Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
029632fb PZ |
2 | |
3 | #include <linux/sched.h> | |
dfc3401a | 4 | #include <linux/sched/autogroup.h> |
cf4aebc2 | 5 | #include <linux/sched/sysctl.h> |
105ab3d8 | 6 | #include <linux/sched/topology.h> |
8bd75c77 | 7 | #include <linux/sched/rt.h> |
ef8bd77f | 8 | #include <linux/sched/deadline.h> |
e6017571 | 9 | #include <linux/sched/clock.h> |
84f001e1 | 10 | #include <linux/sched/wake_q.h> |
3f07c014 | 11 | #include <linux/sched/signal.h> |
6a3827d7 | 12 | #include <linux/sched/numa_balancing.h> |
6e84f315 | 13 | #include <linux/sched/mm.h> |
55687da1 | 14 | #include <linux/sched/cpufreq.h> |
03441a34 | 15 | #include <linux/sched/stat.h> |
370c9135 | 16 | #include <linux/sched/nohz.h> |
b17b0153 | 17 | #include <linux/sched/debug.h> |
ef8bd77f | 18 | #include <linux/sched/hotplug.h> |
29930025 | 19 | #include <linux/sched/task.h> |
68db0cf1 | 20 | #include <linux/sched/task_stack.h> |
32ef5517 | 21 | #include <linux/sched/cputime.h> |
1777e463 | 22 | #include <linux/sched/init.h> |
ef8bd77f | 23 | |
19d23dbf | 24 | #include <linux/u64_stats_sync.h> |
a499a5a1 | 25 | #include <linux/kernel_stat.h> |
3866e845 | 26 | #include <linux/binfmts.h> |
029632fb PZ |
27 | #include <linux/mutex.h> |
28 | #include <linux/spinlock.h> | |
29 | #include <linux/stop_machine.h> | |
b6366f04 | 30 | #include <linux/irq_work.h> |
9f3660c2 | 31 | #include <linux/tick.h> |
f809ca9a | 32 | #include <linux/slab.h> |
d2cc5ed6 | 33 | #include <linux/cgroup.h> |
029632fb | 34 | |
7fce777c IM |
35 | #ifdef CONFIG_PARAVIRT |
36 | #include <asm/paravirt.h> | |
37 | #endif | |
38 | ||
391e43da | 39 | #include "cpupri.h" |
6bfd6d72 | 40 | #include "cpudeadline.h" |
029632fb | 41 | |
9148a3a1 | 42 | #ifdef CONFIG_SCHED_DEBUG |
6d3aed3d | 43 | # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) |
9148a3a1 | 44 | #else |
6d3aed3d | 45 | # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) |
9148a3a1 PZ |
46 | #endif |
47 | ||
45ceebf7 | 48 | struct rq; |
442bf3aa | 49 | struct cpuidle_state; |
45ceebf7 | 50 | |
da0c1e65 KT |
51 | /* task_struct::on_rq states: */ |
52 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 53 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 54 | |
029632fb PZ |
55 | extern __read_mostly int scheduler_running; |
56 | ||
45ceebf7 PG |
57 | extern unsigned long calc_load_update; |
58 | extern atomic_long_t calc_load_tasks; | |
59 | ||
3289bdb4 | 60 | extern void calc_global_load_tick(struct rq *this_rq); |
d60585c5 | 61 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
3289bdb4 PZ |
62 | |
63 | #ifdef CONFIG_SMP | |
cee1afce | 64 | extern void cpu_load_update_active(struct rq *this_rq); |
3289bdb4 | 65 | #else |
cee1afce | 66 | static inline void cpu_load_update_active(struct rq *this_rq) { } |
3289bdb4 | 67 | #endif |
45ceebf7 | 68 | |
029632fb PZ |
69 | /* |
70 | * Helpers for converting nanosecond timing to jiffy resolution | |
71 | */ | |
72 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
73 | ||
cc1f4b1f LZ |
74 | /* |
75 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
76 | * The extra resolution improves shares distribution and load balancing of | |
77 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
78 | * hierarchies, especially on larger systems. This is not a user-visible change | |
79 | * and does not change the user-interface for setting shares/weights. | |
80 | * | |
81 | * We increase resolution only if we have enough bits to allow this increased | |
2159197d PZ |
82 | * resolution (i.e. 64bit). The costs for increasing resolution when 32bit are |
83 | * pretty high and the returns do not justify the increased costs. | |
84 | * | |
85 | * Really only required when CONFIG_FAIR_GROUP_SCHED is also set, but to | |
86 | * increase coverage and consistency always enable it on 64bit platforms. | |
cc1f4b1f | 87 | */ |
2159197d | 88 | #ifdef CONFIG_64BIT |
172895e6 | 89 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
6ecdd749 YD |
90 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
91 | # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) | |
cc1f4b1f | 92 | #else |
172895e6 | 93 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
cc1f4b1f LZ |
94 | # define scale_load(w) (w) |
95 | # define scale_load_down(w) (w) | |
96 | #endif | |
97 | ||
6ecdd749 | 98 | /* |
172895e6 YD |
99 | * Task weight (visible to users) and its load (invisible to users) have |
100 | * independent resolution, but they should be well calibrated. We use | |
101 | * scale_load() and scale_load_down(w) to convert between them. The | |
102 | * following must be true: | |
103 | * | |
104 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD | |
105 | * | |
6ecdd749 | 106 | */ |
172895e6 | 107 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
029632fb | 108 | |
332ac17e DF |
109 | /* |
110 | * Single value that decides SCHED_DEADLINE internal math precision. | |
111 | * 10 -> just above 1us | |
112 | * 9 -> just above 0.5us | |
113 | */ | |
114 | #define DL_SCALE (10) | |
115 | ||
029632fb PZ |
116 | /* |
117 | * These are the 'tuning knobs' of the scheduler: | |
029632fb | 118 | */ |
029632fb PZ |
119 | |
120 | /* | |
121 | * single value that denotes runtime == period, ie unlimited time. | |
122 | */ | |
123 | #define RUNTIME_INF ((u64)~0ULL) | |
124 | ||
20f9cd2a HA |
125 | static inline int idle_policy(int policy) |
126 | { | |
127 | return policy == SCHED_IDLE; | |
128 | } | |
d50dde5a DF |
129 | static inline int fair_policy(int policy) |
130 | { | |
131 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
132 | } | |
133 | ||
029632fb PZ |
134 | static inline int rt_policy(int policy) |
135 | { | |
d50dde5a | 136 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
137 | } |
138 | ||
aab03e05 DF |
139 | static inline int dl_policy(int policy) |
140 | { | |
141 | return policy == SCHED_DEADLINE; | |
142 | } | |
20f9cd2a HA |
143 | static inline bool valid_policy(int policy) |
144 | { | |
145 | return idle_policy(policy) || fair_policy(policy) || | |
146 | rt_policy(policy) || dl_policy(policy); | |
147 | } | |
aab03e05 | 148 | |
029632fb PZ |
149 | static inline int task_has_rt_policy(struct task_struct *p) |
150 | { | |
151 | return rt_policy(p->policy); | |
152 | } | |
153 | ||
aab03e05 DF |
154 | static inline int task_has_dl_policy(struct task_struct *p) |
155 | { | |
156 | return dl_policy(p->policy); | |
157 | } | |
158 | ||
07881166 JL |
159 | #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) |
160 | ||
794a56eb JL |
161 | /* |
162 | * !! For sched_setattr_nocheck() (kernel) only !! | |
163 | * | |
164 | * This is actually gross. :( | |
165 | * | |
166 | * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE | |
167 | * tasks, but still be able to sleep. We need this on platforms that cannot | |
168 | * atomically change clock frequency. Remove once fast switching will be | |
169 | * available on such platforms. | |
170 | * | |
171 | * SUGOV stands for SchedUtil GOVernor. | |
172 | */ | |
173 | #define SCHED_FLAG_SUGOV 0x10000000 | |
174 | ||
175 | static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) | |
176 | { | |
177 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL | |
178 | return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); | |
179 | #else | |
180 | return false; | |
181 | #endif | |
182 | } | |
183 | ||
2d3d891d DF |
184 | /* |
185 | * Tells if entity @a should preempt entity @b. | |
186 | */ | |
332ac17e DF |
187 | static inline bool |
188 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d | 189 | { |
794a56eb JL |
190 | return dl_entity_is_special(a) || |
191 | dl_time_before(a->deadline, b->deadline); | |
2d3d891d DF |
192 | } |
193 | ||
029632fb PZ |
194 | /* |
195 | * This is the priority-queue data structure of the RT scheduling class: | |
196 | */ | |
197 | struct rt_prio_array { | |
198 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
199 | struct list_head queue[MAX_RT_PRIO]; | |
200 | }; | |
201 | ||
202 | struct rt_bandwidth { | |
203 | /* nests inside the rq lock: */ | |
204 | raw_spinlock_t rt_runtime_lock; | |
205 | ktime_t rt_period; | |
206 | u64 rt_runtime; | |
207 | struct hrtimer rt_period_timer; | |
4cfafd30 | 208 | unsigned int rt_period_active; |
029632fb | 209 | }; |
a5e7be3b JL |
210 | |
211 | void __dl_clear_params(struct task_struct *p); | |
212 | ||
332ac17e DF |
213 | /* |
214 | * To keep the bandwidth of -deadline tasks and groups under control | |
215 | * we need some place where: | |
216 | * - store the maximum -deadline bandwidth of the system (the group); | |
217 | * - cache the fraction of that bandwidth that is currently allocated. | |
218 | * | |
219 | * This is all done in the data structure below. It is similar to the | |
220 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
221 | * that, since here we are only interested in admission control, we | |
222 | * do not decrease any runtime while the group "executes", neither we | |
223 | * need a timer to replenish it. | |
224 | * | |
225 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
226 | * meaning that: | |
227 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
228 | * - dl_total_bw array contains, in the i-eth element, the currently | |
229 | * allocated bandwidth on the i-eth CPU. | |
230 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
231 | * consume bandwidth on the CPU they're running on. | |
232 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
233 | * that will be shown the next time the proc or cgroup controls will | |
234 | * be red. It on its turn can be changed by writing on its own | |
235 | * control. | |
236 | */ | |
237 | struct dl_bandwidth { | |
238 | raw_spinlock_t dl_runtime_lock; | |
239 | u64 dl_runtime; | |
240 | u64 dl_period; | |
241 | }; | |
242 | ||
243 | static inline int dl_bandwidth_enabled(void) | |
244 | { | |
1724813d | 245 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
246 | } |
247 | ||
332ac17e DF |
248 | struct dl_bw { |
249 | raw_spinlock_t lock; | |
250 | u64 bw, total_bw; | |
251 | }; | |
252 | ||
daec5798 LA |
253 | static inline void __dl_update(struct dl_bw *dl_b, s64 bw); |
254 | ||
7f51412a | 255 | static inline |
8c0944ce | 256 | void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
257 | { |
258 | dl_b->total_bw -= tsk_bw; | |
daec5798 | 259 | __dl_update(dl_b, (s32)tsk_bw / cpus); |
7f51412a JL |
260 | } |
261 | ||
262 | static inline | |
daec5798 | 263 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
264 | { |
265 | dl_b->total_bw += tsk_bw; | |
daec5798 | 266 | __dl_update(dl_b, -((s32)tsk_bw / cpus)); |
7f51412a JL |
267 | } |
268 | ||
269 | static inline | |
270 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
271 | { | |
272 | return dl_b->bw != -1 && | |
273 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
274 | } | |
275 | ||
209a0cbd | 276 | void dl_change_utilization(struct task_struct *p, u64 new_bw); |
f2cb1360 | 277 | extern void init_dl_bw(struct dl_bw *dl_b); |
06a76fe0 NP |
278 | extern int sched_dl_global_validate(void); |
279 | extern void sched_dl_do_global(void); | |
280 | extern int sched_dl_overflow(struct task_struct *p, int policy, | |
281 | const struct sched_attr *attr); | |
282 | extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); | |
283 | extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); | |
284 | extern bool __checkparam_dl(const struct sched_attr *attr); | |
06a76fe0 NP |
285 | extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); |
286 | extern int dl_task_can_attach(struct task_struct *p, | |
287 | const struct cpumask *cs_cpus_allowed); | |
288 | extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, | |
289 | const struct cpumask *trial); | |
290 | extern bool dl_cpu_busy(unsigned int cpu); | |
029632fb PZ |
291 | |
292 | #ifdef CONFIG_CGROUP_SCHED | |
293 | ||
294 | #include <linux/cgroup.h> | |
295 | ||
296 | struct cfs_rq; | |
297 | struct rt_rq; | |
298 | ||
35cf4e50 | 299 | extern struct list_head task_groups; |
029632fb PZ |
300 | |
301 | struct cfs_bandwidth { | |
302 | #ifdef CONFIG_CFS_BANDWIDTH | |
303 | raw_spinlock_t lock; | |
304 | ktime_t period; | |
305 | u64 quota, runtime; | |
9c58c79a | 306 | s64 hierarchical_quota; |
029632fb PZ |
307 | u64 runtime_expires; |
308 | ||
4cfafd30 | 309 | int idle, period_active; |
029632fb PZ |
310 | struct hrtimer period_timer, slack_timer; |
311 | struct list_head throttled_cfs_rq; | |
312 | ||
313 | /* statistics */ | |
314 | int nr_periods, nr_throttled; | |
315 | u64 throttled_time; | |
316 | #endif | |
317 | }; | |
318 | ||
319 | /* task group related information */ | |
320 | struct task_group { | |
321 | struct cgroup_subsys_state css; | |
322 | ||
323 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
324 | /* schedulable entities of this group on each cpu */ | |
325 | struct sched_entity **se; | |
326 | /* runqueue "owned" by this group on each cpu */ | |
327 | struct cfs_rq **cfs_rq; | |
328 | unsigned long shares; | |
329 | ||
fa6bddeb | 330 | #ifdef CONFIG_SMP |
b0367629 WL |
331 | /* |
332 | * load_avg can be heavily contended at clock tick time, so put | |
333 | * it in its own cacheline separated from the fields above which | |
334 | * will also be accessed at each tick. | |
335 | */ | |
336 | atomic_long_t load_avg ____cacheline_aligned; | |
029632fb | 337 | #endif |
fa6bddeb | 338 | #endif |
029632fb PZ |
339 | |
340 | #ifdef CONFIG_RT_GROUP_SCHED | |
341 | struct sched_rt_entity **rt_se; | |
342 | struct rt_rq **rt_rq; | |
343 | ||
344 | struct rt_bandwidth rt_bandwidth; | |
345 | #endif | |
346 | ||
347 | struct rcu_head rcu; | |
348 | struct list_head list; | |
349 | ||
350 | struct task_group *parent; | |
351 | struct list_head siblings; | |
352 | struct list_head children; | |
353 | ||
354 | #ifdef CONFIG_SCHED_AUTOGROUP | |
355 | struct autogroup *autogroup; | |
356 | #endif | |
357 | ||
358 | struct cfs_bandwidth cfs_bandwidth; | |
359 | }; | |
360 | ||
361 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
362 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
363 | ||
364 | /* | |
365 | * A weight of 0 or 1 can cause arithmetics problems. | |
366 | * A weight of a cfs_rq is the sum of weights of which entities | |
367 | * are queued on this cfs_rq, so a weight of a entity should not be | |
368 | * too large, so as the shares value of a task group. | |
369 | * (The default weight is 1024 - so there's no practical | |
370 | * limitation from this.) | |
371 | */ | |
372 | #define MIN_SHARES (1UL << 1) | |
373 | #define MAX_SHARES (1UL << 18) | |
374 | #endif | |
375 | ||
029632fb PZ |
376 | typedef int (*tg_visitor)(struct task_group *, void *); |
377 | ||
378 | extern int walk_tg_tree_from(struct task_group *from, | |
379 | tg_visitor down, tg_visitor up, void *data); | |
380 | ||
381 | /* | |
382 | * Iterate the full tree, calling @down when first entering a node and @up when | |
383 | * leaving it for the final time. | |
384 | * | |
385 | * Caller must hold rcu_lock or sufficient equivalent. | |
386 | */ | |
387 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
388 | { | |
389 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
390 | } | |
391 | ||
392 | extern int tg_nop(struct task_group *tg, void *data); | |
393 | ||
394 | extern void free_fair_sched_group(struct task_group *tg); | |
395 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
8663e24d | 396 | extern void online_fair_sched_group(struct task_group *tg); |
6fe1f348 | 397 | extern void unregister_fair_sched_group(struct task_group *tg); |
029632fb PZ |
398 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
399 | struct sched_entity *se, int cpu, | |
400 | struct sched_entity *parent); | |
401 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
029632fb PZ |
402 | |
403 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 404 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
405 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
406 | ||
407 | extern void free_rt_sched_group(struct task_group *tg); | |
408 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
409 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
410 | struct sched_rt_entity *rt_se, int cpu, | |
411 | struct sched_rt_entity *parent); | |
8887cd99 NP |
412 | extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); |
413 | extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); | |
414 | extern long sched_group_rt_runtime(struct task_group *tg); | |
415 | extern long sched_group_rt_period(struct task_group *tg); | |
416 | extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); | |
029632fb | 417 | |
25cc7da7 LZ |
418 | extern struct task_group *sched_create_group(struct task_group *parent); |
419 | extern void sched_online_group(struct task_group *tg, | |
420 | struct task_group *parent); | |
421 | extern void sched_destroy_group(struct task_group *tg); | |
422 | extern void sched_offline_group(struct task_group *tg); | |
423 | ||
424 | extern void sched_move_task(struct task_struct *tsk); | |
425 | ||
426 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
427 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
ad936d86 BP |
428 | |
429 | #ifdef CONFIG_SMP | |
430 | extern void set_task_rq_fair(struct sched_entity *se, | |
431 | struct cfs_rq *prev, struct cfs_rq *next); | |
432 | #else /* !CONFIG_SMP */ | |
433 | static inline void set_task_rq_fair(struct sched_entity *se, | |
434 | struct cfs_rq *prev, struct cfs_rq *next) { } | |
435 | #endif /* CONFIG_SMP */ | |
436 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
25cc7da7 | 437 | |
029632fb PZ |
438 | #else /* CONFIG_CGROUP_SCHED */ |
439 | ||
440 | struct cfs_bandwidth { }; | |
441 | ||
442 | #endif /* CONFIG_CGROUP_SCHED */ | |
443 | ||
444 | /* CFS-related fields in a runqueue */ | |
445 | struct cfs_rq { | |
446 | struct load_weight load; | |
1ea6c46a | 447 | unsigned long runnable_weight; |
c82513e5 | 448 | unsigned int nr_running, h_nr_running; |
029632fb PZ |
449 | |
450 | u64 exec_clock; | |
451 | u64 min_vruntime; | |
452 | #ifndef CONFIG_64BIT | |
453 | u64 min_vruntime_copy; | |
454 | #endif | |
455 | ||
bfb06889 | 456 | struct rb_root_cached tasks_timeline; |
029632fb | 457 | |
029632fb PZ |
458 | /* |
459 | * 'curr' points to currently running entity on this cfs_rq. | |
460 | * It is set to NULL otherwise (i.e when none are currently running). | |
461 | */ | |
462 | struct sched_entity *curr, *next, *last, *skip; | |
463 | ||
464 | #ifdef CONFIG_SCHED_DEBUG | |
465 | unsigned int nr_spread_over; | |
466 | #endif | |
467 | ||
2dac754e PT |
468 | #ifdef CONFIG_SMP |
469 | /* | |
9d89c257 | 470 | * CFS load tracking |
2dac754e | 471 | */ |
9d89c257 | 472 | struct sched_avg avg; |
2a2f5d4e PZ |
473 | #ifndef CONFIG_64BIT |
474 | u64 load_last_update_time_copy; | |
9d89c257 | 475 | #endif |
2a2f5d4e PZ |
476 | struct { |
477 | raw_spinlock_t lock ____cacheline_aligned; | |
478 | int nr; | |
479 | unsigned long load_avg; | |
480 | unsigned long util_avg; | |
0e2d2aaa | 481 | unsigned long runnable_sum; |
2a2f5d4e | 482 | } removed; |
82958366 | 483 | |
9d89c257 | 484 | #ifdef CONFIG_FAIR_GROUP_SCHED |
0e2d2aaa PZ |
485 | unsigned long tg_load_avg_contrib; |
486 | long propagate; | |
487 | long prop_runnable_sum; | |
488 | ||
82958366 PT |
489 | /* |
490 | * h_load = weight * f(tg) | |
491 | * | |
492 | * Where f(tg) is the recursive weight fraction assigned to | |
493 | * this group. | |
494 | */ | |
495 | unsigned long h_load; | |
68520796 VD |
496 | u64 last_h_load_update; |
497 | struct sched_entity *h_load_next; | |
498 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
82958366 PT |
499 | #endif /* CONFIG_SMP */ |
500 | ||
029632fb PZ |
501 | #ifdef CONFIG_FAIR_GROUP_SCHED |
502 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
503 | ||
504 | /* | |
505 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
506 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
507 | * (like users, containers etc.) | |
508 | * | |
509 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
510 | * list is used during load balance. | |
511 | */ | |
512 | int on_list; | |
513 | struct list_head leaf_cfs_rq_list; | |
514 | struct task_group *tg; /* group that "owns" this runqueue */ | |
515 | ||
029632fb PZ |
516 | #ifdef CONFIG_CFS_BANDWIDTH |
517 | int runtime_enabled; | |
518 | u64 runtime_expires; | |
519 | s64 runtime_remaining; | |
520 | ||
f1b17280 PT |
521 | u64 throttled_clock, throttled_clock_task; |
522 | u64 throttled_clock_task_time; | |
55e16d30 | 523 | int throttled, throttle_count; |
029632fb PZ |
524 | struct list_head throttled_list; |
525 | #endif /* CONFIG_CFS_BANDWIDTH */ | |
526 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
527 | }; | |
528 | ||
529 | static inline int rt_bandwidth_enabled(void) | |
530 | { | |
531 | return sysctl_sched_rt_runtime >= 0; | |
532 | } | |
533 | ||
b6366f04 | 534 | /* RT IPI pull logic requires IRQ_WORK */ |
4bdced5c | 535 | #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) |
b6366f04 SR |
536 | # define HAVE_RT_PUSH_IPI |
537 | #endif | |
538 | ||
029632fb PZ |
539 | /* Real-Time classes' related field in a runqueue: */ |
540 | struct rt_rq { | |
541 | struct rt_prio_array active; | |
c82513e5 | 542 | unsigned int rt_nr_running; |
01d36d0a | 543 | unsigned int rr_nr_running; |
029632fb PZ |
544 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
545 | struct { | |
546 | int curr; /* highest queued rt task prio */ | |
547 | #ifdef CONFIG_SMP | |
548 | int next; /* next highest */ | |
549 | #endif | |
550 | } highest_prio; | |
551 | #endif | |
552 | #ifdef CONFIG_SMP | |
553 | unsigned long rt_nr_migratory; | |
554 | unsigned long rt_nr_total; | |
555 | int overloaded; | |
556 | struct plist_head pushable_tasks; | |
b6366f04 | 557 | #endif /* CONFIG_SMP */ |
f4ebcbc0 KT |
558 | int rt_queued; |
559 | ||
029632fb PZ |
560 | int rt_throttled; |
561 | u64 rt_time; | |
562 | u64 rt_runtime; | |
563 | /* Nests inside the rq lock: */ | |
564 | raw_spinlock_t rt_runtime_lock; | |
565 | ||
566 | #ifdef CONFIG_RT_GROUP_SCHED | |
567 | unsigned long rt_nr_boosted; | |
568 | ||
569 | struct rq *rq; | |
029632fb PZ |
570 | struct task_group *tg; |
571 | #endif | |
572 | }; | |
573 | ||
aab03e05 DF |
574 | /* Deadline class' related fields in a runqueue */ |
575 | struct dl_rq { | |
576 | /* runqueue is an rbtree, ordered by deadline */ | |
2161573e | 577 | struct rb_root_cached root; |
aab03e05 DF |
578 | |
579 | unsigned long dl_nr_running; | |
1baca4ce JL |
580 | |
581 | #ifdef CONFIG_SMP | |
582 | /* | |
583 | * Deadline values of the currently executing and the | |
584 | * earliest ready task on this rq. Caching these facilitates | |
585 | * the decision wether or not a ready but not running task | |
586 | * should migrate somewhere else. | |
587 | */ | |
588 | struct { | |
589 | u64 curr; | |
590 | u64 next; | |
591 | } earliest_dl; | |
592 | ||
593 | unsigned long dl_nr_migratory; | |
1baca4ce JL |
594 | int overloaded; |
595 | ||
596 | /* | |
597 | * Tasks on this rq that can be pushed away. They are kept in | |
598 | * an rb-tree, ordered by tasks' deadlines, with caching | |
599 | * of the leftmost (earliest deadline) element. | |
600 | */ | |
2161573e | 601 | struct rb_root_cached pushable_dl_tasks_root; |
332ac17e DF |
602 | #else |
603 | struct dl_bw dl_bw; | |
1baca4ce | 604 | #endif |
e36d8677 LA |
605 | /* |
606 | * "Active utilization" for this runqueue: increased when a | |
607 | * task wakes up (becomes TASK_RUNNING) and decreased when a | |
608 | * task blocks | |
609 | */ | |
610 | u64 running_bw; | |
4da3abce | 611 | |
8fd27231 LA |
612 | /* |
613 | * Utilization of the tasks "assigned" to this runqueue (including | |
614 | * the tasks that are in runqueue and the tasks that executed on this | |
615 | * CPU and blocked). Increased when a task moves to this runqueue, and | |
616 | * decreased when the task moves away (migrates, changes scheduling | |
617 | * policy, or terminates). | |
618 | * This is needed to compute the "inactive utilization" for the | |
619 | * runqueue (inactive utilization = this_bw - running_bw). | |
620 | */ | |
621 | u64 this_bw; | |
daec5798 | 622 | u64 extra_bw; |
8fd27231 | 623 | |
4da3abce LA |
624 | /* |
625 | * Inverse of the fraction of CPU utilization that can be reclaimed | |
626 | * by the GRUB algorithm. | |
627 | */ | |
628 | u64 bw_ratio; | |
aab03e05 DF |
629 | }; |
630 | ||
029632fb PZ |
631 | #ifdef CONFIG_SMP |
632 | ||
afe06efd TC |
633 | static inline bool sched_asym_prefer(int a, int b) |
634 | { | |
635 | return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); | |
636 | } | |
637 | ||
029632fb PZ |
638 | /* |
639 | * We add the notion of a root-domain which will be used to define per-domain | |
640 | * variables. Each exclusive cpuset essentially defines an island domain by | |
641 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
642 | * exclusive cpuset is created, we also create and attach a new root-domain | |
643 | * object. | |
644 | * | |
645 | */ | |
646 | struct root_domain { | |
647 | atomic_t refcount; | |
648 | atomic_t rto_count; | |
649 | struct rcu_head rcu; | |
650 | cpumask_var_t span; | |
651 | cpumask_var_t online; | |
652 | ||
4486edd1 TC |
653 | /* Indicate more than one runnable task for any CPU */ |
654 | bool overload; | |
655 | ||
1baca4ce JL |
656 | /* |
657 | * The bit corresponding to a CPU gets set here if such CPU has more | |
658 | * than one runnable -deadline task (as it is below for RT tasks). | |
659 | */ | |
660 | cpumask_var_t dlo_mask; | |
661 | atomic_t dlo_count; | |
332ac17e | 662 | struct dl_bw dl_bw; |
6bfd6d72 | 663 | struct cpudl cpudl; |
1baca4ce | 664 | |
4bdced5c SRRH |
665 | #ifdef HAVE_RT_PUSH_IPI |
666 | /* | |
667 | * For IPI pull requests, loop across the rto_mask. | |
668 | */ | |
669 | struct irq_work rto_push_work; | |
670 | raw_spinlock_t rto_lock; | |
671 | /* These are only updated and read within rto_lock */ | |
672 | int rto_loop; | |
673 | int rto_cpu; | |
674 | /* These atomics are updated outside of a lock */ | |
675 | atomic_t rto_loop_next; | |
676 | atomic_t rto_loop_start; | |
677 | #endif | |
029632fb PZ |
678 | /* |
679 | * The "RT overload" flag: it gets set if a CPU has more than | |
680 | * one runnable RT task. | |
681 | */ | |
682 | cpumask_var_t rto_mask; | |
683 | struct cpupri cpupri; | |
cd92bfd3 DE |
684 | |
685 | unsigned long max_cpu_capacity; | |
029632fb PZ |
686 | }; |
687 | ||
688 | extern struct root_domain def_root_domain; | |
f2cb1360 | 689 | extern struct mutex sched_domains_mutex; |
f2cb1360 IM |
690 | |
691 | extern void init_defrootdomain(void); | |
8d5dc512 | 692 | extern int sched_init_domains(const struct cpumask *cpu_map); |
f2cb1360 | 693 | extern void rq_attach_root(struct rq *rq, struct root_domain *rd); |
364f5665 SRV |
694 | extern void sched_get_rd(struct root_domain *rd); |
695 | extern void sched_put_rd(struct root_domain *rd); | |
029632fb | 696 | |
4bdced5c SRRH |
697 | #ifdef HAVE_RT_PUSH_IPI |
698 | extern void rto_push_irq_work_func(struct irq_work *work); | |
699 | #endif | |
029632fb PZ |
700 | #endif /* CONFIG_SMP */ |
701 | ||
702 | /* | |
703 | * This is the main, per-CPU runqueue data structure. | |
704 | * | |
705 | * Locking rule: those places that want to lock multiple runqueues | |
706 | * (such as the load balancing or the thread migration code), lock | |
707 | * acquire operations must be ordered by ascending &runqueue. | |
708 | */ | |
709 | struct rq { | |
710 | /* runqueue lock: */ | |
711 | raw_spinlock_t lock; | |
712 | ||
713 | /* | |
714 | * nr_running and cpu_load should be in the same cacheline because | |
715 | * remote CPUs use both these fields when doing load calculation. | |
716 | */ | |
c82513e5 | 717 | unsigned int nr_running; |
0ec8aa00 PZ |
718 | #ifdef CONFIG_NUMA_BALANCING |
719 | unsigned int nr_numa_running; | |
720 | unsigned int nr_preferred_running; | |
721 | #endif | |
029632fb PZ |
722 | #define CPU_LOAD_IDX_MAX 5 |
723 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
3451d024 | 724 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 FW |
725 | #ifdef CONFIG_SMP |
726 | unsigned long last_load_update_tick; | |
727 | #endif /* CONFIG_SMP */ | |
1c792db7 | 728 | unsigned long nohz_flags; |
9fd81dd5 | 729 | #endif /* CONFIG_NO_HZ_COMMON */ |
dcdedb24 | 730 | |
029632fb PZ |
731 | /* capture load from *all* tasks on this cpu: */ |
732 | struct load_weight load; | |
733 | unsigned long nr_load_updates; | |
734 | u64 nr_switches; | |
735 | ||
736 | struct cfs_rq cfs; | |
737 | struct rt_rq rt; | |
aab03e05 | 738 | struct dl_rq dl; |
029632fb PZ |
739 | |
740 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
741 | /* list of leaf cfs_rq on this cpu: */ | |
742 | struct list_head leaf_cfs_rq_list; | |
9c2791f9 | 743 | struct list_head *tmp_alone_branch; |
a35b6466 PZ |
744 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
745 | ||
029632fb PZ |
746 | /* |
747 | * This is part of a global counter where only the total sum | |
748 | * over all CPUs matters. A task can increase this counter on | |
749 | * one CPU and if it got migrated afterwards it may decrease | |
750 | * it on another CPU. Always updated under the runqueue lock: | |
751 | */ | |
752 | unsigned long nr_uninterruptible; | |
753 | ||
754 | struct task_struct *curr, *idle, *stop; | |
755 | unsigned long next_balance; | |
756 | struct mm_struct *prev_mm; | |
757 | ||
cb42c9a3 | 758 | unsigned int clock_update_flags; |
029632fb PZ |
759 | u64 clock; |
760 | u64 clock_task; | |
761 | ||
762 | atomic_t nr_iowait; | |
763 | ||
764 | #ifdef CONFIG_SMP | |
765 | struct root_domain *rd; | |
766 | struct sched_domain *sd; | |
767 | ||
ced549fa | 768 | unsigned long cpu_capacity; |
ca6d75e6 | 769 | unsigned long cpu_capacity_orig; |
029632fb | 770 | |
e3fca9e7 PZ |
771 | struct callback_head *balance_callback; |
772 | ||
029632fb PZ |
773 | unsigned char idle_balance; |
774 | /* For active balancing */ | |
029632fb PZ |
775 | int active_balance; |
776 | int push_cpu; | |
777 | struct cpu_stop_work active_balance_work; | |
778 | /* cpu of this runqueue: */ | |
779 | int cpu; | |
780 | int online; | |
781 | ||
367456c7 PZ |
782 | struct list_head cfs_tasks; |
783 | ||
029632fb PZ |
784 | u64 rt_avg; |
785 | u64 age_stamp; | |
786 | u64 idle_stamp; | |
787 | u64 avg_idle; | |
9bd721c5 JL |
788 | |
789 | /* This is used to determine avg_idle's max value */ | |
790 | u64 max_idle_balance_cost; | |
029632fb PZ |
791 | #endif |
792 | ||
793 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
794 | u64 prev_irq_time; | |
795 | #endif | |
796 | #ifdef CONFIG_PARAVIRT | |
797 | u64 prev_steal_time; | |
798 | #endif | |
799 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
800 | u64 prev_steal_time_rq; | |
801 | #endif | |
802 | ||
803 | /* calc_load related fields */ | |
804 | unsigned long calc_load_update; | |
805 | long calc_load_active; | |
806 | ||
807 | #ifdef CONFIG_SCHED_HRTICK | |
808 | #ifdef CONFIG_SMP | |
809 | int hrtick_csd_pending; | |
966a9671 | 810 | call_single_data_t hrtick_csd; |
029632fb PZ |
811 | #endif |
812 | struct hrtimer hrtick_timer; | |
813 | #endif | |
814 | ||
815 | #ifdef CONFIG_SCHEDSTATS | |
816 | /* latency stats */ | |
817 | struct sched_info rq_sched_info; | |
818 | unsigned long long rq_cpu_time; | |
819 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
820 | ||
821 | /* sys_sched_yield() stats */ | |
822 | unsigned int yld_count; | |
823 | ||
824 | /* schedule() stats */ | |
029632fb PZ |
825 | unsigned int sched_count; |
826 | unsigned int sched_goidle; | |
827 | ||
828 | /* try_to_wake_up() stats */ | |
829 | unsigned int ttwu_count; | |
830 | unsigned int ttwu_local; | |
831 | #endif | |
832 | ||
833 | #ifdef CONFIG_SMP | |
834 | struct llist_head wake_list; | |
835 | #endif | |
442bf3aa DL |
836 | |
837 | #ifdef CONFIG_CPU_IDLE | |
838 | /* Must be inspected within a rcu lock section */ | |
839 | struct cpuidle_state *idle_state; | |
840 | #endif | |
029632fb PZ |
841 | }; |
842 | ||
843 | static inline int cpu_of(struct rq *rq) | |
844 | { | |
845 | #ifdef CONFIG_SMP | |
846 | return rq->cpu; | |
847 | #else | |
848 | return 0; | |
849 | #endif | |
850 | } | |
851 | ||
1b568f0a PZ |
852 | |
853 | #ifdef CONFIG_SCHED_SMT | |
854 | ||
855 | extern struct static_key_false sched_smt_present; | |
856 | ||
857 | extern void __update_idle_core(struct rq *rq); | |
858 | ||
859 | static inline void update_idle_core(struct rq *rq) | |
860 | { | |
861 | if (static_branch_unlikely(&sched_smt_present)) | |
862 | __update_idle_core(rq); | |
863 | } | |
864 | ||
865 | #else | |
866 | static inline void update_idle_core(struct rq *rq) { } | |
867 | #endif | |
868 | ||
8b06c55b | 869 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 870 | |
518cd623 | 871 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 872 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
873 | #define task_rq(p) cpu_rq(task_cpu(p)) |
874 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 875 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 876 | |
cebde6d6 PZ |
877 | static inline u64 __rq_clock_broken(struct rq *rq) |
878 | { | |
316c1608 | 879 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
880 | } |
881 | ||
cb42c9a3 MF |
882 | /* |
883 | * rq::clock_update_flags bits | |
884 | * | |
885 | * %RQCF_REQ_SKIP - will request skipping of clock update on the next | |
886 | * call to __schedule(). This is an optimisation to avoid | |
887 | * neighbouring rq clock updates. | |
888 | * | |
889 | * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is | |
890 | * in effect and calls to update_rq_clock() are being ignored. | |
891 | * | |
892 | * %RQCF_UPDATED - is a debug flag that indicates whether a call has been | |
893 | * made to update_rq_clock() since the last time rq::lock was pinned. | |
894 | * | |
895 | * If inside of __schedule(), clock_update_flags will have been | |
896 | * shifted left (a left shift is a cheap operation for the fast path | |
897 | * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, | |
898 | * | |
899 | * if (rq-clock_update_flags >= RQCF_UPDATED) | |
900 | * | |
901 | * to check if %RQCF_UPADTED is set. It'll never be shifted more than | |
902 | * one position though, because the next rq_unpin_lock() will shift it | |
903 | * back. | |
904 | */ | |
905 | #define RQCF_REQ_SKIP 0x01 | |
906 | #define RQCF_ACT_SKIP 0x02 | |
907 | #define RQCF_UPDATED 0x04 | |
908 | ||
909 | static inline void assert_clock_updated(struct rq *rq) | |
910 | { | |
911 | /* | |
912 | * The only reason for not seeing a clock update since the | |
913 | * last rq_pin_lock() is if we're currently skipping updates. | |
914 | */ | |
915 | SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); | |
916 | } | |
917 | ||
78becc27 FW |
918 | static inline u64 rq_clock(struct rq *rq) |
919 | { | |
cebde6d6 | 920 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
921 | assert_clock_updated(rq); |
922 | ||
78becc27 FW |
923 | return rq->clock; |
924 | } | |
925 | ||
926 | static inline u64 rq_clock_task(struct rq *rq) | |
927 | { | |
cebde6d6 | 928 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
929 | assert_clock_updated(rq); |
930 | ||
78becc27 FW |
931 | return rq->clock_task; |
932 | } | |
933 | ||
9edfbfed PZ |
934 | static inline void rq_clock_skip_update(struct rq *rq, bool skip) |
935 | { | |
936 | lockdep_assert_held(&rq->lock); | |
937 | if (skip) | |
cb42c9a3 | 938 | rq->clock_update_flags |= RQCF_REQ_SKIP; |
9edfbfed | 939 | else |
cb42c9a3 | 940 | rq->clock_update_flags &= ~RQCF_REQ_SKIP; |
9edfbfed PZ |
941 | } |
942 | ||
d8ac8971 MF |
943 | struct rq_flags { |
944 | unsigned long flags; | |
945 | struct pin_cookie cookie; | |
cb42c9a3 MF |
946 | #ifdef CONFIG_SCHED_DEBUG |
947 | /* | |
948 | * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the | |
949 | * current pin context is stashed here in case it needs to be | |
950 | * restored in rq_repin_lock(). | |
951 | */ | |
952 | unsigned int clock_update_flags; | |
953 | #endif | |
d8ac8971 MF |
954 | }; |
955 | ||
956 | static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) | |
957 | { | |
958 | rf->cookie = lockdep_pin_lock(&rq->lock); | |
cb42c9a3 MF |
959 | |
960 | #ifdef CONFIG_SCHED_DEBUG | |
961 | rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); | |
962 | rf->clock_update_flags = 0; | |
963 | #endif | |
d8ac8971 MF |
964 | } |
965 | ||
966 | static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) | |
967 | { | |
cb42c9a3 MF |
968 | #ifdef CONFIG_SCHED_DEBUG |
969 | if (rq->clock_update_flags > RQCF_ACT_SKIP) | |
970 | rf->clock_update_flags = RQCF_UPDATED; | |
971 | #endif | |
972 | ||
d8ac8971 MF |
973 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
974 | } | |
975 | ||
976 | static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) | |
977 | { | |
978 | lockdep_repin_lock(&rq->lock, rf->cookie); | |
cb42c9a3 MF |
979 | |
980 | #ifdef CONFIG_SCHED_DEBUG | |
981 | /* | |
982 | * Restore the value we stashed in @rf for this pin context. | |
983 | */ | |
984 | rq->clock_update_flags |= rf->clock_update_flags; | |
985 | #endif | |
d8ac8971 MF |
986 | } |
987 | ||
9942f79b | 988 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
989 | enum numa_topology_type { |
990 | NUMA_DIRECT, | |
991 | NUMA_GLUELESS_MESH, | |
992 | NUMA_BACKPLANE, | |
993 | }; | |
994 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
995 | extern int sched_max_numa_distance; |
996 | extern bool find_numa_distance(int distance); | |
997 | #endif | |
998 | ||
f2cb1360 IM |
999 | #ifdef CONFIG_NUMA |
1000 | extern void sched_init_numa(void); | |
1001 | extern void sched_domains_numa_masks_set(unsigned int cpu); | |
1002 | extern void sched_domains_numa_masks_clear(unsigned int cpu); | |
1003 | #else | |
1004 | static inline void sched_init_numa(void) { } | |
1005 | static inline void sched_domains_numa_masks_set(unsigned int cpu) { } | |
1006 | static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
1007 | #endif | |
1008 | ||
f809ca9a | 1009 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
1010 | /* The regions in numa_faults array from task_struct */ |
1011 | enum numa_faults_stats { | |
1012 | NUMA_MEM = 0, | |
1013 | NUMA_CPU, | |
1014 | NUMA_MEMBUF, | |
1015 | NUMA_CPUBUF | |
1016 | }; | |
0ec8aa00 | 1017 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 1018 | extern int migrate_task_to(struct task_struct *p, int cpu); |
ac66f547 | 1019 | extern int migrate_swap(struct task_struct *, struct task_struct *); |
f809ca9a MG |
1020 | #endif /* CONFIG_NUMA_BALANCING */ |
1021 | ||
518cd623 PZ |
1022 | #ifdef CONFIG_SMP |
1023 | ||
e3fca9e7 PZ |
1024 | static inline void |
1025 | queue_balance_callback(struct rq *rq, | |
1026 | struct callback_head *head, | |
1027 | void (*func)(struct rq *rq)) | |
1028 | { | |
1029 | lockdep_assert_held(&rq->lock); | |
1030 | ||
1031 | if (unlikely(head->next)) | |
1032 | return; | |
1033 | ||
1034 | head->func = (void (*)(struct callback_head *))func; | |
1035 | head->next = rq->balance_callback; | |
1036 | rq->balance_callback = head; | |
1037 | } | |
1038 | ||
e3baac47 PZ |
1039 | extern void sched_ttwu_pending(void); |
1040 | ||
029632fb PZ |
1041 | #define rcu_dereference_check_sched_domain(p) \ |
1042 | rcu_dereference_check((p), \ | |
1043 | lockdep_is_held(&sched_domains_mutex)) | |
1044 | ||
1045 | /* | |
1046 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1047 | * See detach_destroy_domains: synchronize_sched for details. | |
1048 | * | |
1049 | * The domain tree of any CPU may only be accessed from within | |
1050 | * preempt-disabled sections. | |
1051 | */ | |
1052 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
1053 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
1054 | __sd; __sd = __sd->parent) | |
029632fb | 1055 | |
77e81365 SS |
1056 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
1057 | ||
518cd623 PZ |
1058 | /** |
1059 | * highest_flag_domain - Return highest sched_domain containing flag. | |
1060 | * @cpu: The cpu whose highest level of sched domain is to | |
1061 | * be returned. | |
1062 | * @flag: The flag to check for the highest sched_domain | |
1063 | * for the given cpu. | |
1064 | * | |
1065 | * Returns the highest sched_domain of a cpu which contains the given flag. | |
1066 | */ | |
1067 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
1068 | { | |
1069 | struct sched_domain *sd, *hsd = NULL; | |
1070 | ||
1071 | for_each_domain(cpu, sd) { | |
1072 | if (!(sd->flags & flag)) | |
1073 | break; | |
1074 | hsd = sd; | |
1075 | } | |
1076 | ||
1077 | return hsd; | |
1078 | } | |
1079 | ||
fb13c7ee MG |
1080 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
1081 | { | |
1082 | struct sched_domain *sd; | |
1083 | ||
1084 | for_each_domain(cpu, sd) { | |
1085 | if (sd->flags & flag) | |
1086 | break; | |
1087 | } | |
1088 | ||
1089 | return sd; | |
1090 | } | |
1091 | ||
518cd623 | 1092 | DECLARE_PER_CPU(struct sched_domain *, sd_llc); |
7d9ffa89 | 1093 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 1094 | DECLARE_PER_CPU(int, sd_llc_id); |
0e369d75 | 1095 | DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
fb13c7ee | 1096 | DECLARE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 | 1097 | DECLARE_PER_CPU(struct sched_domain *, sd_asym); |
518cd623 | 1098 | |
63b2ca30 | 1099 | struct sched_group_capacity { |
5e6521ea LZ |
1100 | atomic_t ref; |
1101 | /* | |
172895e6 | 1102 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
63b2ca30 | 1103 | * for a single CPU. |
5e6521ea | 1104 | */ |
bf475ce0 MR |
1105 | unsigned long capacity; |
1106 | unsigned long min_capacity; /* Min per-CPU capacity in group */ | |
5e6521ea | 1107 | unsigned long next_update; |
63b2ca30 | 1108 | int imbalance; /* XXX unrelated to capacity but shared group state */ |
5e6521ea | 1109 | |
005f874d PZ |
1110 | #ifdef CONFIG_SCHED_DEBUG |
1111 | int id; | |
1112 | #endif | |
1113 | ||
e5c14b1f | 1114 | unsigned long cpumask[0]; /* balance mask */ |
5e6521ea LZ |
1115 | }; |
1116 | ||
1117 | struct sched_group { | |
1118 | struct sched_group *next; /* Must be a circular list */ | |
1119 | atomic_t ref; | |
1120 | ||
1121 | unsigned int group_weight; | |
63b2ca30 | 1122 | struct sched_group_capacity *sgc; |
afe06efd | 1123 | int asym_prefer_cpu; /* cpu of highest priority in group */ |
5e6521ea LZ |
1124 | |
1125 | /* | |
1126 | * The CPUs this group covers. | |
1127 | * | |
1128 | * NOTE: this field is variable length. (Allocated dynamically | |
1129 | * by attaching extra space to the end of the structure, | |
1130 | * depending on how many CPUs the kernel has booted up with) | |
1131 | */ | |
1132 | unsigned long cpumask[0]; | |
1133 | }; | |
1134 | ||
ae4df9d6 | 1135 | static inline struct cpumask *sched_group_span(struct sched_group *sg) |
5e6521ea LZ |
1136 | { |
1137 | return to_cpumask(sg->cpumask); | |
1138 | } | |
1139 | ||
1140 | /* | |
e5c14b1f | 1141 | * See build_balance_mask(). |
5e6521ea | 1142 | */ |
e5c14b1f | 1143 | static inline struct cpumask *group_balance_mask(struct sched_group *sg) |
5e6521ea | 1144 | { |
63b2ca30 | 1145 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
1146 | } |
1147 | ||
1148 | /** | |
1149 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
1150 | * @group: The group whose first cpu is to be returned. | |
1151 | */ | |
1152 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
1153 | { | |
ae4df9d6 | 1154 | return cpumask_first(sched_group_span(group)); |
5e6521ea LZ |
1155 | } |
1156 | ||
c1174876 PZ |
1157 | extern int group_balance_cpu(struct sched_group *sg); |
1158 | ||
3866e845 SRRH |
1159 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
1160 | void register_sched_domain_sysctl(void); | |
bbdacdfe | 1161 | void dirty_sched_domain_sysctl(int cpu); |
3866e845 SRRH |
1162 | void unregister_sched_domain_sysctl(void); |
1163 | #else | |
1164 | static inline void register_sched_domain_sysctl(void) | |
1165 | { | |
1166 | } | |
bbdacdfe PZ |
1167 | static inline void dirty_sched_domain_sysctl(int cpu) |
1168 | { | |
1169 | } | |
3866e845 SRRH |
1170 | static inline void unregister_sched_domain_sysctl(void) |
1171 | { | |
1172 | } | |
1173 | #endif | |
1174 | ||
e3baac47 PZ |
1175 | #else |
1176 | ||
1177 | static inline void sched_ttwu_pending(void) { } | |
1178 | ||
518cd623 | 1179 | #endif /* CONFIG_SMP */ |
029632fb | 1180 | |
391e43da | 1181 | #include "stats.h" |
1051408f | 1182 | #include "autogroup.h" |
029632fb PZ |
1183 | |
1184 | #ifdef CONFIG_CGROUP_SCHED | |
1185 | ||
1186 | /* | |
1187 | * Return the group to which this tasks belongs. | |
1188 | * | |
8af01f56 TH |
1189 | * We cannot use task_css() and friends because the cgroup subsystem |
1190 | * changes that value before the cgroup_subsys::attach() method is called, | |
1191 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
1192 | * |
1193 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
1194 | * core changes this before calling sched_move_task(). | |
1195 | * | |
1196 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
1197 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
1198 | */ |
1199 | static inline struct task_group *task_group(struct task_struct *p) | |
1200 | { | |
8323f26c | 1201 | return p->sched_task_group; |
029632fb PZ |
1202 | } |
1203 | ||
1204 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
1205 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
1206 | { | |
1207 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
1208 | struct task_group *tg = task_group(p); | |
1209 | #endif | |
1210 | ||
1211 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
ad936d86 | 1212 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
029632fb PZ |
1213 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
1214 | p->se.parent = tg->se[cpu]; | |
1215 | #endif | |
1216 | ||
1217 | #ifdef CONFIG_RT_GROUP_SCHED | |
1218 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
1219 | p->rt.parent = tg->rt_se[cpu]; | |
1220 | #endif | |
1221 | } | |
1222 | ||
1223 | #else /* CONFIG_CGROUP_SCHED */ | |
1224 | ||
1225 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
1226 | static inline struct task_group *task_group(struct task_struct *p) | |
1227 | { | |
1228 | return NULL; | |
1229 | } | |
1230 | ||
1231 | #endif /* CONFIG_CGROUP_SCHED */ | |
1232 | ||
1233 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1234 | { | |
1235 | set_task_rq(p, cpu); | |
1236 | #ifdef CONFIG_SMP | |
1237 | /* | |
1238 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1239 | * successfuly executed on another CPU. We must ensure that updates of | |
1240 | * per-task data have been completed by this moment. | |
1241 | */ | |
1242 | smp_wmb(); | |
c65eacbe AL |
1243 | #ifdef CONFIG_THREAD_INFO_IN_TASK |
1244 | p->cpu = cpu; | |
1245 | #else | |
029632fb | 1246 | task_thread_info(p)->cpu = cpu; |
c65eacbe | 1247 | #endif |
ac66f547 | 1248 | p->wake_cpu = cpu; |
029632fb PZ |
1249 | #endif |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
1254 | */ | |
1255 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 1256 | # include <linux/static_key.h> |
029632fb PZ |
1257 | # define const_debug __read_mostly |
1258 | #else | |
1259 | # define const_debug const | |
1260 | #endif | |
1261 | ||
029632fb PZ |
1262 | #define SCHED_FEAT(name, enabled) \ |
1263 | __SCHED_FEAT_##name , | |
1264 | ||
1265 | enum { | |
391e43da | 1266 | #include "features.h" |
f8b6d1cc | 1267 | __SCHED_FEAT_NR, |
029632fb PZ |
1268 | }; |
1269 | ||
1270 | #undef SCHED_FEAT | |
1271 | ||
f8b6d1cc | 1272 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
765cc3a4 PB |
1273 | |
1274 | /* | |
1275 | * To support run-time toggling of sched features, all the translation units | |
1276 | * (but core.c) reference the sysctl_sched_features defined in core.c. | |
1277 | */ | |
1278 | extern const_debug unsigned int sysctl_sched_features; | |
1279 | ||
f8b6d1cc | 1280 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1281 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1282 | { \ |
6e76ea8a | 1283 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1284 | } |
1285 | ||
1286 | #include "features.h" | |
f8b6d1cc PZ |
1287 | #undef SCHED_FEAT |
1288 | ||
c5905afb | 1289 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc | 1290 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
765cc3a4 | 1291 | |
f8b6d1cc | 1292 | #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ |
765cc3a4 PB |
1293 | |
1294 | /* | |
1295 | * Each translation unit has its own copy of sysctl_sched_features to allow | |
1296 | * constants propagation at compile time and compiler optimization based on | |
1297 | * features default. | |
1298 | */ | |
1299 | #define SCHED_FEAT(name, enabled) \ | |
1300 | (1UL << __SCHED_FEAT_##name) * enabled | | |
1301 | static const_debug __maybe_unused unsigned int sysctl_sched_features = | |
1302 | #include "features.h" | |
1303 | 0; | |
1304 | #undef SCHED_FEAT | |
1305 | ||
029632fb | 1306 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
765cc3a4 | 1307 | |
f8b6d1cc | 1308 | #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ |
029632fb | 1309 | |
2a595721 | 1310 | extern struct static_key_false sched_numa_balancing; |
cb251765 | 1311 | extern struct static_key_false sched_schedstats; |
cbee9f88 | 1312 | |
029632fb PZ |
1313 | static inline u64 global_rt_period(void) |
1314 | { | |
1315 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1316 | } | |
1317 | ||
1318 | static inline u64 global_rt_runtime(void) | |
1319 | { | |
1320 | if (sysctl_sched_rt_runtime < 0) | |
1321 | return RUNTIME_INF; | |
1322 | ||
1323 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1324 | } | |
1325 | ||
029632fb PZ |
1326 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1327 | { | |
1328 | return rq->curr == p; | |
1329 | } | |
1330 | ||
1331 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1332 | { | |
1333 | #ifdef CONFIG_SMP | |
1334 | return p->on_cpu; | |
1335 | #else | |
1336 | return task_current(rq, p); | |
1337 | #endif | |
1338 | } | |
1339 | ||
da0c1e65 KT |
1340 | static inline int task_on_rq_queued(struct task_struct *p) |
1341 | { | |
1342 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1343 | } | |
029632fb | 1344 | |
cca26e80 KT |
1345 | static inline int task_on_rq_migrating(struct task_struct *p) |
1346 | { | |
1347 | return p->on_rq == TASK_ON_RQ_MIGRATING; | |
1348 | } | |
1349 | ||
029632fb PZ |
1350 | #ifndef prepare_arch_switch |
1351 | # define prepare_arch_switch(next) do { } while (0) | |
1352 | #endif | |
01f23e16 CM |
1353 | #ifndef finish_arch_post_lock_switch |
1354 | # define finish_arch_post_lock_switch() do { } while (0) | |
1355 | #endif | |
029632fb | 1356 | |
b13095f0 LZ |
1357 | /* |
1358 | * wake flags | |
1359 | */ | |
1360 | #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */ | |
1361 | #define WF_FORK 0x02 /* child wakeup after fork */ | |
1362 | #define WF_MIGRATED 0x4 /* internal use, task got migrated */ | |
1363 | ||
029632fb PZ |
1364 | /* |
1365 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1366 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1367 | * each task makes to its run queue's load is weighted according to its | |
1368 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1369 | * scaled version of the new time slice allocation that they receive on time | |
1370 | * slice expiry etc. | |
1371 | */ | |
1372 | ||
1373 | #define WEIGHT_IDLEPRIO 3 | |
1374 | #define WMULT_IDLEPRIO 1431655765 | |
1375 | ||
ed82b8a1 AK |
1376 | extern const int sched_prio_to_weight[40]; |
1377 | extern const u32 sched_prio_to_wmult[40]; | |
029632fb | 1378 | |
ff77e468 PZ |
1379 | /* |
1380 | * {de,en}queue flags: | |
1381 | * | |
1382 | * DEQUEUE_SLEEP - task is no longer runnable | |
1383 | * ENQUEUE_WAKEUP - task just became runnable | |
1384 | * | |
1385 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks | |
1386 | * are in a known state which allows modification. Such pairs | |
1387 | * should preserve as much state as possible. | |
1388 | * | |
1389 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location | |
1390 | * in the runqueue. | |
1391 | * | |
1392 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) | |
1393 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) | |
59efa0ba | 1394 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
ff77e468 PZ |
1395 | * |
1396 | */ | |
1397 | ||
1398 | #define DEQUEUE_SLEEP 0x01 | |
1399 | #define DEQUEUE_SAVE 0x02 /* matches ENQUEUE_RESTORE */ | |
1400 | #define DEQUEUE_MOVE 0x04 /* matches ENQUEUE_MOVE */ | |
0a67d1ee | 1401 | #define DEQUEUE_NOCLOCK 0x08 /* matches ENQUEUE_NOCLOCK */ |
ff77e468 | 1402 | |
1de64443 | 1403 | #define ENQUEUE_WAKEUP 0x01 |
ff77e468 PZ |
1404 | #define ENQUEUE_RESTORE 0x02 |
1405 | #define ENQUEUE_MOVE 0x04 | |
0a67d1ee | 1406 | #define ENQUEUE_NOCLOCK 0x08 |
ff77e468 | 1407 | |
0a67d1ee PZ |
1408 | #define ENQUEUE_HEAD 0x10 |
1409 | #define ENQUEUE_REPLENISH 0x20 | |
c82ba9fa | 1410 | #ifdef CONFIG_SMP |
0a67d1ee | 1411 | #define ENQUEUE_MIGRATED 0x40 |
c82ba9fa | 1412 | #else |
59efa0ba | 1413 | #define ENQUEUE_MIGRATED 0x00 |
c82ba9fa | 1414 | #endif |
c82ba9fa | 1415 | |
37e117c0 PZ |
1416 | #define RETRY_TASK ((void *)-1UL) |
1417 | ||
c82ba9fa LZ |
1418 | struct sched_class { |
1419 | const struct sched_class *next; | |
1420 | ||
1421 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1422 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
1423 | void (*yield_task) (struct rq *rq); | |
1424 | bool (*yield_to_task) (struct rq *rq, struct task_struct *p, bool preempt); | |
1425 | ||
1426 | void (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags); | |
1427 | ||
606dba2e PZ |
1428 | /* |
1429 | * It is the responsibility of the pick_next_task() method that will | |
1430 | * return the next task to call put_prev_task() on the @prev task or | |
1431 | * something equivalent. | |
37e117c0 PZ |
1432 | * |
1433 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1434 | * tasks. | |
606dba2e PZ |
1435 | */ |
1436 | struct task_struct * (*pick_next_task) (struct rq *rq, | |
e7904a28 | 1437 | struct task_struct *prev, |
d8ac8971 | 1438 | struct rq_flags *rf); |
c82ba9fa LZ |
1439 | void (*put_prev_task) (struct rq *rq, struct task_struct *p); |
1440 | ||
1441 | #ifdef CONFIG_SMP | |
ac66f547 | 1442 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
5a4fd036 | 1443 | void (*migrate_task_rq)(struct task_struct *p); |
c82ba9fa | 1444 | |
c82ba9fa LZ |
1445 | void (*task_woken) (struct rq *this_rq, struct task_struct *task); |
1446 | ||
1447 | void (*set_cpus_allowed)(struct task_struct *p, | |
1448 | const struct cpumask *newmask); | |
1449 | ||
1450 | void (*rq_online)(struct rq *rq); | |
1451 | void (*rq_offline)(struct rq *rq); | |
1452 | #endif | |
1453 | ||
1454 | void (*set_curr_task) (struct rq *rq); | |
1455 | void (*task_tick) (struct rq *rq, struct task_struct *p, int queued); | |
1456 | void (*task_fork) (struct task_struct *p); | |
e6c390f2 | 1457 | void (*task_dead) (struct task_struct *p); |
c82ba9fa | 1458 | |
67dfa1b7 KT |
1459 | /* |
1460 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1461 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1462 | * rq->lock. They are however serialized by p->pi_lock. | |
1463 | */ | |
c82ba9fa LZ |
1464 | void (*switched_from) (struct rq *this_rq, struct task_struct *task); |
1465 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
1466 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, | |
1467 | int oldprio); | |
1468 | ||
1469 | unsigned int (*get_rr_interval) (struct rq *rq, | |
1470 | struct task_struct *task); | |
1471 | ||
6e998916 SG |
1472 | void (*update_curr) (struct rq *rq); |
1473 | ||
ea86cb4b VG |
1474 | #define TASK_SET_GROUP 0 |
1475 | #define TASK_MOVE_GROUP 1 | |
1476 | ||
c82ba9fa | 1477 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b | 1478 | void (*task_change_group) (struct task_struct *p, int type); |
c82ba9fa LZ |
1479 | #endif |
1480 | }; | |
029632fb | 1481 | |
3f1d2a31 PZ |
1482 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1483 | { | |
1484 | prev->sched_class->put_prev_task(rq, prev); | |
1485 | } | |
1486 | ||
b2bf6c31 PZ |
1487 | static inline void set_curr_task(struct rq *rq, struct task_struct *curr) |
1488 | { | |
1489 | curr->sched_class->set_curr_task(rq); | |
1490 | } | |
1491 | ||
f5832c19 | 1492 | #ifdef CONFIG_SMP |
029632fb | 1493 | #define sched_class_highest (&stop_sched_class) |
f5832c19 NP |
1494 | #else |
1495 | #define sched_class_highest (&dl_sched_class) | |
1496 | #endif | |
029632fb PZ |
1497 | #define for_each_class(class) \ |
1498 | for (class = sched_class_highest; class; class = class->next) | |
1499 | ||
1500 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1501 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1502 | extern const struct sched_class rt_sched_class; |
1503 | extern const struct sched_class fair_sched_class; | |
1504 | extern const struct sched_class idle_sched_class; | |
1505 | ||
1506 | ||
1507 | #ifdef CONFIG_SMP | |
1508 | ||
63b2ca30 | 1509 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1510 | |
7caff66f | 1511 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1512 | |
c5b28038 PZ |
1513 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
1514 | ||
029632fb PZ |
1515 | #endif |
1516 | ||
442bf3aa DL |
1517 | #ifdef CONFIG_CPU_IDLE |
1518 | static inline void idle_set_state(struct rq *rq, | |
1519 | struct cpuidle_state *idle_state) | |
1520 | { | |
1521 | rq->idle_state = idle_state; | |
1522 | } | |
1523 | ||
1524 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1525 | { | |
9148a3a1 | 1526 | SCHED_WARN_ON(!rcu_read_lock_held()); |
442bf3aa DL |
1527 | return rq->idle_state; |
1528 | } | |
1529 | #else | |
1530 | static inline void idle_set_state(struct rq *rq, | |
1531 | struct cpuidle_state *idle_state) | |
1532 | { | |
1533 | } | |
1534 | ||
1535 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1536 | { | |
1537 | return NULL; | |
1538 | } | |
1539 | #endif | |
1540 | ||
8663effb SRV |
1541 | extern void schedule_idle(void); |
1542 | ||
029632fb PZ |
1543 | extern void sysrq_sched_debug_show(void); |
1544 | extern void sched_init_granularity(void); | |
1545 | extern void update_max_interval(void); | |
1baca4ce JL |
1546 | |
1547 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1548 | extern void init_sched_rt_class(void); |
1549 | extern void init_sched_fair_class(void); | |
1550 | ||
9059393e VG |
1551 | extern void reweight_task(struct task_struct *p, int prio); |
1552 | ||
8875125e | 1553 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1554 | extern void resched_cpu(int cpu); |
1555 | ||
1556 | extern struct rt_bandwidth def_rt_bandwidth; | |
1557 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1558 | ||
332ac17e DF |
1559 | extern struct dl_bandwidth def_dl_bandwidth; |
1560 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 | 1561 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
209a0cbd | 1562 | extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); |
4da3abce | 1563 | extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq); |
aab03e05 | 1564 | |
c52f14d3 LA |
1565 | #define BW_SHIFT 20 |
1566 | #define BW_UNIT (1 << BW_SHIFT) | |
4da3abce | 1567 | #define RATIO_SHIFT 8 |
332ac17e DF |
1568 | unsigned long to_ratio(u64 period, u64 runtime); |
1569 | ||
540247fb | 1570 | extern void init_entity_runnable_average(struct sched_entity *se); |
2b8c41da | 1571 | extern void post_init_entity_util_avg(struct sched_entity *se); |
a75cdaa9 | 1572 | |
76d92ac3 FW |
1573 | #ifdef CONFIG_NO_HZ_FULL |
1574 | extern bool sched_can_stop_tick(struct rq *rq); | |
d84b3131 | 1575 | extern int __init sched_tick_offload_init(void); |
76d92ac3 FW |
1576 | |
1577 | /* | |
1578 | * Tick may be needed by tasks in the runqueue depending on their policy and | |
1579 | * requirements. If tick is needed, lets send the target an IPI to kick it out of | |
1580 | * nohz mode if necessary. | |
1581 | */ | |
1582 | static inline void sched_update_tick_dependency(struct rq *rq) | |
1583 | { | |
1584 | int cpu; | |
1585 | ||
1586 | if (!tick_nohz_full_enabled()) | |
1587 | return; | |
1588 | ||
1589 | cpu = cpu_of(rq); | |
1590 | ||
1591 | if (!tick_nohz_full_cpu(cpu)) | |
1592 | return; | |
1593 | ||
1594 | if (sched_can_stop_tick(rq)) | |
1595 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1596 | else | |
1597 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1598 | } | |
1599 | #else | |
d84b3131 | 1600 | static inline int sched_tick_offload_init(void) { return 0; } |
76d92ac3 FW |
1601 | static inline void sched_update_tick_dependency(struct rq *rq) { } |
1602 | #endif | |
1603 | ||
72465447 | 1604 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1605 | { |
72465447 KT |
1606 | unsigned prev_nr = rq->nr_running; |
1607 | ||
1608 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1609 | |
72465447 | 1610 | if (prev_nr < 2 && rq->nr_running >= 2) { |
4486edd1 TC |
1611 | #ifdef CONFIG_SMP |
1612 | if (!rq->rd->overload) | |
1613 | rq->rd->overload = true; | |
1614 | #endif | |
4486edd1 | 1615 | } |
76d92ac3 FW |
1616 | |
1617 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1618 | } |
1619 | ||
72465447 | 1620 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1621 | { |
72465447 | 1622 | rq->nr_running -= count; |
76d92ac3 FW |
1623 | /* Check if we still need preemption */ |
1624 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1625 | } |
1626 | ||
1627 | extern void update_rq_clock(struct rq *rq); | |
1628 | ||
1629 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); | |
1630 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1631 | ||
1632 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1633 | ||
1634 | extern const_debug unsigned int sysctl_sched_time_avg; | |
1635 | extern const_debug unsigned int sysctl_sched_nr_migrate; | |
1636 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1637 | ||
1638 | static inline u64 sched_avg_period(void) | |
1639 | { | |
1640 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1641 | } | |
1642 | ||
029632fb PZ |
1643 | #ifdef CONFIG_SCHED_HRTICK |
1644 | ||
1645 | /* | |
1646 | * Use hrtick when: | |
1647 | * - enabled by features | |
1648 | * - hrtimer is actually high res | |
1649 | */ | |
1650 | static inline int hrtick_enabled(struct rq *rq) | |
1651 | { | |
1652 | if (!sched_feat(HRTICK)) | |
1653 | return 0; | |
1654 | if (!cpu_active(cpu_of(rq))) | |
1655 | return 0; | |
1656 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1657 | } | |
1658 | ||
1659 | void hrtick_start(struct rq *rq, u64 delay); | |
1660 | ||
b39e66ea MG |
1661 | #else |
1662 | ||
1663 | static inline int hrtick_enabled(struct rq *rq) | |
1664 | { | |
1665 | return 0; | |
1666 | } | |
1667 | ||
029632fb PZ |
1668 | #endif /* CONFIG_SCHED_HRTICK */ |
1669 | ||
dfbca41f PZ |
1670 | #ifndef arch_scale_freq_capacity |
1671 | static __always_inline | |
7673c8a4 | 1672 | unsigned long arch_scale_freq_capacity(int cpu) |
dfbca41f PZ |
1673 | { |
1674 | return SCHED_CAPACITY_SCALE; | |
1675 | } | |
1676 | #endif | |
b5b4860d | 1677 | |
7e1a9208 JL |
1678 | #ifdef CONFIG_SMP |
1679 | extern void sched_avg_update(struct rq *rq); | |
1680 | ||
8cd5601c MR |
1681 | #ifndef arch_scale_cpu_capacity |
1682 | static __always_inline | |
1683 | unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) | |
1684 | { | |
e3279a2e | 1685 | if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) |
8cd5601c MR |
1686 | return sd->smt_gain / sd->span_weight; |
1687 | ||
1688 | return SCHED_CAPACITY_SCALE; | |
1689 | } | |
1690 | #endif | |
1691 | ||
029632fb PZ |
1692 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) |
1693 | { | |
7673c8a4 | 1694 | rq->rt_avg += rt_delta * arch_scale_freq_capacity(cpu_of(rq)); |
029632fb PZ |
1695 | sched_avg_update(rq); |
1696 | } | |
1697 | #else | |
7e1a9208 JL |
1698 | #ifndef arch_scale_cpu_capacity |
1699 | static __always_inline | |
1700 | unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu) | |
1701 | { | |
1702 | return SCHED_CAPACITY_SCALE; | |
1703 | } | |
1704 | #endif | |
029632fb PZ |
1705 | static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { } |
1706 | static inline void sched_avg_update(struct rq *rq) { } | |
1707 | #endif | |
1708 | ||
eb580751 | 1709 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 | 1710 | __acquires(rq->lock); |
8a8c69c3 | 1711 | |
eb580751 | 1712 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3960c8c0 | 1713 | __acquires(p->pi_lock) |
3e71a462 | 1714 | __acquires(rq->lock); |
3960c8c0 | 1715 | |
eb580751 | 1716 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) |
3960c8c0 PZ |
1717 | __releases(rq->lock) |
1718 | { | |
d8ac8971 | 1719 | rq_unpin_lock(rq, rf); |
3960c8c0 PZ |
1720 | raw_spin_unlock(&rq->lock); |
1721 | } | |
1722 | ||
1723 | static inline void | |
eb580751 | 1724 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) |
3960c8c0 PZ |
1725 | __releases(rq->lock) |
1726 | __releases(p->pi_lock) | |
1727 | { | |
d8ac8971 | 1728 | rq_unpin_lock(rq, rf); |
3960c8c0 | 1729 | raw_spin_unlock(&rq->lock); |
eb580751 | 1730 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3960c8c0 PZ |
1731 | } |
1732 | ||
8a8c69c3 PZ |
1733 | static inline void |
1734 | rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) | |
1735 | __acquires(rq->lock) | |
1736 | { | |
1737 | raw_spin_lock_irqsave(&rq->lock, rf->flags); | |
1738 | rq_pin_lock(rq, rf); | |
1739 | } | |
1740 | ||
1741 | static inline void | |
1742 | rq_lock_irq(struct rq *rq, struct rq_flags *rf) | |
1743 | __acquires(rq->lock) | |
1744 | { | |
1745 | raw_spin_lock_irq(&rq->lock); | |
1746 | rq_pin_lock(rq, rf); | |
1747 | } | |
1748 | ||
1749 | static inline void | |
1750 | rq_lock(struct rq *rq, struct rq_flags *rf) | |
1751 | __acquires(rq->lock) | |
1752 | { | |
1753 | raw_spin_lock(&rq->lock); | |
1754 | rq_pin_lock(rq, rf); | |
1755 | } | |
1756 | ||
1757 | static inline void | |
1758 | rq_relock(struct rq *rq, struct rq_flags *rf) | |
1759 | __acquires(rq->lock) | |
1760 | { | |
1761 | raw_spin_lock(&rq->lock); | |
1762 | rq_repin_lock(rq, rf); | |
1763 | } | |
1764 | ||
1765 | static inline void | |
1766 | rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) | |
1767 | __releases(rq->lock) | |
1768 | { | |
1769 | rq_unpin_lock(rq, rf); | |
1770 | raw_spin_unlock_irqrestore(&rq->lock, rf->flags); | |
1771 | } | |
1772 | ||
1773 | static inline void | |
1774 | rq_unlock_irq(struct rq *rq, struct rq_flags *rf) | |
1775 | __releases(rq->lock) | |
1776 | { | |
1777 | rq_unpin_lock(rq, rf); | |
1778 | raw_spin_unlock_irq(&rq->lock); | |
1779 | } | |
1780 | ||
1781 | static inline void | |
1782 | rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1783 | __releases(rq->lock) | |
1784 | { | |
1785 | rq_unpin_lock(rq, rf); | |
1786 | raw_spin_unlock(&rq->lock); | |
1787 | } | |
1788 | ||
029632fb PZ |
1789 | #ifdef CONFIG_SMP |
1790 | #ifdef CONFIG_PREEMPT | |
1791 | ||
1792 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1793 | ||
1794 | /* | |
1795 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1796 | * way at the expense of forcing extra atomic operations in all | |
1797 | * invocations. This assures that the double_lock is acquired using the | |
1798 | * same underlying policy as the spinlock_t on this architecture, which | |
1799 | * reduces latency compared to the unfair variant below. However, it | |
1800 | * also adds more overhead and therefore may reduce throughput. | |
1801 | */ | |
1802 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1803 | __releases(this_rq->lock) | |
1804 | __acquires(busiest->lock) | |
1805 | __acquires(this_rq->lock) | |
1806 | { | |
1807 | raw_spin_unlock(&this_rq->lock); | |
1808 | double_rq_lock(this_rq, busiest); | |
1809 | ||
1810 | return 1; | |
1811 | } | |
1812 | ||
1813 | #else | |
1814 | /* | |
1815 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1816 | * latency by eliminating extra atomic operations when the locks are | |
1817 | * already in proper order on entry. This favors lower cpu-ids and will | |
1818 | * grant the double lock to lower cpus over higher ids under contention, | |
1819 | * regardless of entry order into the function. | |
1820 | */ | |
1821 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1822 | __releases(this_rq->lock) | |
1823 | __acquires(busiest->lock) | |
1824 | __acquires(this_rq->lock) | |
1825 | { | |
1826 | int ret = 0; | |
1827 | ||
1828 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1829 | if (busiest < this_rq) { | |
1830 | raw_spin_unlock(&this_rq->lock); | |
1831 | raw_spin_lock(&busiest->lock); | |
1832 | raw_spin_lock_nested(&this_rq->lock, | |
1833 | SINGLE_DEPTH_NESTING); | |
1834 | ret = 1; | |
1835 | } else | |
1836 | raw_spin_lock_nested(&busiest->lock, | |
1837 | SINGLE_DEPTH_NESTING); | |
1838 | } | |
1839 | return ret; | |
1840 | } | |
1841 | ||
1842 | #endif /* CONFIG_PREEMPT */ | |
1843 | ||
1844 | /* | |
1845 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1846 | */ | |
1847 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1848 | { | |
1849 | if (unlikely(!irqs_disabled())) { | |
1850 | /* printk() doesn't work good under rq->lock */ | |
1851 | raw_spin_unlock(&this_rq->lock); | |
1852 | BUG_ON(1); | |
1853 | } | |
1854 | ||
1855 | return _double_lock_balance(this_rq, busiest); | |
1856 | } | |
1857 | ||
1858 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1859 | __releases(busiest->lock) | |
1860 | { | |
1861 | raw_spin_unlock(&busiest->lock); | |
1862 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1863 | } | |
1864 | ||
74602315 PZ |
1865 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
1866 | { | |
1867 | if (l1 > l2) | |
1868 | swap(l1, l2); | |
1869 | ||
1870 | spin_lock(l1); | |
1871 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1872 | } | |
1873 | ||
60e69eed MG |
1874 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
1875 | { | |
1876 | if (l1 > l2) | |
1877 | swap(l1, l2); | |
1878 | ||
1879 | spin_lock_irq(l1); | |
1880 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1881 | } | |
1882 | ||
74602315 PZ |
1883 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
1884 | { | |
1885 | if (l1 > l2) | |
1886 | swap(l1, l2); | |
1887 | ||
1888 | raw_spin_lock(l1); | |
1889 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
1890 | } | |
1891 | ||
029632fb PZ |
1892 | /* |
1893 | * double_rq_lock - safely lock two runqueues | |
1894 | * | |
1895 | * Note this does not disable interrupts like task_rq_lock, | |
1896 | * you need to do so manually before calling. | |
1897 | */ | |
1898 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1899 | __acquires(rq1->lock) | |
1900 | __acquires(rq2->lock) | |
1901 | { | |
1902 | BUG_ON(!irqs_disabled()); | |
1903 | if (rq1 == rq2) { | |
1904 | raw_spin_lock(&rq1->lock); | |
1905 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1906 | } else { | |
1907 | if (rq1 < rq2) { | |
1908 | raw_spin_lock(&rq1->lock); | |
1909 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1910 | } else { | |
1911 | raw_spin_lock(&rq2->lock); | |
1912 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1913 | } | |
1914 | } | |
1915 | } | |
1916 | ||
1917 | /* | |
1918 | * double_rq_unlock - safely unlock two runqueues | |
1919 | * | |
1920 | * Note this does not restore interrupts like task_rq_unlock, | |
1921 | * you need to do so manually after calling. | |
1922 | */ | |
1923 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1924 | __releases(rq1->lock) | |
1925 | __releases(rq2->lock) | |
1926 | { | |
1927 | raw_spin_unlock(&rq1->lock); | |
1928 | if (rq1 != rq2) | |
1929 | raw_spin_unlock(&rq2->lock); | |
1930 | else | |
1931 | __release(rq2->lock); | |
1932 | } | |
1933 | ||
f2cb1360 IM |
1934 | extern void set_rq_online (struct rq *rq); |
1935 | extern void set_rq_offline(struct rq *rq); | |
1936 | extern bool sched_smp_initialized; | |
1937 | ||
029632fb PZ |
1938 | #else /* CONFIG_SMP */ |
1939 | ||
1940 | /* | |
1941 | * double_rq_lock - safely lock two runqueues | |
1942 | * | |
1943 | * Note this does not disable interrupts like task_rq_lock, | |
1944 | * you need to do so manually before calling. | |
1945 | */ | |
1946 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1947 | __acquires(rq1->lock) | |
1948 | __acquires(rq2->lock) | |
1949 | { | |
1950 | BUG_ON(!irqs_disabled()); | |
1951 | BUG_ON(rq1 != rq2); | |
1952 | raw_spin_lock(&rq1->lock); | |
1953 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1954 | } | |
1955 | ||
1956 | /* | |
1957 | * double_rq_unlock - safely unlock two runqueues | |
1958 | * | |
1959 | * Note this does not restore interrupts like task_rq_unlock, | |
1960 | * you need to do so manually after calling. | |
1961 | */ | |
1962 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1963 | __releases(rq1->lock) | |
1964 | __releases(rq2->lock) | |
1965 | { | |
1966 | BUG_ON(rq1 != rq2); | |
1967 | raw_spin_unlock(&rq1->lock); | |
1968 | __release(rq2->lock); | |
1969 | } | |
1970 | ||
1971 | #endif | |
1972 | ||
1973 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
1974 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
1975 | |
1976 | #ifdef CONFIG_SCHED_DEBUG | |
9469eb01 PZ |
1977 | extern bool sched_debug_enabled; |
1978 | ||
029632fb PZ |
1979 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
1980 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 1981 | extern void print_dl_stats(struct seq_file *m, int cpu); |
6b55c965 SD |
1982 | extern void |
1983 | print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); | |
397f2378 SD |
1984 | #ifdef CONFIG_NUMA_BALANCING |
1985 | extern void | |
1986 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
1987 | extern void | |
1988 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
1989 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
1990 | #endif /* CONFIG_NUMA_BALANCING */ | |
1991 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
1992 | |
1993 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
1994 | extern void init_rt_rq(struct rt_rq *rt_rq); |
1995 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 1996 | |
1ee14e6c BS |
1997 | extern void cfs_bandwidth_usage_inc(void); |
1998 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 1999 | |
3451d024 | 2000 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 SS |
2001 | enum rq_nohz_flag_bits { |
2002 | NOHZ_TICK_STOPPED, | |
2003 | NOHZ_BALANCE_KICK, | |
2004 | }; | |
2005 | ||
2006 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
20a5c8cc TG |
2007 | |
2008 | extern void nohz_balance_exit_idle(unsigned int cpu); | |
2009 | #else | |
2010 | static inline void nohz_balance_exit_idle(unsigned int cpu) { } | |
1c792db7 | 2011 | #endif |
73fbec60 | 2012 | |
daec5798 LA |
2013 | |
2014 | #ifdef CONFIG_SMP | |
2015 | static inline | |
2016 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2017 | { | |
2018 | struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); | |
2019 | int i; | |
2020 | ||
2021 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), | |
2022 | "sched RCU must be held"); | |
2023 | for_each_cpu_and(i, rd->span, cpu_active_mask) { | |
2024 | struct rq *rq = cpu_rq(i); | |
2025 | ||
2026 | rq->dl.extra_bw += bw; | |
2027 | } | |
2028 | } | |
2029 | #else | |
2030 | static inline | |
2031 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2032 | { | |
2033 | struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); | |
2034 | ||
2035 | dl->extra_bw += bw; | |
2036 | } | |
2037 | #endif | |
2038 | ||
2039 | ||
73fbec60 | 2040 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
19d23dbf | 2041 | struct irqtime { |
25e2d8c1 | 2042 | u64 total; |
a499a5a1 | 2043 | u64 tick_delta; |
19d23dbf FW |
2044 | u64 irq_start_time; |
2045 | struct u64_stats_sync sync; | |
2046 | }; | |
73fbec60 | 2047 | |
19d23dbf | 2048 | DECLARE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 2049 | |
25e2d8c1 FW |
2050 | /* |
2051 | * Returns the irqtime minus the softirq time computed by ksoftirqd. | |
2052 | * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime | |
2053 | * and never move forward. | |
2054 | */ | |
73fbec60 FW |
2055 | static inline u64 irq_time_read(int cpu) |
2056 | { | |
19d23dbf FW |
2057 | struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); |
2058 | unsigned int seq; | |
2059 | u64 total; | |
73fbec60 FW |
2060 | |
2061 | do { | |
19d23dbf | 2062 | seq = __u64_stats_fetch_begin(&irqtime->sync); |
25e2d8c1 | 2063 | total = irqtime->total; |
19d23dbf | 2064 | } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); |
73fbec60 | 2065 | |
19d23dbf | 2066 | return total; |
73fbec60 | 2067 | } |
73fbec60 | 2068 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
adaf9fcd RW |
2069 | |
2070 | #ifdef CONFIG_CPU_FREQ | |
2071 | DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); | |
2072 | ||
2073 | /** | |
2074 | * cpufreq_update_util - Take a note about CPU utilization changes. | |
12bde33d | 2075 | * @rq: Runqueue to carry out the update for. |
58919e83 | 2076 | * @flags: Update reason flags. |
adaf9fcd | 2077 | * |
58919e83 RW |
2078 | * This function is called by the scheduler on the CPU whose utilization is |
2079 | * being updated. | |
adaf9fcd RW |
2080 | * |
2081 | * It can only be called from RCU-sched read-side critical sections. | |
adaf9fcd RW |
2082 | * |
2083 | * The way cpufreq is currently arranged requires it to evaluate the CPU | |
2084 | * performance state (frequency/voltage) on a regular basis to prevent it from | |
2085 | * being stuck in a completely inadequate performance level for too long. | |
e0367b12 JL |
2086 | * That is not guaranteed to happen if the updates are only triggered from CFS |
2087 | * and DL, though, because they may not be coming in if only RT tasks are | |
2088 | * active all the time (or there are RT tasks only). | |
adaf9fcd | 2089 | * |
e0367b12 JL |
2090 | * As a workaround for that issue, this function is called periodically by the |
2091 | * RT sched class to trigger extra cpufreq updates to prevent it from stalling, | |
adaf9fcd | 2092 | * but that really is a band-aid. Going forward it should be replaced with |
e0367b12 | 2093 | * solutions targeted more specifically at RT tasks. |
adaf9fcd | 2094 | */ |
12bde33d | 2095 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) |
adaf9fcd | 2096 | { |
58919e83 RW |
2097 | struct update_util_data *data; |
2098 | ||
674e7541 VK |
2099 | data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, |
2100 | cpu_of(rq))); | |
58919e83 | 2101 | if (data) |
12bde33d RW |
2102 | data->func(data, rq_clock(rq), flags); |
2103 | } | |
adaf9fcd | 2104 | #else |
12bde33d | 2105 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} |
adaf9fcd | 2106 | #endif /* CONFIG_CPU_FREQ */ |
be53f58f | 2107 | |
9bdcb44e RW |
2108 | #ifdef arch_scale_freq_capacity |
2109 | #ifndef arch_scale_freq_invariant | |
2110 | #define arch_scale_freq_invariant() (true) | |
2111 | #endif | |
2112 | #else /* arch_scale_freq_capacity */ | |
2113 | #define arch_scale_freq_invariant() (false) | |
2114 | #endif | |
d4edd662 | 2115 | |
794a56eb JL |
2116 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL |
2117 | ||
d4edd662 JL |
2118 | static inline unsigned long cpu_util_dl(struct rq *rq) |
2119 | { | |
2120 | return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; | |
2121 | } | |
2122 | ||
2123 | static inline unsigned long cpu_util_cfs(struct rq *rq) | |
2124 | { | |
2125 | return rq->cfs.avg.util_avg; | |
2126 | } | |
794a56eb JL |
2127 | |
2128 | #endif |