]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
97fb7a0a IM |
2 | /* |
3 | * Scheduler internal types and methods: | |
4 | */ | |
029632fb | 5 | #include <linux/sched.h> |
325ea10c | 6 | |
dfc3401a | 7 | #include <linux/sched/autogroup.h> |
e6017571 | 8 | #include <linux/sched/clock.h> |
325ea10c | 9 | #include <linux/sched/coredump.h> |
55687da1 | 10 | #include <linux/sched/cpufreq.h> |
325ea10c IM |
11 | #include <linux/sched/cputime.h> |
12 | #include <linux/sched/deadline.h> | |
b17b0153 | 13 | #include <linux/sched/debug.h> |
ef8bd77f | 14 | #include <linux/sched/hotplug.h> |
325ea10c IM |
15 | #include <linux/sched/idle.h> |
16 | #include <linux/sched/init.h> | |
17 | #include <linux/sched/isolation.h> | |
18 | #include <linux/sched/jobctl.h> | |
19 | #include <linux/sched/loadavg.h> | |
20 | #include <linux/sched/mm.h> | |
21 | #include <linux/sched/nohz.h> | |
22 | #include <linux/sched/numa_balancing.h> | |
23 | #include <linux/sched/prio.h> | |
24 | #include <linux/sched/rt.h> | |
25 | #include <linux/sched/signal.h> | |
321a874a | 26 | #include <linux/sched/smt.h> |
325ea10c IM |
27 | #include <linux/sched/stat.h> |
28 | #include <linux/sched/sysctl.h> | |
29930025 | 29 | #include <linux/sched/task.h> |
68db0cf1 | 30 | #include <linux/sched/task_stack.h> |
325ea10c IM |
31 | #include <linux/sched/topology.h> |
32 | #include <linux/sched/user.h> | |
33 | #include <linux/sched/wake_q.h> | |
34 | #include <linux/sched/xacct.h> | |
35 | ||
36 | #include <uapi/linux/sched/types.h> | |
ef8bd77f | 37 | |
3866e845 | 38 | #include <linux/binfmts.h> |
325ea10c IM |
39 | #include <linux/blkdev.h> |
40 | #include <linux/compat.h> | |
41 | #include <linux/context_tracking.h> | |
42 | #include <linux/cpufreq.h> | |
43 | #include <linux/cpuidle.h> | |
44 | #include <linux/cpuset.h> | |
45 | #include <linux/ctype.h> | |
46 | #include <linux/debugfs.h> | |
47 | #include <linux/delayacct.h> | |
6aa140fa | 48 | #include <linux/energy_model.h> |
325ea10c IM |
49 | #include <linux/init_task.h> |
50 | #include <linux/kprobes.h> | |
51 | #include <linux/kthread.h> | |
52 | #include <linux/membarrier.h> | |
53 | #include <linux/migrate.h> | |
54 | #include <linux/mmu_context.h> | |
55 | #include <linux/nmi.h> | |
56 | #include <linux/proc_fs.h> | |
57 | #include <linux/prefetch.h> | |
58 | #include <linux/profile.h> | |
eb414681 | 59 | #include <linux/psi.h> |
325ea10c IM |
60 | #include <linux/rcupdate_wait.h> |
61 | #include <linux/security.h> | |
029632fb | 62 | #include <linux/stop_machine.h> |
325ea10c IM |
63 | #include <linux/suspend.h> |
64 | #include <linux/swait.h> | |
65 | #include <linux/syscalls.h> | |
66 | #include <linux/task_work.h> | |
67 | #include <linux/tsacct_kern.h> | |
68 | ||
69 | #include <asm/tlb.h> | |
029632fb | 70 | |
7fce777c | 71 | #ifdef CONFIG_PARAVIRT |
325ea10c | 72 | # include <asm/paravirt.h> |
7fce777c IM |
73 | #endif |
74 | ||
391e43da | 75 | #include "cpupri.h" |
6bfd6d72 | 76 | #include "cpudeadline.h" |
029632fb | 77 | |
9148a3a1 | 78 | #ifdef CONFIG_SCHED_DEBUG |
6d3aed3d | 79 | # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) |
9148a3a1 | 80 | #else |
6d3aed3d | 81 | # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) |
9148a3a1 PZ |
82 | #endif |
83 | ||
45ceebf7 | 84 | struct rq; |
442bf3aa | 85 | struct cpuidle_state; |
45ceebf7 | 86 | |
da0c1e65 KT |
87 | /* task_struct::on_rq states: */ |
88 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 89 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 90 | |
029632fb PZ |
91 | extern __read_mostly int scheduler_running; |
92 | ||
45ceebf7 PG |
93 | extern unsigned long calc_load_update; |
94 | extern atomic_long_t calc_load_tasks; | |
95 | ||
3289bdb4 | 96 | extern void calc_global_load_tick(struct rq *this_rq); |
d60585c5 | 97 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
3289bdb4 | 98 | |
029632fb PZ |
99 | /* |
100 | * Helpers for converting nanosecond timing to jiffy resolution | |
101 | */ | |
102 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
103 | ||
cc1f4b1f LZ |
104 | /* |
105 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
106 | * The extra resolution improves shares distribution and load balancing of | |
107 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
108 | * hierarchies, especially on larger systems. This is not a user-visible change | |
109 | * and does not change the user-interface for setting shares/weights. | |
110 | * | |
111 | * We increase resolution only if we have enough bits to allow this increased | |
97fb7a0a IM |
112 | * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit |
113 | * are pretty high and the returns do not justify the increased costs. | |
2159197d | 114 | * |
97fb7a0a IM |
115 | * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to |
116 | * increase coverage and consistency always enable it on 64-bit platforms. | |
cc1f4b1f | 117 | */ |
2159197d | 118 | #ifdef CONFIG_64BIT |
172895e6 | 119 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
6ecdd749 YD |
120 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
121 | # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) | |
cc1f4b1f | 122 | #else |
172895e6 | 123 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
cc1f4b1f LZ |
124 | # define scale_load(w) (w) |
125 | # define scale_load_down(w) (w) | |
126 | #endif | |
127 | ||
6ecdd749 | 128 | /* |
172895e6 YD |
129 | * Task weight (visible to users) and its load (invisible to users) have |
130 | * independent resolution, but they should be well calibrated. We use | |
131 | * scale_load() and scale_load_down(w) to convert between them. The | |
132 | * following must be true: | |
133 | * | |
134 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD | |
135 | * | |
6ecdd749 | 136 | */ |
172895e6 | 137 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
029632fb | 138 | |
332ac17e DF |
139 | /* |
140 | * Single value that decides SCHED_DEADLINE internal math precision. | |
141 | * 10 -> just above 1us | |
142 | * 9 -> just above 0.5us | |
143 | */ | |
97fb7a0a | 144 | #define DL_SCALE 10 |
029632fb PZ |
145 | |
146 | /* | |
97fb7a0a | 147 | * Single value that denotes runtime == period, ie unlimited time. |
029632fb | 148 | */ |
97fb7a0a | 149 | #define RUNTIME_INF ((u64)~0ULL) |
029632fb | 150 | |
20f9cd2a HA |
151 | static inline int idle_policy(int policy) |
152 | { | |
153 | return policy == SCHED_IDLE; | |
154 | } | |
d50dde5a DF |
155 | static inline int fair_policy(int policy) |
156 | { | |
157 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
158 | } | |
159 | ||
029632fb PZ |
160 | static inline int rt_policy(int policy) |
161 | { | |
d50dde5a | 162 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
163 | } |
164 | ||
aab03e05 DF |
165 | static inline int dl_policy(int policy) |
166 | { | |
167 | return policy == SCHED_DEADLINE; | |
168 | } | |
20f9cd2a HA |
169 | static inline bool valid_policy(int policy) |
170 | { | |
171 | return idle_policy(policy) || fair_policy(policy) || | |
172 | rt_policy(policy) || dl_policy(policy); | |
173 | } | |
aab03e05 | 174 | |
1da1843f VK |
175 | static inline int task_has_idle_policy(struct task_struct *p) |
176 | { | |
177 | return idle_policy(p->policy); | |
178 | } | |
179 | ||
029632fb PZ |
180 | static inline int task_has_rt_policy(struct task_struct *p) |
181 | { | |
182 | return rt_policy(p->policy); | |
183 | } | |
184 | ||
aab03e05 DF |
185 | static inline int task_has_dl_policy(struct task_struct *p) |
186 | { | |
187 | return dl_policy(p->policy); | |
188 | } | |
189 | ||
07881166 JL |
190 | #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) |
191 | ||
794a56eb JL |
192 | /* |
193 | * !! For sched_setattr_nocheck() (kernel) only !! | |
194 | * | |
195 | * This is actually gross. :( | |
196 | * | |
197 | * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE | |
198 | * tasks, but still be able to sleep. We need this on platforms that cannot | |
199 | * atomically change clock frequency. Remove once fast switching will be | |
200 | * available on such platforms. | |
201 | * | |
202 | * SUGOV stands for SchedUtil GOVernor. | |
203 | */ | |
204 | #define SCHED_FLAG_SUGOV 0x10000000 | |
205 | ||
206 | static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) | |
207 | { | |
208 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL | |
209 | return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); | |
210 | #else | |
211 | return false; | |
212 | #endif | |
213 | } | |
214 | ||
2d3d891d DF |
215 | /* |
216 | * Tells if entity @a should preempt entity @b. | |
217 | */ | |
332ac17e DF |
218 | static inline bool |
219 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d | 220 | { |
794a56eb JL |
221 | return dl_entity_is_special(a) || |
222 | dl_time_before(a->deadline, b->deadline); | |
2d3d891d DF |
223 | } |
224 | ||
029632fb PZ |
225 | /* |
226 | * This is the priority-queue data structure of the RT scheduling class: | |
227 | */ | |
228 | struct rt_prio_array { | |
229 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
230 | struct list_head queue[MAX_RT_PRIO]; | |
231 | }; | |
232 | ||
233 | struct rt_bandwidth { | |
234 | /* nests inside the rq lock: */ | |
235 | raw_spinlock_t rt_runtime_lock; | |
236 | ktime_t rt_period; | |
237 | u64 rt_runtime; | |
238 | struct hrtimer rt_period_timer; | |
4cfafd30 | 239 | unsigned int rt_period_active; |
029632fb | 240 | }; |
a5e7be3b JL |
241 | |
242 | void __dl_clear_params(struct task_struct *p); | |
243 | ||
332ac17e DF |
244 | /* |
245 | * To keep the bandwidth of -deadline tasks and groups under control | |
246 | * we need some place where: | |
247 | * - store the maximum -deadline bandwidth of the system (the group); | |
248 | * - cache the fraction of that bandwidth that is currently allocated. | |
249 | * | |
250 | * This is all done in the data structure below. It is similar to the | |
251 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
252 | * that, since here we are only interested in admission control, we | |
253 | * do not decrease any runtime while the group "executes", neither we | |
254 | * need a timer to replenish it. | |
255 | * | |
256 | * With respect to SMP, the bandwidth is given on a per-CPU basis, | |
257 | * meaning that: | |
258 | * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; | |
259 | * - dl_total_bw array contains, in the i-eth element, the currently | |
260 | * allocated bandwidth on the i-eth CPU. | |
261 | * Moreover, groups consume bandwidth on each CPU, while tasks only | |
262 | * consume bandwidth on the CPU they're running on. | |
263 | * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw | |
264 | * that will be shown the next time the proc or cgroup controls will | |
265 | * be red. It on its turn can be changed by writing on its own | |
266 | * control. | |
267 | */ | |
268 | struct dl_bandwidth { | |
97fb7a0a IM |
269 | raw_spinlock_t dl_runtime_lock; |
270 | u64 dl_runtime; | |
271 | u64 dl_period; | |
332ac17e DF |
272 | }; |
273 | ||
274 | static inline int dl_bandwidth_enabled(void) | |
275 | { | |
1724813d | 276 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
277 | } |
278 | ||
332ac17e | 279 | struct dl_bw { |
97fb7a0a IM |
280 | raw_spinlock_t lock; |
281 | u64 bw; | |
282 | u64 total_bw; | |
332ac17e DF |
283 | }; |
284 | ||
daec5798 LA |
285 | static inline void __dl_update(struct dl_bw *dl_b, s64 bw); |
286 | ||
7f51412a | 287 | static inline |
8c0944ce | 288 | void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
289 | { |
290 | dl_b->total_bw -= tsk_bw; | |
daec5798 | 291 | __dl_update(dl_b, (s32)tsk_bw / cpus); |
7f51412a JL |
292 | } |
293 | ||
294 | static inline | |
daec5798 | 295 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
296 | { |
297 | dl_b->total_bw += tsk_bw; | |
daec5798 | 298 | __dl_update(dl_b, -((s32)tsk_bw / cpus)); |
7f51412a JL |
299 | } |
300 | ||
301 | static inline | |
302 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
303 | { | |
304 | return dl_b->bw != -1 && | |
305 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
306 | } | |
307 | ||
97fb7a0a | 308 | extern void dl_change_utilization(struct task_struct *p, u64 new_bw); |
f2cb1360 | 309 | extern void init_dl_bw(struct dl_bw *dl_b); |
97fb7a0a | 310 | extern int sched_dl_global_validate(void); |
06a76fe0 | 311 | extern void sched_dl_do_global(void); |
97fb7a0a | 312 | extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); |
06a76fe0 NP |
313 | extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); |
314 | extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); | |
315 | extern bool __checkparam_dl(const struct sched_attr *attr); | |
06a76fe0 | 316 | extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); |
97fb7a0a IM |
317 | extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); |
318 | extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); | |
06a76fe0 | 319 | extern bool dl_cpu_busy(unsigned int cpu); |
029632fb PZ |
320 | |
321 | #ifdef CONFIG_CGROUP_SCHED | |
322 | ||
323 | #include <linux/cgroup.h> | |
eb414681 | 324 | #include <linux/psi.h> |
029632fb PZ |
325 | |
326 | struct cfs_rq; | |
327 | struct rt_rq; | |
328 | ||
35cf4e50 | 329 | extern struct list_head task_groups; |
029632fb PZ |
330 | |
331 | struct cfs_bandwidth { | |
332 | #ifdef CONFIG_CFS_BANDWIDTH | |
97fb7a0a IM |
333 | raw_spinlock_t lock; |
334 | ktime_t period; | |
335 | u64 quota; | |
336 | u64 runtime; | |
337 | s64 hierarchical_quota; | |
97fb7a0a | 338 | |
66567fcb | 339 | u8 idle; |
340 | u8 period_active; | |
341 | u8 distribute_running; | |
342 | u8 slack_started; | |
97fb7a0a IM |
343 | struct hrtimer period_timer; |
344 | struct hrtimer slack_timer; | |
345 | struct list_head throttled_cfs_rq; | |
346 | ||
347 | /* Statistics: */ | |
348 | int nr_periods; | |
349 | int nr_throttled; | |
350 | u64 throttled_time; | |
029632fb PZ |
351 | #endif |
352 | }; | |
353 | ||
97fb7a0a | 354 | /* Task group related information */ |
029632fb PZ |
355 | struct task_group { |
356 | struct cgroup_subsys_state css; | |
357 | ||
358 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
97fb7a0a IM |
359 | /* schedulable entities of this group on each CPU */ |
360 | struct sched_entity **se; | |
361 | /* runqueue "owned" by this group on each CPU */ | |
362 | struct cfs_rq **cfs_rq; | |
363 | unsigned long shares; | |
029632fb | 364 | |
fa6bddeb | 365 | #ifdef CONFIG_SMP |
b0367629 WL |
366 | /* |
367 | * load_avg can be heavily contended at clock tick time, so put | |
368 | * it in its own cacheline separated from the fields above which | |
369 | * will also be accessed at each tick. | |
370 | */ | |
97fb7a0a | 371 | atomic_long_t load_avg ____cacheline_aligned; |
029632fb | 372 | #endif |
fa6bddeb | 373 | #endif |
029632fb PZ |
374 | |
375 | #ifdef CONFIG_RT_GROUP_SCHED | |
97fb7a0a IM |
376 | struct sched_rt_entity **rt_se; |
377 | struct rt_rq **rt_rq; | |
029632fb | 378 | |
97fb7a0a | 379 | struct rt_bandwidth rt_bandwidth; |
029632fb PZ |
380 | #endif |
381 | ||
97fb7a0a IM |
382 | struct rcu_head rcu; |
383 | struct list_head list; | |
029632fb | 384 | |
97fb7a0a IM |
385 | struct task_group *parent; |
386 | struct list_head siblings; | |
387 | struct list_head children; | |
029632fb PZ |
388 | |
389 | #ifdef CONFIG_SCHED_AUTOGROUP | |
97fb7a0a | 390 | struct autogroup *autogroup; |
029632fb PZ |
391 | #endif |
392 | ||
97fb7a0a | 393 | struct cfs_bandwidth cfs_bandwidth; |
029632fb PZ |
394 | }; |
395 | ||
396 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
397 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
398 | ||
399 | /* | |
400 | * A weight of 0 or 1 can cause arithmetics problems. | |
401 | * A weight of a cfs_rq is the sum of weights of which entities | |
402 | * are queued on this cfs_rq, so a weight of a entity should not be | |
403 | * too large, so as the shares value of a task group. | |
404 | * (The default weight is 1024 - so there's no practical | |
405 | * limitation from this.) | |
406 | */ | |
97fb7a0a IM |
407 | #define MIN_SHARES (1UL << 1) |
408 | #define MAX_SHARES (1UL << 18) | |
029632fb PZ |
409 | #endif |
410 | ||
029632fb PZ |
411 | typedef int (*tg_visitor)(struct task_group *, void *); |
412 | ||
413 | extern int walk_tg_tree_from(struct task_group *from, | |
414 | tg_visitor down, tg_visitor up, void *data); | |
415 | ||
416 | /* | |
417 | * Iterate the full tree, calling @down when first entering a node and @up when | |
418 | * leaving it for the final time. | |
419 | * | |
420 | * Caller must hold rcu_lock or sufficient equivalent. | |
421 | */ | |
422 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
423 | { | |
424 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
425 | } | |
426 | ||
427 | extern int tg_nop(struct task_group *tg, void *data); | |
428 | ||
429 | extern void free_fair_sched_group(struct task_group *tg); | |
430 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
8663e24d | 431 | extern void online_fair_sched_group(struct task_group *tg); |
6fe1f348 | 432 | extern void unregister_fair_sched_group(struct task_group *tg); |
029632fb PZ |
433 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
434 | struct sched_entity *se, int cpu, | |
435 | struct sched_entity *parent); | |
436 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
029632fb PZ |
437 | |
438 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 439 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
440 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
441 | ||
442 | extern void free_rt_sched_group(struct task_group *tg); | |
443 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
444 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
445 | struct sched_rt_entity *rt_se, int cpu, | |
446 | struct sched_rt_entity *parent); | |
8887cd99 NP |
447 | extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); |
448 | extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); | |
449 | extern long sched_group_rt_runtime(struct task_group *tg); | |
450 | extern long sched_group_rt_period(struct task_group *tg); | |
451 | extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); | |
029632fb | 452 | |
25cc7da7 LZ |
453 | extern struct task_group *sched_create_group(struct task_group *parent); |
454 | extern void sched_online_group(struct task_group *tg, | |
455 | struct task_group *parent); | |
456 | extern void sched_destroy_group(struct task_group *tg); | |
457 | extern void sched_offline_group(struct task_group *tg); | |
458 | ||
459 | extern void sched_move_task(struct task_struct *tsk); | |
460 | ||
461 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
462 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
ad936d86 BP |
463 | |
464 | #ifdef CONFIG_SMP | |
465 | extern void set_task_rq_fair(struct sched_entity *se, | |
466 | struct cfs_rq *prev, struct cfs_rq *next); | |
467 | #else /* !CONFIG_SMP */ | |
468 | static inline void set_task_rq_fair(struct sched_entity *se, | |
469 | struct cfs_rq *prev, struct cfs_rq *next) { } | |
470 | #endif /* CONFIG_SMP */ | |
471 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
25cc7da7 | 472 | |
029632fb PZ |
473 | #else /* CONFIG_CGROUP_SCHED */ |
474 | ||
475 | struct cfs_bandwidth { }; | |
476 | ||
477 | #endif /* CONFIG_CGROUP_SCHED */ | |
478 | ||
479 | /* CFS-related fields in a runqueue */ | |
480 | struct cfs_rq { | |
97fb7a0a IM |
481 | struct load_weight load; |
482 | unsigned long runnable_weight; | |
483 | unsigned int nr_running; | |
43e9f7f2 VK |
484 | unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */ |
485 | unsigned int idle_h_nr_running; /* SCHED_IDLE */ | |
029632fb | 486 | |
97fb7a0a IM |
487 | u64 exec_clock; |
488 | u64 min_vruntime; | |
029632fb | 489 | #ifndef CONFIG_64BIT |
97fb7a0a | 490 | u64 min_vruntime_copy; |
029632fb PZ |
491 | #endif |
492 | ||
97fb7a0a | 493 | struct rb_root_cached tasks_timeline; |
029632fb | 494 | |
029632fb PZ |
495 | /* |
496 | * 'curr' points to currently running entity on this cfs_rq. | |
497 | * It is set to NULL otherwise (i.e when none are currently running). | |
498 | */ | |
97fb7a0a IM |
499 | struct sched_entity *curr; |
500 | struct sched_entity *next; | |
501 | struct sched_entity *last; | |
502 | struct sched_entity *skip; | |
029632fb PZ |
503 | |
504 | #ifdef CONFIG_SCHED_DEBUG | |
97fb7a0a | 505 | unsigned int nr_spread_over; |
029632fb PZ |
506 | #endif |
507 | ||
2dac754e PT |
508 | #ifdef CONFIG_SMP |
509 | /* | |
9d89c257 | 510 | * CFS load tracking |
2dac754e | 511 | */ |
97fb7a0a | 512 | struct sched_avg avg; |
2a2f5d4e | 513 | #ifndef CONFIG_64BIT |
97fb7a0a | 514 | u64 load_last_update_time_copy; |
9d89c257 | 515 | #endif |
2a2f5d4e PZ |
516 | struct { |
517 | raw_spinlock_t lock ____cacheline_aligned; | |
518 | int nr; | |
519 | unsigned long load_avg; | |
520 | unsigned long util_avg; | |
0e2d2aaa | 521 | unsigned long runnable_sum; |
2a2f5d4e | 522 | } removed; |
82958366 | 523 | |
9d89c257 | 524 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a IM |
525 | unsigned long tg_load_avg_contrib; |
526 | long propagate; | |
527 | long prop_runnable_sum; | |
0e2d2aaa | 528 | |
82958366 PT |
529 | /* |
530 | * h_load = weight * f(tg) | |
531 | * | |
532 | * Where f(tg) is the recursive weight fraction assigned to | |
533 | * this group. | |
534 | */ | |
97fb7a0a IM |
535 | unsigned long h_load; |
536 | u64 last_h_load_update; | |
537 | struct sched_entity *h_load_next; | |
68520796 | 538 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
82958366 PT |
539 | #endif /* CONFIG_SMP */ |
540 | ||
029632fb | 541 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a | 542 | struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ |
029632fb PZ |
543 | |
544 | /* | |
545 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
546 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
547 | * (like users, containers etc.) | |
548 | * | |
97fb7a0a IM |
549 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. |
550 | * This list is used during load balance. | |
029632fb | 551 | */ |
97fb7a0a IM |
552 | int on_list; |
553 | struct list_head leaf_cfs_rq_list; | |
554 | struct task_group *tg; /* group that "owns" this runqueue */ | |
029632fb | 555 | |
029632fb | 556 | #ifdef CONFIG_CFS_BANDWIDTH |
97fb7a0a | 557 | int runtime_enabled; |
97fb7a0a IM |
558 | s64 runtime_remaining; |
559 | ||
560 | u64 throttled_clock; | |
561 | u64 throttled_clock_task; | |
562 | u64 throttled_clock_task_time; | |
563 | int throttled; | |
564 | int throttle_count; | |
565 | struct list_head throttled_list; | |
029632fb PZ |
566 | #endif /* CONFIG_CFS_BANDWIDTH */ |
567 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
568 | }; | |
569 | ||
570 | static inline int rt_bandwidth_enabled(void) | |
571 | { | |
572 | return sysctl_sched_rt_runtime >= 0; | |
573 | } | |
574 | ||
b6366f04 | 575 | /* RT IPI pull logic requires IRQ_WORK */ |
4bdced5c | 576 | #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) |
b6366f04 SR |
577 | # define HAVE_RT_PUSH_IPI |
578 | #endif | |
579 | ||
029632fb PZ |
580 | /* Real-Time classes' related field in a runqueue: */ |
581 | struct rt_rq { | |
97fb7a0a IM |
582 | struct rt_prio_array active; |
583 | unsigned int rt_nr_running; | |
584 | unsigned int rr_nr_running; | |
029632fb PZ |
585 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
586 | struct { | |
97fb7a0a | 587 | int curr; /* highest queued rt task prio */ |
029632fb | 588 | #ifdef CONFIG_SMP |
97fb7a0a | 589 | int next; /* next highest */ |
029632fb PZ |
590 | #endif |
591 | } highest_prio; | |
592 | #endif | |
593 | #ifdef CONFIG_SMP | |
97fb7a0a IM |
594 | unsigned long rt_nr_migratory; |
595 | unsigned long rt_nr_total; | |
596 | int overloaded; | |
597 | struct plist_head pushable_tasks; | |
371bf427 | 598 | |
b6366f04 | 599 | #endif /* CONFIG_SMP */ |
97fb7a0a | 600 | int rt_queued; |
f4ebcbc0 | 601 | |
97fb7a0a IM |
602 | int rt_throttled; |
603 | u64 rt_time; | |
604 | u64 rt_runtime; | |
029632fb | 605 | /* Nests inside the rq lock: */ |
97fb7a0a | 606 | raw_spinlock_t rt_runtime_lock; |
029632fb PZ |
607 | |
608 | #ifdef CONFIG_RT_GROUP_SCHED | |
97fb7a0a | 609 | unsigned long rt_nr_boosted; |
029632fb | 610 | |
97fb7a0a IM |
611 | struct rq *rq; |
612 | struct task_group *tg; | |
029632fb PZ |
613 | #endif |
614 | }; | |
615 | ||
296b2ffe VG |
616 | static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) |
617 | { | |
618 | return rt_rq->rt_queued && rt_rq->rt_nr_running; | |
619 | } | |
620 | ||
aab03e05 DF |
621 | /* Deadline class' related fields in a runqueue */ |
622 | struct dl_rq { | |
623 | /* runqueue is an rbtree, ordered by deadline */ | |
97fb7a0a | 624 | struct rb_root_cached root; |
aab03e05 | 625 | |
97fb7a0a | 626 | unsigned long dl_nr_running; |
1baca4ce JL |
627 | |
628 | #ifdef CONFIG_SMP | |
629 | /* | |
630 | * Deadline values of the currently executing and the | |
631 | * earliest ready task on this rq. Caching these facilitates | |
dfcb245e | 632 | * the decision whether or not a ready but not running task |
1baca4ce JL |
633 | * should migrate somewhere else. |
634 | */ | |
635 | struct { | |
97fb7a0a IM |
636 | u64 curr; |
637 | u64 next; | |
1baca4ce JL |
638 | } earliest_dl; |
639 | ||
97fb7a0a IM |
640 | unsigned long dl_nr_migratory; |
641 | int overloaded; | |
1baca4ce JL |
642 | |
643 | /* | |
644 | * Tasks on this rq that can be pushed away. They are kept in | |
645 | * an rb-tree, ordered by tasks' deadlines, with caching | |
646 | * of the leftmost (earliest deadline) element. | |
647 | */ | |
97fb7a0a | 648 | struct rb_root_cached pushable_dl_tasks_root; |
332ac17e | 649 | #else |
97fb7a0a | 650 | struct dl_bw dl_bw; |
1baca4ce | 651 | #endif |
e36d8677 LA |
652 | /* |
653 | * "Active utilization" for this runqueue: increased when a | |
654 | * task wakes up (becomes TASK_RUNNING) and decreased when a | |
655 | * task blocks | |
656 | */ | |
97fb7a0a | 657 | u64 running_bw; |
4da3abce | 658 | |
8fd27231 LA |
659 | /* |
660 | * Utilization of the tasks "assigned" to this runqueue (including | |
661 | * the tasks that are in runqueue and the tasks that executed on this | |
662 | * CPU and blocked). Increased when a task moves to this runqueue, and | |
663 | * decreased when the task moves away (migrates, changes scheduling | |
664 | * policy, or terminates). | |
665 | * This is needed to compute the "inactive utilization" for the | |
666 | * runqueue (inactive utilization = this_bw - running_bw). | |
667 | */ | |
97fb7a0a IM |
668 | u64 this_bw; |
669 | u64 extra_bw; | |
8fd27231 | 670 | |
4da3abce LA |
671 | /* |
672 | * Inverse of the fraction of CPU utilization that can be reclaimed | |
673 | * by the GRUB algorithm. | |
674 | */ | |
97fb7a0a | 675 | u64 bw_ratio; |
aab03e05 DF |
676 | }; |
677 | ||
c0796298 VG |
678 | #ifdef CONFIG_FAIR_GROUP_SCHED |
679 | /* An entity is a task if it doesn't "own" a runqueue */ | |
680 | #define entity_is_task(se) (!se->my_q) | |
681 | #else | |
682 | #define entity_is_task(se) 1 | |
683 | #endif | |
684 | ||
029632fb | 685 | #ifdef CONFIG_SMP |
c0796298 VG |
686 | /* |
687 | * XXX we want to get rid of these helpers and use the full load resolution. | |
688 | */ | |
689 | static inline long se_weight(struct sched_entity *se) | |
690 | { | |
691 | return scale_load_down(se->load.weight); | |
692 | } | |
693 | ||
694 | static inline long se_runnable(struct sched_entity *se) | |
695 | { | |
696 | return scale_load_down(se->runnable_weight); | |
697 | } | |
029632fb | 698 | |
afe06efd TC |
699 | static inline bool sched_asym_prefer(int a, int b) |
700 | { | |
701 | return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); | |
702 | } | |
703 | ||
6aa140fa QP |
704 | struct perf_domain { |
705 | struct em_perf_domain *em_pd; | |
706 | struct perf_domain *next; | |
707 | struct rcu_head rcu; | |
708 | }; | |
709 | ||
630246a0 QP |
710 | /* Scheduling group status flags */ |
711 | #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */ | |
2802bf3c | 712 | #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */ |
630246a0 | 713 | |
029632fb PZ |
714 | /* |
715 | * We add the notion of a root-domain which will be used to define per-domain | |
716 | * variables. Each exclusive cpuset essentially defines an island domain by | |
97fb7a0a | 717 | * fully partitioning the member CPUs from any other cpuset. Whenever a new |
029632fb PZ |
718 | * exclusive cpuset is created, we also create and attach a new root-domain |
719 | * object. | |
720 | * | |
721 | */ | |
722 | struct root_domain { | |
97fb7a0a IM |
723 | atomic_t refcount; |
724 | atomic_t rto_count; | |
725 | struct rcu_head rcu; | |
726 | cpumask_var_t span; | |
727 | cpumask_var_t online; | |
029632fb | 728 | |
757ffdd7 VS |
729 | /* |
730 | * Indicate pullable load on at least one CPU, e.g: | |
731 | * - More than one runnable task | |
732 | * - Running task is misfit | |
733 | */ | |
575638d1 | 734 | int overload; |
4486edd1 | 735 | |
2802bf3c MR |
736 | /* Indicate one or more cpus over-utilized (tipping point) */ |
737 | int overutilized; | |
738 | ||
1baca4ce JL |
739 | /* |
740 | * The bit corresponding to a CPU gets set here if such CPU has more | |
741 | * than one runnable -deadline task (as it is below for RT tasks). | |
742 | */ | |
97fb7a0a IM |
743 | cpumask_var_t dlo_mask; |
744 | atomic_t dlo_count; | |
745 | struct dl_bw dl_bw; | |
746 | struct cpudl cpudl; | |
1baca4ce | 747 | |
4bdced5c SRRH |
748 | #ifdef HAVE_RT_PUSH_IPI |
749 | /* | |
750 | * For IPI pull requests, loop across the rto_mask. | |
751 | */ | |
97fb7a0a IM |
752 | struct irq_work rto_push_work; |
753 | raw_spinlock_t rto_lock; | |
4bdced5c | 754 | /* These are only updated and read within rto_lock */ |
97fb7a0a IM |
755 | int rto_loop; |
756 | int rto_cpu; | |
4bdced5c | 757 | /* These atomics are updated outside of a lock */ |
97fb7a0a IM |
758 | atomic_t rto_loop_next; |
759 | atomic_t rto_loop_start; | |
4bdced5c | 760 | #endif |
029632fb PZ |
761 | /* |
762 | * The "RT overload" flag: it gets set if a CPU has more than | |
763 | * one runnable RT task. | |
764 | */ | |
97fb7a0a IM |
765 | cpumask_var_t rto_mask; |
766 | struct cpupri cpupri; | |
cd92bfd3 | 767 | |
97fb7a0a | 768 | unsigned long max_cpu_capacity; |
6aa140fa QP |
769 | |
770 | /* | |
771 | * NULL-terminated list of performance domains intersecting with the | |
772 | * CPUs of the rd. Protected by RCU. | |
773 | */ | |
7ba7319f | 774 | struct perf_domain __rcu *pd; |
029632fb PZ |
775 | }; |
776 | ||
f2cb1360 | 777 | extern void init_defrootdomain(void); |
8d5dc512 | 778 | extern int sched_init_domains(const struct cpumask *cpu_map); |
f2cb1360 | 779 | extern void rq_attach_root(struct rq *rq, struct root_domain *rd); |
364f5665 SRV |
780 | extern void sched_get_rd(struct root_domain *rd); |
781 | extern void sched_put_rd(struct root_domain *rd); | |
029632fb | 782 | |
4bdced5c SRRH |
783 | #ifdef HAVE_RT_PUSH_IPI |
784 | extern void rto_push_irq_work_func(struct irq_work *work); | |
785 | #endif | |
029632fb PZ |
786 | #endif /* CONFIG_SMP */ |
787 | ||
69842cba PB |
788 | #ifdef CONFIG_UCLAMP_TASK |
789 | /* | |
790 | * struct uclamp_bucket - Utilization clamp bucket | |
791 | * @value: utilization clamp value for tasks on this clamp bucket | |
792 | * @tasks: number of RUNNABLE tasks on this clamp bucket | |
793 | * | |
794 | * Keep track of how many tasks are RUNNABLE for a given utilization | |
795 | * clamp value. | |
796 | */ | |
797 | struct uclamp_bucket { | |
798 | unsigned long value : bits_per(SCHED_CAPACITY_SCALE); | |
799 | unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE); | |
800 | }; | |
801 | ||
802 | /* | |
803 | * struct uclamp_rq - rq's utilization clamp | |
804 | * @value: currently active clamp values for a rq | |
805 | * @bucket: utilization clamp buckets affecting a rq | |
806 | * | |
807 | * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values. | |
808 | * A clamp value is affecting a rq when there is at least one task RUNNABLE | |
809 | * (or actually running) with that value. | |
810 | * | |
811 | * There are up to UCLAMP_CNT possible different clamp values, currently there | |
812 | * are only two: minimum utilization and maximum utilization. | |
813 | * | |
814 | * All utilization clamping values are MAX aggregated, since: | |
815 | * - for util_min: we want to run the CPU at least at the max of the minimum | |
816 | * utilization required by its currently RUNNABLE tasks. | |
817 | * - for util_max: we want to allow the CPU to run up to the max of the | |
818 | * maximum utilization allowed by its currently RUNNABLE tasks. | |
819 | * | |
820 | * Since on each system we expect only a limited number of different | |
821 | * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track | |
822 | * the metrics required to compute all the per-rq utilization clamp values. | |
823 | */ | |
824 | struct uclamp_rq { | |
825 | unsigned int value; | |
826 | struct uclamp_bucket bucket[UCLAMP_BUCKETS]; | |
827 | }; | |
828 | #endif /* CONFIG_UCLAMP_TASK */ | |
829 | ||
029632fb PZ |
830 | /* |
831 | * This is the main, per-CPU runqueue data structure. | |
832 | * | |
833 | * Locking rule: those places that want to lock multiple runqueues | |
834 | * (such as the load balancing or the thread migration code), lock | |
835 | * acquire operations must be ordered by ascending &runqueue. | |
836 | */ | |
837 | struct rq { | |
838 | /* runqueue lock: */ | |
97fb7a0a | 839 | raw_spinlock_t lock; |
029632fb PZ |
840 | |
841 | /* | |
842 | * nr_running and cpu_load should be in the same cacheline because | |
843 | * remote CPUs use both these fields when doing load calculation. | |
844 | */ | |
97fb7a0a | 845 | unsigned int nr_running; |
0ec8aa00 | 846 | #ifdef CONFIG_NUMA_BALANCING |
97fb7a0a IM |
847 | unsigned int nr_numa_running; |
848 | unsigned int nr_preferred_running; | |
a4739eca | 849 | unsigned int numa_migrate_on; |
0ec8aa00 | 850 | #endif |
3451d024 | 851 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 852 | #ifdef CONFIG_SMP |
97fb7a0a | 853 | unsigned long last_load_update_tick; |
e022e0d3 | 854 | unsigned long last_blocked_load_update_tick; |
f643ea22 | 855 | unsigned int has_blocked_load; |
9fd81dd5 | 856 | #endif /* CONFIG_SMP */ |
00357f5e | 857 | unsigned int nohz_tick_stopped; |
a22e47a4 | 858 | atomic_t nohz_flags; |
9fd81dd5 | 859 | #endif /* CONFIG_NO_HZ_COMMON */ |
dcdedb24 | 860 | |
97fb7a0a IM |
861 | unsigned long nr_load_updates; |
862 | u64 nr_switches; | |
029632fb | 863 | |
69842cba PB |
864 | #ifdef CONFIG_UCLAMP_TASK |
865 | /* Utilization clamp values based on CPU's RUNNABLE tasks */ | |
866 | struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned; | |
e496187d PB |
867 | unsigned int uclamp_flags; |
868 | #define UCLAMP_FLAG_IDLE 0x01 | |
69842cba PB |
869 | #endif |
870 | ||
97fb7a0a IM |
871 | struct cfs_rq cfs; |
872 | struct rt_rq rt; | |
873 | struct dl_rq dl; | |
029632fb PZ |
874 | |
875 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
97fb7a0a IM |
876 | /* list of leaf cfs_rq on this CPU: */ |
877 | struct list_head leaf_cfs_rq_list; | |
878 | struct list_head *tmp_alone_branch; | |
a35b6466 PZ |
879 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
880 | ||
029632fb PZ |
881 | /* |
882 | * This is part of a global counter where only the total sum | |
883 | * over all CPUs matters. A task can increase this counter on | |
884 | * one CPU and if it got migrated afterwards it may decrease | |
885 | * it on another CPU. Always updated under the runqueue lock: | |
886 | */ | |
97fb7a0a | 887 | unsigned long nr_uninterruptible; |
029632fb | 888 | |
97fb7a0a IM |
889 | struct task_struct *curr; |
890 | struct task_struct *idle; | |
891 | struct task_struct *stop; | |
892 | unsigned long next_balance; | |
893 | struct mm_struct *prev_mm; | |
029632fb | 894 | |
97fb7a0a IM |
895 | unsigned int clock_update_flags; |
896 | u64 clock; | |
23127296 VG |
897 | /* Ensure that all clocks are in the same cache line */ |
898 | u64 clock_task ____cacheline_aligned; | |
899 | u64 clock_pelt; | |
900 | unsigned long lost_idle_time; | |
029632fb | 901 | |
97fb7a0a | 902 | atomic_t nr_iowait; |
029632fb PZ |
903 | |
904 | #ifdef CONFIG_SMP | |
994aeb7a JFG |
905 | struct root_domain *rd; |
906 | struct sched_domain __rcu *sd; | |
97fb7a0a IM |
907 | |
908 | unsigned long cpu_capacity; | |
909 | unsigned long cpu_capacity_orig; | |
029632fb | 910 | |
97fb7a0a | 911 | struct callback_head *balance_callback; |
029632fb | 912 | |
97fb7a0a | 913 | unsigned char idle_balance; |
e3fca9e7 | 914 | |
3b1baa64 MR |
915 | unsigned long misfit_task_load; |
916 | ||
029632fb | 917 | /* For active balancing */ |
97fb7a0a IM |
918 | int active_balance; |
919 | int push_cpu; | |
920 | struct cpu_stop_work active_balance_work; | |
921 | ||
922 | /* CPU of this runqueue: */ | |
923 | int cpu; | |
924 | int online; | |
029632fb | 925 | |
367456c7 PZ |
926 | struct list_head cfs_tasks; |
927 | ||
371bf427 | 928 | struct sched_avg avg_rt; |
3727e0e1 | 929 | struct sched_avg avg_dl; |
11d4afd4 | 930 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
91c27493 VG |
931 | struct sched_avg avg_irq; |
932 | #endif | |
97fb7a0a IM |
933 | u64 idle_stamp; |
934 | u64 avg_idle; | |
9bd721c5 JL |
935 | |
936 | /* This is used to determine avg_idle's max value */ | |
97fb7a0a | 937 | u64 max_idle_balance_cost; |
029632fb PZ |
938 | #endif |
939 | ||
940 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
97fb7a0a | 941 | u64 prev_irq_time; |
029632fb PZ |
942 | #endif |
943 | #ifdef CONFIG_PARAVIRT | |
97fb7a0a | 944 | u64 prev_steal_time; |
029632fb PZ |
945 | #endif |
946 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
97fb7a0a | 947 | u64 prev_steal_time_rq; |
029632fb PZ |
948 | #endif |
949 | ||
950 | /* calc_load related fields */ | |
97fb7a0a IM |
951 | unsigned long calc_load_update; |
952 | long calc_load_active; | |
029632fb PZ |
953 | |
954 | #ifdef CONFIG_SCHED_HRTICK | |
955 | #ifdef CONFIG_SMP | |
97fb7a0a IM |
956 | int hrtick_csd_pending; |
957 | call_single_data_t hrtick_csd; | |
029632fb | 958 | #endif |
97fb7a0a | 959 | struct hrtimer hrtick_timer; |
029632fb PZ |
960 | #endif |
961 | ||
962 | #ifdef CONFIG_SCHEDSTATS | |
963 | /* latency stats */ | |
97fb7a0a IM |
964 | struct sched_info rq_sched_info; |
965 | unsigned long long rq_cpu_time; | |
029632fb PZ |
966 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ |
967 | ||
968 | /* sys_sched_yield() stats */ | |
97fb7a0a | 969 | unsigned int yld_count; |
029632fb PZ |
970 | |
971 | /* schedule() stats */ | |
97fb7a0a IM |
972 | unsigned int sched_count; |
973 | unsigned int sched_goidle; | |
029632fb PZ |
974 | |
975 | /* try_to_wake_up() stats */ | |
97fb7a0a IM |
976 | unsigned int ttwu_count; |
977 | unsigned int ttwu_local; | |
029632fb PZ |
978 | #endif |
979 | ||
980 | #ifdef CONFIG_SMP | |
97fb7a0a | 981 | struct llist_head wake_list; |
029632fb | 982 | #endif |
442bf3aa DL |
983 | |
984 | #ifdef CONFIG_CPU_IDLE | |
985 | /* Must be inspected within a rcu lock section */ | |
97fb7a0a | 986 | struct cpuidle_state *idle_state; |
442bf3aa | 987 | #endif |
029632fb PZ |
988 | }; |
989 | ||
62478d99 VG |
990 | #ifdef CONFIG_FAIR_GROUP_SCHED |
991 | ||
992 | /* CPU runqueue to which this cfs_rq is attached */ | |
993 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
994 | { | |
995 | return cfs_rq->rq; | |
996 | } | |
997 | ||
998 | #else | |
999 | ||
1000 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
1001 | { | |
1002 | return container_of(cfs_rq, struct rq, cfs); | |
1003 | } | |
1004 | #endif | |
1005 | ||
029632fb PZ |
1006 | static inline int cpu_of(struct rq *rq) |
1007 | { | |
1008 | #ifdef CONFIG_SMP | |
1009 | return rq->cpu; | |
1010 | #else | |
1011 | return 0; | |
1012 | #endif | |
1013 | } | |
1014 | ||
1b568f0a PZ |
1015 | |
1016 | #ifdef CONFIG_SCHED_SMT | |
1b568f0a PZ |
1017 | extern void __update_idle_core(struct rq *rq); |
1018 | ||
1019 | static inline void update_idle_core(struct rq *rq) | |
1020 | { | |
1021 | if (static_branch_unlikely(&sched_smt_present)) | |
1022 | __update_idle_core(rq); | |
1023 | } | |
1024 | ||
1025 | #else | |
1026 | static inline void update_idle_core(struct rq *rq) { } | |
1027 | #endif | |
1028 | ||
8b06c55b | 1029 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 1030 | |
518cd623 | 1031 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 1032 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
1033 | #define task_rq(p) cpu_rq(task_cpu(p)) |
1034 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 1035 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 1036 | |
1f351d7f JW |
1037 | extern void update_rq_clock(struct rq *rq); |
1038 | ||
cebde6d6 PZ |
1039 | static inline u64 __rq_clock_broken(struct rq *rq) |
1040 | { | |
316c1608 | 1041 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
1042 | } |
1043 | ||
cb42c9a3 MF |
1044 | /* |
1045 | * rq::clock_update_flags bits | |
1046 | * | |
1047 | * %RQCF_REQ_SKIP - will request skipping of clock update on the next | |
1048 | * call to __schedule(). This is an optimisation to avoid | |
1049 | * neighbouring rq clock updates. | |
1050 | * | |
1051 | * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is | |
1052 | * in effect and calls to update_rq_clock() are being ignored. | |
1053 | * | |
1054 | * %RQCF_UPDATED - is a debug flag that indicates whether a call has been | |
1055 | * made to update_rq_clock() since the last time rq::lock was pinned. | |
1056 | * | |
1057 | * If inside of __schedule(), clock_update_flags will have been | |
1058 | * shifted left (a left shift is a cheap operation for the fast path | |
1059 | * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, | |
1060 | * | |
1061 | * if (rq-clock_update_flags >= RQCF_UPDATED) | |
1062 | * | |
1063 | * to check if %RQCF_UPADTED is set. It'll never be shifted more than | |
1064 | * one position though, because the next rq_unpin_lock() will shift it | |
1065 | * back. | |
1066 | */ | |
97fb7a0a IM |
1067 | #define RQCF_REQ_SKIP 0x01 |
1068 | #define RQCF_ACT_SKIP 0x02 | |
1069 | #define RQCF_UPDATED 0x04 | |
cb42c9a3 MF |
1070 | |
1071 | static inline void assert_clock_updated(struct rq *rq) | |
1072 | { | |
1073 | /* | |
1074 | * The only reason for not seeing a clock update since the | |
1075 | * last rq_pin_lock() is if we're currently skipping updates. | |
1076 | */ | |
1077 | SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); | |
1078 | } | |
1079 | ||
78becc27 FW |
1080 | static inline u64 rq_clock(struct rq *rq) |
1081 | { | |
cebde6d6 | 1082 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
1083 | assert_clock_updated(rq); |
1084 | ||
78becc27 FW |
1085 | return rq->clock; |
1086 | } | |
1087 | ||
1088 | static inline u64 rq_clock_task(struct rq *rq) | |
1089 | { | |
cebde6d6 | 1090 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
1091 | assert_clock_updated(rq); |
1092 | ||
78becc27 FW |
1093 | return rq->clock_task; |
1094 | } | |
1095 | ||
adcc8da8 | 1096 | static inline void rq_clock_skip_update(struct rq *rq) |
9edfbfed PZ |
1097 | { |
1098 | lockdep_assert_held(&rq->lock); | |
adcc8da8 DB |
1099 | rq->clock_update_flags |= RQCF_REQ_SKIP; |
1100 | } | |
1101 | ||
1102 | /* | |
595058b6 | 1103 | * See rt task throttling, which is the only time a skip |
adcc8da8 DB |
1104 | * request is cancelled. |
1105 | */ | |
1106 | static inline void rq_clock_cancel_skipupdate(struct rq *rq) | |
1107 | { | |
1108 | lockdep_assert_held(&rq->lock); | |
1109 | rq->clock_update_flags &= ~RQCF_REQ_SKIP; | |
9edfbfed PZ |
1110 | } |
1111 | ||
d8ac8971 MF |
1112 | struct rq_flags { |
1113 | unsigned long flags; | |
1114 | struct pin_cookie cookie; | |
cb42c9a3 MF |
1115 | #ifdef CONFIG_SCHED_DEBUG |
1116 | /* | |
1117 | * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the | |
1118 | * current pin context is stashed here in case it needs to be | |
1119 | * restored in rq_repin_lock(). | |
1120 | */ | |
1121 | unsigned int clock_update_flags; | |
1122 | #endif | |
d8ac8971 MF |
1123 | }; |
1124 | ||
1125 | static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) | |
1126 | { | |
1127 | rf->cookie = lockdep_pin_lock(&rq->lock); | |
cb42c9a3 MF |
1128 | |
1129 | #ifdef CONFIG_SCHED_DEBUG | |
1130 | rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); | |
1131 | rf->clock_update_flags = 0; | |
1132 | #endif | |
d8ac8971 MF |
1133 | } |
1134 | ||
1135 | static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) | |
1136 | { | |
cb42c9a3 MF |
1137 | #ifdef CONFIG_SCHED_DEBUG |
1138 | if (rq->clock_update_flags > RQCF_ACT_SKIP) | |
1139 | rf->clock_update_flags = RQCF_UPDATED; | |
1140 | #endif | |
1141 | ||
d8ac8971 MF |
1142 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
1143 | } | |
1144 | ||
1145 | static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) | |
1146 | { | |
1147 | lockdep_repin_lock(&rq->lock, rf->cookie); | |
cb42c9a3 MF |
1148 | |
1149 | #ifdef CONFIG_SCHED_DEBUG | |
1150 | /* | |
1151 | * Restore the value we stashed in @rf for this pin context. | |
1152 | */ | |
1153 | rq->clock_update_flags |= rf->clock_update_flags; | |
1154 | #endif | |
d8ac8971 MF |
1155 | } |
1156 | ||
1f351d7f JW |
1157 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
1158 | __acquires(rq->lock); | |
1159 | ||
1160 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) | |
1161 | __acquires(p->pi_lock) | |
1162 | __acquires(rq->lock); | |
1163 | ||
1164 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1165 | __releases(rq->lock) | |
1166 | { | |
1167 | rq_unpin_lock(rq, rf); | |
1168 | raw_spin_unlock(&rq->lock); | |
1169 | } | |
1170 | ||
1171 | static inline void | |
1172 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) | |
1173 | __releases(rq->lock) | |
1174 | __releases(p->pi_lock) | |
1175 | { | |
1176 | rq_unpin_lock(rq, rf); | |
1177 | raw_spin_unlock(&rq->lock); | |
1178 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); | |
1179 | } | |
1180 | ||
1181 | static inline void | |
1182 | rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) | |
1183 | __acquires(rq->lock) | |
1184 | { | |
1185 | raw_spin_lock_irqsave(&rq->lock, rf->flags); | |
1186 | rq_pin_lock(rq, rf); | |
1187 | } | |
1188 | ||
1189 | static inline void | |
1190 | rq_lock_irq(struct rq *rq, struct rq_flags *rf) | |
1191 | __acquires(rq->lock) | |
1192 | { | |
1193 | raw_spin_lock_irq(&rq->lock); | |
1194 | rq_pin_lock(rq, rf); | |
1195 | } | |
1196 | ||
1197 | static inline void | |
1198 | rq_lock(struct rq *rq, struct rq_flags *rf) | |
1199 | __acquires(rq->lock) | |
1200 | { | |
1201 | raw_spin_lock(&rq->lock); | |
1202 | rq_pin_lock(rq, rf); | |
1203 | } | |
1204 | ||
1205 | static inline void | |
1206 | rq_relock(struct rq *rq, struct rq_flags *rf) | |
1207 | __acquires(rq->lock) | |
1208 | { | |
1209 | raw_spin_lock(&rq->lock); | |
1210 | rq_repin_lock(rq, rf); | |
1211 | } | |
1212 | ||
1213 | static inline void | |
1214 | rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) | |
1215 | __releases(rq->lock) | |
1216 | { | |
1217 | rq_unpin_lock(rq, rf); | |
1218 | raw_spin_unlock_irqrestore(&rq->lock, rf->flags); | |
1219 | } | |
1220 | ||
1221 | static inline void | |
1222 | rq_unlock_irq(struct rq *rq, struct rq_flags *rf) | |
1223 | __releases(rq->lock) | |
1224 | { | |
1225 | rq_unpin_lock(rq, rf); | |
1226 | raw_spin_unlock_irq(&rq->lock); | |
1227 | } | |
1228 | ||
1229 | static inline void | |
1230 | rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1231 | __releases(rq->lock) | |
1232 | { | |
1233 | rq_unpin_lock(rq, rf); | |
1234 | raw_spin_unlock(&rq->lock); | |
1235 | } | |
1236 | ||
246b3b33 JW |
1237 | static inline struct rq * |
1238 | this_rq_lock_irq(struct rq_flags *rf) | |
1239 | __acquires(rq->lock) | |
1240 | { | |
1241 | struct rq *rq; | |
1242 | ||
1243 | local_irq_disable(); | |
1244 | rq = this_rq(); | |
1245 | rq_lock(rq, rf); | |
1246 | return rq; | |
1247 | } | |
1248 | ||
9942f79b | 1249 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
1250 | enum numa_topology_type { |
1251 | NUMA_DIRECT, | |
1252 | NUMA_GLUELESS_MESH, | |
1253 | NUMA_BACKPLANE, | |
1254 | }; | |
1255 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
1256 | extern int sched_max_numa_distance; |
1257 | extern bool find_numa_distance(int distance); | |
f2cb1360 IM |
1258 | extern void sched_init_numa(void); |
1259 | extern void sched_domains_numa_masks_set(unsigned int cpu); | |
1260 | extern void sched_domains_numa_masks_clear(unsigned int cpu); | |
e0e8d491 | 1261 | extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu); |
f2cb1360 IM |
1262 | #else |
1263 | static inline void sched_init_numa(void) { } | |
1264 | static inline void sched_domains_numa_masks_set(unsigned int cpu) { } | |
1265 | static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
e0e8d491 WL |
1266 | static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu) |
1267 | { | |
1268 | return nr_cpu_ids; | |
1269 | } | |
f2cb1360 IM |
1270 | #endif |
1271 | ||
f809ca9a | 1272 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
1273 | /* The regions in numa_faults array from task_struct */ |
1274 | enum numa_faults_stats { | |
1275 | NUMA_MEM = 0, | |
1276 | NUMA_CPU, | |
1277 | NUMA_MEMBUF, | |
1278 | NUMA_CPUBUF | |
1279 | }; | |
0ec8aa00 | 1280 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 1281 | extern int migrate_task_to(struct task_struct *p, int cpu); |
0ad4e3df SD |
1282 | extern int migrate_swap(struct task_struct *p, struct task_struct *t, |
1283 | int cpu, int scpu); | |
13784475 MG |
1284 | extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); |
1285 | #else | |
1286 | static inline void | |
1287 | init_numa_balancing(unsigned long clone_flags, struct task_struct *p) | |
1288 | { | |
1289 | } | |
f809ca9a MG |
1290 | #endif /* CONFIG_NUMA_BALANCING */ |
1291 | ||
518cd623 PZ |
1292 | #ifdef CONFIG_SMP |
1293 | ||
e3fca9e7 PZ |
1294 | static inline void |
1295 | queue_balance_callback(struct rq *rq, | |
1296 | struct callback_head *head, | |
1297 | void (*func)(struct rq *rq)) | |
1298 | { | |
1299 | lockdep_assert_held(&rq->lock); | |
1300 | ||
1301 | if (unlikely(head->next)) | |
1302 | return; | |
1303 | ||
1304 | head->func = (void (*)(struct callback_head *))func; | |
1305 | head->next = rq->balance_callback; | |
1306 | rq->balance_callback = head; | |
1307 | } | |
1308 | ||
e3baac47 PZ |
1309 | extern void sched_ttwu_pending(void); |
1310 | ||
029632fb PZ |
1311 | #define rcu_dereference_check_sched_domain(p) \ |
1312 | rcu_dereference_check((p), \ | |
1313 | lockdep_is_held(&sched_domains_mutex)) | |
1314 | ||
1315 | /* | |
1316 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
337e9b07 | 1317 | * See destroy_sched_domains: call_rcu for details. |
029632fb PZ |
1318 | * |
1319 | * The domain tree of any CPU may only be accessed from within | |
1320 | * preempt-disabled sections. | |
1321 | */ | |
1322 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
1323 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
1324 | __sd; __sd = __sd->parent) | |
029632fb | 1325 | |
77e81365 SS |
1326 | #define for_each_lower_domain(sd) for (; sd; sd = sd->child) |
1327 | ||
518cd623 PZ |
1328 | /** |
1329 | * highest_flag_domain - Return highest sched_domain containing flag. | |
97fb7a0a | 1330 | * @cpu: The CPU whose highest level of sched domain is to |
518cd623 PZ |
1331 | * be returned. |
1332 | * @flag: The flag to check for the highest sched_domain | |
97fb7a0a | 1333 | * for the given CPU. |
518cd623 | 1334 | * |
97fb7a0a | 1335 | * Returns the highest sched_domain of a CPU which contains the given flag. |
518cd623 PZ |
1336 | */ |
1337 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
1338 | { | |
1339 | struct sched_domain *sd, *hsd = NULL; | |
1340 | ||
1341 | for_each_domain(cpu, sd) { | |
1342 | if (!(sd->flags & flag)) | |
1343 | break; | |
1344 | hsd = sd; | |
1345 | } | |
1346 | ||
1347 | return hsd; | |
1348 | } | |
1349 | ||
fb13c7ee MG |
1350 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
1351 | { | |
1352 | struct sched_domain *sd; | |
1353 | ||
1354 | for_each_domain(cpu, sd) { | |
1355 | if (sd->flags & flag) | |
1356 | break; | |
1357 | } | |
1358 | ||
1359 | return sd; | |
1360 | } | |
1361 | ||
994aeb7a | 1362 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc); |
7d9ffa89 | 1363 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 1364 | DECLARE_PER_CPU(int, sd_llc_id); |
994aeb7a JFG |
1365 | DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); |
1366 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa); | |
1367 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); | |
1368 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); | |
df054e84 | 1369 | extern struct static_key_false sched_asym_cpucapacity; |
518cd623 | 1370 | |
63b2ca30 | 1371 | struct sched_group_capacity { |
97fb7a0a | 1372 | atomic_t ref; |
5e6521ea | 1373 | /* |
172895e6 | 1374 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
63b2ca30 | 1375 | * for a single CPU. |
5e6521ea | 1376 | */ |
97fb7a0a IM |
1377 | unsigned long capacity; |
1378 | unsigned long min_capacity; /* Min per-CPU capacity in group */ | |
e3d6d0cb | 1379 | unsigned long max_capacity; /* Max per-CPU capacity in group */ |
97fb7a0a IM |
1380 | unsigned long next_update; |
1381 | int imbalance; /* XXX unrelated to capacity but shared group state */ | |
5e6521ea | 1382 | |
005f874d | 1383 | #ifdef CONFIG_SCHED_DEBUG |
97fb7a0a | 1384 | int id; |
005f874d PZ |
1385 | #endif |
1386 | ||
97fb7a0a | 1387 | unsigned long cpumask[0]; /* Balance mask */ |
5e6521ea LZ |
1388 | }; |
1389 | ||
1390 | struct sched_group { | |
97fb7a0a IM |
1391 | struct sched_group *next; /* Must be a circular list */ |
1392 | atomic_t ref; | |
5e6521ea | 1393 | |
97fb7a0a | 1394 | unsigned int group_weight; |
63b2ca30 | 1395 | struct sched_group_capacity *sgc; |
97fb7a0a | 1396 | int asym_prefer_cpu; /* CPU of highest priority in group */ |
5e6521ea LZ |
1397 | |
1398 | /* | |
1399 | * The CPUs this group covers. | |
1400 | * | |
1401 | * NOTE: this field is variable length. (Allocated dynamically | |
1402 | * by attaching extra space to the end of the structure, | |
1403 | * depending on how many CPUs the kernel has booted up with) | |
1404 | */ | |
97fb7a0a | 1405 | unsigned long cpumask[0]; |
5e6521ea LZ |
1406 | }; |
1407 | ||
ae4df9d6 | 1408 | static inline struct cpumask *sched_group_span(struct sched_group *sg) |
5e6521ea LZ |
1409 | { |
1410 | return to_cpumask(sg->cpumask); | |
1411 | } | |
1412 | ||
1413 | /* | |
e5c14b1f | 1414 | * See build_balance_mask(). |
5e6521ea | 1415 | */ |
e5c14b1f | 1416 | static inline struct cpumask *group_balance_mask(struct sched_group *sg) |
5e6521ea | 1417 | { |
63b2ca30 | 1418 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
1419 | } |
1420 | ||
1421 | /** | |
97fb7a0a IM |
1422 | * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. |
1423 | * @group: The group whose first CPU is to be returned. | |
5e6521ea LZ |
1424 | */ |
1425 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
1426 | { | |
ae4df9d6 | 1427 | return cpumask_first(sched_group_span(group)); |
5e6521ea LZ |
1428 | } |
1429 | ||
c1174876 PZ |
1430 | extern int group_balance_cpu(struct sched_group *sg); |
1431 | ||
3866e845 SRRH |
1432 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
1433 | void register_sched_domain_sysctl(void); | |
bbdacdfe | 1434 | void dirty_sched_domain_sysctl(int cpu); |
3866e845 SRRH |
1435 | void unregister_sched_domain_sysctl(void); |
1436 | #else | |
1437 | static inline void register_sched_domain_sysctl(void) | |
1438 | { | |
1439 | } | |
bbdacdfe PZ |
1440 | static inline void dirty_sched_domain_sysctl(int cpu) |
1441 | { | |
1442 | } | |
3866e845 SRRH |
1443 | static inline void unregister_sched_domain_sysctl(void) |
1444 | { | |
1445 | } | |
1446 | #endif | |
1447 | ||
e3baac47 PZ |
1448 | #else |
1449 | ||
1450 | static inline void sched_ttwu_pending(void) { } | |
1451 | ||
518cd623 | 1452 | #endif /* CONFIG_SMP */ |
029632fb | 1453 | |
391e43da | 1454 | #include "stats.h" |
1051408f | 1455 | #include "autogroup.h" |
029632fb PZ |
1456 | |
1457 | #ifdef CONFIG_CGROUP_SCHED | |
1458 | ||
1459 | /* | |
1460 | * Return the group to which this tasks belongs. | |
1461 | * | |
8af01f56 TH |
1462 | * We cannot use task_css() and friends because the cgroup subsystem |
1463 | * changes that value before the cgroup_subsys::attach() method is called, | |
1464 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
1465 | * |
1466 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
1467 | * core changes this before calling sched_move_task(). | |
1468 | * | |
1469 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
1470 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
1471 | */ |
1472 | static inline struct task_group *task_group(struct task_struct *p) | |
1473 | { | |
8323f26c | 1474 | return p->sched_task_group; |
029632fb PZ |
1475 | } |
1476 | ||
1477 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
1478 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
1479 | { | |
1480 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
1481 | struct task_group *tg = task_group(p); | |
1482 | #endif | |
1483 | ||
1484 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
ad936d86 | 1485 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
029632fb PZ |
1486 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
1487 | p->se.parent = tg->se[cpu]; | |
1488 | #endif | |
1489 | ||
1490 | #ifdef CONFIG_RT_GROUP_SCHED | |
1491 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
1492 | p->rt.parent = tg->rt_se[cpu]; | |
1493 | #endif | |
1494 | } | |
1495 | ||
1496 | #else /* CONFIG_CGROUP_SCHED */ | |
1497 | ||
1498 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
1499 | static inline struct task_group *task_group(struct task_struct *p) | |
1500 | { | |
1501 | return NULL; | |
1502 | } | |
1503 | ||
1504 | #endif /* CONFIG_CGROUP_SCHED */ | |
1505 | ||
1506 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1507 | { | |
1508 | set_task_rq(p, cpu); | |
1509 | #ifdef CONFIG_SMP | |
1510 | /* | |
1511 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
dfcb245e | 1512 | * successfully executed on another CPU. We must ensure that updates of |
029632fb PZ |
1513 | * per-task data have been completed by this moment. |
1514 | */ | |
1515 | smp_wmb(); | |
c65eacbe | 1516 | #ifdef CONFIG_THREAD_INFO_IN_TASK |
c546951d | 1517 | WRITE_ONCE(p->cpu, cpu); |
c65eacbe | 1518 | #else |
c546951d | 1519 | WRITE_ONCE(task_thread_info(p)->cpu, cpu); |
c65eacbe | 1520 | #endif |
ac66f547 | 1521 | p->wake_cpu = cpu; |
029632fb PZ |
1522 | #endif |
1523 | } | |
1524 | ||
1525 | /* | |
1526 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
1527 | */ | |
1528 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 1529 | # include <linux/static_key.h> |
029632fb PZ |
1530 | # define const_debug __read_mostly |
1531 | #else | |
1532 | # define const_debug const | |
1533 | #endif | |
1534 | ||
029632fb PZ |
1535 | #define SCHED_FEAT(name, enabled) \ |
1536 | __SCHED_FEAT_##name , | |
1537 | ||
1538 | enum { | |
391e43da | 1539 | #include "features.h" |
f8b6d1cc | 1540 | __SCHED_FEAT_NR, |
029632fb PZ |
1541 | }; |
1542 | ||
1543 | #undef SCHED_FEAT | |
1544 | ||
e9666d10 | 1545 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL) |
765cc3a4 PB |
1546 | |
1547 | /* | |
1548 | * To support run-time toggling of sched features, all the translation units | |
1549 | * (but core.c) reference the sysctl_sched_features defined in core.c. | |
1550 | */ | |
1551 | extern const_debug unsigned int sysctl_sched_features; | |
1552 | ||
f8b6d1cc | 1553 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1554 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1555 | { \ |
6e76ea8a | 1556 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1557 | } |
1558 | ||
1559 | #include "features.h" | |
f8b6d1cc PZ |
1560 | #undef SCHED_FEAT |
1561 | ||
c5905afb | 1562 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc | 1563 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
765cc3a4 | 1564 | |
e9666d10 | 1565 | #else /* !(SCHED_DEBUG && CONFIG_JUMP_LABEL) */ |
765cc3a4 PB |
1566 | |
1567 | /* | |
1568 | * Each translation unit has its own copy of sysctl_sched_features to allow | |
1569 | * constants propagation at compile time and compiler optimization based on | |
1570 | * features default. | |
1571 | */ | |
1572 | #define SCHED_FEAT(name, enabled) \ | |
1573 | (1UL << __SCHED_FEAT_##name) * enabled | | |
1574 | static const_debug __maybe_unused unsigned int sysctl_sched_features = | |
1575 | #include "features.h" | |
1576 | 0; | |
1577 | #undef SCHED_FEAT | |
1578 | ||
7e6f4c5d | 1579 | #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
765cc3a4 | 1580 | |
e9666d10 | 1581 | #endif /* SCHED_DEBUG && CONFIG_JUMP_LABEL */ |
029632fb | 1582 | |
2a595721 | 1583 | extern struct static_key_false sched_numa_balancing; |
cb251765 | 1584 | extern struct static_key_false sched_schedstats; |
cbee9f88 | 1585 | |
029632fb PZ |
1586 | static inline u64 global_rt_period(void) |
1587 | { | |
1588 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1589 | } | |
1590 | ||
1591 | static inline u64 global_rt_runtime(void) | |
1592 | { | |
1593 | if (sysctl_sched_rt_runtime < 0) | |
1594 | return RUNTIME_INF; | |
1595 | ||
1596 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1597 | } | |
1598 | ||
029632fb PZ |
1599 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1600 | { | |
1601 | return rq->curr == p; | |
1602 | } | |
1603 | ||
1604 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1605 | { | |
1606 | #ifdef CONFIG_SMP | |
1607 | return p->on_cpu; | |
1608 | #else | |
1609 | return task_current(rq, p); | |
1610 | #endif | |
1611 | } | |
1612 | ||
da0c1e65 KT |
1613 | static inline int task_on_rq_queued(struct task_struct *p) |
1614 | { | |
1615 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1616 | } | |
029632fb | 1617 | |
cca26e80 KT |
1618 | static inline int task_on_rq_migrating(struct task_struct *p) |
1619 | { | |
c546951d | 1620 | return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING; |
cca26e80 KT |
1621 | } |
1622 | ||
b13095f0 LZ |
1623 | /* |
1624 | * wake flags | |
1625 | */ | |
97fb7a0a IM |
1626 | #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */ |
1627 | #define WF_FORK 0x02 /* Child wakeup after fork */ | |
1628 | #define WF_MIGRATED 0x4 /* Internal use, task got migrated */ | |
b13095f0 | 1629 | |
029632fb PZ |
1630 | /* |
1631 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1632 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1633 | * each task makes to its run queue's load is weighted according to its | |
1634 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1635 | * scaled version of the new time slice allocation that they receive on time | |
1636 | * slice expiry etc. | |
1637 | */ | |
1638 | ||
97fb7a0a IM |
1639 | #define WEIGHT_IDLEPRIO 3 |
1640 | #define WMULT_IDLEPRIO 1431655765 | |
029632fb | 1641 | |
97fb7a0a IM |
1642 | extern const int sched_prio_to_weight[40]; |
1643 | extern const u32 sched_prio_to_wmult[40]; | |
029632fb | 1644 | |
ff77e468 PZ |
1645 | /* |
1646 | * {de,en}queue flags: | |
1647 | * | |
1648 | * DEQUEUE_SLEEP - task is no longer runnable | |
1649 | * ENQUEUE_WAKEUP - task just became runnable | |
1650 | * | |
1651 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks | |
1652 | * are in a known state which allows modification. Such pairs | |
1653 | * should preserve as much state as possible. | |
1654 | * | |
1655 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location | |
1656 | * in the runqueue. | |
1657 | * | |
1658 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) | |
1659 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) | |
59efa0ba | 1660 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
ff77e468 PZ |
1661 | * |
1662 | */ | |
1663 | ||
1664 | #define DEQUEUE_SLEEP 0x01 | |
97fb7a0a IM |
1665 | #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ |
1666 | #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ | |
1667 | #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ | |
ff77e468 | 1668 | |
1de64443 | 1669 | #define ENQUEUE_WAKEUP 0x01 |
ff77e468 PZ |
1670 | #define ENQUEUE_RESTORE 0x02 |
1671 | #define ENQUEUE_MOVE 0x04 | |
0a67d1ee | 1672 | #define ENQUEUE_NOCLOCK 0x08 |
ff77e468 | 1673 | |
0a67d1ee PZ |
1674 | #define ENQUEUE_HEAD 0x10 |
1675 | #define ENQUEUE_REPLENISH 0x20 | |
c82ba9fa | 1676 | #ifdef CONFIG_SMP |
0a67d1ee | 1677 | #define ENQUEUE_MIGRATED 0x40 |
c82ba9fa | 1678 | #else |
59efa0ba | 1679 | #define ENQUEUE_MIGRATED 0x00 |
c82ba9fa | 1680 | #endif |
c82ba9fa | 1681 | |
37e117c0 PZ |
1682 | #define RETRY_TASK ((void *)-1UL) |
1683 | ||
c82ba9fa LZ |
1684 | struct sched_class { |
1685 | const struct sched_class *next; | |
1686 | ||
69842cba PB |
1687 | #ifdef CONFIG_UCLAMP_TASK |
1688 | int uclamp_enabled; | |
1689 | #endif | |
1690 | ||
c82ba9fa LZ |
1691 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); |
1692 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
97fb7a0a IM |
1693 | void (*yield_task) (struct rq *rq); |
1694 | bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt); | |
c82ba9fa | 1695 | |
97fb7a0a | 1696 | void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); |
c82ba9fa | 1697 | |
606dba2e PZ |
1698 | /* |
1699 | * It is the responsibility of the pick_next_task() method that will | |
1700 | * return the next task to call put_prev_task() on the @prev task or | |
1701 | * something equivalent. | |
37e117c0 PZ |
1702 | * |
1703 | * May return RETRY_TASK when it finds a higher prio class has runnable | |
1704 | * tasks. | |
606dba2e | 1705 | */ |
97fb7a0a IM |
1706 | struct task_struct * (*pick_next_task)(struct rq *rq, |
1707 | struct task_struct *prev, | |
1708 | struct rq_flags *rf); | |
1709 | void (*put_prev_task)(struct rq *rq, struct task_struct *p); | |
c82ba9fa LZ |
1710 | |
1711 | #ifdef CONFIG_SMP | |
ac66f547 | 1712 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); |
1327237a | 1713 | void (*migrate_task_rq)(struct task_struct *p, int new_cpu); |
c82ba9fa | 1714 | |
97fb7a0a | 1715 | void (*task_woken)(struct rq *this_rq, struct task_struct *task); |
c82ba9fa LZ |
1716 | |
1717 | void (*set_cpus_allowed)(struct task_struct *p, | |
1718 | const struct cpumask *newmask); | |
1719 | ||
1720 | void (*rq_online)(struct rq *rq); | |
1721 | void (*rq_offline)(struct rq *rq); | |
1722 | #endif | |
1723 | ||
97fb7a0a IM |
1724 | void (*set_curr_task)(struct rq *rq); |
1725 | void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); | |
1726 | void (*task_fork)(struct task_struct *p); | |
1727 | void (*task_dead)(struct task_struct *p); | |
c82ba9fa | 1728 | |
67dfa1b7 KT |
1729 | /* |
1730 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1731 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1732 | * rq->lock. They are however serialized by p->pi_lock. | |
1733 | */ | |
97fb7a0a IM |
1734 | void (*switched_from)(struct rq *this_rq, struct task_struct *task); |
1735 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
c82ba9fa | 1736 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, |
97fb7a0a | 1737 | int oldprio); |
c82ba9fa | 1738 | |
97fb7a0a IM |
1739 | unsigned int (*get_rr_interval)(struct rq *rq, |
1740 | struct task_struct *task); | |
c82ba9fa | 1741 | |
97fb7a0a | 1742 | void (*update_curr)(struct rq *rq); |
6e998916 | 1743 | |
97fb7a0a IM |
1744 | #define TASK_SET_GROUP 0 |
1745 | #define TASK_MOVE_GROUP 1 | |
ea86cb4b | 1746 | |
c82ba9fa | 1747 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a | 1748 | void (*task_change_group)(struct task_struct *p, int type); |
c82ba9fa LZ |
1749 | #endif |
1750 | }; | |
029632fb | 1751 | |
3f1d2a31 PZ |
1752 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1753 | { | |
10e7071b | 1754 | WARN_ON_ONCE(rq->curr != prev); |
3f1d2a31 PZ |
1755 | prev->sched_class->put_prev_task(rq, prev); |
1756 | } | |
1757 | ||
b2bf6c31 PZ |
1758 | static inline void set_curr_task(struct rq *rq, struct task_struct *curr) |
1759 | { | |
1760 | curr->sched_class->set_curr_task(rq); | |
1761 | } | |
1762 | ||
f5832c19 | 1763 | #ifdef CONFIG_SMP |
029632fb | 1764 | #define sched_class_highest (&stop_sched_class) |
f5832c19 NP |
1765 | #else |
1766 | #define sched_class_highest (&dl_sched_class) | |
1767 | #endif | |
029632fb PZ |
1768 | #define for_each_class(class) \ |
1769 | for (class = sched_class_highest; class; class = class->next) | |
1770 | ||
1771 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1772 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1773 | extern const struct sched_class rt_sched_class; |
1774 | extern const struct sched_class fair_sched_class; | |
1775 | extern const struct sched_class idle_sched_class; | |
1776 | ||
1777 | ||
1778 | #ifdef CONFIG_SMP | |
1779 | ||
63b2ca30 | 1780 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1781 | |
7caff66f | 1782 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1783 | |
c5b28038 PZ |
1784 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); |
1785 | ||
029632fb PZ |
1786 | #endif |
1787 | ||
442bf3aa DL |
1788 | #ifdef CONFIG_CPU_IDLE |
1789 | static inline void idle_set_state(struct rq *rq, | |
1790 | struct cpuidle_state *idle_state) | |
1791 | { | |
1792 | rq->idle_state = idle_state; | |
1793 | } | |
1794 | ||
1795 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1796 | { | |
9148a3a1 | 1797 | SCHED_WARN_ON(!rcu_read_lock_held()); |
97fb7a0a | 1798 | |
442bf3aa DL |
1799 | return rq->idle_state; |
1800 | } | |
1801 | #else | |
1802 | static inline void idle_set_state(struct rq *rq, | |
1803 | struct cpuidle_state *idle_state) | |
1804 | { | |
1805 | } | |
1806 | ||
1807 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1808 | { | |
1809 | return NULL; | |
1810 | } | |
1811 | #endif | |
1812 | ||
8663effb SRV |
1813 | extern void schedule_idle(void); |
1814 | ||
029632fb PZ |
1815 | extern void sysrq_sched_debug_show(void); |
1816 | extern void sched_init_granularity(void); | |
1817 | extern void update_max_interval(void); | |
1baca4ce JL |
1818 | |
1819 | extern void init_sched_dl_class(void); | |
029632fb PZ |
1820 | extern void init_sched_rt_class(void); |
1821 | extern void init_sched_fair_class(void); | |
1822 | ||
9059393e VG |
1823 | extern void reweight_task(struct task_struct *p, int prio); |
1824 | ||
8875125e | 1825 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
1826 | extern void resched_cpu(int cpu); |
1827 | ||
1828 | extern struct rt_bandwidth def_rt_bandwidth; | |
1829 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
1830 | ||
332ac17e DF |
1831 | extern struct dl_bandwidth def_dl_bandwidth; |
1832 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 | 1833 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
209a0cbd | 1834 | extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); |
4da3abce | 1835 | extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq); |
aab03e05 | 1836 | |
97fb7a0a IM |
1837 | #define BW_SHIFT 20 |
1838 | #define BW_UNIT (1 << BW_SHIFT) | |
1839 | #define RATIO_SHIFT 8 | |
332ac17e DF |
1840 | unsigned long to_ratio(u64 period, u64 runtime); |
1841 | ||
540247fb | 1842 | extern void init_entity_runnable_average(struct sched_entity *se); |
d0fe0b9c | 1843 | extern void post_init_entity_util_avg(struct task_struct *p); |
a75cdaa9 | 1844 | |
76d92ac3 FW |
1845 | #ifdef CONFIG_NO_HZ_FULL |
1846 | extern bool sched_can_stop_tick(struct rq *rq); | |
d84b3131 | 1847 | extern int __init sched_tick_offload_init(void); |
76d92ac3 FW |
1848 | |
1849 | /* | |
1850 | * Tick may be needed by tasks in the runqueue depending on their policy and | |
1851 | * requirements. If tick is needed, lets send the target an IPI to kick it out of | |
1852 | * nohz mode if necessary. | |
1853 | */ | |
1854 | static inline void sched_update_tick_dependency(struct rq *rq) | |
1855 | { | |
1856 | int cpu; | |
1857 | ||
1858 | if (!tick_nohz_full_enabled()) | |
1859 | return; | |
1860 | ||
1861 | cpu = cpu_of(rq); | |
1862 | ||
1863 | if (!tick_nohz_full_cpu(cpu)) | |
1864 | return; | |
1865 | ||
1866 | if (sched_can_stop_tick(rq)) | |
1867 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1868 | else | |
1869 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); | |
1870 | } | |
1871 | #else | |
d84b3131 | 1872 | static inline int sched_tick_offload_init(void) { return 0; } |
76d92ac3 FW |
1873 | static inline void sched_update_tick_dependency(struct rq *rq) { } |
1874 | #endif | |
1875 | ||
72465447 | 1876 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 1877 | { |
72465447 KT |
1878 | unsigned prev_nr = rq->nr_running; |
1879 | ||
1880 | rq->nr_running = prev_nr + count; | |
9f3660c2 | 1881 | |
4486edd1 | 1882 | #ifdef CONFIG_SMP |
3e184501 | 1883 | if (prev_nr < 2 && rq->nr_running >= 2) { |
e90c8fe1 VS |
1884 | if (!READ_ONCE(rq->rd->overload)) |
1885 | WRITE_ONCE(rq->rd->overload, 1); | |
4486edd1 | 1886 | } |
3e184501 | 1887 | #endif |
76d92ac3 FW |
1888 | |
1889 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1890 | } |
1891 | ||
72465447 | 1892 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 1893 | { |
72465447 | 1894 | rq->nr_running -= count; |
76d92ac3 FW |
1895 | /* Check if we still need preemption */ |
1896 | sched_update_tick_dependency(rq); | |
029632fb PZ |
1897 | } |
1898 | ||
029632fb PZ |
1899 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); |
1900 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
1901 | ||
1902 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
1903 | ||
029632fb PZ |
1904 | extern const_debug unsigned int sysctl_sched_nr_migrate; |
1905 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
1906 | ||
029632fb PZ |
1907 | #ifdef CONFIG_SCHED_HRTICK |
1908 | ||
1909 | /* | |
1910 | * Use hrtick when: | |
1911 | * - enabled by features | |
1912 | * - hrtimer is actually high res | |
1913 | */ | |
1914 | static inline int hrtick_enabled(struct rq *rq) | |
1915 | { | |
1916 | if (!sched_feat(HRTICK)) | |
1917 | return 0; | |
1918 | if (!cpu_active(cpu_of(rq))) | |
1919 | return 0; | |
1920 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
1921 | } | |
1922 | ||
1923 | void hrtick_start(struct rq *rq, u64 delay); | |
1924 | ||
b39e66ea MG |
1925 | #else |
1926 | ||
1927 | static inline int hrtick_enabled(struct rq *rq) | |
1928 | { | |
1929 | return 0; | |
1930 | } | |
1931 | ||
029632fb PZ |
1932 | #endif /* CONFIG_SCHED_HRTICK */ |
1933 | ||
dfbca41f PZ |
1934 | #ifndef arch_scale_freq_capacity |
1935 | static __always_inline | |
7673c8a4 | 1936 | unsigned long arch_scale_freq_capacity(int cpu) |
dfbca41f PZ |
1937 | { |
1938 | return SCHED_CAPACITY_SCALE; | |
1939 | } | |
1940 | #endif | |
b5b4860d | 1941 | |
029632fb PZ |
1942 | #ifdef CONFIG_SMP |
1943 | #ifdef CONFIG_PREEMPT | |
1944 | ||
1945 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
1946 | ||
1947 | /* | |
1948 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
1949 | * way at the expense of forcing extra atomic operations in all | |
1950 | * invocations. This assures that the double_lock is acquired using the | |
1951 | * same underlying policy as the spinlock_t on this architecture, which | |
1952 | * reduces latency compared to the unfair variant below. However, it | |
1953 | * also adds more overhead and therefore may reduce throughput. | |
1954 | */ | |
1955 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1956 | __releases(this_rq->lock) | |
1957 | __acquires(busiest->lock) | |
1958 | __acquires(this_rq->lock) | |
1959 | { | |
1960 | raw_spin_unlock(&this_rq->lock); | |
1961 | double_rq_lock(this_rq, busiest); | |
1962 | ||
1963 | return 1; | |
1964 | } | |
1965 | ||
1966 | #else | |
1967 | /* | |
1968 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1969 | * latency by eliminating extra atomic operations when the locks are | |
97fb7a0a IM |
1970 | * already in proper order on entry. This favors lower CPU-ids and will |
1971 | * grant the double lock to lower CPUs over higher ids under contention, | |
029632fb PZ |
1972 | * regardless of entry order into the function. |
1973 | */ | |
1974 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1975 | __releases(this_rq->lock) | |
1976 | __acquires(busiest->lock) | |
1977 | __acquires(this_rq->lock) | |
1978 | { | |
1979 | int ret = 0; | |
1980 | ||
1981 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
1982 | if (busiest < this_rq) { | |
1983 | raw_spin_unlock(&this_rq->lock); | |
1984 | raw_spin_lock(&busiest->lock); | |
1985 | raw_spin_lock_nested(&this_rq->lock, | |
1986 | SINGLE_DEPTH_NESTING); | |
1987 | ret = 1; | |
1988 | } else | |
1989 | raw_spin_lock_nested(&busiest->lock, | |
1990 | SINGLE_DEPTH_NESTING); | |
1991 | } | |
1992 | return ret; | |
1993 | } | |
1994 | ||
1995 | #endif /* CONFIG_PREEMPT */ | |
1996 | ||
1997 | /* | |
1998 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1999 | */ | |
2000 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
2001 | { | |
2002 | if (unlikely(!irqs_disabled())) { | |
97fb7a0a | 2003 | /* printk() doesn't work well under rq->lock */ |
029632fb PZ |
2004 | raw_spin_unlock(&this_rq->lock); |
2005 | BUG_ON(1); | |
2006 | } | |
2007 | ||
2008 | return _double_lock_balance(this_rq, busiest); | |
2009 | } | |
2010 | ||
2011 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
2012 | __releases(busiest->lock) | |
2013 | { | |
2014 | raw_spin_unlock(&busiest->lock); | |
2015 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2016 | } | |
2017 | ||
74602315 PZ |
2018 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
2019 | { | |
2020 | if (l1 > l2) | |
2021 | swap(l1, l2); | |
2022 | ||
2023 | spin_lock(l1); | |
2024 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2025 | } | |
2026 | ||
60e69eed MG |
2027 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
2028 | { | |
2029 | if (l1 > l2) | |
2030 | swap(l1, l2); | |
2031 | ||
2032 | spin_lock_irq(l1); | |
2033 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2034 | } | |
2035 | ||
74602315 PZ |
2036 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
2037 | { | |
2038 | if (l1 > l2) | |
2039 | swap(l1, l2); | |
2040 | ||
2041 | raw_spin_lock(l1); | |
2042 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2043 | } | |
2044 | ||
029632fb PZ |
2045 | /* |
2046 | * double_rq_lock - safely lock two runqueues | |
2047 | * | |
2048 | * Note this does not disable interrupts like task_rq_lock, | |
2049 | * you need to do so manually before calling. | |
2050 | */ | |
2051 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
2052 | __acquires(rq1->lock) | |
2053 | __acquires(rq2->lock) | |
2054 | { | |
2055 | BUG_ON(!irqs_disabled()); | |
2056 | if (rq1 == rq2) { | |
2057 | raw_spin_lock(&rq1->lock); | |
2058 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2059 | } else { | |
2060 | if (rq1 < rq2) { | |
2061 | raw_spin_lock(&rq1->lock); | |
2062 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
2063 | } else { | |
2064 | raw_spin_lock(&rq2->lock); | |
2065 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
2066 | } | |
2067 | } | |
2068 | } | |
2069 | ||
2070 | /* | |
2071 | * double_rq_unlock - safely unlock two runqueues | |
2072 | * | |
2073 | * Note this does not restore interrupts like task_rq_unlock, | |
2074 | * you need to do so manually after calling. | |
2075 | */ | |
2076 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
2077 | __releases(rq1->lock) | |
2078 | __releases(rq2->lock) | |
2079 | { | |
2080 | raw_spin_unlock(&rq1->lock); | |
2081 | if (rq1 != rq2) | |
2082 | raw_spin_unlock(&rq2->lock); | |
2083 | else | |
2084 | __release(rq2->lock); | |
2085 | } | |
2086 | ||
f2cb1360 IM |
2087 | extern void set_rq_online (struct rq *rq); |
2088 | extern void set_rq_offline(struct rq *rq); | |
2089 | extern bool sched_smp_initialized; | |
2090 | ||
029632fb PZ |
2091 | #else /* CONFIG_SMP */ |
2092 | ||
2093 | /* | |
2094 | * double_rq_lock - safely lock two runqueues | |
2095 | * | |
2096 | * Note this does not disable interrupts like task_rq_lock, | |
2097 | * you need to do so manually before calling. | |
2098 | */ | |
2099 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
2100 | __acquires(rq1->lock) | |
2101 | __acquires(rq2->lock) | |
2102 | { | |
2103 | BUG_ON(!irqs_disabled()); | |
2104 | BUG_ON(rq1 != rq2); | |
2105 | raw_spin_lock(&rq1->lock); | |
2106 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * double_rq_unlock - safely unlock two runqueues | |
2111 | * | |
2112 | * Note this does not restore interrupts like task_rq_unlock, | |
2113 | * you need to do so manually after calling. | |
2114 | */ | |
2115 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
2116 | __releases(rq1->lock) | |
2117 | __releases(rq2->lock) | |
2118 | { | |
2119 | BUG_ON(rq1 != rq2); | |
2120 | raw_spin_unlock(&rq1->lock); | |
2121 | __release(rq2->lock); | |
2122 | } | |
2123 | ||
2124 | #endif | |
2125 | ||
2126 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
2127 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
2128 | |
2129 | #ifdef CONFIG_SCHED_DEBUG | |
9469eb01 PZ |
2130 | extern bool sched_debug_enabled; |
2131 | ||
029632fb PZ |
2132 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
2133 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 2134 | extern void print_dl_stats(struct seq_file *m, int cpu); |
f6a34630 MM |
2135 | extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); |
2136 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2137 | extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); | |
397f2378 SD |
2138 | #ifdef CONFIG_NUMA_BALANCING |
2139 | extern void | |
2140 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
2141 | extern void | |
2142 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
2143 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
2144 | #endif /* CONFIG_NUMA_BALANCING */ | |
2145 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
2146 | |
2147 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
2148 | extern void init_rt_rq(struct rt_rq *rt_rq); |
2149 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 2150 | |
1ee14e6c BS |
2151 | extern void cfs_bandwidth_usage_inc(void); |
2152 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 2153 | |
3451d024 | 2154 | #ifdef CONFIG_NO_HZ_COMMON |
00357f5e PZ |
2155 | #define NOHZ_BALANCE_KICK_BIT 0 |
2156 | #define NOHZ_STATS_KICK_BIT 1 | |
a22e47a4 | 2157 | |
a22e47a4 | 2158 | #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) |
b7031a02 PZ |
2159 | #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) |
2160 | ||
2161 | #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) | |
1c792db7 SS |
2162 | |
2163 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
20a5c8cc | 2164 | |
00357f5e | 2165 | extern void nohz_balance_exit_idle(struct rq *rq); |
20a5c8cc | 2166 | #else |
00357f5e | 2167 | static inline void nohz_balance_exit_idle(struct rq *rq) { } |
1c792db7 | 2168 | #endif |
73fbec60 | 2169 | |
daec5798 LA |
2170 | |
2171 | #ifdef CONFIG_SMP | |
2172 | static inline | |
2173 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2174 | { | |
2175 | struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); | |
2176 | int i; | |
2177 | ||
2178 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), | |
2179 | "sched RCU must be held"); | |
2180 | for_each_cpu_and(i, rd->span, cpu_active_mask) { | |
2181 | struct rq *rq = cpu_rq(i); | |
2182 | ||
2183 | rq->dl.extra_bw += bw; | |
2184 | } | |
2185 | } | |
2186 | #else | |
2187 | static inline | |
2188 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2189 | { | |
2190 | struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); | |
2191 | ||
2192 | dl->extra_bw += bw; | |
2193 | } | |
2194 | #endif | |
2195 | ||
2196 | ||
73fbec60 | 2197 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
19d23dbf | 2198 | struct irqtime { |
25e2d8c1 | 2199 | u64 total; |
a499a5a1 | 2200 | u64 tick_delta; |
19d23dbf FW |
2201 | u64 irq_start_time; |
2202 | struct u64_stats_sync sync; | |
2203 | }; | |
73fbec60 | 2204 | |
19d23dbf | 2205 | DECLARE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 2206 | |
25e2d8c1 FW |
2207 | /* |
2208 | * Returns the irqtime minus the softirq time computed by ksoftirqd. | |
2209 | * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime | |
2210 | * and never move forward. | |
2211 | */ | |
73fbec60 FW |
2212 | static inline u64 irq_time_read(int cpu) |
2213 | { | |
19d23dbf FW |
2214 | struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); |
2215 | unsigned int seq; | |
2216 | u64 total; | |
73fbec60 FW |
2217 | |
2218 | do { | |
19d23dbf | 2219 | seq = __u64_stats_fetch_begin(&irqtime->sync); |
25e2d8c1 | 2220 | total = irqtime->total; |
19d23dbf | 2221 | } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); |
73fbec60 | 2222 | |
19d23dbf | 2223 | return total; |
73fbec60 | 2224 | } |
73fbec60 | 2225 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
adaf9fcd RW |
2226 | |
2227 | #ifdef CONFIG_CPU_FREQ | |
b10abd0a | 2228 | DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data); |
adaf9fcd RW |
2229 | |
2230 | /** | |
2231 | * cpufreq_update_util - Take a note about CPU utilization changes. | |
12bde33d | 2232 | * @rq: Runqueue to carry out the update for. |
58919e83 | 2233 | * @flags: Update reason flags. |
adaf9fcd | 2234 | * |
58919e83 RW |
2235 | * This function is called by the scheduler on the CPU whose utilization is |
2236 | * being updated. | |
adaf9fcd RW |
2237 | * |
2238 | * It can only be called from RCU-sched read-side critical sections. | |
adaf9fcd RW |
2239 | * |
2240 | * The way cpufreq is currently arranged requires it to evaluate the CPU | |
2241 | * performance state (frequency/voltage) on a regular basis to prevent it from | |
2242 | * being stuck in a completely inadequate performance level for too long. | |
e0367b12 JL |
2243 | * That is not guaranteed to happen if the updates are only triggered from CFS |
2244 | * and DL, though, because they may not be coming in if only RT tasks are | |
2245 | * active all the time (or there are RT tasks only). | |
adaf9fcd | 2246 | * |
e0367b12 JL |
2247 | * As a workaround for that issue, this function is called periodically by the |
2248 | * RT sched class to trigger extra cpufreq updates to prevent it from stalling, | |
adaf9fcd | 2249 | * but that really is a band-aid. Going forward it should be replaced with |
e0367b12 | 2250 | * solutions targeted more specifically at RT tasks. |
adaf9fcd | 2251 | */ |
12bde33d | 2252 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) |
adaf9fcd | 2253 | { |
58919e83 RW |
2254 | struct update_util_data *data; |
2255 | ||
674e7541 VK |
2256 | data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, |
2257 | cpu_of(rq))); | |
58919e83 | 2258 | if (data) |
12bde33d RW |
2259 | data->func(data, rq_clock(rq), flags); |
2260 | } | |
adaf9fcd | 2261 | #else |
12bde33d | 2262 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} |
adaf9fcd | 2263 | #endif /* CONFIG_CPU_FREQ */ |
be53f58f | 2264 | |
982d9cdc | 2265 | #ifdef CONFIG_UCLAMP_TASK |
9d20ad7d PB |
2266 | unsigned int uclamp_eff_value(struct task_struct *p, unsigned int clamp_id); |
2267 | ||
2268 | static __always_inline | |
2269 | unsigned int uclamp_util_with(struct rq *rq, unsigned int util, | |
2270 | struct task_struct *p) | |
982d9cdc PB |
2271 | { |
2272 | unsigned int min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value); | |
2273 | unsigned int max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value); | |
2274 | ||
9d20ad7d PB |
2275 | if (p) { |
2276 | min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN)); | |
2277 | max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX)); | |
2278 | } | |
2279 | ||
982d9cdc PB |
2280 | /* |
2281 | * Since CPU's {min,max}_util clamps are MAX aggregated considering | |
2282 | * RUNNABLE tasks with _different_ clamps, we can end up with an | |
2283 | * inversion. Fix it now when the clamps are applied. | |
2284 | */ | |
2285 | if (unlikely(min_util >= max_util)) | |
2286 | return min_util; | |
2287 | ||
2288 | return clamp(util, min_util, max_util); | |
2289 | } | |
9d20ad7d PB |
2290 | |
2291 | static inline unsigned int uclamp_util(struct rq *rq, unsigned int util) | |
2292 | { | |
2293 | return uclamp_util_with(rq, util, NULL); | |
2294 | } | |
982d9cdc | 2295 | #else /* CONFIG_UCLAMP_TASK */ |
9d20ad7d PB |
2296 | static inline unsigned int uclamp_util_with(struct rq *rq, unsigned int util, |
2297 | struct task_struct *p) | |
2298 | { | |
2299 | return util; | |
2300 | } | |
982d9cdc PB |
2301 | static inline unsigned int uclamp_util(struct rq *rq, unsigned int util) |
2302 | { | |
2303 | return util; | |
2304 | } | |
2305 | #endif /* CONFIG_UCLAMP_TASK */ | |
2306 | ||
9bdcb44e | 2307 | #ifdef arch_scale_freq_capacity |
97fb7a0a IM |
2308 | # ifndef arch_scale_freq_invariant |
2309 | # define arch_scale_freq_invariant() true | |
2310 | # endif | |
2311 | #else | |
2312 | # define arch_scale_freq_invariant() false | |
9bdcb44e | 2313 | #endif |
d4edd662 | 2314 | |
10a35e68 VG |
2315 | #ifdef CONFIG_SMP |
2316 | static inline unsigned long capacity_orig_of(int cpu) | |
2317 | { | |
2318 | return cpu_rq(cpu)->cpu_capacity_orig; | |
2319 | } | |
2320 | #endif | |
2321 | ||
938e5e4b QP |
2322 | /** |
2323 | * enum schedutil_type - CPU utilization type | |
2324 | * @FREQUENCY_UTIL: Utilization used to select frequency | |
2325 | * @ENERGY_UTIL: Utilization used during energy calculation | |
2326 | * | |
2327 | * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time | |
2328 | * need to be aggregated differently depending on the usage made of them. This | |
2329 | * enum is used within schedutil_freq_util() to differentiate the types of | |
2330 | * utilization expected by the callers, and adjust the aggregation accordingly. | |
2331 | */ | |
2332 | enum schedutil_type { | |
2333 | FREQUENCY_UTIL, | |
2334 | ENERGY_UTIL, | |
2335 | }; | |
2336 | ||
af24bde8 | 2337 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL |
938e5e4b | 2338 | |
af24bde8 PB |
2339 | unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, |
2340 | unsigned long max, enum schedutil_type type, | |
2341 | struct task_struct *p); | |
938e5e4b | 2342 | |
8cc90515 | 2343 | static inline unsigned long cpu_bw_dl(struct rq *rq) |
d4edd662 JL |
2344 | { |
2345 | return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; | |
2346 | } | |
2347 | ||
8cc90515 VG |
2348 | static inline unsigned long cpu_util_dl(struct rq *rq) |
2349 | { | |
2350 | return READ_ONCE(rq->avg_dl.util_avg); | |
2351 | } | |
2352 | ||
d4edd662 JL |
2353 | static inline unsigned long cpu_util_cfs(struct rq *rq) |
2354 | { | |
a07630b8 PB |
2355 | unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); |
2356 | ||
2357 | if (sched_feat(UTIL_EST)) { | |
2358 | util = max_t(unsigned long, util, | |
2359 | READ_ONCE(rq->cfs.avg.util_est.enqueued)); | |
2360 | } | |
2361 | ||
2362 | return util; | |
d4edd662 | 2363 | } |
371bf427 VG |
2364 | |
2365 | static inline unsigned long cpu_util_rt(struct rq *rq) | |
2366 | { | |
dfa444dc | 2367 | return READ_ONCE(rq->avg_rt.util_avg); |
371bf427 | 2368 | } |
938e5e4b | 2369 | #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
af24bde8 PB |
2370 | static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, |
2371 | unsigned long max, enum schedutil_type type, | |
2372 | struct task_struct *p) | |
938e5e4b | 2373 | { |
af24bde8 | 2374 | return 0; |
938e5e4b | 2375 | } |
af24bde8 | 2376 | #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
9033ea11 | 2377 | |
11d4afd4 | 2378 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
9033ea11 VG |
2379 | static inline unsigned long cpu_util_irq(struct rq *rq) |
2380 | { | |
2381 | return rq->avg_irq.util_avg; | |
2382 | } | |
2e62c474 VG |
2383 | |
2384 | static inline | |
2385 | unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) | |
2386 | { | |
2387 | util *= (max - irq); | |
2388 | util /= max; | |
2389 | ||
2390 | return util; | |
2391 | ||
2392 | } | |
9033ea11 VG |
2393 | #else |
2394 | static inline unsigned long cpu_util_irq(struct rq *rq) | |
2395 | { | |
2396 | return 0; | |
2397 | } | |
2398 | ||
2e62c474 VG |
2399 | static inline |
2400 | unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) | |
2401 | { | |
2402 | return util; | |
2403 | } | |
794a56eb | 2404 | #endif |
6aa140fa | 2405 | |
531b5c9f | 2406 | #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) |
f8a696f2 | 2407 | |
6aa140fa | 2408 | #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus))) |
f8a696f2 PZ |
2409 | |
2410 | DECLARE_STATIC_KEY_FALSE(sched_energy_present); | |
2411 | ||
2412 | static inline bool sched_energy_enabled(void) | |
2413 | { | |
2414 | return static_branch_unlikely(&sched_energy_present); | |
2415 | } | |
2416 | ||
2417 | #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */ | |
2418 | ||
6aa140fa | 2419 | #define perf_domain_span(pd) NULL |
f8a696f2 | 2420 | static inline bool sched_energy_enabled(void) { return false; } |
1f74de87 | 2421 | |
f8a696f2 | 2422 | #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |