Commit | Line | Data |
---|---|---|
029632fb PZ |
1 | |
2 | #include <linux/sched.h> | |
cf4aebc2 | 3 | #include <linux/sched/sysctl.h> |
8bd75c77 | 4 | #include <linux/sched/rt.h> |
aab03e05 | 5 | #include <linux/sched/deadline.h> |
029632fb PZ |
6 | #include <linux/mutex.h> |
7 | #include <linux/spinlock.h> | |
8 | #include <linux/stop_machine.h> | |
b6366f04 | 9 | #include <linux/irq_work.h> |
9f3660c2 | 10 | #include <linux/tick.h> |
f809ca9a | 11 | #include <linux/slab.h> |
029632fb | 12 | |
391e43da | 13 | #include "cpupri.h" |
6bfd6d72 | 14 | #include "cpudeadline.h" |
60fed789 | 15 | #include "cpuacct.h" |
029632fb | 16 | |
45ceebf7 | 17 | struct rq; |
442bf3aa | 18 | struct cpuidle_state; |
45ceebf7 | 19 | |
da0c1e65 KT |
20 | /* task_struct::on_rq states: */ |
21 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 22 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 23 | |
029632fb PZ |
24 | extern __read_mostly int scheduler_running; |
25 | ||
45ceebf7 PG |
26 | extern unsigned long calc_load_update; |
27 | extern atomic_long_t calc_load_tasks; | |
28 | ||
3289bdb4 | 29 | extern void calc_global_load_tick(struct rq *this_rq); |
45ceebf7 | 30 | extern long calc_load_fold_active(struct rq *this_rq); |
3289bdb4 PZ |
31 | |
32 | #ifdef CONFIG_SMP | |
45ceebf7 | 33 | extern void update_cpu_load_active(struct rq *this_rq); |
3289bdb4 PZ |
34 | #else |
35 | static inline void update_cpu_load_active(struct rq *this_rq) { } | |
36 | #endif | |
45ceebf7 | 37 | |
029632fb PZ |
38 | /* |
39 | * Helpers for converting nanosecond timing to jiffy resolution | |
40 | */ | |
41 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
42 | ||
cc1f4b1f LZ |
43 | /* |
44 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
45 | * The extra resolution improves shares distribution and load balancing of | |
46 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
47 | * hierarchies, especially on larger systems. This is not a user-visible change | |
48 | * and does not change the user-interface for setting shares/weights. | |
49 | * | |
50 | * We increase resolution only if we have enough bits to allow this increased | |
51 | * resolution (i.e. BITS_PER_LONG > 32). The costs for increasing resolution | |
52 | * when BITS_PER_LONG <= 32 are pretty high and the returns do not justify the | |
53 | * increased costs. | |
54 | */ | |
55 | #if 0 /* BITS_PER_LONG > 32 -- currently broken: it increases power usage under light load */ | |
56 | # define SCHED_LOAD_RESOLUTION 10 | |
57 | # define scale_load(w) ((w) << SCHED_LOAD_RESOLUTION) | |
58 | # define scale_load_down(w) ((w) >> SCHED_LOAD_RESOLUTION) | |
59 | #else | |
60 | # define SCHED_LOAD_RESOLUTION 0 | |
61 | # define scale_load(w) (w) | |
62 | # define scale_load_down(w) (w) | |
63 | #endif | |
64 | ||
65 | #define SCHED_LOAD_SHIFT (10 + SCHED_LOAD_RESOLUTION) | |
66 | #define SCHED_LOAD_SCALE (1L << SCHED_LOAD_SHIFT) | |
67 | ||
029632fb PZ |
68 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
69 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
70 | ||
332ac17e DF |
71 | /* |
72 | * Single value that decides SCHED_DEADLINE internal math precision. | |
73 | * 10 -> just above 1us | |
74 | * 9 -> just above 0.5us | |
75 | */ | |
76 | #define DL_SCALE (10) | |
77 | ||
029632fb PZ |
78 | /* |
79 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 80 | */ |
029632fb PZ |
81 | |
82 | /* | |
83 | * single value that denotes runtime == period, ie unlimited time. | |
84 | */ | |
85 | #define RUNTIME_INF ((u64)~0ULL) | |
86 | ||
20f9cd2a HA |
87 | static inline int idle_policy(int policy) |
88 | { | |
89 | return policy == SCHED_IDLE; | |
90 | } | |
d50dde5a DF |
91 | static inline int fair_policy(int policy) |
92 | { | |
93 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
94 | } | |
95 | ||
029632fb PZ |
96 | static inline int rt_policy(int policy) |
97 | { | |
d50dde5a | 98 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
99 | } |
100 | ||
aab03e05 DF |
101 | static inline int dl_policy(int policy) |
102 | { | |
103 | return policy == SCHED_DEADLINE; | |
104 | } | |
20f9cd2a HA |
105 | static inline bool valid_policy(int policy) |
106 | { | |
107 | return idle_policy(policy) || fair_policy(policy) || | |
108 | rt_policy(policy) || dl_policy(policy); | |
109 | } | |
aab03e05 | 110 | |
029632fb PZ |
111 | static inline int task_has_rt_policy(struct task_struct *p) |
112 | { | |
113 | return rt_policy(p->policy); | |
114 | } | |
115 | ||
aab03e05 DF |
116 | static inline int task_has_dl_policy(struct task_struct *p) |
117 | { | |
118 | return dl_policy(p->policy); | |
119 | } | |
120 | ||
2d3d891d DF |
121 | /* |
122 | * Tells if entity @a should preempt entity @b. | |
123 | */ | |
332ac17e DF |
124 | static inline bool |
125 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
126 | { |
127 | return dl_time_before(a->deadline, b->deadline); | |
128 | } | |
129 | ||
029632fb PZ |
130 | /* |
131 | * This is the priority-queue data structure of the RT scheduling class: | |
132 | */ | |
133 | struct rt_prio_array { | |
134 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
135 | struct list_head queue[MAX_RT_PRIO]; | |
136 | }; | |
137 | ||
138 | struct rt_bandwidth { | |
139 | /* nests inside the rq lock: */ | |
140 | raw_spinlock_t rt_runtime_lock; | |
141 | ktime_t rt_period; | |
142 | u64 rt_runtime; | |
143 | struct hrtimer rt_period_timer; | |
4cfafd30 | 144 | unsigned int rt_period_active; |
029632fb | 145 | }; |
a5e7be3b JL |
146 | |
147 | void __dl_clear_params(struct task_struct *p); | |
148 | ||
332ac17e DF |
149 | /* |
150 | * To keep the bandwidth of -deadline tasks and groups under control | |
151 | * we need some place where: | |
152 | * - store the maximum -deadline bandwidth of the system (the group); | |
153 | * - cache the fraction of that bandwidth that is currently allocated. | |
154 | * | |
155 | * This is all done in the data structure below. It is similar to the | |
156 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
157 | * that, since here we are only interested in admission control, we | |
158 | * do not decrease any runtime while the group "executes", neither we | |
159 | * need a timer to replenish it. | |
160 | * | |
161 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
162 | * meaning that: | |
163 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
164 | * - dl_total_bw array contains, in the i-eth element, the currently | |
165 | * allocated bandwidth on the i-eth CPU. | |
166 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
167 | * consume bandwidth on the CPU they're running on. | |
168 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
169 | * that will be shown the next time the proc or cgroup controls will | |
170 | * be red. It on its turn can be changed by writing on its own | |
171 | * control. | |
172 | */ | |
173 | struct dl_bandwidth { | |
174 | raw_spinlock_t dl_runtime_lock; | |
175 | u64 dl_runtime; | |
176 | u64 dl_period; | |
177 | }; | |
178 | ||
179 | static inline int dl_bandwidth_enabled(void) | |
180 | { | |
1724813d | 181 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
182 | } |
183 | ||
184 | extern struct dl_bw *dl_bw_of(int i); | |
185 | ||
186 | struct dl_bw { | |
187 | raw_spinlock_t lock; | |
188 | u64 bw, total_bw; | |
189 | }; | |
190 | ||
7f51412a JL |
191 | static inline |
192 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
193 | { | |
194 | dl_b->total_bw -= tsk_bw; | |
195 | } | |
196 | ||
197 | static inline | |
198 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
199 | { | |
200 | dl_b->total_bw += tsk_bw; | |
201 | } | |
202 | ||
203 | static inline | |
204 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
205 | { | |
206 | return dl_b->bw != -1 && | |
207 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
208 | } | |
209 | ||
029632fb PZ |
210 | extern struct mutex sched_domains_mutex; |
211 | ||
212 | #ifdef CONFIG_CGROUP_SCHED | |
213 | ||
214 | #include <linux/cgroup.h> | |
215 | ||
216 | struct cfs_rq; | |
217 | struct rt_rq; | |
218 | ||
35cf4e50 | 219 | extern struct list_head task_groups; |
029632fb PZ |
220 | |
221 | struct cfs_bandwidth { | |
222 | #ifdef CONFIG_CFS_BANDWIDTH | |
223 | raw_spinlock_t lock; | |
224 | ktime_t period; | |
225 | u64 quota, runtime; | |
9c58c79a | 226 | s64 hierarchical_quota; |
029632fb PZ |
227 | u64 runtime_expires; |
228 | ||
4cfafd30 | 229 | int idle, period_active; |
029632fb PZ |
230 | struct hrtimer period_timer, slack_timer; |
231 | struct list_head throttled_cfs_rq; | |
232 | ||
233 | /* statistics */ | |
234 | int nr_periods, nr_throttled; | |
235 | u64 throttled_time; | |
236 | #endif | |
237 | }; | |
238 | ||
239 | /* task group related information */ | |
240 | struct task_group { | |
241 | struct cgroup_subsys_state css; | |
242 | ||
243 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
244 | /* schedulable entities of this group on each cpu */ | |
245 | struct sched_entity **se; | |
246 | /* runqueue "owned" by this group on each cpu */ | |
247 | struct cfs_rq **cfs_rq; | |
248 | unsigned long shares; | |
249 | ||
fa6bddeb | 250 | #ifdef CONFIG_SMP |
bf5b986e | 251 | atomic_long_t load_avg; |
029632fb | 252 | #endif |
fa6bddeb | 253 | #endif |
029632fb PZ |
254 | |
255 | #ifdef CONFIG_RT_GROUP_SCHED | |
256 | struct sched_rt_entity **rt_se; | |
257 | struct rt_rq **rt_rq; | |
258 | ||
259 | struct rt_bandwidth rt_bandwidth; | |
260 | #endif | |
261 | ||
262 | struct rcu_head rcu; | |
263 | struct list_head list; | |
264 | ||
265 | struct task_group *parent; | |
266 | struct list_head siblings; | |
267 | struct list_head children; | |
268 | ||
269 | #ifdef CONFIG_SCHED_AUTOGROUP | |
270 | struct autogroup *autogroup; | |
271 | #endif | |
272 | ||
273 | struct cfs_bandwidth cfs_bandwidth; | |
274 | }; | |
275 | ||
276 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
277 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
278 | ||
279 | /* | |
280 | * A weight of 0 or 1 can cause arithmetics problems. | |
281 | * A weight of a cfs_rq is the sum of weights of which entities | |
282 | * are queued on this cfs_rq, so a weight of a entity should not be | |
283 | * too large, so as the shares value of a task group. | |
284 | * (The default weight is 1024 - so there's no practical | |
285 | * limitation from this.) | |
286 | */ | |
287 | #define MIN_SHARES (1UL << 1) | |
288 | #define MAX_SHARES (1UL << 18) | |
289 | #endif | |
290 | ||
029632fb PZ |
291 | typedef int (*tg_visitor)(struct task_group *, void *); |
292 | ||
293 | extern int walk_tg_tree_from(struct task_group *from, | |
294 | tg_visitor down, tg_visitor up, void *data); | |
295 | ||
296 | /* | |
297 | * Iterate the full tree, calling @down when first entering a node and @up when | |
298 | * leaving it for the final time. | |
299 | * | |
300 | * Caller must hold rcu_lock or sufficient equivalent. | |
301 | */ | |
302 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
303 | { | |
304 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
305 | } | |
306 | ||
307 | extern int tg_nop(struct task_group *tg, void *data); | |
308 | ||
309 | extern void free_fair_sched_group(struct task_group *tg); | |
310 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
311 | extern void unregister_fair_sched_group(struct task_group *tg, int cpu); | |
312 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, | |
313 | struct sched_entity *se, int cpu, | |
314 | struct sched_entity *parent); | |
315 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
316 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
317 | ||
318 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 319 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
320 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
321 | ||
322 | extern void free_rt_sched_group(struct task_group *tg); | |
323 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
324 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
325 | struct sched_rt_entity *rt_se, int cpu, | |
326 | struct sched_rt_entity *parent); | |
327 | ||
25cc7da7 LZ |
328 | extern struct task_group *sched_create_group(struct task_group *parent); |
329 | extern void sched_online_group(struct task_group *tg, | |
330 | struct task_group *parent); | |
331 | extern void sched_destroy_group(struct task_group *tg); | |
332 | extern void sched_offline_group(struct task_group *tg); | |
333 | ||
334 | extern void sched_move_task(struct task_struct *tsk); | |
335 | ||
336 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
337 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
338 | #endif | |
339 | ||
029632fb PZ |
340 | #else /* CONFIG_CGROUP_SCHED */ |
341 | ||
342 | struct cfs_bandwidth { }; | |
343 | ||
344 | #endif /* CONFIG_CGROUP_SCHED */ | |
345 | ||
346 | /* CFS-related fields in a runqueue */ | |
347 | struct cfs_rq { | |
348 | struct load_weight load; | |
c82513e5 | 349 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
350 | |
351 | u64 exec_clock; | |
352 | u64 min_vruntime; | |
353 | #ifndef CONFIG_64BIT | |
354 | u64 min_vruntime_copy; | |
355 | #endif | |
356 | ||
357 | struct rb_root tasks_timeline; | |
358 | struct rb_node *rb_leftmost; | |
359 | ||
029632fb PZ |
360 | /* |
361 | * 'curr' points to currently running entity on this cfs_rq. | |
362 | * It is set to NULL otherwise (i.e when none are currently running). | |
363 | */ | |
364 | struct sched_entity *curr, *next, *last, *skip; | |
365 | ||
366 | #ifdef CONFIG_SCHED_DEBUG | |
367 | unsigned int nr_spread_over; | |
368 | #endif | |
369 | ||
2dac754e PT |
370 | #ifdef CONFIG_SMP |
371 | /* | |
9d89c257 | 372 | * CFS load tracking |
2dac754e | 373 | */ |
9d89c257 | 374 | struct sched_avg avg; |
13962234 YD |
375 | u64 runnable_load_sum; |
376 | unsigned long runnable_load_avg; | |
c566e8e9 | 377 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9d89c257 YD |
378 | unsigned long tg_load_avg_contrib; |
379 | #endif | |
380 | atomic_long_t removed_load_avg, removed_util_avg; | |
381 | #ifndef CONFIG_64BIT | |
382 | u64 load_last_update_time_copy; | |
383 | #endif | |
82958366 | 384 | |
9d89c257 | 385 | #ifdef CONFIG_FAIR_GROUP_SCHED |
82958366 PT |
386 | /* |
387 | * h_load = weight * f(tg) | |
388 | * | |
389 | * Where f(tg) is the recursive weight fraction assigned to | |
390 | * this group. | |
391 | */ | |
392 | unsigned long h_load; | |
68520796 VD |
393 | u64 last_h_load_update; |
394 | struct sched_entity *h_load_next; | |
395 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
396 | #endif /* CONFIG_SMP */ |
397 | ||
029632fb PZ |
398 | #ifdef CONFIG_FAIR_GROUP_SCHED |
399 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
400 | ||
401 | /* | |
402 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
403 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
404 | * (like users, containers etc.) | |
405 | * | |
406 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
407 | * list is used during load balance. | |
408 | */ | |
409 | int on_list; | |
410 | struct list_head leaf_cfs_rq_list; | |
411 | struct task_group *tg; /* group that "owns" this runqueue */ | |
412 | ||
029632fb PZ |
413 | #ifdef CONFIG_CFS_BANDWIDTH |
414 | int runtime_enabled; | |
415 | u64 runtime_expires; | |
416 | s64 runtime_remaining; | |
417 | ||
f1b17280 PT |
418 | u64 throttled_clock, throttled_clock_task; |
419 | u64 throttled_clock_task_time; | |
029632fb PZ |
420 | int throttled, throttle_count; |
421 | struct list_head throttled_list; | |
422 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
423 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
424 | }; | |
425 | ||
426 | static inline int rt_bandwidth_enabled(void) | |
427 | { | |
428 | return sysctl_sched_rt_runtime >= 0; | |
429 | } | |
430 | ||
b6366f04 SR |
431 | /* RT IPI pull logic requires IRQ_WORK */ |
432 | #ifdef CONFIG_IRQ_WORK | |
433 | # define HAVE_RT_PUSH_IPI | |
434 | #endif | |
435 | ||
029632fb PZ |
436 | /* Real-Time classes' related field in a runqueue: */ |
437 | struct rt_rq { | |
438 | struct rt_prio_array active; | |
c82513e5 | 439 | unsigned int rt_nr_running; |
029632fb PZ |
440 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
441 | struct { | |
442 | int curr; /* highest queued rt task prio */ | |
443 | #ifdef CONFIG_SMP | |
444 | int next; /* next highest */ | |
445 | #endif | |
446 | } highest_prio; | |
447 | #endif | |
448 | #ifdef CONFIG_SMP | |
449 | unsigned long rt_nr_migratory; | |
450 | unsigned long rt_nr_total; | |
451 | int overloaded; | |
452 | struct plist_head pushable_tasks; | |
b6366f04 SR |
453 | #ifdef HAVE_RT_PUSH_IPI |
454 | int push_flags; | |
455 | int push_cpu; | |
456 | struct irq_work push_work; | |
457 | raw_spinlock_t push_lock; | |
029632fb | 458 | #endif |
b6366f04 | 459 | #endif /* CONFIG_SMP */ |
f4ebcbc0 KT |
460 | int rt_queued; |
461 | ||
029632fb PZ |
462 | int rt_throttled; |
463 | u64 rt_time; | |
464 | u64 rt_runtime; | |
465 | /* Nests inside the rq lock: */ | |
466 | raw_spinlock_t rt_runtime_lock; | |
467 | ||
468 | #ifdef CONFIG_RT_GROUP_SCHED | |
469 | unsigned long rt_nr_boosted; | |
470 | ||
471 | struct rq *rq; | |
029632fb PZ |
472 | struct task_group *tg; |
473 | #endif | |
474 | }; | |
475 | ||
aab03e05 DF |
476 | /* Deadline class' related fields in a runqueue */ |
477 | struct dl_rq { | |
478 | /* runqueue is an rbtree, ordered by deadline */ | |
479 | struct rb_root rb_root; | |
480 | struct rb_node *rb_leftmost; | |
481 | ||
482 | unsigned long dl_nr_running; | |
1baca4ce JL |
483 | |
484 | #ifdef CONFIG_SMP | |
485 | /* | |
486 | * Deadline values of the currently executing and the | |
487 | * earliest ready task on this rq. Caching these facilitates | |
488 | * the decision wether or not a ready but not running task | |
489 | * should migrate somewhere else. | |
490 | */ | |
491 | struct { | |
492 | u64 curr; | |
493 | u64 next; | |
494 | } earliest_dl; | |
495 | ||
496 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
497 | int overloaded; |
498 | ||
499 | /* | |
500 | * Tasks on this rq that can be pushed away. They are kept in | |
501 | * an rb-tree, ordered by tasks' deadlines, with caching | |
502 | * of the leftmost (earliest deadline) element. | |
503 | */ | |
504 | struct rb_root pushable_dl_tasks_root; | |
505 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
506 | #else |
507 | struct dl_bw dl_bw; | |
1baca4ce | 508 | #endif |
aab03e05 DF |
509 | }; |
510 | ||
029632fb PZ |
511 | #ifdef CONFIG_SMP |
512 | ||
513 | /* | |
514 | * We add the notion of a root-domain which will be used to define per-domain | |
515 | * variables. Each exclusive cpuset essentially defines an island domain by | |
516 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
517 | * exclusive cpuset is created, we also create and attach a new root-domain | |
518 | * object. | |
519 | * | |
520 | */ | |
521 | struct root_domain { | |
522 | atomic_t refcount; | |
523 | atomic_t rto_count; | |
524 | struct rcu_head rcu; | |
525 | cpumask_var_t span; | |
526 | cpumask_var_t online; | |
527 | ||
4486edd1 TC |
528 | /* Indicate more than one runnable task for any CPU */ |
529 | bool overload; | |
530 | ||
1baca4ce JL |
531 | /* |
532 | * The bit corresponding to a CPU gets set here if such CPU has more | |
533 | * than one runnable -deadline task (as it is below for RT tasks). | |
534 | */ | |
535 | cpumask_var_t dlo_mask; | |
536 | atomic_t dlo_count; | |
332ac17e | 537 | struct dl_bw dl_bw; |
6bfd6d72 | 538 | struct cpudl cpudl; |
1baca4ce | 539 | |
029632fb PZ |
540 | /* |
541 | * The "RT overload" flag: it gets set if a CPU has more than | |
542 | * one runnable RT task. | |
543 | */ | |
544 | cpumask_var_t rto_mask; | |
545 | struct cpupri cpupri; | |
546 | }; | |
547 | ||
548 | extern struct root_domain def_root_domain; | |
549 | ||
550 | #endif /* CONFIG_SMP */ | |
551 | ||
552 | /* | |
553 | * This is the main, per-CPU runqueue data structure. | |
554 | * | |
555 | * Locking rule: those places that want to lock multiple runqueues | |
556 | * (such as the load balancing or the thread migration code), lock | |
557 | * acquire operations must be ordered by ascending &runqueue. | |
558 | */ | |
559 | struct rq { | |
560 | /* runqueue lock: */ | |
561 | raw_spinlock_t lock; | |
562 | ||
563 | /* | |
564 | * nr_running and cpu_load should be in the same cacheline because | |
565 | * remote CPUs use both these fields when doing load calculation. | |
566 | */ | |
c82513e5 | 567 | unsigned int nr_running; |
0ec8aa00 PZ |
568 | #ifdef CONFIG_NUMA_BALANCING |
569 | unsigned int nr_numa_running; | |
570 | unsigned int nr_preferred_running; | |
571 | #endif | |
029632fb PZ |
572 | #define CPU_LOAD_IDX_MAX 5 |
573 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
574 | unsigned long last_load_update_tick; | |
3451d024 | 575 | #ifdef CONFIG_NO_HZ_COMMON |
029632fb | 576 | u64 nohz_stamp; |
1c792db7 | 577 | unsigned long nohz_flags; |
265f22a9 FW |
578 | #endif |
579 | #ifdef CONFIG_NO_HZ_FULL | |
580 | unsigned long last_sched_tick; | |
029632fb | 581 | #endif |
029632fb PZ |
582 | /* capture load from *all* tasks on this cpu: */ |
583 | struct load_weight load; | |
584 | unsigned long nr_load_updates; | |
585 | u64 nr_switches; | |
586 | ||
587 | struct cfs_rq cfs; | |
588 | struct rt_rq rt; | |
aab03e05 | 589 | struct dl_rq dl; |
029632fb PZ |
590 | |
591 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
592 | /* list of leaf cfs_rq on this cpu: */ | |
593 | struct list_head leaf_cfs_rq_list; | |
a35b6466 PZ |
594 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
595 | ||
029632fb PZ |
596 | /* |
597 | * This is part of a global counter where only the total sum | |
598 | * over all CPUs matters. A task can increase this counter on | |
599 | * one CPU and if it got migrated afterwards it may decrease | |
600 | * it on another CPU. Always updated under the runqueue lock: | |
601 | */ | |
602 | unsigned long nr_uninterruptible; | |
603 | ||
604 | struct task_struct *curr, *idle, *stop; | |
605 | unsigned long next_balance; | |
606 | struct mm_struct *prev_mm; | |
607 | ||
9edfbfed | 608 | unsigned int clock_skip_update; |
029632fb PZ |
609 | u64 clock; |
610 | u64 clock_task; | |
611 | ||
612 | atomic_t nr_iowait; | |
613 | ||
614 | #ifdef CONFIG_SMP | |
615 | struct root_domain *rd; | |
616 | struct sched_domain *sd; | |
617 | ||
ced549fa | 618 | unsigned long cpu_capacity; |
ca6d75e6 | 619 | unsigned long cpu_capacity_orig; |
029632fb | 620 | |
e3fca9e7 PZ |
621 | struct callback_head *balance_callback; |
622 | ||
029632fb PZ |
623 | unsigned char idle_balance; |
624 | /* For active balancing */ | |
029632fb PZ |
625 | int active_balance; |
626 | int push_cpu; | |
627 | struct cpu_stop_work active_balance_work; | |
628 | /* cpu of this runqueue: */ | |
629 | int cpu; | |
630 | int online; | |
631 | ||
367456c7 PZ |
632 | struct list_head cfs_tasks; |
633 | ||
029632fb PZ |
634 | u64 rt_avg; |
635 | u64 age_stamp; | |
636 | u64 idle_stamp; | |
637 | u64 avg_idle; | |
9bd721c5 JL |
638 | |
639 | /* This is used to determine avg_idle's max value */ | |
640 | u64 max_idle_balance_cost; | |
029632fb PZ |
641 | #endif |
642 | ||
643 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
644 | u64 prev_irq_time; | |
645 | #endif | |
646 | #ifdef CONFIG_PARAVIRT | |
647 | u64 prev_steal_time; | |
648 | #endif | |
649 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
650 | u64 prev_steal_time_rq; | |
651 | #endif | |
652 | ||
653 | /* calc_load related fields */ | |
654 | unsigned long calc_load_update; | |
655 | long calc_load_active; | |
656 | ||
657 | #ifdef CONFIG_SCHED_HRTICK | |
658 | #ifdef CONFIG_SMP | |
659 | int hrtick_csd_pending; | |
660 | struct call_single_data hrtick_csd; | |
661 | #endif | |
662 | struct hrtimer hrtick_timer; | |
663 | #endif | |
664 | ||
665 | #ifdef CONFIG_SCHEDSTATS | |
666 | /* latency stats */ | |
667 | struct sched_info rq_sched_info; | |
668 | unsigned long long rq_cpu_time; | |
669 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
670 | ||
671 | /* sys_sched_yield() stats */ | |
672 | unsigned int yld_count; | |
673 | ||
674 | /* schedule() stats */ | |
029632fb PZ |
675 | unsigned int sched_count; |
676 | unsigned int sched_goidle; | |
677 | ||
678 | /* try_to_wake_up() stats */ | |
679 | unsigned int ttwu_count; | |
680 | unsigned int ttwu_local; | |
681 | #endif | |
682 | ||
683 | #ifdef CONFIG_SMP | |
684 | struct llist_head wake_list; | |
685 | #endif | |
442bf3aa DL |
686 | |
687 | #ifdef CONFIG_CPU_IDLE | |
688 | /* Must be inspected within a rcu lock section */ | |
689 | struct cpuidle_state *idle_state; | |
690 | #endif | |
029632fb PZ |
691 | }; |
692 | ||
693 | static inline int cpu_of(struct rq *rq) | |
694 | { | |
695 | #ifdef CONFIG_SMP | |
696 | return rq->cpu; | |
697 | #else | |
698 | return 0; | |
699 | #endif | |
700 | } | |
701 | ||
8b06c55b | 702 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 703 | |
518cd623 | 704 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 705 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
706 | #define task_rq(p) cpu_rq(task_cpu(p)) |
707 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 708 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 709 | |
cebde6d6 PZ |
710 | static inline u64 __rq_clock_broken(struct rq *rq) |
711 | { | |
316c1608 | 712 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
713 | } |
714 | ||
78becc27 FW |
715 | static inline u64 rq_clock(struct rq *rq) |
716 | { | |
cebde6d6 | 717 | lockdep_assert_held(&rq->lock); |
78becc27 FW |
718 | return rq->clock; |
719 | } | |
720 | ||
721 | static inline u64 rq_clock_task(struct rq *rq) | |
722 | { | |
cebde6d6 | 723 | lockdep_assert_held(&rq->lock); |
78becc27 FW |
724 | return rq->clock_task; |
725 | } | |
726 | ||
9edfbfed PZ |
727 | #define RQCF_REQ_SKIP 0x01 |
728 | #define RQCF_ACT_SKIP 0x02 | |
729 | ||
730 | static inline void rq_clock_skip_update(struct rq *rq, bool skip) | |
731 | { | |
732 | lockdep_assert_held(&rq->lock); | |
733 | if (skip) | |
734 | rq->clock_skip_update |= RQCF_REQ_SKIP; | |
735 | else | |
736 | rq->clock_skip_update &= ~RQCF_REQ_SKIP; | |
737 | } | |
738 | ||
9942f79b | 739 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
740 | enum numa_topology_type { |
741 | NUMA_DIRECT, | |
742 | NUMA_GLUELESS_MESH, | |
743 | NUMA_BACKPLANE, | |
744 | }; | |
745 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
746 | extern int sched_max_numa_distance; |
747 | extern bool find_numa_distance(int distance); | |
748 | #endif | |
749 | ||
f809ca9a | 750 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
751 | /* The regions in numa_faults array from task_struct */ |
752 | enum numa_faults_stats { | |
753 | NUMA_MEM = 0, | |
754 | NUMA_CPU, | |
755 | NUMA_MEMBUF, | |
756 | NUMA_CPUBUF | |
757 | }; | |
0ec8aa00 | 758 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 759 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 760 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
761 | #endif /* CONFIG_NUMA_BALANCING */ |
762 | ||
518cd623 PZ |
763 | #ifdef CONFIG_SMP |
764 | ||
e3fca9e7 PZ |
765 | static inline void |
766 | queue_balance_callback(struct rq *rq, | |
767 | struct callback_head *head, | |
768 | void (*func)(struct rq *rq)) | |
769 | { | |
770 | lockdep_assert_held(&rq->lock); | |
771 | ||
772 | if (unlikely(head->next)) | |
773 | return; | |
774 | ||
775 | head->func = (void (*)(struct callback_head *))func; | |
776 | head->next = rq->balance_callback; | |
777 | rq->balance_callback = head; | |
778 | } | |
779 | ||
e3baac47 PZ |
780 | extern void sched_ttwu_pending(void); |
781 | ||
029632fb PZ |
782 | #define rcu_dereference_check_sched_domain(p) \ |
783 | rcu_dereference_check((p), \ | |
784 | lockdep_is_held(&sched_domains_mutex)) | |
785 | ||
786 | /* | |
787 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
788 | * See detach_destroy_domains: synchronize_sched for details. | |
789 | * | |
790 | * The domain tree of any CPU may only be accessed from within | |
791 | * preempt-disabled sections. | |
792 | */ | |
793 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
794 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
795 | __sd; __sd = __sd->parent) | |
029632fb | 796 | |
77e81365 SS |
797 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
798 | ||
518cd623 PZ |
799 | /** |
800 | * highest_flag_domain - Return highest sched_domain containing flag. | |
801 | * @cpu: The cpu whose highest level of sched domain is to | |
802 | * be returned. | |
803 | * @flag: The flag to check for the highest sched_domain | |
804 | * for the given cpu. | |
805 | * | |
806 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
807 | */ | |
808 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
809 | { | |
810 | struct sched_domain *sd, *hsd = NULL; | |
811 | ||
812 | for_each_domain(cpu, sd) { | |
813 | if (!(sd->flags & flag)) | |
814 | break; | |
815 | hsd = sd; | |
816 | } | |
817 | ||
818 | return hsd; | |
819 | } | |
820 | ||
fb13c7ee MG |
821 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
822 | { | |
823 | struct sched_domain *sd; | |
824 | ||
825 | for_each_domain(cpu, sd) { | |
826 | if (sd->flags & flag) | |
827 | break; | |
828 | } | |
829 | ||
830 | return sd; | |
831 | } | |
832 | ||
518cd623 | 833 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 834 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 835 | DECLARE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 836 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
837 | DECLARE_PER_CPU(struct sched_domain *, sd_busy); |
838 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 | 839 | |
63b2ca30 | 840 | struct sched_group_capacity { |
5e6521ea LZ |
841 | atomic_t ref; |
842 | /* | |
63b2ca30 NP |
843 | * CPU capacity of this group, SCHED_LOAD_SCALE being max capacity |
844 | * for a single CPU. | |
5e6521ea | 845 | */ |
dc7ff76e | 846 | unsigned int capacity; |
5e6521ea | 847 | unsigned long next_update; |
63b2ca30 | 848 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea LZ |
849 | /* |
850 | * Number of busy cpus in this group. | |
851 | */ | |
852 | atomic_t nr_busy_cpus; | |
853 | ||
854 | unsigned long cpumask[0]; /* iteration mask */ | |
855 | }; | |
856 | ||
857 | struct sched_group { | |
858 | struct sched_group *next; /* Must be a circular list */ | |
859 | atomic_t ref; | |
860 | ||
861 | unsigned int group_weight; | |
63b2ca30 | 862 | struct sched_group_capacity *sgc; |
5e6521ea LZ |
863 | |
864 | /* | |
865 | * The CPUs this group covers. | |
866 | * | |
867 | * NOTE: this field is variable length. (Allocated dynamically | |
868 | * by attaching extra space to the end of the structure, | |
869 | * depending on how many CPUs the kernel has booted up with) | |
870 | */ | |
871 | unsigned long cpumask[0]; | |
872 | }; | |
873 | ||
874 | static inline struct cpumask *sched_group_cpus(struct sched_group *sg) | |
875 | { | |
876 | return to_cpumask(sg->cpumask); | |
877 | } | |
878 | ||
879 | /* | |
880 | * cpumask masking which cpus in the group are allowed to iterate up the domain | |
881 | * tree. | |
882 | */ | |
883 | static inline struct cpumask *sched_group_mask(struct sched_group *sg) | |
884 | { | |
63b2ca30 | 885 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
886 | } |
887 | ||
888 | /** | |
889 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
890 | * @group: The group whose first cpu is to be returned. | |
891 | */ | |
892 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
893 | { | |
894 | return cpumask_first(sched_group_cpus(group)); | |
895 | } | |
896 | ||
c1174876 PZ |
897 | extern int group_balance_cpu(struct sched_group *sg); |
898 | ||
e3baac47 PZ |
899 | #else |
900 | ||
901 | static inline void sched_ttwu_pending(void) { } | |
902 | ||
518cd623 | 903 | #endif /* CONFIG_SMP */ |
029632fb | 904 | |
391e43da PZ |
905 | #include "stats.h" |
906 | #include "auto_group.h" | |
029632fb PZ |
907 | |
908 | #ifdef CONFIG_CGROUP_SCHED | |
909 | ||
910 | /* | |
911 | * Return the group to which this tasks belongs. | |
912 | * | |
8af01f56 TH |
913 | * We cannot use task_css() and friends because the cgroup subsystem |
914 | * changes that value before the cgroup_subsys::attach() method is called, | |
915 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
916 | * |
917 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
918 | * core changes this before calling sched_move_task(). | |
919 | * | |
920 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
921 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
922 | */ |
923 | static inline struct task_group *task_group(struct task_struct *p) | |
924 | { | |
8323f26c | 925 | return p->sched_task_group; |
029632fb PZ |
926 | } |
927 | ||
928 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
929 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
930 | { | |
931 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
932 | struct task_group *tg = task_group(p); | |
933 | #endif | |
934 | ||
935 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
936 | p->se.cfs_rq = tg->cfs_rq[cpu]; | |
937 | p->se.parent = tg->se[cpu]; | |
938 | #endif | |
939 | ||
940 | #ifdef CONFIG_RT_GROUP_SCHED | |
941 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
942 | p->rt.parent = tg->rt_se[cpu]; | |
943 | #endif | |
944 | } | |
945 | ||
946 | #else /* CONFIG_CGROUP_SCHED */ | |
947 | ||
948 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
949 | static inline struct task_group *task_group(struct task_struct *p) | |
950 | { | |
951 | return NULL; | |
952 | } | |
953 | ||
954 | #endif /* CONFIG_CGROUP_SCHED */ | |
955 | ||
956 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
957 | { | |
958 | set_task_rq(p, cpu); | |
959 | #ifdef CONFIG_SMP | |
960 | /* | |
961 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
962 | * successfuly executed on another CPU. We must ensure that updates of | |
963 | * per-task data have been completed by this moment. | |
964 | */ | |
965 | smp_wmb(); | |
966 | task_thread_info(p)->cpu = cpu; | |
ac66f547 | 967 | p->wake_cpu = cpu; |
029632fb PZ |
968 | #endif |
969 | } | |
970 | ||
971 | /* | |
972 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
973 | */ | |
974 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 975 | # include <linux/static_key.h> |
029632fb PZ |
976 | # define const_debug __read_mostly |
977 | #else | |
978 | # define const_debug const | |
979 | #endif | |
980 | ||
981 | extern const_debug unsigned int sysctl_sched_features; | |
982 | ||
983 | #define SCHED_FEAT(name, enabled) \ | |
984 | __SCHED_FEAT_##name , | |
985 | ||
986 | enum { | |
391e43da | 987 | #include "features.h" |
f8b6d1cc | 988 | __SCHED_FEAT_NR, |
029632fb PZ |
989 | }; |
990 | ||
991 | #undef SCHED_FEAT | |
992 | ||
f8b6d1cc | 993 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
f8b6d1cc | 994 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 995 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 996 | { \ |
6e76ea8a | 997 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
998 | } |
999 | ||
1000 | #include "features.h" | |
1001 | ||
1002 | #undef SCHED_FEAT | |
1003 | ||
c5905afb | 1004 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
1005 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
1006 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 1007 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 1008 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 1009 | |
2a595721 | 1010 | extern struct static_key_false sched_numa_balancing; |
cbee9f88 | 1011 | |
029632fb PZ |
1012 | static inline u64 global_rt_period(void) |
1013 | { | |
1014 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1015 | } | |
1016 | ||
1017 | static inline u64 global_rt_runtime(void) | |
1018 | { | |
1019 | if (sysctl_sched_rt_runtime < 0) | |
1020 | return RUNTIME_INF; | |
1021 | ||
1022 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1023 | } | |
1024 | ||
029632fb PZ |
1025 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1026 | { | |
1027 | return rq->curr == p; | |
1028 | } | |
1029 | ||
1030 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1031 | { | |
1032 | #ifdef CONFIG_SMP | |
1033 | return p->on_cpu; | |
1034 | #else | |
1035 | return task_current(rq, p); | |
1036 | #endif | |
1037 | } | |
1038 | ||
da0c1e65 KT |
1039 | static inline int task_on_rq_queued(struct task_struct *p) |
1040 | { | |
1041 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1042 | } | |
029632fb | 1043 | |
cca26e80 KT |
1044 | static inline int task_on_rq_migrating(struct task_struct *p) |
1045 | { | |
1046 | return p->on_rq == TASK_ON_RQ_MIGRATING; | |
1047 | } | |
1048 | ||
029632fb PZ |
1049 | #ifndef prepare_arch_switch |
1050 | # define prepare_arch_switch(next) do { } while (0) | |
1051 | #endif | |
01f23e16 CM |
1052 | #ifndef finish_arch_post_lock_switch |
1053 | # define finish_arch_post_lock_switch() do { } while (0) | |
1054 | #endif | |
029632fb | 1055 | |
029632fb PZ |
1056 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
1057 | { | |
1058 | #ifdef CONFIG_SMP | |
1059 | /* | |
1060 | * We can optimise this out completely for !SMP, because the | |
1061 | * SMP rebalancing from interrupt is the only thing that cares | |
1062 | * here. | |
1063 | */ | |
1064 | next->on_cpu = 1; | |
1065 | #endif | |
1066 | } | |
1067 | ||
1068 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1069 | { | |
1070 | #ifdef CONFIG_SMP | |
1071 | /* | |
1072 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1073 | * We must ensure this doesn't happen until the switch is completely | |
1074 | * finished. | |
95913d97 PZ |
1075 | * |
1076 | * Pairs with the control dependency and rmb in try_to_wake_up(). | |
029632fb | 1077 | */ |
95913d97 | 1078 | smp_store_release(&prev->on_cpu, 0); |
029632fb PZ |
1079 | #endif |
1080 | #ifdef CONFIG_DEBUG_SPINLOCK | |
1081 | /* this is a valid case when another task releases the spinlock */ | |
1082 | rq->lock.owner = current; | |
1083 | #endif | |
1084 | /* | |
1085 | * If we are tracking spinlock dependencies then we have to | |
1086 | * fix up the runqueue lock - which gets 'carried over' from | |
1087 | * prev into current: | |
1088 | */ | |
1089 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1090 | ||
1091 | raw_spin_unlock_irq(&rq->lock); | |
1092 | } | |
1093 | ||
b13095f0 LZ |
1094 | /* |
1095 | * wake flags | |
1096 | */ | |
1097 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1098 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1099 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1100 | ||
029632fb PZ |
1101 | /* |
1102 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1103 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1104 | * each task makes to its run queue's load is weighted according to its | |
1105 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1106 | * scaled version of the new time slice allocation that they receive on time | |
1107 | * slice expiry etc. | |
1108 | */ | |
1109 | ||
1110 | #define WEIGHT_IDLEPRIO 3 | |
1111 | #define WMULT_IDLEPRIO 1431655765 | |
1112 | ||
1113 | /* | |
1114 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1115 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1116 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1117 | * that remained on nice 0. | |
1118 | * | |
1119 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1120 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
1121 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
1122 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1123 | * the relative distance between them is ~25%.) | |
1124 | */ | |
1125 | static const int prio_to_weight[40] = { | |
1126 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
1127 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1128 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1129 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1130 | /* 0 */ 1024, 820, 655, 526, 423, | |
1131 | /* 5 */ 335, 272, 215, 172, 137, | |
1132 | /* 10 */ 110, 87, 70, 56, 45, | |
1133 | /* 15 */ 36, 29, 23, 18, 15, | |
1134 | }; | |
1135 | ||
1136 | /* | |
1137 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1138 | * | |
1139 | * In cases where the weight does not change often, we can use the | |
1140 | * precalculated inverse to speed up arithmetics by turning divisions | |
1141 | * into multiplications: | |
1142 | */ | |
1143 | static const u32 prio_to_wmult[40] = { | |
1144 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
1145 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1146 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1147 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1148 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1149 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1150 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1151 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
1152 | }; | |
1153 | ||
1de64443 PZ |
1154 | #define ENQUEUE_WAKEUP 0x01 |
1155 | #define ENQUEUE_HEAD 0x02 | |
c82ba9fa | 1156 | #ifdef CONFIG_SMP |
1de64443 | 1157 | #define ENQUEUE_WAKING 0x04 /* sched_class::task_waking was called */ |
c82ba9fa | 1158 | #else |
1de64443 | 1159 | #define ENQUEUE_WAKING 0x00 |
c82ba9fa | 1160 | #endif |
1de64443 PZ |
1161 | #define ENQUEUE_REPLENISH 0x08 |
1162 | #define ENQUEUE_RESTORE 0x10 | |
c82ba9fa | 1163 | |
1de64443 PZ |
1164 | #define DEQUEUE_SLEEP 0x01 |
1165 | #define DEQUEUE_SAVE 0x02 | |
c82ba9fa | 1166 | |
37e117c0 PZ |
1167 | #define RETRY_TASK ((void *)-1UL) |
1168 | ||
c82ba9fa LZ |
1169 | struct sched_class { |
1170 | const struct sched_class *next; | |
1171 | ||
1172 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1173 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1174 | void (*yield_task) (struct rq *rq); | |
1175 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1176 | ||
1177 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1178 | ||
606dba2e PZ |
1179 | /* |
1180 | * It is the responsibility of the pick_next_task() method that will | |
1181 | * return the next task to call put_prev_task() on the @prev task or | |
1182 | * something equivalent. | |
37e117c0 PZ |
1183 | * |
1184 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1185 | * tasks. | |
606dba2e PZ |
1186 | */ |
1187 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
1188 | struct task_struct *prev); | |
c82ba9fa LZ |
1189 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1190 | ||
1191 | #ifdef CONFIG_SMP | |
ac66f547 | 1192 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
5a4fd036 | 1193 | void (*migrate_task_rq)(struct task_struct *p); |
c82ba9fa | 1194 | |
c82ba9fa LZ |
1195 | void (*task_waking) (struct task_struct *task); |
1196 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); | |
1197 | ||
1198 | void (*set_cpus_allowed)(struct task_struct *p, | |
1199 | const struct cpumask *newmask); | |
1200 | ||
1201 | void (*rq_online)(struct rq *rq); | |
1202 | void (*rq_offline)(struct rq *rq); | |
1203 | #endif | |
1204 | ||
1205 | void (*set_curr_task) (struct rq *rq); | |
1206 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1207 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1208 | void (*task_dead) (struct task_struct *p); |
c82ba9fa | 1209 | |
67dfa1b7 KT |
1210 | /* |
1211 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1212 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1213 | * rq->lock. They are however serialized by p->pi_lock. | |
1214 | */ | |
c82ba9fa LZ |
1215 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); |
1216 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1217 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1218 | int oldprio); | |
1219 | ||
1220 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1221 | struct task_struct *task); | |
1222 | ||
6e998916 SG |
1223 | void (*update_curr) (struct rq *rq); |
1224 | ||
c82ba9fa | 1225 | #ifdef CONFIG_FAIR_GROUP_SCHED |
bc54da21 | 1226 | void (*task_move_group) (struct task_struct *p); |
c82ba9fa LZ |
1227 | #endif |
1228 | }; | |
029632fb | 1229 | |
3f1d2a31 PZ |
1230 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1231 | { | |
1232 | prev->sched_class->put_prev_task(rq, prev); | |
1233 | } | |
1234 | ||
029632fb PZ |
1235 | #define sched_class_highest (&stop_sched_class) |
1236 | #define for_each_class(class) \ | |
1237 | for (class = sched_class_highest; class; class = class->next) | |
1238 | ||
1239 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1240 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1241 | extern const struct sched_class rt_sched_class; |
1242 | extern const struct sched_class fair_sched_class; | |
1243 | extern const struct sched_class idle_sched_class; | |
1244 | ||
1245 | ||
1246 | #ifdef CONFIG_SMP | |
1247 | ||
63b2ca30 | 1248 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1249 | |
7caff66f | 1250 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1251 | |
642dbc39 VG |
1252 | extern void idle_enter_fair(struct rq *this_rq); |
1253 | extern void idle_exit_fair(struct rq *this_rq); | |
642dbc39 | 1254 | |
c5b28038 PZ |
1255 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
1256 | ||
dc877341 PZ |
1257 | #else |
1258 | ||
1259 | static inline void idle_enter_fair(struct rq *rq) { } | |
1260 | static inline void idle_exit_fair(struct rq *rq) { } | |
1261 | ||
029632fb PZ |
1262 | #endif |
1263 | ||
442bf3aa DL |
1264 | #ifdef CONFIG_CPU_IDLE |
1265 | static inline void idle_set_state(struct rq *rq, | |
1266 | struct cpuidle_state *idle_state) | |
1267 | { | |
1268 | rq->idle_state = idle_state; | |
1269 | } | |
1270 | ||
1271 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1272 | { | |
1273 | WARN_ON(!rcu_read_lock_held()); | |
1274 | return rq->idle_state; | |
1275 | } | |
1276 | #else | |
1277 | static inline void idle_set_state(struct rq *rq, | |
1278 | struct cpuidle_state *idle_state) | |
1279 | { | |
1280 | } | |
1281 | ||
1282 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1283 | { | |
1284 | return NULL; | |
1285 | } | |
1286 | #endif | |
1287 | ||
029632fb PZ |
1288 | extern void sysrq_sched_debug_show(void); |
1289 | extern void sched_init_granularity(void); | |
1290 | extern void update_max_interval(void); | |
1baca4ce JL |
1291 | |
1292 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1293 | extern void init_sched_rt_class(void); |
1294 | extern void init_sched_fair_class(void); | |
1295 | ||
8875125e | 1296 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1297 | extern void resched_cpu(int cpu); |
1298 | ||
1299 | extern struct rt_bandwidth def_rt_bandwidth; | |
1300 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1301 | ||
332ac17e DF |
1302 | extern struct dl_bandwidth def_dl_bandwidth; |
1303 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 DF |
1304 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
1305 | ||
332ac17e DF |
1306 | unsigned long to_ratio(u64 period, u64 runtime); |
1307 | ||
540247fb | 1308 | extern void init_entity_runnable_average(struct sched_entity *se); |
a75cdaa9 | 1309 | |
72465447 | 1310 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1311 | { |
72465447 KT |
1312 | unsigned prev_nr = rq->nr_running; |
1313 | ||
1314 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1315 | |
72465447 | 1316 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1317 | #ifdef CONFIG_SMP |
1318 | if (!rq->rd->overload) | |
1319 | rq->rd->overload = true; | |
1320 | #endif | |
1321 | ||
1322 | #ifdef CONFIG_NO_HZ_FULL | |
9f3660c2 | 1323 | if (tick_nohz_full_cpu(rq->cpu)) { |
3882ec64 FW |
1324 | /* |
1325 | * Tick is needed if more than one task runs on a CPU. | |
1326 | * Send the target an IPI to kick it out of nohz mode. | |
1327 | * | |
1328 | * We assume that IPI implies full memory barrier and the | |
1329 | * new value of rq->nr_running is visible on reception | |
1330 | * from the target. | |
1331 | */ | |
fd2ac4f4 | 1332 | tick_nohz_full_kick_cpu(rq->cpu); |
9f3660c2 | 1333 | } |
9f3660c2 | 1334 | #endif |
4486edd1 | 1335 | } |
029632fb PZ |
1336 | } |
1337 | ||
72465447 | 1338 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1339 | { |
72465447 | 1340 | rq->nr_running -= count; |
029632fb PZ |
1341 | } |
1342 | ||
265f22a9 FW |
1343 | static inline void rq_last_tick_reset(struct rq *rq) |
1344 | { | |
1345 | #ifdef CONFIG_NO_HZ_FULL | |
1346 | rq->last_sched_tick = jiffies; | |
1347 | #endif | |
1348 | } | |
1349 | ||
029632fb PZ |
1350 | extern void update_rq_clock(struct rq *rq); |
1351 | ||
1352 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1353 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1354 | ||
1355 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1356 | ||
1357 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1358 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1359 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1360 | ||
1361 | static inline u64 sched_avg_period(void) | |
1362 | { | |
1363 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1364 | } | |
1365 | ||
029632fb PZ |
1366 | #ifdef CONFIG_SCHED_HRTICK |
1367 | ||
1368 | /* | |
1369 | * Use hrtick when: | |
1370 | * - enabled by features | |
1371 | * - hrtimer is actually high res | |
1372 | */ | |
1373 | static inline int hrtick_enabled(struct rq *rq) | |
1374 | { | |
1375 | if (!sched_feat(HRTICK)) | |
1376 | return 0; | |
1377 | if (!cpu_active(cpu_of(rq))) | |
1378 | return 0; | |
1379 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1380 | } | |
1381 | ||
1382 | void hrtick_start(struct rq *rq, u64 delay); | |
1383 | ||
b39e66ea MG |
1384 | #else |
1385 | ||
1386 | static inline int hrtick_enabled(struct rq *rq) | |
1387 | { | |
1388 | return 0; | |
1389 | } | |
1390 | ||
029632fb PZ |
1391 | #endif /* CONFIG_SCHED_HRTICK */ |
1392 | ||
1393 | #ifdef CONFIG_SMP | |
1394 | extern void sched_avg_update(struct rq *rq); | |
dfbca41f PZ |
1395 | |
1396 | #ifndef arch_scale_freq_capacity | |
1397 | static __always_inline | |
1398 | unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) | |
1399 | { | |
1400 | return SCHED_CAPACITY_SCALE; | |
1401 | } | |
1402 | #endif | |
b5b4860d | 1403 | |
8cd5601c MR |
1404 | #ifndef arch_scale_cpu_capacity |
1405 | static __always_inline | |
1406 | unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) | |
1407 | { | |
e3279a2e | 1408 | if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) |
8cd5601c MR |
1409 | return sd->smt_gain / sd->span_weight; |
1410 | ||
1411 | return SCHED_CAPACITY_SCALE; | |
1412 | } | |
1413 | #endif | |
1414 | ||
029632fb PZ |
1415 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) |
1416 | { | |
b5b4860d | 1417 | rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); |
029632fb PZ |
1418 | sched_avg_update(rq); |
1419 | } | |
1420 | #else | |
1421 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1422 | static inline void sched_avg_update(struct rq *rq) { } | |
1423 | #endif | |
1424 | ||
3960c8c0 PZ |
1425 | /* |
1426 | * __task_rq_lock - lock the rq @p resides on. | |
1427 | */ | |
1428 | static inline struct rq *__task_rq_lock(struct task_struct *p) | |
1429 | __acquires(rq->lock) | |
1430 | { | |
1431 | struct rq *rq; | |
1432 | ||
1433 | lockdep_assert_held(&p->pi_lock); | |
1434 | ||
1435 | for (;;) { | |
1436 | rq = task_rq(p); | |
1437 | raw_spin_lock(&rq->lock); | |
cbce1a68 PZ |
1438 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
1439 | lockdep_pin_lock(&rq->lock); | |
3960c8c0 | 1440 | return rq; |
cbce1a68 | 1441 | } |
3960c8c0 PZ |
1442 | raw_spin_unlock(&rq->lock); |
1443 | ||
1444 | while (unlikely(task_on_rq_migrating(p))) | |
1445 | cpu_relax(); | |
1446 | } | |
1447 | } | |
1448 | ||
1449 | /* | |
1450 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
1451 | */ | |
1452 | static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) | |
1453 | __acquires(p->pi_lock) | |
1454 | __acquires(rq->lock) | |
1455 | { | |
1456 | struct rq *rq; | |
1457 | ||
1458 | for (;;) { | |
1459 | raw_spin_lock_irqsave(&p->pi_lock, *flags); | |
1460 | rq = task_rq(p); | |
1461 | raw_spin_lock(&rq->lock); | |
1462 | /* | |
1463 | * move_queued_task() task_rq_lock() | |
1464 | * | |
1465 | * ACQUIRE (rq->lock) | |
1466 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
1467 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
1468 | * [S] ->cpu = new_cpu [L] task_rq() | |
1469 | * [L] ->on_rq | |
1470 | * RELEASE (rq->lock) | |
1471 | * | |
1472 | * If we observe the old cpu in task_rq_lock, the acquire of | |
1473 | * the old rq->lock will fully serialize against the stores. | |
1474 | * | |
1475 | * If we observe the new cpu in task_rq_lock, the acquire will | |
1476 | * pair with the WMB to ensure we must then also see migrating. | |
1477 | */ | |
cbce1a68 PZ |
1478 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
1479 | lockdep_pin_lock(&rq->lock); | |
3960c8c0 | 1480 | return rq; |
cbce1a68 | 1481 | } |
3960c8c0 PZ |
1482 | raw_spin_unlock(&rq->lock); |
1483 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1484 | ||
1485 | while (unlikely(task_on_rq_migrating(p))) | |
1486 | cpu_relax(); | |
1487 | } | |
1488 | } | |
1489 | ||
1490 | static inline void __task_rq_unlock(struct rq *rq) | |
1491 | __releases(rq->lock) | |
1492 | { | |
cbce1a68 | 1493 | lockdep_unpin_lock(&rq->lock); |
3960c8c0 PZ |
1494 | raw_spin_unlock(&rq->lock); |
1495 | } | |
1496 | ||
1497 | static inline void | |
1498 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1499 | __releases(rq->lock) | |
1500 | __releases(p->pi_lock) | |
1501 | { | |
cbce1a68 | 1502 | lockdep_unpin_lock(&rq->lock); |
3960c8c0 PZ |
1503 | raw_spin_unlock(&rq->lock); |
1504 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1505 | } | |
1506 | ||
029632fb PZ |
1507 | #ifdef CONFIG_SMP |
1508 | #ifdef CONFIG_PREEMPT | |
1509 | ||
1510 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1511 | ||
1512 | /* | |
1513 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1514 | * way at the expense of forcing extra atomic operations in all | |
1515 | * invocations. This assures that the double_lock is acquired using the | |
1516 | * same underlying policy as the spinlock_t on this architecture, which | |
1517 | * reduces latency compared to the unfair variant below. However, it | |
1518 | * also adds more overhead and therefore may reduce throughput. | |
1519 | */ | |
1520 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1521 | __releases(this_rq->lock) | |
1522 | __acquires(busiest->lock) | |
1523 | __acquires(this_rq->lock) | |
1524 | { | |
1525 | raw_spin_unlock(&this_rq->lock); | |
1526 | double_rq_lock(this_rq, busiest); | |
1527 | ||
1528 | return 1; | |
1529 | } | |
1530 | ||
1531 | #else | |
1532 | /* | |
1533 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1534 | * latency by eliminating extra atomic operations when the locks are | |
1535 | * already in proper order on entry. This favors lower cpu-ids and will | |
1536 | * grant the double lock to lower cpus over higher ids under contention, | |
1537 | * regardless of entry order into the function. | |
1538 | */ | |
1539 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1540 | __releases(this_rq->lock) | |
1541 | __acquires(busiest->lock) | |
1542 | __acquires(this_rq->lock) | |
1543 | { | |
1544 | int ret = 0; | |
1545 | ||
1546 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1547 | if (busiest < this_rq) { | |
1548 | raw_spin_unlock(&this_rq->lock); | |
1549 | raw_spin_lock(&busiest->lock); | |
1550 | raw_spin_lock_nested(&this_rq->lock, | |
1551 | SINGLE_DEPTH_NESTING); | |
1552 | ret = 1; | |
1553 | } else | |
1554 | raw_spin_lock_nested(&busiest->lock, | |
1555 | SINGLE_DEPTH_NESTING); | |
1556 | } | |
1557 | return ret; | |
1558 | } | |
1559 | ||
1560 | #endif /* CONFIG_PREEMPT */ | |
1561 | ||
1562 | /* | |
1563 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1564 | */ | |
1565 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1566 | { | |
1567 | if (unlikely(!irqs_disabled())) { | |
1568 | /* printk() doesn't work good under rq->lock */ | |
1569 | raw_spin_unlock(&this_rq->lock); | |
1570 | BUG_ON(1); | |
1571 | } | |
1572 | ||
1573 | return _double_lock_balance(this_rq, busiest); | |
1574 | } | |
1575 | ||
1576 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1577 | __releases(busiest->lock) | |
1578 | { | |
1579 | raw_spin_unlock(&busiest->lock); | |
1580 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1581 | } | |
1582 | ||
74602315 PZ |
1583 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1584 | { | |
1585 | if (l1 > l2) | |
1586 | swap(l1, l2); | |
1587 | ||
1588 | spin_lock(l1); | |
1589 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1590 | } | |
1591 | ||
60e69eed MG |
1592 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1593 | { | |
1594 | if (l1 > l2) | |
1595 | swap(l1, l2); | |
1596 | ||
1597 | spin_lock_irq(l1); | |
1598 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1599 | } | |
1600 | ||
74602315 PZ |
1601 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1602 | { | |
1603 | if (l1 > l2) | |
1604 | swap(l1, l2); | |
1605 | ||
1606 | raw_spin_lock(l1); | |
1607 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1608 | } | |
1609 | ||
029632fb PZ |
1610 | /* |
1611 | * double_rq_lock - safely lock two runqueues | |
1612 | * | |
1613 | * Note this does not disable interrupts like task_rq_lock, | |
1614 | * you need to do so manually before calling. | |
1615 | */ | |
1616 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1617 | __acquires(rq1->lock) | |
1618 | __acquires(rq2->lock) | |
1619 | { | |
1620 | BUG_ON(!irqs_disabled()); | |
1621 | if (rq1 == rq2) { | |
1622 | raw_spin_lock(&rq1->lock); | |
1623 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1624 | } else { | |
1625 | if (rq1 < rq2) { | |
1626 | raw_spin_lock(&rq1->lock); | |
1627 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1628 | } else { | |
1629 | raw_spin_lock(&rq2->lock); | |
1630 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1631 | } | |
1632 | } | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * double_rq_unlock - safely unlock two runqueues | |
1637 | * | |
1638 | * Note this does not restore interrupts like task_rq_unlock, | |
1639 | * you need to do so manually after calling. | |
1640 | */ | |
1641 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1642 | __releases(rq1->lock) | |
1643 | __releases(rq2->lock) | |
1644 | { | |
1645 | raw_spin_unlock(&rq1->lock); | |
1646 | if (rq1 != rq2) | |
1647 | raw_spin_unlock(&rq2->lock); | |
1648 | else | |
1649 | __release(rq2->lock); | |
1650 | } | |
1651 | ||
1652 | #else /* CONFIG_SMP */ | |
1653 | ||
1654 | /* | |
1655 | * double_rq_lock - safely lock two runqueues | |
1656 | * | |
1657 | * Note this does not disable interrupts like task_rq_lock, | |
1658 | * you need to do so manually before calling. | |
1659 | */ | |
1660 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1661 | __acquires(rq1->lock) | |
1662 | __acquires(rq2->lock) | |
1663 | { | |
1664 | BUG_ON(!irqs_disabled()); | |
1665 | BUG_ON(rq1 != rq2); | |
1666 | raw_spin_lock(&rq1->lock); | |
1667 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1668 | } | |
1669 | ||
1670 | /* | |
1671 | * double_rq_unlock - safely unlock two runqueues | |
1672 | * | |
1673 | * Note this does not restore interrupts like task_rq_unlock, | |
1674 | * you need to do so manually after calling. | |
1675 | */ | |
1676 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1677 | __releases(rq1->lock) | |
1678 | __releases(rq2->lock) | |
1679 | { | |
1680 | BUG_ON(rq1 != rq2); | |
1681 | raw_spin_unlock(&rq1->lock); | |
1682 | __release(rq2->lock); | |
1683 | } | |
1684 | ||
1685 | #endif | |
1686 | ||
1687 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1688 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
1689 | |
1690 | #ifdef CONFIG_SCHED_DEBUG | |
029632fb PZ |
1691 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
1692 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 1693 | extern void print_dl_stats(struct seq_file *m, int cpu); |
6b55c965 SD |
1694 | extern void |
1695 | print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); | |
397f2378 SD |
1696 | |
1697 | #ifdef CONFIG_NUMA_BALANCING | |
1698 | extern void | |
1699 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
1700 | extern void | |
1701 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
1702 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
1703 | #endif /* CONFIG_NUMA_BALANCING */ | |
1704 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
1705 | |
1706 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
1707 | extern void init_rt_rq(struct rt_rq *rt_rq); |
1708 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 1709 | |
1ee14e6c BS |
1710 | extern void cfs_bandwidth_usage_inc(void); |
1711 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1712 | |
3451d024 | 1713 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1714 | enum rq_nohz_flag_bits { |
1715 | NOHZ_TICK_STOPPED, | |
1716 | NOHZ_BALANCE_KICK, | |
1717 | }; | |
1718 | ||
1719 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
1720 | #endif | |
73fbec60 FW |
1721 | |
1722 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
1723 | ||
1724 | DECLARE_PER_CPU(u64, cpu_hardirq_time); | |
1725 | DECLARE_PER_CPU(u64, cpu_softirq_time); | |
1726 | ||
1727 | #ifndef CONFIG_64BIT | |
1728 | DECLARE_PER_CPU(seqcount_t, irq_time_seq); | |
1729 | ||
1730 | static inline void irq_time_write_begin(void) | |
1731 | { | |
1732 | __this_cpu_inc(irq_time_seq.sequence); | |
1733 | smp_wmb(); | |
1734 | } | |
1735 | ||
1736 | static inline void irq_time_write_end(void) | |
1737 | { | |
1738 | smp_wmb(); | |
1739 | __this_cpu_inc(irq_time_seq.sequence); | |
1740 | } | |
1741 | ||
1742 | static inline u64 irq_time_read(int cpu) | |
1743 | { | |
1744 | u64 irq_time; | |
1745 | unsigned seq; | |
1746 | ||
1747 | do { | |
1748 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1749 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1750 | per_cpu(cpu_hardirq_time, cpu); | |
1751 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1752 | ||
1753 | return irq_time; | |
1754 | } | |
1755 | #else /* CONFIG_64BIT */ | |
1756 | static inline void irq_time_write_begin(void) | |
1757 | { | |
1758 | } | |
1759 | ||
1760 | static inline void irq_time_write_end(void) | |
1761 | { | |
1762 | } | |
1763 | ||
1764 | static inline u64 irq_time_read(int cpu) | |
1765 | { | |
1766 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1767 | } | |
1768 | #endif /* CONFIG_64BIT */ | |
1769 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |