Commit | Line | Data |
---|---|---|
029632fb PZ |
1 | |
2 | #include <linux/sched.h> | |
dfc3401a | 3 | #include <linux/sched/autogroup.h> |
cf4aebc2 | 4 | #include <linux/sched/sysctl.h> |
105ab3d8 | 5 | #include <linux/sched/topology.h> |
8bd75c77 | 6 | #include <linux/sched/rt.h> |
ef8bd77f | 7 | #include <linux/sched/deadline.h> |
e6017571 | 8 | #include <linux/sched/clock.h> |
84f001e1 | 9 | #include <linux/sched/wake_q.h> |
3f07c014 | 10 | #include <linux/sched/signal.h> |
6a3827d7 | 11 | #include <linux/sched/numa_balancing.h> |
6e84f315 | 12 | #include <linux/sched/mm.h> |
55687da1 | 13 | #include <linux/sched/cpufreq.h> |
03441a34 | 14 | #include <linux/sched/stat.h> |
370c9135 | 15 | #include <linux/sched/nohz.h> |
b17b0153 | 16 | #include <linux/sched/debug.h> |
ef8bd77f | 17 | #include <linux/sched/hotplug.h> |
29930025 | 18 | #include <linux/sched/task.h> |
68db0cf1 | 19 | #include <linux/sched/task_stack.h> |
32ef5517 | 20 | #include <linux/sched/cputime.h> |
1777e463 | 21 | #include <linux/sched/init.h> |
ef8bd77f | 22 | |
19d23dbf | 23 | #include <linux/u64_stats_sync.h> |
a499a5a1 | 24 | #include <linux/kernel_stat.h> |
3866e845 | 25 | #include <linux/binfmts.h> |
029632fb PZ |
26 | #include <linux/mutex.h> |
27 | #include <linux/spinlock.h> | |
28 | #include <linux/stop_machine.h> | |
b6366f04 | 29 | #include <linux/irq_work.h> |
9f3660c2 | 30 | #include <linux/tick.h> |
f809ca9a | 31 | #include <linux/slab.h> |
029632fb | 32 | |
7fce777c IM |
33 | #ifdef CONFIG_PARAVIRT |
34 | #include <asm/paravirt.h> | |
35 | #endif | |
36 | ||
391e43da | 37 | #include "cpupri.h" |
6bfd6d72 | 38 | #include "cpudeadline.h" |
60fed789 | 39 | #include "cpuacct.h" |
029632fb | 40 | |
9148a3a1 PZ |
41 | #ifdef CONFIG_SCHED_DEBUG |
42 | #define SCHED_WARN_ON(x) WARN_ONCE(x, #x) | |
43 | #else | |
44 | #define SCHED_WARN_ON(x) ((void)(x)) | |
45 | #endif | |
46 | ||
45ceebf7 | 47 | struct rq; |
442bf3aa | 48 | struct cpuidle_state; |
45ceebf7 | 49 | |
da0c1e65 KT |
50 | /* task_struct::on_rq states: */ |
51 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 52 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 53 | |
029632fb PZ |
54 | extern __read_mostly int scheduler_running; |
55 | ||
45ceebf7 PG |
56 | extern unsigned long calc_load_update; |
57 | extern atomic_long_t calc_load_tasks; | |
58 | ||
3289bdb4 | 59 | extern void calc_global_load_tick(struct rq *this_rq); |
d60585c5 | 60 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
3289bdb4 PZ |
61 | |
62 | #ifdef CONFIG_SMP | |
cee1afce | 63 | extern void cpu_load_update_active(struct rq *this_rq); |
3289bdb4 | 64 | #else |
cee1afce | 65 | static inline void cpu_load_update_active(struct rq *this_rq) { } |
3289bdb4 | 66 | #endif |
45ceebf7 | 67 | |
029632fb PZ |
68 | /* |
69 | * Helpers for converting nanosecond timing to jiffy resolution | |
70 | */ | |
71 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
72 | ||
cc1f4b1f LZ |
73 | /* |
74 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
75 | * The extra resolution improves shares distribution and load balancing of | |
76 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
77 | * hierarchies, especially on larger systems. This is not a user-visible change | |
78 | * and does not change the user-interface for setting shares/weights. | |
79 | * | |
80 | * We increase resolution only if we have enough bits to allow this increased | |
2159197d PZ |
81 | * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are |
82 | * pretty high and the returns do not justify the increased costs. | |
83 | * | |
84 | * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to | |
85 | * increase coverage and consistency always enable it on 64bit platforms. | |
cc1f4b1f | 86 | */ |
2159197d | 87 | #ifdef CONFIG_64BIT |
172895e6 | 88 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
6ecdd749 YD |
89 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
90 | # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) | |
cc1f4b1f | 91 | #else |
172895e6 | 92 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
cc1f4b1f LZ |
93 | # define scale_load(w) (w) |
94 | # define scale_load_down(w) (w) | |
95 | #endif | |
96 | ||
6ecdd749 | 97 | /* |
172895e6 YD |
98 | * Task weight (visible to users) and its load (invisible to users) have |
99 | * independent resolution, but they should be well calibrated. We use | |
100 | * scale_load() and scale_load_down(w) to convert between them. The | |
101 | * following must be true: | |
102 | * | |
103 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD | |
104 | * | |
6ecdd749 | 105 | */ |
172895e6 | 106 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
029632fb | 107 | |
332ac17e DF |
108 | /* |
109 | * Single value that decides SCHED_DEADLINE internal math precision. | |
110 | * 10 -> just above 1us | |
111 | * 9 -> just above 0.5us | |
112 | */ | |
113 | #define DL_SCALE (10) | |
114 | ||
029632fb PZ |
115 | /* |
116 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 117 | */ |
029632fb PZ |
118 | |
119 | /* | |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
20f9cd2a HA |
124 | static inline int idle_policy(int policy) |
125 | { | |
126 | return policy == SCHED_IDLE; | |
127 | } | |
d50dde5a DF |
128 | static inline int fair_policy(int policy) |
129 | { | |
130 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
131 | } | |
132 | ||
029632fb PZ |
133 | static inline int rt_policy(int policy) |
134 | { | |
d50dde5a | 135 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
136 | } |
137 | ||
aab03e05 DF |
138 | static inline int dl_policy(int policy) |
139 | { | |
140 | return policy == SCHED_DEADLINE; | |
141 | } | |
20f9cd2a HA |
142 | static inline bool valid_policy(int policy) |
143 | { | |
144 | return idle_policy(policy) || fair_policy(policy) || | |
145 | rt_policy(policy) || dl_policy(policy); | |
146 | } | |
aab03e05 | 147 | |
029632fb PZ |
148 | static inline int task_has_rt_policy(struct task_struct *p) |
149 | { | |
150 | return rt_policy(p->policy); | |
151 | } | |
152 | ||
aab03e05 DF |
153 | static inline int task_has_dl_policy(struct task_struct *p) |
154 | { | |
155 | return dl_policy(p->policy); | |
156 | } | |
157 | ||
2d3d891d DF |
158 | /* |
159 | * Tells if entity @a should preempt entity @b. | |
160 | */ | |
332ac17e DF |
161 | static inline bool |
162 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d DF |
163 | { |
164 | return dl_time_before(a->deadline, b->deadline); | |
165 | } | |
166 | ||
029632fb PZ |
167 | /* |
168 | * This is the priority-queue data structure of the RT scheduling class: | |
169 | */ | |
170 | struct rt_prio_array { | |
171 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
172 | struct list_head queue[MAX_RT_PRIO]; | |
173 | }; | |
174 | ||
175 | struct rt_bandwidth { | |
176 | /* nests inside the rq lock: */ | |
177 | raw_spinlock_t rt_runtime_lock; | |
178 | ktime_t rt_period; | |
179 | u64 rt_runtime; | |
180 | struct hrtimer rt_period_timer; | |
4cfafd30 | 181 | unsigned int rt_period_active; |
029632fb | 182 | }; |
a5e7be3b JL |
183 | |
184 | void __dl_clear_params(struct task_struct *p); | |
185 | ||
332ac17e DF |
186 | /* |
187 | * To keep the bandwidth of -deadline tasks and groups under control | |
188 | * we need some place where: | |
189 | * - store the maximum -deadline bandwidth of the system (the group); | |
190 | * - cache the fraction of that bandwidth that is currently allocated. | |
191 | * | |
192 | * This is all done in the data structure below. It is similar to the | |
193 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
194 | * that, since here we are only interested in admission control, we | |
195 | * do not decrease any runtime while the group "executes", neither we | |
196 | * need a timer to replenish it. | |
197 | * | |
198 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
199 | * meaning that: | |
200 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
201 | * - dl_total_bw array contains, in the i-eth element, the currently | |
202 | * allocated bandwidth on the i-eth CPU. | |
203 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
204 | * consume bandwidth on the CPU they're running on. | |
205 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
206 | * that will be shown the next time the proc or cgroup controls will | |
207 | * be red. It on its turn can be changed by writing on its own | |
208 | * control. | |
209 | */ | |
210 | struct dl_bandwidth { | |
211 | raw_spinlock_t dl_runtime_lock; | |
212 | u64 dl_runtime; | |
213 | u64 dl_period; | |
214 | }; | |
215 | ||
216 | static inline int dl_bandwidth_enabled(void) | |
217 | { | |
1724813d | 218 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
219 | } |
220 | ||
221 | extern struct dl_bw *dl_bw_of(int i); | |
daec5798 | 222 | extern int dl_bw_cpus(int i); |
332ac17e DF |
223 | |
224 | struct dl_bw { | |
225 | raw_spinlock_t lock; | |
226 | u64 bw, total_bw; | |
227 | }; | |
228 | ||
daec5798 LA |
229 | static inline void __dl_update(struct dl_bw *dl_b, s64 bw); |
230 | ||
7f51412a | 231 | static inline |
daec5798 | 232 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
233 | { |
234 | dl_b->total_bw -= tsk_bw; | |
daec5798 | 235 | __dl_update(dl_b, (s32)tsk_bw / cpus); |
7f51412a JL |
236 | } |
237 | ||
238 | static inline | |
daec5798 | 239 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
240 | { |
241 | dl_b->total_bw += tsk_bw; | |
daec5798 | 242 | __dl_update(dl_b, -((s32)tsk_bw / cpus)); |
7f51412a JL |
243 | } |
244 | ||
245 | static inline | |
246 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
247 | { | |
248 | return dl_b->bw != -1 && | |
249 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
250 | } | |
251 | ||
209a0cbd | 252 | void dl_change_utilization(struct task_struct *p, u64 new_bw); |
f2cb1360 | 253 | extern void init_dl_bw(struct dl_bw *dl_b); |
029632fb PZ |
254 | |
255 | #ifdef CONFIG_CGROUP_SCHED | |
256 | ||
257 | #include <linux/cgroup.h> | |
258 | ||
259 | struct cfs_rq; | |
260 | struct rt_rq; | |
261 | ||
35cf4e50 | 262 | extern struct list_head task_groups; |
029632fb PZ |
263 | |
264 | struct cfs_bandwidth { | |
265 | #ifdef CONFIG_CFS_BANDWIDTH | |
266 | raw_spinlock_t lock; | |
267 | ktime_t period; | |
268 | u64 quota, runtime; | |
9c58c79a | 269 | s64 hierarchical_quota; |
029632fb PZ |
270 | u64 runtime_expires; |
271 | ||
4cfafd30 | 272 | int idle, period_active; |
029632fb PZ |
273 | struct hrtimer period_timer, slack_timer; |
274 | struct list_head throttled_cfs_rq; | |
275 | ||
276 | /* statistics */ | |
277 | int nr_periods, nr_throttled; | |
278 | u64 throttled_time; | |
279 | #endif | |
280 | }; | |
281 | ||
282 | /* task group related information */ | |
283 | struct task_group { | |
284 | struct cgroup_subsys_state css; | |
285 | ||
286 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
287 | /* schedulable entities of this group on each cpu */ | |
288 | struct sched_entity **se; | |
289 | /* runqueue "owned" by this group on each cpu */ | |
290 | struct cfs_rq **cfs_rq; | |
291 | unsigned long shares; | |
292 | ||
fa6bddeb | 293 | #ifdef CONFIG_SMP |
b0367629 WL |
294 | /* |
295 | * load_avg can be heavily contended at clock tick time, so put | |
296 | * it in its own cacheline separated from the fields above which | |
297 | * will also be accessed at each tick. | |
298 | */ | |
299 | atomic_long_t load_avg ____cacheline_aligned; | |
029632fb | 300 | #endif |
fa6bddeb | 301 | #endif |
029632fb PZ |
302 | |
303 | #ifdef CONFIG_RT_GROUP_SCHED | |
304 | struct sched_rt_entity **rt_se; | |
305 | struct rt_rq **rt_rq; | |
306 | ||
307 | struct rt_bandwidth rt_bandwidth; | |
308 | #endif | |
309 | ||
310 | struct rcu_head rcu; | |
311 | struct list_head list; | |
312 | ||
313 | struct task_group *parent; | |
314 | struct list_head siblings; | |
315 | struct list_head children; | |
316 | ||
317 | #ifdef CONFIG_SCHED_AUTOGROUP | |
318 | struct autogroup *autogroup; | |
319 | #endif | |
320 | ||
321 | struct cfs_bandwidth cfs_bandwidth; | |
322 | }; | |
323 | ||
324 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
325 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
326 | ||
327 | /* | |
328 | * A weight of 0 or 1 can cause arithmetics problems. | |
329 | * A weight of a cfs_rq is the sum of weights of which entities | |
330 | * are queued on this cfs_rq, so a weight of a entity should not be | |
331 | * too large, so as the shares value of a task group. | |
332 | * (The default weight is 1024 - so there's no practical | |
333 | * limitation from this.) | |
334 | */ | |
335 | #define MIN_SHARES (1UL << 1) | |
336 | #define MAX_SHARES (1UL << 18) | |
337 | #endif | |
338 | ||
029632fb PZ |
339 | typedef int (*tg_visitor)(struct task_group *, void *); |
340 | ||
341 | extern int walk_tg_tree_from(struct task_group *from, | |
342 | tg_visitor down, tg_visitor up, void *data); | |
343 | ||
344 | /* | |
345 | * Iterate the full tree, calling @down when first entering a node and @up when | |
346 | * leaving it for the final time. | |
347 | * | |
348 | * Caller must hold rcu_lock or sufficient equivalent. | |
349 | */ | |
350 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
351 | { | |
352 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
353 | } | |
354 | ||
355 | extern int tg_nop(struct task_group *tg, void *data); | |
356 | ||
357 | extern void free_fair_sched_group(struct task_group *tg); | |
358 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
8663e24d | 359 | extern void online_fair_sched_group(struct task_group *tg); |
6fe1f348 | 360 | extern void unregister_fair_sched_group(struct task_group *tg); |
029632fb PZ |
361 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
362 | struct sched_entity *se, int cpu, | |
363 | struct sched_entity *parent); | |
364 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
029632fb PZ |
365 | |
366 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 367 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
368 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
369 | ||
370 | extern void free_rt_sched_group(struct task_group *tg); | |
371 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
372 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
373 | struct sched_rt_entity *rt_se, int cpu, | |
374 | struct sched_rt_entity *parent); | |
375 | ||
25cc7da7 LZ |
376 | extern struct task_group *sched_create_group(struct task_group *parent); |
377 | extern void sched_online_group(struct task_group *tg, | |
378 | struct task_group *parent); | |
379 | extern void sched_destroy_group(struct task_group *tg); | |
380 | extern void sched_offline_group(struct task_group *tg); | |
381 | ||
382 | extern void sched_move_task(struct task_struct *tsk); | |
383 | ||
384 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
385 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
ad936d86 BP |
386 | |
387 | #ifdef CONFIG_SMP | |
388 | extern void set_task_rq_fair(struct sched_entity *se, | |
389 | struct cfs_rq *prev, struct cfs_rq *next); | |
390 | #else /* !CONFIG_SMP */ | |
391 | static inline void set_task_rq_fair(struct sched_entity *se, | |
392 | struct cfs_rq *prev, struct cfs_rq *next) { } | |
393 | #endif /* CONFIG_SMP */ | |
394 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
25cc7da7 | 395 | |
029632fb PZ |
396 | #else /* CONFIG_CGROUP_SCHED */ |
397 | ||
398 | struct cfs_bandwidth { }; | |
399 | ||
400 | #endif /* CONFIG_CGROUP_SCHED */ | |
401 | ||
402 | /* CFS-related fields in a runqueue */ | |
403 | struct cfs_rq { | |
404 | struct load_weight load; | |
c82513e5 | 405 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
406 | |
407 | u64 exec_clock; | |
408 | u64 min_vruntime; | |
409 | #ifndef CONFIG_64BIT | |
410 | u64 min_vruntime_copy; | |
411 | #endif | |
412 | ||
413 | struct rb_root tasks_timeline; | |
414 | struct rb_node *rb_leftmost; | |
415 | ||
029632fb PZ |
416 | /* |
417 | * 'curr' points to currently running entity on this cfs_rq. | |
418 | * It is set to NULL otherwise (i.e when none are currently running). | |
419 | */ | |
420 | struct sched_entity *curr, *next, *last, *skip; | |
421 | ||
422 | #ifdef CONFIG_SCHED_DEBUG | |
423 | unsigned int nr_spread_over; | |
424 | #endif | |
425 | ||
2dac754e PT |
426 | #ifdef CONFIG_SMP |
427 | /* | |
9d89c257 | 428 | * CFS load tracking |
2dac754e | 429 | */ |
9d89c257 | 430 | struct sched_avg avg; |
13962234 YD |
431 | u64 runnable_load_sum; |
432 | unsigned long runnable_load_avg; | |
c566e8e9 | 433 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9d89c257 | 434 | unsigned long tg_load_avg_contrib; |
09a43ace | 435 | unsigned long propagate_avg; |
9d89c257 YD |
436 | #endif |
437 | atomic_long_t removed_load_avg, removed_util_avg; | |
438 | #ifndef CONFIG_64BIT | |
439 | u64 load_last_update_time_copy; | |
440 | #endif | |
82958366 | 441 | |
9d89c257 | 442 | #ifdef CONFIG_FAIR_GROUP_SCHED |
82958366 PT |
443 | /* |
444 | * h_load = weight * f(tg) | |
445 | * | |
446 | * Where f(tg) is the recursive weight fraction assigned to | |
447 | * this group. | |
448 | */ | |
449 | unsigned long h_load; | |
68520796 VD |
450 | u64 last_h_load_update; |
451 | struct sched_entity *h_load_next; | |
452 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
453 | #endif /* CONFIG_SMP */ |
454 | ||
029632fb PZ |
455 | #ifdef CONFIG_FAIR_GROUP_SCHED |
456 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
457 | ||
458 | /* | |
459 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
460 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
461 | * (like users, containers etc.) | |
462 | * | |
463 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
464 | * list is used during load balance. | |
465 | */ | |
466 | int on_list; | |
467 | struct list_head leaf_cfs_rq_list; | |
468 | struct task_group *tg; /* group that "owns" this runqueue */ | |
469 | ||
029632fb PZ |
470 | #ifdef CONFIG_CFS_BANDWIDTH |
471 | int runtime_enabled; | |
472 | u64 runtime_expires; | |
473 | s64 runtime_remaining; | |
474 | ||
f1b17280 PT |
475 | u64 throttled_clock, throttled_clock_task; |
476 | u64 throttled_clock_task_time; | |
55e16d30 | 477 | int throttled, throttle_count; |
029632fb PZ |
478 | struct list_head throttled_list; |
479 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
480 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
481 | }; | |
482 | ||
483 | static inline int rt_bandwidth_enabled(void) | |
484 | { | |
485 | return sysctl_sched_rt_runtime >= 0; | |
486 | } | |
487 | ||
b6366f04 SR |
488 | /* RT IPI pull logic requires IRQ_WORK */ |
489 | #ifdef CONFIG_IRQ_WORK | |
490 | # define HAVE_RT_PUSH_IPI | |
491 | #endif | |
492 | ||
029632fb PZ |
493 | /* Real-Time classes' related field in a runqueue: */ |
494 | struct rt_rq { | |
495 | struct rt_prio_array active; | |
c82513e5 | 496 | unsigned int rt_nr_running; |
01d36d0a | 497 | unsigned int rr_nr_running; |
029632fb PZ |
498 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
499 | struct { | |
500 | int curr; /* highest queued rt task prio */ | |
501 | #ifdef CONFIG_SMP | |
502 | int next; /* next highest */ | |
503 | #endif | |
504 | } highest_prio; | |
505 | #endif | |
506 | #ifdef CONFIG_SMP | |
507 | unsigned long rt_nr_migratory; | |
508 | unsigned long rt_nr_total; | |
509 | int overloaded; | |
510 | struct plist_head pushable_tasks; | |
b6366f04 SR |
511 | #ifdef HAVE_RT_PUSH_IPI |
512 | int push_flags; | |
513 | int push_cpu; | |
514 | struct irq_work push_work; | |
515 | raw_spinlock_t push_lock; | |
029632fb | 516 | #endif |
b6366f04 | 517 | #endif /* CONFIG_SMP */ |
f4ebcbc0 KT |
518 | int rt_queued; |
519 | ||
029632fb PZ |
520 | int rt_throttled; |
521 | u64 rt_time; | |
522 | u64 rt_runtime; | |
523 | /* Nests inside the rq lock: */ | |
524 | raw_spinlock_t rt_runtime_lock; | |
525 | ||
526 | #ifdef CONFIG_RT_GROUP_SCHED | |
527 | unsigned long rt_nr_boosted; | |
528 | ||
529 | struct rq *rq; | |
029632fb PZ |
530 | struct task_group *tg; |
531 | #endif | |
532 | }; | |
533 | ||
aab03e05 DF |
534 | /* Deadline class' related fields in a runqueue */ |
535 | struct dl_rq { | |
536 | /* runqueue is an rbtree, ordered by deadline */ | |
537 | struct rb_root rb_root; | |
538 | struct rb_node *rb_leftmost; | |
539 | ||
540 | unsigned long dl_nr_running; | |
1baca4ce JL |
541 | |
542 | #ifdef CONFIG_SMP | |
543 | /* | |
544 | * Deadline values of the currently executing and the | |
545 | * earliest ready task on this rq. Caching these facilitates | |
546 | * the decision wether or not a ready but not running task | |
547 | * should migrate somewhere else. | |
548 | */ | |
549 | struct { | |
550 | u64 curr; | |
551 | u64 next; | |
552 | } earliest_dl; | |
553 | ||
554 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
555 | int overloaded; |
556 | ||
557 | /* | |
558 | * Tasks on this rq that can be pushed away. They are kept in | |
559 | * an rb-tree, ordered by tasks' deadlines, with caching | |
560 | * of the leftmost (earliest deadline) element. | |
561 | */ | |
562 | struct rb_root pushable_dl_tasks_root; | |
563 | struct rb_node *pushable_dl_tasks_leftmost; | |
332ac17e DF |
564 | #else |
565 | struct dl_bw dl_bw; | |
1baca4ce | 566 | #endif |
e36d8677 LA |
567 | /* |
568 | * "Active utilization" for this runqueue: increased when a | |
569 | * task wakes up (becomes TASK_RUNNING) and decreased when a | |
570 | * task blocks | |
571 | */ | |
572 | u64 running_bw; | |
4da3abce | 573 | |
8fd27231 LA |
574 | /* |
575 | * Utilization of the tasks "assigned" to this runqueue (including | |
576 | * the tasks that are in runqueue and the tasks that executed on this | |
577 | * CPU and blocked). Increased when a task moves to this runqueue, and | |
578 | * decreased when the task moves away (migrates, changes scheduling | |
579 | * policy, or terminates). | |
580 | * This is needed to compute the "inactive utilization" for the | |
581 | * runqueue (inactive utilization = this_bw - running_bw). | |
582 | */ | |
583 | u64 this_bw; | |
daec5798 | 584 | u64 extra_bw; |
8fd27231 | 585 | |
4da3abce LA |
586 | /* |
587 | * Inverse of the fraction of CPU utilization that can be reclaimed | |
588 | * by the GRUB algorithm. | |
589 | */ | |
590 | u64 bw_ratio; | |
aab03e05 DF |
591 | }; |
592 | ||
029632fb PZ |
593 | #ifdef CONFIG_SMP |
594 | ||
afe06efd TC |
595 | static inline bool sched_asym_prefer(int a, int b) |
596 | { | |
597 | return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); | |
598 | } | |
599 | ||
029632fb PZ |
600 | /* |
601 | * We add the notion of a root-domain which will be used to define per-domain | |
602 | * variables. Each exclusive cpuset essentially defines an island domain by | |
603 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
604 | * exclusive cpuset is created, we also create and attach a new root-domain | |
605 | * object. | |
606 | * | |
607 | */ | |
608 | struct root_domain { | |
609 | atomic_t refcount; | |
610 | atomic_t rto_count; | |
611 | struct rcu_head rcu; | |
612 | cpumask_var_t span; | |
613 | cpumask_var_t online; | |
614 | ||
4486edd1 TC |
615 | /* Indicate more than one runnable task for any CPU */ |
616 | bool overload; | |
617 | ||
1baca4ce JL |
618 | /* |
619 | * The bit corresponding to a CPU gets set here if such CPU has more | |
620 | * than one runnable -deadline task (as it is below for RT tasks). | |
621 | */ | |
622 | cpumask_var_t dlo_mask; | |
623 | atomic_t dlo_count; | |
332ac17e | 624 | struct dl_bw dl_bw; |
6bfd6d72 | 625 | struct cpudl cpudl; |
1baca4ce | 626 | |
029632fb PZ |
627 | /* |
628 | * The "RT overload" flag: it gets set if a CPU has more than | |
629 | * one runnable RT task. | |
630 | */ | |
631 | cpumask_var_t rto_mask; | |
632 | struct cpupri cpupri; | |
cd92bfd3 DE |
633 | |
634 | unsigned long max_cpu_capacity; | |
029632fb PZ |
635 | }; |
636 | ||
637 | extern struct root_domain def_root_domain; | |
f2cb1360 | 638 | extern struct mutex sched_domains_mutex; |
f2cb1360 IM |
639 | |
640 | extern void init_defrootdomain(void); | |
8d5dc512 | 641 | extern int sched_init_domains(const struct cpumask *cpu_map); |
f2cb1360 | 642 | extern void rq_attach_root(struct rq *rq, struct root_domain *rd); |
029632fb PZ |
643 | |
644 | #endif /* CONFIG_SMP */ | |
645 | ||
646 | /* | |
647 | * This is the main, per-CPU runqueue data structure. | |
648 | * | |
649 | * Locking rule: those places that want to lock multiple runqueues | |
650 | * (such as the load balancing or the thread migration code), lock | |
651 | * acquire operations must be ordered by ascending &runqueue. | |
652 | */ | |
653 | struct rq { | |
654 | /* runqueue lock: */ | |
655 | raw_spinlock_t lock; | |
656 | ||
657 | /* | |
658 | * nr_running and cpu_load should be in the same cacheline because | |
659 | * remote CPUs use both these fields when doing load calculation. | |
660 | */ | |
c82513e5 | 661 | unsigned int nr_running; |
0ec8aa00 PZ |
662 | #ifdef CONFIG_NUMA_BALANCING |
663 | unsigned int nr_numa_running; | |
664 | unsigned int nr_preferred_running; | |
665 | #endif | |
029632fb PZ |
666 | #define CPU_LOAD_IDX_MAX 5 |
667 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
3451d024 | 668 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 FW |
669 | #ifdef CONFIG_SMP |
670 | unsigned long last_load_update_tick; | |
671 | #endif /* CONFIG_SMP */ | |
1c792db7 | 672 | unsigned long nohz_flags; |
9fd81dd5 | 673 | #endif /* CONFIG_NO_HZ_COMMON */ |
265f22a9 FW |
674 | #ifdef CONFIG_NO_HZ_FULL |
675 | unsigned long last_sched_tick; | |
029632fb | 676 | #endif |
029632fb PZ |
677 | /* capture load from *all* tasks on this cpu: */ |
678 | struct load_weight load; | |
679 | unsigned long nr_load_updates; | |
680 | u64 nr_switches; | |
681 | ||
682 | struct cfs_rq cfs; | |
683 | struct rt_rq rt; | |
aab03e05 | 684 | struct dl_rq dl; |
029632fb PZ |
685 | |
686 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
687 | /* list of leaf cfs_rq on this cpu: */ | |
688 | struct list_head leaf_cfs_rq_list; | |
9c2791f9 | 689 | struct list_head *tmp_alone_branch; |
a35b6466 PZ |
690 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
691 | ||
029632fb PZ |
692 | /* |
693 | * This is part of a global counter where only the total sum | |
694 | * over all CPUs matters. A task can increase this counter on | |
695 | * one CPU and if it got migrated afterwards it may decrease | |
696 | * it on another CPU. Always updated under the runqueue lock: | |
697 | */ | |
698 | unsigned long nr_uninterruptible; | |
699 | ||
700 | struct task_struct *curr, *idle, *stop; | |
701 | unsigned long next_balance; | |
702 | struct mm_struct *prev_mm; | |
703 | ||
cb42c9a3 | 704 | unsigned int clock_update_flags; |
029632fb PZ |
705 | u64 clock; |
706 | u64 clock_task; | |
707 | ||
708 | atomic_t nr_iowait; | |
709 | ||
710 | #ifdef CONFIG_SMP | |
711 | struct root_domain *rd; | |
712 | struct sched_domain *sd; | |
713 | ||
ced549fa | 714 | unsigned long cpu_capacity; |
ca6d75e6 | 715 | unsigned long cpu_capacity_orig; |
029632fb | 716 | |
e3fca9e7 PZ |
717 | struct callback_head *balance_callback; |
718 | ||
029632fb PZ |
719 | unsigned char idle_balance; |
720 | /* For active balancing */ | |
029632fb PZ |
721 | int active_balance; |
722 | int push_cpu; | |
723 | struct cpu_stop_work active_balance_work; | |
724 | /* cpu of this runqueue: */ | |
725 | int cpu; | |
726 | int online; | |
727 | ||
367456c7 PZ |
728 | struct list_head cfs_tasks; |
729 | ||
029632fb PZ |
730 | u64 rt_avg; |
731 | u64 age_stamp; | |
732 | u64 idle_stamp; | |
733 | u64 avg_idle; | |
9bd721c5 JL |
734 | |
735 | /* This is used to determine avg_idle's max value */ | |
736 | u64 max_idle_balance_cost; | |
029632fb PZ |
737 | #endif |
738 | ||
739 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
740 | u64 prev_irq_time; | |
741 | #endif | |
742 | #ifdef CONFIG_PARAVIRT | |
743 | u64 prev_steal_time; | |
744 | #endif | |
745 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
746 | u64 prev_steal_time_rq; | |
747 | #endif | |
748 | ||
749 | /* calc_load related fields */ | |
750 | unsigned long calc_load_update; | |
751 | long calc_load_active; | |
752 | ||
753 | #ifdef CONFIG_SCHED_HRTICK | |
754 | #ifdef CONFIG_SMP | |
755 | int hrtick_csd_pending; | |
756 | struct call_single_data hrtick_csd; | |
757 | #endif | |
758 | struct hrtimer hrtick_timer; | |
759 | #endif | |
760 | ||
761 | #ifdef CONFIG_SCHEDSTATS | |
762 | /* latency stats */ | |
763 | struct sched_info rq_sched_info; | |
764 | unsigned long long rq_cpu_time; | |
765 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
766 | ||
767 | /* sys_sched_yield() stats */ | |
768 | unsigned int yld_count; | |
769 | ||
770 | /* schedule() stats */ | |
029632fb PZ |
771 | unsigned int sched_count; |
772 | unsigned int sched_goidle; | |
773 | ||
774 | /* try_to_wake_up() stats */ | |
775 | unsigned int ttwu_count; | |
776 | unsigned int ttwu_local; | |
777 | #endif | |
778 | ||
779 | #ifdef CONFIG_SMP | |
780 | struct llist_head wake_list; | |
781 | #endif | |
442bf3aa DL |
782 | |
783 | #ifdef CONFIG_CPU_IDLE | |
784 | /* Must be inspected within a rcu lock section */ | |
785 | struct cpuidle_state *idle_state; | |
786 | #endif | |
029632fb PZ |
787 | }; |
788 | ||
789 | static inline int cpu_of(struct rq *rq) | |
790 | { | |
791 | #ifdef CONFIG_SMP | |
792 | return rq->cpu; | |
793 | #else | |
794 | return 0; | |
795 | #endif | |
796 | } | |
797 | ||
1b568f0a PZ |
798 | |
799 | #ifdef CONFIG_SCHED_SMT | |
800 | ||
801 | extern struct static_key_false sched_smt_present; | |
802 | ||
803 | extern void __update_idle_core(struct rq *rq); | |
804 | ||
805 | static inline void update_idle_core(struct rq *rq) | |
806 | { | |
807 | if (static_branch_unlikely(&sched_smt_present)) | |
808 | __update_idle_core(rq); | |
809 | } | |
810 | ||
811 | #else | |
812 | static inline void update_idle_core(struct rq *rq) { } | |
813 | #endif | |
814 | ||
8b06c55b | 815 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 816 | |
518cd623 | 817 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 818 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
819 | #define task_rq(p) cpu_rq(task_cpu(p)) |
820 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 821 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 822 | |
cebde6d6 PZ |
823 | static inline u64 __rq_clock_broken(struct rq *rq) |
824 | { | |
316c1608 | 825 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
826 | } |
827 | ||
cb42c9a3 MF |
828 | /* |
829 | * rq::clock_update_flags bits | |
830 | * | |
831 | * %RQCF_REQ_SKIP - will request skipping of clock update on the next | |
832 | * call to __schedule(). This is an optimisation to avoid | |
833 | * neighbouring rq clock updates. | |
834 | * | |
835 | * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is | |
836 | * in effect and calls to update_rq_clock() are being ignored. | |
837 | * | |
838 | * %RQCF_UPDATED - is a debug flag that indicates whether a call has been | |
839 | * made to update_rq_clock() since the last time rq::lock was pinned. | |
840 | * | |
841 | * If inside of __schedule(), clock_update_flags will have been | |
842 | * shifted left (a left shift is a cheap operation for the fast path | |
843 | * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, | |
844 | * | |
845 | * if (rq-clock_update_flags >= RQCF_UPDATED) | |
846 | * | |
847 | * to check if %RQCF_UPADTED is set. It'll never be shifted more than | |
848 | * one position though, because the next rq_unpin_lock() will shift it | |
849 | * back. | |
850 | */ | |
851 | #define RQCF_REQ_SKIP 0x01 | |
852 | #define RQCF_ACT_SKIP 0x02 | |
853 | #define RQCF_UPDATED 0x04 | |
854 | ||
855 | static inline void assert_clock_updated(struct rq *rq) | |
856 | { | |
857 | /* | |
858 | * The only reason for not seeing a clock update since the | |
859 | * last rq_pin_lock() is if we're currently skipping updates. | |
860 | */ | |
861 | SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); | |
862 | } | |
863 | ||
78becc27 FW |
864 | static inline u64 rq_clock(struct rq *rq) |
865 | { | |
cebde6d6 | 866 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
867 | assert_clock_updated(rq); |
868 | ||
78becc27 FW |
869 | return rq->clock; |
870 | } | |
871 | ||
872 | static inline u64 rq_clock_task(struct rq *rq) | |
873 | { | |
cebde6d6 | 874 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
875 | assert_clock_updated(rq); |
876 | ||
78becc27 FW |
877 | return rq->clock_task; |
878 | } | |
879 | ||
9edfbfed PZ |
880 | static inline void rq_clock_skip_update(struct rq *rq, bool skip) |
881 | { | |
882 | lockdep_assert_held(&rq->lock); | |
883 | if (skip) | |
cb42c9a3 | 884 | rq->clock_update_flags |= RQCF_REQ_SKIP; |
9edfbfed | 885 | else |
cb42c9a3 | 886 | rq->clock_update_flags &= ~RQCF_REQ_SKIP; |
9edfbfed PZ |
887 | } |
888 | ||
d8ac8971 MF |
889 | struct rq_flags { |
890 | unsigned long flags; | |
891 | struct pin_cookie cookie; | |
cb42c9a3 MF |
892 | #ifdef CONFIG_SCHED_DEBUG |
893 | /* | |
894 | * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the | |
895 | * current pin context is stashed here in case it needs to be | |
896 | * restored in rq_repin_lock(). | |
897 | */ | |
898 | unsigned int clock_update_flags; | |
899 | #endif | |
d8ac8971 MF |
900 | }; |
901 | ||
902 | static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) | |
903 | { | |
904 | rf->cookie = lockdep_pin_lock(&rq->lock); | |
cb42c9a3 MF |
905 | |
906 | #ifdef CONFIG_SCHED_DEBUG | |
907 | rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); | |
908 | rf->clock_update_flags = 0; | |
909 | #endif | |
d8ac8971 MF |
910 | } |
911 | ||
912 | static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) | |
913 | { | |
cb42c9a3 MF |
914 | #ifdef CONFIG_SCHED_DEBUG |
915 | if (rq->clock_update_flags > RQCF_ACT_SKIP) | |
916 | rf->clock_update_flags = RQCF_UPDATED; | |
917 | #endif | |
918 | ||
d8ac8971 MF |
919 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
920 | } | |
921 | ||
922 | static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) | |
923 | { | |
924 | lockdep_repin_lock(&rq->lock, rf->cookie); | |
cb42c9a3 MF |
925 | |
926 | #ifdef CONFIG_SCHED_DEBUG | |
927 | /* | |
928 | * Restore the value we stashed in @rf for this pin context. | |
929 | */ | |
930 | rq->clock_update_flags |= rf->clock_update_flags; | |
931 | #endif | |
d8ac8971 MF |
932 | } |
933 | ||
9942f79b | 934 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
935 | enum numa_topology_type { |
936 | NUMA_DIRECT, | |
937 | NUMA_GLUELESS_MESH, | |
938 | NUMA_BACKPLANE, | |
939 | }; | |
940 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
941 | extern int sched_max_numa_distance; |
942 | extern bool find_numa_distance(int distance); | |
943 | #endif | |
944 | ||
f2cb1360 IM |
945 | #ifdef CONFIG_NUMA |
946 | extern void sched_init_numa(void); | |
947 | extern void sched_domains_numa_masks_set(unsigned int cpu); | |
948 | extern void sched_domains_numa_masks_clear(unsigned int cpu); | |
949 | #else | |
950 | static inline void sched_init_numa(void) { } | |
951 | static inline void sched_domains_numa_masks_set(unsigned int cpu) { } | |
952 | static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
953 | #endif | |
954 | ||
f809ca9a | 955 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
956 | /* The regions in numa_faults array from task_struct */ |
957 | enum numa_faults_stats { | |
958 | NUMA_MEM = 0, | |
959 | NUMA_CPU, | |
960 | NUMA_MEMBUF, | |
961 | NUMA_CPUBUF | |
962 | }; | |
0ec8aa00 | 963 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 964 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 965 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
966 | #endif /* CONFIG_NUMA_BALANCING */ |
967 | ||
518cd623 PZ |
968 | #ifdef CONFIG_SMP |
969 | ||
e3fca9e7 PZ |
970 | static inline void |
971 | queue_balance_callback(struct rq *rq, | |
972 | struct callback_head *head, | |
973 | void (*func)(struct rq *rq)) | |
974 | { | |
975 | lockdep_assert_held(&rq->lock); | |
976 | ||
977 | if (unlikely(head->next)) | |
978 | return; | |
979 | ||
980 | head->func = (void (*)(struct callback_head *))func; | |
981 | head->next = rq->balance_callback; | |
982 | rq->balance_callback = head; | |
983 | } | |
984 | ||
e3baac47 PZ |
985 | extern void sched_ttwu_pending(void); |
986 | ||
029632fb PZ |
987 | #define rcu_dereference_check_sched_domain(p) \ |
988 | rcu_dereference_check((p), \ | |
989 | lockdep_is_held(&sched_domains_mutex)) | |
990 | ||
991 | /* | |
992 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
993 | * See detach_destroy_domains: synchronize_sched for details. | |
994 | * | |
995 | * The domain tree of any CPU may only be accessed from within | |
996 | * preempt-disabled sections. | |
997 | */ | |
998 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
999 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
1000 | __sd; __sd = __sd->parent) | |
029632fb | 1001 | |
77e81365 SS |
1002 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
1003 | ||
518cd623 PZ |
1004 | /** |
1005 | * highest_flag_domain - Return highest sched_domain containing flag. | |
1006 | * @cpu: The cpu whose highest level of sched domain is to | |
1007 | * be returned. | |
1008 | * @flag: The flag to check for the highest sched_domain | |
1009 | * for the given cpu. | |
1010 | * | |
1011 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
1012 | */ | |
1013 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
1014 | { | |
1015 | struct sched_domain *sd, *hsd = NULL; | |
1016 | ||
1017 | for_each_domain(cpu, sd) { | |
1018 | if (!(sd->flags & flag)) | |
1019 | break; | |
1020 | hsd = sd; | |
1021 | } | |
1022 | ||
1023 | return hsd; | |
1024 | } | |
1025 | ||
fb13c7ee MG |
1026 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
1027 | { | |
1028 | struct sched_domain *sd; | |
1029 | ||
1030 | for_each_domain(cpu, sd) { | |
1031 | if (sd->flags & flag) | |
1032 | break; | |
1033 | } | |
1034 | ||
1035 | return sd; | |
1036 | } | |
1037 | ||
518cd623 | 1038 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 1039 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 1040 | DECLARE_PER_CPU(int, sd_llc_id); |
0e369d75 | 1041 | DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
fb13c7ee | 1042 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 | 1043 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); |
518cd623 | 1044 | |
63b2ca30 | 1045 | struct sched_group_capacity { |
5e6521ea LZ |
1046 | atomic_t ref; |
1047 | /* | |
172895e6 | 1048 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
63b2ca30 | 1049 | * for a single CPU. |
5e6521ea | 1050 | */ |
bf475ce0 MR |
1051 | unsigned long capacity; |
1052 | unsigned long min_capacity; /* Min per-CPU capacity in group */ | |
5e6521ea | 1053 | unsigned long next_update; |
63b2ca30 | 1054 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea | 1055 | |
005f874d PZ |
1056 | #ifdef CONFIG_SCHED_DEBUG |
1057 | int id; | |
1058 | #endif | |
1059 | ||
e5c14b1f | 1060 | unsigned long cpumask[0]; /* balance mask */ |
5e6521ea LZ |
1061 | }; |
1062 | ||
1063 | struct sched_group { | |
1064 | struct sched_group *next; /* Must be a circular list */ | |
1065 | atomic_t ref; | |
1066 | ||
1067 | unsigned int group_weight; | |
63b2ca30 | 1068 | struct sched_group_capacity *sgc; |
afe06efd | 1069 | int asym_prefer_cpu; /* cpu of highest priority in group */ |
5e6521ea LZ |
1070 | |
1071 | /* | |
1072 | * The CPUs this group covers. | |
1073 | * | |
1074 | * NOTE: this field is variable length. (Allocated dynamically | |
1075 | * by attaching extra space to the end of the structure, | |
1076 | * depending on how many CPUs the kernel has booted up with) | |
1077 | */ | |
1078 | unsigned long cpumask[0]; | |
1079 | }; | |
1080 | ||
ae4df9d6 | 1081 | static inline struct cpumask *sched_group_span(struct sched_group *sg) |
5e6521ea LZ |
1082 | { |
1083 | return to_cpumask(sg->cpumask); | |
1084 | } | |
1085 | ||
1086 | /* | |
e5c14b1f | 1087 | * See build_balance_mask(). |
5e6521ea | 1088 | */ |
e5c14b1f | 1089 | static inline struct cpumask *group_balance_mask(struct sched_group *sg) |
5e6521ea | 1090 | { |
63b2ca30 | 1091 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
1092 | } |
1093 | ||
1094 | /** | |
1095 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
1096 | * @group: The group whose first cpu is to be returned. | |
1097 | */ | |
1098 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
1099 | { | |
ae4df9d6 | 1100 | return cpumask_first(sched_group_span(group)); |
5e6521ea LZ |
1101 | } |
1102 | ||
c1174876 PZ |
1103 | extern int group_balance_cpu(struct sched_group *sg); |
1104 | ||
3866e845 SRRH |
1105 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
1106 | void register_sched_domain_sysctl(void); | |
1107 | void unregister_sched_domain_sysctl(void); | |
1108 | #else | |
1109 | static inline void register_sched_domain_sysctl(void) | |
1110 | { | |
1111 | } | |
1112 | static inline void unregister_sched_domain_sysctl(void) | |
1113 | { | |
1114 | } | |
1115 | #endif | |
1116 | ||
e3baac47 PZ |
1117 | #else |
1118 | ||
1119 | static inline void sched_ttwu_pending(void) { } | |
1120 | ||
518cd623 | 1121 | #endif /* CONFIG_SMP */ |
029632fb | 1122 | |
391e43da | 1123 | #include "stats.h" |
1051408f | 1124 | #include "autogroup.h" |
029632fb PZ |
1125 | |
1126 | #ifdef CONFIG_CGROUP_SCHED | |
1127 | ||
1128 | /* | |
1129 | * Return the group to which this tasks belongs. | |
1130 | * | |
8af01f56 TH |
1131 | * We cannot use task_css() and friends because the cgroup subsystem |
1132 | * changes that value before the cgroup_subsys::attach() method is called, | |
1133 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
1134 | * |
1135 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
1136 | * core changes this before calling sched_move_task(). | |
1137 | * | |
1138 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
1139 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
1140 | */ |
1141 | static inline struct task_group *task_group(struct task_struct *p) | |
1142 | { | |
8323f26c | 1143 | return p->sched_task_group; |
029632fb PZ |
1144 | } |
1145 | ||
1146 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
1147 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
1148 | { | |
1149 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
1150 | struct task_group *tg = task_group(p); | |
1151 | #endif | |
1152 | ||
1153 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
ad936d86 | 1154 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
029632fb PZ |
1155 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
1156 | p->se.parent = tg->se[cpu]; | |
1157 | #endif | |
1158 | ||
1159 | #ifdef CONFIG_RT_GROUP_SCHED | |
1160 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
1161 | p->rt.parent = tg->rt_se[cpu]; | |
1162 | #endif | |
1163 | } | |
1164 | ||
1165 | #else /* CONFIG_CGROUP_SCHED */ | |
1166 | ||
1167 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
1168 | static inline struct task_group *task_group(struct task_struct *p) | |
1169 | { | |
1170 | return NULL; | |
1171 | } | |
1172 | ||
1173 | #endif /* CONFIG_CGROUP_SCHED */ | |
1174 | ||
1175 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1176 | { | |
1177 | set_task_rq(p, cpu); | |
1178 | #ifdef CONFIG_SMP | |
1179 | /* | |
1180 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1181 | * successfuly executed on another CPU. We must ensure that updates of | |
1182 | * per-task data have been completed by this moment. | |
1183 | */ | |
1184 | smp_wmb(); | |
c65eacbe AL |
1185 | #ifdef CONFIG_THREAD_INFO_IN_TASK |
1186 | p->cpu = cpu; | |
1187 | #else | |
029632fb | 1188 | task_thread_info(p)->cpu = cpu; |
c65eacbe | 1189 | #endif |
ac66f547 | 1190 | p->wake_cpu = cpu; |
029632fb PZ |
1191 | #endif |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
1196 | */ | |
1197 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 1198 | # include <linux/static_key.h> |
029632fb PZ |
1199 | # define const_debug __read_mostly |
1200 | #else | |
1201 | # define const_debug const | |
1202 | #endif | |
1203 | ||
1204 | extern const_debug unsigned int sysctl_sched_features; | |
1205 | ||
1206 | #define SCHED_FEAT(name, enabled) \ | |
1207 | __SCHED_FEAT_##name , | |
1208 | ||
1209 | enum { | |
391e43da | 1210 | #include "features.h" |
f8b6d1cc | 1211 | __SCHED_FEAT_NR, |
029632fb PZ |
1212 | }; |
1213 | ||
1214 | #undef SCHED_FEAT | |
1215 | ||
f8b6d1cc | 1216 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
f8b6d1cc | 1217 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1218 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1219 | { \ |
6e76ea8a | 1220 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1221 | } |
1222 | ||
1223 | #include "features.h" | |
1224 | ||
1225 | #undef SCHED_FEAT | |
1226 | ||
c5905afb | 1227 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc PZ |
1228 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
1229 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ | |
029632fb | 1230 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
f8b6d1cc | 1231 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 1232 | |
2a595721 | 1233 | extern struct static_key_false sched_numa_balancing; |
cb251765 | 1234 | extern struct static_key_false sched_schedstats; |
cbee9f88 | 1235 | |
029632fb PZ |
1236 | static inline u64 global_rt_period(void) |
1237 | { | |
1238 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1239 | } | |
1240 | ||
1241 | static inline u64 global_rt_runtime(void) | |
1242 | { | |
1243 | if (sysctl_sched_rt_runtime < 0) | |
1244 | return RUNTIME_INF; | |
1245 | ||
1246 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1247 | } | |
1248 | ||
029632fb PZ |
1249 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1250 | { | |
1251 | return rq->curr == p; | |
1252 | } | |
1253 | ||
1254 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1255 | { | |
1256 | #ifdef CONFIG_SMP | |
1257 | return p->on_cpu; | |
1258 | #else | |
1259 | return task_current(rq, p); | |
1260 | #endif | |
1261 | } | |
1262 | ||
da0c1e65 KT |
1263 | static inline int task_on_rq_queued(struct task_struct *p) |
1264 | { | |
1265 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1266 | } | |
029632fb | 1267 | |
cca26e80 KT |
1268 | static inline int task_on_rq_migrating(struct task_struct *p) |
1269 | { | |
1270 | return p->on_rq == TASK_ON_RQ_MIGRATING; | |
1271 | } | |
1272 | ||
029632fb PZ |
1273 | #ifndef prepare_arch_switch |
1274 | # define prepare_arch_switch(next) do { } while (0) | |
1275 | #endif | |
01f23e16 CM |
1276 | #ifndef finish_arch_post_lock_switch |
1277 | # define finish_arch_post_lock_switch() do { } while (0) | |
1278 | #endif | |
029632fb | 1279 | |
029632fb PZ |
1280 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
1281 | { | |
1282 | #ifdef CONFIG_SMP | |
1283 | /* | |
1284 | * We can optimise this out completely for !SMP, because the | |
1285 | * SMP rebalancing from interrupt is the only thing that cares | |
1286 | * here. | |
1287 | */ | |
1288 | next->on_cpu = 1; | |
1289 | #endif | |
1290 | } | |
1291 | ||
1292 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) | |
1293 | { | |
1294 | #ifdef CONFIG_SMP | |
1295 | /* | |
1296 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
1297 | * We must ensure this doesn't happen until the switch is completely | |
1298 | * finished. | |
95913d97 | 1299 | * |
b75a2253 PZ |
1300 | * In particular, the load of prev->state in finish_task_switch() must |
1301 | * happen before this. | |
1302 | * | |
1f03e8d2 | 1303 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). |
029632fb | 1304 | */ |
95913d97 | 1305 | smp_store_release(&prev->on_cpu, 0); |
029632fb PZ |
1306 | #endif |
1307 | #ifdef CONFIG_DEBUG_SPINLOCK | |
1308 | /* this is a valid case when another task releases the spinlock */ | |
1309 | rq->lock.owner = current; | |
1310 | #endif | |
1311 | /* | |
1312 | * If we are tracking spinlock dependencies then we have to | |
1313 | * fix up the runqueue lock - which gets 'carried over' from | |
1314 | * prev into current: | |
1315 | */ | |
1316 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
1317 | ||
1318 | raw_spin_unlock_irq(&rq->lock); | |
1319 | } | |
1320 | ||
b13095f0 LZ |
1321 | /* |
1322 | * wake flags | |
1323 | */ | |
1324 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1325 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1326 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1327 | ||
029632fb PZ |
1328 | /* |
1329 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1330 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1331 | * each task makes to its run queue's load is weighted according to its | |
1332 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1333 | * scaled version of the new time slice allocation that they receive on time | |
1334 | * slice expiry etc. | |
1335 | */ | |
1336 | ||
1337 | #define WEIGHT_IDLEPRIO 3 | |
1338 | #define WMULT_IDLEPRIO 1431655765 | |
1339 | ||
ed82b8a1 AK |
1340 | extern const int sched_prio_to_weight[40]; |
1341 | extern const u32 sched_prio_to_wmult[40]; | |
029632fb | 1342 | |
ff77e468 PZ |
1343 | /* |
1344 | * {de,en}queue flags: | |
1345 | * | |
1346 | * DEQUEUE_SLEEP - task is no longer runnable | |
1347 | * ENQUEUE_WAKEUP - task just became runnable | |
1348 | * | |
1349 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks | |
1350 | * are in a known state which allows modification. Such pairs | |
1351 | * should preserve as much state as possible. | |
1352 | * | |
1353 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location | |
1354 | * in the runqueue. | |
1355 | * | |
1356 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) | |
1357 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) | |
59efa0ba | 1358 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
ff77e468 PZ |
1359 | * |
1360 | */ | |
1361 | ||
1362 | #define DEQUEUE_SLEEP 0x01 | |
1363 | #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */ | |
1364 | #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */ | |
0a67d1ee | 1365 | #define DEQUEUE_NOCLOCK 0x08 /* matches ENQUEUE_NOCLOCK */ |
ff77e468 | 1366 | |
1de64443 | 1367 | #define ENQUEUE_WAKEUP 0x01 |
ff77e468 PZ |
1368 | #define ENQUEUE_RESTORE 0x02 |
1369 | #define ENQUEUE_MOVE 0x04 | |
0a67d1ee | 1370 | #define ENQUEUE_NOCLOCK 0x08 |
ff77e468 | 1371 | |
0a67d1ee PZ |
1372 | #define ENQUEUE_HEAD 0x10 |
1373 | #define ENQUEUE_REPLENISH 0x20 | |
c82ba9fa | 1374 | #ifdef CONFIG_SMP |
0a67d1ee | 1375 | #define ENQUEUE_MIGRATED 0x40 |
c82ba9fa | 1376 | #else |
59efa0ba | 1377 | #define ENQUEUE_MIGRATED 0x00 |
c82ba9fa | 1378 | #endif |
c82ba9fa | 1379 | |
37e117c0 PZ |
1380 | #define RETRY_TASK ((void *)-1UL) |
1381 | ||
c82ba9fa LZ |
1382 | struct sched_class { |
1383 | const struct sched_class *next; | |
1384 | ||
1385 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1386 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1387 | void (*yield_task) (struct rq *rq); | |
1388 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1389 | ||
1390 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1391 | ||
606dba2e PZ |
1392 | /* |
1393 | * It is the responsibility of the pick_next_task() method that will | |
1394 | * return the next task to call put_prev_task() on the @prev task or | |
1395 | * something equivalent. | |
37e117c0 PZ |
1396 | * |
1397 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1398 | * tasks. | |
606dba2e PZ |
1399 | */ |
1400 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
e7904a28 | 1401 | struct task_struct *prev, |
d8ac8971 | 1402 | struct rq_flags *rf); |
c82ba9fa LZ |
1403 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1404 | ||
1405 | #ifdef CONFIG_SMP | |
ac66f547 | 1406 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
5a4fd036 | 1407 | void (*migrate_task_rq)(struct task_struct *p); |
c82ba9fa | 1408 | |
c82ba9fa LZ |
1409 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); |
1410 | ||
1411 | void (*set_cpus_allowed)(struct task_struct *p, | |
1412 | const struct cpumask *newmask); | |
1413 | ||
1414 | void (*rq_online)(struct rq *rq); | |
1415 | void (*rq_offline)(struct rq *rq); | |
1416 | #endif | |
1417 | ||
1418 | void (*set_curr_task) (struct rq *rq); | |
1419 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1420 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1421 | void (*task_dead) (struct task_struct *p); |
c82ba9fa | 1422 | |
67dfa1b7 KT |
1423 | /* |
1424 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1425 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1426 | * rq->lock. They are however serialized by p->pi_lock. | |
1427 | */ | |
c82ba9fa LZ |
1428 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); |
1429 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1430 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1431 | int oldprio); | |
1432 | ||
1433 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1434 | struct task_struct *task); | |
1435 | ||
6e998916 SG |
1436 | void (*update_curr) (struct rq *rq); |
1437 | ||
ea86cb4b VG |
1438 | #define TASK_SET_GROUP 0 |
1439 | #define TASK_MOVE_GROUP 1 | |
1440 | ||
c82ba9fa | 1441 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b | 1442 | void (*task_change_group) (struct task_struct *p, int type); |
c82ba9fa LZ |
1443 | #endif |
1444 | }; | |
029632fb | 1445 | |
3f1d2a31 PZ |
1446 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1447 | { | |
1448 | prev->sched_class->put_prev_task(rq, prev); | |
1449 | } | |
1450 | ||
b2bf6c31 PZ |
1451 | static inline void set_curr_task(struct rq *rq, struct task_struct *curr) |
1452 | { | |
1453 | curr->sched_class->set_curr_task(rq); | |
1454 | } | |
1455 | ||
029632fb PZ |
1456 | #define sched_class_highest (&stop_sched_class) |
1457 | #define for_each_class(class) \ | |
1458 | for (class = sched_class_highest; class; class = class->next) | |
1459 | ||
1460 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1461 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1462 | extern const struct sched_class rt_sched_class; |
1463 | extern const struct sched_class fair_sched_class; | |
1464 | extern const struct sched_class idle_sched_class; | |
1465 | ||
1466 | ||
1467 | #ifdef CONFIG_SMP | |
1468 | ||
63b2ca30 | 1469 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1470 | |
7caff66f | 1471 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1472 | |
c5b28038 PZ |
1473 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
1474 | ||
029632fb PZ |
1475 | #endif |
1476 | ||
442bf3aa DL |
1477 | #ifdef CONFIG_CPU_IDLE |
1478 | static inline void idle_set_state(struct rq *rq, | |
1479 | struct cpuidle_state *idle_state) | |
1480 | { | |
1481 | rq->idle_state = idle_state; | |
1482 | } | |
1483 | ||
1484 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1485 | { | |
9148a3a1 | 1486 | SCHED_WARN_ON(!rcu_read_lock_held()); |
442bf3aa DL |
1487 | return rq->idle_state; |
1488 | } | |
1489 | #else | |
1490 | static inline void idle_set_state(struct rq *rq, | |
1491 | struct cpuidle_state *idle_state) | |
1492 | { | |
1493 | } | |
1494 | ||
1495 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1496 | { | |
1497 | return NULL; | |
1498 | } | |
1499 | #endif | |
1500 | ||
8663effb SRV |
1501 | extern void schedule_idle(void); |
1502 | ||
029632fb PZ |
1503 | extern void sysrq_sched_debug_show(void); |
1504 | extern void sched_init_granularity(void); | |
1505 | extern void update_max_interval(void); | |
1baca4ce JL |
1506 | |
1507 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1508 | extern void init_sched_rt_class(void); |
1509 | extern void init_sched_fair_class(void); | |
1510 | ||
8875125e | 1511 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1512 | extern void resched_cpu(int cpu); |
1513 | ||
1514 | extern struct rt_bandwidth def_rt_bandwidth; | |
1515 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1516 | ||
332ac17e DF |
1517 | extern struct dl_bandwidth def_dl_bandwidth; |
1518 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 | 1519 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
209a0cbd | 1520 | extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); |
4da3abce | 1521 | extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq); |
aab03e05 | 1522 | |
c52f14d3 LA |
1523 | #define BW_SHIFT 20 |
1524 | #define BW_UNIT (1 << BW_SHIFT) | |
4da3abce | 1525 | #define RATIO_SHIFT 8 |
332ac17e DF |
1526 | unsigned long to_ratio(u64 period, u64 runtime); |
1527 | ||
540247fb | 1528 | extern void init_entity_runnable_average(struct sched_entity *se); |
2b8c41da | 1529 | extern void post_init_entity_util_avg(struct sched_entity *se); |
a75cdaa9 | 1530 | |
76d92ac3 FW |
1531 | #ifdef CONFIG_NO_HZ_FULL |
1532 | extern bool sched_can_stop_tick(struct rq *rq); | |
1533 | ||
1534 | /* | |
1535 | * Tick may be needed by tasks in the runqueue depending on their policy and | |
1536 | * requirements. If tick is needed, lets send the target an IPI to kick it out of | |
1537 | * nohz mode if necessary. | |
1538 | */ | |
1539 | static inline void sched_update_tick_dependency(struct rq *rq) | |
1540 | { | |
1541 | int cpu; | |
1542 | ||
1543 | if (!tick_nohz_full_enabled()) | |
1544 | return; | |
1545 | ||
1546 | cpu = cpu_of(rq); | |
1547 | ||
1548 | if (!tick_nohz_full_cpu(cpu)) | |
1549 | return; | |
1550 | ||
1551 | if (sched_can_stop_tick(rq)) | |
1552 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1553 | else | |
1554 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1555 | } | |
1556 | #else | |
1557 | static inline void sched_update_tick_dependency(struct rq *rq) { } | |
1558 | #endif | |
1559 | ||
72465447 | 1560 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1561 | { |
72465447 KT |
1562 | unsigned prev_nr = rq->nr_running; |
1563 | ||
1564 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1565 | |
72465447 | 1566 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1567 | #ifdef CONFIG_SMP |
1568 | if (!rq->rd->overload) | |
1569 | rq->rd->overload = true; | |
1570 | #endif | |
4486edd1 | 1571 | } |
76d92ac3 FW |
1572 | |
1573 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1574 | } |
1575 | ||
72465447 | 1576 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1577 | { |
72465447 | 1578 | rq->nr_running -= count; |
76d92ac3 FW |
1579 | /* Check if we still need preemption */ |
1580 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1581 | } |
1582 | ||
265f22a9 FW |
1583 | static inline void rq_last_tick_reset(struct rq *rq) |
1584 | { | |
1585 | #ifdef CONFIG_NO_HZ_FULL | |
1586 | rq->last_sched_tick = jiffies; | |
1587 | #endif | |
1588 | } | |
1589 | ||
029632fb PZ |
1590 | extern void update_rq_clock(struct rq *rq); |
1591 | ||
1592 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1593 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1594 | ||
1595 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1596 | ||
1597 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1598 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1599 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1600 | ||
1601 | static inline u64 sched_avg_period(void) | |
1602 | { | |
1603 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1604 | } | |
1605 | ||
029632fb PZ |
1606 | #ifdef CONFIG_SCHED_HRTICK |
1607 | ||
1608 | /* | |
1609 | * Use hrtick when: | |
1610 | * - enabled by features | |
1611 | * - hrtimer is actually high res | |
1612 | */ | |
1613 | static inline int hrtick_enabled(struct rq *rq) | |
1614 | { | |
1615 | if (!sched_feat(HRTICK)) | |
1616 | return 0; | |
1617 | if (!cpu_active(cpu_of(rq))) | |
1618 | return 0; | |
1619 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1620 | } | |
1621 | ||
1622 | void hrtick_start(struct rq *rq, u64 delay); | |
1623 | ||
b39e66ea MG |
1624 | #else |
1625 | ||
1626 | static inline int hrtick_enabled(struct rq *rq) | |
1627 | { | |
1628 | return 0; | |
1629 | } | |
1630 | ||
029632fb PZ |
1631 | #endif /* CONFIG_SCHED_HRTICK */ |
1632 | ||
1633 | #ifdef CONFIG_SMP | |
1634 | extern void sched_avg_update(struct rq *rq); | |
dfbca41f PZ |
1635 | |
1636 | #ifndef arch_scale_freq_capacity | |
1637 | static __always_inline | |
1638 | unsigned long arch_scale_freq_capacity(struct sched_domain *sd, int cpu) | |
1639 | { | |
1640 | return SCHED_CAPACITY_SCALE; | |
1641 | } | |
1642 | #endif | |
b5b4860d | 1643 | |
8cd5601c MR |
1644 | #ifndef arch_scale_cpu_capacity |
1645 | static __always_inline | |
1646 | unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) | |
1647 | { | |
e3279a2e | 1648 | if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) |
8cd5601c MR |
1649 | return sd->smt_gain / sd->span_weight; |
1650 | ||
1651 | return SCHED_CAPACITY_SCALE; | |
1652 | } | |
1653 | #endif | |
1654 | ||
029632fb PZ |
1655 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) |
1656 | { | |
b5b4860d | 1657 | rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq)); |
029632fb PZ |
1658 | sched_avg_update(rq); |
1659 | } | |
1660 | #else | |
1661 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } | |
1662 | static inline void sched_avg_update(struct rq *rq) { } | |
1663 | #endif | |
1664 | ||
eb580751 | 1665 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 | 1666 | __acquires(rq->lock); |
8a8c69c3 | 1667 | |
eb580751 | 1668 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3960c8c0 | 1669 | __acquires(p->pi_lock) |
3e71a462 | 1670 | __acquires(rq->lock); |
3960c8c0 | 1671 | |
eb580751 | 1672 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) |
3960c8c0 PZ |
1673 | __releases(rq->lock) |
1674 | { | |
d8ac8971 | 1675 | rq_unpin_lock(rq, rf); |
3960c8c0 PZ |
1676 | raw_spin_unlock(&rq->lock); |
1677 | } | |
1678 | ||
1679 | static inline void | |
eb580751 | 1680 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) |
3960c8c0 PZ |
1681 | __releases(rq->lock) |
1682 | __releases(p->pi_lock) | |
1683 | { | |
d8ac8971 | 1684 | rq_unpin_lock(rq, rf); |
3960c8c0 | 1685 | raw_spin_unlock(&rq->lock); |
eb580751 | 1686 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3960c8c0 PZ |
1687 | } |
1688 | ||
8a8c69c3 PZ |
1689 | static inline void |
1690 | rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) | |
1691 | __acquires(rq->lock) | |
1692 | { | |
1693 | raw_spin_lock_irqsave(&rq->lock, rf->flags); | |
1694 | rq_pin_lock(rq, rf); | |
1695 | } | |
1696 | ||
1697 | static inline void | |
1698 | rq_lock_irq(struct rq *rq, struct rq_flags *rf) | |
1699 | __acquires(rq->lock) | |
1700 | { | |
1701 | raw_spin_lock_irq(&rq->lock); | |
1702 | rq_pin_lock(rq, rf); | |
1703 | } | |
1704 | ||
1705 | static inline void | |
1706 | rq_lock(struct rq *rq, struct rq_flags *rf) | |
1707 | __acquires(rq->lock) | |
1708 | { | |
1709 | raw_spin_lock(&rq->lock); | |
1710 | rq_pin_lock(rq, rf); | |
1711 | } | |
1712 | ||
1713 | static inline void | |
1714 | rq_relock(struct rq *rq, struct rq_flags *rf) | |
1715 | __acquires(rq->lock) | |
1716 | { | |
1717 | raw_spin_lock(&rq->lock); | |
1718 | rq_repin_lock(rq, rf); | |
1719 | } | |
1720 | ||
1721 | static inline void | |
1722 | rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) | |
1723 | __releases(rq->lock) | |
1724 | { | |
1725 | rq_unpin_lock(rq, rf); | |
1726 | raw_spin_unlock_irqrestore(&rq->lock, rf->flags); | |
1727 | } | |
1728 | ||
1729 | static inline void | |
1730 | rq_unlock_irq(struct rq *rq, struct rq_flags *rf) | |
1731 | __releases(rq->lock) | |
1732 | { | |
1733 | rq_unpin_lock(rq, rf); | |
1734 | raw_spin_unlock_irq(&rq->lock); | |
1735 | } | |
1736 | ||
1737 | static inline void | |
1738 | rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1739 | __releases(rq->lock) | |
1740 | { | |
1741 | rq_unpin_lock(rq, rf); | |
1742 | raw_spin_unlock(&rq->lock); | |
1743 | } | |
1744 | ||
029632fb PZ |
1745 | #ifdef CONFIG_SMP |
1746 | #ifdef CONFIG_PREEMPT | |
1747 | ||
1748 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1749 | ||
1750 | /* | |
1751 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1752 | * way at the expense of forcing extra atomic operations in all | |
1753 | * invocations. This assures that the double_lock is acquired using the | |
1754 | * same underlying policy as the spinlock_t on this architecture, which | |
1755 | * reduces latency compared to the unfair variant below. However, it | |
1756 | * also adds more overhead and therefore may reduce throughput. | |
1757 | */ | |
1758 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1759 | __releases(this_rq->lock) | |
1760 | __acquires(busiest->lock) | |
1761 | __acquires(this_rq->lock) | |
1762 | { | |
1763 | raw_spin_unlock(&this_rq->lock); | |
1764 | double_rq_lock(this_rq, busiest); | |
1765 | ||
1766 | return 1; | |
1767 | } | |
1768 | ||
1769 | #else | |
1770 | /* | |
1771 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1772 | * latency by eliminating extra atomic operations when the locks are | |
1773 | * already in proper order on entry. This favors lower cpu-ids and will | |
1774 | * grant the double lock to lower cpus over higher ids under contention, | |
1775 | * regardless of entry order into the function. | |
1776 | */ | |
1777 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1778 | __releases(this_rq->lock) | |
1779 | __acquires(busiest->lock) | |
1780 | __acquires(this_rq->lock) | |
1781 | { | |
1782 | int ret = 0; | |
1783 | ||
1784 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1785 | if (busiest < this_rq) { | |
1786 | raw_spin_unlock(&this_rq->lock); | |
1787 | raw_spin_lock(&busiest->lock); | |
1788 | raw_spin_lock_nested(&this_rq->lock, | |
1789 | SINGLE_DEPTH_NESTING); | |
1790 | ret = 1; | |
1791 | } else | |
1792 | raw_spin_lock_nested(&busiest->lock, | |
1793 | SINGLE_DEPTH_NESTING); | |
1794 | } | |
1795 | return ret; | |
1796 | } | |
1797 | ||
1798 | #endif /* CONFIG_PREEMPT */ | |
1799 | ||
1800 | /* | |
1801 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1802 | */ | |
1803 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1804 | { | |
1805 | if (unlikely(!irqs_disabled())) { | |
1806 | /* printk() doesn't work good under rq->lock */ | |
1807 | raw_spin_unlock(&this_rq->lock); | |
1808 | BUG_ON(1); | |
1809 | } | |
1810 | ||
1811 | return _double_lock_balance(this_rq, busiest); | |
1812 | } | |
1813 | ||
1814 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1815 | __releases(busiest->lock) | |
1816 | { | |
1817 | raw_spin_unlock(&busiest->lock); | |
1818 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1819 | } | |
1820 | ||
74602315 PZ |
1821 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1822 | { | |
1823 | if (l1 > l2) | |
1824 | swap(l1, l2); | |
1825 | ||
1826 | spin_lock(l1); | |
1827 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1828 | } | |
1829 | ||
60e69eed MG |
1830 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1831 | { | |
1832 | if (l1 > l2) | |
1833 | swap(l1, l2); | |
1834 | ||
1835 | spin_lock_irq(l1); | |
1836 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1837 | } | |
1838 | ||
74602315 PZ |
1839 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1840 | { | |
1841 | if (l1 > l2) | |
1842 | swap(l1, l2); | |
1843 | ||
1844 | raw_spin_lock(l1); | |
1845 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1846 | } | |
1847 | ||
029632fb PZ |
1848 | /* |
1849 | * double_rq_lock - safely lock two runqueues | |
1850 | * | |
1851 | * Note this does not disable interrupts like task_rq_lock, | |
1852 | * you need to do so manually before calling. | |
1853 | */ | |
1854 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1855 | __acquires(rq1->lock) | |
1856 | __acquires(rq2->lock) | |
1857 | { | |
1858 | BUG_ON(!irqs_disabled()); | |
1859 | if (rq1 == rq2) { | |
1860 | raw_spin_lock(&rq1->lock); | |
1861 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1862 | } else { | |
1863 | if (rq1 < rq2) { | |
1864 | raw_spin_lock(&rq1->lock); | |
1865 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1866 | } else { | |
1867 | raw_spin_lock(&rq2->lock); | |
1868 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1869 | } | |
1870 | } | |
1871 | } | |
1872 | ||
1873 | /* | |
1874 | * double_rq_unlock - safely unlock two runqueues | |
1875 | * | |
1876 | * Note this does not restore interrupts like task_rq_unlock, | |
1877 | * you need to do so manually after calling. | |
1878 | */ | |
1879 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1880 | __releases(rq1->lock) | |
1881 | __releases(rq2->lock) | |
1882 | { | |
1883 | raw_spin_unlock(&rq1->lock); | |
1884 | if (rq1 != rq2) | |
1885 | raw_spin_unlock(&rq2->lock); | |
1886 | else | |
1887 | __release(rq2->lock); | |
1888 | } | |
1889 | ||
f2cb1360 IM |
1890 | extern void set_rq_online (struct rq *rq); |
1891 | extern void set_rq_offline(struct rq *rq); | |
1892 | extern bool sched_smp_initialized; | |
1893 | ||
029632fb PZ |
1894 | #else /* CONFIG_SMP */ |
1895 | ||
1896 | /* | |
1897 | * double_rq_lock - safely lock two runqueues | |
1898 | * | |
1899 | * Note this does not disable interrupts like task_rq_lock, | |
1900 | * you need to do so manually before calling. | |
1901 | */ | |
1902 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1903 | __acquires(rq1->lock) | |
1904 | __acquires(rq2->lock) | |
1905 | { | |
1906 | BUG_ON(!irqs_disabled()); | |
1907 | BUG_ON(rq1 != rq2); | |
1908 | raw_spin_lock(&rq1->lock); | |
1909 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1910 | } | |
1911 | ||
1912 | /* | |
1913 | * double_rq_unlock - safely unlock two runqueues | |
1914 | * | |
1915 | * Note this does not restore interrupts like task_rq_unlock, | |
1916 | * you need to do so manually after calling. | |
1917 | */ | |
1918 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1919 | __releases(rq1->lock) | |
1920 | __releases(rq2->lock) | |
1921 | { | |
1922 | BUG_ON(rq1 != rq2); | |
1923 | raw_spin_unlock(&rq1->lock); | |
1924 | __release(rq2->lock); | |
1925 | } | |
1926 | ||
1927 | #endif | |
1928 | ||
1929 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1930 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
1931 | |
1932 | #ifdef CONFIG_SCHED_DEBUG | |
029632fb PZ |
1933 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
1934 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 1935 | extern void print_dl_stats(struct seq_file *m, int cpu); |
6b55c965 SD |
1936 | extern void |
1937 | print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); | |
397f2378 SD |
1938 | #ifdef CONFIG_NUMA_BALANCING |
1939 | extern void | |
1940 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
1941 | extern void | |
1942 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
1943 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
1944 | #endif /* CONFIG_NUMA_BALANCING */ | |
1945 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
1946 | |
1947 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
1948 | extern void init_rt_rq(struct rt_rq *rt_rq); |
1949 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 1950 | |
1ee14e6c BS |
1951 | extern void cfs_bandwidth_usage_inc(void); |
1952 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1953 | |
3451d024 | 1954 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
1955 | enum rq_nohz_flag_bits { |
1956 | NOHZ_TICK_STOPPED, | |
1957 | NOHZ_BALANCE_KICK, | |
1958 | }; | |
1959 | ||
1960 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
20a5c8cc TG |
1961 | |
1962 | extern void nohz_balance_exit_idle(unsigned int cpu); | |
1963 | #else | |
1964 | static inline void nohz_balance_exit_idle(unsigned int cpu) { } | |
1c792db7 | 1965 | #endif |
73fbec60 | 1966 | |
daec5798 LA |
1967 | |
1968 | #ifdef CONFIG_SMP | |
1969 | static inline | |
1970 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
1971 | { | |
1972 | struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); | |
1973 | int i; | |
1974 | ||
1975 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), | |
1976 | "sched RCU must be held"); | |
1977 | for_each_cpu_and(i, rd->span, cpu_active_mask) { | |
1978 | struct rq *rq = cpu_rq(i); | |
1979 | ||
1980 | rq->dl.extra_bw += bw; | |
1981 | } | |
1982 | } | |
1983 | #else | |
1984 | static inline | |
1985 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
1986 | { | |
1987 | struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); | |
1988 | ||
1989 | dl->extra_bw += bw; | |
1990 | } | |
1991 | #endif | |
1992 | ||
1993 | ||
73fbec60 | 1994 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
19d23dbf | 1995 | struct irqtime { |
25e2d8c1 | 1996 | u64 total; |
a499a5a1 | 1997 | u64 tick_delta; |
19d23dbf FW |
1998 | u64 irq_start_time; |
1999 | struct u64_stats_sync sync; | |
2000 | }; | |
73fbec60 | 2001 | |
19d23dbf | 2002 | DECLARE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 2003 | |
25e2d8c1 FW |
2004 | /* |
2005 | * Returns the irqtime minus the softirq time computed by ksoftirqd. | |
2006 | * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime | |
2007 | * and never move forward. | |
2008 | */ | |
73fbec60 FW |
2009 | static inline u64 irq_time_read(int cpu) |
2010 | { | |
19d23dbf FW |
2011 | struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); |
2012 | unsigned int seq; | |
2013 | u64 total; | |
73fbec60 FW |
2014 | |
2015 | do { | |
19d23dbf | 2016 | seq = __u64_stats_fetch_begin(&irqtime->sync); |
25e2d8c1 | 2017 | total = irqtime->total; |
19d23dbf | 2018 | } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); |
73fbec60 | 2019 | |
19d23dbf | 2020 | return total; |
73fbec60 | 2021 | } |
73fbec60 | 2022 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
adaf9fcd RW |
2023 | |
2024 | #ifdef CONFIG_CPU_FREQ | |
2025 | DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); | |
2026 | ||
2027 | /** | |
2028 | * cpufreq_update_util - Take a note about CPU utilization changes. | |
12bde33d | 2029 | * @rq: Runqueue to carry out the update for. |
58919e83 | 2030 | * @flags: Update reason flags. |
adaf9fcd | 2031 | * |
58919e83 RW |
2032 | * This function is called by the scheduler on the CPU whose utilization is |
2033 | * being updated. | |
adaf9fcd RW |
2034 | * |
2035 | * It can only be called from RCU-sched read-side critical sections. | |
adaf9fcd RW |
2036 | * |
2037 | * The way cpufreq is currently arranged requires it to evaluate the CPU | |
2038 | * performance state (frequency/voltage) on a regular basis to prevent it from | |
2039 | * being stuck in a completely inadequate performance level for too long. | |
2040 | * That is not guaranteed to happen if the updates are only triggered from CFS, | |
2041 | * though, because they may not be coming in if RT or deadline tasks are active | |
2042 | * all the time (or there are RT and DL tasks only). | |
2043 | * | |
2044 | * As a workaround for that issue, this function is called by the RT and DL | |
2045 | * sched classes to trigger extra cpufreq updates to prevent it from stalling, | |
2046 | * but that really is a band-aid. Going forward it should be replaced with | |
2047 | * solutions targeted more specifically at RT and DL tasks. | |
2048 | */ | |
12bde33d | 2049 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) |
adaf9fcd | 2050 | { |
58919e83 RW |
2051 | struct update_util_data *data; |
2052 | ||
2053 | data = rcu_dereference_sched(*this_cpu_ptr(&cpufreq_update_util_data)); | |
2054 | if (data) | |
12bde33d RW |
2055 | data->func(data, rq_clock(rq), flags); |
2056 | } | |
2057 | ||
2058 | static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) | |
2059 | { | |
2060 | if (cpu_of(rq) == smp_processor_id()) | |
2061 | cpufreq_update_util(rq, flags); | |
adaf9fcd RW |
2062 | } |
2063 | #else | |
12bde33d RW |
2064 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} |
2065 | static inline void cpufreq_update_this_cpu(struct rq *rq, unsigned int flags) {} | |
adaf9fcd | 2066 | #endif /* CONFIG_CPU_FREQ */ |
be53f58f | 2067 | |
9bdcb44e RW |
2068 | #ifdef arch_scale_freq_capacity |
2069 | #ifndef arch_scale_freq_invariant | |
2070 | #define arch_scale_freq_invariant() (true) | |
2071 | #endif | |
2072 | #else /* arch_scale_freq_capacity */ | |
2073 | #define arch_scale_freq_invariant() (false) | |
2074 | #endif |