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