | int hrtick_csd_pending; | |
707 | struct call_single_data hrtick_csd; | |
708 | #endif | |
709 | struct hrtimer hrtick_timer; | |
710 | #endif | |
711 | ||
712 | #ifdef CONFIG_SCHEDSTATS | |
713 | /* latency stats */ | |
714 | struct sched_info rq_sched_info; | |
715 | unsigned long long rq_cpu_time; | |
716 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
717 | ||
718 | /* sys_sched_yield() stats */ | |
719 | unsigned int yld_count; | |
720 | ||
721 | /* schedule() stats */ | |
029632fb PZ |
722 | unsigned int sched_count; |
723 | unsigned int sched_goidle; | |
724 | ||
725 | /* try_to_wake_up() stats */ | |
726 | unsigned int ttwu_count; | |
727 | unsigned int ttwu_local; | |
728 | #endif | |
729 | ||
730 | #ifdef CONFIG_SMP | |
731 | struct llist_head wake_list; | |
732 | #endif | |
442bf3aa DL |
733 | |
734 | #ifdef CONFIG_CPU_IDLE | |
735 | /* Must be inspected within a rcu lock section */ | |
736 | struct cpuidle_state *idle_state; | |
737 | #endif | |
029632fb PZ |
738 | }; |
739 | ||
740 | static inline int cpu_of(struct rq *rq) | |
741 | { | |
742 | #ifdef CONFIG_SMP | |
743 | return rq->cpu; | |
744 | #else | |
745 | return 0; | |
746 | #endif | |
747 | } | |
748 | ||
1b568f0a PZ |
749 | |
750 | #ifdef CONFIG_SCHED_SMT | |
751 | ||
752 | extern struct static_key_false sched_smt_present; | |
753 | ||
754 | extern void __update_idle_core(struct rq *rq); | |
755 | ||
756 | static inline void update_idle_core(struct rq *rq) | |
757 | { | |
758 | if (static_branch_unlikely(&sched_smt_present)) | |
759 | __update_idle_core(rq); | |
760 | } | |
761 | ||
762 | #else | |
763 | static inline void update_idle_core(struct rq *rq) { } | |
764 | #endif | |
765 | ||
8b06c55b | 766 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 767 | |
518cd623 | 768 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 769 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
770 | #define task_rq(p) cpu_rq(task_cpu(p)) |
771 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 772 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 773 | |
cebde6d6 PZ |
774 | static inline u64 __rq_clock_broken(struct rq *rq) |
775 | { | |
316c1608 | 776 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
777 | } |
778 | ||
cb42c9a3 MF |
779 | /* |
780 | * rq::clock_update_flags bits | |
781 | * | |
782 | * %RQCF_REQ_SKIP - will request skipping of clock update on the next | |
783 | * call to __schedule(). This is an optimisation to avoid | |
784 | * neighbouring rq clock updates. | |
785 | * | |
786 | * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is | |
787 | * in effect and calls to update_rq_clock() are being ignored. | |
788 | * | |
789 | * %RQCF_UPDATED - is a debug flag that indicates whether a call has been | |
790 | * made to update_rq_clock() since the last time rq::lock was pinned. | |
791 | * | |
792 | * If inside of __schedule(), clock_update_flags will have been | |
793 | * shifted left (a left shift is a cheap operation for the fast path | |
794 | * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, | |
795 | * | |
796 | * if (rq-clock_update_flags >= RQCF_UPDATED) | |
797 | * | |
798 | * to check if %RQCF_UPADTED is set. It'll never be shifted more than | |
799 | * one position though, because the next rq_unpin_lock() will shift it | |
800 | * back. | |
801 | */ | |
802 | #define RQCF_REQ_SKIP 0x01 | |
803 | #define RQCF_ACT_SKIP 0x02 | |
804 | #define RQCF_UPDATED 0x04 | |
805 | ||
806 | static inline void assert_clock_updated(struct rq *rq) | |
807 | { | |
808 | /* | |
809 | * The only reason for not seeing a clock update since the | |
810 | * last rq_pin_lock() is if we're currently skipping updates. | |
811 | */ | |
812 | SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); | |
813 | } | |
814 | ||
78becc27 FW |
815 | static inline u64 rq_clock(struct rq *rq) |
816 | { | |
cebde6d6 | 817 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
818 | assert_clock_updated(rq); |
819 | ||
78becc27 FW |
820 | return rq->clock; |
821 | } | |
822 | ||
823 | static inline u64 rq_clock_task(struct rq *rq) | |
824 | { | |
cebde6d6 | 825 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
826 | assert_clock_updated(rq); |
827 | ||
78becc27 FW |
828 | return rq->clock_task; |
829 | } | |
830 | ||
9edfbfed PZ |
831 | static inline void rq_clock_skip_update(struct rq *rq, bool skip) |
832 | { | |
833 | lockdep_assert_held(&rq->lock); | |
834 | if (skip) | |
cb42c9a3 | 835 | rq->clock_update_flags |= RQCF_REQ_SKIP; |
9edfbfed | 836 | else |
cb42c9a3 | 837 | rq->clock_update_flags &= ~RQCF_REQ_SKIP; |
9edfbfed PZ |
838 | } |
839 | ||
d8ac8971 MF |
840 | struct rq_flags { |
841 | unsigned long flags; | |
842 | struct pin_cookie cookie; | |
cb42c9a3 MF |
843 | #ifdef CONFIG_SCHED_DEBUG |
844 | /* | |
845 | * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the | |
846 | * current pin context is stashed here in case it needs to be | |
847 | * restored in rq_repin_lock(). | |
848 | */ | |
849 | unsigned int clock_update_flags; | |
850 | #endif | |
d8ac8971 MF |
851 | }; |
852 | ||
853 | static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) | |
854 | { | |
855 | rf->cookie = lockdep_pin_lock(&rq->lock); | |
cb42c9a3 MF |
856 | |
857 | #ifdef CONFIG_SCHED_DEBUG | |
858 | rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); | |
859 | rf->clock_update_flags = 0; | |
860 | #endif | |
d8ac8971 MF |
861 | } |
862 | ||
863 | static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) | |
864 | { | |
cb42c9a3 MF |
865 | #ifdef CONFIG_SCHED_DEBUG |
866 | if (rq->clock_update_flags > RQCF_ACT_SKIP) | |
867 | rf->clock_update_flags = RQCF_UPDATED; | |
868 | #endif | |
869 | ||
d8ac8971 MF |
870 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
871 | } | |
872 | ||
873 | static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) | |
874 | { | |
875 | lockdep_repin_lock(&rq->lock, rf->cookie); | |
cb42c9a3 MF |
876 | |
877 | #ifdef CONFIG_SCHED_DEBUG | |
878 | /* | |
879 | * Restore the value we stashed in @rf for this pin context. | |
880 | */ | |
881 | rq->clock_update_flags |= rf->clock_update_flags; | |
882 | #endif | |
d8ac8971 MF |
883 | } |
884 | ||
9942f79b | 885 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
886 | enum numa_topology_type { |
887 | NUMA_DIRECT, | |
888 | NUMA_GLUELESS_MESH, | |
889 | NUMA_BACKPLANE, | |
890 | }; | |
891 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
892 | extern int sched_max_numa_distance; |
893 | extern bool find_numa_distance(int distance); | |
894 | #endif | |
895 | ||
f2cb1360 IM |
896 | #ifdef CONFIG_NUMA |
897 | extern void sched_init_numa(void); | |
898 | extern void sched_domains_numa_masks_set(unsigned int cpu); | |
899 | extern void sched_domains_numa_masks_clear(unsigned int cpu); | |
900 | #else | |
901 | static inline void sched_init_numa(void) { } | |
902 | static inline void sched_domains_numa_masks_set(unsigned int cpu) { } | |
903 | static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
904 | #endif | |
905 | ||
f809ca9a | 906 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
907 | /* The regions in numa_faults array from task_struct */ |
908 | enum numa_faults_stats { | |
909 | NUMA_MEM = 0, | |
910 | NUMA_CPU, | |
911 | NUMA_MEMBUF, | |
912 | NUMA_CPUBUF | |
913 | }; | |
0ec8aa00 | 914 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 915 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 916 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
917 | #endif /* CONFIG_NUMA_BALANCING */ |
918 | ||
518cd623 PZ |
919 | #ifdef CONFIG_SMP |
920 | ||
e3fca9e7 PZ |
921 | static inline void |
922 | queue_balance_callback(struct rq *rq, | |
923 | struct callback_head *head, | |
924 | void (*func)(struct rq *rq)) | |
925 | { | |
926 | lockdep_assert_held(&rq->lock); | |
927 | ||
928 | if (unlikely(head->next)) | |
929 | return; | |
930 | ||
931 | head->func = (void (*)(struct callback_head *))func; | |
932 | head->next = rq->balance_callback; | |
933 | rq->balance_callback = head; | |
934 | } | |
935 | ||
e3baac47 PZ |
936 | extern void sched_ttwu_pending(void); |
937 | ||
029632fb PZ |
938 | #define rcu_dereference_check_sched_domain(p) \ |
939 | rcu_dereference_check((p), \ | |
940 | lockdep_is_held(&sched_domains_mutex)) | |
941 | ||
942 | /* | |
943 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
944 | * See detach_destroy_domains: synchronize_sched for details. | |
945 | * | |
946 | * The domain tree of any CPU may only be accessed from within | |
947 | * preempt-disabled sections. | |
948 | */ | |
949 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
950 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
951 | __sd; __sd = __sd->parent) | |
029632fb | 952 | |
77e81365 SS |
953 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
954 | ||
518cd623 PZ |
955 | /** |
956 | * highest_flag_domain - Return highest sched_domain containing flag. | |
957 | * @cpu: The cpu whose highest level of sched domain is to | |
958 | * be returned. | |
959 | * @flag: The flag to check for the highest sched_domain | |
960 | * for the given cpu. | |
961 | * | |
962 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
963 | */ | |
964 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
965 | { | |
966 | struct sched_domain *sd, *hsd = NULL; | |
967 | ||
968 | for_each_domain(cpu, sd) { | |
969 | if (!(sd->flags & flag)) | |
970 | break; | |
971 | hsd = sd; | |
972 | } | |
973 | ||
974 | return hsd; | |
975 | } | |
976 | ||
fb13c7ee MG |
977 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
978 | { | |
979 | struct sched_domain *sd; | |
980 | ||
981 | for_each_domain(cpu, sd) { | |
982 | if (sd->flags & flag) | |
983 | break; | |
984 | } | |
985 | ||
986 | return sd; | |
987 | } | |
988 | ||
518cd623 | 989 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 990 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 991 | DECLARE_PER_CPU(int, sd_llc_id); |
0e369d75 | 992 | DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
fb13c7ee | 993 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 | 994 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); |
518cd623 | 995 | |
63b2ca30 | 996 | struct sched_group_capacity { |
5e6521ea LZ |
997 | atomic_t ref; |
998 | /* | |
172895e6 | 999 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
63b2ca30 | 1000 | * for a single CPU. |
5e6521ea | 1001 | */ |
bf475ce0 MR |
1002 | unsigned long capacity; |
1003 | unsigned long min_capacity; /* Min per-CPU capacity in group */ | |
5e6521ea | 1004 | unsigned long next_update; |
63b2ca30 | 1005 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea LZ |
1006 | |
1007 | unsigned long cpumask[0]; /* iteration mask */ | |
1008 | }; | |
1009 | ||
1010 | struct sched_group { | |
1011 | struct sched_group *next; /* Must be a circular list */ | |
1012 | atomic_t ref; | |
1013 | ||
1014 | unsigned int group_weight; | |
63b2ca30 | 1015 | struct sched_group_capacity *sgc; |
afe06efd | 1016 | int asym_prefer_cpu; /* cpu of highest priority in group */ |
5e6521ea LZ |
1017 | |
1018 | /* | |
1019 | * The CPUs this group covers. | |
1020 | * | |
1021 | * NOTE: this field is variable length. (Allocated dynamically | |
1022 | * by attaching extra space to the end of the structure, | |
1023 | * depending on how many CPUs the kernel has booted up with) | |
1024 | */ | |
1025 | unsigned long cpumask[0]; | |
1026 | }; | |
1027 | ||
1028 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) | |
1029 | { | |
1030 | return to_cpumask(sg->cpumask); | |
1031 | } | |
1032 | ||
1033 | /* | |
1034 | * cpumask masking which cpus in the group are allowed to iterate up the domain | |
1035 | * tree. | |
1036 | */ | |
1037 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) | |
1038 | { | |
63b2ca30 | 1039 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
1040 | } |
1041 | ||
1042 | /** | |
1043 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
1044 | * @group: The group whose first cpu is to be returned. | |
1045 | */ | |
1046 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
1047 | { | |
1048 | return cpumask_first(sched_group_cpus(group)); | |
1049 | } | |
1050 | ||
c1174876 PZ |
1051 | extern int group_balance_cpu(struct sched_group *sg); |
1052 | ||
3866e845 SRRH |
1053 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
1054 | void register_sched_domain_sysctl(void); | |
1055 | void unregister_sched_domain_sysctl(void); | |
1056 | #else | |
1057 | static inline void register_sched_domain_sysctl(void) | |
1058 | { | |
1059 | } | |
1060 | static inline void unregister_sched_domain_sysctl(void) | |
1061 | { | |
1062 | } | |
1063 | #endif | |
1064 | ||
e3baac47 PZ |
1065 | #else |
1066 | ||
1067 | static inline void sched_ttwu_pending(void) { } | |
1068 | ||
518cd623 | 1069 | #endif /* CONFIG_SMP */ |
029632fb | 1070 | |
391e43da | 1071 | #include "stats.h" |
1051408f | 1072 | #include "autogroup.h" |
029632fb PZ |
1073 | |
1074 | #ifdef CONFIG_CGROUP_SCHED | |
1075 | ||
1076 | /* | |
1077 | * Return the group to which this tasks belongs. | |
1078 | * | |
8af01f56 TH |
1079 | * We cannot use task_css() and friends because the cgroup subsystem |
1080 | * changes that value before the cgroup_subsys::attach() method is called, | |
1081 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
1082 | * |
1083 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
1084 | * core changes this before calling sched_move_task(). | |
1085 | * | |
1086 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
1087 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
1088 | */ |
1089 | static inline struct task_group *task_group(struct task_struct *p) | |
1090 | { | |
8323f26c | 1091 | return p->sched_task_group; |
029632fb PZ |
1092 | } |
1093 | ||
1094 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
1095 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
1096 | { | |
1097 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
1098 | struct task_group *tg = task_group(p); | |
1099 | #endif | |
1100 | ||
1101 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
ad936d86 | 1102 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
029632fb PZ |
1103 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
1104 | p->se.parent = tg->se[cpu]; | |
1105 | #endif | |
1106 | ||
1107 | #ifdef CONFIG_RT_GROUP_SCHED | |
1108 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
1109 | p->rt.parent = tg->rt_se[cpu]; | |
1110 | #endif | |
1111 | } | |
1112 | ||
1113 | #else /* CONFIG_CGROUP_SCHED */ | |
1114 | ||
1115 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
1116 | static inline struct task_group *task_group(struct task_struct *p) | |
1117 | { | |
1118 | return NULL; | |
1119 | } | |
1120 | ||
1121 | #endif /* CONFIG_CGROUP_SCHED */ | |
1122 | ||
1123 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1124 | { | |
1125 | set_task_rq(p, cpu); | |
1126 | #ifdef CONFIG_SMP | |
1127 | /* | |
1128 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1129 | * successfuly executed on another CPU. We must ensure that updates of | |
1130 | * per-task data have been completed by this moment. | |
1131 | */ | |
1132 | smp_wmb(); | |
c65eacbe AL |
1133 | #ifdef CONFIG_THREAD_INFO_IN_TASK |
1134 | p->cpu = cpu; | |
1135 | #else | |
029632fb | 1136 | task_thread_info(p)->cpu = cpu; |
c65eacbe | 1137 | #endif |
ac66f547 | 1138 | p->wake_cpu = cpu; |
029632fb PZ |
1139 | #endif |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
1144 | */ | |
1145 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 1146 | # include <linux/static_key.h> |
029632fb PZ |
1147 | # define const_debug __read_mostly |
1148 | #else | |
1149 | # define const_debug const | |
1150 | #endif | |
1151 | ||
1152 | extern const_debug unsigned int sysctl_sched_features; | |
1153 | ||
1154 | #define SCHED_FEAT(name, enabled) \ | |
1155 | __SCHED_FEAT_##name , | |
1156 | ||
1157 | enum { | |
391e43da | 1158 | #include "features.h" |
f8b6d1cc | 1159 | __SCHED_FEAT_NR, |
029632fb PZ |
1160 | }; |
1161 | ||
1162 | #undef SCHED_FEAT | |
1163 | ||
f8b6d1cc | 1164 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
f8b6d1cc | 1165 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1166 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1167 | { \ |
6e76ea8a | 1168 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1169 | } |
1170 | ||
1171 | #include "features.h" | |
1172 | ||
1173 | #undef SCHED_FEAT | |
1174 | ||
c5905afb | 1175 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
1176 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
1177 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 1178 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 1179 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 1180 | |
2a595721 | 1181 | extern struct static_key_false sched_numa_balancing; |
cb251765 | 1182 | extern struct static_key_false sched_schedstats; |
cbee9f88 | 1183 | |
029632fb PZ |
1184 | static inline u64 global_rt_period(void) |
1185 | { | |
1186 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1187 | } | |
1188 | ||
1189 | static inline u64 global_rt_runtime(void) | |
1190 | { | |
1191 | if (sysctl_sched_rt_runtime < 0) | |
1192 | return RUNTIME_INF; | |
1193 | ||
1194 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1195 | } | |
1196 | ||
029632fb PZ |
1197 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1198 | { | |
1199 | return rq->curr == p; | |
1200 | } | |
1201 | ||
1202 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1203 | { | |
1204 | #ifdef CONFIG_SMP | |
1205 | return p->on_cpu; | |
1206 | #else | |
1207 | return task_current(rq, p); | |
1208 | #endif | |
1209 | } | |
1210 | ||
da0c1e65 KT |
1211 | static inline int task_on_rq_queued(struct task_struct *p) |
1212 | { | |
1213 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1214 | } | |
029632fb | 1215 | |
cca26e80 KT |
1216 | static inline int task_on_rq_migrating(struct task_struct *p) |
1217 | { | |
1218 | return p->on_rq == TASK_ON_RQ_MIGRATING; | |
1219 | } | |
1220 | ||
029632fb PZ |
1221 | #ifndef prepare_arch_switch |
1222 | # define prepare_arch_switch(next) do { } while (0) | |
1223 | #endif | |
01f23e16 CM |
1224 | #ifndef finish_arch_post_lock_switch |
1225 | # define finish_arch_post_lock_switch() do { } while (0) | |
1226 | #endif | |
029632fb | 1227 | |
029632fb PZ |
1228 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
1229 | { | |
1230 | #ifdef CONFIG_SMP | |
1231 | /* | |
1232 | * We can optimise this out completely for !SMP, because the | |
1233 | * SMP rebalancing from interrupt is the only thing that cares | |
1234 | * here. | |
1235 | */ | |
1236 | next->on_cpu = 1; | |
1237 | #endif | |
1238 | } | |
1239 | ||
1240 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1241 | { | |
1242 | #ifdef CONFIG_SMP | |
1243 | /* | |
1244 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1245 | * We must ensure this doesn't happen until the switch is completely | |
1246 | * finished. | |
95913d97 | 1247 | * |
b75a2253 PZ |
1248 | * In particular, the load of prev->state in finish_task_switch() must |
1249 | * happen before this. | |
1250 | * | |
1f03e8d2 | 1251 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). |
029632fb | 1252 | */ |
95913d97 | 1253 | smp_store_release(&prev->on_cpu, 0); |
029632fb PZ |
1254 | #endif |
1255 | #ifdef CONFIG_DEBUG_SPINLOCK | |
1256 | /* this is a valid case when another task releases the spinlock */ | |
1257 | rq->lock.owner = current; | |
1258 | #endif | |
1259 | /* | |
1260 | * If we are tracking spinlock dependencies then we have to | |
1261 | * fix up the runqueue lock - which gets 'carried over' from | |
1262 | * prev into current: | |
1263 | */ | |
1264 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1265 | ||
1266 | raw_spin_unlock_irq(&rq->lock); | |
1267 | } | |
1268 | ||
b13095f0 LZ |
1269 | /* |
1270 | * wake flags | |
1271 | */ | |
1272 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1273 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1274 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1275 | ||
029632fb PZ |
1276 | /* |
1277 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1278 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1279 | * each task makes to its run queue's load is weighted according to its | |
1280 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1281 | * scaled version of the new time slice allocation that they receive on time | |
1282 | * slice expiry etc. | |
1283 | */ | |
1284 | ||
1285 | #define WEIGHT_IDLEPRIO 3 | |
1286 | #define WMULT_IDLEPRIO 1431655765 | |
1287 | ||
ed82b8a1 AK |
1288 | extern const int sched_prio_to_weight[40]; |
1289 | extern const u32 sched_prio_to_wmult[40]; | |
029632fb | 1290 | |
ff77e468 PZ |
1291 | /* |
1292 | * {de,en}queue flags: | |
1293 | * | |
1294 | * DEQUEUE_SLEEP - task is no longer runnable | |
1295 | * ENQUEUE_WAKEUP - task just became runnable | |
1296 | * | |
1297 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks | |
1298 | * are in a known state which allows modification. Such pairs | |
1299 | * should preserve as much state as possible. | |
1300 | * | |
1301 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location | |
1302 | * in the runqueue. | |
1303 | * | |
1304 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) | |
1305 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) | |
59efa0ba | 1306 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
ff77e468 PZ |
1307 | * |
1308 | */ | |
1309 | ||
1310 | #define DEQUEUE_SLEEP 0x01 | |
1311 | #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */ | |
1312 | #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */ | |
1313 | ||
1de64443 | 1314 | #define ENQUEUE_WAKEUP 0x01 |
ff77e468 PZ |
1315 | #define ENQUEUE_RESTORE 0x02 |
1316 | #define ENQUEUE_MOVE 0x04 | |
1317 | ||
1318 | #define ENQUEUE_HEAD 0x08 | |
1319 | #define ENQUEUE_REPLENISH 0x10 | |
c82ba9fa | 1320 | #ifdef CONFIG_SMP |
59efa0ba | 1321 | #define ENQUEUE_MIGRATED 0x20 |
c82ba9fa | 1322 | #else |
59efa0ba | 1323 | #define ENQUEUE_MIGRATED 0x00 |
c82ba9fa | 1324 | #endif |
c82ba9fa | 1325 | |
37e117c0 PZ |
1326 | #define RETRY_TASK ((void *)-1UL) |
1327 | ||
c82ba9fa LZ |
1328 | struct sched_class { |
1329 | const struct sched_class *next; | |
1330 | ||
1331 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1332 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1333 | void (*yield_task) (struct rq *rq); | |
1334 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1335 | ||
1336 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1337 | ||
606dba2e PZ |
1338 | /* |
1339 | * It is the responsibility of the pick_next_task() method that will | |
1340 | * return the next task to call put_prev_task() on the @prev task or | |
1341 | * something equivalent. | |
37e117c0 PZ |
1342 | * |
1343 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1344 | * tasks. | |
606dba2e PZ |
1345 | */ |
1346 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
e7904a28 | 1347 | struct task_struct *prev, |
d8ac8971 | 1348 | struct rq_flags *rf); |
c82ba9fa LZ |
1349 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1350 | ||
1351 | #ifdef CONFIG_SMP | |
ac66f547 | 1352 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
5a4fd036 | 1353 | void (*migrate_task_rq)(struct task_struct *p); |
c82ba9fa | 1354 | |
c82ba9fa LZ |
1355 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); |
1356 | ||
1357 | void (*set_cpus_allowed)(struct task_struct *p, | |
1358 | const struct cpumask *newmask); | |
1359 | ||
1360 | void (*rq_online)(struct rq *rq); | |
1361 | void (*rq_offline)(struct rq *rq); | |
1362 | #endif | |
1363 | ||
1364 | void (*set_curr_task) (struct rq *rq); | |
1365 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1366 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1367 | void (*task_dead) (struct task_struct *p); |
c82ba9fa | 1368 | |
67dfa1b7 KT |
1369 | /* |
1370 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1371 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1372 | * rq->lock. They are however serialized by p->pi_lock. | |
1373 | */ | |
c82ba9fa LZ |
1374 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); |
1375 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1376 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1377 | int oldprio); | |
1378 | ||
1379 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1380 | struct task_struct *task); | |
1381 | ||
6e998916 SG |
1382 | void (*update_curr) (struct rq *rq); |
1383 | ||
ea86cb4b VG |
1384 | #define TASK_SET_GROUP 0 |
1385 | #define TASK_MOVE_GROUP 1 | |
1386 | ||
c82ba9fa | 1387 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b | 1388 | void (*task_change_group) (struct task_struct *p, int type); |
c82ba9fa LZ |
1389 | #endif |
1390 | }; | |
029632fb | 1391 | |
3f1d2a31 PZ |
1392 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1393 | { | |
1394 | prev->sched_class->put_prev_task(rq, prev); | |
1395 | } | |
1396 | ||
b2bf6c31 PZ |
1397 | static inline void set_curr_task(struct rq *rq, struct task_struct *curr) |
1398 | { | |
1399 | curr->sched_class->set_curr_task(rq); | |
1400 | } | |
1401 | ||
029632fb PZ |
1402 | #define sched_class_highest (&stop_sched_class) |
1403 | #define for_each_class(class) \ | |
1404 | for (class = sched_class_highest; class; class = class->next) | |
1405 | ||
1406 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1407 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1408 | extern const struct sched_class rt_sched_class; |
1409 | extern const struct sched_class fair_sched_class; | |
1410 | extern const struct sched_class idle_sched_class; | |
1411 | ||
1412 | ||
1413 | #ifdef CONFIG_SMP | |
1414 | ||
63b2ca30 | 1415 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1416 | |
7caff66f | 1417 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1418 | |
c5b28038 PZ |
1419 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
1420 | ||
029632fb PZ |
1421 | #endif |
1422 | ||
442bf3aa DL |
1423 | #ifdef CONFIG_CPU_IDLE |
1424 | static inline void idle_set_state(struct rq *rq, | |
1425 | struct cpuidle_state *idle_state) | |
1426 | { | |
1427 | rq->idle_state = idle_state; | |
1428 | } | |
1429 | ||
1430 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1431 | { | |
9148a3a1 | 1432 | SCHED_WARN_ON(!rcu_read_lock_held()); |
442bf3aa DL |
1433 | return rq->idle_state; |
1434 | } | |
1435 | #else | |
1436 | static inline void idle_set_state(struct rq *rq, | |
1437 | struct cpuidle_state *idle_state) | |
1438 | { | |
1439 | } | |
1440 | ||
1441 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1442 | { | |
1443 | return NULL; | |
1444 | } | |
1445 | #endif | |
1446 | ||
029632fb PZ |
1447 | extern void sysrq_sched_debug_show(void); |
1448 | extern void sched_init_granularity(void); | |
1449 | extern void update_max_interval(void); | |
1baca4ce JL |
1450 | |
1451 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1452 | extern void init_sched_rt_class(void); |
1453 | extern void init_sched_fair_class(void); | |
1454 | ||
8875125e | 1455 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1456 | extern void resched_cpu(int cpu); |
1457 | ||
1458 | extern struct rt_bandwidth def_rt_bandwidth; | |
1459 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1460 | ||
332ac17e DF |
1461 | extern struct dl_bandwidth def_dl_bandwidth; |
1462 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 DF |
1463 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
1464 | ||
332ac17e DF |
1465 | unsigned long to_ratio(u64 period, u64 runtime); |
1466 | ||
540247fb | 1467 | extern void init_entity_runnable_average(struct sched_entity *se); |
2b8c41da | 1468 | extern void post_init_entity_util_avg(struct sched_entity *se); |
a75cdaa9 | 1469 | |
76d92ac3 FW |
1470 | #ifdef CONFIG_NO_HZ_FULL |
1471 | extern bool sched_can_stop_tick(struct rq *rq); | |
1472 | ||
1473 | /* | |
1474 | * Tick may be needed by tasks in the runqueue depending on their policy and | |
1475 | * requirements. If tick is needed, lets send the target an IPI to kick it out of | |
1476 | * nohz mode if necessary. | |
1477 | */ | |
1478 | static inline void sched_update_tick_dependency(struct rq *rq) | |
1479 | { | |
1480 | int cpu; | |
1481 | ||
1482 | if (!tick_nohz_full_enabled()) | |
1483 | return; | |
1484 | ||
1485 | cpu = cpu_of(rq); | |
1486 | ||
1487 | if (!tick_nohz_full_cpu(cpu)) | |
1488 | return; | |
1489 | ||
1490 | if (sched_can_stop_tick(rq)) | |
1491 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1492 | else | |
1493 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1494 | } | |
1495 | #else | |
1496 | static inline void sched_update_tick_dependency(struct rq *rq) { } | |
1497 | #endif | |
1498 | ||
72465447 | 1499 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1500 | { |
72465447 KT |
1501 | unsigned prev_nr = rq->nr_running; |
1502 | ||
1503 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1504 | |
72465447 | 1505 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1506 | #ifdef CONFIG_SMP |
1507 | if (!rq->rd->overload) | |
1508 | rq->rd->overload = true; | |
1509 | #endif | |
4486edd1 | 1510 | } |
76d92ac3 FW |
1511 | |
1512 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1513 | } |
1514 | ||
72465447 | 1515 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1516 | { |
72465447 | 1517 | rq->nr_running -= count; |
76d92ac3 FW |
1518 | /* Check if we still need preemption */ |
1519 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1520 | } |
1521 | ||
265f22a9 FW |
1522 | static inline void rq_last_tick_reset(struct rq *rq) |
1523 | { | |
1524 | #ifdef CONFIG_NO_HZ_FULL | |
1525 | rq->last_sched_tick = jiffies; | |
1526 | #endif | |
1527 | } | |
1528 | ||
029632fb PZ |
1529 | extern void update_rq_clock(struct rq *rq); |
1530 | ||
1531 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1532 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1533 | ||
1534 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1535 | ||
1536 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1537 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1538 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1539 | ||
1540 | static inline u64 sched_avg_period(void) | |
1541 | { | |
1542 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1543 | } | |
1544 | ||
029632fb PZ |
1545 | #ifdef CONFIG_SCHED_HRTICK |
1546 | ||
1547 | /* | |
1548 | * Use hrtick when: | |
1549 | * - enabled by features | |
1550 | * - hrtimer is actually high res | |
1551 | */ | |
1552 | static inline int hrtick_enabled(struct rq *rq) | |
1553 | { | |
1554 | if (!sched_feat(HRTICK)) | |
1555 | return 0; | |
1556 | if (!cpu_active(cpu_of(rq))) | |
1557 | return 0; | |
1558 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1559 | } | |
1560 | ||
1561 | void hrtick_start(struct rq *rq, u64 delay); | |
1562 | ||
b39e66ea MG |
1563 | #else |
1564 | ||
1565 | static inline int hrtick_enabled(struct rq *rq) | |
1566 | { | |
1567 | return 0; | |
1568 | } | |
1569 | ||
029632fb PZ |
1570 | #endif /* CONFIG_SCHED_HRTICK */ |
1571 | ||
1572 | #ifdef CONFIG_SMP | |
1573 | extern void sched_avg_update(struct rq *rq); | |
dfbca41f PZ |
1574 | |
1575 | #ifndef arch_scale_freq_capacity | |
1576 | static __always_inline | |
1577 | unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) | |
1578 | { | |
1579 | return SCHED_CAPACITY_SCALE; | |
1580 | } | |
1581 | #endif | |
b5b4860d | 1582 | |
8cd5601c MR |
1583 | #ifndef arch_scale_cpu_capacity |
1584 | static __always_inline | |
1585 | unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) | |
1586 | { | |
e3279a2e | 1587 | if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) |
8cd5601c MR |
1588 | return sd->smt_gain / sd->span_weight; |
1589 | ||
1590 | return SCHED_CAPACITY_SCALE; | |
1591 | } | |
1592 | #endif | |
1593 | ||
029632fb PZ |
1594 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) |
1595 | { | |
b5b4860d | 1596 | rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); |
029632fb PZ |
1597 | sched_avg_update(rq); |
1598 | } | |
1599 | #else | |
1600 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1601 | static inline void sched_avg_update(struct rq *rq) { } | |
1602 | #endif | |
1603 | ||
eb580751 | 1604 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 | 1605 | __acquires(rq->lock); |
eb580751 | 1606 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3960c8c0 | 1607 | __acquires(p->pi_lock) |
3e71a462 | 1608 | __acquires(rq->lock); |
3960c8c0 | 1609 | |
eb580751 | 1610 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) |
3960c8c0 PZ |
1611 | __releases(rq->lock) |
1612 | { | |
d8ac8971 | 1613 | rq_unpin_lock(rq, rf); |
3960c8c0 PZ |
1614 | raw_spin_unlock(&rq->lock); |
1615 | } | |
1616 | ||
1617 | static inline void | |
eb580751 | 1618 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) |
3960c8c0 PZ |
1619 | __releases(rq->lock) |
1620 | __releases(p->pi_lock) | |
1621 | { | |
d8ac8971 | 1622 | rq_unpin_lock(rq, rf); |
3960c8c0 | 1623 | raw_spin_unlock(&rq->lock); |
eb580751 | 1624 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3960c8c0 PZ |
1625 | } |
1626 | ||
029632fb PZ |
1627 | #ifdef CONFIG_SMP |
1628 | #ifdef CONFIG_PREEMPT | |
1629 | ||
1630 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1631 | ||
1632 | /* | |
1633 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1634 | * way at the expense of forcing extra atomic operations in all | |
1635 | * invocations. This assures that the double_lock is acquired using the | |
1636 | * same underlying policy as the spinlock_t on this architecture, which | |
1637 | * reduces latency compared to the unfair variant below. However, it | |
1638 | * also adds more overhead and therefore may reduce throughput. | |
1639 | */ | |
1640 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1641 | __releases(this_rq->lock) | |
1642 | __acquires(busiest->lock) | |
1643 | __acquires(this_rq->lock) | |
1644 | { | |
1645 | raw_spin_unlock(&this_rq->lock); | |
1646 | double_rq_lock(this_rq, busiest); | |
1647 | ||
1648 | return 1; | |
1649 | } | |
1650 | ||
1651 | #else | |
1652 | /* | |
1653 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1654 | * latency by eliminating extra atomic operations when the locks are | |
1655 | * already in proper order on entry. This favors lower cpu-ids and will | |
1656 | * grant the double lock to lower cpus over higher ids under contention, | |
1657 | * regardless of entry order into the function. | |
1658 | */ | |
1659 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1660 | __releases(this_rq->lock) | |
1661 | __acquires(busiest->lock) | |
1662 | __acquires(this_rq->lock) | |
1663 | { | |
1664 | int ret = 0; | |
1665 | ||
1666 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1667 | if (busiest < this_rq) { | |
1668 | raw_spin_unlock(&this_rq->lock); | |
1669 | raw_spin_lock(&busiest->lock); | |
1670 | raw_spin_lock_nested(&this_rq->lock, | |
1671 | SINGLE_DEPTH_NESTING); | |
1672 | ret = 1; | |
1673 | } else | |
1674 | raw_spin_lock_nested(&busiest->lock, | |
1675 | SINGLE_DEPTH_NESTING); | |
1676 | } | |
1677 | return ret; | |
1678 | } | |
1679 | ||
1680 | #endif /* CONFIG_PREEMPT */ | |
1681 | ||
1682 | /* | |
1683 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1684 | */ | |
1685 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1686 | { | |
1687 | if (unlikely(!irqs_disabled())) { | |
1688 | /* printk() doesn't work good under rq->lock */ | |
1689 | raw_spin_unlock(&this_rq->lock); | |
1690 | BUG_ON(1); | |
1691 | } | |
1692 | ||
1693 | return _double_lock_balance(this_rq, busiest); | |
1694 | } | |
1695 | ||
1696 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1697 | __releases(busiest->lock) | |
1698 | { | |
1699 | raw_spin_unlock(&busiest->lock); | |
1700 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1701 | } | |
1702 | ||
74602315 PZ |
1703 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1704 | { | |
1705 | if (l1 > l2) | |
1706 | swap(l1, l2); | |
1707 | ||
1708 | spin_lock(l1); | |
1709 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1710 | } | |
1711 | ||
60e69eed MG |
1712 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1713 | { | |
1714 | if (l1 > l2) | |
1715 | swap(l1, l2); | |
1716 | ||
1717 | spin_lock_irq(l1); | |
1718 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1719 | } | |
1720 | ||
74602315 PZ |
1721 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1722 | { | |
1723 | if (l1 > l2) | |
1724 | swap(l1, l2); | |
1725 | ||
1726 | raw_spin_lock(l1); | |
1727 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1728 | } | |
1729 | ||
029632fb PZ |
1730 | /* |
1731 | * double_rq_lock - safely lock two runqueues | |
1732 | * | |
1733 | * Note this does not disable interrupts like task_rq_lock, | |
1734 | * you need to do so manually before calling. | |
1735 | */ | |
1736 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1737 | __acquires(rq1->lock) | |
1738 | __acquires(rq2->lock) | |
1739 | { | |
1740 | BUG_ON(!irqs_disabled()); | |
1741 | if (rq1 == rq2) { | |
1742 | raw_spin_lock(&rq1->lock); | |
1743 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1744 | } else { | |
1745 | if (rq1 < rq2) { | |
1746 | raw_spin_lock(&rq1->lock); | |
1747 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1748 | } else { | |
1749 | raw_spin_lock(&rq2->lock); | |
1750 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1751 | } | |
1752 | } | |
1753 | } | |
1754 | ||
1755 | /* | |
1756 | * double_rq_unlock - safely unlock two runqueues | |
1757 | * | |
1758 | * Note this does not restore interrupts like task_rq_unlock, | |
1759 | * you need to do so manually after calling. | |
1760 | */ | |
1761 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1762 | __releases(rq1->lock) | |
1763 | __releases(rq2->lock) | |
1764 | { | |
1765 | raw_spin_unlock(&rq1->lock); | |
1766 | if (rq1 != rq2) | |
1767 | raw_spin_unlock(&rq2->lock); | |
1768 | else | |
1769 | __release(rq2->lock); | |
1770 | } | |
1771 | ||
f2cb1360 IM |
1772 | extern void set_rq_online (struct rq *rq); |
1773 | extern void set_rq_offline(struct rq *rq); | |
1774 | extern bool sched_smp_initialized; | |
1775 | ||
029632fb PZ |
1776 | #else /* CONFIG_SMP */ |
1777 | ||
1778 | /* | |
1779 | * double_rq_lock - safely lock two runqueues | |
1780 | * | |
1781 | * Note this does not disable interrupts like task_rq_lock, | |
1782 | * you need to do so manually before calling. | |
1783 | */ | |
1784 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1785 | __acquires(rq1->lock) | |
1786 | __acquires(rq2->lock) | |
1787 | { | |
1788 | BUG_ON(!irqs_disabled()); | |
1789 | BUG_ON(rq1 != rq2); | |
1790 | raw_spin_lock(&rq1->lock); | |
1791 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1792 | } | |
1793 | ||
1794 | /* | |
1795 | * double_rq_unlock - safely unlock two runqueues | |
1796 | * | |
1797 | * Note this does not restore interrupts like task_rq_unlock, | |
1798 | * you need to do so manually after calling. | |
1799 | */ | |
1800 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1801 | __releases(rq1->lock) | |
1802 | __releases(rq2->lock) | |
1803 | { | |
1804 | BUG_ON(rq1 != rq2); | |
1805 | raw_spin_unlock(&rq1->lock); | |
1806 | __release(rq2->lock); | |
1807 | } | |
1808 | ||
1809 | #endif | |
1810 | ||
1811 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1812 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
1813 | |
1814 | #ifdef CONFIG_SCHED_DEBUG | |
029632fb PZ |
1815 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
1816 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 1817 | extern void print_dl_stats(struct seq_file *m, int cpu); |
6b55c965 SD |
1818 | extern void |
1819 | print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); | |
397f2378 SD |
1820 | |
1821 | #ifdef CONFIG_NUMA_BALANCING | |
1822 | extern void | |
1823 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
1824 | extern void | |
1825 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
1826 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
1827 | #endif /* CONFIG_NUMA_BALANCING */ | |
1828 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
1829 | |
1830 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
1831 | extern void init_rt_rq(struct rt_rq *rt_rq); |
1832 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 1833 | |
1ee14e6c BS |
1834 | extern void cfs_bandwidth_usage_inc(void); |
1835 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1836 | |
3451d024 | 1837 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1838 | enum rq_nohz_flag_bits { |
1839 | NOHZ_TICK_STOPPED, | |
1840 | NOHZ_BALANCE_KICK, | |
1841 | }; | |
1842 | ||
1843 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
20a5c8cc TG |
1844 | |
1845 | extern void nohz_balance_exit_idle(unsigned int cpu); | |
1846 | #else | |
1847 | static inline void nohz_balance_exit_idle(unsigned int cpu) { } | |
1c792db7 | 1848 | #endif |
73fbec60 FW |
1849 | |
1850 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
19d23dbf | 1851 | struct irqtime { |
a499a5a1 | 1852 | u64 tick_delta; |
19d23dbf FW |
1853 | u64 irq_start_time; |
1854 | struct u64_stats_sync sync; | |
1855 | }; | |
73fbec60 | 1856 | |
19d23dbf | 1857 | DECLARE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 FW |
1858 | |
1859 | static inline u64 irq_time_read(int cpu) | |
1860 | { | |
19d23dbf | 1861 | struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); |
a499a5a1 | 1862 | u64 *cpustat = kcpustat_cpu(cpu).cpustat; |
19d23dbf FW |
1863 | unsigned int seq; |
1864 | u64 total; | |
73fbec60 FW |
1865 | |
1866 | do { | |
19d23dbf | 1867 | seq = __u64_stats_fetch_begin(&irqtime->sync); |
a499a5a1 | 1868 | total = cpustat[CPUTIME_SOFTIRQ] + cpustat[CPUTIME_IRQ]; |
19d23dbf | 1869 | } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); |
73fbec60 | 1870 | |
19d23dbf | 1871 | return total; |
73fbec60 | 1872 | } |
73fbec60 | 1873 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
adaf9fcd RW |
1874 | |
1875 | #ifdef CONFIG_CPU_FREQ | |
1876 | DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); | |
1877 | ||
1878 | /** | |
1879 | * cpufreq_update_util - Take a note about CPU utilization changes. | |
12bde33d | 1880 | * @rq: Runqueue to carry out the update for. |
58919e83 | 1881 | * @flags: Update reason flags. |
adaf9fcd | 1882 | * |
58919e83 RW |
1883 | * This function is called by the scheduler on the CPU whose utilization is |
1884 | * being updated. | |
adaf9fcd RW |
1885 | * |
1886 | * It can only be called from RCU-sched read-side critical sections. | |
adaf9fcd RW |
1887 | * |
1888 | * The way cpufreq is currently arranged requires it to evaluate the CPU | |
1889 | * performance state (frequency/voltage) on a regular basis to prevent it from | |
1890 | * being stuck in a completely inadequate performance level for too long. | |
1891 | * That is not guaranteed to happen if the updates are only triggered from CFS, | |
1892 | * though, because they may not be coming in if RT or deadline tasks are active | |
1893 | * all the time (or there are RT and DL tasks only). | |
1894 | * | |
1895 | * As a workaround for that issue, this function is called by the RT and DL | |
1896 | * sched classes to trigger extra cpufreq updates to prevent it from stalling, | |
1897 | * but that really is a band-aid. Going forward it should be replaced with | |
1898 | * solutions targeted more specifically at RT and DL tasks. | |
1899 | */ | |
12bde33d | 1900 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) |
adaf9fcd | 1901 | { |
58919e83 RW |
1902 | struct update_util_data *data; |
1903 | ||
1904 | data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data)); | |
1905 | if (data) | |
12bde33d RW |
1906 | data->func(data, rq_clock(rq), flags); |
1907 | } | |
1908 | ||
1909 | static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) | |
1910 | { | |
1911 | if (cpu_of(rq) == smp_processor_id()) | |
1912 | cpufreq_update_util(rq, flags); | |
adaf9fcd RW |
1913 | } |
1914 | #else | |
12bde33d RW |
1915 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} |
1916 | static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) {} | |
adaf9fcd | 1917 | #endif /* CONFIG_CPU_FREQ */ |
be53f58f | 1918 | |
9bdcb44e RW |
1919 | #ifdef arch_scale_freq_capacity |
1920 | #ifndef arch_scale_freq_invariant | |
1921 | #define arch_scale_freq_invariant() (true) | |
1922 | #endif | |
1923 | #else /* arch_scale_freq_capacity */ | |
1924 | #define arch_scale_freq_invariant() (false) | |
1925 | #endif |