]>
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 | |
9d246053 PA |
78 | #include <trace/events/sched.h> |
79 | ||
9148a3a1 | 80 | #ifdef CONFIG_SCHED_DEBUG |
6d3aed3d | 81 | # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) |
9148a3a1 | 82 | #else |
6d3aed3d | 83 | # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) |
9148a3a1 PZ |
84 | #endif |
85 | ||
45ceebf7 | 86 | struct rq; |
442bf3aa | 87 | struct cpuidle_state; |
45ceebf7 | 88 | |
da0c1e65 KT |
89 | /* task_struct::on_rq states: */ |
90 | #define TASK_ON_RQ_QUEUED 1 | |
cca26e80 | 91 | #define TASK_ON_RQ_MIGRATING 2 |
da0c1e65 | 92 | |
029632fb PZ |
93 | extern __read_mostly int scheduler_running; |
94 | ||
45ceebf7 PG |
95 | extern unsigned long calc_load_update; |
96 | extern atomic_long_t calc_load_tasks; | |
97 | ||
3289bdb4 | 98 | extern void calc_global_load_tick(struct rq *this_rq); |
d60585c5 | 99 | extern long calc_load_fold_active(struct rq *this_rq, long adjust); |
3289bdb4 | 100 | |
9d246053 | 101 | extern void call_trace_sched_update_nr_running(struct rq *rq, int count); |
029632fb PZ |
102 | /* |
103 | * Helpers for converting nanosecond timing to jiffy resolution | |
104 | */ | |
105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) | |
106 | ||
cc1f4b1f LZ |
107 | /* |
108 | * Increase resolution of nice-level calculations for 64-bit architectures. | |
109 | * The extra resolution improves shares distribution and load balancing of | |
110 | * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup | |
111 | * hierarchies, especially on larger systems. This is not a user-visible change | |
112 | * and does not change the user-interface for setting shares/weights. | |
113 | * | |
114 | * We increase resolution only if we have enough bits to allow this increased | |
97fb7a0a IM |
115 | * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit |
116 | * are pretty high and the returns do not justify the increased costs. | |
2159197d | 117 | * |
97fb7a0a IM |
118 | * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to |
119 | * increase coverage and consistency always enable it on 64-bit platforms. | |
cc1f4b1f | 120 | */ |
2159197d | 121 | #ifdef CONFIG_64BIT |
172895e6 | 122 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) |
6ecdd749 | 123 | # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) |
26cf5222 MW |
124 | # define scale_load_down(w) \ |
125 | ({ \ | |
126 | unsigned long __w = (w); \ | |
127 | if (__w) \ | |
128 | __w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \ | |
129 | __w; \ | |
130 | }) | |
cc1f4b1f | 131 | #else |
172895e6 | 132 | # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) |
cc1f4b1f LZ |
133 | # define scale_load(w) (w) |
134 | # define scale_load_down(w) (w) | |
135 | #endif | |
136 | ||
6ecdd749 | 137 | /* |
172895e6 YD |
138 | * Task weight (visible to users) and its load (invisible to users) have |
139 | * independent resolution, but they should be well calibrated. We use | |
140 | * scale_load() and scale_load_down(w) to convert between them. The | |
141 | * following must be true: | |
142 | * | |
143 | * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD | |
144 | * | |
6ecdd749 | 145 | */ |
172895e6 | 146 | #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) |
029632fb | 147 | |
332ac17e DF |
148 | /* |
149 | * Single value that decides SCHED_DEADLINE internal math precision. | |
150 | * 10 -> just above 1us | |
151 | * 9 -> just above 0.5us | |
152 | */ | |
97fb7a0a | 153 | #define DL_SCALE 10 |
029632fb PZ |
154 | |
155 | /* | |
97fb7a0a | 156 | * Single value that denotes runtime == period, ie unlimited time. |
029632fb | 157 | */ |
97fb7a0a | 158 | #define RUNTIME_INF ((u64)~0ULL) |
029632fb | 159 | |
20f9cd2a HA |
160 | static inline int idle_policy(int policy) |
161 | { | |
162 | return policy == SCHED_IDLE; | |
163 | } | |
d50dde5a DF |
164 | static inline int fair_policy(int policy) |
165 | { | |
166 | return policy == SCHED_NORMAL || policy == SCHED_BATCH; | |
167 | } | |
168 | ||
029632fb PZ |
169 | static inline int rt_policy(int policy) |
170 | { | |
d50dde5a | 171 | return policy == SCHED_FIFO || policy == SCHED_RR; |
029632fb PZ |
172 | } |
173 | ||
aab03e05 DF |
174 | static inline int dl_policy(int policy) |
175 | { | |
176 | return policy == SCHED_DEADLINE; | |
177 | } | |
20f9cd2a HA |
178 | static inline bool valid_policy(int policy) |
179 | { | |
180 | return idle_policy(policy) || fair_policy(policy) || | |
181 | rt_policy(policy) || dl_policy(policy); | |
182 | } | |
aab03e05 | 183 | |
1da1843f VK |
184 | static inline int task_has_idle_policy(struct task_struct *p) |
185 | { | |
186 | return idle_policy(p->policy); | |
187 | } | |
188 | ||
029632fb PZ |
189 | static inline int task_has_rt_policy(struct task_struct *p) |
190 | { | |
191 | return rt_policy(p->policy); | |
192 | } | |
193 | ||
aab03e05 DF |
194 | static inline int task_has_dl_policy(struct task_struct *p) |
195 | { | |
196 | return dl_policy(p->policy); | |
197 | } | |
198 | ||
07881166 JL |
199 | #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) |
200 | ||
d76343c6 VS |
201 | static inline void update_avg(u64 *avg, u64 sample) |
202 | { | |
203 | s64 diff = sample - *avg; | |
204 | *avg += diff / 8; | |
205 | } | |
206 | ||
914497a1 VS |
207 | /* |
208 | * Shifting a value by an exponent greater *or equal* to the size of said value | |
209 | * is UB; cap at size-1. | |
210 | */ | |
211 | #define shr_bound(val, shift) \ | |
212 | (val >> min_t(typeof(shift), shift, BITS_PER_TYPE(typeof(val)) - 1)) | |
213 | ||
794a56eb JL |
214 | /* |
215 | * !! For sched_setattr_nocheck() (kernel) only !! | |
216 | * | |
217 | * This is actually gross. :( | |
218 | * | |
219 | * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE | |
220 | * tasks, but still be able to sleep. We need this on platforms that cannot | |
221 | * atomically change clock frequency. Remove once fast switching will be | |
222 | * available on such platforms. | |
223 | * | |
224 | * SUGOV stands for SchedUtil GOVernor. | |
225 | */ | |
226 | #define SCHED_FLAG_SUGOV 0x10000000 | |
227 | ||
228 | static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) | |
229 | { | |
230 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL | |
231 | return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); | |
232 | #else | |
233 | return false; | |
234 | #endif | |
235 | } | |
236 | ||
2d3d891d DF |
237 | /* |
238 | * Tells if entity @a should preempt entity @b. | |
239 | */ | |
332ac17e DF |
240 | static inline bool |
241 | dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) | |
2d3d891d | 242 | { |
794a56eb JL |
243 | return dl_entity_is_special(a) || |
244 | dl_time_before(a->deadline, b->deadline); | |
2d3d891d DF |
245 | } |
246 | ||
029632fb PZ |
247 | /* |
248 | * This is the priority-queue data structure of the RT scheduling class: | |
249 | */ | |
250 | struct rt_prio_array { | |
251 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
252 | struct list_head queue[MAX_RT_PRIO]; | |
253 | }; | |
254 | ||
255 | struct rt_bandwidth { | |
256 | /* nests inside the rq lock: */ | |
257 | raw_spinlock_t rt_runtime_lock; | |
258 | ktime_t rt_period; | |
259 | u64 rt_runtime; | |
260 | struct hrtimer rt_period_timer; | |
4cfafd30 | 261 | unsigned int rt_period_active; |
029632fb | 262 | }; |
a5e7be3b JL |
263 | |
264 | void __dl_clear_params(struct task_struct *p); | |
265 | ||
332ac17e | 266 | struct dl_bandwidth { |
97fb7a0a IM |
267 | raw_spinlock_t dl_runtime_lock; |
268 | u64 dl_runtime; | |
269 | u64 dl_period; | |
332ac17e DF |
270 | }; |
271 | ||
272 | static inline int dl_bandwidth_enabled(void) | |
273 | { | |
1724813d | 274 | return sysctl_sched_rt_runtime >= 0; |
332ac17e DF |
275 | } |
276 | ||
a57415f5 PL |
277 | /* |
278 | * To keep the bandwidth of -deadline tasks under control | |
279 | * we need some place where: | |
280 | * - store the maximum -deadline bandwidth of each cpu; | |
281 | * - cache the fraction of bandwidth that is currently allocated in | |
282 | * each root domain; | |
283 | * | |
284 | * This is all done in the data structure below. It is similar to the | |
285 | * one used for RT-throttling (rt_bandwidth), with the main difference | |
286 | * that, since here we are only interested in admission control, we | |
287 | * do not decrease any runtime while the group "executes", neither we | |
288 | * need a timer to replenish it. | |
289 | * | |
290 | * With respect to SMP, bandwidth is given on a per root domain basis, | |
291 | * meaning that: | |
292 | * - bw (< 100%) is the deadline bandwidth of each CPU; | |
293 | * - total_bw is the currently allocated bandwidth in each root domain; | |
294 | */ | |
332ac17e | 295 | struct dl_bw { |
97fb7a0a IM |
296 | raw_spinlock_t lock; |
297 | u64 bw; | |
298 | u64 total_bw; | |
332ac17e DF |
299 | }; |
300 | ||
daec5798 LA |
301 | static inline void __dl_update(struct dl_bw *dl_b, s64 bw); |
302 | ||
7f51412a | 303 | static inline |
8c0944ce | 304 | void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
305 | { |
306 | dl_b->total_bw -= tsk_bw; | |
daec5798 | 307 | __dl_update(dl_b, (s32)tsk_bw / cpus); |
7f51412a JL |
308 | } |
309 | ||
310 | static inline | |
daec5798 | 311 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) |
7f51412a JL |
312 | { |
313 | dl_b->total_bw += tsk_bw; | |
daec5798 | 314 | __dl_update(dl_b, -((s32)tsk_bw / cpus)); |
7f51412a JL |
315 | } |
316 | ||
60ffd5ed LA |
317 | static inline bool __dl_overflow(struct dl_bw *dl_b, unsigned long cap, |
318 | u64 old_bw, u64 new_bw) | |
7f51412a JL |
319 | { |
320 | return dl_b->bw != -1 && | |
60ffd5ed | 321 | cap_scale(dl_b->bw, cap) < dl_b->total_bw - old_bw + new_bw; |
7f51412a JL |
322 | } |
323 | ||
b4118988 LA |
324 | /* |
325 | * Verify the fitness of task @p to run on @cpu taking into account the | |
326 | * CPU original capacity and the runtime/deadline ratio of the task. | |
327 | * | |
328 | * The function will return true if the CPU original capacity of the | |
329 | * @cpu scaled by SCHED_CAPACITY_SCALE >= runtime/deadline ratio of the | |
330 | * task and false otherwise. | |
331 | */ | |
332 | static inline bool dl_task_fits_capacity(struct task_struct *p, int cpu) | |
333 | { | |
334 | unsigned long cap = arch_scale_cpu_capacity(cpu); | |
335 | ||
336 | return cap_scale(p->dl.dl_deadline, cap) >= p->dl.dl_runtime; | |
337 | } | |
338 | ||
f2cb1360 | 339 | extern void init_dl_bw(struct dl_bw *dl_b); |
97fb7a0a | 340 | extern int sched_dl_global_validate(void); |
06a76fe0 | 341 | extern void sched_dl_do_global(void); |
97fb7a0a | 342 | extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); |
06a76fe0 NP |
343 | extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); |
344 | extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); | |
345 | extern bool __checkparam_dl(const struct sched_attr *attr); | |
06a76fe0 | 346 | extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); |
97fb7a0a IM |
347 | extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); |
348 | extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); | |
06a76fe0 | 349 | extern bool dl_cpu_busy(unsigned int cpu); |
029632fb PZ |
350 | |
351 | #ifdef CONFIG_CGROUP_SCHED | |
352 | ||
353 | #include <linux/cgroup.h> | |
eb414681 | 354 | #include <linux/psi.h> |
029632fb PZ |
355 | |
356 | struct cfs_rq; | |
357 | struct rt_rq; | |
358 | ||
35cf4e50 | 359 | extern struct list_head task_groups; |
029632fb PZ |
360 | |
361 | struct cfs_bandwidth { | |
362 | #ifdef CONFIG_CFS_BANDWIDTH | |
97fb7a0a IM |
363 | raw_spinlock_t lock; |
364 | ktime_t period; | |
365 | u64 quota; | |
366 | u64 runtime; | |
367 | s64 hierarchical_quota; | |
97fb7a0a | 368 | |
66567fcb | 369 | u8 idle; |
370 | u8 period_active; | |
66567fcb | 371 | u8 slack_started; |
97fb7a0a IM |
372 | struct hrtimer period_timer; |
373 | struct hrtimer slack_timer; | |
374 | struct list_head throttled_cfs_rq; | |
375 | ||
376 | /* Statistics: */ | |
377 | int nr_periods; | |
378 | int nr_throttled; | |
379 | u64 throttled_time; | |
029632fb PZ |
380 | #endif |
381 | }; | |
382 | ||
97fb7a0a | 383 | /* Task group related information */ |
029632fb PZ |
384 | struct task_group { |
385 | struct cgroup_subsys_state css; | |
386 | ||
387 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
97fb7a0a IM |
388 | /* schedulable entities of this group on each CPU */ |
389 | struct sched_entity **se; | |
390 | /* runqueue "owned" by this group on each CPU */ | |
391 | struct cfs_rq **cfs_rq; | |
392 | unsigned long shares; | |
029632fb | 393 | |
fa6bddeb | 394 | #ifdef CONFIG_SMP |
b0367629 WL |
395 | /* |
396 | * load_avg can be heavily contended at clock tick time, so put | |
397 | * it in its own cacheline separated from the fields above which | |
398 | * will also be accessed at each tick. | |
399 | */ | |
97fb7a0a | 400 | atomic_long_t load_avg ____cacheline_aligned; |
029632fb | 401 | #endif |
fa6bddeb | 402 | #endif |
029632fb PZ |
403 | |
404 | #ifdef CONFIG_RT_GROUP_SCHED | |
97fb7a0a IM |
405 | struct sched_rt_entity **rt_se; |
406 | struct rt_rq **rt_rq; | |
029632fb | 407 | |
97fb7a0a | 408 | struct rt_bandwidth rt_bandwidth; |
029632fb PZ |
409 | #endif |
410 | ||
97fb7a0a IM |
411 | struct rcu_head rcu; |
412 | struct list_head list; | |
029632fb | 413 | |
97fb7a0a IM |
414 | struct task_group *parent; |
415 | struct list_head siblings; | |
416 | struct list_head children; | |
029632fb PZ |
417 | |
418 | #ifdef CONFIG_SCHED_AUTOGROUP | |
97fb7a0a | 419 | struct autogroup *autogroup; |
029632fb PZ |
420 | #endif |
421 | ||
97fb7a0a | 422 | struct cfs_bandwidth cfs_bandwidth; |
2480c093 PB |
423 | |
424 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
425 | /* The two decimal precision [%] value requested from user-space */ | |
426 | unsigned int uclamp_pct[UCLAMP_CNT]; | |
427 | /* Clamp values requested for a task group */ | |
428 | struct uclamp_se uclamp_req[UCLAMP_CNT]; | |
0b60ba2d PB |
429 | /* Effective clamp values used for a task group */ |
430 | struct uclamp_se uclamp[UCLAMP_CNT]; | |
2480c093 PB |
431 | #endif |
432 | ||
029632fb PZ |
433 | }; |
434 | ||
435 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
436 | #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD | |
437 | ||
438 | /* | |
439 | * A weight of 0 or 1 can cause arithmetics problems. | |
440 | * A weight of a cfs_rq is the sum of weights of which entities | |
441 | * are queued on this cfs_rq, so a weight of a entity should not be | |
442 | * too large, so as the shares value of a task group. | |
443 | * (The default weight is 1024 - so there's no practical | |
444 | * limitation from this.) | |
445 | */ | |
97fb7a0a IM |
446 | #define MIN_SHARES (1UL << 1) |
447 | #define MAX_SHARES (1UL << 18) | |
029632fb PZ |
448 | #endif |
449 | ||
029632fb PZ |
450 | typedef int (*tg_visitor)(struct task_group *, void *); |
451 | ||
452 | extern int walk_tg_tree_from(struct task_group *from, | |
453 | tg_visitor down, tg_visitor up, void *data); | |
454 | ||
455 | /* | |
456 | * Iterate the full tree, calling @down when first entering a node and @up when | |
457 | * leaving it for the final time. | |
458 | * | |
459 | * Caller must hold rcu_lock or sufficient equivalent. | |
460 | */ | |
461 | static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) | |
462 | { | |
463 | return walk_tg_tree_from(&root_task_group, down, up, data); | |
464 | } | |
465 | ||
466 | extern int tg_nop(struct task_group *tg, void *data); | |
467 | ||
468 | extern void free_fair_sched_group(struct task_group *tg); | |
469 | extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); | |
8663e24d | 470 | extern void online_fair_sched_group(struct task_group *tg); |
6fe1f348 | 471 | extern void unregister_fair_sched_group(struct task_group *tg); |
029632fb PZ |
472 | extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
473 | struct sched_entity *se, int cpu, | |
474 | struct sched_entity *parent); | |
475 | extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); | |
029632fb PZ |
476 | |
477 | extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); | |
77a4d1a1 | 478 | extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); |
029632fb PZ |
479 | extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); |
480 | ||
481 | extern void free_rt_sched_group(struct task_group *tg); | |
482 | extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); | |
483 | extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, | |
484 | struct sched_rt_entity *rt_se, int cpu, | |
485 | struct sched_rt_entity *parent); | |
8887cd99 NP |
486 | extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); |
487 | extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); | |
488 | extern long sched_group_rt_runtime(struct task_group *tg); | |
489 | extern long sched_group_rt_period(struct task_group *tg); | |
490 | extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); | |
029632fb | 491 | |
25cc7da7 LZ |
492 | extern struct task_group *sched_create_group(struct task_group *parent); |
493 | extern void sched_online_group(struct task_group *tg, | |
494 | struct task_group *parent); | |
495 | extern void sched_destroy_group(struct task_group *tg); | |
496 | extern void sched_offline_group(struct task_group *tg); | |
497 | ||
498 | extern void sched_move_task(struct task_struct *tsk); | |
499 | ||
500 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
501 | extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); | |
ad936d86 BP |
502 | |
503 | #ifdef CONFIG_SMP | |
504 | extern void set_task_rq_fair(struct sched_entity *se, | |
505 | struct cfs_rq *prev, struct cfs_rq *next); | |
506 | #else /* !CONFIG_SMP */ | |
507 | static inline void set_task_rq_fair(struct sched_entity *se, | |
508 | struct cfs_rq *prev, struct cfs_rq *next) { } | |
509 | #endif /* CONFIG_SMP */ | |
510 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
25cc7da7 | 511 | |
029632fb PZ |
512 | #else /* CONFIG_CGROUP_SCHED */ |
513 | ||
514 | struct cfs_bandwidth { }; | |
515 | ||
516 | #endif /* CONFIG_CGROUP_SCHED */ | |
517 | ||
518 | /* CFS-related fields in a runqueue */ | |
519 | struct cfs_rq { | |
97fb7a0a | 520 | struct load_weight load; |
97fb7a0a | 521 | unsigned int nr_running; |
43e9f7f2 VK |
522 | unsigned int h_nr_running; /* SCHED_{NORMAL,BATCH,IDLE} */ |
523 | unsigned int idle_h_nr_running; /* SCHED_IDLE */ | |
029632fb | 524 | |
97fb7a0a IM |
525 | u64 exec_clock; |
526 | u64 min_vruntime; | |
029632fb | 527 | #ifndef CONFIG_64BIT |
97fb7a0a | 528 | u64 min_vruntime_copy; |
029632fb PZ |
529 | #endif |
530 | ||
97fb7a0a | 531 | struct rb_root_cached tasks_timeline; |
029632fb | 532 | |
029632fb PZ |
533 | /* |
534 | * 'curr' points to currently running entity on this cfs_rq. | |
535 | * It is set to NULL otherwise (i.e when none are currently running). | |
536 | */ | |
97fb7a0a IM |
537 | struct sched_entity *curr; |
538 | struct sched_entity *next; | |
539 | struct sched_entity *last; | |
540 | struct sched_entity *skip; | |
029632fb PZ |
541 | |
542 | #ifdef CONFIG_SCHED_DEBUG | |
97fb7a0a | 543 | unsigned int nr_spread_over; |
029632fb PZ |
544 | #endif |
545 | ||
2dac754e PT |
546 | #ifdef CONFIG_SMP |
547 | /* | |
9d89c257 | 548 | * CFS load tracking |
2dac754e | 549 | */ |
97fb7a0a | 550 | struct sched_avg avg; |
2a2f5d4e | 551 | #ifndef CONFIG_64BIT |
97fb7a0a | 552 | u64 load_last_update_time_copy; |
9d89c257 | 553 | #endif |
2a2f5d4e PZ |
554 | struct { |
555 | raw_spinlock_t lock ____cacheline_aligned; | |
556 | int nr; | |
557 | unsigned long load_avg; | |
558 | unsigned long util_avg; | |
9f683953 | 559 | unsigned long runnable_avg; |
2a2f5d4e | 560 | } removed; |
82958366 | 561 | |
9d89c257 | 562 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a IM |
563 | unsigned long tg_load_avg_contrib; |
564 | long propagate; | |
565 | long prop_runnable_sum; | |
0e2d2aaa | 566 | |
82958366 PT |
567 | /* |
568 | * h_load = weight * f(tg) | |
569 | * | |
570 | * Where f(tg) is the recursive weight fraction assigned to | |
571 | * this group. | |
572 | */ | |
97fb7a0a IM |
573 | unsigned long h_load; |
574 | u64 last_h_load_update; | |
575 | struct sched_entity *h_load_next; | |
68520796 | 576 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
82958366 PT |
577 | #endif /* CONFIG_SMP */ |
578 | ||
029632fb | 579 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a | 580 | struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ |
029632fb PZ |
581 | |
582 | /* | |
583 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
584 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
585 | * (like users, containers etc.) | |
586 | * | |
97fb7a0a IM |
587 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. |
588 | * This list is used during load balance. | |
029632fb | 589 | */ |
97fb7a0a IM |
590 | int on_list; |
591 | struct list_head leaf_cfs_rq_list; | |
592 | struct task_group *tg; /* group that "owns" this runqueue */ | |
029632fb | 593 | |
029632fb | 594 | #ifdef CONFIG_CFS_BANDWIDTH |
97fb7a0a | 595 | int runtime_enabled; |
97fb7a0a IM |
596 | s64 runtime_remaining; |
597 | ||
598 | u64 throttled_clock; | |
599 | u64 throttled_clock_task; | |
600 | u64 throttled_clock_task_time; | |
601 | int throttled; | |
602 | int throttle_count; | |
603 | struct list_head throttled_list; | |
029632fb PZ |
604 | #endif /* CONFIG_CFS_BANDWIDTH */ |
605 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
606 | }; | |
607 | ||
608 | static inline int rt_bandwidth_enabled(void) | |
609 | { | |
610 | return sysctl_sched_rt_runtime >= 0; | |
611 | } | |
612 | ||
b6366f04 | 613 | /* RT IPI pull logic requires IRQ_WORK */ |
4bdced5c | 614 | #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) |
b6366f04 SR |
615 | # define HAVE_RT_PUSH_IPI |
616 | #endif | |
617 | ||
029632fb PZ |
618 | /* Real-Time classes' related field in a runqueue: */ |
619 | struct rt_rq { | |
97fb7a0a IM |
620 | struct rt_prio_array active; |
621 | unsigned int rt_nr_running; | |
622 | unsigned int rr_nr_running; | |
029632fb PZ |
623 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
624 | struct { | |
97fb7a0a | 625 | int curr; /* highest queued rt task prio */ |
029632fb | 626 | #ifdef CONFIG_SMP |
97fb7a0a | 627 | int next; /* next highest */ |
029632fb PZ |
628 | #endif |
629 | } highest_prio; | |
630 | #endif | |
631 | #ifdef CONFIG_SMP | |
97fb7a0a IM |
632 | unsigned long rt_nr_migratory; |
633 | unsigned long rt_nr_total; | |
634 | int overloaded; | |
635 | struct plist_head pushable_tasks; | |
371bf427 | 636 | |
b6366f04 | 637 | #endif /* CONFIG_SMP */ |
97fb7a0a | 638 | int rt_queued; |
f4ebcbc0 | 639 | |
97fb7a0a IM |
640 | int rt_throttled; |
641 | u64 rt_time; | |
642 | u64 rt_runtime; | |
029632fb | 643 | /* Nests inside the rq lock: */ |
97fb7a0a | 644 | raw_spinlock_t rt_runtime_lock; |
029632fb PZ |
645 | |
646 | #ifdef CONFIG_RT_GROUP_SCHED | |
97fb7a0a | 647 | unsigned long rt_nr_boosted; |
029632fb | 648 | |
97fb7a0a IM |
649 | struct rq *rq; |
650 | struct task_group *tg; | |
029632fb PZ |
651 | #endif |
652 | }; | |
653 | ||
296b2ffe VG |
654 | static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) |
655 | { | |
656 | return rt_rq->rt_queued && rt_rq->rt_nr_running; | |
657 | } | |
658 | ||
aab03e05 DF |
659 | /* Deadline class' related fields in a runqueue */ |
660 | struct dl_rq { | |
661 | /* runqueue is an rbtree, ordered by deadline */ | |
97fb7a0a | 662 | struct rb_root_cached root; |
aab03e05 | 663 | |
97fb7a0a | 664 | unsigned long dl_nr_running; |
1baca4ce JL |
665 | |
666 | #ifdef CONFIG_SMP | |
667 | /* | |
668 | * Deadline values of the currently executing and the | |
669 | * earliest ready task on this rq. Caching these facilitates | |
dfcb245e | 670 | * the decision whether or not a ready but not running task |
1baca4ce JL |
671 | * should migrate somewhere else. |
672 | */ | |
673 | struct { | |
97fb7a0a IM |
674 | u64 curr; |
675 | u64 next; | |
1baca4ce JL |
676 | } earliest_dl; |
677 | ||
97fb7a0a IM |
678 | unsigned long dl_nr_migratory; |
679 | int overloaded; | |
1baca4ce JL |
680 | |
681 | /* | |
682 | * Tasks on this rq that can be pushed away. They are kept in | |
683 | * an rb-tree, ordered by tasks' deadlines, with caching | |
684 | * of the leftmost (earliest deadline) element. | |
685 | */ | |
97fb7a0a | 686 | struct rb_root_cached pushable_dl_tasks_root; |
332ac17e | 687 | #else |
97fb7a0a | 688 | struct dl_bw dl_bw; |
1baca4ce | 689 | #endif |
e36d8677 LA |
690 | /* |
691 | * "Active utilization" for this runqueue: increased when a | |
692 | * task wakes up (becomes TASK_RUNNING) and decreased when a | |
693 | * task blocks | |
694 | */ | |
97fb7a0a | 695 | u64 running_bw; |
4da3abce | 696 | |
8fd27231 LA |
697 | /* |
698 | * Utilization of the tasks "assigned" to this runqueue (including | |
699 | * the tasks that are in runqueue and the tasks that executed on this | |
700 | * CPU and blocked). Increased when a task moves to this runqueue, and | |
701 | * decreased when the task moves away (migrates, changes scheduling | |
702 | * policy, or terminates). | |
703 | * This is needed to compute the "inactive utilization" for the | |
704 | * runqueue (inactive utilization = this_bw - running_bw). | |
705 | */ | |
97fb7a0a IM |
706 | u64 this_bw; |
707 | u64 extra_bw; | |
8fd27231 | 708 | |
4da3abce LA |
709 | /* |
710 | * Inverse of the fraction of CPU utilization that can be reclaimed | |
711 | * by the GRUB algorithm. | |
712 | */ | |
97fb7a0a | 713 | u64 bw_ratio; |
aab03e05 DF |
714 | }; |
715 | ||
c0796298 VG |
716 | #ifdef CONFIG_FAIR_GROUP_SCHED |
717 | /* An entity is a task if it doesn't "own" a runqueue */ | |
718 | #define entity_is_task(se) (!se->my_q) | |
0dacee1b | 719 | |
9f683953 VG |
720 | static inline void se_update_runnable(struct sched_entity *se) |
721 | { | |
722 | if (!entity_is_task(se)) | |
723 | se->runnable_weight = se->my_q->h_nr_running; | |
724 | } | |
725 | ||
726 | static inline long se_runnable(struct sched_entity *se) | |
727 | { | |
728 | if (entity_is_task(se)) | |
729 | return !!se->on_rq; | |
730 | else | |
731 | return se->runnable_weight; | |
732 | } | |
733 | ||
c0796298 VG |
734 | #else |
735 | #define entity_is_task(se) 1 | |
0dacee1b | 736 | |
9f683953 VG |
737 | static inline void se_update_runnable(struct sched_entity *se) {} |
738 | ||
739 | static inline long se_runnable(struct sched_entity *se) | |
740 | { | |
741 | return !!se->on_rq; | |
742 | } | |
c0796298 VG |
743 | #endif |
744 | ||
029632fb | 745 | #ifdef CONFIG_SMP |
c0796298 VG |
746 | /* |
747 | * XXX we want to get rid of these helpers and use the full load resolution. | |
748 | */ | |
749 | static inline long se_weight(struct sched_entity *se) | |
750 | { | |
751 | return scale_load_down(se->load.weight); | |
752 | } | |
753 | ||
029632fb | 754 | |
afe06efd TC |
755 | static inline bool sched_asym_prefer(int a, int b) |
756 | { | |
757 | return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); | |
758 | } | |
759 | ||
6aa140fa QP |
760 | struct perf_domain { |
761 | struct em_perf_domain *em_pd; | |
762 | struct perf_domain *next; | |
763 | struct rcu_head rcu; | |
764 | }; | |
765 | ||
630246a0 QP |
766 | /* Scheduling group status flags */ |
767 | #define SG_OVERLOAD 0x1 /* More than one runnable task on a CPU. */ | |
2802bf3c | 768 | #define SG_OVERUTILIZED 0x2 /* One or more CPUs are over-utilized. */ |
630246a0 | 769 | |
029632fb PZ |
770 | /* |
771 | * We add the notion of a root-domain which will be used to define per-domain | |
772 | * variables. Each exclusive cpuset essentially defines an island domain by | |
97fb7a0a | 773 | * fully partitioning the member CPUs from any other cpuset. Whenever a new |
029632fb PZ |
774 | * exclusive cpuset is created, we also create and attach a new root-domain |
775 | * object. | |
776 | * | |
777 | */ | |
778 | struct root_domain { | |
97fb7a0a IM |
779 | atomic_t refcount; |
780 | atomic_t rto_count; | |
781 | struct rcu_head rcu; | |
782 | cpumask_var_t span; | |
783 | cpumask_var_t online; | |
029632fb | 784 | |
757ffdd7 VS |
785 | /* |
786 | * Indicate pullable load on at least one CPU, e.g: | |
787 | * - More than one runnable task | |
788 | * - Running task is misfit | |
789 | */ | |
575638d1 | 790 | int overload; |
4486edd1 | 791 | |
2802bf3c MR |
792 | /* Indicate one or more cpus over-utilized (tipping point) */ |
793 | int overutilized; | |
794 | ||
1baca4ce JL |
795 | /* |
796 | * The bit corresponding to a CPU gets set here if such CPU has more | |
797 | * than one runnable -deadline task (as it is below for RT tasks). | |
798 | */ | |
97fb7a0a IM |
799 | cpumask_var_t dlo_mask; |
800 | atomic_t dlo_count; | |
801 | struct dl_bw dl_bw; | |
802 | struct cpudl cpudl; | |
1baca4ce | 803 | |
26762423 PL |
804 | /* |
805 | * Indicate whether a root_domain's dl_bw has been checked or | |
806 | * updated. It's monotonously increasing value. | |
807 | * | |
808 | * Also, some corner cases, like 'wrap around' is dangerous, but given | |
809 | * that u64 is 'big enough'. So that shouldn't be a concern. | |
810 | */ | |
811 | u64 visit_gen; | |
812 | ||
4bdced5c SRRH |
813 | #ifdef HAVE_RT_PUSH_IPI |
814 | /* | |
815 | * For IPI pull requests, loop across the rto_mask. | |
816 | */ | |
97fb7a0a IM |
817 | struct irq_work rto_push_work; |
818 | raw_spinlock_t rto_lock; | |
4bdced5c | 819 | /* These are only updated and read within rto_lock */ |
97fb7a0a IM |
820 | int rto_loop; |
821 | int rto_cpu; | |
4bdced5c | 822 | /* These atomics are updated outside of a lock */ |
97fb7a0a IM |
823 | atomic_t rto_loop_next; |
824 | atomic_t rto_loop_start; | |
4bdced5c | 825 | #endif |
029632fb PZ |
826 | /* |
827 | * The "RT overload" flag: it gets set if a CPU has more than | |
828 | * one runnable RT task. | |
829 | */ | |
97fb7a0a IM |
830 | cpumask_var_t rto_mask; |
831 | struct cpupri cpupri; | |
cd92bfd3 | 832 | |
97fb7a0a | 833 | unsigned long max_cpu_capacity; |
6aa140fa QP |
834 | |
835 | /* | |
836 | * NULL-terminated list of performance domains intersecting with the | |
837 | * CPUs of the rd. Protected by RCU. | |
838 | */ | |
7ba7319f | 839 | struct perf_domain __rcu *pd; |
029632fb PZ |
840 | }; |
841 | ||
f2cb1360 | 842 | extern void init_defrootdomain(void); |
8d5dc512 | 843 | extern int sched_init_domains(const struct cpumask *cpu_map); |
f2cb1360 | 844 | extern void rq_attach_root(struct rq *rq, struct root_domain *rd); |
364f5665 SRV |
845 | extern void sched_get_rd(struct root_domain *rd); |
846 | extern void sched_put_rd(struct root_domain *rd); | |
029632fb | 847 | |
4bdced5c SRRH |
848 | #ifdef HAVE_RT_PUSH_IPI |
849 | extern void rto_push_irq_work_func(struct irq_work *work); | |
850 | #endif | |
029632fb PZ |
851 | #endif /* CONFIG_SMP */ |
852 | ||
69842cba PB |
853 | #ifdef CONFIG_UCLAMP_TASK |
854 | /* | |
855 | * struct uclamp_bucket - Utilization clamp bucket | |
856 | * @value: utilization clamp value for tasks on this clamp bucket | |
857 | * @tasks: number of RUNNABLE tasks on this clamp bucket | |
858 | * | |
859 | * Keep track of how many tasks are RUNNABLE for a given utilization | |
860 | * clamp value. | |
861 | */ | |
862 | struct uclamp_bucket { | |
863 | unsigned long value : bits_per(SCHED_CAPACITY_SCALE); | |
864 | unsigned long tasks : BITS_PER_LONG - bits_per(SCHED_CAPACITY_SCALE); | |
865 | }; | |
866 | ||
867 | /* | |
868 | * struct uclamp_rq - rq's utilization clamp | |
869 | * @value: currently active clamp values for a rq | |
870 | * @bucket: utilization clamp buckets affecting a rq | |
871 | * | |
872 | * Keep track of RUNNABLE tasks on a rq to aggregate their clamp values. | |
873 | * A clamp value is affecting a rq when there is at least one task RUNNABLE | |
874 | * (or actually running) with that value. | |
875 | * | |
876 | * There are up to UCLAMP_CNT possible different clamp values, currently there | |
877 | * are only two: minimum utilization and maximum utilization. | |
878 | * | |
879 | * All utilization clamping values are MAX aggregated, since: | |
880 | * - for util_min: we want to run the CPU at least at the max of the minimum | |
881 | * utilization required by its currently RUNNABLE tasks. | |
882 | * - for util_max: we want to allow the CPU to run up to the max of the | |
883 | * maximum utilization allowed by its currently RUNNABLE tasks. | |
884 | * | |
885 | * Since on each system we expect only a limited number of different | |
886 | * utilization clamp values (UCLAMP_BUCKETS), use a simple array to track | |
887 | * the metrics required to compute all the per-rq utilization clamp values. | |
888 | */ | |
889 | struct uclamp_rq { | |
890 | unsigned int value; | |
891 | struct uclamp_bucket bucket[UCLAMP_BUCKETS]; | |
892 | }; | |
46609ce2 QY |
893 | |
894 | DECLARE_STATIC_KEY_FALSE(sched_uclamp_used); | |
69842cba PB |
895 | #endif /* CONFIG_UCLAMP_TASK */ |
896 | ||
029632fb PZ |
897 | /* |
898 | * This is the main, per-CPU runqueue data structure. | |
899 | * | |
900 | * Locking rule: those places that want to lock multiple runqueues | |
901 | * (such as the load balancing or the thread migration code), lock | |
902 | * acquire operations must be ordered by ascending &runqueue. | |
903 | */ | |
904 | struct rq { | |
905 | /* runqueue lock: */ | |
97fb7a0a | 906 | raw_spinlock_t lock; |
029632fb PZ |
907 | |
908 | /* | |
909 | * nr_running and cpu_load should be in the same cacheline because | |
910 | * remote CPUs use both these fields when doing load calculation. | |
911 | */ | |
97fb7a0a | 912 | unsigned int nr_running; |
0ec8aa00 | 913 | #ifdef CONFIG_NUMA_BALANCING |
97fb7a0a IM |
914 | unsigned int nr_numa_running; |
915 | unsigned int nr_preferred_running; | |
a4739eca | 916 | unsigned int numa_migrate_on; |
0ec8aa00 | 917 | #endif |
3451d024 | 918 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 919 | #ifdef CONFIG_SMP |
e022e0d3 | 920 | unsigned long last_blocked_load_update_tick; |
f643ea22 | 921 | unsigned int has_blocked_load; |
90b5363a | 922 | call_single_data_t nohz_csd; |
9fd81dd5 | 923 | #endif /* CONFIG_SMP */ |
00357f5e | 924 | unsigned int nohz_tick_stopped; |
90b5363a | 925 | atomic_t nohz_flags; |
9fd81dd5 | 926 | #endif /* CONFIG_NO_HZ_COMMON */ |
dcdedb24 | 927 | |
126c2092 PZ |
928 | #ifdef CONFIG_SMP |
929 | unsigned int ttwu_pending; | |
930 | #endif | |
97fb7a0a | 931 | u64 nr_switches; |
029632fb | 932 | |
69842cba PB |
933 | #ifdef CONFIG_UCLAMP_TASK |
934 | /* Utilization clamp values based on CPU's RUNNABLE tasks */ | |
935 | struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned; | |
e496187d PB |
936 | unsigned int uclamp_flags; |
937 | #define UCLAMP_FLAG_IDLE 0x01 | |
69842cba PB |
938 | #endif |
939 | ||
97fb7a0a IM |
940 | struct cfs_rq cfs; |
941 | struct rt_rq rt; | |
942 | struct dl_rq dl; | |
029632fb PZ |
943 | |
944 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
97fb7a0a IM |
945 | /* list of leaf cfs_rq on this CPU: */ |
946 | struct list_head leaf_cfs_rq_list; | |
947 | struct list_head *tmp_alone_branch; | |
a35b6466 PZ |
948 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
949 | ||
029632fb PZ |
950 | /* |
951 | * This is part of a global counter where only the total sum | |
952 | * over all CPUs matters. A task can increase this counter on | |
953 | * one CPU and if it got migrated afterwards it may decrease | |
954 | * it on another CPU. Always updated under the runqueue lock: | |
955 | */ | |
97fb7a0a | 956 | unsigned long nr_uninterruptible; |
029632fb | 957 | |
4104a562 | 958 | struct task_struct __rcu *curr; |
97fb7a0a IM |
959 | struct task_struct *idle; |
960 | struct task_struct *stop; | |
961 | unsigned long next_balance; | |
962 | struct mm_struct *prev_mm; | |
029632fb | 963 | |
97fb7a0a IM |
964 | unsigned int clock_update_flags; |
965 | u64 clock; | |
23127296 VG |
966 | /* Ensure that all clocks are in the same cache line */ |
967 | u64 clock_task ____cacheline_aligned; | |
968 | u64 clock_pelt; | |
969 | unsigned long lost_idle_time; | |
029632fb | 970 | |
97fb7a0a | 971 | atomic_t nr_iowait; |
029632fb | 972 | |
227a4aad MD |
973 | #ifdef CONFIG_MEMBARRIER |
974 | int membarrier_state; | |
975 | #endif | |
976 | ||
029632fb | 977 | #ifdef CONFIG_SMP |
994aeb7a JFG |
978 | struct root_domain *rd; |
979 | struct sched_domain __rcu *sd; | |
97fb7a0a IM |
980 | |
981 | unsigned long cpu_capacity; | |
982 | unsigned long cpu_capacity_orig; | |
029632fb | 983 | |
97fb7a0a | 984 | struct callback_head *balance_callback; |
975707f2 | 985 | unsigned char balance_push; |
029632fb | 986 | |
19a1f5ec | 987 | unsigned char nohz_idle_balance; |
97fb7a0a | 988 | unsigned char idle_balance; |
e3fca9e7 | 989 | |
3b1baa64 MR |
990 | unsigned long misfit_task_load; |
991 | ||
029632fb | 992 | /* For active balancing */ |
97fb7a0a IM |
993 | int active_balance; |
994 | int push_cpu; | |
995 | struct cpu_stop_work active_balance_work; | |
996 | ||
997 | /* CPU of this runqueue: */ | |
998 | int cpu; | |
999 | int online; | |
029632fb | 1000 | |
367456c7 PZ |
1001 | struct list_head cfs_tasks; |
1002 | ||
371bf427 | 1003 | struct sched_avg avg_rt; |
3727e0e1 | 1004 | struct sched_avg avg_dl; |
11d4afd4 | 1005 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
91c27493 | 1006 | struct sched_avg avg_irq; |
76504793 TG |
1007 | #endif |
1008 | #ifdef CONFIG_SCHED_THERMAL_PRESSURE | |
1009 | struct sched_avg avg_thermal; | |
91c27493 | 1010 | #endif |
97fb7a0a IM |
1011 | u64 idle_stamp; |
1012 | u64 avg_idle; | |
9bd721c5 JL |
1013 | |
1014 | /* This is used to determine avg_idle's max value */ | |
97fb7a0a | 1015 | u64 max_idle_balance_cost; |
f2469a1f TG |
1016 | |
1017 | #ifdef CONFIG_HOTPLUG_CPU | |
1018 | struct rcuwait hotplug_wait; | |
1019 | #endif | |
90b5363a | 1020 | #endif /* CONFIG_SMP */ |
029632fb PZ |
1021 | |
1022 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
97fb7a0a | 1023 | u64 prev_irq_time; |
029632fb PZ |
1024 | #endif |
1025 | #ifdef CONFIG_PARAVIRT | |
97fb7a0a | 1026 | u64 prev_steal_time; |
029632fb PZ |
1027 | #endif |
1028 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
97fb7a0a | 1029 | u64 prev_steal_time_rq; |
029632fb PZ |
1030 | #endif |
1031 | ||
1032 | /* calc_load related fields */ | |
97fb7a0a IM |
1033 | unsigned long calc_load_update; |
1034 | long calc_load_active; | |
029632fb PZ |
1035 | |
1036 | #ifdef CONFIG_SCHED_HRTICK | |
1037 | #ifdef CONFIG_SMP | |
97fb7a0a | 1038 | call_single_data_t hrtick_csd; |
029632fb | 1039 | #endif |
97fb7a0a | 1040 | struct hrtimer hrtick_timer; |
c30ac5e3 | 1041 | ktime_t hrtick_time; |
029632fb PZ |
1042 | #endif |
1043 | ||
1044 | #ifdef CONFIG_SCHEDSTATS | |
1045 | /* latency stats */ | |
97fb7a0a IM |
1046 | struct sched_info rq_sched_info; |
1047 | unsigned long long rq_cpu_time; | |
029632fb PZ |
1048 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ |
1049 | ||
1050 | /* sys_sched_yield() stats */ | |
97fb7a0a | 1051 | unsigned int yld_count; |
029632fb PZ |
1052 | |
1053 | /* schedule() stats */ | |
97fb7a0a IM |
1054 | unsigned int sched_count; |
1055 | unsigned int sched_goidle; | |
029632fb PZ |
1056 | |
1057 | /* try_to_wake_up() stats */ | |
97fb7a0a IM |
1058 | unsigned int ttwu_count; |
1059 | unsigned int ttwu_local; | |
029632fb PZ |
1060 | #endif |
1061 | ||
442bf3aa DL |
1062 | #ifdef CONFIG_CPU_IDLE |
1063 | /* Must be inspected within a rcu lock section */ | |
97fb7a0a | 1064 | struct cpuidle_state *idle_state; |
442bf3aa | 1065 | #endif |
3015ef4b | 1066 | |
74d862b6 | 1067 | #ifdef CONFIG_SMP |
3015ef4b TG |
1068 | unsigned int nr_pinned; |
1069 | #endif | |
a7c81556 PZ |
1070 | unsigned int push_busy; |
1071 | struct cpu_stop_work push_work; | |
029632fb PZ |
1072 | }; |
1073 | ||
62478d99 VG |
1074 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1075 | ||
1076 | /* CPU runqueue to which this cfs_rq is attached */ | |
1077 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
1078 | { | |
1079 | return cfs_rq->rq; | |
1080 | } | |
1081 | ||
1082 | #else | |
1083 | ||
1084 | static inline struct rq *rq_of(struct cfs_rq *cfs_rq) | |
1085 | { | |
1086 | return container_of(cfs_rq, struct rq, cfs); | |
1087 | } | |
1088 | #endif | |
1089 | ||
029632fb PZ |
1090 | static inline int cpu_of(struct rq *rq) |
1091 | { | |
1092 | #ifdef CONFIG_SMP | |
1093 | return rq->cpu; | |
1094 | #else | |
1095 | return 0; | |
1096 | #endif | |
1097 | } | |
1098 | ||
a7c81556 PZ |
1099 | #define MDF_PUSH 0x01 |
1100 | ||
1101 | static inline bool is_migration_disabled(struct task_struct *p) | |
1102 | { | |
74d862b6 | 1103 | #ifdef CONFIG_SMP |
a7c81556 PZ |
1104 | return p->migration_disabled; |
1105 | #else | |
1106 | return false; | |
1107 | #endif | |
1108 | } | |
1b568f0a PZ |
1109 | |
1110 | #ifdef CONFIG_SCHED_SMT | |
1b568f0a PZ |
1111 | extern void __update_idle_core(struct rq *rq); |
1112 | ||
1113 | static inline void update_idle_core(struct rq *rq) | |
1114 | { | |
1115 | if (static_branch_unlikely(&sched_smt_present)) | |
1116 | __update_idle_core(rq); | |
1117 | } | |
1118 | ||
1119 | #else | |
1120 | static inline void update_idle_core(struct rq *rq) { } | |
1121 | #endif | |
1122 | ||
8b06c55b | 1123 | DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
029632fb | 1124 | |
518cd623 | 1125 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
4a32fea9 | 1126 | #define this_rq() this_cpu_ptr(&runqueues) |
518cd623 PZ |
1127 | #define task_rq(p) cpu_rq(task_cpu(p)) |
1128 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
4a32fea9 | 1129 | #define raw_rq() raw_cpu_ptr(&runqueues) |
518cd623 | 1130 | |
1f351d7f JW |
1131 | extern void update_rq_clock(struct rq *rq); |
1132 | ||
cebde6d6 PZ |
1133 | static inline u64 __rq_clock_broken(struct rq *rq) |
1134 | { | |
316c1608 | 1135 | return READ_ONCE(rq->clock); |
cebde6d6 PZ |
1136 | } |
1137 | ||
cb42c9a3 MF |
1138 | /* |
1139 | * rq::clock_update_flags bits | |
1140 | * | |
1141 | * %RQCF_REQ_SKIP - will request skipping of clock update on the next | |
1142 | * call to __schedule(). This is an optimisation to avoid | |
1143 | * neighbouring rq clock updates. | |
1144 | * | |
1145 | * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is | |
1146 | * in effect and calls to update_rq_clock() are being ignored. | |
1147 | * | |
1148 | * %RQCF_UPDATED - is a debug flag that indicates whether a call has been | |
1149 | * made to update_rq_clock() since the last time rq::lock was pinned. | |
1150 | * | |
1151 | * If inside of __schedule(), clock_update_flags will have been | |
1152 | * shifted left (a left shift is a cheap operation for the fast path | |
1153 | * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, | |
1154 | * | |
1155 | * if (rq-clock_update_flags >= RQCF_UPDATED) | |
1156 | * | |
1157 | * to check if %RQCF_UPADTED is set. It'll never be shifted more than | |
1158 | * one position though, because the next rq_unpin_lock() will shift it | |
1159 | * back. | |
1160 | */ | |
97fb7a0a IM |
1161 | #define RQCF_REQ_SKIP 0x01 |
1162 | #define RQCF_ACT_SKIP 0x02 | |
1163 | #define RQCF_UPDATED 0x04 | |
cb42c9a3 MF |
1164 | |
1165 | static inline void assert_clock_updated(struct rq *rq) | |
1166 | { | |
1167 | /* | |
1168 | * The only reason for not seeing a clock update since the | |
1169 | * last rq_pin_lock() is if we're currently skipping updates. | |
1170 | */ | |
1171 | SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); | |
1172 | } | |
1173 | ||
78becc27 FW |
1174 | static inline u64 rq_clock(struct rq *rq) |
1175 | { | |
cebde6d6 | 1176 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
1177 | assert_clock_updated(rq); |
1178 | ||
78becc27 FW |
1179 | return rq->clock; |
1180 | } | |
1181 | ||
1182 | static inline u64 rq_clock_task(struct rq *rq) | |
1183 | { | |
cebde6d6 | 1184 | lockdep_assert_held(&rq->lock); |
cb42c9a3 MF |
1185 | assert_clock_updated(rq); |
1186 | ||
78becc27 FW |
1187 | return rq->clock_task; |
1188 | } | |
1189 | ||
05289b90 TG |
1190 | /** |
1191 | * By default the decay is the default pelt decay period. | |
1192 | * The decay shift can change the decay period in | |
1193 | * multiples of 32. | |
1194 | * Decay shift Decay period(ms) | |
1195 | * 0 32 | |
1196 | * 1 64 | |
1197 | * 2 128 | |
1198 | * 3 256 | |
1199 | * 4 512 | |
1200 | */ | |
1201 | extern int sched_thermal_decay_shift; | |
1202 | ||
1203 | static inline u64 rq_clock_thermal(struct rq *rq) | |
1204 | { | |
1205 | return rq_clock_task(rq) >> sched_thermal_decay_shift; | |
1206 | } | |
1207 | ||
adcc8da8 | 1208 | static inline void rq_clock_skip_update(struct rq *rq) |
9edfbfed PZ |
1209 | { |
1210 | lockdep_assert_held(&rq->lock); | |
adcc8da8 DB |
1211 | rq->clock_update_flags |= RQCF_REQ_SKIP; |
1212 | } | |
1213 | ||
1214 | /* | |
595058b6 | 1215 | * See rt task throttling, which is the only time a skip |
adcc8da8 DB |
1216 | * request is cancelled. |
1217 | */ | |
1218 | static inline void rq_clock_cancel_skipupdate(struct rq *rq) | |
1219 | { | |
1220 | lockdep_assert_held(&rq->lock); | |
1221 | rq->clock_update_flags &= ~RQCF_REQ_SKIP; | |
9edfbfed PZ |
1222 | } |
1223 | ||
d8ac8971 MF |
1224 | struct rq_flags { |
1225 | unsigned long flags; | |
1226 | struct pin_cookie cookie; | |
cb42c9a3 MF |
1227 | #ifdef CONFIG_SCHED_DEBUG |
1228 | /* | |
1229 | * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the | |
1230 | * current pin context is stashed here in case it needs to be | |
1231 | * restored in rq_repin_lock(). | |
1232 | */ | |
1233 | unsigned int clock_update_flags; | |
1234 | #endif | |
d8ac8971 MF |
1235 | }; |
1236 | ||
ae792702 PZ |
1237 | extern struct callback_head balance_push_callback; |
1238 | ||
58877d34 PZ |
1239 | /* |
1240 | * Lockdep annotation that avoids accidental unlocks; it's like a | |
1241 | * sticky/continuous lockdep_assert_held(). | |
1242 | * | |
1243 | * This avoids code that has access to 'struct rq *rq' (basically everything in | |
1244 | * the scheduler) from accidentally unlocking the rq if they do not also have a | |
1245 | * copy of the (on-stack) 'struct rq_flags rf'. | |
1246 | * | |
1247 | * Also see Documentation/locking/lockdep-design.rst. | |
1248 | */ | |
d8ac8971 MF |
1249 | static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) |
1250 | { | |
1251 | rf->cookie = lockdep_pin_lock(&rq->lock); | |
cb42c9a3 MF |
1252 | |
1253 | #ifdef CONFIG_SCHED_DEBUG | |
1254 | rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); | |
1255 | rf->clock_update_flags = 0; | |
565790d2 | 1256 | #ifdef CONFIG_SMP |
ae792702 PZ |
1257 | SCHED_WARN_ON(rq->balance_callback && rq->balance_callback != &balance_push_callback); |
1258 | #endif | |
565790d2 | 1259 | #endif |
d8ac8971 MF |
1260 | } |
1261 | ||
1262 | static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) | |
1263 | { | |
cb42c9a3 MF |
1264 | #ifdef CONFIG_SCHED_DEBUG |
1265 | if (rq->clock_update_flags > RQCF_ACT_SKIP) | |
1266 | rf->clock_update_flags = RQCF_UPDATED; | |
1267 | #endif | |
1268 | ||
d8ac8971 MF |
1269 | lockdep_unpin_lock(&rq->lock, rf->cookie); |
1270 | } | |
1271 | ||
1272 | static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) | |
1273 | { | |
1274 | lockdep_repin_lock(&rq->lock, rf->cookie); | |
cb42c9a3 MF |
1275 | |
1276 | #ifdef CONFIG_SCHED_DEBUG | |
1277 | /* | |
1278 | * Restore the value we stashed in @rf for this pin context. | |
1279 | */ | |
1280 | rq->clock_update_flags |= rf->clock_update_flags; | |
1281 | #endif | |
d8ac8971 MF |
1282 | } |
1283 | ||
1f351d7f JW |
1284 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
1285 | __acquires(rq->lock); | |
1286 | ||
1287 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) | |
1288 | __acquires(p->pi_lock) | |
1289 | __acquires(rq->lock); | |
1290 | ||
1291 | static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1292 | __releases(rq->lock) | |
1293 | { | |
1294 | rq_unpin_lock(rq, rf); | |
1295 | raw_spin_unlock(&rq->lock); | |
1296 | } | |
1297 | ||
1298 | static inline void | |
1299 | task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) | |
1300 | __releases(rq->lock) | |
1301 | __releases(p->pi_lock) | |
1302 | { | |
1303 | rq_unpin_lock(rq, rf); | |
1304 | raw_spin_unlock(&rq->lock); | |
1305 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); | |
1306 | } | |
1307 | ||
1308 | static inline void | |
1309 | rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) | |
1310 | __acquires(rq->lock) | |
1311 | { | |
1312 | raw_spin_lock_irqsave(&rq->lock, rf->flags); | |
1313 | rq_pin_lock(rq, rf); | |
1314 | } | |
1315 | ||
1316 | static inline void | |
1317 | rq_lock_irq(struct rq *rq, struct rq_flags *rf) | |
1318 | __acquires(rq->lock) | |
1319 | { | |
1320 | raw_spin_lock_irq(&rq->lock); | |
1321 | rq_pin_lock(rq, rf); | |
1322 | } | |
1323 | ||
1324 | static inline void | |
1325 | rq_lock(struct rq *rq, struct rq_flags *rf) | |
1326 | __acquires(rq->lock) | |
1327 | { | |
1328 | raw_spin_lock(&rq->lock); | |
1329 | rq_pin_lock(rq, rf); | |
1330 | } | |
1331 | ||
1332 | static inline void | |
1333 | rq_relock(struct rq *rq, struct rq_flags *rf) | |
1334 | __acquires(rq->lock) | |
1335 | { | |
1336 | raw_spin_lock(&rq->lock); | |
1337 | rq_repin_lock(rq, rf); | |
1338 | } | |
1339 | ||
1340 | static inline void | |
1341 | rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) | |
1342 | __releases(rq->lock) | |
1343 | { | |
1344 | rq_unpin_lock(rq, rf); | |
1345 | raw_spin_unlock_irqrestore(&rq->lock, rf->flags); | |
1346 | } | |
1347 | ||
1348 | static inline void | |
1349 | rq_unlock_irq(struct rq *rq, struct rq_flags *rf) | |
1350 | __releases(rq->lock) | |
1351 | { | |
1352 | rq_unpin_lock(rq, rf); | |
1353 | raw_spin_unlock_irq(&rq->lock); | |
1354 | } | |
1355 | ||
1356 | static inline void | |
1357 | rq_unlock(struct rq *rq, struct rq_flags *rf) | |
1358 | __releases(rq->lock) | |
1359 | { | |
1360 | rq_unpin_lock(rq, rf); | |
1361 | raw_spin_unlock(&rq->lock); | |
1362 | } | |
1363 | ||
246b3b33 JW |
1364 | static inline struct rq * |
1365 | this_rq_lock_irq(struct rq_flags *rf) | |
1366 | __acquires(rq->lock) | |
1367 | { | |
1368 | struct rq *rq; | |
1369 | ||
1370 | local_irq_disable(); | |
1371 | rq = this_rq(); | |
1372 | rq_lock(rq, rf); | |
1373 | return rq; | |
1374 | } | |
1375 | ||
9942f79b | 1376 | #ifdef CONFIG_NUMA |
e3fe70b1 RR |
1377 | enum numa_topology_type { |
1378 | NUMA_DIRECT, | |
1379 | NUMA_GLUELESS_MESH, | |
1380 | NUMA_BACKPLANE, | |
1381 | }; | |
1382 | extern enum numa_topology_type sched_numa_topology_type; | |
9942f79b RR |
1383 | extern int sched_max_numa_distance; |
1384 | extern bool find_numa_distance(int distance); | |
f2cb1360 IM |
1385 | extern void sched_init_numa(void); |
1386 | extern void sched_domains_numa_masks_set(unsigned int cpu); | |
1387 | extern void sched_domains_numa_masks_clear(unsigned int cpu); | |
e0e8d491 | 1388 | extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu); |
f2cb1360 IM |
1389 | #else |
1390 | static inline void sched_init_numa(void) { } | |
1391 | static inline void sched_domains_numa_masks_set(unsigned int cpu) { } | |
1392 | static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } | |
e0e8d491 WL |
1393 | static inline int sched_numa_find_closest(const struct cpumask *cpus, int cpu) |
1394 | { | |
1395 | return nr_cpu_ids; | |
1396 | } | |
f2cb1360 IM |
1397 | #endif |
1398 | ||
f809ca9a | 1399 | #ifdef CONFIG_NUMA_BALANCING |
44dba3d5 IM |
1400 | /* The regions in numa_faults array from task_struct */ |
1401 | enum numa_faults_stats { | |
1402 | NUMA_MEM = 0, | |
1403 | NUMA_CPU, | |
1404 | NUMA_MEMBUF, | |
1405 | NUMA_CPUBUF | |
1406 | }; | |
0ec8aa00 | 1407 | extern void sched_setnuma(struct task_struct *p, int node); |
e6628d5b | 1408 | extern int migrate_task_to(struct task_struct *p, int cpu); |
0ad4e3df SD |
1409 | extern int migrate_swap(struct task_struct *p, struct task_struct *t, |
1410 | int cpu, int scpu); | |
13784475 MG |
1411 | extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); |
1412 | #else | |
1413 | static inline void | |
1414 | init_numa_balancing(unsigned long clone_flags, struct task_struct *p) | |
1415 | { | |
1416 | } | |
f809ca9a MG |
1417 | #endif /* CONFIG_NUMA_BALANCING */ |
1418 | ||
518cd623 PZ |
1419 | #ifdef CONFIG_SMP |
1420 | ||
e3fca9e7 PZ |
1421 | static inline void |
1422 | queue_balance_callback(struct rq *rq, | |
1423 | struct callback_head *head, | |
1424 | void (*func)(struct rq *rq)) | |
1425 | { | |
1426 | lockdep_assert_held(&rq->lock); | |
1427 | ||
ae792702 | 1428 | if (unlikely(head->next || rq->balance_callback == &balance_push_callback)) |
e3fca9e7 PZ |
1429 | return; |
1430 | ||
1431 | head->func = (void (*)(struct callback_head *))func; | |
1432 | head->next = rq->balance_callback; | |
1433 | rq->balance_callback = head; | |
1434 | } | |
1435 | ||
029632fb PZ |
1436 | #define rcu_dereference_check_sched_domain(p) \ |
1437 | rcu_dereference_check((p), \ | |
1438 | lockdep_is_held(&sched_domains_mutex)) | |
1439 | ||
1440 | /* | |
1441 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
337e9b07 | 1442 | * See destroy_sched_domains: call_rcu for details. |
029632fb PZ |
1443 | * |
1444 | * The domain tree of any CPU may only be accessed from within | |
1445 | * preempt-disabled sections. | |
1446 | */ | |
1447 | #define for_each_domain(cpu, __sd) \ | |
518cd623 PZ |
1448 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ |
1449 | __sd; __sd = __sd->parent) | |
029632fb | 1450 | |
518cd623 PZ |
1451 | /** |
1452 | * highest_flag_domain - Return highest sched_domain containing flag. | |
97fb7a0a | 1453 | * @cpu: The CPU whose highest level of sched domain is to |
518cd623 PZ |
1454 | * be returned. |
1455 | * @flag: The flag to check for the highest sched_domain | |
97fb7a0a | 1456 | * for the given CPU. |
518cd623 | 1457 | * |
97fb7a0a | 1458 | * Returns the highest sched_domain of a CPU which contains the given flag. |
518cd623 PZ |
1459 | */ |
1460 | static inline struct sched_domain *highest_flag_domain(int cpu, int flag) | |
1461 | { | |
1462 | struct sched_domain *sd, *hsd = NULL; | |
1463 | ||
1464 | for_each_domain(cpu, sd) { | |
1465 | if (!(sd->flags & flag)) | |
1466 | break; | |
1467 | hsd = sd; | |
1468 | } | |
1469 | ||
1470 | return hsd; | |
1471 | } | |
1472 | ||
fb13c7ee MG |
1473 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) |
1474 | { | |
1475 | struct sched_domain *sd; | |
1476 | ||
1477 | for_each_domain(cpu, sd) { | |
1478 | if (sd->flags & flag) | |
1479 | break; | |
1480 | } | |
1481 | ||
1482 | return sd; | |
1483 | } | |
1484 | ||
994aeb7a | 1485 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_llc); |
7d9ffa89 | 1486 | DECLARE_PER_CPU(int, sd_llc_size); |
518cd623 | 1487 | DECLARE_PER_CPU(int, sd_llc_id); |
994aeb7a JFG |
1488 | DECLARE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); |
1489 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_numa); | |
1490 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); | |
1491 | DECLARE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); | |
df054e84 | 1492 | extern struct static_key_false sched_asym_cpucapacity; |
518cd623 | 1493 | |
63b2ca30 | 1494 | struct sched_group_capacity { |
97fb7a0a | 1495 | atomic_t ref; |
5e6521ea | 1496 | /* |
172895e6 | 1497 | * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity |
63b2ca30 | 1498 | * for a single CPU. |
5e6521ea | 1499 | */ |
97fb7a0a IM |
1500 | unsigned long capacity; |
1501 | unsigned long min_capacity; /* Min per-CPU capacity in group */ | |
e3d6d0cb | 1502 | unsigned long max_capacity; /* Max per-CPU capacity in group */ |
97fb7a0a IM |
1503 | unsigned long next_update; |
1504 | int imbalance; /* XXX unrelated to capacity but shared group state */ | |
5e6521ea | 1505 | |
005f874d | 1506 | #ifdef CONFIG_SCHED_DEBUG |
97fb7a0a | 1507 | int id; |
005f874d PZ |
1508 | #endif |
1509 | ||
eba9f082 | 1510 | unsigned long cpumask[]; /* Balance mask */ |
5e6521ea LZ |
1511 | }; |
1512 | ||
1513 | struct sched_group { | |
97fb7a0a IM |
1514 | struct sched_group *next; /* Must be a circular list */ |
1515 | atomic_t ref; | |
5e6521ea | 1516 | |
97fb7a0a | 1517 | unsigned int group_weight; |
63b2ca30 | 1518 | struct sched_group_capacity *sgc; |
97fb7a0a | 1519 | int asym_prefer_cpu; /* CPU of highest priority in group */ |
5e6521ea LZ |
1520 | |
1521 | /* | |
1522 | * The CPUs this group covers. | |
1523 | * | |
1524 | * NOTE: this field is variable length. (Allocated dynamically | |
1525 | * by attaching extra space to the end of the structure, | |
1526 | * depending on how many CPUs the kernel has booted up with) | |
1527 | */ | |
04f5c362 | 1528 | unsigned long cpumask[]; |
5e6521ea LZ |
1529 | }; |
1530 | ||
ae4df9d6 | 1531 | static inline struct cpumask *sched_group_span(struct sched_group *sg) |
5e6521ea LZ |
1532 | { |
1533 | return to_cpumask(sg->cpumask); | |
1534 | } | |
1535 | ||
1536 | /* | |
e5c14b1f | 1537 | * See build_balance_mask(). |
5e6521ea | 1538 | */ |
e5c14b1f | 1539 | static inline struct cpumask *group_balance_mask(struct sched_group *sg) |
5e6521ea | 1540 | { |
63b2ca30 | 1541 | return to_cpumask(sg->sgc->cpumask); |
5e6521ea LZ |
1542 | } |
1543 | ||
1544 | /** | |
97fb7a0a IM |
1545 | * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. |
1546 | * @group: The group whose first CPU is to be returned. | |
5e6521ea LZ |
1547 | */ |
1548 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
1549 | { | |
ae4df9d6 | 1550 | return cpumask_first(sched_group_span(group)); |
5e6521ea LZ |
1551 | } |
1552 | ||
c1174876 PZ |
1553 | extern int group_balance_cpu(struct sched_group *sg); |
1554 | ||
3866e845 SRRH |
1555 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
1556 | void register_sched_domain_sysctl(void); | |
bbdacdfe | 1557 | void dirty_sched_domain_sysctl(int cpu); |
3866e845 SRRH |
1558 | void unregister_sched_domain_sysctl(void); |
1559 | #else | |
1560 | static inline void register_sched_domain_sysctl(void) | |
1561 | { | |
1562 | } | |
bbdacdfe PZ |
1563 | static inline void dirty_sched_domain_sysctl(int cpu) |
1564 | { | |
1565 | } | |
3866e845 SRRH |
1566 | static inline void unregister_sched_domain_sysctl(void) |
1567 | { | |
1568 | } | |
1569 | #endif | |
1570 | ||
b2a02fc4 | 1571 | extern void flush_smp_call_function_from_idle(void); |
e3baac47 | 1572 | |
b2a02fc4 PZ |
1573 | #else /* !CONFIG_SMP: */ |
1574 | static inline void flush_smp_call_function_from_idle(void) { } | |
b2a02fc4 | 1575 | #endif |
029632fb | 1576 | |
391e43da | 1577 | #include "stats.h" |
1051408f | 1578 | #include "autogroup.h" |
029632fb PZ |
1579 | |
1580 | #ifdef CONFIG_CGROUP_SCHED | |
1581 | ||
1582 | /* | |
1583 | * Return the group to which this tasks belongs. | |
1584 | * | |
8af01f56 TH |
1585 | * We cannot use task_css() and friends because the cgroup subsystem |
1586 | * changes that value before the cgroup_subsys::attach() method is called, | |
1587 | * therefore we cannot pin it and might observe the wrong value. | |
8323f26c PZ |
1588 | * |
1589 | * The same is true for autogroup's p->signal->autogroup->tg, the autogroup | |
1590 | * core changes this before calling sched_move_task(). | |
1591 | * | |
1592 | * Instead we use a 'copy' which is updated from sched_move_task() while | |
1593 | * holding both task_struct::pi_lock and rq::lock. | |
029632fb PZ |
1594 | */ |
1595 | static inline struct task_group *task_group(struct task_struct *p) | |
1596 | { | |
8323f26c | 1597 | return p->sched_task_group; |
029632fb PZ |
1598 | } |
1599 | ||
1600 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
1601 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
1602 | { | |
1603 | #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) | |
1604 | struct task_group *tg = task_group(p); | |
1605 | #endif | |
1606 | ||
1607 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
ad936d86 | 1608 | set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); |
029632fb PZ |
1609 | p->se.cfs_rq = tg->cfs_rq[cpu]; |
1610 | p->se.parent = tg->se[cpu]; | |
1611 | #endif | |
1612 | ||
1613 | #ifdef CONFIG_RT_GROUP_SCHED | |
1614 | p->rt.rt_rq = tg->rt_rq[cpu]; | |
1615 | p->rt.parent = tg->rt_se[cpu]; | |
1616 | #endif | |
1617 | } | |
1618 | ||
1619 | #else /* CONFIG_CGROUP_SCHED */ | |
1620 | ||
1621 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
1622 | static inline struct task_group *task_group(struct task_struct *p) | |
1623 | { | |
1624 | return NULL; | |
1625 | } | |
1626 | ||
1627 | #endif /* CONFIG_CGROUP_SCHED */ | |
1628 | ||
1629 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1630 | { | |
1631 | set_task_rq(p, cpu); | |
1632 | #ifdef CONFIG_SMP | |
1633 | /* | |
1634 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
dfcb245e | 1635 | * successfully executed on another CPU. We must ensure that updates of |
029632fb PZ |
1636 | * per-task data have been completed by this moment. |
1637 | */ | |
1638 | smp_wmb(); | |
c65eacbe | 1639 | #ifdef CONFIG_THREAD_INFO_IN_TASK |
c546951d | 1640 | WRITE_ONCE(p->cpu, cpu); |
c65eacbe | 1641 | #else |
c546951d | 1642 | WRITE_ONCE(task_thread_info(p)->cpu, cpu); |
c65eacbe | 1643 | #endif |
ac66f547 | 1644 | p->wake_cpu = cpu; |
029632fb PZ |
1645 | #endif |
1646 | } | |
1647 | ||
1648 | /* | |
1649 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
1650 | */ | |
1651 | #ifdef CONFIG_SCHED_DEBUG | |
c5905afb | 1652 | # include <linux/static_key.h> |
029632fb PZ |
1653 | # define const_debug __read_mostly |
1654 | #else | |
1655 | # define const_debug const | |
1656 | #endif | |
1657 | ||
029632fb PZ |
1658 | #define SCHED_FEAT(name, enabled) \ |
1659 | __SCHED_FEAT_##name , | |
1660 | ||
1661 | enum { | |
391e43da | 1662 | #include "features.h" |
f8b6d1cc | 1663 | __SCHED_FEAT_NR, |
029632fb PZ |
1664 | }; |
1665 | ||
1666 | #undef SCHED_FEAT | |
1667 | ||
a73f863a | 1668 | #ifdef CONFIG_SCHED_DEBUG |
765cc3a4 PB |
1669 | |
1670 | /* | |
1671 | * To support run-time toggling of sched features, all the translation units | |
1672 | * (but core.c) reference the sysctl_sched_features defined in core.c. | |
1673 | */ | |
1674 | extern const_debug unsigned int sysctl_sched_features; | |
1675 | ||
a73f863a | 1676 | #ifdef CONFIG_JUMP_LABEL |
f8b6d1cc | 1677 | #define SCHED_FEAT(name, enabled) \ |
c5905afb | 1678 | static __always_inline bool static_branch_##name(struct static_key *key) \ |
f8b6d1cc | 1679 | { \ |
6e76ea8a | 1680 | return static_key_##enabled(key); \ |
f8b6d1cc PZ |
1681 | } |
1682 | ||
1683 | #include "features.h" | |
f8b6d1cc PZ |
1684 | #undef SCHED_FEAT |
1685 | ||
c5905afb | 1686 | extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; |
f8b6d1cc | 1687 | #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) |
765cc3a4 | 1688 | |
a73f863a JL |
1689 | #else /* !CONFIG_JUMP_LABEL */ |
1690 | ||
1691 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
1692 | ||
1693 | #endif /* CONFIG_JUMP_LABEL */ | |
1694 | ||
1695 | #else /* !SCHED_DEBUG */ | |
765cc3a4 PB |
1696 | |
1697 | /* | |
1698 | * Each translation unit has its own copy of sysctl_sched_features to allow | |
1699 | * constants propagation at compile time and compiler optimization based on | |
1700 | * features default. | |
1701 | */ | |
1702 | #define SCHED_FEAT(name, enabled) \ | |
1703 | (1UL << __SCHED_FEAT_##name) * enabled | | |
1704 | static const_debug __maybe_unused unsigned int sysctl_sched_features = | |
1705 | #include "features.h" | |
1706 | 0; | |
1707 | #undef SCHED_FEAT | |
1708 | ||
7e6f4c5d | 1709 | #define sched_feat(x) !!(sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) |
765cc3a4 | 1710 | |
a73f863a | 1711 | #endif /* SCHED_DEBUG */ |
029632fb | 1712 | |
2a595721 | 1713 | extern struct static_key_false sched_numa_balancing; |
cb251765 | 1714 | extern struct static_key_false sched_schedstats; |
cbee9f88 | 1715 | |
029632fb PZ |
1716 | static inline u64 global_rt_period(void) |
1717 | { | |
1718 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
1719 | } | |
1720 | ||
1721 | static inline u64 global_rt_runtime(void) | |
1722 | { | |
1723 | if (sysctl_sched_rt_runtime < 0) | |
1724 | return RUNTIME_INF; | |
1725 | ||
1726 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
1727 | } | |
1728 | ||
029632fb PZ |
1729 | static inline int task_current(struct rq *rq, struct task_struct *p) |
1730 | { | |
1731 | return rq->curr == p; | |
1732 | } | |
1733 | ||
1734 | static inline int task_running(struct rq *rq, struct task_struct *p) | |
1735 | { | |
1736 | #ifdef CONFIG_SMP | |
1737 | return p->on_cpu; | |
1738 | #else | |
1739 | return task_current(rq, p); | |
1740 | #endif | |
1741 | } | |
1742 | ||
da0c1e65 KT |
1743 | static inline int task_on_rq_queued(struct task_struct *p) |
1744 | { | |
1745 | return p->on_rq == TASK_ON_RQ_QUEUED; | |
1746 | } | |
029632fb | 1747 | |
cca26e80 KT |
1748 | static inline int task_on_rq_migrating(struct task_struct *p) |
1749 | { | |
c546951d | 1750 | return READ_ONCE(p->on_rq) == TASK_ON_RQ_MIGRATING; |
cca26e80 KT |
1751 | } |
1752 | ||
17770579 VS |
1753 | /* Wake flags. The first three directly map to some SD flag value */ |
1754 | #define WF_EXEC 0x02 /* Wakeup after exec; maps to SD_BALANCE_EXEC */ | |
1755 | #define WF_FORK 0x04 /* Wakeup after fork; maps to SD_BALANCE_FORK */ | |
1756 | #define WF_TTWU 0x08 /* Wakeup; maps to SD_BALANCE_WAKE */ | |
1757 | ||
1758 | #define WF_SYNC 0x10 /* Waker goes to sleep after wakeup */ | |
1759 | #define WF_MIGRATED 0x20 /* Internal use, task got migrated */ | |
1760 | #define WF_ON_CPU 0x40 /* Wakee is on_cpu */ | |
1761 | ||
1762 | #ifdef CONFIG_SMP | |
1763 | static_assert(WF_EXEC == SD_BALANCE_EXEC); | |
1764 | static_assert(WF_FORK == SD_BALANCE_FORK); | |
1765 | static_assert(WF_TTWU == SD_BALANCE_WAKE); | |
1766 | #endif | |
b13095f0 | 1767 | |
029632fb PZ |
1768 | /* |
1769 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1770 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1771 | * each task makes to its run queue's load is weighted according to its | |
1772 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
1773 | * scaled version of the new time slice allocation that they receive on time | |
1774 | * slice expiry etc. | |
1775 | */ | |
1776 | ||
97fb7a0a IM |
1777 | #define WEIGHT_IDLEPRIO 3 |
1778 | #define WMULT_IDLEPRIO 1431655765 | |
029632fb | 1779 | |
97fb7a0a IM |
1780 | extern const int sched_prio_to_weight[40]; |
1781 | extern const u32 sched_prio_to_wmult[40]; | |
029632fb | 1782 | |
ff77e468 PZ |
1783 | /* |
1784 | * {de,en}queue flags: | |
1785 | * | |
1786 | * DEQUEUE_SLEEP - task is no longer runnable | |
1787 | * ENQUEUE_WAKEUP - task just became runnable | |
1788 | * | |
1789 | * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks | |
1790 | * are in a known state which allows modification. Such pairs | |
1791 | * should preserve as much state as possible. | |
1792 | * | |
1793 | * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location | |
1794 | * in the runqueue. | |
1795 | * | |
1796 | * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) | |
1797 | * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) | |
59efa0ba | 1798 | * ENQUEUE_MIGRATED - the task was migrated during wakeup |
ff77e468 PZ |
1799 | * |
1800 | */ | |
1801 | ||
1802 | #define DEQUEUE_SLEEP 0x01 | |
97fb7a0a IM |
1803 | #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ |
1804 | #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ | |
1805 | #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ | |
ff77e468 | 1806 | |
1de64443 | 1807 | #define ENQUEUE_WAKEUP 0x01 |
ff77e468 PZ |
1808 | #define ENQUEUE_RESTORE 0x02 |
1809 | #define ENQUEUE_MOVE 0x04 | |
0a67d1ee | 1810 | #define ENQUEUE_NOCLOCK 0x08 |
ff77e468 | 1811 | |
0a67d1ee PZ |
1812 | #define ENQUEUE_HEAD 0x10 |
1813 | #define ENQUEUE_REPLENISH 0x20 | |
c82ba9fa | 1814 | #ifdef CONFIG_SMP |
0a67d1ee | 1815 | #define ENQUEUE_MIGRATED 0x40 |
c82ba9fa | 1816 | #else |
59efa0ba | 1817 | #define ENQUEUE_MIGRATED 0x00 |
c82ba9fa | 1818 | #endif |
c82ba9fa | 1819 | |
37e117c0 PZ |
1820 | #define RETRY_TASK ((void *)-1UL) |
1821 | ||
c82ba9fa | 1822 | struct sched_class { |
c82ba9fa | 1823 | |
69842cba PB |
1824 | #ifdef CONFIG_UCLAMP_TASK |
1825 | int uclamp_enabled; | |
1826 | #endif | |
1827 | ||
c82ba9fa LZ |
1828 | void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); |
1829 | void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); | |
97fb7a0a | 1830 | void (*yield_task) (struct rq *rq); |
0900acf2 | 1831 | bool (*yield_to_task)(struct rq *rq, struct task_struct *p); |
c82ba9fa | 1832 | |
97fb7a0a | 1833 | void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); |
c82ba9fa | 1834 | |
98c2f700 PZ |
1835 | struct task_struct *(*pick_next_task)(struct rq *rq); |
1836 | ||
6e2df058 | 1837 | void (*put_prev_task)(struct rq *rq, struct task_struct *p); |
a0e813f2 | 1838 | void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first); |
c82ba9fa LZ |
1839 | |
1840 | #ifdef CONFIG_SMP | |
6e2df058 | 1841 | int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); |
3aef1551 | 1842 | int (*select_task_rq)(struct task_struct *p, int task_cpu, int flags); |
1327237a | 1843 | void (*migrate_task_rq)(struct task_struct *p, int new_cpu); |
c82ba9fa | 1844 | |
97fb7a0a | 1845 | void (*task_woken)(struct rq *this_rq, struct task_struct *task); |
c82ba9fa LZ |
1846 | |
1847 | void (*set_cpus_allowed)(struct task_struct *p, | |
9cfc3e18 PZ |
1848 | const struct cpumask *newmask, |
1849 | u32 flags); | |
c82ba9fa LZ |
1850 | |
1851 | void (*rq_online)(struct rq *rq); | |
1852 | void (*rq_offline)(struct rq *rq); | |
a7c81556 PZ |
1853 | |
1854 | struct rq *(*find_lock_rq)(struct task_struct *p, struct rq *rq); | |
c82ba9fa LZ |
1855 | #endif |
1856 | ||
97fb7a0a IM |
1857 | void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); |
1858 | void (*task_fork)(struct task_struct *p); | |
1859 | void (*task_dead)(struct task_struct *p); | |
c82ba9fa | 1860 | |
67dfa1b7 KT |
1861 | /* |
1862 | * The switched_from() call is allowed to drop rq->lock, therefore we | |
1863 | * cannot assume the switched_from/switched_to pair is serliazed by | |
1864 | * rq->lock. They are however serialized by p->pi_lock. | |
1865 | */ | |
97fb7a0a IM |
1866 | void (*switched_from)(struct rq *this_rq, struct task_struct *task); |
1867 | void (*switched_to) (struct rq *this_rq, struct task_struct *task); | |
c82ba9fa | 1868 | void (*prio_changed) (struct rq *this_rq, struct task_struct *task, |
97fb7a0a | 1869 | int oldprio); |
c82ba9fa | 1870 | |
97fb7a0a IM |
1871 | unsigned int (*get_rr_interval)(struct rq *rq, |
1872 | struct task_struct *task); | |
c82ba9fa | 1873 | |
97fb7a0a | 1874 | void (*update_curr)(struct rq *rq); |
6e998916 | 1875 | |
97fb7a0a IM |
1876 | #define TASK_SET_GROUP 0 |
1877 | #define TASK_MOVE_GROUP 1 | |
ea86cb4b | 1878 | |
c82ba9fa | 1879 | #ifdef CONFIG_FAIR_GROUP_SCHED |
97fb7a0a | 1880 | void (*task_change_group)(struct task_struct *p, int type); |
c82ba9fa | 1881 | #endif |
43c31ac0 | 1882 | }; |
029632fb | 1883 | |
3f1d2a31 PZ |
1884 | static inline void put_prev_task(struct rq *rq, struct task_struct *prev) |
1885 | { | |
10e7071b | 1886 | WARN_ON_ONCE(rq->curr != prev); |
6e2df058 | 1887 | prev->sched_class->put_prev_task(rq, prev); |
3f1d2a31 PZ |
1888 | } |
1889 | ||
03b7fad1 | 1890 | static inline void set_next_task(struct rq *rq, struct task_struct *next) |
b2bf6c31 | 1891 | { |
03b7fad1 | 1892 | WARN_ON_ONCE(rq->curr != next); |
a0e813f2 | 1893 | next->sched_class->set_next_task(rq, next, false); |
b2bf6c31 PZ |
1894 | } |
1895 | ||
43c31ac0 PZ |
1896 | |
1897 | /* | |
1898 | * Helper to define a sched_class instance; each one is placed in a separate | |
1899 | * section which is ordered by the linker script: | |
1900 | * | |
1901 | * include/asm-generic/vmlinux.lds.h | |
1902 | * | |
1903 | * Also enforce alignment on the instance, not the type, to guarantee layout. | |
1904 | */ | |
1905 | #define DEFINE_SCHED_CLASS(name) \ | |
1906 | const struct sched_class name##_sched_class \ | |
1907 | __aligned(__alignof__(struct sched_class)) \ | |
1908 | __section("__" #name "_sched_class") | |
1909 | ||
c3a340f7 SRV |
1910 | /* Defined in include/asm-generic/vmlinux.lds.h */ |
1911 | extern struct sched_class __begin_sched_classes[]; | |
1912 | extern struct sched_class __end_sched_classes[]; | |
1913 | ||
1914 | #define sched_class_highest (__end_sched_classes - 1) | |
1915 | #define sched_class_lowest (__begin_sched_classes - 1) | |
6e2df058 PZ |
1916 | |
1917 | #define for_class_range(class, _from, _to) \ | |
c3a340f7 | 1918 | for (class = (_from); class != (_to); class--) |
6e2df058 | 1919 | |
029632fb | 1920 | #define for_each_class(class) \ |
c3a340f7 | 1921 | for_class_range(class, sched_class_highest, sched_class_lowest) |
029632fb PZ |
1922 | |
1923 | extern const struct sched_class stop_sched_class; | |
aab03e05 | 1924 | extern const struct sched_class dl_sched_class; |
029632fb PZ |
1925 | extern const struct sched_class rt_sched_class; |
1926 | extern const struct sched_class fair_sched_class; | |
1927 | extern const struct sched_class idle_sched_class; | |
1928 | ||
6e2df058 PZ |
1929 | static inline bool sched_stop_runnable(struct rq *rq) |
1930 | { | |
1931 | return rq->stop && task_on_rq_queued(rq->stop); | |
1932 | } | |
1933 | ||
1934 | static inline bool sched_dl_runnable(struct rq *rq) | |
1935 | { | |
1936 | return rq->dl.dl_nr_running > 0; | |
1937 | } | |
1938 | ||
1939 | static inline bool sched_rt_runnable(struct rq *rq) | |
1940 | { | |
1941 | return rq->rt.rt_queued > 0; | |
1942 | } | |
1943 | ||
1944 | static inline bool sched_fair_runnable(struct rq *rq) | |
1945 | { | |
1946 | return rq->cfs.nr_running > 0; | |
1947 | } | |
029632fb | 1948 | |
5d7d6056 | 1949 | extern struct task_struct *pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf); |
98c2f700 | 1950 | extern struct task_struct *pick_next_task_idle(struct rq *rq); |
5d7d6056 | 1951 | |
af449901 PZ |
1952 | #define SCA_CHECK 0x01 |
1953 | #define SCA_MIGRATE_DISABLE 0x02 | |
1954 | #define SCA_MIGRATE_ENABLE 0x04 | |
1955 | ||
029632fb PZ |
1956 | #ifdef CONFIG_SMP |
1957 | ||
63b2ca30 | 1958 | extern void update_group_capacity(struct sched_domain *sd, int cpu); |
b719203b | 1959 | |
7caff66f | 1960 | extern void trigger_load_balance(struct rq *rq); |
029632fb | 1961 | |
9cfc3e18 | 1962 | extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags); |
c5b28038 | 1963 | |
a7c81556 PZ |
1964 | static inline struct task_struct *get_push_task(struct rq *rq) |
1965 | { | |
1966 | struct task_struct *p = rq->curr; | |
1967 | ||
1968 | lockdep_assert_held(&rq->lock); | |
1969 | ||
1970 | if (rq->push_busy) | |
1971 | return NULL; | |
1972 | ||
1973 | if (p->nr_cpus_allowed == 1) | |
1974 | return NULL; | |
1975 | ||
1976 | rq->push_busy = true; | |
1977 | return get_task_struct(p); | |
1978 | } | |
1979 | ||
1980 | extern int push_cpu_stop(void *arg); | |
c5b28038 | 1981 | |
029632fb PZ |
1982 | #endif |
1983 | ||
442bf3aa DL |
1984 | #ifdef CONFIG_CPU_IDLE |
1985 | static inline void idle_set_state(struct rq *rq, | |
1986 | struct cpuidle_state *idle_state) | |
1987 | { | |
1988 | rq->idle_state = idle_state; | |
1989 | } | |
1990 | ||
1991 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
1992 | { | |
9148a3a1 | 1993 | SCHED_WARN_ON(!rcu_read_lock_held()); |
97fb7a0a | 1994 | |
442bf3aa DL |
1995 | return rq->idle_state; |
1996 | } | |
1997 | #else | |
1998 | static inline void idle_set_state(struct rq *rq, | |
1999 | struct cpuidle_state *idle_state) | |
2000 | { | |
2001 | } | |
2002 | ||
2003 | static inline struct cpuidle_state *idle_get_state(struct rq *rq) | |
2004 | { | |
2005 | return NULL; | |
2006 | } | |
2007 | #endif | |
2008 | ||
8663effb SRV |
2009 | extern void schedule_idle(void); |
2010 | ||
029632fb PZ |
2011 | extern void sysrq_sched_debug_show(void); |
2012 | extern void sched_init_granularity(void); | |
2013 | extern void update_max_interval(void); | |
1baca4ce JL |
2014 | |
2015 | extern void init_sched_dl_class(void); | |
029632fb PZ |
2016 | extern void init_sched_rt_class(void); |
2017 | extern void init_sched_fair_class(void); | |
2018 | ||
9059393e VG |
2019 | extern void reweight_task(struct task_struct *p, int prio); |
2020 | ||
8875125e | 2021 | extern void resched_curr(struct rq *rq); |
029632fb PZ |
2022 | extern void resched_cpu(int cpu); |
2023 | ||
2024 | extern struct rt_bandwidth def_rt_bandwidth; | |
2025 | extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); | |
2026 | ||
332ac17e DF |
2027 | extern struct dl_bandwidth def_dl_bandwidth; |
2028 | extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); | |
aab03e05 | 2029 | extern void init_dl_task_timer(struct sched_dl_entity *dl_se); |
209a0cbd | 2030 | extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); |
aab03e05 | 2031 | |
97fb7a0a IM |
2032 | #define BW_SHIFT 20 |
2033 | #define BW_UNIT (1 << BW_SHIFT) | |
2034 | #define RATIO_SHIFT 8 | |
d505b8af HC |
2035 | #define MAX_BW_BITS (64 - BW_SHIFT) |
2036 | #define MAX_BW ((1ULL << MAX_BW_BITS) - 1) | |
332ac17e DF |
2037 | unsigned long to_ratio(u64 period, u64 runtime); |
2038 | ||
540247fb | 2039 | extern void init_entity_runnable_average(struct sched_entity *se); |
d0fe0b9c | 2040 | extern void post_init_entity_util_avg(struct task_struct *p); |
a75cdaa9 | 2041 | |
76d92ac3 FW |
2042 | #ifdef CONFIG_NO_HZ_FULL |
2043 | extern bool sched_can_stop_tick(struct rq *rq); | |
d84b3131 | 2044 | extern int __init sched_tick_offload_init(void); |
76d92ac3 FW |
2045 | |
2046 | /* | |
2047 | * Tick may be needed by tasks in the runqueue depending on their policy and | |
2048 | * requirements. If tick is needed, lets send the target an IPI to kick it out of | |
2049 | * nohz mode if necessary. | |
2050 | */ | |
2051 | static inline void sched_update_tick_dependency(struct rq *rq) | |
2052 | { | |
21a6ee14 | 2053 | int cpu = cpu_of(rq); |
76d92ac3 FW |
2054 | |
2055 | if (!tick_nohz_full_cpu(cpu)) | |
2056 | return; | |
2057 | ||
2058 | if (sched_can_stop_tick(rq)) | |
2059 | tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); | |
2060 | else | |
2061 | tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); | |
2062 | } | |
2063 | #else | |
d84b3131 | 2064 | static inline int sched_tick_offload_init(void) { return 0; } |
76d92ac3 FW |
2065 | static inline void sched_update_tick_dependency(struct rq *rq) { } |
2066 | #endif | |
2067 | ||
72465447 | 2068 | static inline void add_nr_running(struct rq *rq, unsigned count) |
029632fb | 2069 | { |
72465447 KT |
2070 | unsigned prev_nr = rq->nr_running; |
2071 | ||
2072 | rq->nr_running = prev_nr + count; | |
9d246053 PA |
2073 | if (trace_sched_update_nr_running_tp_enabled()) { |
2074 | call_trace_sched_update_nr_running(rq, count); | |
2075 | } | |
9f3660c2 | 2076 | |
4486edd1 | 2077 | #ifdef CONFIG_SMP |
3e184501 | 2078 | if (prev_nr < 2 && rq->nr_running >= 2) { |
e90c8fe1 VS |
2079 | if (!READ_ONCE(rq->rd->overload)) |
2080 | WRITE_ONCE(rq->rd->overload, 1); | |
4486edd1 | 2081 | } |
3e184501 | 2082 | #endif |
76d92ac3 FW |
2083 | |
2084 | sched_update_tick_dependency(rq); | |
029632fb PZ |
2085 | } |
2086 | ||
72465447 | 2087 | static inline void sub_nr_running(struct rq *rq, unsigned count) |
029632fb | 2088 | { |
72465447 | 2089 | rq->nr_running -= count; |
9d246053 | 2090 | if (trace_sched_update_nr_running_tp_enabled()) { |
a1bd0685 | 2091 | call_trace_sched_update_nr_running(rq, -count); |
9d246053 PA |
2092 | } |
2093 | ||
76d92ac3 FW |
2094 | /* Check if we still need preemption */ |
2095 | sched_update_tick_dependency(rq); | |
029632fb PZ |
2096 | } |
2097 | ||
029632fb PZ |
2098 | extern void activate_task(struct rq *rq, struct task_struct *p, int flags); |
2099 | extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); | |
2100 | ||
2101 | extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); | |
2102 | ||
029632fb PZ |
2103 | extern const_debug unsigned int sysctl_sched_nr_migrate; |
2104 | extern const_debug unsigned int sysctl_sched_migration_cost; | |
2105 | ||
029632fb PZ |
2106 | #ifdef CONFIG_SCHED_HRTICK |
2107 | ||
2108 | /* | |
2109 | * Use hrtick when: | |
2110 | * - enabled by features | |
2111 | * - hrtimer is actually high res | |
2112 | */ | |
2113 | static inline int hrtick_enabled(struct rq *rq) | |
2114 | { | |
2115 | if (!sched_feat(HRTICK)) | |
2116 | return 0; | |
2117 | if (!cpu_active(cpu_of(rq))) | |
2118 | return 0; | |
2119 | return hrtimer_is_hres_active(&rq->hrtick_timer); | |
2120 | } | |
2121 | ||
2122 | void hrtick_start(struct rq *rq, u64 delay); | |
2123 | ||
b39e66ea MG |
2124 | #else |
2125 | ||
2126 | static inline int hrtick_enabled(struct rq *rq) | |
2127 | { | |
2128 | return 0; | |
2129 | } | |
2130 | ||
029632fb PZ |
2131 | #endif /* CONFIG_SCHED_HRTICK */ |
2132 | ||
1567c3e3 GG |
2133 | #ifndef arch_scale_freq_tick |
2134 | static __always_inline | |
2135 | void arch_scale_freq_tick(void) | |
2136 | { | |
2137 | } | |
2138 | #endif | |
2139 | ||
dfbca41f | 2140 | #ifndef arch_scale_freq_capacity |
f4470cdf VS |
2141 | /** |
2142 | * arch_scale_freq_capacity - get the frequency scale factor of a given CPU. | |
2143 | * @cpu: the CPU in question. | |
2144 | * | |
2145 | * Return: the frequency scale factor normalized against SCHED_CAPACITY_SCALE, i.e. | |
2146 | * | |
2147 | * f_curr | |
2148 | * ------ * SCHED_CAPACITY_SCALE | |
2149 | * f_max | |
2150 | */ | |
dfbca41f | 2151 | static __always_inline |
7673c8a4 | 2152 | unsigned long arch_scale_freq_capacity(int cpu) |
dfbca41f PZ |
2153 | { |
2154 | return SCHED_CAPACITY_SCALE; | |
2155 | } | |
2156 | #endif | |
b5b4860d | 2157 | |
029632fb | 2158 | #ifdef CONFIG_SMP |
c1a280b6 | 2159 | #ifdef CONFIG_PREEMPTION |
029632fb PZ |
2160 | |
2161 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
2162 | ||
2163 | /* | |
2164 | * fair double_lock_balance: Safely acquires both rq->locks in a fair | |
2165 | * way at the expense of forcing extra atomic operations in all | |
2166 | * invocations. This assures that the double_lock is acquired using the | |
2167 | * same underlying policy as the spinlock_t on this architecture, which | |
2168 | * reduces latency compared to the unfair variant below. However, it | |
2169 | * also adds more overhead and therefore may reduce throughput. | |
2170 | */ | |
2171 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
2172 | __releases(this_rq->lock) | |
2173 | __acquires(busiest->lock) | |
2174 | __acquires(this_rq->lock) | |
2175 | { | |
2176 | raw_spin_unlock(&this_rq->lock); | |
2177 | double_rq_lock(this_rq, busiest); | |
2178 | ||
2179 | return 1; | |
2180 | } | |
2181 | ||
2182 | #else | |
2183 | /* | |
2184 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
2185 | * latency by eliminating extra atomic operations when the locks are | |
97fb7a0a IM |
2186 | * already in proper order on entry. This favors lower CPU-ids and will |
2187 | * grant the double lock to lower CPUs over higher ids under contention, | |
029632fb PZ |
2188 | * regardless of entry order into the function. |
2189 | */ | |
2190 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
2191 | __releases(this_rq->lock) | |
2192 | __acquires(busiest->lock) | |
2193 | __acquires(this_rq->lock) | |
2194 | { | |
2195 | int ret = 0; | |
2196 | ||
2197 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { | |
2198 | if (busiest < this_rq) { | |
2199 | raw_spin_unlock(&this_rq->lock); | |
2200 | raw_spin_lock(&busiest->lock); | |
2201 | raw_spin_lock_nested(&this_rq->lock, | |
2202 | SINGLE_DEPTH_NESTING); | |
2203 | ret = 1; | |
2204 | } else | |
2205 | raw_spin_lock_nested(&busiest->lock, | |
2206 | SINGLE_DEPTH_NESTING); | |
2207 | } | |
2208 | return ret; | |
2209 | } | |
2210 | ||
c1a280b6 | 2211 | #endif /* CONFIG_PREEMPTION */ |
029632fb PZ |
2212 | |
2213 | /* | |
2214 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2215 | */ | |
2216 | static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
2217 | { | |
2218 | if (unlikely(!irqs_disabled())) { | |
97fb7a0a | 2219 | /* printk() doesn't work well under rq->lock */ |
029632fb PZ |
2220 | raw_spin_unlock(&this_rq->lock); |
2221 | BUG_ON(1); | |
2222 | } | |
2223 | ||
2224 | return _double_lock_balance(this_rq, busiest); | |
2225 | } | |
2226 | ||
2227 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
2228 | __releases(busiest->lock) | |
2229 | { | |
2230 | raw_spin_unlock(&busiest->lock); | |
2231 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
2232 | } | |
2233 | ||
74602315 PZ |
2234 | static inline void double_lock(spinlock_t *l1, spinlock_t *l2) |
2235 | { | |
2236 | if (l1 > l2) | |
2237 | swap(l1, l2); | |
2238 | ||
2239 | spin_lock(l1); | |
2240 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2241 | } | |
2242 | ||
60e69eed MG |
2243 | static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) |
2244 | { | |
2245 | if (l1 > l2) | |
2246 | swap(l1, l2); | |
2247 | ||
2248 | spin_lock_irq(l1); | |
2249 | spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2250 | } | |
2251 | ||
74602315 PZ |
2252 | static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) |
2253 | { | |
2254 | if (l1 > l2) | |
2255 | swap(l1, l2); | |
2256 | ||
2257 | raw_spin_lock(l1); | |
2258 | raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); | |
2259 | } | |
2260 | ||
029632fb PZ |
2261 | /* |
2262 | * double_rq_lock - safely lock two runqueues | |
2263 | * | |
2264 | * Note this does not disable interrupts like task_rq_lock, | |
2265 | * you need to do so manually before calling. | |
2266 | */ | |
2267 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
2268 | __acquires(rq1->lock) | |
2269 | __acquires(rq2->lock) | |
2270 | { | |
2271 | BUG_ON(!irqs_disabled()); | |
2272 | if (rq1 == rq2) { | |
2273 | raw_spin_lock(&rq1->lock); | |
2274 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2275 | } else { | |
2276 | if (rq1 < rq2) { | |
2277 | raw_spin_lock(&rq1->lock); | |
2278 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
2279 | } else { | |
2280 | raw_spin_lock(&rq2->lock); | |
2281 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
2282 | } | |
2283 | } | |
2284 | } | |
2285 | ||
2286 | /* | |
2287 | * double_rq_unlock - safely unlock two runqueues | |
2288 | * | |
2289 | * Note this does not restore interrupts like task_rq_unlock, | |
2290 | * you need to do so manually after calling. | |
2291 | */ | |
2292 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
2293 | __releases(rq1->lock) | |
2294 | __releases(rq2->lock) | |
2295 | { | |
2296 | raw_spin_unlock(&rq1->lock); | |
2297 | if (rq1 != rq2) | |
2298 | raw_spin_unlock(&rq2->lock); | |
2299 | else | |
2300 | __release(rq2->lock); | |
2301 | } | |
2302 | ||
f2cb1360 IM |
2303 | extern void set_rq_online (struct rq *rq); |
2304 | extern void set_rq_offline(struct rq *rq); | |
2305 | extern bool sched_smp_initialized; | |
2306 | ||
029632fb PZ |
2307 | #else /* CONFIG_SMP */ |
2308 | ||
2309 | /* | |
2310 | * double_rq_lock - safely lock two runqueues | |
2311 | * | |
2312 | * Note this does not disable interrupts like task_rq_lock, | |
2313 | * you need to do so manually before calling. | |
2314 | */ | |
2315 | static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
2316 | __acquires(rq1->lock) | |
2317 | __acquires(rq2->lock) | |
2318 | { | |
2319 | BUG_ON(!irqs_disabled()); | |
2320 | BUG_ON(rq1 != rq2); | |
2321 | raw_spin_lock(&rq1->lock); | |
2322 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2323 | } | |
2324 | ||
2325 | /* | |
2326 | * double_rq_unlock - safely unlock two runqueues | |
2327 | * | |
2328 | * Note this does not restore interrupts like task_rq_unlock, | |
2329 | * you need to do so manually after calling. | |
2330 | */ | |
2331 | static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
2332 | __releases(rq1->lock) | |
2333 | __releases(rq2->lock) | |
2334 | { | |
2335 | BUG_ON(rq1 != rq2); | |
2336 | raw_spin_unlock(&rq1->lock); | |
2337 | __release(rq2->lock); | |
2338 | } | |
2339 | ||
2340 | #endif | |
2341 | ||
2342 | extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); | |
2343 | extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); | |
6b55c965 SD |
2344 | |
2345 | #ifdef CONFIG_SCHED_DEBUG | |
9469eb01 PZ |
2346 | extern bool sched_debug_enabled; |
2347 | ||
029632fb PZ |
2348 | extern void print_cfs_stats(struct seq_file *m, int cpu); |
2349 | extern void print_rt_stats(struct seq_file *m, int cpu); | |
acb32132 | 2350 | extern void print_dl_stats(struct seq_file *m, int cpu); |
f6a34630 MM |
2351 | extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); |
2352 | extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); | |
2353 | extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); | |
397f2378 SD |
2354 | #ifdef CONFIG_NUMA_BALANCING |
2355 | extern void | |
2356 | show_numa_stats(struct task_struct *p, struct seq_file *m); | |
2357 | extern void | |
2358 | print_numa_stats(struct seq_file *m, int node, unsigned long tsf, | |
2359 | unsigned long tpf, unsigned long gsf, unsigned long gpf); | |
2360 | #endif /* CONFIG_NUMA_BALANCING */ | |
2361 | #endif /* CONFIG_SCHED_DEBUG */ | |
029632fb PZ |
2362 | |
2363 | extern void init_cfs_rq(struct cfs_rq *cfs_rq); | |
07c54f7a AV |
2364 | extern void init_rt_rq(struct rt_rq *rt_rq); |
2365 | extern void init_dl_rq(struct dl_rq *dl_rq); | |
029632fb | 2366 | |
1ee14e6c BS |
2367 | extern void cfs_bandwidth_usage_inc(void); |
2368 | extern void cfs_bandwidth_usage_dec(void); | |
1c792db7 | 2369 | |
3451d024 | 2370 | #ifdef CONFIG_NO_HZ_COMMON |
00357f5e PZ |
2371 | #define NOHZ_BALANCE_KICK_BIT 0 |
2372 | #define NOHZ_STATS_KICK_BIT 1 | |
a22e47a4 | 2373 | |
a22e47a4 | 2374 | #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) |
b7031a02 PZ |
2375 | #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) |
2376 | ||
2377 | #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) | |
1c792db7 SS |
2378 | |
2379 | #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) | |
20a5c8cc | 2380 | |
00357f5e | 2381 | extern void nohz_balance_exit_idle(struct rq *rq); |
20a5c8cc | 2382 | #else |
00357f5e | 2383 | static inline void nohz_balance_exit_idle(struct rq *rq) { } |
1c792db7 | 2384 | #endif |
73fbec60 | 2385 | |
daec5798 LA |
2386 | |
2387 | #ifdef CONFIG_SMP | |
2388 | static inline | |
2389 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2390 | { | |
2391 | struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); | |
2392 | int i; | |
2393 | ||
2394 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), | |
2395 | "sched RCU must be held"); | |
2396 | for_each_cpu_and(i, rd->span, cpu_active_mask) { | |
2397 | struct rq *rq = cpu_rq(i); | |
2398 | ||
2399 | rq->dl.extra_bw += bw; | |
2400 | } | |
2401 | } | |
2402 | #else | |
2403 | static inline | |
2404 | void __dl_update(struct dl_bw *dl_b, s64 bw) | |
2405 | { | |
2406 | struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); | |
2407 | ||
2408 | dl->extra_bw += bw; | |
2409 | } | |
2410 | #endif | |
2411 | ||
2412 | ||
73fbec60 | 2413 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
19d23dbf | 2414 | struct irqtime { |
25e2d8c1 | 2415 | u64 total; |
a499a5a1 | 2416 | u64 tick_delta; |
19d23dbf FW |
2417 | u64 irq_start_time; |
2418 | struct u64_stats_sync sync; | |
2419 | }; | |
73fbec60 | 2420 | |
19d23dbf | 2421 | DECLARE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 2422 | |
25e2d8c1 FW |
2423 | /* |
2424 | * Returns the irqtime minus the softirq time computed by ksoftirqd. | |
2425 | * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime | |
2426 | * and never move forward. | |
2427 | */ | |
73fbec60 FW |
2428 | static inline u64 irq_time_read(int cpu) |
2429 | { | |
19d23dbf FW |
2430 | struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); |
2431 | unsigned int seq; | |
2432 | u64 total; | |
73fbec60 FW |
2433 | |
2434 | do { | |
19d23dbf | 2435 | seq = __u64_stats_fetch_begin(&irqtime->sync); |
25e2d8c1 | 2436 | total = irqtime->total; |
19d23dbf | 2437 | } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); |
73fbec60 | 2438 | |
19d23dbf | 2439 | return total; |
73fbec60 | 2440 | } |
73fbec60 | 2441 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
adaf9fcd RW |
2442 | |
2443 | #ifdef CONFIG_CPU_FREQ | |
b10abd0a | 2444 | DECLARE_PER_CPU(struct update_util_data __rcu *, cpufreq_update_util_data); |
adaf9fcd RW |
2445 | |
2446 | /** | |
2447 | * cpufreq_update_util - Take a note about CPU utilization changes. | |
12bde33d | 2448 | * @rq: Runqueue to carry out the update for. |
58919e83 | 2449 | * @flags: Update reason flags. |
adaf9fcd | 2450 | * |
58919e83 RW |
2451 | * This function is called by the scheduler on the CPU whose utilization is |
2452 | * being updated. | |
adaf9fcd RW |
2453 | * |
2454 | * It can only be called from RCU-sched read-side critical sections. | |
adaf9fcd RW |
2455 | * |
2456 | * The way cpufreq is currently arranged requires it to evaluate the CPU | |
2457 | * performance state (frequency/voltage) on a regular basis to prevent it from | |
2458 | * being stuck in a completely inadequate performance level for too long. | |
e0367b12 JL |
2459 | * That is not guaranteed to happen if the updates are only triggered from CFS |
2460 | * and DL, though, because they may not be coming in if only RT tasks are | |
2461 | * active all the time (or there are RT tasks only). | |
adaf9fcd | 2462 | * |
e0367b12 JL |
2463 | * As a workaround for that issue, this function is called periodically by the |
2464 | * RT sched class to trigger extra cpufreq updates to prevent it from stalling, | |
adaf9fcd | 2465 | * but that really is a band-aid. Going forward it should be replaced with |
e0367b12 | 2466 | * solutions targeted more specifically at RT tasks. |
adaf9fcd | 2467 | */ |
12bde33d | 2468 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) |
adaf9fcd | 2469 | { |
58919e83 RW |
2470 | struct update_util_data *data; |
2471 | ||
674e7541 VK |
2472 | data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, |
2473 | cpu_of(rq))); | |
58919e83 | 2474 | if (data) |
12bde33d RW |
2475 | data->func(data, rq_clock(rq), flags); |
2476 | } | |
adaf9fcd | 2477 | #else |
12bde33d | 2478 | static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} |
adaf9fcd | 2479 | #endif /* CONFIG_CPU_FREQ */ |
be53f58f | 2480 | |
982d9cdc | 2481 | #ifdef CONFIG_UCLAMP_TASK |
686516b5 | 2482 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id); |
9d20ad7d | 2483 | |
46609ce2 QY |
2484 | /** |
2485 | * uclamp_rq_util_with - clamp @util with @rq and @p effective uclamp values. | |
2486 | * @rq: The rq to clamp against. Must not be NULL. | |
2487 | * @util: The util value to clamp. | |
2488 | * @p: The task to clamp against. Can be NULL if you want to clamp | |
2489 | * against @rq only. | |
2490 | * | |
2491 | * Clamps the passed @util to the max(@rq, @p) effective uclamp values. | |
2492 | * | |
2493 | * If sched_uclamp_used static key is disabled, then just return the util | |
2494 | * without any clamping since uclamp aggregation at the rq level in the fast | |
2495 | * path is disabled, rendering this operation a NOP. | |
2496 | * | |
2497 | * Use uclamp_eff_value() if you don't care about uclamp values at rq level. It | |
2498 | * will return the correct effective uclamp value of the task even if the | |
2499 | * static key is disabled. | |
2500 | */ | |
9d20ad7d | 2501 | static __always_inline |
d2b58a28 VS |
2502 | unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, |
2503 | struct task_struct *p) | |
982d9cdc | 2504 | { |
46609ce2 QY |
2505 | unsigned long min_util; |
2506 | unsigned long max_util; | |
2507 | ||
2508 | if (!static_branch_likely(&sched_uclamp_used)) | |
2509 | return util; | |
2510 | ||
2511 | min_util = READ_ONCE(rq->uclamp[UCLAMP_MIN].value); | |
2512 | max_util = READ_ONCE(rq->uclamp[UCLAMP_MAX].value); | |
982d9cdc | 2513 | |
9d20ad7d PB |
2514 | if (p) { |
2515 | min_util = max(min_util, uclamp_eff_value(p, UCLAMP_MIN)); | |
2516 | max_util = max(max_util, uclamp_eff_value(p, UCLAMP_MAX)); | |
2517 | } | |
2518 | ||
982d9cdc PB |
2519 | /* |
2520 | * Since CPU's {min,max}_util clamps are MAX aggregated considering | |
2521 | * RUNNABLE tasks with _different_ clamps, we can end up with an | |
2522 | * inversion. Fix it now when the clamps are applied. | |
2523 | */ | |
2524 | if (unlikely(min_util >= max_util)) | |
2525 | return min_util; | |
2526 | ||
2527 | return clamp(util, min_util, max_util); | |
2528 | } | |
46609ce2 QY |
2529 | |
2530 | /* | |
2531 | * When uclamp is compiled in, the aggregation at rq level is 'turned off' | |
2532 | * by default in the fast path and only gets turned on once userspace performs | |
2533 | * an operation that requires it. | |
2534 | * | |
2535 | * Returns true if userspace opted-in to use uclamp and aggregation at rq level | |
2536 | * hence is active. | |
2537 | */ | |
2538 | static inline bool uclamp_is_used(void) | |
2539 | { | |
2540 | return static_branch_likely(&sched_uclamp_used); | |
2541 | } | |
982d9cdc | 2542 | #else /* CONFIG_UCLAMP_TASK */ |
d2b58a28 VS |
2543 | static inline |
2544 | unsigned long uclamp_rq_util_with(struct rq *rq, unsigned long util, | |
2545 | struct task_struct *p) | |
9d20ad7d PB |
2546 | { |
2547 | return util; | |
2548 | } | |
46609ce2 QY |
2549 | |
2550 | static inline bool uclamp_is_used(void) | |
2551 | { | |
2552 | return false; | |
2553 | } | |
982d9cdc PB |
2554 | #endif /* CONFIG_UCLAMP_TASK */ |
2555 | ||
9bdcb44e | 2556 | #ifdef arch_scale_freq_capacity |
97fb7a0a IM |
2557 | # ifndef arch_scale_freq_invariant |
2558 | # define arch_scale_freq_invariant() true | |
2559 | # endif | |
2560 | #else | |
2561 | # define arch_scale_freq_invariant() false | |
9bdcb44e | 2562 | #endif |
d4edd662 | 2563 | |
10a35e68 VG |
2564 | #ifdef CONFIG_SMP |
2565 | static inline unsigned long capacity_orig_of(int cpu) | |
2566 | { | |
2567 | return cpu_rq(cpu)->cpu_capacity_orig; | |
2568 | } | |
2569 | #endif | |
2570 | ||
938e5e4b QP |
2571 | /** |
2572 | * enum schedutil_type - CPU utilization type | |
2573 | * @FREQUENCY_UTIL: Utilization used to select frequency | |
2574 | * @ENERGY_UTIL: Utilization used during energy calculation | |
2575 | * | |
2576 | * The utilization signals of all scheduling classes (CFS/RT/DL) and IRQ time | |
2577 | * need to be aggregated differently depending on the usage made of them. This | |
2578 | * enum is used within schedutil_freq_util() to differentiate the types of | |
2579 | * utilization expected by the callers, and adjust the aggregation accordingly. | |
2580 | */ | |
2581 | enum schedutil_type { | |
2582 | FREQUENCY_UTIL, | |
2583 | ENERGY_UTIL, | |
2584 | }; | |
2585 | ||
af24bde8 | 2586 | #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL |
938e5e4b | 2587 | |
af24bde8 PB |
2588 | unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, |
2589 | unsigned long max, enum schedutil_type type, | |
2590 | struct task_struct *p); | |
938e5e4b | 2591 | |
8cc90515 | 2592 | static inline unsigned long cpu_bw_dl(struct rq *rq) |
d4edd662 JL |
2593 | { |
2594 | return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; | |
2595 | } | |
2596 | ||
8cc90515 VG |
2597 | static inline unsigned long cpu_util_dl(struct rq *rq) |
2598 | { | |
2599 | return READ_ONCE(rq->avg_dl.util_avg); | |
2600 | } | |
2601 | ||
d4edd662 JL |
2602 | static inline unsigned long cpu_util_cfs(struct rq *rq) |
2603 | { | |
a07630b8 PB |
2604 | unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); |
2605 | ||
2606 | if (sched_feat(UTIL_EST)) { | |
2607 | util = max_t(unsigned long, util, | |
2608 | READ_ONCE(rq->cfs.avg.util_est.enqueued)); | |
2609 | } | |
2610 | ||
2611 | return util; | |
d4edd662 | 2612 | } |
371bf427 VG |
2613 | |
2614 | static inline unsigned long cpu_util_rt(struct rq *rq) | |
2615 | { | |
dfa444dc | 2616 | return READ_ONCE(rq->avg_rt.util_avg); |
371bf427 | 2617 | } |
938e5e4b | 2618 | #else /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
af24bde8 PB |
2619 | static inline unsigned long schedutil_cpu_util(int cpu, unsigned long util_cfs, |
2620 | unsigned long max, enum schedutil_type type, | |
2621 | struct task_struct *p) | |
938e5e4b | 2622 | { |
af24bde8 | 2623 | return 0; |
938e5e4b | 2624 | } |
af24bde8 | 2625 | #endif /* CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
9033ea11 | 2626 | |
11d4afd4 | 2627 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
9033ea11 VG |
2628 | static inline unsigned long cpu_util_irq(struct rq *rq) |
2629 | { | |
2630 | return rq->avg_irq.util_avg; | |
2631 | } | |
2e62c474 VG |
2632 | |
2633 | static inline | |
2634 | unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) | |
2635 | { | |
2636 | util *= (max - irq); | |
2637 | util /= max; | |
2638 | ||
2639 | return util; | |
2640 | ||
2641 | } | |
9033ea11 VG |
2642 | #else |
2643 | static inline unsigned long cpu_util_irq(struct rq *rq) | |
2644 | { | |
2645 | return 0; | |
2646 | } | |
2647 | ||
2e62c474 VG |
2648 | static inline |
2649 | unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) | |
2650 | { | |
2651 | return util; | |
2652 | } | |
794a56eb | 2653 | #endif |
6aa140fa | 2654 | |
531b5c9f | 2655 | #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) |
f8a696f2 | 2656 | |
6aa140fa | 2657 | #define perf_domain_span(pd) (to_cpumask(((pd)->em_pd->cpus))) |
f8a696f2 PZ |
2658 | |
2659 | DECLARE_STATIC_KEY_FALSE(sched_energy_present); | |
2660 | ||
2661 | static inline bool sched_energy_enabled(void) | |
2662 | { | |
2663 | return static_branch_unlikely(&sched_energy_present); | |
2664 | } | |
2665 | ||
2666 | #else /* ! (CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL) */ | |
2667 | ||
6aa140fa | 2668 | #define perf_domain_span(pd) NULL |
f8a696f2 | 2669 | static inline bool sched_energy_enabled(void) { return false; } |
1f74de87 | 2670 | |
f8a696f2 | 2671 | #endif /* CONFIG_ENERGY_MODEL && CONFIG_CPU_FREQ_GOV_SCHEDUTIL */ |
227a4aad MD |
2672 | |
2673 | #ifdef CONFIG_MEMBARRIER | |
2674 | /* | |
2675 | * The scheduler provides memory barriers required by membarrier between: | |
2676 | * - prior user-space memory accesses and store to rq->membarrier_state, | |
2677 | * - store to rq->membarrier_state and following user-space memory accesses. | |
2678 | * In the same way it provides those guarantees around store to rq->curr. | |
2679 | */ | |
2680 | static inline void membarrier_switch_mm(struct rq *rq, | |
2681 | struct mm_struct *prev_mm, | |
2682 | struct mm_struct *next_mm) | |
2683 | { | |
2684 | int membarrier_state; | |
2685 | ||
2686 | if (prev_mm == next_mm) | |
2687 | return; | |
2688 | ||
2689 | membarrier_state = atomic_read(&next_mm->membarrier_state); | |
2690 | if (READ_ONCE(rq->membarrier_state) == membarrier_state) | |
2691 | return; | |
2692 | ||
2693 | WRITE_ONCE(rq->membarrier_state, membarrier_state); | |
2694 | } | |
2695 | #else | |
2696 | static inline void membarrier_switch_mm(struct rq *rq, | |
2697 | struct mm_struct *prev_mm, | |
2698 | struct mm_struct *next_mm) | |
2699 | { | |
2700 | } | |
2701 | #endif | |
52262ee5 MG |
2702 | |
2703 | #ifdef CONFIG_SMP | |
2704 | static inline bool is_per_cpu_kthread(struct task_struct *p) | |
2705 | { | |
2706 | if (!(p->flags & PF_KTHREAD)) | |
2707 | return false; | |
2708 | ||
2709 | if (p->nr_cpus_allowed != 1) | |
2710 | return false; | |
2711 | ||
2712 | return true; | |
2713 | } | |
2714 | #endif | |
b3212fe2 TG |
2715 | |
2716 | void swake_up_all_locked(struct swait_queue_head *q); | |
2717 | void __prepare_to_swait(struct swait_queue_head *q, struct swait_queue *wait); |