root/kernel/sched/core.c

DEFINITIONS

1. __task_rq_lock
2. task_rq_lock
3. update_rq_clock_task
4. update_rq_clock
5. hrtick_clear
6. hrtick
7. __hrtick_restart
8. __hrtick_start
9. hrtick_start
10. hrtick_start
11. hrtick_rq_init
12. hrtick_clear
13. hrtick_rq_init
14. set_nr_and_not_polling
15. set_nr_if_polling
16. set_nr_and_not_polling
17. set_nr_if_polling
18. __wake_q_add
19. wake_q_add
20. wake_q_add_safe
21. wake_up_q
22. resched_curr
23. resched_cpu
24. get_nohz_timer_target
25. wake_up_idle_cpu
26. wake_up_full_nohz_cpu
27. wake_up_nohz_cpu
28. got_nohz_idle_kick
29. got_nohz_idle_kick
30. sched_can_stop_tick
31. walk_tg_tree_from
32. tg_nop
33. set_load_weight
34. uclamp_bucket_id
35. uclamp_bucket_base_value
36. uclamp_none
37. uclamp_se_set
38. uclamp_idle_value
39. uclamp_idle_reset
40. uclamp_rq_max_value
41. uclamp_tg_restrict
42. uclamp_eff_get
43. uclamp_eff_value
44. uclamp_rq_inc_id
45. uclamp_rq_dec_id
46. uclamp_rq_inc
47. uclamp_rq_dec
48. uclamp_update_active
49. uclamp_update_active_tasks
50. uclamp_update_root_tg
51. uclamp_update_root_tg
52. sysctl_sched_uclamp_handler
53. uclamp_validate
54. __setscheduler_uclamp
55. uclamp_fork
56. init_uclamp
57. uclamp_rq_inc
58. uclamp_rq_dec
59. uclamp_validate
60. __setscheduler_uclamp
61. uclamp_fork
62. init_uclamp
63. enqueue_task
64. dequeue_task
65. activate_task
66. deactivate_task
67. __normal_prio
68. normal_prio
69. effective_prio
70. task_curr
71. check_class_changed
72. check_preempt_curr
73. is_per_cpu_kthread
74. is_cpu_allowed
75. move_queued_task
76. __migrate_task
77. migration_cpu_stop
78. set_cpus_allowed_common
79. do_set_cpus_allowed
80. __set_cpus_allowed_ptr
81. set_cpus_allowed_ptr
82. set_task_cpu
83. __migrate_swap_task
84. migrate_swap_stop
85. migrate_swap
86. wait_task_inactive
87. kick_process
88. select_fallback_rq
89. select_task_rq
90. update_avg
91. sched_set_stop_task
92. __set_cpus_allowed_ptr
93. ttwu_stat
94. ttwu_do_wakeup
95. ttwu_do_activate
96. ttwu_remote
97. sched_ttwu_pending
98. scheduler_ipi
99. ttwu_queue_remote
100. wake_up_if_idle
101. cpus_share_cache
102. ttwu_queue
103. try_to_wake_up
104. wake_up_process
105. wake_up_state
106. __sched_fork
107. set_numabalancing_state
108. sysctl_numa_balancing
109. set_schedstats
110. force_schedstat_enabled
111. setup_schedstats
112. init_schedstats
113. sysctl_schedstats
114. init_schedstats
115. sched_fork
116. to_ratio
117. wake_up_new_task
118. preempt_notifier_inc
119. preempt_notifier_dec
120. preempt_notifier_register
121. preempt_notifier_unregister
122. __fire_sched_in_preempt_notifiers
123. fire_sched_in_preempt_notifiers
124. __fire_sched_out_preempt_notifiers
125. fire_sched_out_preempt_notifiers
126. fire_sched_in_preempt_notifiers
127. fire_sched_out_preempt_notifiers
128. prepare_task
129. finish_task
130. prepare_lock_switch
131. finish_lock_switch
132. prepare_task_switch
133. finish_task_switch
134. __balance_callback
135. balance_callback
136. balance_callback
137. schedule_tail
138. context_switch
139. nr_running
140. single_task_running
141. nr_context_switches
142. nr_iowait_cpu
143. nr_iowait
144. sched_exec
145. prefetch_curr_exec_start
146. task_sched_runtime
147. scheduler_tick
148. sched_tick_remote
149. sched_tick_start
150. sched_tick_stop
151. sched_tick_offload_init
152. sched_tick_start
153. sched_tick_stop
154. preempt_latency_start
155. preempt_count_add
156. preempt_latency_stop
157. preempt_count_sub
158. preempt_latency_start
159. preempt_latency_stop
160. get_preempt_disable_ip
161. __schedule_bug
162. schedule_debug
163. pick_next_task
164. __schedule
165. do_task_dead
166. sched_submit_work
167. sched_update_worker
168. schedule
169. schedule_idle
170. schedule_user
171. schedule_preempt_disabled
172. preempt_schedule_common
173. preempt_schedule
174. preempt_schedule_notrace
175. preempt_schedule_irq
176. default_wake_function
177. __rt_effective_prio
178. rt_effective_prio
179. rt_mutex_setprio
180. rt_effective_prio
181. set_user_nice
182. can_nice
183. SYSCALL_DEFINE1
184. task_prio
185. idle_cpu
186. available_idle_cpu
187. idle_task
188. find_process_by_pid
189. __setscheduler_params
190. __setscheduler
191. check_same_owner
192. __sched_setscheduler
193. _sched_setscheduler
194. sched_setscheduler
195. sched_setattr
196. sched_setattr_nocheck
197. sched_setscheduler_nocheck
198. do_sched_setscheduler
199. sched_copy_attr
200. SYSCALL_DEFINE3
201. SYSCALL_DEFINE2
202. SYSCALL_DEFINE3
203. SYSCALL_DEFINE1
204. SYSCALL_DEFINE2
205. sched_attr_copy_to_user
206. SYSCALL_DEFINE4
207. sched_setaffinity
208. get_user_cpu_mask
209. SYSCALL_DEFINE3
210. sched_getaffinity
211. SYSCALL_DEFINE3
212. do_sched_yield
213. SYSCALL_DEFINE0
214. _cond_resched
215. __cond_resched_lock
216. yield
217. yield_to
218. io_schedule_prepare
219. io_schedule_finish
220. io_schedule_timeout
221. io_schedule
222. SYSCALL_DEFINE1
223. SYSCALL_DEFINE1
224. sched_rr_get_interval
225. SYSCALL_DEFINE2
226. SYSCALL_DEFINE2
227. sched_show_task
228. state_filter_match
229. show_state_filter
230. init_idle
231. cpuset_cpumask_can_shrink
232. task_can_attach
233. migrate_task_to
234. sched_setnuma
235. idle_task_exit
236. calc_load_migrate
237. __pick_migrate_task
238. migrate_tasks
239. set_rq_online
240. set_rq_offline
241. cpuset_cpu_active
242. cpuset_cpu_inactive
243. sched_cpu_activate
244. sched_cpu_deactivate
245. sched_rq_cpu_starting
246. sched_cpu_starting
247. sched_cpu_dying
248. sched_init_smp
249. migration_init
250. sched_init_smp
251. in_sched_functions
252. sched_init
253. preempt_count_equals
254. __might_sleep
255. ___might_sleep
256. __cant_sleep
257. normalize_rt_tasks
258. curr_task
259. ia64_set_curr_task
260. alloc_uclamp_sched_group
261. sched_free_group
262. sched_create_group
263. sched_online_group
264. sched_free_group_rcu
265. sched_destroy_group
266. sched_offline_group
267. sched_change_group
268. sched_move_task
269. css_tg
270. cpu_cgroup_css_alloc
271. cpu_cgroup_css_online
272. cpu_cgroup_css_released
273. cpu_cgroup_css_free
274. cpu_cgroup_fork
275. cpu_cgroup_can_attach
276. cpu_cgroup_attach
277. cpu_util_update_eff
278. capacity_from_percent
279. cpu_uclamp_write
280. cpu_uclamp_min_write
281. cpu_uclamp_max_write
282. cpu_uclamp_print
283. cpu_uclamp_min_show
284. cpu_uclamp_max_show
285. cpu_shares_write_u64
286. cpu_shares_read_u64
287. tg_set_cfs_bandwidth
288. tg_set_cfs_quota
289. tg_get_cfs_quota
290. tg_set_cfs_period
291. tg_get_cfs_period
292. cpu_cfs_quota_read_s64
293. cpu_cfs_quota_write_s64
294. cpu_cfs_period_read_u64
295. cpu_cfs_period_write_u64
296. normalize_cfs_quota
297. tg_cfs_schedulable_down
298. __cfs_schedulable
299. cpu_cfs_stat_show
300. cpu_rt_runtime_write
301. cpu_rt_runtime_read
302. cpu_rt_period_write_uint
303. cpu_rt_period_read_uint
304. cpu_extra_stat_show
305. cpu_weight_read_u64
306. cpu_weight_write_u64
307. cpu_weight_nice_read_s64
308. cpu_weight_nice_write_s64
309. cpu_period_quota_print
310. cpu_period_quota_parse
311. cpu_max_show
312. cpu_max_write
313. dump_cpu_task

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  kernel/sched/core.c
   4  *
   5  *  Core kernel scheduler code and related syscalls
   6  *
   7  *  Copyright (C) 1991-2002  Linus Torvalds
   8  */
   9 #include "sched.h"
  10 
  11 #include <linux/nospec.h>
  12 
  13 #include <linux/kcov.h>
  14 
  15 #include <asm/switch_to.h>
  16 #include <asm/tlb.h>
  17 
  18 #include "../workqueue_internal.h"
  19 #include "../smpboot.h"
  20 
  21 #include "pelt.h"
  22 
  23 #define CREATE_TRACE_POINTS
  24 #include <trace/events/sched.h>
  25 
  26 /*
  27  * Export tracepoints that act as a bare tracehook (ie: have no trace event
  28  * associated with them) to allow external modules to probe them.
  29  */
  30 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp);
  31 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp);
  32 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp);
  33 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp);
  34 EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp);
  35 EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp);
  36 
  37 DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
  38 
  39 #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
  40 /*
  41  * Debugging: various feature bits
  42  *
  43  * If SCHED_DEBUG is disabled, each compilation unit has its own copy of
  44  * sysctl_sched_features, defined in sched.h, to allow constants propagation
  45  * at compile time and compiler optimization based on features default.
  46  */
  47 #define SCHED_FEAT(name, enabled)       \
  48         (1UL << __SCHED_FEAT_##name) * enabled |
  49 const_debug unsigned int sysctl_sched_features =
  50 #include "features.h"
  51         0;
  52 #undef SCHED_FEAT
  53 #endif
  54 
  55 /*
  56  * Number of tasks to iterate in a single balance run.
  57  * Limited because this is done with IRQs disabled.
  58  */
  59 const_debug unsigned int sysctl_sched_nr_migrate = 32;
  60 
  61 /*
  62  * period over which we measure -rt task CPU usage in us.
  63  * default: 1s
  64  */
  65 unsigned int sysctl_sched_rt_period = 1000000;
  66 
  67 __read_mostly int scheduler_running;
  68 
  69 /*
  70  * part of the period that we allow rt tasks to run in us.
  71  * default: 0.95s
  72  */
  73 int sysctl_sched_rt_runtime = 950000;
  74 
  75 /*
  76  * __task_rq_lock - lock the rq @p resides on.
  77  */
  78 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
  79         __acquires(rq->lock)
  80 {
  81         struct rq *rq;
  82 
  83         lockdep_assert_held(&p->pi_lock);
  84 
  85         for (;;) {
  86                 rq = task_rq(p);
  87                 raw_spin_lock(&rq->lock);
  88                 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
  89                         rq_pin_lock(rq, rf);
  90                         return rq;
  91                 }
  92                 raw_spin_unlock(&rq->lock);
  93 
  94                 while (unlikely(task_on_rq_migrating(p)))
  95                         cpu_relax();
  96         }
  97 }
  98 
  99 /*
 100  * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
 101  */
 102 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
 103         __acquires(p->pi_lock)
 104         __acquires(rq->lock)
 105 {
 106         struct rq *rq;
 107 
 108         for (;;) {
 109                 raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
 110                 rq = task_rq(p);
 111                 raw_spin_lock(&rq->lock);
 112                 /*
 113                  *      move_queued_task()              task_rq_lock()
 114                  *
 115                  *      ACQUIRE (rq->lock)
 116                  *      [S] ->on_rq = MIGRATING         [L] rq = task_rq()
 117                  *      WMB (__set_task_cpu())          ACQUIRE (rq->lock);
 118                  *      [S] ->cpu = new_cpu             [L] task_rq()
 119                  *                                      [L] ->on_rq
 120                  *      RELEASE (rq->lock)
 121                  *
 122                  * If we observe the old CPU in task_rq_lock(), the acquire of
 123                  * the old rq->lock will fully serialize against the stores.
 124                  *
 125                  * If we observe the new CPU in task_rq_lock(), the address
 126                  * dependency headed by '[L] rq = task_rq()' and the acquire
 127                  * will pair with the WMB to ensure we then also see migrating.
 128                  */
 129                 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
 130                         rq_pin_lock(rq, rf);
 131                         return rq;
 132                 }
 133                 raw_spin_unlock(&rq->lock);
 134                 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
 135 
 136                 while (unlikely(task_on_rq_migrating(p)))
 137                         cpu_relax();
 138         }
 139 }
 140 
 141 /*
 142  * RQ-clock updating methods:
 143  */
 144 
 145 static void update_rq_clock_task(struct rq *rq, s64 delta)
 146 {
 147 /*
 148  * In theory, the compile should just see 0 here, and optimize out the call
 149  * to sched_rt_avg_update. But I don't trust it...
 150  */
 151         s64 __maybe_unused steal = 0, irq_delta = 0;
 152 
 153 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
 154         irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
 155 
 156         /*
 157          * Since irq_time is only updated on {soft,}irq_exit, we might run into
 158          * this case when a previous update_rq_clock() happened inside a
 159          * {soft,}irq region.
 160          *
 161          * When this happens, we stop ->clock_task and only update the
 162          * prev_irq_time stamp to account for the part that fit, so that a next
 163          * update will consume the rest. This ensures ->clock_task is
 164          * monotonic.
 165          *
 166          * It does however cause some slight miss-attribution of {soft,}irq
 167          * time, a more accurate solution would be to update the irq_time using
 168          * the current rq->clock timestamp, except that would require using
 169          * atomic ops.
 170          */
 171         if (irq_delta > delta)
 172                 irq_delta = delta;
 173 
 174         rq->prev_irq_time += irq_delta;
 175         delta -= irq_delta;
 176 #endif
 177 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
 178         if (static_key_false((&paravirt_steal_rq_enabled))) {
 179                 steal = paravirt_steal_clock(cpu_of(rq));
 180                 steal -= rq->prev_steal_time_rq;
 181 
 182                 if (unlikely(steal > delta))
 183                         steal = delta;
 184 
 185                 rq->prev_steal_time_rq += steal;
 186                 delta -= steal;
 187         }
 188 #endif
 189 
 190         rq->clock_task += delta;
 191 
 192 #ifdef CONFIG_HAVE_SCHED_AVG_IRQ
 193         if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
 194                 update_irq_load_avg(rq, irq_delta + steal);
 195 #endif
 196         update_rq_clock_pelt(rq, delta);
 197 }
 198 
 199 void update_rq_clock(struct rq *rq)
 200 {
 201         s64 delta;
 202 
 203         lockdep_assert_held(&rq->lock);
 204 
 205         if (rq->clock_update_flags & RQCF_ACT_SKIP)
 206                 return;
 207 
 208 #ifdef CONFIG_SCHED_DEBUG
 209         if (sched_feat(WARN_DOUBLE_CLOCK))
 210                 SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED);
 211         rq->clock_update_flags |= RQCF_UPDATED;
 212 #endif
 213 
 214         delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
 215         if (delta < 0)
 216                 return;
 217         rq->clock += delta;
 218         update_rq_clock_task(rq, delta);
 219 }
 220 
 221 
 222 #ifdef CONFIG_SCHED_HRTICK
 223 /*
 224  * Use HR-timers to deliver accurate preemption points.
 225  */
 226 
 227 static void hrtick_clear(struct rq *rq)
 228 {
 229         if (hrtimer_active(&rq->hrtick_timer))
 230                 hrtimer_cancel(&rq->hrtick_timer);
 231 }
 232 
 233 /*
 234  * High-resolution timer tick.
 235  * Runs from hardirq context with interrupts disabled.
 236  */
 237 static enum hrtimer_restart hrtick(struct hrtimer *timer)
 238 {
 239         struct rq *rq = container_of(timer, struct rq, hrtick_timer);
 240         struct rq_flags rf;
 241 
 242         WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
 243 
 244         rq_lock(rq, &rf);
 245         update_rq_clock(rq);
 246         rq->curr->sched_class->task_tick(rq, rq->curr, 1);
 247         rq_unlock(rq, &rf);
 248 
 249         return HRTIMER_NORESTART;
 250 }
 251 
 252 #ifdef CONFIG_SMP
 253 
 254 static void __hrtick_restart(struct rq *rq)
 255 {
 256         struct hrtimer *timer = &rq->hrtick_timer;
 257 
 258         hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED_HARD);
 259 }
 260 
 261 /*
 262  * called from hardirq (IPI) context
 263  */
 264 static void __hrtick_start(void *arg)
 265 {
 266         struct rq *rq = arg;
 267         struct rq_flags rf;
 268 
 269         rq_lock(rq, &rf);
 270         __hrtick_restart(rq);
 271         rq->hrtick_csd_pending = 0;
 272         rq_unlock(rq, &rf);
 273 }
 274 
 275 /*
 276  * Called to set the hrtick timer state.
 277  *
 278  * called with rq->lock held and irqs disabled
 279  */
 280 void hrtick_start(struct rq *rq, u64 delay)
 281 {
 282         struct hrtimer *timer = &rq->hrtick_timer;
 283         ktime_t time;
 284         s64 delta;
 285 
 286         /*
 287          * Don't schedule slices shorter than 10000ns, that just
 288          * doesn't make sense and can cause timer DoS.
 289          */
 290         delta = max_t(s64, delay, 10000LL);
 291         time = ktime_add_ns(timer->base->get_time(), delta);
 292 
 293         hrtimer_set_expires(timer, time);
 294 
 295         if (rq == this_rq()) {
 296                 __hrtick_restart(rq);
 297         } else if (!rq->hrtick_csd_pending) {
 298                 smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
 299                 rq->hrtick_csd_pending = 1;
 300         }
 301 }
 302 
 303 #else
 304 /*
 305  * Called to set the hrtick timer state.
 306  *
 307  * called with rq->lock held and irqs disabled
 308  */
 309 void hrtick_start(struct rq *rq, u64 delay)
 310 {
 311         /*
 312          * Don't schedule slices shorter than 10000ns, that just
 313          * doesn't make sense. Rely on vruntime for fairness.
 314          */
 315         delay = max_t(u64, delay, 10000LL);
 316         hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
 317                       HRTIMER_MODE_REL_PINNED_HARD);
 318 }
 319 #endif /* CONFIG_SMP */
 320 
 321 static void hrtick_rq_init(struct rq *rq)
 322 {
 323 #ifdef CONFIG_SMP
 324         rq->hrtick_csd_pending = 0;
 325 
 326         rq->hrtick_csd.flags = 0;
 327         rq->hrtick_csd.func = __hrtick_start;
 328         rq->hrtick_csd.info = rq;
 329 #endif
 330 
 331         hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
 332         rq->hrtick_timer.function = hrtick;
 333 }
 334 #else   /* CONFIG_SCHED_HRTICK */
 335 static inline void hrtick_clear(struct rq *rq)
 336 {
 337 }
 338 
 339 static inline void hrtick_rq_init(struct rq *rq)
 340 {
 341 }
 342 #endif  /* CONFIG_SCHED_HRTICK */
 343 
 344 /*
 345  * cmpxchg based fetch_or, macro so it works for different integer types
 346  */
 347 #define fetch_or(ptr, mask)                                             \
 348         ({                                                              \
 349                 typeof(ptr) _ptr = (ptr);                               \
 350                 typeof(mask) _mask = (mask);                            \
 351                 typeof(*_ptr) _old, _val = *_ptr;                       \
 352                                                                         \
 353                 for (;;) {                                              \
 354                         _old = cmpxchg(_ptr, _val, _val | _mask);       \
 355                         if (_old == _val)                               \
 356                                 break;                                  \
 357                         _val = _old;                                    \
 358                 }                                                       \
 359         _old;                                                           \
 360 })
 361 
 362 #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
 363 /*
 364  * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG,
 365  * this avoids any races wrt polling state changes and thereby avoids
 366  * spurious IPIs.
 367  */
 368 static bool set_nr_and_not_polling(struct task_struct *p)
 369 {
 370         struct thread_info *ti = task_thread_info(p);
 371         return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG);
 372 }
 373 
 374 /*
 375  * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set.
 376  *
 377  * If this returns true, then the idle task promises to call
 378  * sched_ttwu_pending() and reschedule soon.
 379  */
 380 static bool set_nr_if_polling(struct task_struct *p)
 381 {
 382         struct thread_info *ti = task_thread_info(p);
 383         typeof(ti->flags) old, val = READ_ONCE(ti->flags);
 384 
 385         for (;;) {
 386                 if (!(val & _TIF_POLLING_NRFLAG))
 387                         return false;
 388                 if (val & _TIF_NEED_RESCHED)
 389                         return true;
 390                 old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED);
 391                 if (old == val)
 392                         break;
 393                 val = old;
 394         }
 395         return true;
 396 }
 397 
 398 #else
 399 static bool set_nr_and_not_polling(struct task_struct *p)
 400 {
 401         set_tsk_need_resched(p);
 402         return true;
 403 }
 404 
 405 #ifdef CONFIG_SMP
 406 static bool set_nr_if_polling(struct task_struct *p)
 407 {
 408         return false;
 409 }
 410 #endif
 411 #endif
 412 
 413 static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task)
 414 {
 415         struct wake_q_node *node = &task->wake_q;
 416 
 417         /*
 418          * Atomically grab the task, if ->wake_q is !nil already it means
 419          * its already queued (either by us or someone else) and will get the
 420          * wakeup due to that.
 421          *
 422          * In order to ensure that a pending wakeup will observe our pending
 423          * state, even in the failed case, an explicit smp_mb() must be used.
 424          */
 425         smp_mb__before_atomic();
 426         if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL)))
 427                 return false;
 428 
 429         /*
 430          * The head is context local, there can be no concurrency.
 431          */
 432         *head->lastp = node;
 433         head->lastp = &node->next;
 434         return true;
 435 }
 436 
 437 /**
 438  * wake_q_add() - queue a wakeup for 'later' waking.
 439  * @head: the wake_q_head to add @task to
 440  * @task: the task to queue for 'later' wakeup
 441  *
 442  * Queue a task for later wakeup, most likely by the wake_up_q() call in the
 443  * same context, _HOWEVER_ this is not guaranteed, the wakeup can come
 444  * instantly.
 445  *
 446  * This function must be used as-if it were wake_up_process(); IOW the task
 447  * must be ready to be woken at this location.
 448  */
 449 void wake_q_add(struct wake_q_head *head, struct task_struct *task)
 450 {
 451         if (__wake_q_add(head, task))
 452                 get_task_struct(task);
 453 }
 454 
 455 /**
 456  * wake_q_add_safe() - safely queue a wakeup for 'later' waking.
 457  * @head: the wake_q_head to add @task to
 458  * @task: the task to queue for 'later' wakeup
 459  *
 460  * Queue a task for later wakeup, most likely by the wake_up_q() call in the
 461  * same context, _HOWEVER_ this is not guaranteed, the wakeup can come
 462  * instantly.
 463  *
 464  * This function must be used as-if it were wake_up_process(); IOW the task
 465  * must be ready to be woken at this location.
 466  *
 467  * This function is essentially a task-safe equivalent to wake_q_add(). Callers
 468  * that already hold reference to @task can call the 'safe' version and trust
 469  * wake_q to do the right thing depending whether or not the @task is already
 470  * queued for wakeup.
 471  */
 472 void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task)
 473 {
 474         if (!__wake_q_add(head, task))
 475                 put_task_struct(task);
 476 }
 477 
 478 void wake_up_q(struct wake_q_head *head)
 479 {
 480         struct wake_q_node *node = head->first;
 481 
 482         while (node != WAKE_Q_TAIL) {
 483                 struct task_struct *task;
 484 
 485                 task = container_of(node, struct task_struct, wake_q);
 486                 BUG_ON(!task);
 487                 /* Task can safely be re-inserted now: */
 488                 node = node->next;
 489                 task->wake_q.next = NULL;
 490 
 491                 /*
 492                  * wake_up_process() executes a full barrier, which pairs with
 493                  * the queueing in wake_q_add() so as not to miss wakeups.
 494                  */
 495                 wake_up_process(task);
 496                 put_task_struct(task);
 497         }
 498 }
 499 
 500 /*
 501  * resched_curr - mark rq's current task 'to be rescheduled now'.
 502  *
 503  * On UP this means the setting of the need_resched flag, on SMP it
 504  * might also involve a cross-CPU call to trigger the scheduler on
 505  * the target CPU.
 506  */
 507 void resched_curr(struct rq *rq)
 508 {
 509         struct task_struct *curr = rq->curr;
 510         int cpu;
 511 
 512         lockdep_assert_held(&rq->lock);
 513 
 514         if (test_tsk_need_resched(curr))
 515                 return;
 516 
 517         cpu = cpu_of(rq);
 518 
 519         if (cpu == smp_processor_id()) {
 520                 set_tsk_need_resched(curr);
 521                 set_preempt_need_resched();
 522                 return;
 523         }
 524 
 525         if (set_nr_and_not_polling(curr))
 526                 smp_send_reschedule(cpu);
 527         else
 528                 trace_sched_wake_idle_without_ipi(cpu);
 529 }
 530 
 531 void resched_cpu(int cpu)
 532 {
 533         struct rq *rq = cpu_rq(cpu);
 534         unsigned long flags;
 535 
 536         raw_spin_lock_irqsave(&rq->lock, flags);
 537         if (cpu_online(cpu) || cpu == smp_processor_id())
 538                 resched_curr(rq);
 539         raw_spin_unlock_irqrestore(&rq->lock, flags);
 540 }
 541 
 542 #ifdef CONFIG_SMP
 543 #ifdef CONFIG_NO_HZ_COMMON
 544 /*
 545  * In the semi idle case, use the nearest busy CPU for migrating timers
 546  * from an idle CPU.  This is good for power-savings.
 547  *
 548  * We don't do similar optimization for completely idle system, as
 549  * selecting an idle CPU will add more delays to the timers than intended
 550  * (as that CPU's timer base may not be uptodate wrt jiffies etc).
 551  */
 552 int get_nohz_timer_target(void)
 553 {
 554         int i, cpu = smp_processor_id();
 555         struct sched_domain *sd;
 556 
 557         if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
 558                 return cpu;
 559 
 560         rcu_read_lock();
 561         for_each_domain(cpu, sd) {
 562                 for_each_cpu(i, sched_domain_span(sd)) {
 563                         if (cpu == i)
 564                                 continue;
 565 
 566                         if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
 567                                 cpu = i;
 568                                 goto unlock;
 569                         }
 570                 }
 571         }
 572 
 573         if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
 574                 cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
 575 unlock:
 576         rcu_read_unlock();
 577         return cpu;
 578 }
 579 
 580 /*
 581  * When add_timer_on() enqueues a timer into the timer wheel of an
 582  * idle CPU then this timer might expire before the next timer event
 583  * which is scheduled to wake up that CPU. In case of a completely
 584  * idle system the next event might even be infinite time into the
 585  * future. wake_up_idle_cpu() ensures that the CPU is woken up and
 586  * leaves the inner idle loop so the newly added timer is taken into
 587  * account when the CPU goes back to idle and evaluates the timer
 588  * wheel for the next timer event.
 589  */
 590 static void wake_up_idle_cpu(int cpu)
 591 {
 592         struct rq *rq = cpu_rq(cpu);
 593 
 594         if (cpu == smp_processor_id())
 595                 return;
 596 
 597         if (set_nr_and_not_polling(rq->idle))
 598                 smp_send_reschedule(cpu);
 599         else
 600                 trace_sched_wake_idle_without_ipi(cpu);
 601 }
 602 
 603 static bool wake_up_full_nohz_cpu(int cpu)
 604 {
 605         /*
 606          * We just need the target to call irq_exit() and re-evaluate
 607          * the next tick. The nohz full kick at least implies that.
 608          * If needed we can still optimize that later with an
 609          * empty IRQ.
 610          */
 611         if (cpu_is_offline(cpu))
 612                 return true;  /* Don't try to wake offline CPUs. */
 613         if (tick_nohz_full_cpu(cpu)) {
 614                 if (cpu != smp_processor_id() ||
 615                     tick_nohz_tick_stopped())
 616                         tick_nohz_full_kick_cpu(cpu);
 617                 return true;
 618         }
 619 
 620         return false;
 621 }
 622 
 623 /*
 624  * Wake up the specified CPU.  If the CPU is going offline, it is the
 625  * caller's responsibility to deal with the lost wakeup, for example,
 626  * by hooking into the CPU_DEAD notifier like timers and hrtimers do.
 627  */
 628 void wake_up_nohz_cpu(int cpu)
 629 {
 630         if (!wake_up_full_nohz_cpu(cpu))
 631                 wake_up_idle_cpu(cpu);
 632 }
 633 
 634 static inline bool got_nohz_idle_kick(void)
 635 {
 636         int cpu = smp_processor_id();
 637 
 638         if (!(atomic_read(nohz_flags(cpu)) & NOHZ_KICK_MASK))
 639                 return false;
 640 
 641         if (idle_cpu(cpu) && !need_resched())
 642                 return true;
 643 
 644         /*
 645          * We can't run Idle Load Balance on this CPU for this time so we
 646          * cancel it and clear NOHZ_BALANCE_KICK
 647          */
 648         atomic_andnot(NOHZ_KICK_MASK, nohz_flags(cpu));
 649         return false;
 650 }
 651 
 652 #else /* CONFIG_NO_HZ_COMMON */
 653 
 654 static inline bool got_nohz_idle_kick(void)
 655 {
 656         return false;
 657 }
 658 
 659 #endif /* CONFIG_NO_HZ_COMMON */
 660 
 661 #ifdef CONFIG_NO_HZ_FULL
 662 bool sched_can_stop_tick(struct rq *rq)
 663 {
 664         int fifo_nr_running;
 665 
 666         /* Deadline tasks, even if single, need the tick */
 667         if (rq->dl.dl_nr_running)
 668                 return false;
 669 
 670         /*
 671          * If there are more than one RR tasks, we need the tick to effect the
 672          * actual RR behaviour.
 673          */
 674         if (rq->rt.rr_nr_running) {
 675                 if (rq->rt.rr_nr_running == 1)
 676                         return true;
 677                 else
 678                         return false;
 679         }
 680 
 681         /*
 682          * If there's no RR tasks, but FIFO tasks, we can skip the tick, no
 683          * forced preemption between FIFO tasks.
 684          */
 685         fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running;
 686         if (fifo_nr_running)
 687                 return true;
 688 
 689         /*
 690          * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left;
 691          * if there's more than one we need the tick for involuntary
 692          * preemption.
 693          */
 694         if (rq->nr_running > 1)
 695                 return false;
 696 
 697         return true;
 698 }
 699 #endif /* CONFIG_NO_HZ_FULL */
 700 #endif /* CONFIG_SMP */
 701 
 702 #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
 703                         (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
 704 /*
 705  * Iterate task_group tree rooted at *from, calling @down when first entering a
 706  * node and @up when leaving it for the final time.
 707  *
 708  * Caller must hold rcu_lock or sufficient equivalent.
 709  */
 710 int walk_tg_tree_from(struct task_group *from,
 711                              tg_visitor down, tg_visitor up, void *data)
 712 {
 713         struct task_group *parent, *child;
 714         int ret;
 715 
 716         parent = from;
 717 
 718 down:
 719         ret = (*down)(parent, data);
 720         if (ret)
 721                 goto out;
 722         list_for_each_entry_rcu(child, &parent->children, siblings) {
 723                 parent = child;
 724                 goto down;
 725 
 726 up:
 727                 continue;
 728         }
 729         ret = (*up)(parent, data);
 730         if (ret || parent == from)
 731                 goto out;
 732 
 733         child = parent;
 734         parent = parent->parent;
 735         if (parent)
 736                 goto up;
 737 out:
 738         return ret;
 739 }
 740 
 741 int tg_nop(struct task_group *tg, void *data)
 742 {
 743         return 0;
 744 }
 745 #endif
 746 
 747 static void set_load_weight(struct task_struct *p, bool update_load)
 748 {
 749         int prio = p->static_prio - MAX_RT_PRIO;
 750         struct load_weight *load = &p->se.load;
 751 
 752         /*
 753          * SCHED_IDLE tasks get minimal weight:
 754          */
 755         if (task_has_idle_policy(p)) {
 756                 load->weight = scale_load(WEIGHT_IDLEPRIO);
 757                 load->inv_weight = WMULT_IDLEPRIO;
 758                 p->se.runnable_weight = load->weight;
 759                 return;
 760         }
 761 
 762         /*
 763          * SCHED_OTHER tasks have to update their load when changing their
 764          * weight
 765          */
 766         if (update_load && p->sched_class == &fair_sched_class) {
 767                 reweight_task(p, prio);
 768         } else {
 769                 load->weight = scale_load(sched_prio_to_weight[prio]);
 770                 load->inv_weight = sched_prio_to_wmult[prio];
 771                 p->se.runnable_weight = load->weight;
 772         }
 773 }
 774 
 775 #ifdef CONFIG_UCLAMP_TASK
 776 /*
 777  * Serializes updates of utilization clamp values
 778  *
 779  * The (slow-path) user-space triggers utilization clamp value updates which
 780  * can require updates on (fast-path) scheduler's data structures used to
 781  * support enqueue/dequeue operations.
 782  * While the per-CPU rq lock protects fast-path update operations, user-space
 783  * requests are serialized using a mutex to reduce the risk of conflicting
 784  * updates or API abuses.
 785  */
 786 static DEFINE_MUTEX(uclamp_mutex);
 787 
 788 /* Max allowed minimum utilization */
 789 unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE;
 790 
 791 /* Max allowed maximum utilization */
 792 unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE;
 793 
 794 /* All clamps are required to be less or equal than these values */
 795 static struct uclamp_se uclamp_default[UCLAMP_CNT];
 796 
 797 /* Integer rounded range for each bucket */
 798 #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
 799 
 800 #define for_each_clamp_id(clamp_id) \
 801         for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
 802 
 803 static inline unsigned int uclamp_bucket_id(unsigned int clamp_value)
 804 {
 805         return clamp_value / UCLAMP_BUCKET_DELTA;
 806 }
 807 
 808 static inline unsigned int uclamp_bucket_base_value(unsigned int clamp_value)
 809 {
 810         return UCLAMP_BUCKET_DELTA * uclamp_bucket_id(clamp_value);
 811 }
 812 
 813 static inline unsigned int uclamp_none(enum uclamp_id clamp_id)
 814 {
 815         if (clamp_id == UCLAMP_MIN)
 816                 return 0;
 817         return SCHED_CAPACITY_SCALE;
 818 }
 819 
 820 static inline void uclamp_se_set(struct uclamp_se *uc_se,
 821                                  unsigned int value, bool user_defined)
 822 {
 823         uc_se->value = value;
 824         uc_se->bucket_id = uclamp_bucket_id(value);
 825         uc_se->user_defined = user_defined;
 826 }
 827 
 828 static inline unsigned int
 829 uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id,
 830                   unsigned int clamp_value)
 831 {
 832         /*
 833          * Avoid blocked utilization pushing up the frequency when we go
 834          * idle (which drops the max-clamp) by retaining the last known
 835          * max-clamp.
 836          */
 837         if (clamp_id == UCLAMP_MAX) {
 838                 rq->uclamp_flags |= UCLAMP_FLAG_IDLE;
 839                 return clamp_value;
 840         }
 841 
 842         return uclamp_none(UCLAMP_MIN);
 843 }
 844 
 845 static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id,
 846                                      unsigned int clamp_value)
 847 {
 848         /* Reset max-clamp retention only on idle exit */
 849         if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE))
 850                 return;
 851 
 852         WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value);
 853 }
 854 
 855 static inline
 856 unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id,
 857                                    unsigned int clamp_value)
 858 {
 859         struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket;
 860         int bucket_id = UCLAMP_BUCKETS - 1;
 861 
 862         /*
 863          * Since both min and max clamps are max aggregated, find the
 864          * top most bucket with tasks in.
 865          */
 866         for ( ; bucket_id >= 0; bucket_id--) {
 867                 if (!bucket[bucket_id].tasks)
 868                         continue;
 869                 return bucket[bucket_id].value;
 870         }
 871 
 872         /* No tasks -- default clamp values */
 873         return uclamp_idle_value(rq, clamp_id, clamp_value);
 874 }
 875 
 876 static inline struct uclamp_se
 877 uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id)
 878 {
 879         struct uclamp_se uc_req = p->uclamp_req[clamp_id];
 880 #ifdef CONFIG_UCLAMP_TASK_GROUP
 881         struct uclamp_se uc_max;
 882 
 883         /*
 884          * Tasks in autogroups or root task group will be
 885          * restricted by system defaults.
 886          */
 887         if (task_group_is_autogroup(task_group(p)))
 888                 return uc_req;
 889         if (task_group(p) == &root_task_group)
 890                 return uc_req;
 891 
 892         uc_max = task_group(p)->uclamp[clamp_id];
 893         if (uc_req.value > uc_max.value || !uc_req.user_defined)
 894                 return uc_max;
 895 #endif
 896 
 897         return uc_req;
 898 }
 899 
 900 /*
 901  * The effective clamp bucket index of a task depends on, by increasing
 902  * priority:
 903  * - the task specific clamp value, when explicitly requested from userspace
 904  * - the task group effective clamp value, for tasks not either in the root
 905  *   group or in an autogroup
 906  * - the system default clamp value, defined by the sysadmin
 907  */
 908 static inline struct uclamp_se
 909 uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id)
 910 {
 911         struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id);
 912         struct uclamp_se uc_max = uclamp_default[clamp_id];
 913 
 914         /* System default restrictions always apply */
 915         if (unlikely(uc_req.value > uc_max.value))
 916                 return uc_max;
 917 
 918         return uc_req;
 919 }
 920 
 921 unsigned int uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id)
 922 {
 923         struct uclamp_se uc_eff;
 924 
 925         /* Task currently refcounted: use back-annotated (effective) value */
 926         if (p->uclamp[clamp_id].active)
 927                 return p->uclamp[clamp_id].value;
 928 
 929         uc_eff = uclamp_eff_get(p, clamp_id);
 930 
 931         return uc_eff.value;
 932 }
 933 
 934 /*
 935  * When a task is enqueued on a rq, the clamp bucket currently defined by the
 936  * task's uclamp::bucket_id is refcounted on that rq. This also immediately
 937  * updates the rq's clamp value if required.
 938  *
 939  * Tasks can have a task-specific value requested from user-space, track
 940  * within each bucket the maximum value for tasks refcounted in it.
 941  * This "local max aggregation" allows to track the exact "requested" value
 942  * for each bucket when all its RUNNABLE tasks require the same clamp.
 943  */
 944 static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p,
 945                                     enum uclamp_id clamp_id)
 946 {
 947         struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id];
 948         struct uclamp_se *uc_se = &p->uclamp[clamp_id];
 949         struct uclamp_bucket *bucket;
 950 
 951         lockdep_assert_held(&rq->lock);
 952 
 953         /* Update task effective clamp */
 954         p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id);
 955 
 956         bucket = &uc_rq->bucket[uc_se->bucket_id];
 957         bucket->tasks++;
 958         uc_se->active = true;
 959 
 960         uclamp_idle_reset(rq, clamp_id, uc_se->value);
 961 
 962         /*
 963          * Local max aggregation: rq buckets always track the max
 964          * "requested" clamp value of its RUNNABLE tasks.
 965          */
 966         if (bucket->tasks == 1 || uc_se->value > bucket->value)
 967                 bucket->value = uc_se->value;
 968 
 969         if (uc_se->value > READ_ONCE(uc_rq->value))
 970                 WRITE_ONCE(uc_rq->value, uc_se->value);
 971 }
 972 
 973 /*
 974  * When a task is dequeued from a rq, the clamp bucket refcounted by the task
 975  * is released. If this is the last task reference counting the rq's max
 976  * active clamp value, then the rq's clamp value is updated.
 977  *
 978  * Both refcounted tasks and rq's cached clamp values are expected to be
 979  * always valid. If it's detected they are not, as defensive programming,
 980  * enforce the expected state and warn.
 981  */
 982 static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p,
 983                                     enum uclamp_id clamp_id)
 984 {
 985         struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id];
 986         struct uclamp_se *uc_se = &p->uclamp[clamp_id];
 987         struct uclamp_bucket *bucket;
 988         unsigned int bkt_clamp;
 989         unsigned int rq_clamp;
 990 
 991         lockdep_assert_held(&rq->lock);
 992 
 993         bucket = &uc_rq->bucket[uc_se->bucket_id];
 994         SCHED_WARN_ON(!bucket->tasks);
 995         if (likely(bucket->tasks))
 996                 bucket->tasks--;
 997         uc_se->active = false;
 998 
 999         /*
1000          * Keep "local max aggregation" simple and accept to (possibly)
1001          * overboost some RUNNABLE tasks in the same bucket.
1002          * The rq clamp bucket value is reset to its base value whenever
1003          * there are no more RUNNABLE tasks refcounting it.
1004          */
1005         if (likely(bucket->tasks))
1006                 return;
1007 
1008         rq_clamp = READ_ONCE(uc_rq->value);
1009         /*
1010          * Defensive programming: this should never happen. If it happens,
1011          * e.g. due to future modification, warn and fixup the expected value.
1012          */
1013         SCHED_WARN_ON(bucket->value > rq_clamp);
1014         if (bucket->value >= rq_clamp) {
1015                 bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value);
1016                 WRITE_ONCE(uc_rq->value, bkt_clamp);
1017         }
1018 }
1019 
1020 static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p)
1021 {
1022         enum uclamp_id clamp_id;
1023 
1024         if (unlikely(!p->sched_class->uclamp_enabled))
1025                 return;
1026 
1027         for_each_clamp_id(clamp_id)
1028                 uclamp_rq_inc_id(rq, p, clamp_id);
1029 
1030         /* Reset clamp idle holding when there is one RUNNABLE task */
1031         if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
1032                 rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE;
1033 }
1034 
1035 static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p)
1036 {
1037         enum uclamp_id clamp_id;
1038 
1039         if (unlikely(!p->sched_class->uclamp_enabled))
1040                 return;
1041 
1042         for_each_clamp_id(clamp_id)
1043                 uclamp_rq_dec_id(rq, p, clamp_id);
1044 }
1045 
1046 static inline void
1047 uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id)
1048 {
1049         struct rq_flags rf;
1050         struct rq *rq;
1051 
1052         /*
1053          * Lock the task and the rq where the task is (or was) queued.
1054          *
1055          * We might lock the (previous) rq of a !RUNNABLE task, but that's the
1056          * price to pay to safely serialize util_{min,max} updates with
1057          * enqueues, dequeues and migration operations.
1058          * This is the same locking schema used by __set_cpus_allowed_ptr().
1059          */
1060         rq = task_rq_lock(p, &rf);
1061 
1062         /*
1063          * Setting the clamp bucket is serialized by task_rq_lock().
1064          * If the task is not yet RUNNABLE and its task_struct is not
1065          * affecting a valid clamp bucket, the next time it's enqueued,
1066          * it will already see the updated clamp bucket value.
1067          */
1068         if (p->uclamp[clamp_id].active) {
1069                 uclamp_rq_dec_id(rq, p, clamp_id);
1070                 uclamp_rq_inc_id(rq, p, clamp_id);
1071         }
1072 
1073         task_rq_unlock(rq, p, &rf);
1074 }
1075 
1076 #ifdef CONFIG_UCLAMP_TASK_GROUP
1077 static inline void
1078 uclamp_update_active_tasks(struct cgroup_subsys_state *css,
1079                            unsigned int clamps)
1080 {
1081         enum uclamp_id clamp_id;
1082         struct css_task_iter it;
1083         struct task_struct *p;
1084 
1085         css_task_iter_start(css, 0, &it);
1086         while ((p = css_task_iter_next(&it))) {
1087                 for_each_clamp_id(clamp_id) {
1088                         if ((0x1 << clamp_id) & clamps)
1089                                 uclamp_update_active(p, clamp_id);
1090                 }
1091         }
1092         css_task_iter_end(&it);
1093 }
1094 
1095 static void cpu_util_update_eff(struct cgroup_subsys_state *css);
1096 static void uclamp_update_root_tg(void)
1097 {
1098         struct task_group *tg = &root_task_group;
1099 
1100         uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN],
1101                       sysctl_sched_uclamp_util_min, false);
1102         uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX],
1103                       sysctl_sched_uclamp_util_max, false);
1104 
1105         rcu_read_lock();
1106         cpu_util_update_eff(&root_task_group.css);
1107         rcu_read_unlock();
1108 }
1109 #else
1110 static void uclamp_update_root_tg(void) { }
1111 #endif
1112 
1113 int sysctl_sched_uclamp_handler(struct ctl_table *table, int write,
1114                                 void __user *buffer, size_t *lenp,
1115                                 loff_t *ppos)
1116 {
1117         bool update_root_tg = false;
1118         int old_min, old_max;
1119         int result;
1120 
1121         mutex_lock(&uclamp_mutex);
1122         old_min = sysctl_sched_uclamp_util_min;
1123         old_max = sysctl_sched_uclamp_util_max;
1124 
1125         result = proc_dointvec(table, write, buffer, lenp, ppos);
1126         if (result)
1127                 goto undo;
1128         if (!write)
1129                 goto done;
1130 
1131         if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max ||
1132             sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE) {
1133                 result = -EINVAL;
1134                 goto undo;
1135         }
1136 
1137         if (old_min != sysctl_sched_uclamp_util_min) {
1138                 uclamp_se_set(&uclamp_default[UCLAMP_MIN],
1139                               sysctl_sched_uclamp_util_min, false);
1140                 update_root_tg = true;
1141         }
1142         if (old_max != sysctl_sched_uclamp_util_max) {
1143                 uclamp_se_set(&uclamp_default[UCLAMP_MAX],
1144                               sysctl_sched_uclamp_util_max, false);
1145                 update_root_tg = true;
1146         }
1147 
1148         if (update_root_tg)
1149                 uclamp_update_root_tg();
1150 
1151         /*
1152          * We update all RUNNABLE tasks only when task groups are in use.
1153          * Otherwise, keep it simple and do just a lazy update at each next
1154          * task enqueue time.
1155          */
1156 
1157         goto done;
1158 
1159 undo:
1160         sysctl_sched_uclamp_util_min = old_min;
1161         sysctl_sched_uclamp_util_max = old_max;
1162 done:
1163         mutex_unlock(&uclamp_mutex);
1164 
1165         return result;
1166 }
1167 
1168 static int uclamp_validate(struct task_struct *p,
1169                            const struct sched_attr *attr)
1170 {
1171         unsigned int lower_bound = p->uclamp_req[UCLAMP_MIN].value;
1172         unsigned int upper_bound = p->uclamp_req[UCLAMP_MAX].value;
1173 
1174         if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN)
1175                 lower_bound = attr->sched_util_min;
1176         if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX)
1177                 upper_bound = attr->sched_util_max;
1178 
1179         if (lower_bound > upper_bound)
1180                 return -EINVAL;
1181         if (upper_bound > SCHED_CAPACITY_SCALE)
1182                 return -EINVAL;
1183 
1184         return 0;
1185 }
1186 
1187 static void __setscheduler_uclamp(struct task_struct *p,
1188                                   const struct sched_attr *attr)
1189 {
1190         enum uclamp_id clamp_id;
1191 
1192         /*
1193          * On scheduling class change, reset to default clamps for tasks
1194          * without a task-specific value.
1195          */
1196         for_each_clamp_id(clamp_id) {
1197                 struct uclamp_se *uc_se = &p->uclamp_req[clamp_id];
1198                 unsigned int clamp_value = uclamp_none(clamp_id);
1199 
1200                 /* Keep using defined clamps across class changes */
1201                 if (uc_se->user_defined)
1202                         continue;
1203 
1204                 /* By default, RT tasks always get 100% boost */
1205                 if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
1206                         clamp_value = uclamp_none(UCLAMP_MAX);
1207 
1208                 uclamp_se_set(uc_se, clamp_value, false);
1209         }
1210 
1211         if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)))
1212                 return;
1213 
1214         if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) {
1215                 uclamp_se_set(&p->uclamp_req[UCLAMP_MIN],
1216                               attr->sched_util_min, true);
1217         }
1218 
1219         if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) {
1220                 uclamp_se_set(&p->uclamp_req[UCLAMP_MAX],
1221                               attr->sched_util_max, true);
1222         }
1223 }
1224 
1225 static void uclamp_fork(struct task_struct *p)
1226 {
1227         enum uclamp_id clamp_id;
1228 
1229         for_each_clamp_id(clamp_id)
1230                 p->uclamp[clamp_id].active = false;
1231 
1232         if (likely(!p->sched_reset_on_fork))
1233                 return;
1234 
1235         for_each_clamp_id(clamp_id) {
1236                 uclamp_se_set(&p->uclamp_req[clamp_id],
1237                               uclamp_none(clamp_id), false);
1238         }
1239 }
1240 
1241 static void __init init_uclamp(void)
1242 {
1243         struct uclamp_se uc_max = {};
1244         enum uclamp_id clamp_id;
1245         int cpu;
1246 
1247         mutex_init(&uclamp_mutex);
1248 
1249         for_each_possible_cpu(cpu) {
1250                 memset(&cpu_rq(cpu)->uclamp, 0,
1251                                 sizeof(struct uclamp_rq)*UCLAMP_CNT);
1252                 cpu_rq(cpu)->uclamp_flags = 0;
1253         }
1254 
1255         for_each_clamp_id(clamp_id) {
1256                 uclamp_se_set(&init_task.uclamp_req[clamp_id],
1257                               uclamp_none(clamp_id), false);
1258         }
1259 
1260         /* System defaults allow max clamp values for both indexes */
1261         uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false);
1262         for_each_clamp_id(clamp_id) {
1263                 uclamp_default[clamp_id] = uc_max;
1264 #ifdef CONFIG_UCLAMP_TASK_GROUP
1265                 root_task_group.uclamp_req[clamp_id] = uc_max;
1266                 root_task_group.uclamp[clamp_id] = uc_max;
1267 #endif
1268         }
1269 }
1270 
1271 #else /* CONFIG_UCLAMP_TASK */
1272 static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { }
1273 static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { }
1274 static inline int uclamp_validate(struct task_struct *p,
1275                                   const struct sched_attr *attr)
1276 {
1277         return -EOPNOTSUPP;
1278 }
1279 static void __setscheduler_uclamp(struct task_struct *p,
1280                                   const struct sched_attr *attr) { }
1281 static inline void uclamp_fork(struct task_struct *p) { }
1282 static inline void init_uclamp(void) { }
1283 #endif /* CONFIG_UCLAMP_TASK */
1284 
1285 static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
1286 {
1287         if (!(flags & ENQUEUE_NOCLOCK))
1288                 update_rq_clock(rq);
1289 
1290         if (!(flags & ENQUEUE_RESTORE)) {
1291                 sched_info_queued(rq, p);
1292                 psi_enqueue(p, flags & ENQUEUE_WAKEUP);
1293         }
1294 
1295         uclamp_rq_inc(rq, p);
1296         p->sched_class->enqueue_task(rq, p, flags);
1297 }
1298 
1299 static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
1300 {
1301         if (!(flags & DEQUEUE_NOCLOCK))
1302                 update_rq_clock(rq);
1303 
1304         if (!(flags & DEQUEUE_SAVE)) {
1305                 sched_info_dequeued(rq, p);
1306                 psi_dequeue(p, flags & DEQUEUE_SLEEP);
1307         }
1308 
1309         uclamp_rq_dec(rq, p);
1310         p->sched_class->dequeue_task(rq, p, flags);
1311 }
1312 
1313 void activate_task(struct rq *rq, struct task_struct *p, int flags)
1314 {
1315         if (task_contributes_to_load(p))
1316                 rq->nr_uninterruptible--;
1317 
1318         enqueue_task(rq, p, flags);
1319 
1320         p->on_rq = TASK_ON_RQ_QUEUED;
1321 }
1322 
1323 void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
1324 {
1325         p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING;
1326 
1327         if (task_contributes_to_load(p))
1328                 rq->nr_uninterruptible++;
1329 
1330         dequeue_task(rq, p, flags);
1331 }
1332 
1333 /*
1334  * __normal_prio - return the priority that is based on the static prio
1335  */
1336 static inline int __normal_prio(struct task_struct *p)
1337 {
1338         return p->static_prio;
1339 }
1340 
1341 /*
1342  * Calculate the expected normal priority: i.e. priority
1343  * without taking RT-inheritance into account. Might be
1344  * boosted by interactivity modifiers. Changes upon fork,
1345  * setprio syscalls, and whenever the interactivity
1346  * estimator recalculates.
1347  */
1348 static inline int normal_prio(struct task_struct *p)
1349 {
1350         int prio;
1351 
1352         if (task_has_dl_policy(p))
1353                 prio = MAX_DL_PRIO-1;
1354         else if (task_has_rt_policy(p))
1355                 prio = MAX_RT_PRIO-1 - p->rt_priority;
1356         else
1357                 prio = __normal_prio(p);
1358         return prio;
1359 }
1360 
1361 /*
1362  * Calculate the current priority, i.e. the priority
1363  * taken into account by the scheduler. This value might
1364  * be boosted by RT tasks, or might be boosted by
1365  * interactivity modifiers. Will be RT if the task got
1366  * RT-boosted. If not then it returns p->normal_prio.
1367  */
1368 static int effective_prio(struct task_struct *p)
1369 {
1370         p->normal_prio = normal_prio(p);
1371         /*
1372          * If we are RT tasks or we were boosted to RT priority,
1373          * keep the priority unchanged. Otherwise, update priority
1374          * to the normal priority:
1375          */
1376         if (!rt_prio(p->prio))
1377                 return p->normal_prio;
1378         return p->prio;
1379 }
1380 
1381 /**
1382  * task_curr - is this task currently executing on a CPU?
1383  * @p: the task in question.
1384  *
1385  * Return: 1 if the task is currently executing. 0 otherwise.
1386  */
1387 inline int task_curr(const struct task_struct *p)
1388 {
1389         return cpu_curr(task_cpu(p)) == p;
1390 }
1391 
1392 /*
1393  * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock,
1394  * use the balance_callback list if you want balancing.
1395  *
1396  * this means any call to check_class_changed() must be followed by a call to
1397  * balance_callback().
1398  */
1399 static inline void check_class_changed(struct rq *rq, struct task_struct *p,
1400                                        const struct sched_class *prev_class,
1401                                        int oldprio)
1402 {
1403         if (prev_class != p->sched_class) {
1404                 if (prev_class->switched_from)
1405                         prev_class->switched_from(rq, p);
1406 
1407                 p->sched_class->switched_to(rq, p);
1408         } else if (oldprio != p->prio || dl_task(p))
1409                 p->sched_class->prio_changed(rq, p, oldprio);
1410 }
1411 
1412 void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
1413 {
1414         const struct sched_class *class;
1415 
1416         if (p->sched_class == rq->curr->sched_class) {
1417                 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
1418         } else {
1419                 for_each_class(class) {
1420                         if (class == rq->curr->sched_class)
1421                                 break;
1422                         if (class == p->sched_class) {
1423                                 resched_curr(rq);
1424                                 break;
1425                         }
1426                 }
1427         }
1428 
1429         /*
1430          * A queue event has occurred, and we're going to schedule.  In
1431          * this case, we can save a useless back to back clock update.
1432          */
1433         if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
1434                 rq_clock_skip_update(rq);
1435 }
1436 
1437 #ifdef CONFIG_SMP
1438 
1439 static inline bool is_per_cpu_kthread(struct task_struct *p)
1440 {
1441         if (!(p->flags & PF_KTHREAD))
1442                 return false;
1443 
1444         if (p->nr_cpus_allowed != 1)
1445                 return false;
1446 
1447         return true;
1448 }
1449 
1450 /*
1451  * Per-CPU kthreads are allowed to run on !active && online CPUs, see
1452  * __set_cpus_allowed_ptr() and select_fallback_rq().
1453  */
1454 static inline bool is_cpu_allowed(struct task_struct *p, int cpu)
1455 {
1456         if (!cpumask_test_cpu(cpu, p->cpus_ptr))
1457                 return false;
1458 
1459         if (is_per_cpu_kthread(p))
1460                 return cpu_online(cpu);
1461 
1462         return cpu_active(cpu);
1463 }
1464 
1465 /*
1466  * This is how migration works:
1467  *
1468  * 1) we invoke migration_cpu_stop() on the target CPU using
1469  *    stop_one_cpu().
1470  * 2) stopper starts to run (implicitly forcing the migrated thread
1471  *    off the CPU)
1472  * 3) it checks whether the migrated task is still in the wrong runqueue.
1473  * 4) if it's in the wrong runqueue then the migration thread removes
1474  *    it and puts it into the right queue.
1475  * 5) stopper completes and stop_one_cpu() returns and the migration
1476  *    is done.
1477  */
1478 
1479 /*
1480  * move_queued_task - move a queued task to new rq.
1481  *
1482  * Returns (locked) new rq. Old rq's lock is released.
1483  */
1484 static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf,
1485                                    struct task_struct *p, int new_cpu)
1486 {
1487         lockdep_assert_held(&rq->lock);
1488 
1489         WRITE_ONCE(p->on_rq, TASK_ON_RQ_MIGRATING);
1490         dequeue_task(rq, p, DEQUEUE_NOCLOCK);
1491         set_task_cpu(p, new_cpu);
1492         rq_unlock(rq, rf);
1493 
1494         rq = cpu_rq(new_cpu);
1495 
1496         rq_lock(rq, rf);
1497         BUG_ON(task_cpu(p) != new_cpu);
1498         enqueue_task(rq, p, 0);
1499         p->on_rq = TASK_ON_RQ_QUEUED;
1500         check_preempt_curr(rq, p, 0);
1501 
1502         return rq;
1503 }
1504 
1505 struct migration_arg {
1506         struct task_struct *task;
1507         int dest_cpu;
1508 };
1509 
1510 /*
1511  * Move (not current) task off this CPU, onto the destination CPU. We're doing
1512  * this because either it can't run here any more (set_cpus_allowed()
1513  * away from this CPU, or CPU going down), or because we're
1514  * attempting to rebalance this task on exec (sched_exec).
1515  *
1516  * So we race with normal scheduler movements, but that's OK, as long
1517  * as the task is no longer on this CPU.
1518  */
1519 static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
1520                                  struct task_struct *p, int dest_cpu)
1521 {
1522         /* Affinity changed (again). */
1523         if (!is_cpu_allowed(p, dest_cpu))
1524                 return rq;
1525 
1526         update_rq_clock(rq);
1527         rq = move_queued_task(rq, rf, p, dest_cpu);
1528 
1529         return rq;
1530 }
1531 
1532 /*
1533  * migration_cpu_stop - this will be executed by a highprio stopper thread
1534  * and performs thread migration by bumping thread off CPU then
1535  * 'pushing' onto another runqueue.
1536  */
1537 static int migration_cpu_stop(void *data)
1538 {
1539         struct migration_arg *arg = data;
1540         struct task_struct *p = arg->task;
1541         struct rq *rq = this_rq();
1542         struct rq_flags rf;
1543 
1544         /*
1545          * The original target CPU might have gone down and we might
1546          * be on another CPU but it doesn't matter.
1547          */
1548         local_irq_disable();
1549         /*
1550          * We need to explicitly wake pending tasks before running
1551          * __migrate_task() such that we will not miss enforcing cpus_ptr
1552          * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test.
1553          */
1554         sched_ttwu_pending();
1555 
1556         raw_spin_lock(&p->pi_lock);
1557         rq_lock(rq, &rf);
1558         /*
1559          * If task_rq(p) != rq, it cannot be migrated here, because we're
1560          * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because
1561          * we're holding p->pi_lock.
1562          */
1563         if (task_rq(p) == rq) {
1564                 if (task_on_rq_queued(p))
1565                         rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
1566                 else
1567                         p->wake_cpu = arg->dest_cpu;
1568         }
1569         rq_unlock(rq, &rf);
1570         raw_spin_unlock(&p->pi_lock);
1571 
1572         local_irq_enable();
1573         return 0;
1574 }
1575 
1576 /*
1577  * sched_class::set_cpus_allowed must do the below, but is not required to
1578  * actually call this function.
1579  */
1580 void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask)
1581 {
1582         cpumask_copy(&p->cpus_mask, new_mask);
1583         p->nr_cpus_allowed = cpumask_weight(new_mask);
1584 }
1585 
1586 void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1587 {
1588         struct rq *rq = task_rq(p);
1589         bool queued, running;
1590 
1591         lockdep_assert_held(&p->pi_lock);
1592 
1593         queued = task_on_rq_queued(p);
1594         running = task_current(rq, p);
1595 
1596         if (queued) {
1597                 /*
1598                  * Because __kthread_bind() calls this on blocked tasks without
1599                  * holding rq->lock.
1600                  */
1601                 lockdep_assert_held(&rq->lock);
1602                 dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
1603         }
1604         if (running)
1605                 put_prev_task(rq, p);
1606 
1607         p->sched_class->set_cpus_allowed(p, new_mask);
1608 
1609         if (queued)
1610                 enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
1611         if (running)
1612                 set_next_task(rq, p);
1613 }
1614 
1615 /*
1616  * Change a given task's CPU affinity. Migrate the thread to a
1617  * proper CPU and schedule it away if the CPU it's executing on
1618  * is removed from the allowed bitmask.
1619  *
1620  * NOTE: the caller must have a valid reference to the task, the
1621  * task must not exit() & deallocate itself prematurely. The
1622  * call is not atomic; no spinlocks may be held.
1623  */
1624 static int __set_cpus_allowed_ptr(struct task_struct *p,
1625                                   const struct cpumask *new_mask, bool check)
1626 {
1627         const struct cpumask *cpu_valid_mask = cpu_active_mask;
1628         unsigned int dest_cpu;
1629         struct rq_flags rf;
1630         struct rq *rq;
1631         int ret = 0;
1632 
1633         rq = task_rq_lock(p, &rf);
1634         update_rq_clock(rq);
1635 
1636         if (p->flags & PF_KTHREAD) {
1637                 /*
1638                  * Kernel threads are allowed on online && !active CPUs
1639                  */
1640                 cpu_valid_mask = cpu_online_mask;
1641         }
1642 
1643         /*
1644          * Must re-check here, to close a race against __kthread_bind(),
1645          * sched_setaffinity() is not guaranteed to observe the flag.
1646          */
1647         if (check && (p->flags & PF_NO_SETAFFINITY)) {
1648                 ret = -EINVAL;
1649                 goto out;
1650         }
1651 
1652         if (cpumask_equal(p->cpus_ptr, new_mask))
1653                 goto out;
1654 
1655         dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
1656         if (dest_cpu >= nr_cpu_ids) {
1657                 ret = -EINVAL;
1658                 goto out;
1659         }
1660 
1661         do_set_cpus_allowed(p, new_mask);
1662 
1663         if (p->flags & PF_KTHREAD) {
1664                 /*
1665                  * For kernel threads that do indeed end up on online &&
1666                  * !active we want to ensure they are strict per-CPU threads.
1667                  */
1668                 WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) &&
1669                         !cpumask_intersects(new_mask, cpu_active_mask) &&
1670                         p->nr_cpus_allowed != 1);
1671         }
1672 
1673         /* Can the task run on the task's current CPU? If so, we're done */
1674         if (cpumask_test_cpu(task_cpu(p), new_mask))
1675                 goto out;
1676 
1677         if (task_running(rq, p) || p->state == TASK_WAKING) {
1678                 struct migration_arg arg = { p, dest_cpu };
1679                 /* Need help from migration thread: drop lock and wait. */
1680                 task_rq_unlock(rq, p, &rf);
1681                 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1682                 return 0;
1683         } else if (task_on_rq_queued(p)) {
1684                 /*
1685                  * OK, since we're going to drop the lock immediately
1686                  * afterwards anyway.
1687                  */
1688                 rq = move_queued_task(rq, &rf, p, dest_cpu);
1689         }
1690 out:
1691         task_rq_unlock(rq, p, &rf);
1692 
1693         return ret;
1694 }
1695 
1696 int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1697 {
1698         return __set_cpus_allowed_ptr(p, new_mask, false);
1699 }
1700 EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
1701 
1702 void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
1703 {
1704 #ifdef CONFIG_SCHED_DEBUG
1705         /*
1706          * We should never call set_task_cpu() on a blocked task,
1707          * ttwu() will sort out the placement.
1708          */
1709         WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
1710                         !p->on_rq);
1711 
1712         /*
1713          * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING,
1714          * because schedstat_wait_{start,end} rebase migrating task's wait_start
1715          * time relying on p->on_rq.
1716          */
1717         WARN_ON_ONCE(p->state == TASK_RUNNING &&
1718                      p->sched_class == &fair_sched_class &&
1719                      (p->on_rq && !task_on_rq_migrating(p)));
1720 
1721 #ifdef CONFIG_LOCKDEP
1722         /*
1723          * The caller should hold either p->pi_lock or rq->lock, when changing
1724          * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks.
1725          *
1726          * sched_move_task() holds both and thus holding either pins the cgroup,
1727          * see task_group().
1728          *
1729          * Furthermore, all task_rq users should acquire both locks, see
1730          * task_rq_lock().
1731          */
1732         WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
1733                                       lockdep_is_held(&task_rq(p)->lock)));
1734 #endif
1735         /*
1736          * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
1737          */
1738         WARN_ON_ONCE(!cpu_online(new_cpu));
1739 #endif
1740 
1741         trace_sched_migrate_task(p, new_cpu);
1742 
1743         if (task_cpu(p) != new_cpu) {
1744                 if (p->sched_class->migrate_task_rq)
1745                         p->sched_class->migrate_task_rq(p, new_cpu);
1746                 p->se.nr_migrations++;
1747                 rseq_migrate(p);
1748                 perf_event_task_migrate(p);
1749         }
1750 
1751         __set_task_cpu(p, new_cpu);
1752 }
1753 
1754 #ifdef CONFIG_NUMA_BALANCING
1755 static void __migrate_swap_task(struct task_struct *p, int cpu)
1756 {
1757         if (task_on_rq_queued(p)) {
1758                 struct rq *src_rq, *dst_rq;
1759                 struct rq_flags srf, drf;
1760 
1761                 src_rq = task_rq(p);
1762                 dst_rq = cpu_rq(cpu);
1763 
1764                 rq_pin_lock(src_rq, &srf);
1765                 rq_pin_lock(dst_rq, &drf);
1766 
1767                 deactivate_task(src_rq, p, 0);
1768                 set_task_cpu(p, cpu);
1769                 activate_task(dst_rq, p, 0);
1770                 check_preempt_curr(dst_rq, p, 0);
1771 
1772                 rq_unpin_lock(dst_rq, &drf);
1773                 rq_unpin_lock(src_rq, &srf);
1774 
1775         } else {
1776                 /*
1777                  * Task isn't running anymore; make it appear like we migrated
1778                  * it before it went to sleep. This means on wakeup we make the
1779                  * previous CPU our target instead of where it really is.
1780                  */
1781                 p->wake_cpu = cpu;
1782         }
1783 }
1784 
1785 struct migration_swap_arg {
1786         struct task_struct *src_task, *dst_task;
1787         int src_cpu, dst_cpu;
1788 };
1789 
1790 static int migrate_swap_stop(void *data)
1791 {
1792         struct migration_swap_arg *arg = data;
1793         struct rq *src_rq, *dst_rq;
1794         int ret = -EAGAIN;
1795 
1796         if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
1797                 return -EAGAIN;
1798 
1799         src_rq = cpu_rq(arg->src_cpu);
1800         dst_rq = cpu_rq(arg->dst_cpu);
1801 
1802         double_raw_lock(&arg->src_task->pi_lock,
1803                         &arg->dst_task->pi_lock);
1804         double_rq_lock(src_rq, dst_rq);
1805 
1806         if (task_cpu(arg->dst_task) != arg->dst_cpu)
1807                 goto unlock;
1808 
1809         if (task_cpu(arg->src_task) != arg->src_cpu)
1810                 goto unlock;
1811 
1812         if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr))
1813                 goto unlock;
1814 
1815         if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr))
1816                 goto unlock;
1817 
1818         __migrate_swap_task(arg->src_task, arg->dst_cpu);
1819         __migrate_swap_task(arg->dst_task, arg->src_cpu);
1820 
1821         ret = 0;
1822 
1823 unlock:
1824         double_rq_unlock(src_rq, dst_rq);
1825         raw_spin_unlock(&arg->dst_task->pi_lock);
1826         raw_spin_unlock(&arg->src_task->pi_lock);
1827 
1828         return ret;
1829 }
1830 
1831 /*
1832  * Cross migrate two tasks
1833  */
1834 int migrate_swap(struct task_struct *cur, struct task_struct *p,
1835                 int target_cpu, int curr_cpu)
1836 {
1837         struct migration_swap_arg arg;
1838         int ret = -EINVAL;
1839 
1840         arg = (struct migration_swap_arg){
1841                 .src_task = cur,
1842                 .src_cpu = curr_cpu,
1843                 .dst_task = p,
1844                 .dst_cpu = target_cpu,
1845         };
1846 
1847         if (arg.src_cpu == arg.dst_cpu)
1848                 goto out;
1849 
1850         /*
1851          * These three tests are all lockless; this is OK since all of them
1852          * will be re-checked with proper locks held further down the line.
1853          */
1854         if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
1855                 goto out;
1856 
1857         if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr))
1858                 goto out;
1859 
1860         if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr))
1861                 goto out;
1862 
1863         trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1864         ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);
1865 
1866 out:
1867         return ret;
1868 }
1869 #endif /* CONFIG_NUMA_BALANCING */
1870 
1871 /*
1872  * wait_task_inactive - wait for a thread to unschedule.
1873  *
1874  * If @match_state is nonzero, it's the @p->state value just checked and
1875  * not expected to change.  If it changes, i.e. @p might have woken up,
1876  * then return zero.  When we succeed in waiting for @p to be off its CPU,
1877  * we return a positive number (its total switch count).  If a second call
1878  * a short while later returns the same number, the caller can be sure that
1879  * @p has remained unscheduled the whole time.
1880  *
1881  * The caller must ensure that the task *will* unschedule sometime soon,
1882  * else this function might spin for a *long* time. This function can't
1883  * be called with interrupts off, or it may introduce deadlock with
1884  * smp_call_function() if an IPI is sent by the same process we are
1885  * waiting to become inactive.
1886  */
1887 unsigned long wait_task_inactive(struct task_struct *p, long match_state)
1888 {
1889         int running, queued;
1890         struct rq_flags rf;
1891         unsigned long ncsw;
1892         struct rq *rq;
1893 
1894         for (;;) {
1895                 /*
1896                  * We do the initial early heuristics without holding
1897                  * any task-queue locks at all. We'll only try to get
1898                  * the runqueue lock when things look like they will
1899                  * work out!
1900                  */
1901                 rq = task_rq(p);
1902 
1903                 /*
1904                  * If the task is actively running on another CPU
1905                  * still, just relax and busy-wait without holding
1906                  * any locks.
1907                  *
1908                  * NOTE! Since we don't hold any locks, it's not
1909                  * even sure that "rq" stays as the right runqueue!
1910                  * But we don't care, since "task_running()" will
1911                  * return false if the runqueue has changed and p
1912                  * is actually now running somewhere else!
1913                  */
1914                 while (task_running(rq, p)) {
1915                         if (match_state && unlikely(p->state != match_state))
1916                                 return 0;
1917                         cpu_relax();
1918                 }
1919 
1920                 /*
1921                  * Ok, time to look more closely! We need the rq
1922                  * lock now, to be *sure*. If we're wrong, we'll
1923                  * just go back and repeat.
1924                  */
1925                 rq = task_rq_lock(p, &rf);
1926                 trace_sched_wait_task(p);
1927                 running = task_running(rq, p);
1928                 queued = task_on_rq_queued(p);
1929                 ncsw = 0;
1930                 if (!match_state || p->state == match_state)
1931                         ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1932                 task_rq_unlock(rq, p, &rf);
1933 
1934                 /*
1935                  * If it changed from the expected state, bail out now.
1936                  */
1937                 if (unlikely(!ncsw))
1938                         break;
1939 
1940                 /*
1941                  * Was it really running after all now that we
1942                  * checked with the proper locks actually held?
1943                  *
1944                  * Oops. Go back and try again..
1945                  */
1946                 if (unlikely(running)) {
1947                         cpu_relax();
1948                         continue;
1949                 }
1950 
1951                 /*
1952                  * It's not enough that it's not actively running,
1953                  * it must be off the runqueue _entirely_, and not
1954                  * preempted!
1955                  *
1956                  * So if it was still runnable (but just not actively
1957                  * running right now), it's preempted, and we should
1958                  * yield - it could be a while.
1959                  */
1960                 if (unlikely(queued)) {
1961                         ktime_t to = NSEC_PER_SEC / HZ;
1962 
1963                         set_current_state(TASK_UNINTERRUPTIBLE);
1964                         schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1965                         continue;
1966                 }
1967 
1968                 /*
1969                  * Ahh, all good. It wasn't running, and it wasn't
1970                  * runnable, which means that it will never become
1971                  * running in the future either. We're all done!
1972                  */
1973                 break;
1974         }
1975 
1976         return ncsw;
1977 }
1978 
1979 /***
1980  * kick_process - kick a running thread to enter/exit the kernel
1981  * @p: the to-be-kicked thread
1982  *
1983  * Cause a process which is running on another CPU to enter
1984  * kernel-mode, without any delay. (to get signals handled.)
1985  *
1986  * NOTE: this function doesn't have to take the runqueue lock,
1987  * because all it wants to ensure is that the remote task enters
1988  * the kernel. If the IPI races and the task has been migrated
1989  * to another CPU then no harm is done and the purpose has been
1990  * achieved as well.
1991  */
1992 void kick_process(struct task_struct *p)
1993 {
1994         int cpu;
1995 
1996         preempt_disable();
1997         cpu = task_cpu(p);
1998         if ((cpu != smp_processor_id()) && task_curr(p))
1999                 smp_send_reschedule(cpu);
2000         preempt_enable();
2001 }
2002 EXPORT_SYMBOL_GPL(kick_process);
2003 
2004 /*
2005  * ->cpus_ptr is protected by both rq->lock and p->pi_lock
2006  *
2007  * A few notes on cpu_active vs cpu_online:
2008  *
2009  *  - cpu_active must be a subset of cpu_online
2010  *
2011  *  - on CPU-up we allow per-CPU kthreads on the online && !active CPU,
2012  *    see __set_cpus_allowed_ptr(). At this point the newly online
2013  *    CPU isn't yet part of the sched domains, and balancing will not
2014  *    see it.
2015  *
2016  *  - on CPU-down we clear cpu_active() to mask the sched domains and
2017  *    avoid the load balancer to place new tasks on the to be removed
2018  *    CPU. Existing tasks will remain running there and will be taken
2019  *    off.
2020  *
2021  * This means that fallback selection must not select !active CPUs.
2022  * And can assume that any active CPU must be online. Conversely
2023  * select_task_rq() below may allow selection of !active CPUs in order
2024  * to satisfy the above rules.
2025  */
2026 static int select_fallback_rq(int cpu, struct task_struct *p)
2027 {
2028         int nid = cpu_to_node(cpu);
2029         const struct cpumask *nodemask = NULL;
2030         enum { cpuset, possible, fail } state = cpuset;
2031         int dest_cpu;
2032 
2033         /*
2034          * If the node that the CPU is on has been offlined, cpu_to_node()
2035          * will return -1. There is no CPU on the node, and we should
2036          * select the CPU on the other node.
2037          */
2038         if (nid != -1) {
2039                 nodemask = cpumask_of_node(nid);
2040 
2041                 /* Look for allowed, online CPU in same node. */
2042                 for_each_cpu(dest_cpu, nodemask) {
2043                         if (!cpu_active(dest_cpu))
2044                                 continue;
2045                         if (cpumask_test_cpu(dest_cpu, p->cpus_ptr))
2046                                 return dest_cpu;
2047                 }
2048         }
2049 
2050         for (;;) {
2051                 /* Any allowed, online CPU? */
2052                 for_each_cpu(dest_cpu, p->cpus_ptr) {
2053                         if (!is_cpu_allowed(p, dest_cpu))
2054                                 continue;
2055 
2056                         goto out;
2057                 }
2058 
2059                 /* No more Mr. Nice Guy. */
2060                 switch (state) {
2061                 case cpuset:
2062                         if (IS_ENABLED(CONFIG_CPUSETS)) {
2063                                 cpuset_cpus_allowed_fallback(p);
2064                                 state = possible;
2065                                 break;
2066                         }
2067                         /* Fall-through */
2068                 case possible:
2069                         do_set_cpus_allowed(p, cpu_possible_mask);
2070                         state = fail;
2071                         break;
2072 
2073                 case fail:
2074                         BUG();
2075                         break;
2076                 }
2077         }
2078 
2079 out:
2080         if (state != cpuset) {
2081                 /*
2082                  * Don't tell them about moving exiting tasks or
2083                  * kernel threads (both mm NULL), since they never
2084                  * leave kernel.
2085                  */
2086                 if (p->mm && printk_ratelimit()) {
2087                         printk_deferred("process %d (%s) no longer affine to cpu%d\n",
2088                                         task_pid_nr(p), p->comm, cpu);
2089                 }
2090         }
2091 
2092         return dest_cpu;
2093 }
2094 
2095 /*
2096  * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable.
2097  */
2098 static inline
2099 int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
2100 {
2101         lockdep_assert_held(&p->pi_lock);
2102 
2103         if (p->nr_cpus_allowed > 1)
2104                 cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
2105         else
2106                 cpu = cpumask_any(p->cpus_ptr);
2107 
2108         /*
2109          * In order not to call set_task_cpu() on a blocking task we need
2110          * to rely on ttwu() to place the task on a valid ->cpus_ptr
2111          * CPU.
2112          *
2113          * Since this is common to all placement strategies, this lives here.
2114          *
2115          * [ this allows ->select_task() to simply return task_cpu(p) and
2116          *   not worry about this generic constraint ]
2117          */
2118         if (unlikely(!is_cpu_allowed(p, cpu)))
2119                 cpu = select_fallback_rq(task_cpu(p), p);
2120 
2121         return cpu;
2122 }
2123 
2124 static void update_avg(u64 *avg, u64 sample)
2125 {
2126         s64 diff = sample - *avg;
2127         *avg += diff >> 3;
2128 }
2129 
2130 void sched_set_stop_task(int cpu, struct task_struct *stop)
2131 {
2132         struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
2133         struct task_struct *old_stop = cpu_rq(cpu)->stop;
2134 
2135         if (stop) {
2136                 /*
2137                  * Make it appear like a SCHED_FIFO task, its something
2138                  * userspace knows about and won't get confused about.
2139                  *
2140                  * Also, it will make PI more or less work without too
2141                  * much confusion -- but then, stop work should not
2142                  * rely on PI working anyway.
2143                  */
2144                 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
2145 
2146                 stop->sched_class = &stop_sched_class;
2147         }
2148 
2149         cpu_rq(cpu)->stop = stop;
2150 
2151         if (old_stop) {
2152                 /*
2153                  * Reset it back to a normal scheduling class so that
2154                  * it can die in pieces.
2155                  */
2156                 old_stop->sched_class = &rt_sched_class;
2157         }
2158 }
2159 
2160 #else
2161 
2162 static inline int __set_cpus_allowed_ptr(struct task_struct *p,
2163                                          const struct cpumask *new_mask, bool check)
2164 {
2165         return set_cpus_allowed_ptr(p, new_mask);
2166 }
2167 
2168 #endif /* CONFIG_SMP */
2169 
2170 static void
2171 ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
2172 {
2173         struct rq *rq;
2174 
2175         if (!schedstat_enabled())
2176                 return;
2177 
2178         rq = this_rq();
2179 
2180 #ifdef CONFIG_SMP
2181         if (cpu == rq->cpu) {
2182                 __schedstat_inc(rq->ttwu_local);
2183                 __schedstat_inc(p->se.statistics.nr_wakeups_local);
2184         } else {
2185                 struct sched_domain *sd;
2186 
2187                 __schedstat_inc(p->se.statistics.nr_wakeups_remote);
2188                 rcu_read_lock();
2189                 for_each_domain(rq->cpu, sd) {
2190                         if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2191                                 __schedstat_inc(sd->ttwu_wake_remote);
2192                                 break;
2193                         }
2194                 }
2195                 rcu_read_unlock();
2196         }
2197 
2198         if (wake_flags & WF_MIGRATED)
2199                 __schedstat_inc(p->se.statistics.nr_wakeups_migrate);
2200 #endif /* CONFIG_SMP */
2201 
2202         __schedstat_inc(rq->ttwu_count);
2203         __schedstat_inc(p->se.statistics.nr_wakeups);
2204 
2205         if (wake_flags & WF_SYNC)
2206                 __schedstat_inc(p->se.statistics.nr_wakeups_sync);
2207 }
2208 
2209 /*
2210  * Mark the task runnable and perform wakeup-preemption.
2211  */
2212 static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
2213                            struct rq_flags *rf)
2214 {
2215         check_preempt_curr(rq, p, wake_flags);
2216         p->state = TASK_RUNNING;
2217         trace_sched_wakeup(p);
2218 
2219 #ifdef CONFIG_SMP
2220         if (p->sched_class->task_woken) {
2221                 /*
2222                  * Our task @p is fully woken up and running; so its safe to
2223                  * drop the rq->lock, hereafter rq is only used for statistics.
2224                  */
2225                 rq_unpin_lock(rq, rf);
2226                 p->sched_class->task_woken(rq, p);
2227                 rq_repin_lock(rq, rf);
2228         }
2229 
2230         if (rq->idle_stamp) {
2231                 u64 delta = rq_clock(rq) - rq->idle_stamp;
2232                 u64 max = 2*rq->max_idle_balance_cost;
2233 
2234                 update_avg(&rq->avg_idle, delta);
2235 
2236                 if (rq->avg_idle > max)
2237                         rq->avg_idle = max;
2238 
2239                 rq->idle_stamp = 0;
2240         }
2241 #endif
2242 }
2243 
2244 static void
2245 ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
2246                  struct rq_flags *rf)
2247 {
2248         int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK;
2249 
2250         lockdep_assert_held(&rq->lock);
2251 
2252 #ifdef CONFIG_SMP
2253         if (p->sched_contributes_to_load)
2254                 rq->nr_uninterruptible--;
2255 
2256         if (wake_flags & WF_MIGRATED)
2257                 en_flags |= ENQUEUE_MIGRATED;
2258 #endif
2259 
2260         activate_task(rq, p, en_flags);
2261         ttwu_do_wakeup(rq, p, wake_flags, rf);
2262 }
2263 
2264 /*
2265  * Called in case the task @p isn't fully descheduled from its runqueue,
2266  * in this case we must do a remote wakeup. Its a 'light' wakeup though,
2267  * since all we need to do is flip p->state to TASK_RUNNING, since
2268  * the task is still ->on_rq.
2269  */
2270 static int ttwu_remote(struct task_struct *p, int wake_flags)
2271 {
2272         struct rq_flags rf;
2273         struct rq *rq;
2274         int ret = 0;
2275 
2276         rq = __task_rq_lock(p, &rf);
2277         if (task_on_rq_queued(p)) {
2278                 /* check_preempt_curr() may use rq clock */
2279                 update_rq_clock(rq);
2280                 ttwu_do_wakeup(rq, p, wake_flags, &rf);
2281                 ret = 1;
2282         }
2283         __task_rq_unlock(rq, &rf);
2284 
2285         return ret;
2286 }
2287 
2288 #ifdef CONFIG_SMP
2289 void sched_ttwu_pending(void)
2290 {
2291         struct rq *rq = this_rq();
2292         struct llist_node *llist = llist_del_all(&rq->wake_list);
2293         struct task_struct *p, *t;
2294         struct rq_flags rf;
2295 
2296         if (!llist)
2297                 return;
2298 
2299         rq_lock_irqsave(rq, &rf);
2300         update_rq_clock(rq);
2301 
2302         llist_for_each_entry_safe(p, t, llist, wake_entry)
2303                 ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
2304 
2305         rq_unlock_irqrestore(rq, &rf);
2306 }
2307 
2308 void scheduler_ipi(void)
2309 {
2310         /*
2311          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
2312          * TIF_NEED_RESCHED remotely (for the first time) will also send
2313          * this IPI.
2314          */
2315         preempt_fold_need_resched();
2316 
2317         if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
2318                 return;
2319 
2320         /*
2321          * Not all reschedule IPI handlers call irq_enter/irq_exit, since
2322          * traditionally all their work was done from the interrupt return
2323          * path. Now that we actually do some work, we need to make sure
2324          * we do call them.
2325          *
2326          * Some archs already do call them, luckily irq_enter/exit nest
2327          * properly.
2328          *
2329          * Arguably we should visit all archs and update all handlers,
2330          * however a fair share of IPIs are still resched only so this would
2331          * somewhat pessimize the simple resched case.
2332          */
2333         irq_enter();
2334         sched_ttwu_pending();
2335 
2336         /*
2337          * Check if someone kicked us for doing the nohz idle load balance.
2338          */
2339         if (unlikely(got_nohz_idle_kick())) {
2340                 this_rq()->idle_balance = 1;
2341                 raise_softirq_irqoff(SCHED_SOFTIRQ);
2342         }
2343         irq_exit();
2344 }
2345 
2346 static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
2347 {
2348         struct rq *rq = cpu_rq(cpu);
2349 
2350         p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);
2351 
2352         if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) {
2353                 if (!set_nr_if_polling(rq->idle))
2354                         smp_send_reschedule(cpu);
2355                 else
2356                         trace_sched_wake_idle_without_ipi(cpu);
2357         }
2358 }
2359 
2360 void wake_up_if_idle(int cpu)
2361 {
2362         struct rq *rq = cpu_rq(cpu);
2363         struct rq_flags rf;
2364 
2365         rcu_read_lock();
2366 
2367         if (!is_idle_task(rcu_dereference(rq->curr)))
2368                 goto out;
2369 
2370         if (set_nr_if_polling(rq->idle)) {
2371                 trace_sched_wake_idle_without_ipi(cpu);
2372         } else {
2373                 rq_lock_irqsave(rq, &rf);
2374                 if (is_idle_task(rq->curr))
2375                         smp_send_reschedule(cpu);
2376                 /* Else CPU is not idle, do nothing here: */
2377                 rq_unlock_irqrestore(rq, &rf);
2378         }
2379 
2380 out:
2381         rcu_read_unlock();
2382 }
2383 
2384 bool cpus_share_cache(int this_cpu, int that_cpu)
2385 {
2386         return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
2387 }
2388 #endif /* CONFIG_SMP */
2389 
2390 static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
2391 {
2392         struct rq *rq = cpu_rq(cpu);
2393         struct rq_flags rf;
2394 
2395 #if defined(CONFIG_SMP)
2396         if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
2397                 sched_clock_cpu(cpu); /* Sync clocks across CPUs */
2398                 ttwu_queue_remote(p, cpu, wake_flags);
2399                 return;
2400         }
2401 #endif
2402 
2403         rq_lock(rq, &rf);
2404         update_rq_clock(rq);
2405         ttwu_do_activate(rq, p, wake_flags, &rf);
2406         rq_unlock(rq, &rf);
2407 }
2408 
2409 /*
2410  * Notes on Program-Order guarantees on SMP systems.
2411  *
2412  *  MIGRATION
2413  *
2414  * The basic program-order guarantee on SMP systems is that when a task [t]
2415  * migrates, all its activity on its old CPU [c0] happens-before any subsequent
2416  * execution on its new CPU [c1].
2417  *
2418  * For migration (of runnable tasks) this is provided by the following means:
2419  *
2420  *  A) UNLOCK of the rq(c0)->lock scheduling out task t
2421  *  B) migration for t is required to synchronize *both* rq(c0)->lock and
2422  *     rq(c1)->lock (if not at the same time, then in that order).
2423  *  C) LOCK of the rq(c1)->lock scheduling in task
2424  *
2425  * Release/acquire chaining guarantees that B happens after A and C after B.
2426  * Note: the CPU doing B need not be c0 or c1
2427  *
2428  * Example:
2429  *
2430  *   CPU0            CPU1            CPU2
2431  *
2432  *   LOCK rq(0)->lock
2433  *   sched-out X
2434  *   sched-in Y
2435  *   UNLOCK rq(0)->lock
2436  *
2437  *                                   LOCK rq(0)->lock // orders against CPU0
2438  *                                   dequeue X
2439  *                                   UNLOCK rq(0)->lock
2440  *
2441  *                                   LOCK rq(1)->lock
2442  *                                   enqueue X
2443  *                                   UNLOCK rq(1)->lock
2444  *
2445  *                   LOCK rq(1)->lock // orders against CPU2
2446  *                   sched-out Z
2447  *                   sched-in X
2448  *                   UNLOCK rq(1)->lock
2449  *
2450  *
2451  *  BLOCKING -- aka. SLEEP + WAKEUP
2452  *
2453  * For blocking we (obviously) need to provide the same guarantee as for
2454  * migration. However the means are completely different as there is no lock
2455  * chain to provide order. Instead we do:
2456  *
2457  *   1) smp_store_release(X->on_cpu, 0)
2458  *   2) smp_cond_load_acquire(!X->on_cpu)
2459  *
2460  * Example:
2461  *
2462  *   CPU0 (schedule)  CPU1 (try_to_wake_up) CPU2 (schedule)
2463  *
2464  *   LOCK rq(0)->lock LOCK X->pi_lock
2465  *   dequeue X
2466  *   sched-out X
2467  *   smp_store_release(X->on_cpu, 0);
2468  *
2469  *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
2470  *                    X->state = WAKING
2471  *                    set_task_cpu(X,2)
2472  *
2473  *                    LOCK rq(2)->lock
2474  *                    enqueue X
2475  *                    X->state = RUNNING
2476  *                    UNLOCK rq(2)->lock
2477  *
2478  *                                          LOCK rq(2)->lock // orders against CPU1
2479  *                                          sched-out Z
2480  *                                          sched-in X
2481  *                                          UNLOCK rq(2)->lock
2482  *
2483  *                    UNLOCK X->pi_lock
2484  *   UNLOCK rq(0)->lock
2485  *
2486  *
2487  * However, for wakeups there is a second guarantee we must provide, namely we
2488  * must ensure that CONDITION=1 done by the caller can not be reordered with
2489  * accesses to the task state; see try_to_wake_up() and set_current_state().
2490  */
2491 
2492 /**
2493  * try_to_wake_up - wake up a thread
2494  * @p: the thread to be awakened
2495  * @state: the mask of task states that can be woken
2496  * @wake_flags: wake modifier flags (WF_*)
2497  *
2498  * If (@state & @p->state) @p->state = TASK_RUNNING.
2499  *
2500  * If the task was not queued/runnable, also place it back on a runqueue.
2501  *
2502  * Atomic against schedule() which would dequeue a task, also see
2503  * set_current_state().
2504  *
2505  * This function executes a full memory barrier before accessing the task
2506  * state; see set_current_state().
2507  *
2508  * Return: %true if @p->state changes (an actual wakeup was done),
2509  *         %false otherwise.
2510  */
2511 static int
2512 try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
2513 {
2514         unsigned long flags;
2515         int cpu, success = 0;
2516 
2517         preempt_disable();
2518         if (p == current) {
2519                 /*
2520                  * We're waking current, this means 'p->on_rq' and 'task_cpu(p)
2521                  * == smp_processor_id()'. Together this means we can special
2522                  * case the whole 'p->on_rq && ttwu_remote()' case below
2523                  * without taking any locks.
2524                  *
2525                  * In particular:
2526                  *  - we rely on Program-Order guarantees for all the ordering,
2527                  *  - we're serialized against set_special_state() by virtue of
2528                  *    it disabling IRQs (this allows not taking ->pi_lock).
2529                  */
2530                 if (!(p->state & state))
2531                         goto out;
2532 
2533                 success = 1;
2534                 cpu = task_cpu(p);
2535                 trace_sched_waking(p);
2536                 p->state = TASK_RUNNING;
2537                 trace_sched_wakeup(p);
2538                 goto out;
2539         }
2540 
2541         /*
2542          * If we are going to wake up a thread waiting for CONDITION we
2543          * need to ensure that CONDITION=1 done by the caller can not be
2544          * reordered with p->state check below. This pairs with mb() in
2545          * set_current_state() the waiting thread does.
2546          */
2547         raw_spin_lock_irqsave(&p->pi_lock, flags);
2548         smp_mb__after_spinlock();
2549         if (!(p->state & state))
2550                 goto unlock;
2551 
2552         trace_sched_waking(p);
2553 
2554         /* We're going to change ->state: */
2555         success = 1;
2556         cpu = task_cpu(p);
2557 
2558         /*
2559          * Ensure we load p->on_rq _after_ p->state, otherwise it would
2560          * be possible to, falsely, observe p->on_rq == 0 and get stuck
2561          * in smp_cond_load_acquire() below.
2562          *
2563          * sched_ttwu_pending()                 try_to_wake_up()
2564          *   STORE p->on_rq = 1                   LOAD p->state
2565          *   UNLOCK rq->lock
2566          *
2567          * __schedule() (switch to task 'p')
2568          *   LOCK rq->lock                        smp_rmb();
2569          *   smp_mb__after_spinlock();
2570          *   UNLOCK rq->lock
2571          *
2572          * [task p]
2573          *   STORE p->state = UNINTERRUPTIBLE     LOAD p->on_rq
2574          *
2575          * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
2576          * __schedule().  See the comment for smp_mb__after_spinlock().
2577          */
2578         smp_rmb();
2579         if (p->on_rq && ttwu_remote(p, wake_flags))
2580                 goto unlock;
2581 
2582 #ifdef CONFIG_SMP
2583         /*
2584          * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be
2585          * possible to, falsely, observe p->on_cpu == 0.
2586          *
2587          * One must be running (->on_cpu == 1) in order to remove oneself
2588          * from the runqueue.
2589          *
2590          * __schedule() (switch to task 'p')    try_to_wake_up()
2591          *   STORE p->on_cpu = 1                  LOAD p->on_rq
2592          *   UNLOCK rq->lock
2593          *
2594          * __schedule() (put 'p' to sleep)
2595          *   LOCK rq->lock                        smp_rmb();
2596          *   smp_mb__after_spinlock();
2597          *   STORE p->on_rq = 0                   LOAD p->on_cpu
2598          *
2599          * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
2600          * __schedule().  See the comment for smp_mb__after_spinlock().
2601          */
2602         smp_rmb();
2603 
2604         /*
2605          * If the owning (remote) CPU is still in the middle of schedule() with
2606          * this task as prev, wait until its done referencing the task.
2607          *
2608          * Pairs with the smp_store_release() in finish_task().
2609          *
2610          * This ensures that tasks getting woken will be fully ordered against
2611          * their previous state and preserve Program Order.
2612          */
2613         smp_cond_load_acquire(&p->on_cpu, !VAL);
2614 
2615         p->sched_contributes_to_load = !!task_contributes_to_load(p);
2616         p->state = TASK_WAKING;
2617 
2618         if (p->in_iowait) {
2619                 delayacct_blkio_end(p);
2620                 atomic_dec(&task_rq(p)->nr_iowait);
2621         }
2622 
2623         cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2624         if (task_cpu(p) != cpu) {
2625                 wake_flags |= WF_MIGRATED;
2626                 psi_ttwu_dequeue(p);
2627                 set_task_cpu(p, cpu);
2628         }
2629 
2630 #else /* CONFIG_SMP */
2631 
2632         if (p->in_iowait) {
2633                 delayacct_blkio_end(p);
2634                 atomic_dec(&task_rq(p)->nr_iowait);
2635         }
2636 
2637 #endif /* CONFIG_SMP */
2638 
2639         ttwu_queue(p, cpu, wake_flags);
2640 unlock:
2641         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2642 out:
2643         if (success)
2644                 ttwu_stat(p, cpu, wake_flags);
2645         preempt_enable();
2646 
2647         return success;
2648 }
2649 
2650 /**
2651  * wake_up_process - Wake up a specific process
2652  * @p: The process to be woken up.
2653  *
2654  * Attempt to wake up the nominated process and move it to the set of runnable
2655  * processes.
2656  *
2657  * Return: 1 if the process was woken up, 0 if it was already running.
2658  *
2659  * This function executes a full memory barrier before accessing the task state.
2660  */
2661 int wake_up_process(struct task_struct *p)
2662 {
2663         return try_to_wake_up(p, TASK_NORMAL, 0);
2664 }
2665 EXPORT_SYMBOL(wake_up_process);
2666 
2667 int wake_up_state(struct task_struct *p, unsigned int state)
2668 {
2669         return try_to_wake_up(p, state, 0);
2670 }
2671 
2672 /*
2673  * Perform scheduler related setup for a newly forked process p.
2674  * p is forked by current.
2675  *
2676  * __sched_fork() is basic setup used by init_idle() too:
2677  */
2678 static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
2679 {
2680         p->on_rq                        = 0;
2681 
2682         p->se.on_rq                     = 0;
2683         p->se.exec_start                = 0;
2684         p->se.sum_exec_runtime          = 0;
2685         p->se.prev_sum_exec_runtime     = 0;
2686         p->se.nr_migrations             = 0;
2687         p->se.vruntime                  = 0;
2688         INIT_LIST_HEAD(&p->se.group_node);
2689 
2690 #ifdef CONFIG_FAIR_GROUP_SCHED
2691         p->se.cfs_rq                    = NULL;
2692 #endif
2693 
2694 #ifdef CONFIG_SCHEDSTATS
2695         /* Even if schedstat is disabled, there should not be garbage */
2696         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
2697 #endif
2698 
2699         RB_CLEAR_NODE(&p->dl.rb_node);
2700         init_dl_task_timer(&p->dl);
2701         init_dl_inactive_task_timer(&p->dl);
2702         __dl_clear_params(p);
2703 
2704         INIT_LIST_HEAD(&p->rt.run_list);
2705         p->rt.timeout           = 0;
2706         p->rt.time_slice        = sched_rr_timeslice;
2707         p->rt.on_rq             = 0;
2708         p->rt.on_list           = 0;
2709 
2710 #ifdef CONFIG_PREEMPT_NOTIFIERS
2711         INIT_HLIST_HEAD(&p->preempt_notifiers);
2712 #endif
2713 
2714 #ifdef CONFIG_COMPACTION
2715         p->capture_control = NULL;
2716 #endif
2717         init_numa_balancing(clone_flags, p);
2718 }
2719 
2720 DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);
2721 
2722 #ifdef CONFIG_NUMA_BALANCING
2723 
2724 void set_numabalancing_state(bool enabled)
2725 {
2726         if (enabled)
2727                 static_branch_enable(&sched_numa_balancing);
2728         else
2729                 static_branch_disable(&sched_numa_balancing);
2730 }
2731 
2732 #ifdef CONFIG_PROC_SYSCTL
2733 int sysctl_numa_balancing(struct ctl_table *table, int write,
2734                          void __user *buffer, size_t *lenp, loff_t *ppos)
2735 {
2736         struct ctl_table t;
2737         int err;
2738         int state = static_branch_likely(&sched_numa_balancing);
2739 
2740         if (write && !capable(CAP_SYS_ADMIN))
2741                 return -EPERM;
2742 
2743         t = *table;
2744         t.data = &state;
2745         err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2746         if (err < 0)
2747                 return err;
2748         if (write)
2749                 set_numabalancing_state(state);
2750         return err;
2751 }
2752 #endif
2753 #endif
2754 
2755 #ifdef CONFIG_SCHEDSTATS
2756 
2757 DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2758 static bool __initdata __sched_schedstats = false;
2759 
2760 static void set_schedstats(bool enabled)
2761 {
2762         if (enabled)
2763                 static_branch_enable(&sched_schedstats);
2764         else
2765                 static_branch_disable(&sched_schedstats);
2766 }
2767 
2768 void force_schedstat_enabled(void)
2769 {
2770         if (!schedstat_enabled()) {
2771                 pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n");
2772                 static_branch_enable(&sched_schedstats);
2773         }
2774 }
2775 
2776 static int __init setup_schedstats(char *str)
2777 {
2778         int ret = 0;
2779         if (!str)
2780                 goto out;
2781 
2782         /*
2783          * This code is called before jump labels have been set up, so we can't
2784          * change the static branch directly just yet.  Instead set a temporary
2785          * variable so init_schedstats() can do it later.
2786          */
2787         if (!strcmp(str, "enable")) {
2788                 __sched_schedstats = true;
2789                 ret = 1;
2790         } else if (!strcmp(str, "disable")) {
2791                 __sched_schedstats = false;
2792                 ret = 1;
2793         }
2794 out:
2795         if (!ret)
2796                 pr_warn("Unable to parse schedstats=\n");
2797 
2798         return ret;
2799 }
2800 __setup("schedstats=", setup_schedstats);
2801 
2802 static void __init init_schedstats(void)
2803 {
2804         set_schedstats(__sched_schedstats);
2805 }
2806 
2807 #ifdef CONFIG_PROC_SYSCTL
2808 int sysctl_schedstats(struct ctl_table *table, int write,
2809                          void __user *buffer, size_t *lenp, loff_t *ppos)
2810 {
2811         struct ctl_table t;
2812         int err;
2813         int state = static_branch_likely(&sched_schedstats);
2814 
2815         if (write && !capable(CAP_SYS_ADMIN))
2816                 return -EPERM;
2817 
2818         t = *table;
2819         t.data = &state;
2820         err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2821         if (err < 0)
2822                 return err;
2823         if (write)
2824                 set_schedstats(state);
2825         return err;
2826 }
2827 #endif /* CONFIG_PROC_SYSCTL */
2828 #else  /* !CONFIG_SCHEDSTATS */
2829 static inline void init_schedstats(void) {}
2830 #endif /* CONFIG_SCHEDSTATS */
2831 
2832 /*
2833  * fork()/clone()-time setup:
2834  */
2835 int sched_fork(unsigned long clone_flags, struct task_struct *p)
2836 {
2837         unsigned long flags;
2838 
2839         __sched_fork(clone_flags, p);
2840         /*
2841          * We mark the process as NEW here. This guarantees that
2842          * nobody will actually run it, and a signal or other external
2843          * event cannot wake it up and insert it on the runqueue either.
2844          */
2845         p->state = TASK_NEW;
2846 
2847         /*
2848          * Make sure we do not leak PI boosting priority to the child.
2849          */
2850         p->prio = current->normal_prio;
2851 
2852         uclamp_fork(p);
2853 
2854         /*
2855          * Revert to default priority/policy on fork if requested.
2856          */
2857         if (unlikely(p->sched_reset_on_fork)) {
2858                 if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2859                         p->policy = SCHED_NORMAL;
2860                         p->static_prio = NICE_TO_PRIO(0);
2861                         p->rt_priority = 0;
2862                 } else if (PRIO_TO_NICE(p->static_prio) < 0)
2863                         p->static_prio = NICE_TO_PRIO(0);
2864 
2865                 p->prio = p->normal_prio = __normal_prio(p);
2866                 set_load_weight(p, false);
2867 
2868                 /*
2869                  * We don't need the reset flag anymore after the fork. It has
2870                  * fulfilled its duty:
2871                  */
2872                 p->sched_reset_on_fork = 0;
2873         }
2874 
2875         if (dl_prio(p->prio))
2876                 return -EAGAIN;
2877         else if (rt_prio(p->prio))
2878                 p->sched_class = &rt_sched_class;
2879         else
2880                 p->sched_class = &fair_sched_class;
2881 
2882         init_entity_runnable_average(&p->se);
2883 
2884         /*
2885          * The child is not yet in the pid-hash so no cgroup attach races,
2886          * and the cgroup is pinned to this child due to cgroup_fork()
2887          * is ran before sched_fork().
2888          *
2889          * Silence PROVE_RCU.
2890          */
2891         raw_spin_lock_irqsave(&p->pi_lock, flags);
2892         /*
2893          * We're setting the CPU for the first time, we don't migrate,
2894          * so use __set_task_cpu().
2895          */
2896         __set_task_cpu(p, smp_processor_id());
2897         if (p->sched_class->task_fork)
2898                 p->sched_class->task_fork(p);
2899         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2900 
2901 #ifdef CONFIG_SCHED_INFO
2902         if (likely(sched_info_on()))
2903                 memset(&p->sched_info, 0, sizeof(p->sched_info));
2904 #endif
2905 #if defined(CONFIG_SMP)
2906         p->on_cpu = 0;
2907 #endif
2908         init_task_preempt_count(p);
2909 #ifdef CONFIG_SMP
2910         plist_node_init(&p->pushable_tasks, MAX_PRIO);
2911         RB_CLEAR_NODE(&p->pushable_dl_tasks);
2912 #endif
2913         return 0;
2914 }
2915 
2916 unsigned long to_ratio(u64 period, u64 runtime)
2917 {
2918         if (runtime == RUNTIME_INF)
2919                 return BW_UNIT;
2920 
2921         /*
2922          * Doing this here saves a lot of checks in all
2923          * the calling paths, and returning zero seems
2924          * safe for them anyway.
2925          */
2926         if (period == 0)
2927                 return 0;
2928 
2929         return div64_u64(runtime << BW_SHIFT, period);
2930 }
2931 
2932 /*
2933  * wake_up_new_task - wake up a newly created task for the first time.
2934  *
2935  * This function will do some initial scheduler statistics housekeeping
2936  * that must be done for every newly created context, then puts the task
2937  * on the runqueue and wakes it.
2938  */
2939 void wake_up_new_task(struct task_struct *p)
2940 {
2941         struct rq_flags rf;
2942         struct rq *rq;
2943 
2944         raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2945         p->state = TASK_RUNNING;
2946 #ifdef CONFIG_SMP
2947         /*
2948          * Fork balancing, do it here and not earlier because:
2949          *  - cpus_ptr can change in the fork path
2950          *  - any previously selected CPU might disappear through hotplug
2951          *
2952          * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
2953          * as we're not fully set-up yet.
2954          */
2955         p->recent_used_cpu = task_cpu(p);
2956         __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2957 #endif
2958         rq = __task_rq_lock(p, &rf);
2959         update_rq_clock(rq);
2960         post_init_entity_util_avg(p);
2961 
2962         activate_task(rq, p, ENQUEUE_NOCLOCK);
2963         trace_sched_wakeup_new(p);
2964         check_preempt_curr(rq, p, WF_FORK);
2965 #ifdef CONFIG_SMP
2966         if (p->sched_class->task_woken) {
2967                 /*
2968                  * Nothing relies on rq->lock after this, so its fine to
2969                  * drop it.
2970                  */
2971                 rq_unpin_lock(rq, &rf);
2972                 p->sched_class->task_woken(rq, p);
2973                 rq_repin_lock(rq, &rf);
2974         }
2975 #endif
2976         task_rq_unlock(rq, p, &rf);
2977 }
2978 
2979 #ifdef CONFIG_PREEMPT_NOTIFIERS
2980 
2981 static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key);
2982 
2983 void preempt_notifier_inc(void)
2984 {
2985         static_branch_inc(&preempt_notifier_key);
2986 }
2987 EXPORT_SYMBOL_GPL(preempt_notifier_inc);
2988 
2989 void preempt_notifier_dec(void)
2990 {
2991         static_branch_dec(&preempt_notifier_key);
2992 }
2993 EXPORT_SYMBOL_GPL(preempt_notifier_dec);
2994 
2995 /**
2996  * preempt_notifier_register - tell me when current is being preempted & rescheduled
2997  * @notifier: notifier struct to register
2998  */
2999 void preempt_notifier_register(struct preempt_notifier *notifier)
3000 {
3001         if (!static_branch_unlikely(&preempt_notifier_key))
3002                 WARN(1, "registering preempt_notifier while notifiers disabled\n");
3003 
3004         hlist_add_head(&notifier->link, &current->preempt_notifiers);
3005 }
3006 EXPORT_SYMBOL_GPL(preempt_notifier_register);
3007 
3008 /**
3009  * preempt_notifier_unregister - no longer interested in preemption notifications
3010  * @notifier: notifier struct to unregister
3011  *
3012  * This is *not* safe to call from within a preemption notifier.
3013  */
3014 void preempt_notifier_unregister(struct preempt_notifier *notifier)
3015 {
3016         hlist_del(&notifier->link);
3017 }
3018 EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
3019 
3020 static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
3021 {
3022         struct preempt_notifier *notifier;
3023 
3024         hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
3025                 notifier->ops->sched_in(notifier, raw_smp_processor_id());
3026 }
3027 
3028 static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
3029 {
3030         if (static_branch_unlikely(&preempt_notifier_key))
3031                 __fire_sched_in_preempt_notifiers(curr);
3032 }
3033 
3034 static void
3035 __fire_sched_out_preempt_notifiers(struct task_struct *curr,
3036                                    struct task_struct *next)
3037 {
3038         struct preempt_notifier *notifier;
3039 
3040         hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
3041                 notifier->ops->sched_out(notifier, next);
3042 }
3043 
3044 static __always_inline void
3045 fire_sched_out_preempt_notifiers(struct task_struct *curr,
3046                                  struct task_struct *next)
3047 {
3048         if (static_branch_unlikely(&preempt_notifier_key))
3049                 __fire_sched_out_preempt_notifiers(curr, next);
3050 }
3051 
3052 #else /* !CONFIG_PREEMPT_NOTIFIERS */
3053 
3054 static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
3055 {
3056 }
3057 
3058 static inline void
3059 fire_sched_out_preempt_notifiers(struct task_struct *curr,
3060                                  struct task_struct *next)
3061 {
3062 }
3063 
3064 #endif /* CONFIG_PREEMPT_NOTIFIERS */
3065 
3066 static inline void prepare_task(struct task_struct *next)
3067 {
3068 #ifdef CONFIG_SMP
3069         /*
3070          * Claim the task as running, we do this before switching to it
3071          * such that any running task will have this set.
3072          */
3073         next->on_cpu = 1;
3074 #endif
3075 }
3076 
3077 static inline void finish_task(struct task_struct *prev)
3078 {
3079 #ifdef CONFIG_SMP
3080         /*
3081          * After ->on_cpu is cleared, the task can be moved to a different CPU.
3082          * We must ensure this doesn't happen until the switch is completely
3083          * finished.
3084          *
3085          * In particular, the load of prev->state in finish_task_switch() must
3086          * happen before this.
3087          *
3088          * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
3089          */
3090         smp_store_release(&prev->on_cpu, 0);
3091 #endif
3092 }
3093 
3094 static inline void
3095 prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf)
3096 {
3097         /*
3098          * Since the runqueue lock will be released by the next
3099          * task (which is an invalid locking op but in the case
3100          * of the scheduler it's an obvious special-case), so we
3101          * do an early lockdep release here:
3102          */
3103         rq_unpin_lock(rq, rf);
3104         spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
3105 #ifdef CONFIG_DEBUG_SPINLOCK
3106         /* this is a valid case when another task releases the spinlock */
3107         rq->lock.owner = next;
3108 #endif
3109 }
3110 
3111 static inline void finish_lock_switch(struct rq *rq)
3112 {
3113         /*
3114          * If we are tracking spinlock dependencies then we have to
3115          * fix up the runqueue lock - which gets 'carried over' from
3116          * prev into current:
3117          */
3118         spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
3119         raw_spin_unlock_irq(&rq->lock);
3120 }
3121 
3122 /*
3123  * NOP if the arch has not defined these:
3124  */
3125 
3126 #ifndef prepare_arch_switch
3127 # define prepare_arch_switch(next)      do { } while (0)
3128 #endif
3129 
3130 #ifndef finish_arch_post_lock_switch
3131 # define finish_arch_post_lock_switch() do { } while (0)
3132 #endif
3133 
3134 /**
3135  * prepare_task_switch - prepare to switch tasks
3136  * @rq: the runqueue preparing to switch
3137  * @prev: the current task that is being switched out
3138  * @next: the task we are going to switch to.
3139  *
3140  * This is called with the rq lock held and interrupts off. It must
3141  * be paired with a subsequent finish_task_switch after the context
3142  * switch.
3143  *
3144  * prepare_task_switch sets up locking and calls architecture specific
3145  * hooks.
3146  */
3147 static inline void
3148 prepare_task_switch(struct rq *rq, struct task_struct *prev,
3149                     struct task_struct *next)
3150 {
3151         kcov_prepare_switch(prev);
3152         sched_info_switch(rq, prev, next);
3153         perf_event_task_sched_out(prev, next);
3154         rseq_preempt(prev);
3155         fire_sched_out_preempt_notifiers(prev, next);
3156         prepare_task(next);
3157         prepare_arch_switch(next);
3158 }
3159 
3160 /**
3161  * finish_task_switch - clean up after a task-switch
3162  * @prev: the thread we just switched away from.
3163  *
3164  * finish_task_switch must be called after the context switch, paired
3165  * with a prepare_task_switch call before the context switch.
3166  * finish_task_switch will reconcile locking set up by prepare_task_switch,
3167  * and do any other architecture-specific cleanup actions.
3168  *
3169  * Note that we may have delayed dropping an mm in context_switch(). If
3170  * so, we finish that here outside of the runqueue lock. (Doing it
3171  * with the lock held can cause deadlocks; see schedule() for
3172  * details.)
3173  *
3174  * The context switch have flipped the stack from under us and restored the
3175  * local variables which were saved when this task called schedule() in the
3176  * past. prev == current is still correct but we need to recalculate this_rq
3177  * because prev may have moved to another CPU.
3178  */
3179 static struct rq *finish_task_switch(struct task_struct *prev)
3180         __releases(rq->lock)
3181 {
3182         struct rq *rq = this_rq();
3183         struct mm_struct *mm = rq->prev_mm;
3184         long prev_state;
3185 
3186         /*
3187          * The previous task will have left us with a preempt_count of 2
3188          * because it left us after:
3189          *
3190          *      schedule()
3191          *        preempt_disable();                    // 1
3192          *        __schedule()
3193          *          raw_spin_lock_irq(&rq->lock)        // 2
3194          *
3195          * Also, see FORK_PREEMPT_COUNT.
3196          */
3197         if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET,
3198                       "corrupted preempt_count: %s/%d/0x%x\n",
3199                       current->comm, current->pid, preempt_count()))
3200                 preempt_count_set(FORK_PREEMPT_COUNT);
3201 
3202         rq->prev_mm = NULL;
3203 
3204         /*
3205          * A task struct has one reference for the use as "current".
3206          * If a task dies, then it sets TASK_DEAD in tsk->state and calls
3207          * schedule one last time. The schedule call will never return, and
3208          * the scheduled task must drop that reference.
3209          *
3210          * We must observe prev->state before clearing prev->on_cpu (in
3211          * finish_task), otherwise a concurrent wakeup can get prev
3212          * running on another CPU and we could rave with its RUNNING -> DEAD
3213          * transition, resulting in a double drop.
3214          */
3215         prev_state = prev->state;
3216         vtime_task_switch(prev);
3217         perf_event_task_sched_in(prev, current);
3218         finish_task(prev);
3219         finish_lock_switch(rq);
3220         finish_arch_post_lock_switch();
3221         kcov_finish_switch(current);
3222 
3223         fire_sched_in_preempt_notifiers(current);
3224         /*
3225          * When switching through a kernel thread, the loop in
3226          * membarrier_{private,global}_expedited() may have observed that
3227          * kernel thread and not issued an IPI. It is therefore possible to
3228          * schedule between user->kernel->user threads without passing though
3229          * switch_mm(). Membarrier requires a barrier after storing to
3230          * rq->curr, before returning to userspace, so provide them here:
3231          *
3232          * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly
3233          *   provided by mmdrop(),
3234          * - a sync_core for SYNC_CORE.
3235          */
3236         if (mm) {
3237                 membarrier_mm_sync_core_before_usermode(mm);
3238                 mmdrop(mm);
3239         }
3240         if (unlikely(prev_state == TASK_DEAD)) {
3241                 if (prev->sched_class->task_dead)
3242                         prev->sched_class->task_dead(prev);
3243 
3244                 /*
3245                  * Remove function-return probe instances associated with this
3246                  * task and put them back on the free list.
3247                  */
3248                 kprobe_flush_task(prev);
3249 
3250                 /* Task is done with its stack. */
3251                 put_task_stack(prev);
3252 
3253                 put_task_struct_rcu_user(prev);
3254         }
3255 
3256         tick_nohz_task_switch();
3257         return rq;
3258 }
3259 
3260 #ifdef CONFIG_SMP
3261 
3262 /* rq->lock is NOT held, but preemption is disabled */
3263 static void __balance_callback(struct rq *rq)
3264 {
3265         struct callback_head *head, *next;
3266         void (*func)(struct rq *rq);
3267         unsigned long flags;
3268 
3269         raw_spin_lock_irqsave(&rq->lock, flags);
3270         head = rq->balance_callback;
3271         rq->balance_callback = NULL;
3272         while (head) {
3273                 func = (void (*)(struct rq *))head->func;
3274                 next = head->next;
3275                 head->next = NULL;
3276                 head = next;
3277 
3278                 func(rq);
3279         }
3280         raw_spin_unlock_irqrestore(&rq->lock, flags);
3281 }
3282 
3283 static inline void balance_callback(struct rq *rq)
3284 {
3285         if (unlikely(rq->balance_callback))
3286                 __balance_callback(rq);
3287 }
3288 
3289 #else
3290 
3291 static inline void balance_callback(struct rq *rq)
3292 {
3293 }
3294 
3295 #endif
3296 
3297 /**
3298  * schedule_tail - first thing a freshly forked thread must call.
3299  * @prev: the thread we just switched away from.
3300  */
3301 asmlinkage __visible void schedule_tail(struct task_struct *prev)
3302         __releases(rq->lock)
3303 {
3304         struct rq *rq;
3305 
3306         /*
3307          * New tasks start with FORK_PREEMPT_COUNT, see there and
3308          * finish_task_switch() for details.
3309          *
3310          * finish_task_switch() will drop rq->lock() and lower preempt_count
3311          * and the preempt_enable() will end up enabling preemption (on
3312          * PREEMPT_COUNT kernels).
3313          */
3314 
3315         rq = finish_task_switch(prev);
3316         balance_callback(rq);
3317         preempt_enable();
3318 
3319         if (current->set_child_tid)
3320                 put_user(task_pid_vnr(current), current->set_child_tid);
3321 
3322         calculate_sigpending();
3323 }
3324 
3325 /*
3326  * context_switch - switch to the new MM and the new thread's register state.
3327  */
3328 static __always_inline struct rq *
3329 context_switch(struct rq *rq, struct task_struct *prev,
3330                struct task_struct *next, struct rq_flags *rf)
3331 {
3332         prepare_task_switch(rq, prev, next);
3333 
3334         /*
3335          * For paravirt, this is coupled with an exit in switch_to to
3336          * combine the page table reload and the switch backend into
3337          * one hypercall.
3338          */
3339         arch_start_context_switch(prev);
3340 
3341         /*
3342          * kernel -> kernel   lazy + transfer active
3343          *   user -> kernel   lazy + mmgrab() active
3344          *
3345          * kernel ->   user   switch + mmdrop() active
3346          *   user ->   user   switch
3347          */
3348         if (!next->mm) {                                // to kernel
3349                 enter_lazy_tlb(prev->active_mm, next);
3350 
3351                 next->active_mm = prev->active_mm;
3352                 if (prev->mm)                           // from user
3353                         mmgrab(prev->active_mm);
3354                 else
3355                         prev->active_mm = NULL;
3356         } else {                                        // to user
3357                 membarrier_switch_mm(rq, prev->active_mm, next->mm);
3358                 /*
3359                  * sys_membarrier() requires an smp_mb() between setting
3360                  * rq->curr / membarrier_switch_mm() and returning to userspace.
3361                  *
3362                  * The below provides this either through switch_mm(), or in
3363                  * case 'prev->active_mm == next->mm' through
3364                  * finish_task_switch()'s mmdrop().
3365                  */
3366                 switch_mm_irqs_off(prev->active_mm, next->mm, next);
3367 
3368                 if (!prev->mm) {                        // from kernel
3369                         /* will mmdrop() in finish_task_switch(). */
3370                         rq->prev_mm = prev->active_mm;
3371                         prev->active_mm = NULL;
3372                 }
3373         }
3374 
3375         rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3376 
3377         prepare_lock_switch(rq, next, rf);
3378 
3379         /* Here we just switch the register state and the stack. */
3380         switch_to(prev, next, prev);
3381         barrier();
3382 
3383         return finish_task_switch(prev);
3384 }
3385 
3386 /*
3387  * nr_running and nr_context_switches:
3388  *
3389  * externally visible scheduler statistics: current number of runnable
3390  * threads, total number of context switches performed since bootup.
3391  */
3392 unsigned long nr_running(void)
3393 {
3394         unsigned long i, sum = 0;
3395 
3396         for_each_online_cpu(i)
3397                 sum += cpu_rq(i)->nr_running;
3398 
3399         return sum;
3400 }
3401 
3402 /*
3403  * Check if only the current task is running on the CPU.
3404  *
3405  * Caution: this function does not check that the caller has disabled
3406  * preemption, thus the result might have a time-of-check-to-time-of-use
3407  * race.  The caller is responsible to use it correctly, for example:
3408  *
3409  * - from a non-preemptible section (of course)
3410  *
3411  * - from a thread that is bound to a single CPU
3412  *
3413  * - in a loop with very short iterations (e.g. a polling loop)
3414  */
3415 bool single_task_running(void)
3416 {
3417         return raw_rq()->nr_running == 1;
3418 }
3419 EXPORT_SYMBOL(single_task_running);
3420 
3421 unsigned long long nr_context_switches(void)
3422 {
3423         int i;
3424         unsigned long long sum = 0;
3425 
3426         for_each_possible_cpu(i)
3427                 sum += cpu_rq(i)->nr_switches;
3428 
3429         return sum;
3430 }
3431 
3432 /*
3433  * Consumers of these two interfaces, like for example the cpuidle menu
3434  * governor, are using nonsensical data. Preferring shallow idle state selection
3435  * for a CPU that has IO-wait which might not even end up running the task when
3436  * it does become runnable.
3437  */
3438 
3439 unsigned long nr_iowait_cpu(int cpu)
3440 {
3441         return atomic_read(&cpu_rq(cpu)->nr_iowait);
3442 }
3443 
3444 /*
3445  * IO-wait accounting, and how its mostly bollocks (on SMP).
3446  *
3447  * The idea behind IO-wait account is to account the idle time that we could
3448  * have spend running if it were not for IO. That is, if we were to improve the
3449  * storage performance, we'd have a proportional reduction in IO-wait time.
3450  *
3451  * This all works nicely on UP, where, when a task blocks on IO, we account
3452  * idle time as IO-wait, because if the storage were faster, it could've been
3453  * running and we'd not be idle.
3454  *
3455  * This has been extended to SMP, by doing the same for each CPU. This however
3456  * is broken.
3457  *
3458  * Imagine for instance the case where two tasks block on one CPU, only the one
3459  * CPU will have IO-wait accounted, while the other has regular idle. Even
3460  * though, if the storage were faster, both could've ran at the same time,
3461  * utilising both CPUs.
3462  *
3463  * This means, that when looking globally, the current IO-wait accounting on
3464  * SMP is a lower bound, by reason of under accounting.
3465  *
3466  * Worse, since the numbers are provided per CPU, they are sometimes
3467  * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly
3468  * associated with any one particular CPU, it can wake to another CPU than it
3469  * blocked on. This means the per CPU IO-wait number is meaningless.
3470  *
3471  * Task CPU affinities can make all that even more 'interesting'.
3472  */
3473 
3474 unsigned long nr_iowait(void)
3475 {
3476         unsigned long i, sum = 0;
3477 
3478         for_each_possible_cpu(i)
3479                 sum += nr_iowait_cpu(i);
3480 
3481         return sum;
3482 }
3483 
3484 #ifdef CONFIG_SMP
3485 
3486 /*
3487  * sched_exec - execve() is a valuable balancing opportunity, because at
3488  * this point the task has the smallest effective memory and cache footprint.
3489  */
3490 void sched_exec(void)
3491 {
3492         struct task_struct *p = current;
3493         unsigned long flags;
3494         int dest_cpu;
3495 
3496         raw_spin_lock_irqsave(&p->pi_lock, flags);
3497         dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
3498         if (dest_cpu == smp_processor_id())
3499                 goto unlock;
3500 
3501         if (likely(cpu_active(dest_cpu))) {
3502                 struct migration_arg arg = { p, dest_cpu };
3503 
3504                 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
3505                 stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
3506                 return;
3507         }
3508 unlock:
3509         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
3510 }
3511 
3512 #endif
3513 
3514 DEFINE_PER_CPU(struct kernel_stat, kstat);
3515 DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
3516 
3517 EXPORT_PER_CPU_SYMBOL(kstat);
3518 EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
3519 
3520 /*
3521  * The function fair_sched_class.update_curr accesses the struct curr
3522  * and its field curr->exec_start; when called from task_sched_runtime(),
3523  * we observe a high rate of cache misses in practice.
3524  * Prefetching this data results in improved performance.
3525  */
3526 static inline void prefetch_curr_exec_start(struct task_struct *p)
3527 {
3528 #ifdef CONFIG_FAIR_GROUP_SCHED
3529         struct sched_entity *curr = (&p->se)->cfs_rq->curr;
3530 #else
3531         struct sched_entity *curr = (&task_rq(p)->cfs)->curr;
3532 #endif
3533         prefetch(curr);
3534         prefetch(&curr->exec_start);
3535 }
3536 
3537 /*
3538  * Return accounted runtime for the task.
3539  * In case the task is currently running, return the runtime plus current's
3540  * pending runtime that have not been accounted yet.
3541  */
3542 unsigned long long task_sched_runtime(struct task_struct *p)
3543 {
3544         struct rq_flags rf;
3545         struct rq *rq;
3546         u64 ns;
3547 
3548 #if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
3549         /*
3550          * 64-bit doesn't need locks to atomically read a 64-bit value.
3551          * So we have a optimization chance when the task's delta_exec is 0.
3552          * Reading ->on_cpu is racy, but this is ok.
3553          *
3554          * If we race with it leaving CPU, we'll take a lock. So we're correct.
3555          * If we race with it entering CPU, unaccounted time is 0. This is
3556          * indistinguishable from the read occurring a few cycles earlier.
3557          * If we see ->on_cpu without ->on_rq, the task is leaving, and has
3558          * been accounted, so we're correct here as well.
3559          */
3560         if (!p->on_cpu || !task_on_rq_queued(p))
3561                 return p->se.sum_exec_runtime;
3562 #endif
3563 
3564         rq = task_rq_lock(p, &rf);
3565         /*
3566          * Must be ->curr _and_ ->on_rq.  If dequeued, we would
3567          * project cycles that may never be accounted to this
3568          * thread, breaking clock_gettime().
3569          */
3570         if (task_current(rq, p) && task_on_rq_queued(p)) {
3571                 prefetch_curr_exec_start(p);
3572                 update_rq_clock(rq);
3573                 p->sched_class->update_curr(rq);
3574         }
3575         ns = p->se.sum_exec_runtime;
3576         task_rq_unlock(rq, p, &rf);
3577 
3578         return ns;
3579 }
3580 
3581 /*
3582  * This function gets called by the timer code, with HZ frequency.
3583  * We call it with interrupts disabled.
3584  */
3585 void scheduler_tick(void)
3586 {
3587         int cpu = smp_processor_id();
3588         struct rq *rq = cpu_rq(cpu);
3589         struct task_struct *curr = rq->curr;
3590         struct rq_flags rf;
3591 
3592         sched_clock_tick();
3593 
3594         rq_lock(rq, &rf);
3595 
3596         update_rq_clock(rq);
3597         curr->sched_class->task_tick(rq, curr, 0);
3598         calc_global_load_tick(rq);
3599         psi_task_tick(rq);
3600 
3601         rq_unlock(rq, &rf);
3602 
3603         perf_event_task_tick();
3604 
3605 #ifdef CONFIG_SMP
3606         rq->idle_balance = idle_cpu(cpu);
3607         trigger_load_balance(rq);
3608 #endif
3609 }
3610 
3611 #ifdef CONFIG_NO_HZ_FULL
3612 
3613 struct tick_work {
3614         int                     cpu;
3615         atomic_t                state;
3616         struct delayed_work     work;
3617 };
3618 /* Values for ->state, see diagram below. */
3619 #define TICK_SCHED_REMOTE_OFFLINE       0
3620 #define TICK_SCHED_REMOTE_OFFLINING     1
3621 #define TICK_SCHED_REMOTE_RUNNING       2
3622 
3623 /*
3624  * State diagram for ->state:
3625  *
3626  *
3627  *          TICK_SCHED_REMOTE_OFFLINE
3628  *                    |   ^
3629  *                    |   |
3630  *                    |   | sched_tick_remote()
3631  *                    |   |
3632  *                    |   |
3633  *                    +--TICK_SCHED_REMOTE_OFFLINING
3634  *                    |   ^
3635  *                    |   |
3636  * sched_tick_start() |   | sched_tick_stop()
3637  *                    |   |
3638  *                    V   |
3639  *          TICK_SCHED_REMOTE_RUNNING
3640  *
3641  *
3642  * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote()
3643  * and sched_tick_start() are happy to leave the state in RUNNING.
3644  */
3645 
3646 static struct tick_work __percpu *tick_work_cpu;
3647 
3648 static void sched_tick_remote(struct work_struct *work)
3649 {
3650         struct delayed_work *dwork = to_delayed_work(work);
3651         struct tick_work *twork = container_of(dwork, struct tick_work, work);
3652         int cpu = twork->cpu;
3653         struct rq *rq = cpu_rq(cpu);
3654         struct task_struct *curr;
3655         struct rq_flags rf;
3656         u64 delta;
3657         int os;
3658 
3659         /*
3660          * Handle the tick only if it appears the remote CPU is running in full
3661          * dynticks mode. The check is racy by nature, but missing a tick or
3662          * having one too much is no big deal because the scheduler tick updates
3663          * statistics and checks timeslices in a time-independent way, regardless
3664          * of when exactly it is running.
3665          */
3666         if (!tick_nohz_tick_stopped_cpu(cpu))
3667                 goto out_requeue;
3668 
3669         rq_lock_irq(rq, &rf);
3670         curr = rq->curr;
3671         if (cpu_is_offline(cpu))
3672                 goto out_unlock;
3673 
3674         update_rq_clock(rq);
3675 
3676         if (!is_idle_task(curr)) {
3677                 /*
3678                  * Make sure the next tick runs within a reasonable
3679                  * amount of time.
3680                  */
3681                 delta = rq_clock_task(rq) - curr->se.exec_start;
3682                 WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
3683         }
3684         curr->sched_class->task_tick(rq, curr, 0);
3685 
3686         calc_load_nohz_remote(rq);
3687 out_unlock:
3688         rq_unlock_irq(rq, &rf);
3689 out_requeue:
3690 
3691         /*
3692          * Run the remote tick once per second (1Hz). This arbitrary
3693          * frequency is large enough to avoid overload but short enough
3694          * to keep scheduler internal stats reasonably up to date.  But
3695          * first update state to reflect hotplug activity if required.
3696          */
3697         os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING);
3698         WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE);
3699         if (os == TICK_SCHED_REMOTE_RUNNING)
3700                 queue_delayed_work(system_unbound_wq, dwork, HZ);
3701 }
3702 
3703 static void sched_tick_start(int cpu)
3704 {
3705         int os;
3706         struct tick_work *twork;
3707 
3708         if (housekeeping_cpu(cpu, HK_FLAG_TICK))
3709                 return;
3710 
3711         WARN_ON_ONCE(!tick_work_cpu);
3712 
3713         twork = per_cpu_ptr(tick_work_cpu, cpu);
3714         os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING);
3715         WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING);
3716         if (os == TICK_SCHED_REMOTE_OFFLINE) {
3717                 twork->cpu = cpu;
3718                 INIT_DELAYED_WORK(&twork->work, sched_tick_remote);
3719                 queue_delayed_work(system_unbound_wq, &twork->work, HZ);
3720         }
3721 }
3722 
3723 #ifdef CONFIG_HOTPLUG_CPU
3724 static void sched_tick_stop(int cpu)
3725 {
3726         struct tick_work *twork;
3727         int os;
3728 
3729         if (housekeeping_cpu(cpu, HK_FLAG_TICK))
3730                 return;
3731 
3732         WARN_ON_ONCE(!tick_work_cpu);
3733 
3734         twork = per_cpu_ptr(tick_work_cpu, cpu);
3735         /* There cannot be competing actions, but don't rely on stop-machine. */
3736         os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING);
3737         WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING);
3738         /* Don't cancel, as this would mess up the state machine. */
3739 }
3740 #endif /* CONFIG_HOTPLUG_CPU */
3741 
3742 int __init sched_tick_offload_init(void)
3743 {
3744         tick_work_cpu = alloc_percpu(struct tick_work);
3745         BUG_ON(!tick_work_cpu);
3746         return 0;
3747 }
3748 
3749 #else /* !CONFIG_NO_HZ_FULL */
3750 static inline void sched_tick_start(int cpu) { }
3751 static inline void sched_tick_stop(int cpu) { }
3752 #endif
3753 
3754 #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \
3755                                 defined(CONFIG_TRACE_PREEMPT_TOGGLE))
3756 /*
3757  * If the value passed in is equal to the current preempt count
3758  * then we just disabled preemption. Start timing the latency.
3759  */
3760 static inline void preempt_latency_start(int val)
3761 {
3762         if (preempt_count() == val) {
3763                 unsigned long ip = get_lock_parent_ip();
3764 #ifdef CONFIG_DEBUG_PREEMPT
3765                 current->preempt_disable_ip = ip;
3766 #endif
3767                 trace_preempt_off(CALLER_ADDR0, ip);
3768         }
3769 }
3770 
3771 void preempt_count_add(int val)
3772 {
3773 #ifdef CONFIG_DEBUG_PREEMPT
3774         /*
3775          * Underflow?
3776          */
3777         if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3778                 return;
3779 #endif
3780         __preempt_count_add(val);
3781 #ifdef CONFIG_DEBUG_PREEMPT
3782         /*
3783          * Spinlock count overflowing soon?
3784          */
3785         DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3786                                 PREEMPT_MASK - 10);
3787 #endif
3788         preempt_latency_start(val);
3789 }
3790 EXPORT_SYMBOL(preempt_count_add);
3791 NOKPROBE_SYMBOL(preempt_count_add);
3792 
3793 /*
3794  * If the value passed in equals to the current preempt count
3795  * then we just enabled preemption. Stop timing the latency.
3796  */
3797 static inline void preempt_latency_stop(int val)
3798 {
3799         if (preempt_count() == val)
3800                 trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
3801 }
3802 
3803 void preempt_count_sub(int val)
3804 {
3805 #ifdef CONFIG_DEBUG_PREEMPT
3806         /*
3807          * Underflow?
3808          */
3809         if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3810                 return;
3811         /*
3812          * Is the spinlock portion underflowing?
3813          */
3814         if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3815                         !(preempt_count() & PREEMPT_MASK)))
3816                 return;
3817 #endif
3818 
3819         preempt_latency_stop(val);
3820         __preempt_count_sub(val);
3821 }
3822 EXPORT_SYMBOL(preempt_count_sub);
3823 NOKPROBE_SYMBOL(preempt_count_sub);
3824 
3825 #else
3826 static inline void preempt_latency_start(int val) { }
3827 static inline void preempt_latency_stop(int val) { }
3828 #endif
3829 
3830 static inline unsigned long get_preempt_disable_ip(struct task_struct *p)
3831 {
3832 #ifdef CONFIG_DEBUG_PREEMPT
3833         return p->preempt_disable_ip;
3834 #else
3835         return 0;
3836 #endif
3837 }
3838 
3839 /*
3840  * Print scheduling while atomic bug:
3841  */
3842 static noinline void __schedule_bug(struct task_struct *prev)
3843 {
3844         /* Save this before calling printk(), since that will clobber it */
3845         unsigned long preempt_disable_ip = get_preempt_disable_ip(current);
3846 
3847         if (oops_in_progress)
3848                 return;
3849 
3850         printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3851                 prev->comm, prev->pid, preempt_count());
3852 
3853         debug_show_held_locks(prev);
3854         print_modules();
3855         if (irqs_disabled())
3856                 print_irqtrace_events(prev);
3857         if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
3858             && in_atomic_preempt_off()) {
3859                 pr_err("Preemption disabled at:");
3860                 print_ip_sym(preempt_disable_ip);
3861                 pr_cont("\n");
3862         }
3863         if (panic_on_warn)
3864                 panic("scheduling while atomic\n");
3865 
3866         dump_stack();
3867         add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
3868 }
3869 
3870 /*
3871  * Various schedule()-time debugging checks and statistics:
3872  */
3873 static inline void schedule_debug(struct task_struct *prev, bool preempt)
3874 {
3875 #ifdef CONFIG_SCHED_STACK_END_CHECK
3876         if (task_stack_end_corrupted(prev))
3877                 panic("corrupted stack end detected inside scheduler\n");
3878 #endif
3879 
3880 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
3881         if (!preempt && prev->state && prev->non_block_count) {
3882                 printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n",
3883                         prev->comm, prev->pid, prev->non_block_count);
3884                 dump_stack();
3885                 add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
3886         }
3887 #endif
3888 
3889         if (unlikely(in_atomic_preempt_off())) {
3890                 __schedule_bug(prev);
3891                 preempt_count_set(PREEMPT_DISABLED);
3892         }
3893         rcu_sleep_check();
3894 
3895         profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3896 
3897         schedstat_inc(this_rq()->sched_count);
3898 }
3899 
3900 /*
3901  * Pick up the highest-prio task:
3902  */
3903 static inline struct task_struct *
3904 pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
3905 {
3906         const struct sched_class *class;
3907         struct task_struct *p;
3908 
3909         /*
3910          * Optimization: we know that if all tasks are in the fair class we can
3911          * call that function directly, but only if the @prev task wasn't of a
3912          * higher scheduling class, because otherwise those loose the
3913          * opportunity to pull in more work from other CPUs.
3914          */
3915         if (likely((prev->sched_class == &idle_sched_class ||
3916                     prev->sched_class == &fair_sched_class) &&
3917                    rq->nr_running == rq->cfs.h_nr_running)) {
3918 
3919                 p = fair_sched_class.pick_next_task(rq, prev, rf);
3920                 if (unlikely(p == RETRY_TASK))
3921                         goto restart;
3922 
3923                 /* Assumes fair_sched_class->next == idle_sched_class */
3924                 if (unlikely(!p))
3925                         p = idle_sched_class.pick_next_task(rq, prev, rf);
3926 
3927                 return p;
3928         }
3929 
3930 restart:
3931 #ifdef CONFIG_SMP
3932         /*
3933          * We must do the balancing pass before put_next_task(), such
3934          * that when we release the rq->lock the task is in the same
3935          * state as before we took rq->lock.
3936          *
3937          * We can terminate the balance pass as soon as we know there is
3938          * a runnable task of @class priority or higher.
3939          */
3940         for_class_range(class, prev->sched_class, &idle_sched_class) {
3941                 if (class->balance(rq, prev, rf))
3942                         break;
3943         }
3944 #endif
3945 
3946         put_prev_task(rq, prev);
3947 
3948         for_each_class(class) {
3949                 p = class->pick_next_task(rq, NULL, NULL);
3950                 if (p)
3951                         return p;
3952         }
3953 
3954         /* The idle class should always have a runnable task: */
3955         BUG();
3956 }
3957 
3958 /*
3959  * __schedule() is the main scheduler function.
3960  *
3961  * The main means of driving the scheduler and thus entering this function are:
3962  *
3963  *   1. Explicit blocking: mutex, semaphore, waitqueue, etc.
3964  *
3965  *   2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return
3966  *      paths. For example, see arch/x86/entry_64.S.
3967  *
3968  *      To drive preemption between tasks, the scheduler sets the flag in timer
3969  *      interrupt handler scheduler_tick().
3970  *
3971  *   3. Wakeups don't really cause entry into schedule(). They add a
3972  *      task to the run-queue and that's it.
3973  *
3974  *      Now, if the new task added to the run-queue preempts the current
3975  *      task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets
3976  *      called on the nearest possible occasion:
3977  *
3978  *       - If the kernel is preemptible (CONFIG_PREEMPTION=y):
3979  *
3980  *         - in syscall or exception context, at the next outmost
3981  *           preempt_enable(). (this might be as soon as the wake_up()'s
3982  *           spin_unlock()!)
3983  *
3984  *         - in IRQ context, return from interrupt-handler to
3985  *           preemptible context
3986  *
3987  *       - If the kernel is not preemptible (CONFIG_PREEMPTION is not set)
3988  *         then at the next:
3989  *
3990  *          - cond_resched() call
3991  *          - explicit schedule() call
3992  *          - return from syscall or exception to user-space
3993  *          - return from interrupt-handler to user-space
3994  *
3995  * WARNING: must be called with preemption disabled!
3996  */
3997 static void __sched notrace __schedule(bool preempt)
3998 {
3999         struct task_struct *prev, *next;
4000         unsigned long *switch_count;
4001         struct rq_flags rf;
4002         struct rq *rq;
4003         int cpu;
4004 
4005         cpu = smp_processor_id();
4006         rq = cpu_rq(cpu);
4007         prev = rq->curr;
4008 
4009         schedule_debug(prev, preempt);
4010 
4011         if (sched_feat(HRTICK))
4012                 hrtick_clear(rq);
4013 
4014         local_irq_disable();
4015         rcu_note_context_switch(preempt);
4016 
4017         /*
4018          * Make sure that signal_pending_state()->signal_pending() below
4019          * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE)
4020          * done by the caller to avoid the race with signal_wake_up().
4021          *
4022          * The membarrier system call requires a full memory barrier
4023          * after coming from user-space, before storing to rq->curr.
4024          */
4025         rq_lock(rq, &rf);
4026         smp_mb__after_spinlock();
4027 
4028         /* Promote REQ to ACT */
4029         rq->clock_update_flags <<= 1;
4030         update_rq_clock(rq);
4031 
4032         switch_count = &prev->nivcsw;
4033         if (!preempt && prev->state) {
4034                 if (signal_pending_state(prev->state, prev)) {
4035                         prev->state = TASK_RUNNING;
4036                 } else {
4037                         deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);
4038 
4039                         if (prev->in_iowait) {
4040                                 atomic_inc(&rq->nr_iowait);
4041                                 delayacct_blkio_start();
4042                         }
4043                 }
4044                 switch_count = &prev->nvcsw;
4045         }
4046 
4047         next = pick_next_task(rq, prev, &rf);
4048         clear_tsk_need_resched(prev);
4049         clear_preempt_need_resched();
4050 
4051         if (likely(prev != next)) {
4052                 rq->nr_switches++;
4053                 /*
4054                  * RCU users of rcu_dereference(rq->curr) may not see
4055                  * changes to task_struct made by pick_next_task().
4056                  */
4057                 RCU_INIT_POINTER(rq->curr, next);
4058                 /*
4059                  * The membarrier system call requires each architecture
4060                  * to have a full memory barrier after updating
4061                  * rq->curr, before returning to user-space.
4062                  *
4063                  * Here are the schemes providing that barrier on the
4064                  * various architectures:
4065                  * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC.
4066                  *   switch_mm() rely on membarrier_arch_switch_mm() on PowerPC.
4067                  * - finish_lock_switch() for weakly-ordered
4068                  *   architectures where spin_unlock is a full barrier,
4069                  * - switch_to() for arm64 (weakly-ordered, spin_unlock
4070                  *   is a RELEASE barrier),
4071                  */
4072                 ++*switch_count;
4073 
4074                 trace_sched_switch(preempt, prev, next);
4075 
4076                 /* Also unlocks the rq: */
4077                 rq = context_switch(rq, prev, next, &rf);
4078         } else {
4079                 rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
4080                 rq_unlock_irq(rq, &rf);
4081         }
4082 
4083         balance_callback(rq);
4084 }
4085 
4086 void __noreturn do_task_dead(void)
4087 {
4088         /* Causes final put_task_struct in finish_task_switch(): */
4089         set_special_state(TASK_DEAD);
4090 
4091         /* Tell freezer to ignore us: */
4092         current->flags |= PF_NOFREEZE;
4093 
4094         __schedule(false);
4095         BUG();
4096 
4097         /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */
4098         for (;;)
4099                 cpu_relax();
4100 }
4101 
4102 static inline void sched_submit_work(struct task_struct *tsk)
4103 {
4104         if (!tsk->state)
4105                 return;
4106 
4107         /*
4108          * If a worker went to sleep, notify and ask workqueue whether
4109          * it wants to wake up a task to maintain concurrency.
4110          * As this function is called inside the schedule() context,
4111          * we disable preemption to avoid it calling schedule() again
4112          * in the possible wakeup of a kworker.
4113          */
4114         if (tsk->flags & PF_WQ_WORKER) {
4115                 preempt_disable();
4116                 wq_worker_sleeping(tsk);
4117                 preempt_enable_no_resched();
4118         }
4119 
4120         if (tsk_is_pi_blocked(tsk))
4121                 return;
4122 
4123         /*
4124          * If we are going to sleep and we have plugged IO queued,
4125          * make sure to submit it to avoid deadlocks.
4126          */
4127         if (blk_needs_flush_plug(tsk))
4128                 blk_schedule_flush_plug(tsk);
4129 }
4130 
4131 static void sched_update_worker(struct task_struct *tsk)
4132 {
4133         if (tsk->flags & PF_WQ_WORKER)
4134                 wq_worker_running(tsk);
4135 }
4136 
4137 asmlinkage __visible void __sched schedule(void)
4138 {
4139         struct task_struct *tsk = current;
4140 
4141         sched_submit_work(tsk);
4142         do {
4143                 preempt_disable();
4144                 __schedule(false);
4145                 sched_preempt_enable_no_resched();
4146         } while (need_resched());
4147         sched_update_worker(tsk);
4148 }
4149 EXPORT_SYMBOL(schedule);
4150 
4151 /*
4152  * synchronize_rcu_tasks() makes sure that no task is stuck in preempted
4153  * state (have scheduled out non-voluntarily) by making sure that all
4154  * tasks have either left the run queue or have gone into user space.
4155  * As idle tasks do not do either, they must not ever be preempted
4156  * (schedule out non-voluntarily).
4157  *
4158  * schedule_idle() is similar to schedule_preempt_disable() except that it
4159  * never enables preemption because it does not call sched_submit_work().
4160  */
4161 void __sched schedule_idle(void)
4162 {
4163         /*
4164          * As this skips calling sched_submit_work(), which the idle task does
4165          * regardless because that function is a nop when the task is in a
4166          * TASK_RUNNING state, make sure this isn't used someplace that the
4167          * current task can be in any other state. Note, idle is always in the
4168          * TASK_RUNNING state.
4169          */
4170         WARN_ON_ONCE(current->state);
4171         do {
4172                 __schedule(false);
4173         } while (need_resched());
4174 }
4175 
4176 #ifdef CONFIG_CONTEXT_TRACKING
4177 asmlinkage __visible void __sched schedule_user(void)
4178 {
4179         /*
4180          * If we come here after a random call to set_need_resched(),
4181          * or we have been woken up remotely but the IPI has not yet arrived,
4182          * we haven't yet exited the RCU idle mode. Do it here manually until
4183          * we find a better solution.
4184          *
4185          * NB: There are buggy callers of this function.  Ideally we
4186          * should warn if prev_state != CONTEXT_USER, but that will trigger
4187          * too frequently to make sense yet.
4188          */
4189         enum ctx_state prev_state = exception_enter();
4190         schedule();
4191         exception_exit(prev_state);
4192 }
4193 #endif
4194 
4195 /**
4196  * schedule_preempt_disabled - called with preemption disabled
4197  *
4198  * Returns with preemption disabled. Note: preempt_count must be 1
4199  */
4200 void __sched schedule_preempt_disabled(void)
4201 {
4202         sched_preempt_enable_no_resched();
4203         schedule();
4204         preempt_disable();
4205 }
4206 
4207 static void __sched notrace preempt_schedule_common(void)
4208 {
4209         do {
4210                 /*
4211                  * Because the function tracer can trace preempt_count_sub()
4212                  * and it also uses preempt_enable/disable_notrace(), if
4213                  * NEED_RESCHED is set, the preempt_enable_notrace() called
4214                  * by the function tracer will call this function again and
4215                  * cause infinite recursion.
4216                  *
4217                  * Preemption must be disabled here before the function
4218                  * tracer can trace. Break up preempt_disable() into two
4219                  * calls. One to disable preemption without fear of being
4220                  * traced. The other to still record the preemption latency,
4221                  * which can also be traced by the function tracer.
4222                  */
4223                 preempt_disable_notrace();
4224                 preempt_latency_start(1);
4225                 __schedule(true);
4226                 preempt_latency_stop(1);
4227                 preempt_enable_no_resched_notrace();
4228 
4229                 /*
4230                  * Check again in case we missed a preemption opportunity
4231                  * between schedule and now.
4232                  */
4233         } while (need_resched());
4234 }
4235 
4236 #ifdef CONFIG_PREEMPTION
4237 /*
4238  * This is the entry point to schedule() from in-kernel preemption
4239  * off of preempt_enable.
4240  */
4241 asmlinkage __visible void __sched notrace preempt_schedule(void)
4242 {
4243         /*
4244          * If there is a non-zero preempt_count or interrupts are disabled,
4245          * we do not want to preempt the current task. Just return..
4246          */
4247         if (likely(!preemptible()))
4248                 return;
4249 
4250         preempt_schedule_common();
4251 }
4252 NOKPROBE_SYMBOL(preempt_schedule);
4253 EXPORT_SYMBOL(preempt_schedule);
4254 
4255 /**
4256  * preempt_schedule_notrace - preempt_schedule called by tracing
4257  *
4258  * The tracing infrastructure uses preempt_enable_notrace to prevent
4259  * recursion and tracing preempt enabling caused by the tracing
4260  * infrastructure itself. But as tracing can happen in areas coming
4261  * from userspace or just about to enter userspace, a preempt enable
4262  * can occur before user_exit() is called. This will cause the scheduler
4263  * to be called when the system is still in usermode.
4264  *
4265  * To prevent this, the preempt_enable_notrace will use this function
4266  * instead of preempt_schedule() to exit user context if needed before
4267  * calling the scheduler.
4268  */
4269 asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
4270 {
4271         enum ctx_state prev_ctx;
4272 
4273         if (likely(!preemptible()))
4274                 return;
4275 
4276         do {
4277                 /*
4278                  * Because the function tracer can trace preempt_count_sub()
4279                  * and it also uses preempt_enable/disable_notrace(), if
4280                  * NEED_RESCHED is set, the preempt_enable_notrace() called
4281                  * by the function tracer will call this function again and
4282                  * cause infinite recursion.
4283                  *
4284                  * Preemption must be disabled here before the function
4285                  * tracer can trace. Break up preempt_disable() into two
4286                  * calls. One to disable preemption without fear of being
4287                  * traced. The other to still record the preemption latency,
4288                  * which can also be traced by the function tracer.
4289                  */
4290                 preempt_disable_notrace();
4291                 preempt_latency_start(1);
4292                 /*
4293                  * Needs preempt disabled in case user_exit() is traced
4294                  * and the tracer calls preempt_enable_notrace() causing
4295                  * an infinite recursion.
4296                  */
4297                 prev_ctx = exception_enter();
4298                 __schedule(true);
4299                 exception_exit(prev_ctx);
4300 
4301                 preempt_latency_stop(1);
4302                 preempt_enable_no_resched_notrace();
4303         } while (need_resched());
4304 }
4305 EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
4306 
4307 #endif /* CONFIG_PREEMPTION */
4308 
4309 /*
4310  * This is the entry point to schedule() from kernel preemption
4311  * off of irq context.
4312  * Note, that this is called and return with irqs disabled. This will
4313  * protect us against recursive calling from irq.
4314  */
4315 asmlinkage __visible void __sched preempt_schedule_irq(void)
4316 {
4317         enum ctx_state prev_state;
4318 
4319         /* Catch callers which need to be fixed */
4320         BUG_ON(preempt_count() || !irqs_disabled());
4321 
4322         prev_state = exception_enter();
4323 
4324         do {
4325                 preempt_disable();
4326                 local_irq_enable();
4327                 __schedule(true);
4328                 local_irq_disable();
4329                 sched_preempt_enable_no_resched();
4330         } while (need_resched());
4331 
4332         exception_exit(prev_state);
4333 }
4334 
4335 int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags,
4336                           void *key)
4337 {
4338         return try_to_wake_up(curr->private, mode, wake_flags);
4339 }
4340 EXPORT_SYMBOL(default_wake_function);
4341 
4342 #ifdef CONFIG_RT_MUTEXES
4343 
4344 static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
4345 {
4346         if (pi_task)
4347                 prio = min(prio, pi_task->prio);
4348 
4349         return prio;
4350 }
4351 
4352 static inline int rt_effective_prio(struct task_struct *p, int prio)
4353 {
4354         struct task_struct *pi_task = rt_mutex_get_top_task(p);
4355 
4356         return __rt_effective_prio(pi_task, prio);
4357 }
4358 
4359 /*
4360  * rt_mutex_setprio - set the current priority of a task
4361  * @p: task to boost
4362  * @pi_task: donor task
4363  *
4364  * This function changes the 'effective' priority of a task. It does
4365  * not touch ->normal_prio like __setscheduler().
4366  *
4367  * Used by the rt_mutex code to implement priority inheritance
4368  * logic. Call site only calls if the priority of the task changed.
4369  */
4370 void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
4371 {
4372         int prio, oldprio, queued, running, queue_flag =
4373                 DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4374         const struct sched_class *prev_class;
4375         struct rq_flags rf;
4376         struct rq *rq;
4377 
4378         /* XXX used to be waiter->prio, not waiter->task->prio */
4379         prio = __rt_effective_prio(pi_task, p->normal_prio);
4380 
4381         /*
4382          * If nothing changed; bail early.
4383          */
4384         if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio))
4385                 return;
4386 
4387         rq = __task_rq_lock(p, &rf);
4388         update_rq_clock(rq);
4389         /*
4390          * Set under pi_lock && rq->lock, such that the value can be used under
4391          * either lock.
4392          *
4393          * Note that there is loads of tricky to make this pointer cache work
4394          * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to
4395          * ensure a task is de-boosted (pi_task is set to NULL) before the
4396          * task is allowed to run again (and can exit). This ensures the pointer
4397          * points to a blocked task -- which guaratees the task is present.
4398          */
4399         p->pi_top_task = pi_task;
4400 
4401         /*
4402          * For FIFO/RR we only need to set prio, if that matches we're done.
4403          */
4404         if (prio == p->prio && !dl_prio(prio))
4405                 goto out_unlock;
4406 
4407         /*
4408          * Idle task boosting is a nono in general. There is one
4409          * exception, when PREEMPT_RT and NOHZ is active:
4410          *
4411          * The idle task calls get_next_timer_interrupt() and holds
4412          * the timer wheel base->lock on the CPU and another CPU wants
4413          * to access the timer (probably to cancel it). We can safely
4414          * ignore the boosting request, as the idle CPU runs this code
4415          * with interrupts disabled and will complete the lock
4416          * protected section without being interrupted. So there is no
4417          * real need to boost.
4418          */
4419         if (unlikely(p == rq->idle)) {
4420                 WARN_ON(p != rq->curr);
4421                 WARN_ON(p->pi_blocked_on);
4422                 goto out_unlock;
4423         }
4424 
4425         trace_sched_pi_setprio(p, pi_task);
4426         oldprio = p->prio;
4427 
4428         if (oldprio == prio)
4429                 queue_flag &= ~DEQUEUE_MOVE;
4430 
4431         prev_class = p->sched_class;
4432         queued = task_on_rq_queued(p);
4433         running = task_current(rq, p);
4434         if (queued)
4435                 dequeue_task(rq, p, queue_flag);
4436         if (running)
4437                 put_prev_task(rq, p);
4438 
4439         /*
4440          * Boosting condition are:
4441          * 1. -rt task is running and holds mutex A
4442          *      --> -dl task blocks on mutex A
4443          *
4444          * 2. -dl task is running and holds mutex A
4445          *      --> -dl task blocks on mutex A and could preempt the
4446          *          running task
4447          */
4448         if (dl_prio(prio)) {
4449                 if (!dl_prio(p->normal_prio) ||
4450                     (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
4451                         p->dl.dl_boosted = 1;
4452                         queue_flag |= ENQUEUE_REPLENISH;
4453                 } else
4454                         p->dl.dl_boosted = 0;
4455                 p->sched_class = &dl_sched_class;
4456         } else if (rt_prio(prio)) {
4457                 if (dl_prio(oldprio))
4458                         p->dl.dl_boosted = 0;
4459                 if (oldprio < prio)
4460                         queue_flag |= ENQUEUE_HEAD;
4461                 p->sched_class = &rt_sched_class;
4462         } else {
4463                 if (dl_prio(oldprio))
4464                         p->dl.dl_boosted = 0;
4465                 if (rt_prio(oldprio))
4466                         p->rt.timeout = 0;
4467                 p->sched_class = &fair_sched_class;
4468         }
4469 
4470         p->prio = prio;
4471 
4472         if (queued)
4473                 enqueue_task(rq, p, queue_flag);
4474         if (running)
4475                 set_next_task(rq, p);
4476 
4477         check_class_changed(rq, p, prev_class, oldprio);
4478 out_unlock:
4479         /* Avoid rq from going away on us: */
4480         preempt_disable();
4481         __task_rq_unlock(rq, &rf);
4482 
4483         balance_callback(rq);
4484         preempt_enable();
4485 }
4486 #else
4487 static inline int rt_effective_prio(struct task_struct *p, int prio)
4488 {
4489         return prio;
4490 }
4491 #endif
4492 
4493 void set_user_nice(struct task_struct *p, long nice)
4494 {
4495         bool queued, running;
4496         int old_prio, delta;
4497         struct rq_flags rf;
4498         struct rq *rq;
4499 
4500         if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
4501                 return;
4502         /*
4503          * We have to be careful, if called from sys_setpriority(),
4504          * the task might be in the middle of scheduling on another CPU.
4505          */
4506         rq = task_rq_lock(p, &rf);
4507         update_rq_clock(rq);
4508 
4509         /*
4510          * The RT priorities are set via sched_setscheduler(), but we still
4511          * allow the 'normal' nice value to be set - but as expected
4512          * it wont have any effect on scheduling until the task is
4513          * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
4514          */
4515         if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
4516                 p->static_prio = NICE_TO_PRIO(nice);
4517                 goto out_unlock;
4518         }
4519         queued = task_on_rq_queued(p);
4520         running = task_current(rq, p);
4521         if (queued)
4522                 dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
4523         if (running)
4524                 put_prev_task(rq, p);
4525 
4526         p->static_prio = NICE_TO_PRIO(nice);
4527         set_load_weight(p, true);
4528         old_prio = p->prio;
4529         p->prio = effective_prio(p);
4530         delta = p->prio - old_prio;
4531 
4532         if (queued) {
4533                 enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
4534                 /*
4535                  * If the task increased its priority or is running and
4536                  * lowered its priority, then reschedule its CPU:
4537                  */
4538                 if (delta < 0 || (delta > 0 && task_running(rq, p)))
4539                         resched_curr(rq);
4540         }
4541         if (running)
4542                 set_next_task(rq, p);
4543 out_unlock:
4544         task_rq_unlock(rq, p, &rf);
4545 }
4546 EXPORT_SYMBOL(set_user_nice);
4547 
4548 /*
4549  * can_nice - check if a task can reduce its nice value
4550  * @p: task
4551  * @nice: nice value
4552  */
4553 int can_nice(const struct task_struct *p, const int nice)
4554 {
4555         /* Convert nice value [19,-20] to rlimit style value [1,40]: */
4556         int nice_rlim = nice_to_rlimit(nice);
4557 
4558         return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
4559                 capable(CAP_SYS_NICE));
4560 }
4561 
4562 #ifdef __ARCH_WANT_SYS_NICE
4563 
4564 /*
4565  * sys_nice - change the priority of the current process.
4566  * @increment: priority increment
4567  *
4568  * sys_setpriority is a more generic, but much slower function that
4569  * does similar things.
4570  */
4571 SYSCALL_DEFINE1(nice, int, increment)
4572 {
4573         long nice, retval;
4574 
4575         /*
4576          * Setpriority might change our priority at the same moment.
4577          * We don't have to worry. Conceptually one call occurs first
4578          * and we have a single winner.
4579          */
4580         increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
4581         nice = task_nice(current) + increment;
4582 
4583         nice = clamp_val(nice, MIN_NICE, MAX_NICE);
4584         if (increment < 0 && !can_nice(current, nice))
4585                 return -EPERM;
4586 
4587         retval = security_task_setnice(current, nice);
4588         if (retval)
4589                 return retval;
4590 
4591         set_user_nice(current, nice);
4592         return 0;
4593 }
4594 
4595 #endif
4596 
4597 /**
4598  * task_prio - return the priority value of a given task.
4599  * @p: the task in question.
4600  *
4601  * Return: The priority value as seen by users in /proc.
4602  * RT tasks are offset by -200. Normal tasks are centered
4603  * around 0, value goes from -16 to +15.
4604  */
4605 int task_prio(const struct task_struct *p)
4606 {
4607         return p->prio - MAX_RT_PRIO;
4608 }
4609 
4610 /**
4611  * idle_cpu - is a given CPU idle currently?
4612  * @cpu: the processor in question.
4613  *
4614  * Return: 1 if the CPU is currently idle. 0 otherwise.
4615  */
4616 int idle_cpu(int cpu)
4617 {
4618         struct rq *rq = cpu_rq(cpu);
4619 
4620         if (rq->curr != rq->idle)
4621                 return 0;
4622 
4623         if (rq->nr_running)
4624                 return 0;
4625 
4626 #ifdef CONFIG_SMP
4627         if (!llist_empty(&rq->wake_list))
4628                 return 0;
4629 #endif
4630 
4631         return 1;
4632 }
4633 
4634 /**
4635  * available_idle_cpu - is a given CPU idle for enqueuing work.
4636  * @cpu: the CPU in question.
4637  *
4638  * Return: 1 if the CPU is currently idle. 0 otherwise.
4639  */
4640 int available_idle_cpu(int cpu)
4641 {
4642         if (!idle_cpu(cpu))
4643                 return 0;
4644 
4645         if (vcpu_is_preempted(cpu))
4646                 return 0;
4647 
4648         return 1;
4649 }
4650 
4651 /**
4652  * idle_task - return the idle task for a given CPU.
4653  * @cpu: the processor in question.
4654  *
4655  * Return: The idle task for the CPU @cpu.
4656  */
4657 struct task_struct *idle_task(int cpu)
4658 {
4659         return cpu_rq(cpu)->idle;
4660 }
4661 
4662 /**
4663  * find_process_by_pid - find a process with a matching PID value.
4664  * @pid: the pid in question.
4665  *
4666  * The task of @pid, if found. %NULL otherwise.
4667  */
4668 static struct task_struct *find_process_by_pid(pid_t pid)
4669 {
4670         return pid ? find_task_by_vpid(pid) : current;
4671 }
4672 
4673 /*
4674  * sched_setparam() passes in -1 for its policy, to let the functions
4675  * it calls know not to change it.
4676  */
4677 #define SETPARAM_POLICY -1
4678 
4679 static void __setscheduler_params(struct task_struct *p,
4680                 const struct sched_attr *attr)
4681 {
4682         int policy = attr->sched_policy;
4683 
4684         if (policy == SETPARAM_POLICY)
4685                 policy = p->policy;
4686 
4687         p->policy = policy;
4688 
4689         if (dl_policy(policy))
4690                 __setparam_dl(p, attr);
4691         else if (fair_policy(policy))
4692                 p->static_prio = NICE_TO_PRIO(attr->sched_nice);
4693 
4694         /*
4695          * __sched_setscheduler() ensures attr->sched_priority == 0 when
4696          * !rt_policy. Always setting this ensures that things like
4697          * getparam()/getattr() don't report silly values for !rt tasks.
4698          */
4699         p->rt_priority = attr->sched_priority;
4700         p->normal_prio = normal_prio(p);
4701         set_load_weight(p, true);
4702 }
4703 
4704 /* Actually do priority change: must hold pi & rq lock. */
4705 static void __setscheduler(struct rq *rq, struct task_struct *p,
4706                            const struct sched_attr *attr, bool keep_boost)
4707 {
4708         /*
4709          * If params can't change scheduling class changes aren't allowed
4710          * either.
4711          */
4712         if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)
4713                 return;
4714 
4715         __setscheduler_params(p, attr);
4716 
4717         /*
4718          * Keep a potential priority boosting if called from
4719          * sched_setscheduler().
4720          */
4721         p->prio = normal_prio(p);
4722         if (keep_boost)
4723                 p->prio = rt_effective_prio(p, p->prio);
4724 
4725         if (dl_prio(p->prio))
4726                 p->sched_class = &dl_sched_class;
4727         else if (rt_prio(p->prio))
4728                 p->sched_class = &rt_sched_class;
4729         else
4730                 p->sched_class = &fair_sched_class;
4731 }
4732 
4733 /*
4734  * Check the target process has a UID that matches the current process's:
4735  */
4736 static bool check_same_owner(struct task_struct *p)
4737 {
4738         const struct cred *cred = current_cred(), *pcred;
4739         bool match;
4740 
4741         rcu_read_lock();
4742         pcred = __task_cred(p);
4743         match = (uid_eq(cred->euid, pcred->euid) ||
4744                  uid_eq(cred->euid, pcred->uid));
4745         rcu_read_unlock();
4746         return match;
4747 }
4748 
4749 static int __sched_setscheduler(struct task_struct *p,
4750                                 const struct sched_attr *attr,
4751                                 bool user, bool pi)
4752 {
4753         int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
4754                       MAX_RT_PRIO - 1 - attr->sched_priority;
4755         int retval, oldprio, oldpolicy = -1, queued, running;
4756         int new_effective_prio, policy = attr->sched_policy;
4757         const struct sched_class *prev_class;
4758         struct rq_flags rf;
4759         int reset_on_fork;
4760         int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4761         struct rq *rq;
4762 
4763         /* The pi code expects interrupts enabled */
4764         BUG_ON(pi && in_interrupt());
4765 recheck:
4766         /* Double check policy once rq lock held: */
4767         if (policy < 0) {
4768                 reset_on_fork = p->sched_reset_on_fork;
4769                 policy = oldpolicy = p->policy;
4770         } else {
4771                 reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4772 
4773                 if (!valid_policy(policy))
4774                         return -EINVAL;
4775         }
4776 
4777         if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
4778                 return -EINVAL;
4779 
4780         /*
4781          * Valid priorities for SCHED_FIFO and SCHED_RR are
4782          * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4783          * SCHED_BATCH and SCHED_IDLE is 0.
4784          */
4785         if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4786             (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
4787                 return -EINVAL;
4788         if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
4789             (rt_policy(policy) != (attr->sched_priority != 0)))
4790                 return -EINVAL;
4791 
4792         /*
4793          * Allow unprivileged RT tasks to decrease priority:
4794          */
4795         if (user && !capable(CAP_SYS_NICE)) {
4796                 if (fair_policy(policy)) {
4797                         if (attr->sched_nice < task_nice(p) &&
4798                             !can_nice(p, attr->sched_nice))
4799                                 return -EPERM;
4800                 }
4801 
4802                 if (rt_policy(policy)) {
4803                         unsigned long rlim_rtprio =
4804                                         task_rlimit(p, RLIMIT_RTPRIO);
4805 
4806                         /* Can't set/change the rt policy: */
4807                         if (policy != p->policy && !rlim_rtprio)
4808                                 return -EPERM;
4809 
4810                         /* Can't increase priority: */
4811                         if (attr->sched_priority > p->rt_priority &&
4812                             attr->sched_priority > rlim_rtprio)
4813                                 return -EPERM;
4814                 }
4815 
4816                  /*
4817                   * Can't set/change SCHED_DEADLINE policy at all for now
4818                   * (safest behavior); in the future we would like to allow
4819                   * unprivileged DL tasks to increase their relative deadline
4820                   * or reduce their runtime (both ways reducing utilization)
4821                   */
4822                 if (dl_policy(policy))
4823                         return -EPERM;
4824 
4825                 /*
4826                  * Treat SCHED_IDLE as nice 20. Only allow a switch to
4827                  * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
4828                  */
4829                 if (task_has_idle_policy(p) && !idle_policy(policy)) {
4830                         if (!can_nice(p, task_nice(p)))
4831                                 return -EPERM;
4832                 }
4833 
4834                 /* Can't change other user's priorities: */
4835                 if (!check_same_owner(p))
4836                         return -EPERM;
4837 
4838                 /* Normal users shall not reset the sched_reset_on_fork flag: */
4839                 if (p->sched_reset_on_fork && !reset_on_fork)
4840                         return -EPERM;
4841         }
4842 
4843         if (user) {
4844                 if (attr->sched_flags & SCHED_FLAG_SUGOV)
4845                         return -EINVAL;
4846 
4847                 retval = security_task_setscheduler(p);
4848                 if (retval)
4849                         return retval;
4850         }
4851 
4852         /* Update task specific "requested" clamps */
4853         if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) {
4854                 retval = uclamp_validate(p, attr);
4855                 if (retval)
4856                         return retval;
4857         }
4858 
4859         if (pi)
4860                 cpuset_read_lock();
4861 
4862         /*
4863          * Make sure no PI-waiters arrive (or leave) while we are
4864          * changing the priority of the task:
4865          *
4866          * To be able to change p->policy safely, the appropriate
4867          * runqueue lock must be held.
4868          */
4869         rq = task_rq_lock(p, &rf);
4870         update_rq_clock(rq);
4871 
4872         /*
4873          * Changing the policy of the stop threads its a very bad idea:
4874          */
4875         if (p == rq->stop) {
4876                 retval = -EINVAL;
4877                 goto unlock;
4878         }
4879 
4880         /*
4881          * If not changing anything there's no need to proceed further,
4882          * but store a possible modification of reset_on_fork.
4883          */
4884         if (unlikely(policy == p->policy)) {
4885                 if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4886                         goto change;
4887                 if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
4888                         goto change;
4889                 if (dl_policy(policy) && dl_param_changed(p, attr))
4890                         goto change;
4891                 if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)
4892                         goto change;
4893 
4894                 p->sched_reset_on_fork = reset_on_fork;
4895                 retval = 0;
4896                 goto unlock;
4897         }
4898 change:
4899 
4900         if (user) {
4901 #ifdef CONFIG_RT_GROUP_SCHED
4902                 /*
4903                  * Do not allow realtime tasks into groups that have no runtime
4904                  * assigned.
4905                  */
4906                 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4907                                 task_group(p)->rt_bandwidth.rt_runtime == 0 &&
4908                                 !task_group_is_autogroup(task_group(p))) {
4909                         retval = -EPERM;
4910                         goto unlock;
4911                 }
4912 #endif
4913 #ifdef CONFIG_SMP
4914                 if (dl_bandwidth_enabled() && dl_policy(policy) &&
4915                                 !(attr->sched_flags & SCHED_FLAG_SUGOV)) {
4916                         cpumask_t *span = rq->rd->span;
4917 
4918                         /*
4919                          * Don't allow tasks with an affinity mask smaller than
4920                          * the entire root_domain to become SCHED_DEADLINE. We
4921                          * will also fail if there's no bandwidth available.
4922                          */
4923                         if (!cpumask_subset(span, p->cpus_ptr) ||
4924                             rq->rd->dl_bw.bw == 0) {
4925                                 retval = -EPERM;
4926                                 goto unlock;
4927                         }
4928                 }
4929 #endif
4930         }
4931 
4932         /* Re-check policy now with rq lock held: */
4933         if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4934                 policy = oldpolicy = -1;
4935                 task_rq_unlock(rq, p, &rf);
4936                 if (pi)
4937                         cpuset_read_unlock();
4938                 goto recheck;
4939         }
4940 
4941         /*
4942          * If setscheduling to SCHED_DEADLINE (or changing the parameters
4943          * of a SCHED_DEADLINE task) we need to check if enough bandwidth
4944          * is available.
4945          */
4946         if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
4947                 retval = -EBUSY;
4948                 goto unlock;
4949         }
4950 
4951         p->sched_reset_on_fork = reset_on_fork;
4952         oldprio = p->prio;
4953 
4954         if (pi) {
4955                 /*
4956                  * Take priority boosted tasks into account. If the new
4957                  * effective priority is unchanged, we just store the new
4958                  * normal parameters and do not touch the scheduler class and
4959                  * the runqueue. This will be done when the task deboost
4960                  * itself.
4961                  */
4962                 new_effective_prio = rt_effective_prio(p, newprio);
4963                 if (new_effective_prio == oldprio)
4964                         queue_flags &= ~DEQUEUE_MOVE;
4965         }
4966 
4967         queued = task_on_rq_queued(p);
4968         running = task_current(rq, p);
4969         if (queued)
4970                 dequeue_task(rq, p, queue_flags);
4971         if (running)
4972                 put_prev_task(rq, p);
4973 
4974         prev_class = p->sched_class;
4975 
4976         __setscheduler(rq, p, attr, pi);
4977         __setscheduler_uclamp(p, attr);
4978 
4979         if (queued) {
4980                 /*
4981                  * We enqueue to tail when the priority of a task is
4982                  * increased (user space view).
4983                  */
4984                 if (oldprio < p->prio)
4985                         queue_flags |= ENQUEUE_HEAD;
4986 
4987                 enqueue_task(rq, p, queue_flags);
4988         }
4989         if (running)
4990                 set_next_task(rq, p);
4991 
4992         check_class_changed(rq, p, prev_class, oldprio);
4993 
4994         /* Avoid rq from going away on us: */
4995         preempt_disable();
4996         task_rq_unlock(rq, p, &rf);
4997 
4998         if (pi) {
4999                 cpuset_read_unlock();
5000                 rt_mutex_adjust_pi(p);
5001         }
5002 
5003         /* Run balance callbacks after we've adjusted the PI chain: */
5004         balance_callback(rq);
5005         preempt_enable();
5006 
5007         return 0;
5008 
5009 unlock:
5010         task_rq_unlock(rq, p, &rf);
5011         if (pi)
5012                 cpuset_read_unlock();
5013         return retval;
5014 }
5015 
5016 static int _sched_setscheduler(struct task_struct *p, int policy,
5017                                const struct sched_param *param, bool check)
5018 {
5019         struct sched_attr attr = {
5020                 .sched_policy   = policy,
5021                 .sched_priority = param->sched_priority,
5022                 .sched_nice     = PRIO_TO_NICE(p->static_prio),
5023         };
5024 
5025         /* Fixup the legacy SCHED_RESET_ON_FORK hack. */
5026         if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
5027                 attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
5028                 policy &= ~SCHED_RESET_ON_FORK;
5029                 attr.sched_policy = policy;
5030         }
5031 
5032         return __sched_setscheduler(p, &attr, check, true);
5033 }
5034 /**
5035  * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
5036  * @p: the task in question.
5037  * @policy: new policy.
5038  * @param: structure containing the new RT priority.
5039  *
5040  * Return: 0 on success. An error code otherwise.
5041  *
5042  * NOTE that the task may be already dead.
5043  */
5044 int sched_setscheduler(struct task_struct *p, int policy,
5045                        const struct sched_param *param)
5046 {
5047         return _sched_setscheduler(p, policy, param, true);
5048 }
5049 EXPORT_SYMBOL_GPL(sched_setscheduler);
5050 
5051 int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
5052 {
5053         return __sched_setscheduler(p, attr, true, true);
5054 }
5055 EXPORT_SYMBOL_GPL(sched_setattr);
5056 
5057 int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
5058 {
5059         return __sched_setscheduler(p, attr, false, true);
5060 }
5061 
5062 /**
5063  * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
5064  * @p: the task in question.
5065  * @policy: new policy.
5066  * @param: structure containing the new RT priority.
5067  *
5068  * Just like sched_setscheduler, only don't bother checking if the
5069  * current context has permission.  For example, this is needed in
5070  * stop_machine(): we create temporary high priority worker threads,
5071  * but our caller might not have that capability.
5072  *
5073  * Return: 0 on success. An error code otherwise.
5074  */
5075 int sched_setscheduler_nocheck(struct task_struct *p, int policy,
5076                                const struct sched_param *param)
5077 {
5078         return _sched_setscheduler(p, policy, param, false);
5079 }
5080 EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
5081 
5082 static int
5083 do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
5084 {
5085         struct sched_param lparam;
5086         struct task_struct *p;
5087         int retval;
5088 
5089         if (!param || pid < 0)
5090                 return -EINVAL;
5091         if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
5092                 return -EFAULT;
5093 
5094         rcu_read_lock();
5095         retval = -ESRCH;
5096         p = find_process_by_pid(pid);
5097         if (likely(p))
5098                 get_task_struct(p);
5099         rcu_read_unlock();
5100 
5101         if (likely(p)) {
5102                 retval = sched_setscheduler(p, policy, &lparam);
5103                 put_task_struct(p);
5104         }
5105 
5106         return retval;
5107 }
5108 
5109 /*
5110  * Mimics kernel/events/core.c perf_copy_attr().
5111  */
5112 static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
5113 {
5114         u32 size;
5115         int ret;
5116 
5117         /* Zero the full structure, so that a short copy will be nice: */
5118         memset(attr, 0, sizeof(*attr));
5119 
5120         ret = get_user(size, &uattr->size);
5121         if (ret)
5122                 return ret;
5123 
5124         /* ABI compatibility quirk: */
5125         if (!size)
5126                 size = SCHED_ATTR_SIZE_VER0;
5127         if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE)
5128                 goto err_size;
5129 
5130         ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size);
5131         if (ret) {
5132                 if (ret == -E2BIG)
5133                         goto err_size;
5134                 return ret;
5135         }
5136 
5137         if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) &&
5138             size < SCHED_ATTR_SIZE_VER1)
5139                 return -EINVAL;
5140 
5141         /*
5142          * XXX: Do we want to be lenient like existing syscalls; or do we want
5143          * to be strict and return an error on out-of-bounds values?
5144          */
5145         attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
5146 
5147         return 0;
5148 
5149 err_size:
5150         put_user(sizeof(*attr), &uattr->size);
5151         return -E2BIG;
5152 }
5153 
5154 /**
5155  * sys_sched_setscheduler - set/change the scheduler policy and RT priority
5156  * @pid: the pid in question.
5157  * @policy: new policy.
5158  * @param: structure containing the new RT priority.
5159  *
5160  * Return: 0 on success. An error code otherwise.
5161  */
5162 SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
5163 {
5164         if (policy < 0)
5165                 return -EINVAL;
5166 
5167         return do_sched_setscheduler(pid, policy, param);
5168 }
5169 
5170 /**
5171  * sys_sched_setparam - set/change the RT priority of a thread
5172  * @pid: the pid in question.
5173  * @param: structure containing the new RT priority.
5174  *
5175  * Return: 0 on success. An error code otherwise.
5176  */
5177 SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
5178 {
5179         return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
5180 }
5181 
5182 /**
5183  * sys_sched_setattr - same as above, but with extended sched_attr
5184  * @pid: the pid in question.
5185  * @uattr: structure containing the extended parameters.
5186  * @flags: for future extension.
5187  */
5188 SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
5189                                unsigned int, flags)
5190 {
5191         struct sched_attr attr;
5192         struct task_struct *p;
5193         int retval;
5194 
5195         if (!uattr || pid < 0 || flags)
5196                 return -EINVAL;
5197 
5198         retval = sched_copy_attr(uattr, &attr);
5199         if (retval)
5200                 return retval;
5201 
5202         if ((int)attr.sched_policy < 0)
5203                 return -EINVAL;
5204         if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY)
5205                 attr.sched_policy = SETPARAM_POLICY;
5206 
5207         rcu_read_lock();
5208         retval = -ESRCH;
5209         p = find_process_by_pid(pid);
5210         if (likely(p))
5211                 get_task_struct(p);
5212         rcu_read_unlock();
5213 
5214         if (likely(p)) {
5215                 retval = sched_setattr(p, &attr);
5216                 put_task_struct(p);
5217         }
5218 
5219         return retval;
5220 }
5221 
5222 /**
5223  * sys_sched_getscheduler - get the policy (scheduling class) of a thread
5224  * @pid: the pid in question.
5225  *
5226  * Return: On success, the policy of the thread. Otherwise, a negative error
5227  * code.
5228  */
5229 SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
5230 {
5231         struct task_struct *p;
5232         int retval;
5233 
5234         if (pid < 0)
5235                 return -EINVAL;
5236 
5237         retval = -ESRCH;
5238         rcu_read_lock();
5239         p = find_process_by_pid(pid);
5240         if (p) {
5241                 retval = security_task_getscheduler(p);
5242                 if (!retval)
5243                         retval = p->policy
5244                                 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
5245         }
5246         rcu_read_unlock();
5247         return retval;
5248 }
5249 
5250 /**
5251  * sys_sched_getparam - get the RT priority of a thread
5252  * @pid: the pid in question.
5253  * @param: structure containing the RT priority.
5254  *
5255  * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
5256  * code.
5257  */
5258 SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
5259 {
5260         struct sched_param lp = { .sched_priority = 0 };
5261         struct task_struct *p;
5262         int retval;
5263 
5264         if (!param || pid < 0)
5265                 return -EINVAL;
5266 
5267         rcu_read_lock();
5268         p = find_process_by_pid(pid);
5269         retval = -ESRCH;
5270         if (!p)
5271                 goto out_unlock;
5272 
5273         retval = security_task_getscheduler(p);
5274         if (retval)
5275                 goto out_unlock;
5276 
5277         if (task_has_rt_policy(p))
5278                 lp.sched_priority = p->rt_priority;
5279         rcu_read_unlock();
5280 
5281         /*
5282          * This one might sleep, we cannot do it with a spinlock held ...
5283          */
5284         retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
5285 
5286         return retval;
5287 
5288 out_unlock:
5289         rcu_read_unlock();
5290         return retval;
5291 }
5292 
5293 /*
5294  * Copy the kernel size attribute structure (which might be larger
5295  * than what user-space knows about) to user-space.
5296  *
5297  * Note that all cases are valid: user-space buffer can be larger or
5298  * smaller than the kernel-space buffer. The usual case is that both
5299  * have the same size.
5300  */
5301 static int
5302 sched_attr_copy_to_user(struct sched_attr __user *uattr,
5303                         struct sched_attr *kattr,
5304                         unsigned int usize)
5305 {
5306         unsigned int ksize = sizeof(*kattr);
5307 
5308         if (!access_ok(uattr, usize))
5309                 return -EFAULT;
5310 
5311         /*
5312          * sched_getattr() ABI forwards and backwards compatibility:
5313          *
5314          * If usize == ksize then we just copy everything to user-space and all is good.
5315          *
5316          * If usize < ksize then we only copy as much as user-space has space for,
5317          * this keeps ABI compatibility as well. We skip the rest.
5318          *
5319          * If usize > ksize then user-space is using a newer version of the ABI,
5320          * which part the kernel doesn't know about. Just ignore it - tooling can
5321          * detect the kernel's knowledge of attributes from the attr->size value
5322          * which is set to ksize in this case.
5323          */
5324         kattr->size = min(usize, ksize);
5325 
5326         if (copy_to_user(uattr, kattr, kattr->size))
5327                 return -EFAULT;
5328 
5329         return 0;
5330 }
5331 
5332 /**
5333  * sys_sched_getattr - similar to sched_getparam, but with sched_attr
5334  * @pid: the pid in question.
5335  * @uattr: structure containing the extended parameters.
5336  * @usize: sizeof(attr) for fwd/bwd comp.
5337  * @flags: for future extension.
5338  */
5339 SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
5340                 unsigned int, usize, unsigned int, flags)
5341 {
5342         struct sched_attr kattr = { };
5343         struct task_struct *p;
5344         int retval;
5345 
5346         if (!uattr || pid < 0 || usize > PAGE_SIZE ||
5347             usize < SCHED_ATTR_SIZE_VER0 || flags)
5348                 return -EINVAL;
5349 
5350         rcu_read_lock();
5351         p = find_process_by_pid(pid);
5352         retval = -ESRCH;
5353         if (!p)
5354                 goto out_unlock;
5355 
5356         retval = security_task_getscheduler(p);
5357         if (retval)
5358                 goto out_unlock;
5359 
5360         kattr.sched_policy = p->policy;
5361         if (p->sched_reset_on_fork)
5362                 kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
5363         if (task_has_dl_policy(p))
5364                 __getparam_dl(p, &kattr);
5365         else if (task_has_rt_policy(p))
5366                 kattr.sched_priority = p->rt_priority;
5367         else
5368                 kattr.sched_nice = task_nice(p);
5369 
5370 #ifdef CONFIG_UCLAMP_TASK
5371         kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value;
5372         kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value;
5373 #endif
5374 
5375         rcu_read_unlock();
5376 
5377         return sched_attr_copy_to_user(uattr, &kattr, usize);
5378 
5379 out_unlock:
5380         rcu_read_unlock();
5381         return retval;
5382 }
5383 
5384 long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
5385 {
5386         cpumask_var_t cpus_allowed, new_mask;
5387         struct task_struct *p;
5388         int retval;
5389 
5390         rcu_read_lock();
5391 
5392         p = find_process_by_pid(pid);
5393         if (!p) {
5394                 rcu_read_unlock();
5395                 return -ESRCH;
5396         }
5397 
5398         /* Prevent p going away */
5399         get_task_struct(p);
5400         rcu_read_unlock();
5401 
5402         if (p->flags & PF_NO_SETAFFINITY) {
5403                 retval = -EINVAL;
5404                 goto out_put_task;
5405         }
5406         if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5407                 retval = -ENOMEM;
5408                 goto out_put_task;
5409         }
5410         if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
5411                 retval = -ENOMEM;
5412                 goto out_free_cpus_allowed;
5413         }
5414         retval = -EPERM;
5415         if (!check_same_owner(p)) {
5416                 rcu_read_lock();
5417                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
5418                         rcu_read_unlock();
5419                         goto out_free_new_mask;
5420                 }
5421                 rcu_read_unlock();
5422         }
5423 
5424         retval = security_task_setscheduler(p);
5425         if (retval)
5426                 goto out_free_new_mask;
5427 
5428 
5429         cpuset_cpus_allowed(p, cpus_allowed);
5430         cpumask_and(new_mask, in_mask, cpus_allowed);
5431 
5432         /*
5433          * Since bandwidth control happens on root_domain basis,
5434          * if admission test is enabled, we only admit -deadline
5435          * tasks allowed to run on all the CPUs in the task's
5436          * root_domain.
5437          */
5438 #ifdef CONFIG_SMP
5439         if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
5440                 rcu_read_lock();
5441                 if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
5442                         retval = -EBUSY;
5443                         rcu_read_unlock();
5444                         goto out_free_new_mask;
5445                 }
5446                 rcu_read_unlock();
5447         }
5448 #endif
5449 again:
5450         retval = __set_cpus_allowed_ptr(p, new_mask, true);
5451 
5452         if (!retval) {
5453                 cpuset_cpus_allowed(p, cpus_allowed);
5454                 if (!cpumask_subset(new_mask, cpus_allowed)) {
5455                         /*
5456                          * We must have raced with a concurrent cpuset
5457                          * update. Just reset the cpus_allowed to the
5458                          * cpuset's cpus_allowed
5459                          */
5460                         cpumask_copy(new_mask, cpus_allowed);
5461                         goto again;
5462                 }
5463         }
5464 out_free_new_mask:
5465         free_cpumask_var(new_mask);
5466 out_free_cpus_allowed:
5467         free_cpumask_var(cpus_allowed);
5468 out_put_task:
5469         put_task_struct(p);
5470         return retval;
5471 }
5472 
5473 static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
5474                              struct cpumask *new_mask)
5475 {
5476         if (len < cpumask_size())
5477                 cpumask_clear(new_mask);
5478         else if (len > cpumask_size())
5479                 len = cpumask_size();
5480 
5481         return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5482 }
5483 
5484 /**
5485  * sys_sched_setaffinity - set the CPU affinity of a process
5486  * @pid: pid of the process
5487  * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5488  * @user_mask_ptr: user-space pointer to the new CPU mask
5489  *
5490  * Return: 0 on success. An error code otherwise.
5491  */
5492 SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
5493                 unsigned long __user *, user_mask_ptr)
5494 {
5495         cpumask_var_t new_mask;
5496         int retval;
5497 
5498         if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
5499                 return -ENOMEM;
5500 
5501         retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
5502         if (retval == 0)
5503                 retval = sched_setaffinity(pid, new_mask);
5504         free_cpumask_var(new_mask);
5505         return retval;
5506 }
5507 
5508 long sched_getaffinity(pid_t pid, struct cpumask *mask)
5509 {
5510         struct task_struct *p;
5511         unsigned long flags;
5512         int retval;
5513 
5514         rcu_read_lock();
5515 
5516         retval = -ESRCH;
5517         p = find_process_by_pid(pid);
5518         if (!p)
5519                 goto out_unlock;
5520 
5521         retval = security_task_getscheduler(p);
5522         if (retval)
5523                 goto out_unlock;
5524 
5525         raw_spin_lock_irqsave(&p->pi_lock, flags);
5526         cpumask_and(mask, &p->cpus_mask, cpu_active_mask);
5527         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
5528 
5529 out_unlock:
5530         rcu_read_unlock();
5531 
5532         return retval;
5533 }
5534 
5535 /**
5536  * sys_sched_getaffinity - get the CPU affinity of a process
5537  * @pid: pid of the process
5538  * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5539  * @user_mask_ptr: user-space pointer to hold the current CPU mask
5540  *
5541  * Return: size of CPU mask copied to user_mask_ptr on success. An
5542  * error code otherwise.
5543  */
5544 SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
5545                 unsigned long __user *, user_mask_ptr)
5546 {
5547         int ret;
5548         cpumask_var_t mask;
5549 
5550         if ((len * BITS_PER_BYTE) < nr_cpu_ids)
5551                 return -EINVAL;
5552         if (len & (sizeof(unsigned long)-1))
5553                 return -EINVAL;
5554 
5555         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5556                 return -ENOMEM;
5557 
5558         ret = sched_getaffinity(pid, mask);
5559         if (ret == 0) {
5560                 unsigned int retlen = min(len, cpumask_size());
5561 
5562                 if (copy_to_user(user_mask_ptr, mask, retlen))
5563                         ret = -EFAULT;
5564                 else
5565                         ret = retlen;
5566         }
5567         free_cpumask_var(mask);
5568 
5569         return ret;
5570 }
5571 
5572 /**
5573  * sys_sched_yield - yield the current processor to other threads.
5574  *
5575  * This function yields the current CPU to other tasks. If there are no
5576  * other threads running on this CPU then this function will return.
5577  *
5578  * Return: 0.
5579  */
5580 static void do_sched_yield(void)
5581 {
5582         struct rq_flags rf;
5583         struct rq *rq;
5584 
5585         rq = this_rq_lock_irq(&rf);
5586 
5587         schedstat_inc(rq->yld_count);
5588         current->sched_class->yield_task(rq);
5589 
5590         /*
5591          * Since we are going to call schedule() anyway, there's
5592          * no need to preempt or enable interrupts:
5593          */
5594         preempt_disable();
5595         rq_unlock(rq, &rf);
5596         sched_preempt_enable_no_resched();
5597 
5598         schedule();
5599 }
5600 
5601 SYSCALL_DEFINE0(sched_yield)
5602 {
5603         do_sched_yield();
5604         return 0;
5605 }
5606 
5607 #ifndef CONFIG_PREEMPTION
5608 int __sched _cond_resched(void)
5609 {
5610         if (should_resched(0)) {
5611                 preempt_schedule_common();
5612                 return 1;
5613         }
5614         rcu_all_qs();
5615         return 0;
5616 }
5617 EXPORT_SYMBOL(_cond_resched);
5618 #endif
5619 
5620 /*
5621  * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
5622  * call schedule, and on return reacquire the lock.
5623  *
5624  * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level
5625  * operations here to prevent schedule() from being called twice (once via
5626  * spin_unlock(), once by hand).
5627  */
5628 int __cond_resched_lock(spinlock_t *lock)
5629 {
5630         int resched = should_resched(PREEMPT_LOCK_OFFSET);
5631         int ret = 0;
5632 
5633         lockdep_assert_held(lock);
5634 
5635         if (spin_needbreak(lock) || resched) {
5636                 spin_unlock(lock);
5637                 if (resched)
5638                         preempt_schedule_common();
5639                 else
5640                         cpu_relax();
5641                 ret = 1;
5642                 spin_lock(lock);
5643         }
5644         return ret;
5645 }
5646 EXPORT_SYMBOL(__cond_resched_lock);
5647 
5648 /**
5649  * yield - yield the current processor to other threads.
5650  *
5651  * Do not ever use this function, there's a 99% chance you're doing it wrong.
5652  *
5653  * The scheduler is at all times free to pick the calling task as the most
5654  * eligible task to run, if removing the yield() call from your code breaks
5655  * it, its already broken.
5656  *
5657  * Typical broken usage is:
5658  *
5659  * while (!event)
5660  *      yield();
5661  *
5662  * where one assumes that yield() will let 'the other' process run that will
5663  * make event true. If the current task is a SCHED_FIFO task that will never
5664  * happen. Never use yield() as a progress guarantee!!
5665  *
5666  * If you want to use yield() to wait for something, use wait_event().
5667  * If you want to use yield() to be 'nice' for others, use cond_resched().
5668  * If you still want to use yield(), do not!
5669  */
5670 void __sched yield(void)
5671 {
5672         set_current_state(TASK_RUNNING);
5673         do_sched_yield();
5674 }
5675 EXPORT_SYMBOL(yield);
5676 
5677 /**
5678  * yield_to - yield the current processor to another thread in
5679  * your thread group, or accelerate that thread toward the
5680  * processor it's on.
5681  * @p: target task
5682  * @preempt: whether task preemption is allowed or not
5683  *
5684  * It's the caller's job to ensure that the target task struct
5685  * can't go away on us before we can do any checks.
5686  *
5687  * Return:
5688  *      true (>0) if we indeed boosted the target task.
5689  *      false (0) if we failed to boost the target.
5690  *      -ESRCH if there's no task to yield to.
5691  */
5692 int __sched yield_to(struct task_struct *p, bool preempt)
5693 {
5694         struct task_struct *curr = current;
5695         struct rq *rq, *p_rq;
5696         unsigned long flags;
5697         int yielded = 0;
5698 
5699         local_irq_save(flags);
5700         rq = this_rq();
5701 
5702 again:
5703         p_rq = task_rq(p);
5704         /*
5705          * If we're the only runnable task on the rq and target rq also
5706          * has only one task, there's absolutely no point in yielding.
5707          */
5708         if (rq->nr_running == 1 && p_rq->nr_running == 1) {
5709                 yielded = -ESRCH;
5710                 goto out_irq;
5711         }
5712 
5713         double_rq_lock(rq, p_rq);
5714         if (task_rq(p) != p_rq) {
5715                 double_rq_unlock(rq, p_rq);
5716                 goto again;
5717         }
5718 
5719         if (!curr->sched_class->yield_to_task)
5720                 goto out_unlock;
5721 
5722         if (curr->sched_class != p->sched_class)
5723                 goto out_unlock;
5724 
5725         if (task_running(p_rq, p) || p->state)
5726                 goto out_unlock;
5727 
5728         yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5729         if (yielded) {
5730                 schedstat_inc(rq->yld_count);
5731                 /*
5732                  * Make p's CPU reschedule; pick_next_entity takes care of
5733                  * fairness.
5734                  */
5735                 if (preempt && rq != p_rq)
5736                         resched_curr(p_rq);
5737         }
5738 
5739 out_unlock:
5740         double_rq_unlock(rq, p_rq);
5741 out_irq:
5742         local_irq_restore(flags);
5743 
5744         if (yielded > 0)
5745                 schedule();
5746 
5747         return yielded;
5748 }
5749 EXPORT_SYMBOL_GPL(yield_to);
5750 
5751 int io_schedule_prepare(void)
5752 {
5753         int old_iowait = current->in_iowait;
5754 
5755         current->in_iowait = 1;
5756         blk_schedule_flush_plug(current);
5757 
5758         return old_iowait;
5759 }
5760 
5761 void io_schedule_finish(int token)
5762 {
5763         current->in_iowait = token;
5764 }
5765 
5766 /*
5767  * This task is about to go to sleep on IO. Increment rq->nr_iowait so
5768  * that process accounting knows that this is a task in IO wait state.
5769  */
5770 long __sched io_schedule_timeout(long timeout)
5771 {
5772         int token;
5773         long ret;
5774 
5775         token = io_schedule_prepare();
5776         ret = schedule_timeout(timeout);
5777         io_schedule_finish(token);
5778 
5779         return ret;
5780 }
5781 EXPORT_SYMBOL(io_schedule_timeout);
5782 
5783 void __sched io_schedule(void)
5784 {
5785         int token;
5786 
5787         token = io_schedule_prepare();
5788         schedule();
5789         io_schedule_finish(token);
5790 }
5791 EXPORT_SYMBOL(io_schedule);
5792 
5793 /**
5794  * sys_sched_get_priority_max - return maximum RT priority.
5795  * @policy: scheduling class.
5796  *
5797  * Return: On success, this syscall returns the maximum
5798  * rt_priority that can be used by a given scheduling class.
5799  * On failure, a negative error code is returned.
5800  */
5801 SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
5802 {
5803         int ret = -EINVAL;
5804 
5805         switch (policy) {
5806         case SCHED_FIFO:
5807         case SCHED_RR:
5808                 ret = MAX_USER_RT_PRIO-1;
5809                 break;
5810         case SCHED_DEADLINE:
5811         case SCHED_NORMAL:
5812         case SCHED_BATCH:
5813         case SCHED_IDLE:
5814                 ret = 0;
5815                 break;
5816         }
5817         return ret;
5818 }
5819 
5820 /**
5821  * sys_sched_get_priority_min - return minimum RT priority.
5822  * @policy: scheduling class.
5823  *
5824  * Return: On success, this syscall returns the minimum
5825  * rt_priority that can be used by a given scheduling class.
5826  * On failure, a negative error code is returned.
5827  */
5828 SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
5829 {
5830         int ret = -EINVAL;
5831 
5832         switch (policy) {
5833         case SCHED_FIFO:
5834         case SCHED_RR:
5835                 ret = 1;
5836                 break;
5837         case SCHED_DEADLINE:
5838         case SCHED_NORMAL:
5839         case SCHED_BATCH:
5840         case SCHED_IDLE:
5841                 ret = 0;
5842         }
5843         return ret;
5844 }
5845 
5846 static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
5847 {
5848         struct task_struct *p;
5849         unsigned int time_slice;
5850         struct rq_flags rf;
5851         struct rq *rq;
5852         int retval;
5853 
5854         if (pid < 0)
5855                 return -EINVAL;
5856 
5857         retval = -ESRCH;
5858         rcu_read_lock();
5859         p = find_process_by_pid(pid);
5860         if (!p)
5861                 goto out_unlock;
5862 
5863         retval = security_task_getscheduler(p);
5864         if (retval)
5865                 goto out_unlock;
5866 
5867         rq = task_rq_lock(p, &rf);
5868         time_slice = 0;
5869         if (p->sched_class->get_rr_interval)
5870                 time_slice = p->sched_class->get_rr_interval(rq, p);
5871         task_rq_unlock(rq, p, &rf);
5872 
5873         rcu_read_unlock();
5874         jiffies_to_timespec64(time_slice, t);
5875         return 0;
5876 
5877 out_unlock:
5878         rcu_read_unlock();
5879         return retval;
5880 }
5881 
5882 /**
5883  * sys_sched_rr_get_interval - return the default timeslice of a process.
5884  * @pid: pid of the process.
5885  * @interval: userspace pointer to the timeslice value.
5886  *
5887  * this syscall writes the default timeslice value of a given process
5888  * into the user-space timespec buffer. A value of '0' means infinity.
5889  *
5890  * Return: On success, 0 and the timeslice is in @interval. Otherwise,
5891  * an error code.
5892  */
5893 SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5894                 struct __kernel_timespec __user *, interval)
5895 {
5896         struct timespec64 t;
5897         int retval = sched_rr_get_interval(pid, &t);
5898 
5899         if (retval == 0)
5900                 retval = put_timespec64(&t, interval);
5901 
5902         return retval;
5903 }
5904 
5905 #ifdef CONFIG_COMPAT_32BIT_TIME
5906 SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid,
5907                 struct old_timespec32 __user *, interval)
5908 {
5909         struct timespec64 t;
5910         int retval = sched_rr_get_interval(pid, &t);
5911 
5912         if (retval == 0)
5913                 retval = put_old_timespec32(&t, interval);
5914         return retval;
5915 }
5916 #endif
5917 
5918 void sched_show_task(struct task_struct *p)
5919 {
5920         unsigned long free = 0;
5921         int ppid;
5922 
5923         if (!try_get_task_stack(p))
5924                 return;
5925 
5926         printk(KERN_INFO "%-15.15s %c", p->comm, task_state_to_char(p));
5927 
5928         if (p->state == TASK_RUNNING)
5929                 printk(KERN_CONT "  running task    ");
5930 #ifdef CONFIG_DEBUG_STACK_USAGE
5931         free = stack_not_used(p);
5932 #endif
5933         ppid = 0;
5934         rcu_read_lock();
5935         if (pid_alive(p))
5936                 ppid = task_pid_nr(rcu_dereference(p->real_parent));
5937         rcu_read_unlock();
5938         printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5939                 task_pid_nr(p), ppid,
5940                 (unsigned long)task_thread_info(p)->flags);
5941 
5942         print_worker_info(KERN_INFO, p);
5943         show_stack(p, NULL);
5944         put_task_stack(p);
5945 }
5946 EXPORT_SYMBOL_GPL(sched_show_task);
5947 
5948 static inline bool
5949 state_filter_match(unsigned long state_filter, struct task_struct *p)
5950 {
5951         /* no filter, everything matches */
5952         if (!state_filter)
5953                 return true;
5954 
5955         /* filter, but doesn't match */
5956         if (!(p->state & state_filter))
5957                 return false;
5958 
5959         /*
5960          * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows
5961          * TASK_KILLABLE).
5962          */
5963         if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE)
5964                 return false;
5965 
5966         return true;
5967 }
5968 
5969 
5970 void show_state_filter(unsigned long state_filter)
5971 {
5972         struct task_struct *g, *p;
5973 
5974 #if BITS_PER_LONG == 32
5975         printk(KERN_INFO
5976                 "  task                PC stack   pid father\n");
5977 #else
5978         printk(KERN_INFO
5979                 "  task                        PC stack   pid father\n");
5980 #endif
5981         rcu_read_lock();
5982         for_each_process_thread(g, p) {
5983                 /*
5984                  * reset the NMI-timeout, listing all files on a slow
5985                  * console might take a lot of time:
5986                  * Also, reset softlockup watchdogs on all CPUs, because
5987                  * another CPU might be blocked waiting for us to process
5988                  * an IPI.
5989                  */
5990                 touch_nmi_watchdog();
5991                 touch_all_softlockup_watchdogs();
5992                 if (state_filter_match(state_filter, p))
5993                         sched_show_task(p);
5994         }
5995 
5996 #ifdef CONFIG_SCHED_DEBUG
5997         if (!state_filter)
5998                 sysrq_sched_debug_show();
5999 #endif
6000         rcu_read_unlock();
6001         /*
6002          * Only show locks if all tasks are dumped:
6003          */
6004         if (!state_filter)
6005                 debug_show_all_locks();
6006 }
6007 
6008 /**
6009  * init_idle - set up an idle thread for a given CPU
6010  * @idle: task in question
6011  * @cpu: CPU the idle task belongs to
6012  *
6013  * NOTE: this function does not set the idle thread's NEED_RESCHED
6014  * flag, to make booting more robust.
6015  */
6016 void init_idle(struct task_struct *idle, int cpu)
6017 {
6018         struct rq *rq = cpu_rq(cpu);
6019         unsigned long flags;
6020 
6021         __sched_fork(0, idle);
6022 
6023         raw_spin_lock_irqsave(&idle->pi_lock, flags);
6024         raw_spin_lock(&rq->lock);
6025 
6026         idle->state = TASK_RUNNING;
6027         idle->se.exec_start = sched_clock();
6028         idle->flags |= PF_IDLE;
6029 
6030         kasan_unpoison_task_stack(idle);
6031 
6032 #ifdef CONFIG_SMP
6033         /*
6034          * Its possible that init_idle() gets called multiple times on a task,
6035          * in that case do_set_cpus_allowed() will not do the right thing.
6036          *
6037          * And since this is boot we can forgo the serialization.
6038          */
6039         set_cpus_allowed_common(idle, cpumask_of(cpu));
6040 #endif
6041         /*
6042          * We're having a chicken and egg problem, even though we are
6043          * holding rq->lock, the CPU isn't yet set to this CPU so the
6044          * lockdep check in task_group() will fail.
6045          *
6046          * Similar case to sched_fork(). / Alternatively we could
6047          * use task_rq_lock() here and obtain the other rq->lock.
6048          *
6049          * Silence PROVE_RCU
6050          */
6051         rcu_read_lock();
6052         __set_task_cpu(idle, cpu);
6053         rcu_read_unlock();
6054 
6055         rq->idle = idle;
6056         rcu_assign_pointer(rq->curr, idle);
6057         idle->on_rq = TASK_ON_RQ_QUEUED;
6058 #ifdef CONFIG_SMP
6059         idle->on_cpu = 1;
6060 #endif
6061         raw_spin_unlock(&rq->lock);
6062         raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
6063 
6064         /* Set the preempt count _outside_ the spinlocks! */
6065         init_idle_preempt_count(idle, cpu);
6066 
6067         /*
6068          * The idle tasks have their own, simple scheduling class:
6069          */
6070         idle->sched_class = &idle_sched_class;
6071         ftrace_graph_init_idle_task(idle, cpu);
6072         vtime_init_idle(idle, cpu);
6073 #ifdef CONFIG_SMP
6074         sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
6075 #endif
6076 }
6077 
6078 #ifdef CONFIG_SMP
6079 
6080 int cpuset_cpumask_can_shrink(const struct cpumask *cur,
6081                               const struct cpumask *trial)
6082 {
6083         int ret = 1;
6084 
6085         if (!cpumask_weight(cur))
6086                 return ret;
6087 
6088         ret = dl_cpuset_cpumask_can_shrink(cur, trial);
6089 
6090         return ret;
6091 }
6092 
6093 int task_can_attach(struct task_struct *p,
6094                     const struct cpumask *cs_cpus_allowed)
6095 {
6096         int ret = 0;
6097 
6098         /*
6099          * Kthreads which disallow setaffinity shouldn't be moved
6100          * to a new cpuset; we don't want to change their CPU
6101          * affinity and isolating such threads by their set of
6102          * allowed nodes is unnecessary.  Thus, cpusets are not
6103          * applicable for such threads.  This prevents checking for
6104          * success of set_cpus_allowed_ptr() on all attached tasks
6105          * before cpus_mask may be changed.
6106          */
6107         if (p->flags & PF_NO_SETAFFINITY) {
6108                 ret = -EINVAL;
6109                 goto out;
6110         }
6111 
6112         if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span,
6113                                               cs_cpus_allowed))
6114                 ret = dl_task_can_attach(p, cs_cpus_allowed);
6115 
6116 out:
6117         return ret;
6118 }
6119 
6120 bool sched_smp_initialized __read_mostly;
6121 
6122 #ifdef CONFIG_NUMA_BALANCING
6123 /* Migrate current task p to target_cpu */
6124 int migrate_task_to(struct task_struct *p, int target_cpu)
6125 {
6126         struct migration_arg arg = { p, target_cpu };
6127         int curr_cpu = task_cpu(p);
6128 
6129         if (curr_cpu == target_cpu)
6130                 return 0;
6131 
6132         if (!cpumask_test_cpu(target_cpu, p->cpus_ptr))
6133                 return -EINVAL;
6134 
6135         /* TODO: This is not properly updating schedstats */
6136 
6137         trace_sched_move_numa(p, curr_cpu, target_cpu);
6138         return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
6139 }
6140 
6141 /*
6142  * Requeue a task on a given node and accurately track the number of NUMA
6143  * tasks on the runqueues
6144  */
6145 void sched_setnuma(struct task_struct *p, int nid)
6146 {
6147         bool queued, running;
6148         struct rq_flags rf;
6149         struct rq *rq;
6150 
6151         rq = task_rq_lock(p, &rf);
6152         queued = task_on_rq_queued(p);
6153         running = task_current(rq, p);
6154 
6155         if (queued)
6156                 dequeue_task(rq, p, DEQUEUE_SAVE);
6157         if (running)
6158                 put_prev_task(rq, p);
6159 
6160         p->numa_preferred_nid = nid;
6161 
6162         if (queued)
6163                 enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
6164         if (running)
6165                 set_next_task(rq, p);
6166         task_rq_unlock(rq, p, &rf);
6167 }
6168 #endif /* CONFIG_NUMA_BALANCING */
6169 
6170 #ifdef CONFIG_HOTPLUG_CPU
6171 /*
6172  * Ensure that the idle task is using init_mm right before its CPU goes
6173  * offline.
6174  */
6175 void idle_task_exit(void)
6176 {
6177         struct mm_struct *mm = current->active_mm;
6178 
6179         BUG_ON(cpu_online(smp_processor_id()));
6180 
6181         if (mm != &init_mm) {
6182                 switch_mm(mm, &init_mm, current);
6183                 current->active_mm = &init_mm;
6184                 finish_arch_post_lock_switch();
6185         }
6186         mmdrop(mm);
6187 }
6188 
6189 /*
6190  * Since this CPU is going 'away' for a while, fold any nr_active delta
6191  * we might have. Assumes we're called after migrate_tasks() so that the
6192  * nr_active count is stable. We need to take the teardown thread which
6193  * is calling this into account, so we hand in adjust = 1 to the load
6194  * calculation.
6195  *
6196  * Also see the comment "Global load-average calculations".
6197  */
6198 static void calc_load_migrate(struct rq *rq)
6199 {
6200         long delta = calc_load_fold_active(rq, 1);
6201         if (delta)
6202                 atomic_long_add(delta, &calc_load_tasks);
6203 }
6204 
6205 static struct task_struct *__pick_migrate_task(struct rq *rq)
6206 {
6207         const struct sched_class *class;
6208         struct task_struct *next;
6209 
6210         for_each_class(class) {
6211                 next = class->pick_next_task(rq, NULL, NULL);
6212                 if (next) {
6213                         next->sched_class->put_prev_task(rq, next);
6214                         return next;
6215                 }
6216         }
6217 
6218         /* The idle class should always have a runnable task */
6219         BUG();
6220 }
6221 
6222 /*
6223  * Migrate all tasks from the rq, sleeping tasks will be migrated by
6224  * try_to_wake_up()->select_task_rq().
6225  *
6226  * Called with rq->lock held even though we'er in stop_machine() and
6227  * there's no concurrency possible, we hold the required locks anyway
6228  * because of lock validation efforts.
6229  */
6230 static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
6231 {
6232         struct rq *rq = dead_rq;
6233         struct task_struct *next, *stop = rq->stop;
6234         struct rq_flags orf = *rf;
6235         int dest_cpu;
6236 
6237         /*
6238          * Fudge the rq selection such that the below task selection loop
6239          * doesn't get stuck on the currently eligible stop task.
6240          *
6241          * We're currently inside stop_machine() and the rq is either stuck
6242          * in the stop_machine_cpu_stop() loop, or we're executing this code,
6243          * either way we should never end up calling schedule() until we're
6244          * done here.
6245          */
6246         rq->stop = NULL;
6247 
6248         /*
6249          * put_prev_task() and pick_next_task() sched
6250          * class method both need to have an up-to-date
6251          * value of rq->clock[_task]
6252          */
6253         update_rq_clock(rq);
6254 
6255         for (;;) {
6256                 /*
6257                  * There's this thread running, bail when that's the only
6258                  * remaining thread:
6259                  */
6260                 if (rq->nr_running == 1)
6261                         break;
6262 
6263                 next = __pick_migrate_task(rq);
6264 
6265                 /*
6266                  * Rules for changing task_struct::cpus_mask are holding
6267                  * both pi_lock and rq->lock, such that holding either
6268                  * stabilizes the mask.
6269                  *
6270                  * Drop rq->lock is not quite as disastrous as it usually is
6271                  * because !cpu_active at this point, which means load-balance
6272                  * will not interfere. Also, stop-machine.
6273                  */
6274                 rq_unlock(rq, rf);
6275                 raw_spin_lock(&next->pi_lock);
6276                 rq_relock(rq, rf);
6277 
6278                 /*
6279                  * Since we're inside stop-machine, _nothing_ should have
6280                  * changed the task, WARN if weird stuff happened, because in
6281                  * that case the above rq->lock drop is a fail too.
6282                  */
6283                 if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) {
6284                         raw_spin_unlock(&next->pi_lock);
6285                         continue;
6286                 }
6287 
6288                 /* Find suitable destination for @next, with force if needed. */
6289                 dest_cpu = select_fallback_rq(dead_rq->cpu, next);
6290                 rq = __migrate_task(rq, rf, next, dest_cpu);
6291                 if (rq != dead_rq) {
6292                         rq_unlock(rq, rf);
6293                         rq = dead_rq;
6294                         *rf = orf;
6295                         rq_relock(rq, rf);
6296                 }
6297                 raw_spin_unlock(&next->pi_lock);
6298         }
6299 
6300         rq->stop = stop;
6301 }
6302 #endif /* CONFIG_HOTPLUG_CPU */
6303 
6304 void set_rq_online(struct rq *rq)
6305 {
6306         if (!rq->online) {
6307                 const struct sched_class *class;
6308 
6309                 cpumask_set_cpu(rq->cpu, rq->rd->online);
6310                 rq->online = 1;
6311 
6312                 for_each_class(class) {
6313                         if (class->rq_online)
6314                                 class->rq_online(rq);
6315                 }
6316         }
6317 }
6318 
6319 void set_rq_offline(struct rq *rq)
6320 {
6321         if (rq->online) {
6322                 const struct sched_class *class;
6323 
6324                 for_each_class(class) {
6325                         if (class->rq_offline)
6326                                 class->rq_offline(rq);
6327                 }
6328 
6329                 cpumask_clear_cpu(rq->cpu, rq->rd->online);
6330                 rq->online = 0;
6331         }
6332 }
6333 
6334 /*
6335  * used to mark begin/end of suspend/resume:
6336  */
6337 static int num_cpus_frozen;
6338 
6339 /*
6340  * Update cpusets according to cpu_active mask.  If cpusets are
6341  * disabled, cpuset_update_active_cpus() becomes a simple wrapper
6342  * around partition_sched_domains().
6343  *
6344  * If we come here as part of a suspend/resume, don't touch cpusets because we
6345  * want to restore it back to its original state upon resume anyway.
6346  */
6347 static void cpuset_cpu_active(void)
6348 {
6349         if (cpuhp_tasks_frozen) {
6350                 /*
6351                  * num_cpus_frozen tracks how many CPUs are involved in suspend
6352                  * resume sequence. As long as this is not the last online
6353                  * operation in the resume sequence, just build a single sched
6354                  * domain, ignoring cpusets.
6355                  */
6356                 partition_sched_domains(1, NULL, NULL);
6357                 if (--num_cpus_frozen)
6358                         return;
6359                 /*
6360                  * This is the last CPU online operation. So fall through and
6361                  * restore the original sched domains by considering the
6362                  * cpuset configurations.
6363                  */
6364                 cpuset_force_rebuild();
6365         }
6366         cpuset_update_active_cpus();
6367 }
6368 
6369 static int cpuset_cpu_inactive(unsigned int cpu)
6370 {
6371         if (!cpuhp_tasks_frozen) {
6372                 if (dl_cpu_busy(cpu))
6373                         return -EBUSY;
6374                 cpuset_update_active_cpus();
6375         } else {
6376                 num_cpus_frozen++;
6377                 partition_sched_domains(1, NULL, NULL);
6378         }
6379         return 0;
6380 }
6381 
6382 int sched_cpu_activate(unsigned int cpu)
6383 {
6384         struct rq *rq = cpu_rq(cpu);
6385         struct rq_flags rf;
6386 
6387 #ifdef CONFIG_SCHED_SMT
6388         /*
6389          * When going up, increment the number of cores with SMT present.
6390          */
6391         if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
6392                 static_branch_inc_cpuslocked(&sched_smt_present);
6393 #endif
6394         set_cpu_active(cpu, true);
6395 
6396         if (sched_smp_initialized) {
6397                 sched_domains_numa_masks_set(cpu);
6398                 cpuset_cpu_active();
6399         }
6400 
6401         /*
6402          * Put the rq online, if not already. This happens:
6403          *
6404          * 1) In the early boot process, because we build the real domains
6405          *    after all CPUs have been brought up.
6406          *
6407          * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
6408          *    domains.
6409          */
6410         rq_lock_irqsave(rq, &rf);
6411         if (rq->rd) {
6412                 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
6413                 set_rq_online(rq);
6414         }
6415         rq_unlock_irqrestore(rq, &rf);
6416 
6417         return 0;
6418 }
6419 
6420 int sched_cpu_deactivate(unsigned int cpu)
6421 {
6422         int ret;
6423 
6424         set_cpu_active(cpu, false);
6425         /*
6426          * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU
6427          * users of this state to go away such that all new such users will
6428          * observe it.
6429          *
6430          * Do sync before park smpboot threads to take care the rcu boost case.
6431          */
6432         synchronize_rcu();
6433 
6434 #ifdef CONFIG_SCHED_SMT
6435         /*
6436          * When going down, decrement the number of cores with SMT present.
6437          */
6438         if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
6439                 static_branch_dec_cpuslocked(&sched_smt_present);
6440 #endif
6441 
6442         if (!sched_smp_initialized)
6443                 return 0;
6444 
6445         ret = cpuset_cpu_inactive(cpu);
6446         if (ret) {
6447                 set_cpu_active(cpu, true);
6448                 return ret;
6449         }
6450         sched_domains_numa_masks_clear(cpu);
6451         return 0;
6452 }
6453 
6454 static void sched_rq_cpu_starting(unsigned int cpu)
6455 {
6456         struct rq *rq = cpu_rq(cpu);
6457 
6458         rq->calc_load_update = calc_load_update;
6459         update_max_interval();
6460 }
6461 
6462 int sched_cpu_starting(unsigned int cpu)
6463 {
6464         sched_rq_cpu_starting(cpu);
6465         sched_tick_start(cpu);
6466         return 0;
6467 }
6468 
6469 #ifdef CONFIG_HOTPLUG_CPU
6470 int sched_cpu_dying(unsigned int cpu)
6471 {
6472         struct rq *rq = cpu_rq(cpu);
6473         struct rq_flags rf;
6474 
6475         /* Handle pending wakeups and then migrate everything off */
6476         sched_ttwu_pending();
6477         sched_tick_stop(cpu);
6478 
6479         rq_lock_irqsave(rq, &rf);
6480         if (rq->rd) {
6481                 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
6482                 set_rq_offline(rq);
6483         }
6484         migrate_tasks(rq, &rf);
6485         BUG_ON(rq->nr_running != 1);
6486         rq_unlock_irqrestore(rq, &rf);
6487 
6488         calc_load_migrate(rq);
6489         update_max_interval();
6490         nohz_balance_exit_idle(rq);
6491         hrtick_clear(rq);
6492         return 0;
6493 }
6494 #endif
6495 
6496 void __init sched_init_smp(void)
6497 {
6498         sched_init_numa();
6499 
6500         /*
6501          * There's no userspace yet to cause hotplug operations; hence all the
6502          * CPU masks are stable and all blatant races in the below code cannot
6503          * happen.
6504          */
6505         mutex_lock(&sched_domains_mutex);
6506         sched_init_domains(cpu_active_mask);
6507         mutex_unlock(&sched_domains_mutex);
6508 
6509         /* Move init over to a non-isolated CPU */
6510         if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0)
6511                 BUG();
6512         sched_init_granularity();
6513 
6514         init_sched_rt_class();
6515         init_sched_dl_class();
6516 
6517         sched_smp_initialized = true;
6518 }
6519 
6520 static int __init migration_init(void)
6521 {
6522         sched_cpu_starting(smp_processor_id());
6523         return 0;
6524 }
6525 early_initcall(migration_init);
6526 
6527 #else
6528 void __init sched_init_smp(void)
6529 {
6530         sched_init_granularity();
6531 }
6532 #endif /* CONFIG_SMP */
6533 
6534 int in_sched_functions(unsigned long addr)
6535 {
6536         return in_lock_functions(addr) ||
6537                 (addr >= (unsigned long)__sched_text_start
6538                 && addr < (unsigned long)__sched_text_end);
6539 }
6540 
6541 #ifdef CONFIG_CGROUP_SCHED
6542 /*
6543  * Default task group.
6544  * Every task in system belongs to this group at bootup.
6545  */
6546 struct task_group root_task_group;
6547 LIST_HEAD(task_groups);
6548 
6549 /* Cacheline aligned slab cache for task_group */
6550 static struct kmem_cache *task_group_cache __read_mostly;
6551 #endif
6552 
6553 DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
6554 DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
6555 
6556 void __init sched_init(void)
6557 {
6558         unsigned long ptr = 0;
6559         int i;
6560 
6561         wait_bit_init();
6562 
6563 #ifdef CONFIG_FAIR_GROUP_SCHED
6564         ptr += 2 * nr_cpu_ids * sizeof(void **);
6565 #endif
6566 #ifdef CONFIG_RT_GROUP_SCHED
6567         ptr += 2 * nr_cpu_ids * sizeof(void **);
6568 #endif
6569         if (ptr) {
6570                 ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT);
6571 
6572 #ifdef CONFIG_FAIR_GROUP_SCHED
6573                 root_task_group.se = (struct sched_entity **)ptr;
6574                 ptr += nr_cpu_ids * sizeof(void **);
6575 
6576                 root_task_group.cfs_rq = (struct cfs_rq **)ptr;
6577                 ptr += nr_cpu_ids * sizeof(void **);
6578 
6579 #endif /* CONFIG_FAIR_GROUP_SCHED */
6580 #ifdef CONFIG_RT_GROUP_SCHED
6581                 root_task_group.rt_se = (struct sched_rt_entity **)ptr;
6582                 ptr += nr_cpu_ids * sizeof(void **);
6583 
6584                 root_task_group.rt_rq = (struct rt_rq **)ptr;
6585                 ptr += nr_cpu_ids * sizeof(void **);
6586 
6587 #endif /* CONFIG_RT_GROUP_SCHED */
6588         }
6589 #ifdef CONFIG_CPUMASK_OFFSTACK
6590         for_each_possible_cpu(i) {
6591                 per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
6592                         cpumask_size(), GFP_KERNEL, cpu_to_node(i));
6593                 per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
6594                         cpumask_size(), GFP_KERNEL, cpu_to_node(i));
6595         }
6596 #endif /* CONFIG_CPUMASK_OFFSTACK */
6597 
6598         init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
6599         init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime());
6600 
6601 #ifdef CONFIG_SMP
6602         init_defrootdomain();
6603 #endif
6604 
6605 #ifdef CONFIG_RT_GROUP_SCHED
6606         init_rt_bandwidth(&root_task_group.rt_bandwidth,
6607                         global_rt_period(), global_rt_runtime());
6608 #endif /* CONFIG_RT_GROUP_SCHED */
6609 
6610 #ifdef CONFIG_CGROUP_SCHED
6611         task_group_cache = KMEM_CACHE(task_group, 0);
6612 
6613         list_add(&root_task_group.list, &task_groups);
6614         INIT_LIST_HEAD(&root_task_group.children);
6615         INIT_LIST_HEAD(&root_task_group.siblings);
6616         autogroup_init(&init_task);
6617 #endif /* CONFIG_CGROUP_SCHED */
6618 
6619         for_each_possible_cpu(i) {
6620                 struct rq *rq;
6621 
6622                 rq = cpu_rq(i);
6623                 raw_spin_lock_init(&rq->lock);
6624                 rq->nr_running = 0;
6625                 rq->calc_load_active = 0;
6626                 rq->calc_load_update = jiffies + LOAD_FREQ;
6627                 init_cfs_rq(&rq->cfs);
6628                 init_rt_rq(&rq->rt);
6629                 init_dl_rq(&rq->dl);
6630 #ifdef CONFIG_FAIR_GROUP_SCHED
6631                 root_task_group.shares = ROOT_TASK_GROUP_LOAD;
6632                 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
6633                 rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
6634                 /*
6635                  * How much CPU bandwidth does root_task_group get?
6636                  *
6637                  * In case of task-groups formed thr' the cgroup filesystem, it
6638                  * gets 100% of the CPU resources in the system. This overall
6639                  * system CPU resource is divided among the tasks of
6640                  * root_task_group and its child task-groups in a fair manner,
6641                  * based on each entity's (task or task-group's) weight
6642                  * (se->load.weight).
6643                  *
6644                  * In other words, if root_task_group has 10 tasks of weight
6645                  * 1024) and two child groups A0 and A1 (of weight 1024 each),
6646                  * then A0's share of the CPU resource is:
6647                  *
6648                  *      A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
6649                  *
6650                  * We achieve this by letting root_task_group's tasks sit
6651                  * directly in rq->cfs (i.e root_task_group->se[] = NULL).
6652                  */
6653                 init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
6654                 init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
6655 #endif /* CONFIG_FAIR_GROUP_SCHED */
6656 
6657                 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
6658 #ifdef CONFIG_RT_GROUP_SCHED
6659                 init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
6660 #endif
6661 #ifdef CONFIG_SMP
6662                 rq->sd = NULL;
6663                 rq->rd = NULL;
6664                 rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
6665                 rq->balance_callback = NULL;
6666                 rq->active_balance = 0;
6667                 rq->next_balance = jiffies;
6668                 rq->push_cpu = 0;
6669                 rq->cpu = i;
6670                 rq->online = 0;
6671                 rq->idle_stamp = 0;
6672                 rq->avg_idle = 2*sysctl_sched_migration_cost;
6673                 rq->max_idle_balance_cost = sysctl_sched_migration_cost;
6674 
6675                 INIT_LIST_HEAD(&rq->cfs_tasks);
6676 
6677                 rq_attach_root(rq, &def_root_domain);
6678 #ifdef CONFIG_NO_HZ_COMMON
6679                 rq->last_load_update_tick = jiffies;
6680                 rq->last_blocked_load_update_tick = jiffies;
6681                 atomic_set(&rq->nohz_flags, 0);
6682 #endif
6683 #endif /* CONFIG_SMP */
6684                 hrtick_rq_init(rq);
6685                 atomic_set(&rq->nr_iowait, 0);
6686         }
6687 
6688         set_load_weight(&init_task, false);
6689 
6690         /*
6691          * The boot idle thread does lazy MMU switching as well:
6692          */
6693         mmgrab(&init_mm);
6694         enter_lazy_tlb(&init_mm, current);
6695 
6696         /*
6697          * Make us the idle thread. Technically, schedule() should not be
6698          * called from this thread, however somewhere below it might be,
6699          * but because we are the idle thread, we just pick up running again
6700          * when this runqueue becomes "idle".
6701          */
6702         init_idle(current, smp_processor_id());
6703 
6704         calc_load_update = jiffies + LOAD_FREQ;
6705 
6706 #ifdef CONFIG_SMP
6707         idle_thread_set_boot_cpu();
6708 #endif
6709         init_sched_fair_class();
6710 
6711         init_schedstats();
6712 
6713         psi_init();
6714 
6715         init_uclamp();
6716 
6717         scheduler_running = 1;
6718 }
6719 
6720 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6721 static inline int preempt_count_equals(int preempt_offset)
6722 {
6723         int nested = preempt_count() + rcu_preempt_depth();
6724 
6725         return (nested == preempt_offset);
6726 }
6727 
6728 void __might_sleep(const char *file, int line, int preempt_offset)
6729 {
6730         /*
6731          * Blocking primitives will set (and therefore destroy) current->state,
6732          * since we will exit with TASK_RUNNING make sure we enter with it,
6733          * otherwise we will destroy state.
6734          */
6735         WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
6736                         "do not call blocking ops when !TASK_RUNNING; "
6737                         "state=%lx set at [<%p>] %pS\n",
6738                         current->state,
6739                         (void *)current->task_state_change,
6740                         (void *)current->task_state_change);
6741 
6742         ___might_sleep(file, line, preempt_offset);
6743 }
6744 EXPORT_SYMBOL(__might_sleep);
6745 
6746 void ___might_sleep(const char *file, int line, int preempt_offset)
6747 {
6748         /* Ratelimiting timestamp: */
6749         static unsigned long prev_jiffy;
6750 
6751         unsigned long preempt_disable_ip;
6752 
6753         /* WARN_ON_ONCE() by default, no rate limit required: */
6754         rcu_sleep_check();
6755 
6756         if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
6757              !is_idle_task(current) && !current->non_block_count) ||
6758             system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING ||
6759             oops_in_progress)
6760                 return;
6761 
6762         if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
6763                 return;
6764         prev_jiffy = jiffies;
6765 
6766         /* Save this before calling printk(), since that will clobber it: */
6767         preempt_disable_ip = get_preempt_disable_ip(current);
6768 
6769         printk(KERN_ERR
6770                 "BUG: sleeping function called from invalid context at %s:%d\n",
6771                         file, line);
6772         printk(KERN_ERR
6773                 "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n",
6774                         in_atomic(), irqs_disabled(), current->non_block_count,
6775                         current->pid, current->comm);
6776 
6777         if (task_stack_end_corrupted(current))
6778                 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
6779 
6780         debug_show_held_locks(current);
6781         if (irqs_disabled())
6782                 print_irqtrace_events(current);
6783         if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
6784             && !preempt_count_equals(preempt_offset)) {
6785                 pr_err("Preemption disabled at:");
6786                 print_ip_sym(preempt_disable_ip);
6787                 pr_cont("\n");
6788         }
6789         dump_stack();
6790         add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
6791 }
6792 EXPORT_SYMBOL(___might_sleep);
6793 
6794 void __cant_sleep(const char *file, int line, int preempt_offset)
6795 {
6796         static unsigned long prev_jiffy;
6797 
6798         if (irqs_disabled())
6799                 return;
6800 
6801         if (!IS_ENABLED(CONFIG_PREEMPT_COUNT))
6802                 return;
6803 
6804         if (preempt_count() > preempt_offset)
6805                 return;
6806 
6807         if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
6808                 return;
6809         prev_jiffy = jiffies;
6810 
6811         printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line);
6812         printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
6813                         in_atomic(), irqs_disabled(),
6814                         current->pid, current->comm);
6815 
6816         debug_show_held_locks(current);
6817         dump_stack();
6818         add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
6819 }
6820 EXPORT_SYMBOL_GPL(__cant_sleep);
6821 #endif
6822 
6823 #ifdef CONFIG_MAGIC_SYSRQ
6824 void normalize_rt_tasks(void)
6825 {
6826         struct task_struct *g, *p;
6827         struct sched_attr attr = {
6828                 .sched_policy = SCHED_NORMAL,
6829         };
6830 
6831         read_lock(&tasklist_lock);
6832         for_each_process_thread(g, p) {
6833                 /*
6834                  * Only normalize user tasks:
6835                  */
6836                 if (p->flags & PF_KTHREAD)
6837                         continue;
6838 
6839                 p->se.exec_start = 0;
6840                 schedstat_set(p->se.statistics.wait_start,  0);
6841                 schedstat_set(p->se.statistics.sleep_start, 0);
6842                 schedstat_set(p->se.statistics.block_start, 0);
6843 
6844                 if (!dl_task(p) && !rt_task(p)) {
6845                         /*
6846                          * Renice negative nice level userspace
6847                          * tasks back to 0:
6848                          */
6849                         if (task_nice(p) < 0)
6850                                 set_user_nice(p, 0);
6851                         continue;
6852                 }
6853 
6854                 __sched_setscheduler(p, &attr, false, false);
6855         }
6856         read_unlock(&tasklist_lock);
6857 }
6858 
6859 #endif /* CONFIG_MAGIC_SYSRQ */
6860 
6861 #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6862 /*
6863  * These functions are only useful for the IA64 MCA handling, or kdb.
6864  *
6865  * They can only be called when the whole system has been
6866  * stopped - every CPU needs to be quiescent, and no scheduling
6867  * activity can take place. Using them for anything else would
6868  * be a serious bug, and as a result, they aren't even visible
6869  * under any other configuration.
6870  */
6871 
6872 /**
6873  * curr_task - return the current task for a given CPU.
6874  * @cpu: the processor in question.
6875  *
6876  * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6877  *
6878  * Return: The current task for @cpu.
6879  */
6880 struct task_struct *curr_task(int cpu)
6881 {
6882         return cpu_curr(cpu);
6883 }
6884 
6885 #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
6886 
6887 #ifdef CONFIG_IA64
6888 /**
6889  * ia64_set_curr_task - set the current task for a given CPU.
6890  * @cpu: the processor in question.
6891  * @p: the task pointer to set.
6892  *
6893  * Description: This function must only be used when non-maskable interrupts
6894  * are serviced on a separate stack. It allows the architecture to switch the
6895  * notion of the current task on a CPU in a non-blocking manner. This function
6896  * must be called with all CPU's synchronized, and interrupts disabled, the
6897  * and caller must save the original value of the current task (see
6898  * curr_task() above) and restore that value before reenabling interrupts and
6899  * re-starting the system.
6900  *
6901  * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6902  */
6903 void ia64_set_curr_task(int cpu, struct task_struct *p)
6904 {
6905         cpu_curr(cpu) = p;
6906 }
6907 
6908 #endif
6909 
6910 #ifdef CONFIG_CGROUP_SCHED
6911 /* task_group_lock serializes the addition/removal of task groups */
6912 static DEFINE_SPINLOCK(task_group_lock);
6913 
6914 static inline void alloc_uclamp_sched_group(struct task_group *tg,
6915                                             struct task_group *parent)
6916 {
6917 #ifdef CONFIG_UCLAMP_TASK_GROUP
6918         enum uclamp_id clamp_id;
6919 
6920         for_each_clamp_id(clamp_id) {
6921                 uclamp_se_set(&tg->uclamp_req[clamp_id],
6922                               uclamp_none(clamp_id), false);
6923                 tg->uclamp[clamp_id] = parent->uclamp[clamp_id];
6924         }
6925 #endif
6926 }
6927 
6928 static void sched_free_group(struct task_group *tg)
6929 {
6930         free_fair_sched_group(tg);
6931         free_rt_sched_group(tg);
6932         autogroup_free(tg);
6933         kmem_cache_free(task_group_cache, tg);
6934 }
6935 
6936 /* allocate runqueue etc for a new task group */
6937 struct task_group *sched_create_group(struct task_group *parent)
6938 {
6939         struct task_group *tg;
6940 
6941         tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
6942         if (!tg)
6943                 return ERR_PTR(-ENOMEM);
6944 
6945         if (!alloc_fair_sched_group(tg, parent))
6946                 goto err;
6947 
6948         if (!alloc_rt_sched_group(tg, parent))
6949                 goto err;
6950 
6951         alloc_uclamp_sched_group(tg, parent);
6952 
6953         return tg;
6954 
6955 err:
6956         sched_free_group(tg);
6957         return ERR_PTR(-ENOMEM);
6958 }
6959 
6960 void sched_online_group(struct task_group *tg, struct task_group *parent)
6961 {
6962         unsigned long flags;
6963 
6964         spin_lock_irqsave(&task_group_lock, flags);
6965         list_add_rcu(&tg->list, &task_groups);
6966 
6967         /* Root should already exist: */
6968         WARN_ON(!parent);
6969 
6970         tg->parent = parent;
6971         INIT_LIST_HEAD(&tg->children);
6972         list_add_rcu(&tg->siblings, &parent->children);
6973         spin_unlock_irqrestore(&task_group_lock, flags);
6974 
6975         online_fair_sched_group(tg);
6976 }
6977 
6978 /* rcu callback to free various structures associated with a task group */
6979 static void sched_free_group_rcu(struct rcu_head *rhp)
6980 {
6981         /* Now it should be safe to free those cfs_rqs: */
6982         sched_free_group(container_of(rhp, struct task_group, rcu));
6983 }
6984 
6985 void sched_destroy_group(struct task_group *tg)
6986 {
6987         /* Wait for possible concurrent references to cfs_rqs complete: */
6988         call_rcu(&tg->rcu, sched_free_group_rcu);
6989 }
6990 
6991 void sched_offline_group(struct task_group *tg)
6992 {
6993         unsigned long flags;
6994 
6995         /* End participation in shares distribution: */
6996         unregister_fair_sched_group(tg);
6997 
6998         spin_lock_irqsave(&task_group_lock, flags);
6999         list_del_rcu(&tg->list);
7000         list_del_rcu(&tg->siblings);
7001         spin_unlock_irqrestore(&task_group_lock, flags);
7002 }
7003 
7004 static void sched_change_group(struct task_struct *tsk, int type)
7005 {
7006         struct task_group *tg;
7007 
7008         /*
7009          * All callers are synchronized by task_rq_lock(); we do not use RCU
7010          * which is pointless here. Thus, we pass "true" to task_css_check()
7011          * to prevent lockdep warnings.
7012          */
7013         tg = container_of(task_css_check(tsk, cpu_cgrp_id, true),
7014                           struct task_group, css);
7015         tg = autogroup_task_group(tsk, tg);
7016         tsk->sched_task_group = tg;
7017 
7018 #ifdef CONFIG_FAIR_GROUP_SCHED
7019         if (tsk->sched_class->task_change_group)
7020                 tsk->sched_class->task_change_group(tsk, type);
7021         else
7022 #endif
7023                 set_task_rq(tsk, task_cpu(tsk));
7024 }
7025 
7026 /*
7027  * Change task's runqueue when it moves between groups.
7028  *
7029  * The caller of this function should have put the task in its new group by
7030  * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect
7031  * its new group.
7032  */
7033 void sched_move_task(struct task_struct *tsk)
7034 {
7035         int queued, running, queue_flags =
7036                 DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
7037         struct rq_flags rf;
7038         struct rq *rq;
7039 
7040         rq = task_rq_lock(tsk, &rf);
7041         update_rq_clock(rq);
7042 
7043         running = task_current(rq, tsk);
7044         queued = task_on_rq_queued(tsk);
7045 
7046         if (queued)
7047                 dequeue_task(rq, tsk, queue_flags);
7048         if (running)
7049                 put_prev_task(rq, tsk);
7050 
7051         sched_change_group(tsk, TASK_MOVE_GROUP);
7052 
7053         if (queued)
7054                 enqueue_task(rq, tsk, queue_flags);
7055         if (running) {
7056                 set_next_task(rq, tsk);
7057                 /*
7058                  * After changing group, the running task may have joined a
7059                  * throttled one but it's still the running task. Trigger a
7060                  * resched to make sure that task can still run.
7061                  */
7062                 resched_curr(rq);
7063         }
7064 
7065         task_rq_unlock(rq, tsk, &rf);
7066 }
7067 
7068 static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
7069 {
7070         return css ? container_of(css, struct task_group, css) : NULL;
7071 }
7072 
7073 static struct cgroup_subsys_state *
7074 cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
7075 {
7076         struct task_group *parent = css_tg(parent_css);
7077         struct task_group *tg;
7078 
7079         if (!parent) {
7080                 /* This is early initialization for the top cgroup */
7081                 return &root_task_group.css;
7082         }
7083 
7084         tg = sched_create_group(parent);
7085         if (IS_ERR(tg))
7086                 return ERR_PTR(-ENOMEM);
7087 
7088         return &tg->css;
7089 }
7090 
7091 /* Expose task group only after completing cgroup initialization */
7092 static int cpu_cgroup_css_online(struct cgroup_subsys_state *css)
7093 {
7094         struct task_group *tg = css_tg(css);
7095         struct task_group *parent = css_tg(css->parent);
7096 
7097         if (parent)
7098                 sched_online_group(tg, parent);
7099 
7100 #ifdef CONFIG_UCLAMP_TASK_GROUP
7101         /* Propagate the effective uclamp value for the new group */
7102         cpu_util_update_eff(css);
7103 #endif
7104 
7105         return 0;
7106 }
7107 
7108 static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
7109 {
7110         struct task_group *tg = css_tg(css);
7111 
7112         sched_offline_group(tg);
7113 }
7114 
7115 static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
7116 {
7117         struct task_group *tg = css_tg(css);
7118 
7119         /*
7120          * Relies on the RCU grace period between css_released() and this.
7121          */
7122         sched_free_group(tg);
7123 }
7124 
7125 /*
7126  * This is called before wake_up_new_task(), therefore we really only
7127  * have to set its group bits, all the other stuff does not apply.
7128  */
7129 static void cpu_cgroup_fork(struct task_struct *task)
7130 {
7131         struct rq_flags rf;
7132         struct rq *rq;
7133 
7134         rq = task_rq_lock(task, &rf);
7135 
7136         update_rq_clock(rq);
7137         sched_change_group(task, TASK_SET_GROUP);
7138 
7139         task_rq_unlock(rq, task, &rf);
7140 }
7141 
7142 static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
7143 {
7144         struct task_struct *task;
7145         struct cgroup_subsys_state *css;
7146         int ret = 0;
7147 
7148         cgroup_taskset_for_each(task, css, tset) {
7149 #ifdef CONFIG_RT_GROUP_SCHED
7150                 if (!sched_rt_can_attach(css_tg(css), task))
7151                         return -EINVAL;
7152 #endif
7153                 /*
7154                  * Serialize against wake_up_new_task() such that if its
7155                  * running, we're sure to observe its full state.
7156                  */
7157                 raw_spin_lock_irq(&task->pi_lock);
7158                 /*
7159                  * Avoid calling sched_move_task() before wake_up_new_task()
7160                  * has happened. This would lead to problems with PELT, due to
7161                  * move wanting to detach+attach while we're not attached yet.
7162                  */
7163                 if (task->state == TASK_NEW)
7164                         ret = -EINVAL;
7165                 raw_spin_unlock_irq(&task->pi_lock);
7166 
7167                 if (ret)
7168                         break;
7169         }
7170         return ret;
7171 }
7172 
7173 static void cpu_cgroup_attach(struct cgroup_taskset *tset)
7174 {
7175         struct task_struct *task;
7176         struct cgroup_subsys_state *css;
7177 
7178         cgroup_taskset_for_each(task, css, tset)
7179                 sched_move_task(task);
7180 }
7181 
7182 #ifdef CONFIG_UCLAMP_TASK_GROUP
7183 static void cpu_util_update_eff(struct cgroup_subsys_state *css)
7184 {
7185         struct cgroup_subsys_state *top_css = css;
7186         struct uclamp_se *uc_parent = NULL;
7187         struct uclamp_se *uc_se = NULL;
7188         unsigned int eff[UCLAMP_CNT];
7189         enum uclamp_id clamp_id;
7190         unsigned int clamps;
7191 
7192         css_for_each_descendant_pre(css, top_css) {
7193                 uc_parent = css_tg(css)->parent
7194                         ? css_tg(css)->parent->uclamp : NULL;
7195 
7196                 for_each_clamp_id(clamp_id) {
7197                         /* Assume effective clamps matches requested clamps */
7198                         eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value;
7199                         /* Cap effective clamps with parent's effective clamps */
7200                         if (uc_parent &&
7201                             eff[clamp_id] > uc_parent[clamp_id].value) {
7202                                 eff[clamp_id] = uc_parent[clamp_id].value;
7203                         }
7204                 }
7205                 /* Ensure protection is always capped by limit */
7206                 eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]);
7207 
7208                 /* Propagate most restrictive effective clamps */
7209                 clamps = 0x0;
7210                 uc_se = css_tg(css)->uclamp;
7211                 for_each_clamp_id(clamp_id) {
7212                         if (eff[clamp_id] == uc_se[clamp_id].value)
7213                                 continue;
7214                         uc_se[clamp_id].value = eff[clamp_id];
7215                         uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]);
7216                         clamps |= (0x1 << clamp_id);
7217                 }
7218                 if (!clamps) {
7219                         css = css_rightmost_descendant(css);
7220                         continue;
7221                 }
7222 
7223                 /* Immediately update descendants RUNNABLE tasks */
7224                 uclamp_update_active_tasks(css, clamps);
7225         }
7226 }
7227 
7228 /*
7229  * Integer 10^N with a given N exponent by casting to integer the literal "1eN"
7230  * C expression. Since there is no way to convert a macro argument (N) into a
7231  * character constant, use two levels of macros.
7232  */
7233 #define _POW10(exp) ((unsigned int)1e##exp)
7234 #define POW10(exp) _POW10(exp)
7235 
7236 struct uclamp_request {
7237 #define UCLAMP_PERCENT_SHIFT    2
7238 #define UCLAMP_PERCENT_SCALE    (100 * POW10(UCLAMP_PERCENT_SHIFT))
7239         s64 percent;
7240         u64 util;
7241         int ret;
7242 };
7243 
7244 static inline struct uclamp_request
7245 capacity_from_percent(char *buf)
7246 {
7247         struct uclamp_request req = {
7248                 .percent = UCLAMP_PERCENT_SCALE,
7249                 .util = SCHED_CAPACITY_SCALE,
7250                 .ret = 0,
7251         };
7252 
7253         buf = strim(buf);
7254         if (strcmp(buf, "max")) {
7255                 req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT,
7256                                              &req.percent);
7257                 if (req.ret)
7258                         return req;
7259                 if ((u64)req.percent > UCLAMP_PERCENT_SCALE) {
7260                         req.ret = -ERANGE;
7261                         return req;
7262                 }
7263 
7264                 req.util = req.percent << SCHED_CAPACITY_SHIFT;
7265                 req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE);
7266         }
7267 
7268         return req;
7269 }
7270 
7271 static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf,
7272                                 size_t nbytes, loff_t off,
7273                                 enum uclamp_id clamp_id)
7274 {
7275         struct uclamp_request req;
7276         struct task_group *tg;
7277 
7278         req = capacity_from_percent(buf);
7279         if (req.ret)
7280                 return req.ret;
7281 
7282         mutex_lock(&uclamp_mutex);
7283         rcu_read_lock();
7284 
7285         tg = css_tg(of_css(of));
7286         if (tg->uclamp_req[clamp_id].value != req.util)
7287                 uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false);
7288 
7289         /*
7290          * Because of not recoverable conversion rounding we keep track of the
7291          * exact requested value
7292          */
7293         tg->uclamp_pct[clamp_id] = req.percent;
7294 
7295         /* Update effective clamps to track the most restrictive value */
7296         cpu_util_update_eff(of_css(of));
7297 
7298         rcu_read_unlock();
7299         mutex_unlock(&uclamp_mutex);
7300 
7301         return nbytes;
7302 }
7303 
7304 static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of,
7305                                     char *buf, size_t nbytes,
7306                                     loff_t off)
7307 {
7308         return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN);
7309 }
7310 
7311 static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of,
7312                                     char *buf, size_t nbytes,
7313                                     loff_t off)
7314 {
7315         return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX);
7316 }
7317 
7318 static inline void cpu_uclamp_print(struct seq_file *sf,
7319                                     enum uclamp_id clamp_id)
7320 {
7321         struct task_group *tg;
7322         u64 util_clamp;
7323         u64 percent;
7324         u32 rem;
7325 
7326         rcu_read_lock();
7327         tg = css_tg(seq_css(sf));
7328         util_clamp = tg->uclamp_req[clamp_id].value;
7329         rcu_read_unlock();
7330 
7331         if (util_clamp == SCHED_CAPACITY_SCALE) {
7332                 seq_puts(sf, "max\n");
7333                 return;
7334         }
7335 
7336         percent = tg->uclamp_pct[clamp_id];
7337         percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem);
7338         seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem);
7339 }
7340 
7341 static int cpu_uclamp_min_show(struct seq_file *sf, void *v)
7342 {
7343         cpu_uclamp_print(sf, UCLAMP_MIN);
7344         return 0;
7345 }
7346 
7347 static int cpu_uclamp_max_show(struct seq_file *sf, void *v)
7348 {
7349         cpu_uclamp_print(sf, UCLAMP_MAX);
7350         return 0;
7351 }
7352 #endif /* CONFIG_UCLAMP_TASK_GROUP */
7353 
7354 #ifdef CONFIG_FAIR_GROUP_SCHED
7355 static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
7356                                 struct cftype *cftype, u64 shareval)
7357 {
7358         if (shareval > scale_load_down(ULONG_MAX))
7359                 shareval = MAX_SHARES;
7360         return sched_group_set_shares(css_tg(css), scale_load(shareval));
7361 }
7362 
7363 static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
7364                                struct cftype *cft)
7365 {
7366         struct task_group *tg = css_tg(css);
7367 
7368         return (u64) scale_load_down(tg->shares);
7369 }
7370 
7371 #ifdef CONFIG_CFS_BANDWIDTH
7372 static DEFINE_MUTEX(cfs_constraints_mutex);
7373 
7374 const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
7375 static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */
7376 
7377 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);
7378 
7379 static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
7380 {
7381         int i, ret = 0, runtime_enabled, runtime_was_enabled;
7382         struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
7383 
7384         if (tg == &root_task_group)
7385                 return -EINVAL;
7386 
7387         /*
7388          * Ensure we have at some amount of bandwidth every period.  This is
7389          * to prevent reaching a state of large arrears when throttled via
7390          * entity_tick() resulting in prolonged exit starvation.
7391          */
7392         if (quota < min_cfs_quota_period || period < min_cfs_quota_period)
7393                 return -EINVAL;
7394 
7395         /*
7396          * Likewise, bound things on the otherside by preventing insane quota
7397          * periods.  This also allows us to normalize in computing quota
7398          * feasibility.
7399          */
7400         if (period > max_cfs_quota_period)
7401                 return -EINVAL;
7402 
7403         /*
7404          * Prevent race between setting of cfs_rq->runtime_enabled and
7405          * unthrottle_offline_cfs_rqs().
7406          */
7407         get_online_cpus();
7408         mutex_lock(&cfs_constraints_mutex);
7409         ret = __cfs_schedulable(tg, period, quota);
7410         if (ret)
7411                 goto out_unlock;
7412 
7413         runtime_enabled = quota != RUNTIME_INF;
7414         runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
7415         /*
7416          * If we need to toggle cfs_bandwidth_used, off->on must occur
7417          * before making related changes, and on->off must occur afterwards
7418          */
7419         if (runtime_enabled && !runtime_was_enabled)
7420                 cfs_bandwidth_usage_inc();
7421         raw_spin_lock_irq(&cfs_b->lock);
7422         cfs_b->period = ns_to_ktime(period);
7423         cfs_b->quota = quota;
7424 
7425         __refill_cfs_bandwidth_runtime(cfs_b);
7426 
7427         /* Restart the period timer (if active) to handle new period expiry: */
7428         if (runtime_enabled)
7429                 start_cfs_bandwidth(cfs_b);
7430 
7431         raw_spin_unlock_irq(&cfs_b->lock);
7432 
7433         for_each_online_cpu(i) {
7434                 struct cfs_rq *cfs_rq = tg->cfs_rq[i];
7435                 struct rq *rq = cfs_rq->rq;
7436                 struct rq_flags rf;
7437 
7438                 rq_lock_irq(rq, &rf);
7439                 cfs_rq->runtime_enabled = runtime_enabled;
7440                 cfs_rq->runtime_remaining = 0;
7441 
7442                 if (cfs_rq->throttled)
7443                         unthrottle_cfs_rq(cfs_rq);
7444                 rq_unlock_irq(rq, &rf);
7445         }
7446         if (runtime_was_enabled && !runtime_enabled)
7447                 cfs_bandwidth_usage_dec();
7448 out_unlock:
7449         mutex_unlock(&cfs_constraints_mutex);
7450         put_online_cpus();
7451 
7452         return ret;
7453 }
7454 
7455 static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
7456 {
7457         u64 quota, period;
7458 
7459         period = ktime_to_ns(tg->cfs_bandwidth.period);
7460         if (cfs_quota_us < 0)
7461                 quota = RUNTIME_INF;
7462         else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC)
7463                 quota = (u64)cfs_quota_us * NSEC_PER_USEC;
7464         else
7465                 return -EINVAL;
7466 
7467         return tg_set_cfs_bandwidth(tg, period, quota);
7468 }
7469 
7470 static long tg_get_cfs_quota(struct task_group *tg)
7471 {
7472         u64 quota_us;
7473 
7474         if (tg->cfs_bandwidth.quota == RUNTIME_INF)
7475                 return -1;
7476 
7477         quota_us = tg->cfs_bandwidth.quota;
7478         do_div(quota_us, NSEC_PER_USEC);
7479 
7480         return quota_us;
7481 }
7482 
7483 static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
7484 {
7485         u64 quota, period;
7486 
7487         if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC)
7488                 return -EINVAL;
7489 
7490         period = (u64)cfs_period_us * NSEC_PER_USEC;
7491         quota = tg->cfs_bandwidth.quota;
7492 
7493         return tg_set_cfs_bandwidth(tg, period, quota);
7494 }
7495 
7496 static long tg_get_cfs_period(struct task_group *tg)
7497 {
7498         u64 cfs_period_us;
7499 
7500         cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
7501         do_div(cfs_period_us, NSEC_PER_USEC);
7502 
7503         return cfs_period_us;
7504 }
7505 
7506 static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
7507                                   struct cftype *cft)
7508 {
7509         return tg_get_cfs_quota(css_tg(css));
7510 }
7511 
7512 static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
7513                                    struct cftype *cftype, s64 cfs_quota_us)
7514 {
7515         return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
7516 }
7517 
7518 static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
7519                                    struct cftype *cft)
7520 {
7521         return tg_get_cfs_period(css_tg(css));
7522 }
7523 
7524 static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
7525                                     struct cftype *cftype, u64 cfs_period_us)
7526 {
7527         return tg_set_cfs_period(css_tg(css), cfs_period_us);
7528 }
7529 
7530 struct cfs_schedulable_data {
7531         struct task_group *tg;
7532         u64 period, quota;
7533 };
7534 
7535 /*
7536  * normalize group quota/period to be quota/max_period
7537  * note: units are usecs
7538  */
7539 static u64 normalize_cfs_quota(struct task_group *tg,
7540                                struct cfs_schedulable_data *d)
7541 {
7542         u64 quota, period;
7543 
7544         if (tg == d->tg) {
7545                 period = d->period;
7546                 quota = d->quota;
7547         } else {
7548                 period = tg_get_cfs_period(tg);
7549                 quota = tg_get_cfs_quota(tg);
7550         }
7551 
7552         /* note: these should typically be equivalent */
7553         if (quota == RUNTIME_INF || quota == -1)
7554                 return RUNTIME_INF;
7555 
7556         return to_ratio(period, quota);
7557 }
7558 
7559 static int tg_cfs_schedulable_down(struct task_group *tg, void *data)
7560 {
7561         struct cfs_schedulable_data *d = data;
7562         struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
7563         s64 quota = 0, parent_quota = -1;
7564 
7565         if (!tg->parent) {
7566                 quota = RUNTIME_INF;
7567         } else {
7568                 struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
7569 
7570                 quota = normalize_cfs_quota(tg, d);
7571                 parent_quota = parent_b->hierarchical_quota;
7572 
7573                 /*
7574                  * Ensure max(child_quota) <= parent_quota.  On cgroup2,
7575                  * always take the min.  On cgroup1, only inherit when no
7576                  * limit is set:
7577                  */
7578                 if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) {
7579                         quota = min(quota, parent_quota);
7580                 } else {
7581                         if (quota == RUNTIME_INF)
7582                                 quota = parent_quota;
7583                         else if (parent_quota != RUNTIME_INF && quota > parent_quota)
7584                                 return -EINVAL;
7585                 }
7586         }
7587         cfs_b->hierarchical_quota = quota;
7588 
7589         return 0;
7590 }
7591 
7592 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
7593 {
7594         int ret;
7595         struct cfs_schedulable_data data = {
7596                 .tg = tg,
7597                 .period = period,
7598                 .quota = quota,
7599         };
7600 
7601         if (quota != RUNTIME_INF) {
7602                 do_div(data.period, NSEC_PER_USEC);
7603                 do_div(data.quota, NSEC_PER_USEC);
7604         }
7605 
7606         rcu_read_lock();
7607         ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
7608         rcu_read_unlock();
7609 
7610         return ret;
7611 }
7612 
7613 static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
7614 {
7615         struct task_group *tg = css_tg(seq_css(sf));
7616         struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
7617 
7618         seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods);
7619         seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled);
7620         seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time);
7621 
7622         if (schedstat_enabled() && tg != &root_task_group) {
7623                 u64 ws = 0;
7624                 int i;
7625 
7626                 for_each_possible_cpu(i)
7627                         ws += schedstat_val(tg->se[i]->statistics.wait_sum);
7628 
7629                 seq_printf(sf, "wait_sum %llu\n", ws);
7630         }
7631 
7632         return 0;
7633 }
7634 #endif /* CONFIG_CFS_BANDWIDTH */
7635 #endif /* CONFIG_FAIR_GROUP_SCHED */
7636 
7637 #ifdef CONFIG_RT_GROUP_SCHED
7638 static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
7639                                 struct cftype *cft, s64 val)
7640 {
7641         return sched_group_set_rt_runtime(css_tg(css), val);
7642 }
7643 
7644 static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
7645                                struct cftype *cft)
7646 {
7647         return sched_group_rt_runtime(css_tg(css));
7648 }
7649 
7650 static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
7651                                     struct cftype *cftype, u64 rt_period_us)
7652 {
7653         return sched_group_set_rt_period(css_tg(css), rt_period_us);
7654 }
7655 
7656 static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
7657                                    struct cftype *cft)
7658 {
7659         return sched_group_rt_period(css_tg(css));
7660 }
7661 #endif /* CONFIG_RT_GROUP_SCHED */
7662 
7663 static struct cftype cpu_legacy_files[] = {
7664 #ifdef CONFIG_FAIR_GROUP_SCHED
7665         {
7666                 .name = "shares",
7667                 .read_u64 = cpu_shares_read_u64,
7668                 .write_u64 = cpu_shares_write_u64,
7669         },
7670 #endif
7671 #ifdef CONFIG_CFS_BANDWIDTH
7672         {
7673                 .name = "cfs_quota_us",
7674                 .read_s64 = cpu_cfs_quota_read_s64,
7675                 .write_s64 = cpu_cfs_quota_write_s64,
7676         },
7677         {
7678                 .name = "cfs_period_us",
7679                 .read_u64 = cpu_cfs_period_read_u64,
7680                 .write_u64 = cpu_cfs_period_write_u64,
7681         },
7682         {
7683                 .name = "stat",
7684                 .seq_show = cpu_cfs_stat_show,
7685         },
7686 #endif
7687 #ifdef CONFIG_RT_GROUP_SCHED
7688         {
7689                 .name = "rt_runtime_us",
7690                 .read_s64 = cpu_rt_runtime_read,
7691                 .write_s64 = cpu_rt_runtime_write,
7692         },
7693         {
7694                 .name = "rt_period_us",
7695                 .read_u64 = cpu_rt_period_read_uint,
7696                 .write_u64 = cpu_rt_period_write_uint,
7697         },
7698 #endif
7699 #ifdef CONFIG_UCLAMP_TASK_GROUP
7700         {
7701                 .name = "uclamp.min",
7702                 .flags = CFTYPE_NOT_ON_ROOT,
7703                 .seq_show = cpu_uclamp_min_show,
7704                 .write = cpu_uclamp_min_write,
7705         },
7706         {
7707                 .name = "uclamp.max",
7708                 .flags = CFTYPE_NOT_ON_ROOT,
7709                 .seq_show = cpu_uclamp_max_show,
7710                 .write = cpu_uclamp_max_write,
7711         },
7712 #endif
7713         { }     /* Terminate */
7714 };
7715 
7716 static int cpu_extra_stat_show(struct seq_file *sf,
7717                                struct cgroup_subsys_state *css)
7718 {
7719 #ifdef CONFIG_CFS_BANDWIDTH
7720         {
7721                 struct task_group *tg = css_tg(css);
7722                 struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
7723                 u64 throttled_usec;
7724 
7725                 throttled_usec = cfs_b->throttled_time;
7726                 do_div(throttled_usec, NSEC_PER_USEC);
7727 
7728                 seq_printf(sf, "nr_periods %d\n"
7729                            "nr_throttled %d\n"
7730                            "throttled_usec %llu\n",
7731                            cfs_b->nr_periods, cfs_b->nr_throttled,
7732                            throttled_usec);
7733         }
7734 #endif
7735         return 0;
7736 }
7737 
7738 #ifdef CONFIG_FAIR_GROUP_SCHED
7739 static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css,
7740                                struct cftype *cft)
7741 {
7742         struct task_group *tg = css_tg(css);
7743         u64 weight = scale_load_down(tg->shares);
7744 
7745         return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024);
7746 }
7747 
7748 static int cpu_weight_write_u64(struct cgroup_subsys_state *css,
7749                                 struct cftype *cft, u64 weight)
7750 {
7751         /*
7752          * cgroup weight knobs should use the common MIN, DFL and MAX
7753          * values which are 1, 100 and 10000 respectively.  While it loses
7754          * a bit of range on both ends, it maps pretty well onto the shares
7755          * value used by scheduler and the round-trip conversions preserve
7756          * the original value over the entire range.
7757          */
7758         if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX)
7759                 return -ERANGE;
7760 
7761         weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL);
7762 
7763         return sched_group_set_shares(css_tg(css), scale_load(weight));
7764 }
7765 
7766 static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css,
7767                                     struct cftype *cft)
7768 {
7769         unsigned long weight = scale_load_down(css_tg(css)->shares);
7770         int last_delta = INT_MAX;
7771         int prio, delta;
7772 
7773         /* find the closest nice value to the current weight */
7774         for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) {
7775                 delta = abs(sched_prio_to_weight[prio] - weight);
7776                 if (delta >= last_delta)
7777                         break;
7778                 last_delta = delta;
7779         }
7780 
7781         return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO);
7782 }
7783 
7784 static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css,
7785                                      struct cftype *cft, s64 nice)
7786 {
7787         unsigned long weight;
7788         int idx;
7789 
7790         if (nice < MIN_NICE || nice > MAX_NICE)
7791                 return -ERANGE;
7792 
7793         idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO;
7794         idx = array_index_nospec(idx, 40);
7795         weight = sched_prio_to_weight[idx];
7796 
7797         return sched_group_set_shares(css_tg(css), scale_load(weight));
7798 }
7799 #endif
7800 
7801 static void __maybe_unused cpu_period_quota_print(struct seq_file *sf,
7802                                                   long period, long quota)
7803 {
7804         if (quota < 0)
7805                 seq_puts(sf, "max");
7806         else
7807                 seq_printf(sf, "%ld", quota);
7808 
7809         seq_printf(sf, " %ld\n", period);
7810 }
7811 
7812 /* caller should put the current value in *@periodp before calling */
7813 static int __maybe_unused cpu_period_quota_parse(char *buf,
7814                                                  u64 *periodp, u64 *quotap)
7815 {
7816         char tok[21];   /* U64_MAX */
7817 
7818         if (sscanf(buf, "%20s %llu", tok, periodp) < 1)
7819                 return -EINVAL;
7820 
7821         *periodp *= NSEC_PER_USEC;
7822 
7823         if (sscanf(tok, "%llu", quotap))
7824                 *quotap *= NSEC_PER_USEC;
7825         else if (!strcmp(tok, "max"))
7826                 *quotap = RUNTIME_INF;
7827         else
7828                 return -EINVAL;
7829 
7830         return 0;
7831 }
7832 
7833 #ifdef CONFIG_CFS_BANDWIDTH
7834 static int cpu_max_show(struct seq_file *sf, void *v)
7835 {
7836         struct task_group *tg = css_tg(seq_css(sf));
7837 
7838         cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg));
7839         return 0;
7840 }
7841 
7842 static ssize_t cpu_max_write(struct kernfs_open_file *of,
7843                              char *buf, size_t nbytes, loff_t off)
7844 {
7845         struct task_group *tg = css_tg(of_css(of));
7846         u64 period = tg_get_cfs_period(tg);
7847         u64 quota;
7848         int ret;
7849 
7850         ret = cpu_period_quota_parse(buf, &period, &quota);
7851         if (!ret)
7852                 ret = tg_set_cfs_bandwidth(tg, period, quota);
7853         return ret ?: nbytes;
7854 }
7855 #endif
7856 
7857 static struct cftype cpu_files[] = {
7858 #ifdef CONFIG_FAIR_GROUP_SCHED
7859         {
7860                 .name = "weight",
7861                 .flags = CFTYPE_NOT_ON_ROOT,
7862                 .read_u64 = cpu_weight_read_u64,
7863                 .write_u64 = cpu_weight_write_u64,
7864         },
7865         {
7866                 .name = "weight.nice",
7867                 .flags = CFTYPE_NOT_ON_ROOT,
7868                 .read_s64 = cpu_weight_nice_read_s64,
7869                 .write_s64 = cpu_weight_nice_write_s64,
7870         },
7871 #endif
7872 #ifdef CONFIG_CFS_BANDWIDTH
7873         {
7874                 .name = "max",
7875                 .flags = CFTYPE_NOT_ON_ROOT,
7876                 .seq_show = cpu_max_show,
7877                 .write = cpu_max_write,
7878         },
7879 #endif
7880 #ifdef CONFIG_UCLAMP_TASK_GROUP
7881         {
7882                 .name = "uclamp.min",
7883                 .flags = CFTYPE_NOT_ON_ROOT,
7884                 .seq_show = cpu_uclamp_min_show,
7885                 .write = cpu_uclamp_min_write,
7886         },
7887         {
7888                 .name = "uclamp.max",
7889                 .flags = CFTYPE_NOT_ON_ROOT,
7890                 .seq_show = cpu_uclamp_max_show,
7891                 .write = cpu_uclamp_max_write,
7892         },
7893 #endif
7894         { }     /* terminate */
7895 };
7896 
7897 struct cgroup_subsys cpu_cgrp_subsys = {
7898         .css_alloc      = cpu_cgroup_css_alloc,
7899         .css_online     = cpu_cgroup_css_online,
7900         .css_released   = cpu_cgroup_css_released,
7901         .css_free       = cpu_cgroup_css_free,
7902         .css_extra_stat_show = cpu_extra_stat_show,
7903         .fork           = cpu_cgroup_fork,
7904         .can_attach     = cpu_cgroup_can_attach,
7905         .attach         = cpu_cgroup_attach,
7906         .legacy_cftypes = cpu_legacy_files,
7907         .dfl_cftypes    = cpu_files,
7908         .early_init     = true,
7909         .threaded       = true,
7910 };
7911 
7912 #endif  /* CONFIG_CGROUP_SCHED */
7913 
7914 void dump_cpu_task(int cpu)
7915 {
7916         pr_info("Task dump for CPU %d:\n", cpu);
7917         sched_show_task(cpu_curr(cpu));
7918 }
7919 
7920 /*
7921  * Nice levels are multiplicative, with a gentle 10% change for every
7922  * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
7923  * nice 1, it will get ~10% less CPU time than another CPU-bound task
7924  * that remained on nice 0.
7925  *
7926  * The "10% effect" is relative and cumulative: from _any_ nice level,
7927  * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
7928  * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
7929  * If a task goes up by ~10% and another task goes down by ~10% then
7930  * the relative distance between them is ~25%.)
7931  */
7932 const int sched_prio_to_weight[40] = {
7933  /* -20 */     88761,     71755,     56483,     46273,     36291,
7934  /* -15 */     29154,     23254,     18705,     14949,     11916,
7935  /* -10 */      9548,      7620,      6100,      4904,      3906,
7936  /*  -5 */      3121,      2501,      1991,      1586,      1277,
7937  /*   0 */      1024,       820,       655,       526,       423,
7938  /*   5 */       335,       272,       215,       172,       137,
7939  /*  10 */       110,        87,        70,        56,        45,
7940  /*  15 */        36,        29,        23,        18,        15,
7941 };
7942 
7943 /*
7944  * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated.
7945  *
7946  * In cases where the weight does not change often, we can use the
7947  * precalculated inverse to speed up arithmetics by turning divisions
7948  * into multiplications:
7949  */
7950 const u32 sched_prio_to_wmult[40] = {
7951  /* -20 */     48388,     59856,     76040,     92818,    118348,
7952  /* -15 */    147320,    184698,    229616,    287308,    360437,
7953  /* -10 */    449829,    563644,    704093,    875809,   1099582,
7954  /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
7955  /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
7956  /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
7957  /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
7958  /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
7959 };
7960 
7961 #undef CREATE_TRACE_POINTS