~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/sched/core.c

Version: ~ [ linux-4.19-rc4 ] ~ [ linux-4.18.8 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.70 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.127 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.156 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.122 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.57 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.31.14 ] ~ [ linux-2.6.30.10 ] ~ [ linux-2.6.29.6 ] ~ [ linux-2.6.28.10 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  kernel/sched/core.c
  3  *
  4  *  Core kernel scheduler code and related syscalls
  5  *
  6  *  Copyright (C) 1991-2002  Linus Torvalds
  7  */
  8 #include "sched.h"
  9 
 10 #include <linux/nospec.h>
 11 
 12 #include <linux/kcov.h>
 13 
 14 #include <asm/switch_to.h>
 15 #include <asm/tlb.h>
 16 
 17 #include "../workqueue_internal.h"
 18 #include "../smpboot.h"
 19 
 20 #include "pelt.h"
 21 
 22 #define CREATE_TRACE_POINTS
 23 #include <trace/events/sched.h>
 24 
 25 DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
 26 
 27 #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
 28 /*
 29  * Debugging: various feature bits
 30  *
 31  * If SCHED_DEBUG is disabled, each compilation unit has its own copy of
 32  * sysctl_sched_features, defined in sched.h, to allow constants propagation
 33  * at compile time and compiler optimization based on features default.
 34  */
 35 #define SCHED_FEAT(name, enabled)       \
 36         (1UL << __SCHED_FEAT_##name) * enabled |
 37 const_debug unsigned int sysctl_sched_features =
 38 #include "features.h"
 39         0;
 40 #undef SCHED_FEAT
 41 #endif
 42 
 43 /*
 44  * Number of tasks to iterate in a single balance run.
 45  * Limited because this is done with IRQs disabled.
 46  */
 47 const_debug unsigned int sysctl_sched_nr_migrate = 32;
 48 
 49 /*
 50  * period over which we measure -rt task CPU usage in us.
 51  * default: 1s
 52  */
 53 unsigned int sysctl_sched_rt_period = 1000000;
 54 
 55 __read_mostly int scheduler_running;
 56 
 57 /*
 58  * part of the period that we allow rt tasks to run in us.
 59  * default: 0.95s
 60  */
 61 int sysctl_sched_rt_runtime = 950000;
 62 
 63 /*
 64  * __task_rq_lock - lock the rq @p resides on.
 65  */
 66 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
 67         __acquires(rq->lock)
 68 {
 69         struct rq *rq;
 70 
 71         lockdep_assert_held(&p->pi_lock);
 72 
 73         for (;;) {
 74                 rq = task_rq(p);
 75                 raw_spin_lock(&rq->lock);
 76                 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
 77                         rq_pin_lock(rq, rf);
 78                         return rq;
 79                 }
 80                 raw_spin_unlock(&rq->lock);
 81 
 82                 while (unlikely(task_on_rq_migrating(p)))
 83                         cpu_relax();
 84         }
 85 }
 86 
 87 /*
 88  * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
 89  */
 90 struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
 91         __acquires(p->pi_lock)
 92         __acquires(rq->lock)
 93 {
 94         struct rq *rq;
 95 
 96         for (;;) {
 97                 raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
 98                 rq = task_rq(p);
 99                 raw_spin_lock(&rq->lock);
100                 /*
101                  *      move_queued_task()              task_rq_lock()
102                  *
103                  *      ACQUIRE (rq->lock)
104                  *      [S] ->on_rq = MIGRATING         [L] rq = task_rq()
105                  *      WMB (__set_task_cpu())          ACQUIRE (rq->lock);
106                  *      [S] ->cpu = new_cpu             [L] task_rq()
107                  *                                      [L] ->on_rq
108                  *      RELEASE (rq->lock)
109                  *
110                  * If we observe the old CPU in task_rq_lock, the acquire of
111                  * the old rq->lock will fully serialize against the stores.
112                  *
113                  * If we observe the new CPU in task_rq_lock, the acquire will
114                  * pair with the WMB to ensure we must then also see migrating.
115                  */
116                 if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
117                         rq_pin_lock(rq, rf);
118                         return rq;
119                 }
120                 raw_spin_unlock(&rq->lock);
121                 raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
122 
123                 while (unlikely(task_on_rq_migrating(p)))
124                         cpu_relax();
125         }
126 }
127 
128 /*
129  * RQ-clock updating methods:
130  */
131 
132 static void update_rq_clock_task(struct rq *rq, s64 delta)
133 {
134 /*
135  * In theory, the compile should just see 0 here, and optimize out the call
136  * to sched_rt_avg_update. But I don't trust it...
137  */
138 #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
139         s64 steal = 0, irq_delta = 0;
140 #endif
141 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
142         irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
143 
144         /*
145          * Since irq_time is only updated on {soft,}irq_exit, we might run into
146          * this case when a previous update_rq_clock() happened inside a
147          * {soft,}irq region.
148          *
149          * When this happens, we stop ->clock_task and only update the
150          * prev_irq_time stamp to account for the part that fit, so that a next
151          * update will consume the rest. This ensures ->clock_task is
152          * monotonic.
153          *
154          * It does however cause some slight miss-attribution of {soft,}irq
155          * time, a more accurate solution would be to update the irq_time using
156          * the current rq->clock timestamp, except that would require using
157          * atomic ops.
158          */
159         if (irq_delta > delta)
160                 irq_delta = delta;
161 
162         rq->prev_irq_time += irq_delta;
163         delta -= irq_delta;
164 #endif
165 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
166         if (static_key_false((&paravirt_steal_rq_enabled))) {
167                 steal = paravirt_steal_clock(cpu_of(rq));
168                 steal -= rq->prev_steal_time_rq;
169 
170                 if (unlikely(steal > delta))
171                         steal = delta;
172 
173                 rq->prev_steal_time_rq += steal;
174                 delta -= steal;
175         }
176 #endif
177 
178         rq->clock_task += delta;
179 
180 #ifdef HAVE_SCHED_AVG_IRQ
181         if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
182                 update_irq_load_avg(rq, irq_delta + steal);
183 #endif
184 }
185 
186 void update_rq_clock(struct rq *rq)
187 {
188         s64 delta;
189 
190         lockdep_assert_held(&rq->lock);
191 
192         if (rq->clock_update_flags & RQCF_ACT_SKIP)
193                 return;
194 
195 #ifdef CONFIG_SCHED_DEBUG
196         if (sched_feat(WARN_DOUBLE_CLOCK))
197                 SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED);
198         rq->clock_update_flags |= RQCF_UPDATED;
199 #endif
200 
201         delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
202         if (delta < 0)
203                 return;
204         rq->clock += delta;
205         update_rq_clock_task(rq, delta);
206 }
207 
208 
209 #ifdef CONFIG_SCHED_HRTICK
210 /*
211  * Use HR-timers to deliver accurate preemption points.
212  */
213 
214 static void hrtick_clear(struct rq *rq)
215 {
216         if (hrtimer_active(&rq->hrtick_timer))
217                 hrtimer_cancel(&rq->hrtick_timer);
218 }
219 
220 /*
221  * High-resolution timer tick.
222  * Runs from hardirq context with interrupts disabled.
223  */
224 static enum hrtimer_restart hrtick(struct hrtimer *timer)
225 {
226         struct rq *rq = container_of(timer, struct rq, hrtick_timer);
227         struct rq_flags rf;
228 
229         WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
230 
231         rq_lock(rq, &rf);
232         update_rq_clock(rq);
233         rq->curr->sched_class->task_tick(rq, rq->curr, 1);
234         rq_unlock(rq, &rf);
235 
236         return HRTIMER_NORESTART;
237 }
238 
239 #ifdef CONFIG_SMP
240 
241 static void __hrtick_restart(struct rq *rq)
242 {
243         struct hrtimer *timer = &rq->hrtick_timer;
244 
245         hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
246 }
247 
248 /*
249  * called from hardirq (IPI) context
250  */
251 static void __hrtick_start(void *arg)
252 {
253         struct rq *rq = arg;
254         struct rq_flags rf;
255 
256         rq_lock(rq, &rf);
257         __hrtick_restart(rq);
258         rq->hrtick_csd_pending = 0;
259         rq_unlock(rq, &rf);
260 }
261 
262 /*
263  * Called to set the hrtick timer state.
264  *
265  * called with rq->lock held and irqs disabled
266  */
267 void hrtick_start(struct rq *rq, u64 delay)
268 {
269         struct hrtimer *timer = &rq->hrtick_timer;
270         ktime_t time;
271         s64 delta;
272 
273         /*
274          * Don't schedule slices shorter than 10000ns, that just
275          * doesn't make sense and can cause timer DoS.
276          */
277         delta = max_t(s64, delay, 10000LL);
278         time = ktime_add_ns(timer->base->get_time(), delta);
279 
280         hrtimer_set_expires(timer, time);
281 
282         if (rq == this_rq()) {
283                 __hrtick_restart(rq);
284         } else if (!rq->hrtick_csd_pending) {
285                 smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
286                 rq->hrtick_csd_pending = 1;
287         }
288 }
289 
290 #else
291 /*
292  * Called to set the hrtick timer state.
293  *
294  * called with rq->lock held and irqs disabled
295  */
296 void hrtick_start(struct rq *rq, u64 delay)
297 {
298         /*
299          * Don't schedule slices shorter than 10000ns, that just
300          * doesn't make sense. Rely on vruntime for fairness.
301          */
302         delay = max_t(u64, delay, 10000LL);
303         hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
304                       HRTIMER_MODE_REL_PINNED);
305 }
306 #endif /* CONFIG_SMP */
307 
308 static void hrtick_rq_init(struct rq *rq)
309 {
310 #ifdef CONFIG_SMP
311         rq->hrtick_csd_pending = 0;
312 
313         rq->hrtick_csd.flags = 0;
314         rq->hrtick_csd.func = __hrtick_start;
315         rq->hrtick_csd.info = rq;
316 #endif
317 
318         hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
319         rq->hrtick_timer.function = hrtick;
320 }
321 #else   /* CONFIG_SCHED_HRTICK */
322 static inline void hrtick_clear(struct rq *rq)
323 {
324 }
325 
326 static inline void hrtick_rq_init(struct rq *rq)
327 {
328 }
329 #endif  /* CONFIG_SCHED_HRTICK */
330 
331 /*
332  * cmpxchg based fetch_or, macro so it works for different integer types
333  */
334 #define fetch_or(ptr, mask)                                             \
335         ({                                                              \
336                 typeof(ptr) _ptr = (ptr);                               \
337                 typeof(mask) _mask = (mask);                            \
338                 typeof(*_ptr) _old, _val = *_ptr;                       \
339                                                                         \
340                 for (;;) {                                              \
341                         _old = cmpxchg(_ptr, _val, _val | _mask);       \
342                         if (_old == _val)                               \
343                                 break;                                  \
344                         _val = _old;                                    \
345                 }                                                       \
346         _old;                                                           \
347 })
348 
349 #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
350 /*
351  * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG,
352  * this avoids any races wrt polling state changes and thereby avoids
353  * spurious IPIs.
354  */
355 static bool set_nr_and_not_polling(struct task_struct *p)
356 {
357         struct thread_info *ti = task_thread_info(p);
358         return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG);
359 }
360 
361 /*
362  * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set.
363  *
364  * If this returns true, then the idle task promises to call
365  * sched_ttwu_pending() and reschedule soon.
366  */
367 static bool set_nr_if_polling(struct task_struct *p)
368 {
369         struct thread_info *ti = task_thread_info(p);
370         typeof(ti->flags) old, val = READ_ONCE(ti->flags);
371 
372         for (;;) {
373                 if (!(val & _TIF_POLLING_NRFLAG))
374                         return false;
375                 if (val & _TIF_NEED_RESCHED)
376                         return true;
377                 old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED);
378                 if (old == val)
379                         break;
380                 val = old;
381         }
382         return true;
383 }
384 
385 #else
386 static bool set_nr_and_not_polling(struct task_struct *p)
387 {
388         set_tsk_need_resched(p);
389         return true;
390 }
391 
392 #ifdef CONFIG_SMP
393 static bool set_nr_if_polling(struct task_struct *p)
394 {
395         return false;
396 }
397 #endif
398 #endif
399 
400 void wake_q_add(struct wake_q_head *head, struct task_struct *task)
401 {
402         struct wake_q_node *node = &task->wake_q;
403 
404         /*
405          * Atomically grab the task, if ->wake_q is !nil already it means
406          * its already queued (either by us or someone else) and will get the
407          * wakeup due to that.
408          *
409          * This cmpxchg() executes a full barrier, which pairs with the full
410          * barrier executed by the wakeup in wake_up_q().
411          */
412         if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL))
413                 return;
414 
415         get_task_struct(task);
416 
417         /*
418          * The head is context local, there can be no concurrency.
419          */
420         *head->lastp = node;
421         head->lastp = &node->next;
422 }
423 
424 void wake_up_q(struct wake_q_head *head)
425 {
426         struct wake_q_node *node = head->first;
427 
428         while (node != WAKE_Q_TAIL) {
429                 struct task_struct *task;
430 
431                 task = container_of(node, struct task_struct, wake_q);
432                 BUG_ON(!task);
433                 /* Task can safely be re-inserted now: */
434                 node = node->next;
435                 task->wake_q.next = NULL;
436 
437                 /*
438                  * wake_up_process() executes a full barrier, which pairs with
439                  * the queueing in wake_q_add() so as not to miss wakeups.
440                  */
441                 wake_up_process(task);
442                 put_task_struct(task);
443         }
444 }
445 
446 /*
447  * resched_curr - mark rq's current task 'to be rescheduled now'.
448  *
449  * On UP this means the setting of the need_resched flag, on SMP it
450  * might also involve a cross-CPU call to trigger the scheduler on
451  * the target CPU.
452  */
453 void resched_curr(struct rq *rq)
454 {
455         struct task_struct *curr = rq->curr;
456         int cpu;
457 
458         lockdep_assert_held(&rq->lock);
459 
460         if (test_tsk_need_resched(curr))
461                 return;
462 
463         cpu = cpu_of(rq);
464 
465         if (cpu == smp_processor_id()) {
466                 set_tsk_need_resched(curr);
467                 set_preempt_need_resched();
468                 return;
469         }
470 
471         if (set_nr_and_not_polling(curr))
472                 smp_send_reschedule(cpu);
473         else
474                 trace_sched_wake_idle_without_ipi(cpu);
475 }
476 
477 void resched_cpu(int cpu)
478 {
479         struct rq *rq = cpu_rq(cpu);
480         unsigned long flags;
481 
482         raw_spin_lock_irqsave(&rq->lock, flags);
483         if (cpu_online(cpu) || cpu == smp_processor_id())
484                 resched_curr(rq);
485         raw_spin_unlock_irqrestore(&rq->lock, flags);
486 }
487 
488 #ifdef CONFIG_SMP
489 #ifdef CONFIG_NO_HZ_COMMON
490 /*
491  * In the semi idle case, use the nearest busy CPU for migrating timers
492  * from an idle CPU.  This is good for power-savings.
493  *
494  * We don't do similar optimization for completely idle system, as
495  * selecting an idle CPU will add more delays to the timers than intended
496  * (as that CPU's timer base may not be uptodate wrt jiffies etc).
497  */
498 int get_nohz_timer_target(void)
499 {
500         int i, cpu = smp_processor_id();
501         struct sched_domain *sd;
502 
503         if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
504                 return cpu;
505 
506         rcu_read_lock();
507         for_each_domain(cpu, sd) {
508                 for_each_cpu(i, sched_domain_span(sd)) {
509                         if (cpu == i)
510                                 continue;
511 
512                         if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
513                                 cpu = i;
514                                 goto unlock;
515                         }
516                 }
517         }
518 
519         if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
520                 cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
521 unlock:
522         rcu_read_unlock();
523         return cpu;
524 }
525 
526 /*
527  * When add_timer_on() enqueues a timer into the timer wheel of an
528  * idle CPU then this timer might expire before the next timer event
529  * which is scheduled to wake up that CPU. In case of a completely
530  * idle system the next event might even be infinite time into the
531  * future. wake_up_idle_cpu() ensures that the CPU is woken up and
532  * leaves the inner idle loop so the newly added timer is taken into
533  * account when the CPU goes back to idle and evaluates the timer
534  * wheel for the next timer event.
535  */
536 static void wake_up_idle_cpu(int cpu)
537 {
538         struct rq *rq = cpu_rq(cpu);
539 
540         if (cpu == smp_processor_id())
541                 return;
542 
543         if (set_nr_and_not_polling(rq->idle))
544                 smp_send_reschedule(cpu);
545         else
546                 trace_sched_wake_idle_without_ipi(cpu);
547 }
548 
549 static bool wake_up_full_nohz_cpu(int cpu)
550 {
551         /*
552          * We just need the target to call irq_exit() and re-evaluate
553          * the next tick. The nohz full kick at least implies that.
554          * If needed we can still optimize that later with an
555          * empty IRQ.
556          */
557         if (cpu_is_offline(cpu))
558                 return true;  /* Don't try to wake offline CPUs. */
559         if (tick_nohz_full_cpu(cpu)) {
560                 if (cpu != smp_processor_id() ||
561                     tick_nohz_tick_stopped())
562                         tick_nohz_full_kick_cpu(cpu);
563                 return true;
564         }
565 
566         return false;
567 }
568 
569 /*
570  * Wake up the specified CPU.  If the CPU is going offline, it is the
571  * caller's responsibility to deal with the lost wakeup, for example,
572  * by hooking into the CPU_DEAD notifier like timers and hrtimers do.
573  */
574 void wake_up_nohz_cpu(int cpu)
575 {
576         if (!wake_up_full_nohz_cpu(cpu))
577                 wake_up_idle_cpu(cpu);
578 }
579 
580 static inline bool got_nohz_idle_kick(void)
581 {
582         int cpu = smp_processor_id();
583 
584         if (!(atomic_read(nohz_flags(cpu)) & NOHZ_KICK_MASK))
585                 return false;
586 
587         if (idle_cpu(cpu) && !need_resched())
588                 return true;
589 
590         /*
591          * We can't run Idle Load Balance on this CPU for this time so we
592          * cancel it and clear NOHZ_BALANCE_KICK
593          */
594         atomic_andnot(NOHZ_KICK_MASK, nohz_flags(cpu));
595         return false;
596 }
597 
598 #else /* CONFIG_NO_HZ_COMMON */
599 
600 static inline bool got_nohz_idle_kick(void)
601 {
602         return false;
603 }
604 
605 #endif /* CONFIG_NO_HZ_COMMON */
606 
607 #ifdef CONFIG_NO_HZ_FULL
608 bool sched_can_stop_tick(struct rq *rq)
609 {
610         int fifo_nr_running;
611 
612         /* Deadline tasks, even if single, need the tick */
613         if (rq->dl.dl_nr_running)
614                 return false;
615 
616         /*
617          * If there are more than one RR tasks, we need the tick to effect the
618          * actual RR behaviour.
619          */
620         if (rq->rt.rr_nr_running) {
621                 if (rq->rt.rr_nr_running == 1)
622                         return true;
623                 else
624                         return false;
625         }
626 
627         /*
628          * If there's no RR tasks, but FIFO tasks, we can skip the tick, no
629          * forced preemption between FIFO tasks.
630          */
631         fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running;
632         if (fifo_nr_running)
633                 return true;
634 
635         /*
636          * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left;
637          * if there's more than one we need the tick for involuntary
638          * preemption.
639          */
640         if (rq->nr_running > 1)
641                 return false;
642 
643         return true;
644 }
645 #endif /* CONFIG_NO_HZ_FULL */
646 #endif /* CONFIG_SMP */
647 
648 #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
649                         (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
650 /*
651  * Iterate task_group tree rooted at *from, calling @down when first entering a
652  * node and @up when leaving it for the final time.
653  *
654  * Caller must hold rcu_lock or sufficient equivalent.
655  */
656 int walk_tg_tree_from(struct task_group *from,
657                              tg_visitor down, tg_visitor up, void *data)
658 {
659         struct task_group *parent, *child;
660         int ret;
661 
662         parent = from;
663 
664 down:
665         ret = (*down)(parent, data);
666         if (ret)
667                 goto out;
668         list_for_each_entry_rcu(child, &parent->children, siblings) {
669                 parent = child;
670                 goto down;
671 
672 up:
673                 continue;
674         }
675         ret = (*up)(parent, data);
676         if (ret || parent == from)
677                 goto out;
678 
679         child = parent;
680         parent = parent->parent;
681         if (parent)
682                 goto up;
683 out:
684         return ret;
685 }
686 
687 int tg_nop(struct task_group *tg, void *data)
688 {
689         return 0;
690 }
691 #endif
692 
693 static void set_load_weight(struct task_struct *p, bool update_load)
694 {
695         int prio = p->static_prio - MAX_RT_PRIO;
696         struct load_weight *load = &p->se.load;
697 
698         /*
699          * SCHED_IDLE tasks get minimal weight:
700          */
701         if (idle_policy(p->policy)) {
702                 load->weight = scale_load(WEIGHT_IDLEPRIO);
703                 load->inv_weight = WMULT_IDLEPRIO;
704                 return;
705         }
706 
707         /*
708          * SCHED_OTHER tasks have to update their load when changing their
709          * weight
710          */
711         if (update_load && p->sched_class == &fair_sched_class) {
712                 reweight_task(p, prio);
713         } else {
714                 load->weight = scale_load(sched_prio_to_weight[prio]);
715                 load->inv_weight = sched_prio_to_wmult[prio];
716         }
717 }
718 
719 static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
720 {
721         if (!(flags & ENQUEUE_NOCLOCK))
722                 update_rq_clock(rq);
723 
724         if (!(flags & ENQUEUE_RESTORE))
725                 sched_info_queued(rq, p);
726 
727         p->sched_class->enqueue_task(rq, p, flags);
728 }
729 
730 static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
731 {
732         if (!(flags & DEQUEUE_NOCLOCK))
733                 update_rq_clock(rq);
734 
735         if (!(flags & DEQUEUE_SAVE))
736                 sched_info_dequeued(rq, p);
737 
738         p->sched_class->dequeue_task(rq, p, flags);
739 }
740 
741 void activate_task(struct rq *rq, struct task_struct *p, int flags)
742 {
743         if (task_contributes_to_load(p))
744                 rq->nr_uninterruptible--;
745 
746         enqueue_task(rq, p, flags);
747 }
748 
749 void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
750 {
751         if (task_contributes_to_load(p))
752                 rq->nr_uninterruptible++;
753 
754         dequeue_task(rq, p, flags);
755 }
756 
757 /*
758  * __normal_prio - return the priority that is based on the static prio
759  */
760 static inline int __normal_prio(struct task_struct *p)
761 {
762         return p->static_prio;
763 }
764 
765 /*
766  * Calculate the expected normal priority: i.e. priority
767  * without taking RT-inheritance into account. Might be
768  * boosted by interactivity modifiers. Changes upon fork,
769  * setprio syscalls, and whenever the interactivity
770  * estimator recalculates.
771  */
772 static inline int normal_prio(struct task_struct *p)
773 {
774         int prio;
775 
776         if (task_has_dl_policy(p))
777                 prio = MAX_DL_PRIO-1;
778         else if (task_has_rt_policy(p))
779                 prio = MAX_RT_PRIO-1 - p->rt_priority;
780         else
781                 prio = __normal_prio(p);
782         return prio;
783 }
784 
785 /*
786  * Calculate the current priority, i.e. the priority
787  * taken into account by the scheduler. This value might
788  * be boosted by RT tasks, or might be boosted by
789  * interactivity modifiers. Will be RT if the task got
790  * RT-boosted. If not then it returns p->normal_prio.
791  */
792 static int effective_prio(struct task_struct *p)
793 {
794         p->normal_prio = normal_prio(p);
795         /*
796          * If we are RT tasks or we were boosted to RT priority,
797          * keep the priority unchanged. Otherwise, update priority
798          * to the normal priority:
799          */
800         if (!rt_prio(p->prio))
801                 return p->normal_prio;
802         return p->prio;
803 }
804 
805 /**
806  * task_curr - is this task currently executing on a CPU?
807  * @p: the task in question.
808  *
809  * Return: 1 if the task is currently executing. 0 otherwise.
810  */
811 inline int task_curr(const struct task_struct *p)
812 {
813         return cpu_curr(task_cpu(p)) == p;
814 }
815 
816 /*
817  * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock,
818  * use the balance_callback list if you want balancing.
819  *
820  * this means any call to check_class_changed() must be followed by a call to
821  * balance_callback().
822  */
823 static inline void check_class_changed(struct rq *rq, struct task_struct *p,
824                                        const struct sched_class *prev_class,
825                                        int oldprio)
826 {
827         if (prev_class != p->sched_class) {
828                 if (prev_class->switched_from)
829                         prev_class->switched_from(rq, p);
830 
831                 p->sched_class->switched_to(rq, p);
832         } else if (oldprio != p->prio || dl_task(p))
833                 p->sched_class->prio_changed(rq, p, oldprio);
834 }
835 
836 void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
837 {
838         const struct sched_class *class;
839 
840         if (p->sched_class == rq->curr->sched_class) {
841                 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
842         } else {
843                 for_each_class(class) {
844                         if (class == rq->curr->sched_class)
845                                 break;
846                         if (class == p->sched_class) {
847                                 resched_curr(rq);
848                                 break;
849                         }
850                 }
851         }
852 
853         /*
854          * A queue event has occurred, and we're going to schedule.  In
855          * this case, we can save a useless back to back clock update.
856          */
857         if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
858                 rq_clock_skip_update(rq);
859 }
860 
861 #ifdef CONFIG_SMP
862 
863 static inline bool is_per_cpu_kthread(struct task_struct *p)
864 {
865         if (!(p->flags & PF_KTHREAD))
866                 return false;
867 
868         if (p->nr_cpus_allowed != 1)
869                 return false;
870 
871         return true;
872 }
873 
874 /*
875  * Per-CPU kthreads are allowed to run on !actie && online CPUs, see
876  * __set_cpus_allowed_ptr() and select_fallback_rq().
877  */
878 static inline bool is_cpu_allowed(struct task_struct *p, int cpu)
879 {
880         if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
881                 return false;
882 
883         if (is_per_cpu_kthread(p))
884                 return cpu_online(cpu);
885 
886         return cpu_active(cpu);
887 }
888 
889 /*
890  * This is how migration works:
891  *
892  * 1) we invoke migration_cpu_stop() on the target CPU using
893  *    stop_one_cpu().
894  * 2) stopper starts to run (implicitly forcing the migrated thread
895  *    off the CPU)
896  * 3) it checks whether the migrated task is still in the wrong runqueue.
897  * 4) if it's in the wrong runqueue then the migration thread removes
898  *    it and puts it into the right queue.
899  * 5) stopper completes and stop_one_cpu() returns and the migration
900  *    is done.
901  */
902 
903 /*
904  * move_queued_task - move a queued task to new rq.
905  *
906  * Returns (locked) new rq. Old rq's lock is released.
907  */
908 static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf,
909                                    struct task_struct *p, int new_cpu)
910 {
911         lockdep_assert_held(&rq->lock);
912 
913         p->on_rq = TASK_ON_RQ_MIGRATING;
914         dequeue_task(rq, p, DEQUEUE_NOCLOCK);
915         set_task_cpu(p, new_cpu);
916         rq_unlock(rq, rf);
917 
918         rq = cpu_rq(new_cpu);
919 
920         rq_lock(rq, rf);
921         BUG_ON(task_cpu(p) != new_cpu);
922         enqueue_task(rq, p, 0);
923         p->on_rq = TASK_ON_RQ_QUEUED;
924         check_preempt_curr(rq, p, 0);
925 
926         return rq;
927 }
928 
929 struct migration_arg {
930         struct task_struct *task;
931         int dest_cpu;
932 };
933 
934 /*
935  * Move (not current) task off this CPU, onto the destination CPU. We're doing
936  * this because either it can't run here any more (set_cpus_allowed()
937  * away from this CPU, or CPU going down), or because we're
938  * attempting to rebalance this task on exec (sched_exec).
939  *
940  * So we race with normal scheduler movements, but that's OK, as long
941  * as the task is no longer on this CPU.
942  */
943 static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
944                                  struct task_struct *p, int dest_cpu)
945 {
946         /* Affinity changed (again). */
947         if (!is_cpu_allowed(p, dest_cpu))
948                 return rq;
949 
950         update_rq_clock(rq);
951         rq = move_queued_task(rq, rf, p, dest_cpu);
952 
953         return rq;
954 }
955 
956 /*
957  * migration_cpu_stop - this will be executed by a highprio stopper thread
958  * and performs thread migration by bumping thread off CPU then
959  * 'pushing' onto another runqueue.
960  */
961 static int migration_cpu_stop(void *data)
962 {
963         struct migration_arg *arg = data;
964         struct task_struct *p = arg->task;
965         struct rq *rq = this_rq();
966         struct rq_flags rf;
967 
968         /*
969          * The original target CPU might have gone down and we might
970          * be on another CPU but it doesn't matter.
971          */
972         local_irq_disable();
973         /*
974          * We need to explicitly wake pending tasks before running
975          * __migrate_task() such that we will not miss enforcing cpus_allowed
976          * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test.
977          */
978         sched_ttwu_pending();
979 
980         raw_spin_lock(&p->pi_lock);
981         rq_lock(rq, &rf);
982         /*
983          * If task_rq(p) != rq, it cannot be migrated here, because we're
984          * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because
985          * we're holding p->pi_lock.
986          */
987         if (task_rq(p) == rq) {
988                 if (task_on_rq_queued(p))
989                         rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
990                 else
991                         p->wake_cpu = arg->dest_cpu;
992         }
993         rq_unlock(rq, &rf);
994         raw_spin_unlock(&p->pi_lock);
995 
996         local_irq_enable();
997         return 0;
998 }
999 
1000 /*
1001  * sched_class::set_cpus_allowed must do the below, but is not required to
1002  * actually call this function.
1003  */
1004 void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask)
1005 {
1006         cpumask_copy(&p->cpus_allowed, new_mask);
1007         p->nr_cpus_allowed = cpumask_weight(new_mask);
1008 }
1009 
1010 void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1011 {
1012         struct rq *rq = task_rq(p);
1013         bool queued, running;
1014 
1015         lockdep_assert_held(&p->pi_lock);
1016 
1017         queued = task_on_rq_queued(p);
1018         running = task_current(rq, p);
1019 
1020         if (queued) {
1021                 /*
1022                  * Because __kthread_bind() calls this on blocked tasks without
1023                  * holding rq->lock.
1024                  */
1025                 lockdep_assert_held(&rq->lock);
1026                 dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
1027         }
1028         if (running)
1029                 put_prev_task(rq, p);
1030 
1031         p->sched_class->set_cpus_allowed(p, new_mask);
1032 
1033         if (queued)
1034                 enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
1035         if (running)
1036                 set_curr_task(rq, p);
1037 }
1038 
1039 /*
1040  * Change a given task's CPU affinity. Migrate the thread to a
1041  * proper CPU and schedule it away if the CPU it's executing on
1042  * is removed from the allowed bitmask.
1043  *
1044  * NOTE: the caller must have a valid reference to the task, the
1045  * task must not exit() & deallocate itself prematurely. The
1046  * call is not atomic; no spinlocks may be held.
1047  */
1048 static int __set_cpus_allowed_ptr(struct task_struct *p,
1049                                   const struct cpumask *new_mask, bool check)
1050 {
1051         const struct cpumask *cpu_valid_mask = cpu_active_mask;
1052         unsigned int dest_cpu;
1053         struct rq_flags rf;
1054         struct rq *rq;
1055         int ret = 0;
1056 
1057         rq = task_rq_lock(p, &rf);
1058         update_rq_clock(rq);
1059 
1060         if (p->flags & PF_KTHREAD) {
1061                 /*
1062                  * Kernel threads are allowed on online && !active CPUs
1063                  */
1064                 cpu_valid_mask = cpu_online_mask;
1065         }
1066 
1067         /*
1068          * Must re-check here, to close a race against __kthread_bind(),
1069          * sched_setaffinity() is not guaranteed to observe the flag.
1070          */
1071         if (check && (p->flags & PF_NO_SETAFFINITY)) {
1072                 ret = -EINVAL;
1073                 goto out;
1074         }
1075 
1076         if (cpumask_equal(&p->cpus_allowed, new_mask))
1077                 goto out;
1078 
1079         if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
1080                 ret = -EINVAL;
1081                 goto out;
1082         }
1083 
1084         do_set_cpus_allowed(p, new_mask);
1085 
1086         if (p->flags & PF_KTHREAD) {
1087                 /*
1088                  * For kernel threads that do indeed end up on online &&
1089                  * !active we want to ensure they are strict per-CPU threads.
1090                  */
1091                 WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) &&
1092                         !cpumask_intersects(new_mask, cpu_active_mask) &&
1093                         p->nr_cpus_allowed != 1);
1094         }
1095 
1096         /* Can the task run on the task's current CPU? If so, we're done */
1097         if (cpumask_test_cpu(task_cpu(p), new_mask))
1098                 goto out;
1099 
1100         dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
1101         if (task_running(rq, p) || p->state == TASK_WAKING) {
1102                 struct migration_arg arg = { p, dest_cpu };
1103                 /* Need help from migration thread: drop lock and wait. */
1104                 task_rq_unlock(rq, p, &rf);
1105                 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1106                 tlb_migrate_finish(p->mm);
1107                 return 0;
1108         } else if (task_on_rq_queued(p)) {
1109                 /*
1110                  * OK, since we're going to drop the lock immediately
1111                  * afterwards anyway.
1112                  */
1113                 rq = move_queued_task(rq, &rf, p, dest_cpu);
1114         }
1115 out:
1116         task_rq_unlock(rq, p, &rf);
1117 
1118         return ret;
1119 }
1120 
1121 int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1122 {
1123         return __set_cpus_allowed_ptr(p, new_mask, false);
1124 }
1125 EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
1126 
1127 void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
1128 {
1129 #ifdef CONFIG_SCHED_DEBUG
1130         /*
1131          * We should never call set_task_cpu() on a blocked task,
1132          * ttwu() will sort out the placement.
1133          */
1134         WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
1135                         !p->on_rq);
1136 
1137         /*
1138          * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING,
1139          * because schedstat_wait_{start,end} rebase migrating task's wait_start
1140          * time relying on p->on_rq.
1141          */
1142         WARN_ON_ONCE(p->state == TASK_RUNNING &&
1143                      p->sched_class == &fair_sched_class &&
1144                      (p->on_rq && !task_on_rq_migrating(p)));
1145 
1146 #ifdef CONFIG_LOCKDEP
1147         /*
1148          * The caller should hold either p->pi_lock or rq->lock, when changing
1149          * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks.
1150          *
1151          * sched_move_task() holds both and thus holding either pins the cgroup,
1152          * see task_group().
1153          *
1154          * Furthermore, all task_rq users should acquire both locks, see
1155          * task_rq_lock().
1156          */
1157         WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
1158                                       lockdep_is_held(&task_rq(p)->lock)));
1159 #endif
1160         /*
1161          * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
1162          */
1163         WARN_ON_ONCE(!cpu_online(new_cpu));
1164 #endif
1165 
1166         trace_sched_migrate_task(p, new_cpu);
1167 
1168         if (task_cpu(p) != new_cpu) {
1169                 if (p->sched_class->migrate_task_rq)
1170                         p->sched_class->migrate_task_rq(p);
1171                 p->se.nr_migrations++;
1172                 rseq_migrate(p);
1173                 perf_event_task_migrate(p);
1174         }
1175 
1176         __set_task_cpu(p, new_cpu);
1177 }
1178 
1179 #ifdef CONFIG_NUMA_BALANCING
1180 static void __migrate_swap_task(struct task_struct *p, int cpu)
1181 {
1182         if (task_on_rq_queued(p)) {
1183                 struct rq *src_rq, *dst_rq;
1184                 struct rq_flags srf, drf;
1185 
1186                 src_rq = task_rq(p);
1187                 dst_rq = cpu_rq(cpu);
1188 
1189                 rq_pin_lock(src_rq, &srf);
1190                 rq_pin_lock(dst_rq, &drf);
1191 
1192                 p->on_rq = TASK_ON_RQ_MIGRATING;
1193                 deactivate_task(src_rq, p, 0);
1194                 set_task_cpu(p, cpu);
1195                 activate_task(dst_rq, p, 0);
1196                 p->on_rq = TASK_ON_RQ_QUEUED;
1197                 check_preempt_curr(dst_rq, p, 0);
1198 
1199                 rq_unpin_lock(dst_rq, &drf);
1200                 rq_unpin_lock(src_rq, &srf);
1201 
1202         } else {
1203                 /*
1204                  * Task isn't running anymore; make it appear like we migrated
1205                  * it before it went to sleep. This means on wakeup we make the
1206                  * previous CPU our target instead of where it really is.
1207                  */
1208                 p->wake_cpu = cpu;
1209         }
1210 }
1211 
1212 struct migration_swap_arg {
1213         struct task_struct *src_task, *dst_task;
1214         int src_cpu, dst_cpu;
1215 };
1216 
1217 static int migrate_swap_stop(void *data)
1218 {
1219         struct migration_swap_arg *arg = data;
1220         struct rq *src_rq, *dst_rq;
1221         int ret = -EAGAIN;
1222 
1223         if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
1224                 return -EAGAIN;
1225 
1226         src_rq = cpu_rq(arg->src_cpu);
1227         dst_rq = cpu_rq(arg->dst_cpu);
1228 
1229         double_raw_lock(&arg->src_task->pi_lock,
1230                         &arg->dst_task->pi_lock);
1231         double_rq_lock(src_rq, dst_rq);
1232 
1233         if (task_cpu(arg->dst_task) != arg->dst_cpu)
1234                 goto unlock;
1235 
1236         if (task_cpu(arg->src_task) != arg->src_cpu)
1237                 goto unlock;
1238 
1239         if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed))
1240                 goto unlock;
1241 
1242         if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed))
1243                 goto unlock;
1244 
1245         __migrate_swap_task(arg->src_task, arg->dst_cpu);
1246         __migrate_swap_task(arg->dst_task, arg->src_cpu);
1247 
1248         ret = 0;
1249 
1250 unlock:
1251         double_rq_unlock(src_rq, dst_rq);
1252         raw_spin_unlock(&arg->dst_task->pi_lock);
1253         raw_spin_unlock(&arg->src_task->pi_lock);
1254 
1255         return ret;
1256 }
1257 
1258 /*
1259  * Cross migrate two tasks
1260  */
1261 int migrate_swap(struct task_struct *cur, struct task_struct *p,
1262                 int target_cpu, int curr_cpu)
1263 {
1264         struct migration_swap_arg arg;
1265         int ret = -EINVAL;
1266 
1267         arg = (struct migration_swap_arg){
1268                 .src_task = cur,
1269                 .src_cpu = curr_cpu,
1270                 .dst_task = p,
1271                 .dst_cpu = target_cpu,
1272         };
1273 
1274         if (arg.src_cpu == arg.dst_cpu)
1275                 goto out;
1276 
1277         /*
1278          * These three tests are all lockless; this is OK since all of them
1279          * will be re-checked with proper locks held further down the line.
1280          */
1281         if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
1282                 goto out;
1283 
1284         if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed))
1285                 goto out;
1286 
1287         if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed))
1288                 goto out;
1289 
1290         trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1291         ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);
1292 
1293 out:
1294         return ret;
1295 }
1296 #endif /* CONFIG_NUMA_BALANCING */
1297 
1298 /*
1299  * wait_task_inactive - wait for a thread to unschedule.
1300  *
1301  * If @match_state is nonzero, it's the @p->state value just checked and
1302  * not expected to change.  If it changes, i.e. @p might have woken up,
1303  * then return zero.  When we succeed in waiting for @p to be off its CPU,
1304  * we return a positive number (its total switch count).  If a second call
1305  * a short while later returns the same number, the caller can be sure that
1306  * @p has remained unscheduled the whole time.
1307  *
1308  * The caller must ensure that the task *will* unschedule sometime soon,
1309  * else this function might spin for a *long* time. This function can't
1310  * be called with interrupts off, or it may introduce deadlock with
1311  * smp_call_function() if an IPI is sent by the same process we are
1312  * waiting to become inactive.
1313  */
1314 unsigned long wait_task_inactive(struct task_struct *p, long match_state)
1315 {
1316         int running, queued;
1317         struct rq_flags rf;
1318         unsigned long ncsw;
1319         struct rq *rq;
1320 
1321         for (;;) {
1322                 /*
1323                  * We do the initial early heuristics without holding
1324                  * any task-queue locks at all. We'll only try to get
1325                  * the runqueue lock when things look like they will
1326                  * work out!
1327                  */
1328                 rq = task_rq(p);
1329 
1330                 /*
1331                  * If the task is actively running on another CPU
1332                  * still, just relax and busy-wait without holding
1333                  * any locks.
1334                  *
1335                  * NOTE! Since we don't hold any locks, it's not
1336                  * even sure that "rq" stays as the right runqueue!
1337                  * But we don't care, since "task_running()" will
1338                  * return false if the runqueue has changed and p
1339                  * is actually now running somewhere else!
1340                  */
1341                 while (task_running(rq, p)) {
1342                         if (match_state && unlikely(p->state != match_state))
1343                                 return 0;
1344                         cpu_relax();
1345                 }
1346 
1347                 /*
1348                  * Ok, time to look more closely! We need the rq
1349                  * lock now, to be *sure*. If we're wrong, we'll
1350                  * just go back and repeat.
1351                  */
1352                 rq = task_rq_lock(p, &rf);
1353                 trace_sched_wait_task(p);
1354                 running = task_running(rq, p);
1355                 queued = task_on_rq_queued(p);
1356                 ncsw = 0;
1357                 if (!match_state || p->state == match_state)
1358                         ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1359                 task_rq_unlock(rq, p, &rf);
1360 
1361                 /*
1362                  * If it changed from the expected state, bail out now.
1363                  */
1364                 if (unlikely(!ncsw))
1365                         break;
1366 
1367                 /*
1368                  * Was it really running after all now that we
1369                  * checked with the proper locks actually held?
1370                  *
1371                  * Oops. Go back and try again..
1372                  */
1373                 if (unlikely(running)) {
1374                         cpu_relax();
1375                         continue;
1376                 }
1377 
1378                 /*
1379                  * It's not enough that it's not actively running,
1380                  * it must be off the runqueue _entirely_, and not
1381                  * preempted!
1382                  *
1383                  * So if it was still runnable (but just not actively
1384                  * running right now), it's preempted, and we should
1385                  * yield - it could be a while.
1386                  */
1387                 if (unlikely(queued)) {
1388                         ktime_t to = NSEC_PER_SEC / HZ;
1389 
1390                         set_current_state(TASK_UNINTERRUPTIBLE);
1391                         schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1392                         continue;
1393                 }
1394 
1395                 /*
1396                  * Ahh, all good. It wasn't running, and it wasn't
1397                  * runnable, which means that it will never become
1398                  * running in the future either. We're all done!
1399                  */
1400                 break;
1401         }
1402 
1403         return ncsw;
1404 }
1405 
1406 /***
1407  * kick_process - kick a running thread to enter/exit the kernel
1408  * @p: the to-be-kicked thread
1409  *
1410  * Cause a process which is running on another CPU to enter
1411  * kernel-mode, without any delay. (to get signals handled.)
1412  *
1413  * NOTE: this function doesn't have to take the runqueue lock,
1414  * because all it wants to ensure is that the remote task enters
1415  * the kernel. If the IPI races and the task has been migrated
1416  * to another CPU then no harm is done and the purpose has been
1417  * achieved as well.
1418  */
1419 void kick_process(struct task_struct *p)
1420 {
1421         int cpu;
1422 
1423         preempt_disable();
1424         cpu = task_cpu(p);
1425         if ((cpu != smp_processor_id()) && task_curr(p))
1426                 smp_send_reschedule(cpu);
1427         preempt_enable();
1428 }
1429 EXPORT_SYMBOL_GPL(kick_process);
1430 
1431 /*
1432  * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1433  *
1434  * A few notes on cpu_active vs cpu_online:
1435  *
1436  *  - cpu_active must be a subset of cpu_online
1437  *
1438  *  - on CPU-up we allow per-CPU kthreads on the online && !active CPU,
1439  *    see __set_cpus_allowed_ptr(). At this point the newly online
1440  *    CPU isn't yet part of the sched domains, and balancing will not
1441  *    see it.
1442  *
1443  *  - on CPU-down we clear cpu_active() to mask the sched domains and
1444  *    avoid the load balancer to place new tasks on the to be removed
1445  *    CPU. Existing tasks will remain running there and will be taken
1446  *    off.
1447  *
1448  * This means that fallback selection must not select !active CPUs.
1449  * And can assume that any active CPU must be online. Conversely
1450  * select_task_rq() below may allow selection of !active CPUs in order
1451  * to satisfy the above rules.
1452  */
1453 static int select_fallback_rq(int cpu, struct task_struct *p)
1454 {
1455         int nid = cpu_to_node(cpu);
1456         const struct cpumask *nodemask = NULL;
1457         enum { cpuset, possible, fail } state = cpuset;
1458         int dest_cpu;
1459 
1460         /*
1461          * If the node that the CPU is on has been offlined, cpu_to_node()
1462          * will return -1. There is no CPU on the node, and we should
1463          * select the CPU on the other node.
1464          */
1465         if (nid != -1) {
1466                 nodemask = cpumask_of_node(nid);
1467 
1468                 /* Look for allowed, online CPU in same node. */
1469                 for_each_cpu(dest_cpu, nodemask) {
1470                         if (!cpu_active(dest_cpu))
1471                                 continue;
1472                         if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
1473                                 return dest_cpu;
1474                 }
1475         }
1476 
1477         for (;;) {
1478                 /* Any allowed, online CPU? */
1479                 for_each_cpu(dest_cpu, &p->cpus_allowed) {
1480                         if (!is_cpu_allowed(p, dest_cpu))
1481                                 continue;
1482 
1483                         goto out;
1484                 }
1485 
1486                 /* No more Mr. Nice Guy. */
1487                 switch (state) {
1488                 case cpuset:
1489                         if (IS_ENABLED(CONFIG_CPUSETS)) {
1490                                 cpuset_cpus_allowed_fallback(p);
1491                                 state = possible;
1492                                 break;
1493                         }
1494                         /* Fall-through */
1495                 case possible:
1496                         do_set_cpus_allowed(p, cpu_possible_mask);
1497                         state = fail;
1498                         break;
1499 
1500                 case fail:
1501                         BUG();
1502                         break;
1503                 }
1504         }
1505 
1506 out:
1507         if (state != cpuset) {
1508                 /*
1509                  * Don't tell them about moving exiting tasks or
1510                  * kernel threads (both mm NULL), since they never
1511                  * leave kernel.
1512                  */
1513                 if (p->mm && printk_ratelimit()) {
1514                         printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1515                                         task_pid_nr(p), p->comm, cpu);
1516                 }
1517         }
1518 
1519         return dest_cpu;
1520 }
1521 
1522 /*
1523  * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1524  */
1525 static inline
1526 int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1527 {
1528         lockdep_assert_held(&p->pi_lock);
1529 
1530         if (p->nr_cpus_allowed > 1)
1531                 cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1532         else
1533                 cpu = cpumask_any(&p->cpus_allowed);
1534 
1535         /*
1536          * In order not to call set_task_cpu() on a blocking task we need
1537          * to rely on ttwu() to place the task on a valid ->cpus_allowed
1538          * CPU.
1539          *
1540          * Since this is common to all placement strategies, this lives here.
1541          *
1542          * [ this allows ->select_task() to simply return task_cpu(p) and
1543          *   not worry about this generic constraint ]
1544          */
1545         if (unlikely(!is_cpu_allowed(p, cpu)))
1546                 cpu = select_fallback_rq(task_cpu(p), p);
1547 
1548         return cpu;
1549 }
1550 
1551 static void update_avg(u64 *avg, u64 sample)
1552 {
1553         s64 diff = sample - *avg;
1554         *avg += diff >> 3;
1555 }
1556 
1557 void sched_set_stop_task(int cpu, struct task_struct *stop)
1558 {
1559         struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
1560         struct task_struct *old_stop = cpu_rq(cpu)->stop;
1561 
1562         if (stop) {
1563                 /*
1564                  * Make it appear like a SCHED_FIFO task, its something
1565                  * userspace knows about and won't get confused about.
1566                  *
1567                  * Also, it will make PI more or less work without too
1568                  * much confusion -- but then, stop work should not
1569                  * rely on PI working anyway.
1570                  */
1571                 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
1572 
1573                 stop->sched_class = &stop_sched_class;
1574         }
1575 
1576         cpu_rq(cpu)->stop = stop;
1577 
1578         if (old_stop) {
1579                 /*
1580                  * Reset it back to a normal scheduling class so that
1581                  * it can die in pieces.
1582                  */
1583                 old_stop->sched_class = &rt_sched_class;
1584         }
1585 }
1586 
1587 #else
1588 
1589 static inline int __set_cpus_allowed_ptr(struct task_struct *p,
1590                                          const struct cpumask *new_mask, bool check)
1591 {
1592         return set_cpus_allowed_ptr(p, new_mask);
1593 }
1594 
1595 #endif /* CONFIG_SMP */
1596 
1597 static void
1598 ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
1599 {
1600         struct rq *rq;
1601 
1602         if (!schedstat_enabled())
1603                 return;
1604 
1605         rq = this_rq();
1606 
1607 #ifdef CONFIG_SMP
1608         if (cpu == rq->cpu) {
1609                 __schedstat_inc(rq->ttwu_local);
1610                 __schedstat_inc(p->se.statistics.nr_wakeups_local);
1611         } else {
1612                 struct sched_domain *sd;
1613 
1614                 __schedstat_inc(p->se.statistics.nr_wakeups_remote);
1615                 rcu_read_lock();
1616                 for_each_domain(rq->cpu, sd) {
1617                         if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1618                                 __schedstat_inc(sd->ttwu_wake_remote);
1619                                 break;
1620                         }
1621                 }
1622                 rcu_read_unlock();
1623         }
1624 
1625         if (wake_flags & WF_MIGRATED)
1626                 __schedstat_inc(p->se.statistics.nr_wakeups_migrate);
1627 #endif /* CONFIG_SMP */
1628 
1629         __schedstat_inc(rq->ttwu_count);
1630         __schedstat_inc(p->se.statistics.nr_wakeups);
1631 
1632         if (wake_flags & WF_SYNC)
1633                 __schedstat_inc(p->se.statistics.nr_wakeups_sync);
1634 }
1635 
1636 static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
1637 {
1638         activate_task(rq, p, en_flags);
1639         p->on_rq = TASK_ON_RQ_QUEUED;
1640 
1641         /* If a worker is waking up, notify the workqueue: */
1642         if (p->flags & PF_WQ_WORKER)
1643                 wq_worker_waking_up(p, cpu_of(rq));
1644 }
1645 
1646 /*
1647  * Mark the task runnable and perform wakeup-preemption.
1648  */
1649 static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
1650                            struct rq_flags *rf)
1651 {
1652         check_preempt_curr(rq, p, wake_flags);
1653         p->state = TASK_RUNNING;
1654         trace_sched_wakeup(p);
1655 
1656 #ifdef CONFIG_SMP
1657         if (p->sched_class->task_woken) {
1658                 /*
1659                  * Our task @p is fully woken up and running; so its safe to
1660                  * drop the rq->lock, hereafter rq is only used for statistics.
1661                  */
1662                 rq_unpin_lock(rq, rf);
1663                 p->sched_class->task_woken(rq, p);
1664                 rq_repin_lock(rq, rf);
1665         }
1666 
1667         if (rq->idle_stamp) {
1668                 u64 delta = rq_clock(rq) - rq->idle_stamp;
1669                 u64 max = 2*rq->max_idle_balance_cost;
1670 
1671                 update_avg(&rq->avg_idle, delta);
1672 
1673                 if (rq->avg_idle > max)
1674                         rq->avg_idle = max;
1675 
1676                 rq->idle_stamp = 0;
1677         }
1678 #endif
1679 }
1680 
1681 static void
1682 ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
1683                  struct rq_flags *rf)
1684 {
1685         int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK;
1686 
1687         lockdep_assert_held(&rq->lock);
1688 
1689 #ifdef CONFIG_SMP
1690         if (p->sched_contributes_to_load)
1691                 rq->nr_uninterruptible--;
1692 
1693         if (wake_flags & WF_MIGRATED)
1694                 en_flags |= ENQUEUE_MIGRATED;
1695 #endif
1696 
1697         ttwu_activate(rq, p, en_flags);
1698         ttwu_do_wakeup(rq, p, wake_flags, rf);
1699 }
1700 
1701 /*
1702  * Called in case the task @p isn't fully descheduled from its runqueue,
1703  * in this case we must do a remote wakeup. Its a 'light' wakeup though,
1704  * since all we need to do is flip p->state to TASK_RUNNING, since
1705  * the task is still ->on_rq.
1706  */
1707 static int ttwu_remote(struct task_struct *p, int wake_flags)
1708 {
1709         struct rq_flags rf;
1710         struct rq *rq;
1711         int ret = 0;
1712 
1713         rq = __task_rq_lock(p, &rf);
1714         if (task_on_rq_queued(p)) {
1715                 /* check_preempt_curr() may use rq clock */
1716                 update_rq_clock(rq);
1717                 ttwu_do_wakeup(rq, p, wake_flags, &rf);
1718                 ret = 1;
1719         }
1720         __task_rq_unlock(rq, &rf);
1721 
1722         return ret;
1723 }
1724 
1725 #ifdef CONFIG_SMP
1726 void sched_ttwu_pending(void)
1727 {
1728         struct rq *rq = this_rq();
1729         struct llist_node *llist = llist_del_all(&rq->wake_list);
1730         struct task_struct *p, *t;
1731         struct rq_flags rf;
1732 
1733         if (!llist)
1734                 return;
1735 
1736         rq_lock_irqsave(rq, &rf);
1737         update_rq_clock(rq);
1738 
1739         llist_for_each_entry_safe(p, t, llist, wake_entry)
1740                 ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
1741 
1742         rq_unlock_irqrestore(rq, &rf);
1743 }
1744 
1745 void scheduler_ipi(void)
1746 {
1747         /*
1748          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1749          * TIF_NEED_RESCHED remotely (for the first time) will also send
1750          * this IPI.
1751          */
1752         preempt_fold_need_resched();
1753 
1754         if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1755                 return;
1756 
1757         /*
1758          * Not all reschedule IPI handlers call irq_enter/irq_exit, since
1759          * traditionally all their work was done from the interrupt return
1760          * path. Now that we actually do some work, we need to make sure
1761          * we do call them.
1762          *
1763          * Some archs already do call them, luckily irq_enter/exit nest
1764          * properly.
1765          *
1766          * Arguably we should visit all archs and update all handlers,
1767          * however a fair share of IPIs are still resched only so this would
1768          * somewhat pessimize the simple resched case.
1769          */
1770         irq_enter();
1771         sched_ttwu_pending();
1772 
1773         /*
1774          * Check if someone kicked us for doing the nohz idle load balance.
1775          */
1776         if (unlikely(got_nohz_idle_kick())) {
1777                 this_rq()->idle_balance = 1;
1778                 raise_softirq_irqoff(SCHED_SOFTIRQ);
1779         }
1780         irq_exit();
1781 }
1782 
1783 static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1784 {
1785         struct rq *rq = cpu_rq(cpu);
1786 
1787         p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);
1788 
1789         if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) {
1790                 if (!set_nr_if_polling(rq->idle))
1791                         smp_send_reschedule(cpu);
1792                 else
1793                         trace_sched_wake_idle_without_ipi(cpu);
1794         }
1795 }
1796 
1797 void wake_up_if_idle(int cpu)
1798 {
1799         struct rq *rq = cpu_rq(cpu);
1800         struct rq_flags rf;
1801 
1802         rcu_read_lock();
1803 
1804         if (!is_idle_task(rcu_dereference(rq->curr)))
1805                 goto out;
1806 
1807         if (set_nr_if_polling(rq->idle)) {
1808                 trace_sched_wake_idle_without_ipi(cpu);
1809         } else {
1810                 rq_lock_irqsave(rq, &rf);
1811                 if (is_idle_task(rq->curr))
1812                         smp_send_reschedule(cpu);
1813                 /* Else CPU is not idle, do nothing here: */
1814                 rq_unlock_irqrestore(rq, &rf);
1815         }
1816 
1817 out:
1818         rcu_read_unlock();
1819 }
1820 
1821 bool cpus_share_cache(int this_cpu, int that_cpu)
1822 {
1823         return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
1824 }
1825 #endif /* CONFIG_SMP */
1826 
1827 static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1828 {
1829         struct rq *rq = cpu_rq(cpu);
1830         struct rq_flags rf;
1831 
1832 #if defined(CONFIG_SMP)
1833         if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
1834                 sched_clock_cpu(cpu); /* Sync clocks across CPUs */
1835                 ttwu_queue_remote(p, cpu, wake_flags);
1836                 return;
1837         }
1838 #endif
1839 
1840         rq_lock(rq, &rf);
1841         update_rq_clock(rq);
1842         ttwu_do_activate(rq, p, wake_flags, &rf);
1843         rq_unlock(rq, &rf);
1844 }
1845 
1846 /*
1847  * Notes on Program-Order guarantees on SMP systems.
1848  *
1849  *  MIGRATION
1850  *
1851  * The basic program-order guarantee on SMP systems is that when a task [t]
1852  * migrates, all its activity on its old CPU [c0] happens-before any subsequent
1853  * execution on its new CPU [c1].
1854  *
1855  * For migration (of runnable tasks) this is provided by the following means:
1856  *
1857  *  A) UNLOCK of the rq(c0)->lock scheduling out task t
1858  *  B) migration for t is required to synchronize *both* rq(c0)->lock and
1859  *     rq(c1)->lock (if not at the same time, then in that order).
1860  *  C) LOCK of the rq(c1)->lock scheduling in task
1861  *
1862  * Release/acquire chaining guarantees that B happens after A and C after B.
1863  * Note: the CPU doing B need not be c0 or c1
1864  *
1865  * Example:
1866  *
1867  *   CPU0            CPU1            CPU2
1868  *
1869  *   LOCK rq(0)->lock
1870  *   sched-out X
1871  *   sched-in Y
1872  *   UNLOCK rq(0)->lock
1873  *
1874  *                                   LOCK rq(0)->lock // orders against CPU0
1875  *                                   dequeue X
1876  *                                   UNLOCK rq(0)->lock
1877  *
1878  *                                   LOCK rq(1)->lock
1879  *                                   enqueue X
1880  *                                   UNLOCK rq(1)->lock
1881  *
1882  *                   LOCK rq(1)->lock // orders against CPU2
1883  *                   sched-out Z
1884  *                   sched-in X
1885  *                   UNLOCK rq(1)->lock
1886  *
1887  *
1888  *  BLOCKING -- aka. SLEEP + WAKEUP
1889  *
1890  * For blocking we (obviously) need to provide the same guarantee as for
1891  * migration. However the means are completely different as there is no lock
1892  * chain to provide order. Instead we do:
1893  *
1894  *   1) smp_store_release(X->on_cpu, 0)
1895  *   2) smp_cond_load_acquire(!X->on_cpu)
1896  *
1897  * Example:
1898  *
1899  *   CPU0 (schedule)  CPU1 (try_to_wake_up) CPU2 (schedule)
1900  *
1901  *   LOCK rq(0)->lock LOCK X->pi_lock
1902  *   dequeue X
1903  *   sched-out X
1904  *   smp_store_release(X->on_cpu, 0);
1905  *
1906  *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1907  *                    X->state = WAKING
1908  *                    set_task_cpu(X,2)
1909  *
1910  *                    LOCK rq(2)->lock
1911  *                    enqueue X
1912  *                    X->state = RUNNING
1913  *                    UNLOCK rq(2)->lock
1914  *
1915  *                                          LOCK rq(2)->lock // orders against CPU1
1916  *                                          sched-out Z
1917  *                                          sched-in X
1918  *                                          UNLOCK rq(2)->lock
1919  *
1920  *                    UNLOCK X->pi_lock
1921  *   UNLOCK rq(0)->lock
1922  *
1923  *
1924  * However, for wakeups there is a second guarantee we must provide, namely we
1925  * must ensure that CONDITION=1 done by the caller can not be reordered with
1926  * accesses to the task state; see try_to_wake_up() and set_current_state().
1927  */
1928 
1929 /**
1930  * try_to_wake_up - wake up a thread
1931  * @p: the thread to be awakened
1932  * @state: the mask of task states that can be woken
1933  * @wake_flags: wake modifier flags (WF_*)
1934  *
1935  * If (@state & @p->state) @p->state = TASK_RUNNING.
1936  *
1937  * If the task was not queued/runnable, also place it back on a runqueue.
1938  *
1939  * Atomic against schedule() which would dequeue a task, also see
1940  * set_current_state().
1941  *
1942  * This function executes a full memory barrier before accessing the task
1943  * state; see set_current_state().
1944  *
1945  * Return: %true if @p->state changes (an actual wakeup was done),
1946  *         %false otherwise.
1947  */
1948 static int
1949 try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
1950 {
1951         unsigned long flags;
1952         int cpu, success = 0;
1953 
1954         /*
1955          * If we are going to wake up a thread waiting for CONDITION we
1956          * need to ensure that CONDITION=1 done by the caller can not be
1957          * reordered with p->state check below. This pairs with mb() in
1958          * set_current_state() the waiting thread does.
1959          */
1960         raw_spin_lock_irqsave(&p->pi_lock, flags);
1961         smp_mb__after_spinlock();
1962         if (!(p->state & state))
1963                 goto out;
1964 
1965         trace_sched_waking(p);
1966 
1967         /* We're going to change ->state: */
1968         success = 1;
1969         cpu = task_cpu(p);
1970 
1971         /*
1972          * Ensure we load p->on_rq _after_ p->state, otherwise it would
1973          * be possible to, falsely, observe p->on_rq == 0 and get stuck
1974          * in smp_cond_load_acquire() below.
1975          *
1976          * sched_ttwu_pending()                 try_to_wake_up()
1977          *   STORE p->on_rq = 1                   LOAD p->state
1978          *   UNLOCK rq->lock
1979          *
1980          * __schedule() (switch to task 'p')
1981          *   LOCK rq->lock                        smp_rmb();
1982          *   smp_mb__after_spinlock();
1983          *   UNLOCK rq->lock
1984          *
1985          * [task p]
1986          *   STORE p->state = UNINTERRUPTIBLE     LOAD p->on_rq
1987          *
1988          * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
1989          * __schedule().  See the comment for smp_mb__after_spinlock().
1990          */
1991         smp_rmb();
1992         if (p->on_rq && ttwu_remote(p, wake_flags))
1993                 goto stat;
1994 
1995 #ifdef CONFIG_SMP
1996         /*
1997          * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be
1998          * possible to, falsely, observe p->on_cpu == 0.
1999          *
2000          * One must be running (->on_cpu == 1) in order to remove oneself
2001          * from the runqueue.
2002          *
2003          * __schedule() (switch to task 'p')    try_to_wake_up()
2004          *   STORE p->on_cpu = 1                  LOAD p->on_rq
2005          *   UNLOCK rq->lock
2006          *
2007          * __schedule() (put 'p' to sleep)
2008          *   LOCK rq->lock                        smp_rmb();
2009          *   smp_mb__after_spinlock();
2010          *   STORE p->on_rq = 0                   LOAD p->on_cpu
2011          *
2012          * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
2013          * __schedule().  See the comment for smp_mb__after_spinlock().
2014          */
2015         smp_rmb();
2016 
2017         /*
2018          * If the owning (remote) CPU is still in the middle of schedule() with
2019          * this task as prev, wait until its done referencing the task.
2020          *
2021          * Pairs with the smp_store_release() in finish_task().
2022          *
2023          * This ensures that tasks getting woken will be fully ordered against
2024          * their previous state and preserve Program Order.
2025          */
2026         smp_cond_load_acquire(&p->on_cpu, !VAL);
2027 
2028         p->sched_contributes_to_load = !!task_contributes_to_load(p);
2029         p->state = TASK_WAKING;
2030 
2031         if (p->in_iowait) {
2032                 delayacct_blkio_end(p);
2033                 atomic_dec(&task_rq(p)->nr_iowait);
2034         }
2035 
2036         cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2037         if (task_cpu(p) != cpu) {
2038                 wake_flags |= WF_MIGRATED;
2039                 set_task_cpu(p, cpu);
2040         }
2041 
2042 #else /* CONFIG_SMP */
2043 
2044         if (p->in_iowait) {
2045                 delayacct_blkio_end(p);
2046                 atomic_dec(&task_rq(p)->nr_iowait);
2047         }
2048 
2049 #endif /* CONFIG_SMP */
2050 
2051         ttwu_queue(p, cpu, wake_flags);
2052 stat:
2053         ttwu_stat(p, cpu, wake_flags);
2054 out:
2055         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2056 
2057         return success;
2058 }
2059 
2060 /**
2061  * try_to_wake_up_local - try to wake up a local task with rq lock held
2062  * @p: the thread to be awakened
2063  * @rf: request-queue flags for pinning
2064  *
2065  * Put @p on the run-queue if it's not already there. The caller must
2066  * ensure that this_rq() is locked, @p is bound to this_rq() and not
2067  * the current task.
2068  */
2069 static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf)
2070 {
2071         struct rq *rq = task_rq(p);
2072 
2073         if (WARN_ON_ONCE(rq != this_rq()) ||
2074             WARN_ON_ONCE(p == current))
2075                 return;
2076 
2077         lockdep_assert_held(&rq->lock);
2078 
2079         if (!raw_spin_trylock(&p->pi_lock)) {
2080                 /*
2081                  * This is OK, because current is on_cpu, which avoids it being
2082                  * picked for load-balance and preemption/IRQs are still
2083                  * disabled avoiding further scheduler activity on it and we've
2084                  * not yet picked a replacement task.
2085                  */
2086                 rq_unlock(rq, rf);
2087                 raw_spin_lock(&p->pi_lock);
2088                 rq_relock(rq, rf);
2089         }
2090 
2091         if (!(p->state & TASK_NORMAL))
2092                 goto out;
2093 
2094         trace_sched_waking(p);
2095 
2096         if (!task_on_rq_queued(p)) {
2097                 if (p->in_iowait) {
2098                         delayacct_blkio_end(p);
2099                         atomic_dec(&rq->nr_iowait);
2100                 }
2101                 ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK);
2102         }
2103 
2104         ttwu_do_wakeup(rq, p, 0, rf);
2105         ttwu_stat(p, smp_processor_id(), 0);
2106 out:
2107         raw_spin_unlock(&p->pi_lock);
2108 }
2109 
2110 /**
2111  * wake_up_process - Wake up a specific process
2112  * @p: The process to be woken up.
2113  *
2114  * Attempt to wake up the nominated process and move it to the set of runnable
2115  * processes.
2116  *
2117  * Return: 1 if the process was woken up, 0 if it was already running.
2118  *
2119  * This function executes a full memory barrier before accessing the task state.
2120  */
2121 int wake_up_process(struct task_struct *p)
2122 {
2123         return try_to_wake_up(p, TASK_NORMAL, 0);
2124 }
2125 EXPORT_SYMBOL(wake_up_process);
2126 
2127 int wake_up_state(struct task_struct *p, unsigned int state)
2128 {
2129         return try_to_wake_up(p, state, 0);
2130 }
2131 
2132 /*
2133  * Perform scheduler related setup for a newly forked process p.
2134  * p is forked by current.
2135  *
2136  * __sched_fork() is basic setup used by init_idle() too:
2137  */
2138 static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
2139 {
2140         p->on_rq                        = 0;
2141 
2142         p->se.on_rq                     = 0;
2143         p->se.exec_start                = 0;
2144         p->se.sum_exec_runtime          = 0;
2145         p->se.prev_sum_exec_runtime     = 0;
2146         p->se.nr_migrations             = 0;
2147         p->se.vruntime                  = 0;
2148         INIT_LIST_HEAD(&p->se.group_node);
2149 
2150 #ifdef CONFIG_FAIR_GROUP_SCHED
2151         p->se.cfs_rq                    = NULL;
2152 #endif
2153 
2154 #ifdef CONFIG_SCHEDSTATS
2155         /* Even if schedstat is disabled, there should not be garbage */
2156         memset(&p->se.statistics, 0, sizeof(p->se.statistics));
2157 #endif
2158 
2159         RB_CLEAR_NODE(&p->dl.rb_node);
2160         init_dl_task_timer(&p->dl);
2161         init_dl_inactive_task_timer(&p->dl);
2162         __dl_clear_params(p);
2163 
2164         INIT_LIST_HEAD(&p->rt.run_list);
2165         p->rt.timeout           = 0;
2166         p->rt.time_slice        = sched_rr_timeslice;
2167         p->rt.on_rq             = 0;
2168         p->rt.on_list           = 0;
2169 
2170 #ifdef CONFIG_PREEMPT_NOTIFIERS
2171         INIT_HLIST_HEAD(&p->preempt_notifiers);
2172 #endif
2173 
2174         init_numa_balancing(clone_flags, p);
2175 }
2176 
2177 DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);
2178 
2179 #ifdef CONFIG_NUMA_BALANCING
2180 
2181 void set_numabalancing_state(bool enabled)
2182 {
2183         if (enabled)
2184                 static_branch_enable(&sched_numa_balancing);
2185         else
2186                 static_branch_disable(&sched_numa_balancing);
2187 }
2188 
2189 #ifdef CONFIG_PROC_SYSCTL
2190 int sysctl_numa_balancing(struct ctl_table *table, int write,
2191                          void __user *buffer, size_t *lenp, loff_t *ppos)
2192 {
2193         struct ctl_table t;
2194         int err;
2195         int state = static_branch_likely(&sched_numa_balancing);
2196 
2197         if (write && !capable(CAP_SYS_ADMIN))
2198                 return -EPERM;
2199 
2200         t = *table;
2201         t.data = &state;
2202         err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2203         if (err < 0)
2204                 return err;
2205         if (write)
2206                 set_numabalancing_state(state);
2207         return err;
2208 }
2209 #endif
2210 #endif
2211 
2212 #ifdef CONFIG_SCHEDSTATS
2213 
2214 DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2215 static bool __initdata __sched_schedstats = false;
2216 
2217 static void set_schedstats(bool enabled)
2218 {
2219         if (enabled)
2220                 static_branch_enable(&sched_schedstats);
2221         else
2222                 static_branch_disable(&sched_schedstats);
2223 }
2224 
2225 void force_schedstat_enabled(void)
2226 {
2227         if (!schedstat_enabled()) {
2228                 pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n");
2229                 static_branch_enable(&sched_schedstats);
2230         }
2231 }
2232 
2233 static int __init setup_schedstats(char *str)
2234 {
2235         int ret = 0;
2236         if (!str)
2237                 goto out;
2238 
2239         /*
2240          * This code is called before jump labels have been set up, so we can't
2241          * change the static branch directly just yet.  Instead set a temporary
2242          * variable so init_schedstats() can do it later.
2243          */
2244         if (!strcmp(str, "enable")) {
2245                 __sched_schedstats = true;
2246                 ret = 1;
2247         } else if (!strcmp(str, "disable")) {
2248                 __sched_schedstats = false;
2249                 ret = 1;
2250         }
2251 out:
2252         if (!ret)
2253                 pr_warn("Unable to parse schedstats=\n");
2254 
2255         return ret;
2256 }
2257 __setup("schedstats=", setup_schedstats);
2258 
2259 static void __init init_schedstats(void)
2260 {
2261         set_schedstats(__sched_schedstats);
2262 }
2263 
2264 #ifdef CONFIG_PROC_SYSCTL
2265 int sysctl_schedstats(struct ctl_table *table, int write,
2266                          void __user *buffer, size_t *lenp, loff_t *ppos)
2267 {
2268         struct ctl_table t;
2269         int err;
2270         int state = static_branch_likely(&sched_schedstats);
2271 
2272         if (write && !capable(CAP_SYS_ADMIN))
2273                 return -EPERM;
2274 
2275         t = *table;
2276         t.data = &state;
2277         err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2278         if (err < 0)
2279                 return err;
2280         if (write)
2281                 set_schedstats(state);
2282         return err;
2283 }
2284 #endif /* CONFIG_PROC_SYSCTL */
2285 #else  /* !CONFIG_SCHEDSTATS */
2286 static inline void init_schedstats(void) {}
2287 #endif /* CONFIG_SCHEDSTATS */
2288 
2289 /*
2290  * fork()/clone()-time setup:
2291  */
2292 int sched_fork(unsigned long clone_flags, struct task_struct *p)
2293 {
2294         unsigned long flags;
2295 
2296         __sched_fork(clone_flags, p);
2297         /*
2298          * We mark the process as NEW here. This guarantees that
2299          * nobody will actually run it, and a signal or other external
2300          * event cannot wake it up and insert it on the runqueue either.
2301          */
2302         p->state = TASK_NEW;
2303 
2304         /*
2305          * Make sure we do not leak PI boosting priority to the child.
2306          */
2307         p->prio = current->normal_prio;
2308 
2309         /*
2310          * Revert to default priority/policy on fork if requested.
2311          */
2312         if (unlikely(p->sched_reset_on_fork)) {
2313                 if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2314                         p->policy = SCHED_NORMAL;
2315                         p->static_prio = NICE_TO_PRIO(0);
2316                         p->rt_priority = 0;
2317                 } else if (PRIO_TO_NICE(p->static_prio) < 0)
2318                         p->static_prio = NICE_TO_PRIO(0);
2319 
2320                 p->prio = p->normal_prio = __normal_prio(p);
2321                 set_load_weight(p, false);
2322 
2323                 /*
2324                  * We don't need the reset flag anymore after the fork. It has
2325                  * fulfilled its duty:
2326                  */
2327                 p->sched_reset_on_fork = 0;
2328         }
2329 
2330         if (dl_prio(p->prio))
2331                 return -EAGAIN;
2332         else if (rt_prio(p->prio))
2333                 p->sched_class = &rt_sched_class;
2334         else
2335                 p->sched_class = &fair_sched_class;
2336 
2337         init_entity_runnable_average(&p->se);
2338 
2339         /*
2340          * The child is not yet in the pid-hash so no cgroup attach races,
2341          * and the cgroup is pinned to this child due to cgroup_fork()
2342          * is ran before sched_fork().
2343          *
2344          * Silence PROVE_RCU.
2345          */
2346         raw_spin_lock_irqsave(&p->pi_lock, flags);
2347         /*
2348          * We're setting the CPU for the first time, we don't migrate,
2349          * so use __set_task_cpu().
2350          */
2351         __set_task_cpu(p, smp_processor_id());
2352         if (p->sched_class->task_fork)
2353                 p->sched_class->task_fork(p);
2354         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2355 
2356 #ifdef CONFIG_SCHED_INFO
2357         if (likely(sched_info_on()))
2358                 memset(&p->sched_info, 0, sizeof(p->sched_info));
2359 #endif
2360 #if defined(CONFIG_SMP)
2361         p->on_cpu = 0;
2362 #endif
2363         init_task_preempt_count(p);
2364 #ifdef CONFIG_SMP
2365         plist_node_init(&p->pushable_tasks, MAX_PRIO);
2366         RB_CLEAR_NODE(&p->pushable_dl_tasks);
2367 #endif
2368         return 0;
2369 }
2370 
2371 unsigned long to_ratio(u64 period, u64 runtime)
2372 {
2373         if (runtime == RUNTIME_INF)
2374                 return BW_UNIT;
2375 
2376         /*
2377          * Doing this here saves a lot of checks in all
2378          * the calling paths, and returning zero seems
2379          * safe for them anyway.
2380          */
2381         if (period == 0)
2382                 return 0;
2383 
2384         return div64_u64(runtime << BW_SHIFT, period);
2385 }
2386 
2387 /*
2388  * wake_up_new_task - wake up a newly created task for the first time.
2389  *
2390  * This function will do some initial scheduler statistics housekeeping
2391  * that must be done for every newly created context, then puts the task
2392  * on the runqueue and wakes it.
2393  */
2394 void wake_up_new_task(struct task_struct *p)
2395 {
2396         struct rq_flags rf;
2397         struct rq *rq;
2398 
2399         raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2400         p->state = TASK_RUNNING;
2401 #ifdef CONFIG_SMP
2402         /*
2403          * Fork balancing, do it here and not earlier because:
2404          *  - cpus_allowed can change in the fork path
2405          *  - any previously selected CPU might disappear through hotplug
2406          *
2407          * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
2408          * as we're not fully set-up yet.
2409          */
2410         p->recent_used_cpu = task_cpu(p);
2411         __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2412 #endif
2413         rq = __task_rq_lock(p, &rf);
2414         update_rq_clock(rq);
2415         post_init_entity_util_avg(&p->se);
2416 
2417         activate_task(rq, p, ENQUEUE_NOCLOCK);
2418         p->on_rq = TASK_ON_RQ_QUEUED;
2419         trace_sched_wakeup_new(p);
2420         check_preempt_curr(rq, p, WF_FORK);
2421 #ifdef CONFIG_SMP
2422         if (p->sched_class->task_woken) {
2423                 /*
2424                  * Nothing relies on rq->lock after this, so its fine to
2425                  * drop it.
2426                  */
2427                 rq_unpin_lock(rq, &rf);
2428                 p->sched_class->task_woken(rq, p);
2429                 rq_repin_lock(rq, &rf);
2430         }
2431 #endif
2432         task_rq_unlock(rq, p, &rf);
2433 }
2434 
2435 #ifdef CONFIG_PREEMPT_NOTIFIERS
2436 
2437 static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key);
2438 
2439 void preempt_notifier_inc(void)
2440 {
2441         static_branch_inc(&preempt_notifier_key);
2442 }
2443 EXPORT_SYMBOL_GPL(preempt_notifier_inc);
2444 
2445 void preempt_notifier_dec(void)
2446 {
2447         static_branch_dec(&preempt_notifier_key);
2448 }
2449 EXPORT_SYMBOL_GPL(preempt_notifier_dec);
2450 
2451 /**
2452  * preempt_notifier_register - tell me when current is being preempted & rescheduled
2453  * @notifier: notifier struct to register
2454  */
2455 void preempt_notifier_register(struct preempt_notifier *notifier)
2456 {
2457         if (!static_branch_unlikely(&preempt_notifier_key))
2458                 WARN(1, "registering preempt_notifier while notifiers disabled\n");
2459 
2460         hlist_add_head(&notifier->link, &current->preempt_notifiers);
2461 }
2462 EXPORT_SYMBOL_GPL(preempt_notifier_register);
2463 
2464 /**
2465  * preempt_notifier_unregister - no longer interested in preemption notifications
2466  * @notifier: notifier struct to unregister
2467  *
2468  * This is *not* safe to call from within a preemption notifier.
2469  */
2470 void preempt_notifier_unregister(struct preempt_notifier *notifier)
2471 {
2472         hlist_del(&notifier->link);
2473 }
2474 EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2475 
2476 static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2477 {
2478         struct preempt_notifier *notifier;
2479 
2480         hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2481                 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2482 }
2483 
2484 static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2485 {
2486         if (static_branch_unlikely(&preempt_notifier_key))
2487                 __fire_sched_in_preempt_notifiers(curr);
2488 }
2489 
2490 static void
2491 __fire_sched_out_preempt_notifiers(struct task_struct *curr,
2492                                    struct task_struct *next)
2493 {
2494         struct preempt_notifier *notifier;
2495 
2496         hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2497                 notifier->ops->sched_out(notifier, next);
2498 }
2499 
2500 static __always_inline void
2501 fire_sched_out_preempt_notifiers(struct task_struct *curr,
2502                                  struct task_struct *next)
2503 {
2504         if (static_branch_unlikely(&preempt_notifier_key))
2505                 __fire_sched_out_preempt_notifiers(curr, next);
2506 }
2507 
2508 #else /* !CONFIG_PREEMPT_NOTIFIERS */
2509 
2510 static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2511 {
2512 }
2513 
2514 static inline void
2515 fire_sched_out_preempt_notifiers(struct task_struct *curr,
2516                                  struct task_struct *next)
2517 {
2518 }
2519 
2520 #endif /* CONFIG_PREEMPT_NOTIFIERS */
2521 
2522 static inline void prepare_task(struct task_struct *next)
2523 {
2524 #ifdef CONFIG_SMP
2525         /*
2526          * Claim the task as running, we do this before switching to it
2527          * such that any running task will have this set.
2528          */
2529         next->on_cpu = 1;
2530 #endif
2531 }
2532 
2533 static inline void finish_task(struct task_struct *prev)
2534 {
2535 #ifdef CONFIG_SMP
2536         /*
2537          * After ->on_cpu is cleared, the task can be moved to a different CPU.
2538          * We must ensure this doesn't happen until the switch is completely
2539          * finished.
2540          *
2541          * In particular, the load of prev->state in finish_task_switch() must
2542          * happen before this.
2543          *
2544          * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
2545          */
2546         smp_store_release(&prev->on_cpu, 0);
2547 #endif
2548 }
2549 
2550 static inline void
2551 prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf)
2552 {
2553         /*
2554          * Since the runqueue lock will be released by the next
2555          * task (which is an invalid locking op but in the case
2556          * of the scheduler it's an obvious special-case), so we
2557          * do an early lockdep release here:
2558          */
2559         rq_unpin_lock(rq, rf);
2560         spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
2561 #ifdef CONFIG_DEBUG_SPINLOCK
2562         /* this is a valid case when another task releases the spinlock */
2563         rq->lock.owner = next;
2564 #endif
2565 }
2566 
2567 static inline void finish_lock_switch(struct rq *rq)
2568 {
2569         /*
2570          * If we are tracking spinlock dependencies then we have to
2571          * fix up the runqueue lock - which gets 'carried over' from
2572          * prev into current:
2573          */
2574         spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
2575         raw_spin_unlock_irq(&rq->lock);
2576 }
2577 
2578 /*
2579  * NOP if the arch has not defined these:
2580  */
2581 
2582 #ifndef prepare_arch_switch
2583 # define prepare_arch_switch(next)      do { } while (0)
2584 #endif
2585 
2586 #ifndef finish_arch_post_lock_switch
2587 # define finish_arch_post_lock_switch() do { } while (0)
2588 #endif
2589 
2590 /**
2591  * prepare_task_switch - prepare to switch tasks
2592  * @rq: the runqueue preparing to switch
2593  * @prev: the current task that is being switched out
2594  * @next: the task we are going to switch to.
2595  *
2596  * This is called with the rq lock held and interrupts off. It must
2597  * be paired with a subsequent finish_task_switch after the context
2598  * switch.
2599  *
2600  * prepare_task_switch sets up locking and calls architecture specific
2601  * hooks.
2602  */
2603 static inline void
2604 prepare_task_switch(struct rq *rq, struct task_struct *prev,
2605                     struct task_struct *next)
2606 {
2607         kcov_prepare_switch(prev);
2608         sched_info_switch(rq, prev, next);
2609         perf_event_task_sched_out(prev, next);
2610         rseq_preempt(prev);
2611         fire_sched_out_preempt_notifiers(prev, next);
2612         prepare_task(next);
2613         prepare_arch_switch(next);
2614 }
2615 
2616 /**
2617  * finish_task_switch - clean up after a task-switch
2618  * @prev: the thread we just switched away from.
2619  *
2620  * finish_task_switch must be called after the context switch, paired
2621  * with a prepare_task_switch call before the context switch.
2622  * finish_task_switch will reconcile locking set up by prepare_task_switch,
2623  * and do any other architecture-specific cleanup actions.
2624  *
2625  * Note that we may have delayed dropping an mm in context_switch(). If
2626  * so, we finish that here outside of the runqueue lock. (Doing it
2627  * with the lock held can cause deadlocks; see schedule() for
2628  * details.)
2629  *
2630  * The context switch have flipped the stack from under us and restored the
2631  * local variables which were saved when this task called schedule() in the
2632  * past. prev == current is still correct but we need to recalculate this_rq
2633  * because prev may have moved to another CPU.
2634  */
2635 static struct rq *finish_task_switch(struct task_struct *prev)
2636         __releases(rq->lock)
2637 {
2638         struct rq *rq = this_rq();
2639         struct mm_struct *mm = rq->prev_mm;
2640         long prev_state;
2641 
2642         /*
2643          * The previous task will have left us with a preempt_count of 2
2644          * because it left us after:
2645          *
2646          *      schedule()
2647          *        preempt_disable();                    // 1
2648          *        __schedule()
2649          *          raw_spin_lock_irq(&rq->lock)        // 2
2650          *
2651          * Also, see FORK_PREEMPT_COUNT.
2652          */
2653         if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET,
2654                       "corrupted preempt_count: %s/%d/0x%x\n",
2655                       current->comm, current->pid, preempt_count()))
2656                 preempt_count_set(FORK_PREEMPT_COUNT);
2657 
2658         rq->prev_mm = NULL;
2659 
2660         /*
2661          * A task struct has one reference for the use as "current".
2662          * If a task dies, then it sets TASK_DEAD in tsk->state and calls
2663          * schedule one last time. The schedule call will never return, and
2664          * the scheduled task must drop that reference.
2665          *
2666          * We must observe prev->state before clearing prev->on_cpu (in
2667          * finish_task), otherwise a concurrent wakeup can get prev
2668          * running on another CPU and we could rave with its RUNNING -> DEAD
2669          * transition, resulting in a double drop.
2670          */
2671         prev_state = prev->state;
2672         vtime_task_switch(prev);
2673         perf_event_task_sched_in(prev, current);
2674         finish_task(prev);
2675         finish_lock_switch(rq);
2676         finish_arch_post_lock_switch();
2677         kcov_finish_switch(current);
2678 
2679         fire_sched_in_preempt_notifiers(current);
2680         /*
2681          * When switching through a kernel thread, the loop in
2682          * membarrier_{private,global}_expedited() may have observed that
2683          * kernel thread and not issued an IPI. It is therefore possible to
2684          * schedule between user->kernel->user threads without passing though
2685          * switch_mm(). Membarrier requires a barrier after storing to
2686          * rq->curr, before returning to userspace, so provide them here:
2687          *
2688          * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly
2689          *   provided by mmdrop(),
2690          * - a sync_core for SYNC_CORE.
2691          */
2692         if (mm) {
2693                 membarrier_mm_sync_core_before_usermode(mm);
2694                 mmdrop(mm);
2695         }
2696         if (unlikely(prev_state == TASK_DEAD)) {
2697                 if (prev->sched_class->task_dead)
2698                         prev->sched_class->task_dead(prev);
2699 
2700                 /*
2701                  * Remove function-return probe instances associated with this
2702                  * task and put them back on the free list.
2703                  */
2704                 kprobe_flush_task(prev);
2705 
2706                 /* Task is done with its stack. */
2707                 put_task_stack(prev);
2708 
2709                 put_task_struct(prev);
2710         }
2711 
2712         tick_nohz_task_switch();
2713         return rq;
2714 }
2715 
2716 #ifdef CONFIG_SMP
2717 
2718 /* rq->lock is NOT held, but preemption is disabled */
2719 static void __balance_callback(struct rq *rq)
2720 {
2721         struct callback_head *head, *next;
2722         void (*func)(struct rq *rq);
2723         unsigned long flags;
2724 
2725         raw_spin_lock_irqsave(&rq->lock, flags);
2726         head = rq->balance_callback;
2727         rq->balance_callback = NULL;
2728         while (head) {
2729                 func = (void (*)(struct rq *))head->func;
2730                 next = head->next;
2731                 head->next = NULL;
2732                 head = next;
2733 
2734                 func(rq);
2735         }
2736         raw_spin_unlock_irqrestore(&rq->lock, flags);
2737 }
2738 
2739 static inline void balance_callback(struct rq *rq)
2740 {
2741         if (unlikely(rq->balance_callback))
2742                 __balance_callback(rq);
2743 }
2744 
2745 #else
2746 
2747 static inline void balance_callback(struct rq *rq)
2748 {
2749 }
2750 
2751 #endif
2752 
2753 /**
2754  * schedule_tail - first thing a freshly forked thread must call.
2755  * @prev: the thread we just switched away from.
2756  */
2757 asmlinkage __visible void schedule_tail(struct task_struct *prev)
2758         __releases(rq->lock)
2759 {
2760         struct rq *rq;
2761 
2762         /*
2763          * New tasks start with FORK_PREEMPT_COUNT, see there and
2764          * finish_task_switch() for details.
2765          *
2766          * finish_task_switch() will drop rq->lock() and lower preempt_count
2767          * and the preempt_enable() will end up enabling preemption (on
2768          * PREEMPT_COUNT kernels).
2769          */
2770 
2771         rq = finish_task_switch(prev);
2772         balance_callback(rq);
2773         preempt_enable();
2774 
2775         if (current->set_child_tid)
2776                 put_user(task_pid_vnr(current), current->set_child_tid);
2777 
2778         calculate_sigpending();
2779 }
2780 
2781 /*
2782  * context_switch - switch to the new MM and the new thread's register state.
2783  */
2784 static __always_inline struct rq *
2785 context_switch(struct rq *rq, struct task_struct *prev,
2786                struct task_struct *next, struct rq_flags *rf)
2787 {
2788         struct mm_struct *mm, *oldmm;
2789 
2790         prepare_task_switch(rq, prev, next);
2791 
2792         mm = next->mm;
2793         oldmm = prev->active_mm;
2794         /*
2795          * For paravirt, this is coupled with an exit in switch_to to
2796          * combine the page table reload and the switch backend into
2797          * one hypercall.
2798          */
2799         arch_start_context_switch(prev);
2800 
2801         /*
2802          * If mm is non-NULL, we pass through switch_mm(). If mm is
2803          * NULL, we will pass through mmdrop() in finish_task_switch().
2804          * Both of these contain the full memory barrier required by
2805          * membarrier after storing to rq->curr, before returning to
2806          * user-space.
2807          */
2808         if (!mm) {
2809                 next->active_mm = oldmm;
2810                 mmgrab(oldmm);
2811                 enter_lazy_tlb(oldmm, next);
2812         } else
2813                 switch_mm_irqs_off(oldmm, mm, next);
2814 
2815         if (!prev->mm) {
2816                 prev->active_mm = NULL;
2817                 rq->prev_mm = oldmm;
2818         }
2819 
2820         rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
2821 
2822         prepare_lock_switch(rq, next, rf);
2823 
2824         /* Here we just switch the register state and the stack. */
2825         switch_to(prev, next, prev);
2826         barrier();
2827 
2828         return finish_task_switch(prev);
2829 }
2830 
2831 /*
2832  * nr_running and nr_context_switches:
2833  *
2834  * externally visible scheduler statistics: current number of runnable
2835  * threads, total number of context switches performed since bootup.
2836  */
2837 unsigned long nr_running(void)
2838 {
2839         unsigned long i, sum = 0;
2840 
2841         for_each_online_cpu(i)
2842                 sum += cpu_rq(i)->nr_running;
2843 
2844         return sum;
2845 }
2846 
2847 /*
2848  * Check if only the current task is running on the CPU.
2849  *
2850  * Caution: this function does not check that the caller has disabled
2851  * preemption, thus the result might have a time-of-check-to-time-of-use
2852  * race.  The caller is responsible to use it correctly, for example:
2853  *
2854  * - from a non-preemptable section (of course)
2855  *
2856  * - from a thread that is bound to a single CPU
2857  *
2858  * - in a loop with very short iterations (e.g. a polling loop)
2859  */
2860 bool single_task_running(void)
2861 {
2862         return raw_rq()->nr_running == 1;
2863 }
2864 EXPORT_SYMBOL(single_task_running);
2865 
2866 unsigned long long nr_context_switches(void)
2867 {
2868         int i;
2869         unsigned long long sum = 0;
2870 
2871         for_each_possible_cpu(i)
2872                 sum += cpu_rq(i)->nr_switches;
2873 
2874         return sum;
2875 }
2876 
2877 /*
2878  * IO-wait accounting, and how its mostly bollocks (on SMP).
2879  *
2880  * The idea behind IO-wait account is to account the idle time that we could
2881  * have spend running if it were not for IO. That is, if we were to improve the
2882  * storage performance, we'd have a proportional reduction in IO-wait time.
2883  *
2884  * This all works nicely on UP, where, when a task blocks on IO, we account
2885  * idle time as IO-wait, because if the storage were faster, it could've been
2886  * running and we'd not be idle.
2887  *
2888  * This has been extended to SMP, by doing the same for each CPU. This however
2889  * is broken.
2890  *
2891  * Imagine for instance the case where two tasks block on one CPU, only the one
2892  * CPU will have IO-wait accounted, while the other has regular idle. Even
2893  * though, if the storage were faster, both could've ran at the same time,
2894  * utilising both CPUs.
2895  *
2896  * This means, that when looking globally, the current IO-wait accounting on
2897  * SMP is a lower bound, by reason of under accounting.
2898  *
2899  * Worse, since the numbers are provided per CPU, they are sometimes
2900  * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly
2901  * associated with any one particular CPU, it can wake to another CPU than it
2902  * blocked on. This means the per CPU IO-wait number is meaningless.
2903  *
2904  * Task CPU affinities can make all that even more 'interesting'.
2905  */
2906 
2907 unsigned long nr_iowait(void)
2908 {
2909         unsigned long i, sum = 0;
2910 
2911         for_each_possible_cpu(i)
2912                 sum += atomic_read(&cpu_rq(i)->nr_iowait);
2913 
2914         return sum;
2915 }
2916 
2917 /*
2918  * Consumers of these two interfaces, like for example the cpufreq menu
2919  * governor are using nonsensical data. Boosting frequency for a CPU that has
2920  * IO-wait which might not even end up running the task when it does become
2921  * runnable.
2922  */
2923 
2924 unsigned long nr_iowait_cpu(int cpu)
2925 {
2926         struct rq *this = cpu_rq(cpu);
2927         return atomic_read(&this->nr_iowait);
2928 }
2929 
2930 void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
2931 {
2932         struct rq *rq = this_rq();
2933         *nr_waiters = atomic_read(&rq->nr_iowait);
2934         *load = rq->load.weight;
2935 }
2936 
2937 #ifdef CONFIG_SMP
2938 
2939 /*
2940  * sched_exec - execve() is a valuable balancing opportunity, because at
2941  * this point the task has the smallest effective memory and cache footprint.
2942  */
2943 void sched_exec(void)
2944 {
2945         struct task_struct *p = current;
2946         unsigned long flags;
2947         int dest_cpu;
2948 
2949         raw_spin_lock_irqsave(&p->pi_lock, flags);
2950         dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2951         if (dest_cpu == smp_processor_id())
2952                 goto unlock;
2953 
2954         if (likely(cpu_active(dest_cpu))) {
2955                 struct migration_arg arg = { p, dest_cpu };
2956 
2957                 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2958                 stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
2959                 return;
2960         }
2961 unlock:
2962         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2963 }
2964 
2965 #endif
2966 
2967 DEFINE_PER_CPU(struct kernel_stat, kstat);
2968 DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
2969 
2970 EXPORT_PER_CPU_SYMBOL(kstat);
2971 EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
2972 
2973 /*
2974  * The function fair_sched_class.update_curr accesses the struct curr
2975  * and its field curr->exec_start; when called from task_sched_runtime(),
2976  * we observe a high rate of cache misses in practice.
2977  * Prefetching this data results in improved performance.
2978  */
2979 static inline void prefetch_curr_exec_start(struct task_struct *p)
2980 {
2981 #ifdef CONFIG_FAIR_GROUP_SCHED
2982         struct sched_entity *curr = (&p->se)->cfs_rq->curr;
2983 #else
2984         struct sched_entity *curr = (&task_rq(p)->cfs)->curr;
2985 #endif
2986         prefetch(curr);
2987         prefetch(&curr->exec_start);
2988 }
2989 
2990 /*
2991  * Return accounted runtime for the task.
2992  * In case the task is currently running, return the runtime plus current's
2993  * pending runtime that have not been accounted yet.
2994  */
2995 unsigned long long task_sched_runtime(struct task_struct *p)
2996 {
2997         struct rq_flags rf;
2998         struct rq *rq;
2999         u64 ns;
3000 
3001 #if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
3002         /*
3003          * 64-bit doesn't need locks to atomically read a 64-bit value.
3004          * So we have a optimization chance when the task's delta_exec is 0.
3005          * Reading ->on_cpu is racy, but this is ok.
3006          *
3007          * If we race with it leaving CPU, we'll take a lock. So we're correct.
3008          * If we race with it entering CPU, unaccounted time is 0. This is
3009          * indistinguishable from the read occurring a few cycles earlier.
3010          * If we see ->on_cpu without ->on_rq, the task is leaving, and has
3011          * been accounted, so we're correct here as well.
3012          */
3013         if (!p->on_cpu || !task_on_rq_queued(p))
3014                 return p->se.sum_exec_runtime;
3015 #endif
3016 
3017         rq = task_rq_lock(p, &rf);
3018         /*
3019          * Must be ->curr _and_ ->on_rq.  If dequeued, we would
3020          * project cycles that may never be accounted to this
3021          * thread, breaking clock_gettime().
3022          */
3023         if (task_current(rq, p) && task_on_rq_queued(p)) {
3024                 prefetch_curr_exec_start(p);
3025                 update_rq_clock(rq);
3026                 p->sched_class->update_curr(rq);
3027         }
3028         ns = p->se.sum_exec_runtime;
3029         task_rq_unlock(rq, p, &rf);
3030 
3031         return ns;
3032 }
3033 
3034 /*
3035  * This function gets called by the timer code, with HZ frequency.
3036  * We call it with interrupts disabled.
3037  */
3038 void scheduler_tick(void)
3039 {
3040         int cpu = smp_processor_id();
3041         struct rq *rq = cpu_rq(cpu);
3042         struct task_struct *curr = rq->curr;
3043         struct rq_flags rf;
3044 
3045         sched_clock_tick();
3046 
3047         rq_lock(rq, &rf);
3048 
3049         update_rq_clock(rq);
3050         curr->sched_class->task_tick(rq, curr, 0);
3051         cpu_load_update_active(rq);
3052         calc_global_load_tick(rq);
3053 
3054         rq_unlock(rq, &rf);
3055 
3056         perf_event_task_tick();
3057 
3058 #ifdef CONFIG_SMP
3059         rq->idle_balance = idle_cpu(cpu);
3060         trigger_load_balance(rq);
3061 #endif
3062 }
3063 
3064 #ifdef CONFIG_NO_HZ_FULL
3065 
3066 struct tick_work {
3067         int                     cpu;
3068         struct delayed_work     work;
3069 };
3070 
3071 static struct tick_work __percpu *tick_work_cpu;
3072 
3073 static void sched_tick_remote(struct work_struct *work)
3074 {
3075         struct delayed_work *dwork = to_delayed_work(work);
3076         struct tick_work *twork = container_of(dwork, struct tick_work, work);
3077         int cpu = twork->cpu;
3078         struct rq *rq = cpu_rq(cpu);
3079         struct task_struct *curr;
3080         struct rq_flags rf;
3081         u64 delta;
3082 
3083         /*
3084          * Handle the tick only if it appears the remote CPU is running in full
3085          * dynticks mode. The check is racy by nature, but missing a tick or
3086          * having one too much is no big deal because the scheduler tick updates
3087          * statistics and checks timeslices in a time-independent way, regardless
3088          * of when exactly it is running.
3089          */
3090         if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu))
3091                 goto out_requeue;
3092 
3093         rq_lock_irq(rq, &rf);
3094         curr = rq->curr;
3095         if (is_idle_task(curr))
3096                 goto out_unlock;
3097 
3098         update_rq_clock(rq);
3099         delta = rq_clock_task(rq) - curr->se.exec_start;
3100 
3101         /*
3102          * Make sure the next tick runs within a reasonable
3103          * amount of time.
3104          */
3105         WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
3106         curr->sched_class->task_tick(rq, curr, 0);
3107 
3108 out_unlock:
3109         rq_unlock_irq(rq, &rf);
3110 
3111 out_requeue:
3112         /*
3113          * Run the remote tick once per second (1Hz). This arbitrary
3114          * frequency is large enough to avoid overload but short enough
3115          * to keep scheduler internal stats reasonably up to date.
3116          */
3117         queue_delayed_work(system_unbound_wq, dwork, HZ);
3118 }
3119 
3120 static void sched_tick_start(int cpu)
3121 {
3122         struct tick_work *twork;
3123 
3124         if (housekeeping_cpu(cpu, HK_FLAG_TICK))
3125                 return;
3126 
3127         WARN_ON_ONCE(!tick_work_cpu);
3128 
3129         twork = per_cpu_ptr(tick_work_cpu, cpu);
3130         twork->cpu = cpu;
3131         INIT_DELAYED_WORK(&twork->work, sched_tick_remote);
3132         queue_delayed_work(system_unbound_wq, &twork->work, HZ);
3133 }
3134 
3135 #ifdef CONFIG_HOTPLUG_CPU
3136 static void sched_tick_stop(int cpu)
3137 {
3138         struct tick_work *twork;
3139 
3140         if (housekeeping_cpu(cpu, HK_FLAG_TICK))
3141                 return;
3142 
3143         WARN_ON_ONCE(!tick_work_cpu);
3144 
3145         twork = per_cpu_ptr(tick_work_cpu, cpu);
3146         cancel_delayed_work_sync(&twork->work);
3147 }
3148 #endif /* CONFIG_HOTPLUG_CPU */
3149 
3150 int __init sched_tick_offload_init(void)
3151 {
3152         tick_work_cpu = alloc_percpu(struct tick_work);
3153         BUG_ON(!tick_work_cpu);
3154 
3155         return 0;
3156 }
3157 
3158 #else /* !CONFIG_NO_HZ_FULL */
3159 static inline void sched_tick_start(int cpu) { }
3160 static inline void sched_tick_stop(int cpu) { }
3161 #endif
3162 
3163 #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3164                                 defined(CONFIG_TRACE_PREEMPT_TOGGLE))
3165 /*
3166  * If the value passed in is equal to the current preempt count
3167  * then we just disabled preemption. Start timing the latency.
3168  */
3169 static inline void preempt_latency_start(int val)
3170 {
3171         if (preempt_count() == val) {
3172                 unsigned long ip = get_lock_parent_ip();
3173 #ifdef CONFIG_DEBUG_PREEMPT
3174                 current->preempt_disable_ip = ip;
3175 #endif
3176                 trace_preempt_off(CALLER_ADDR0, ip);
3177         }
3178 }
3179 
3180 void preempt_count_add(int val)
3181 {
3182 #ifdef CONFIG_DEBUG_PREEMPT
3183         /*
3184          * Underflow?
3185          */
3186         if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3187                 return;
3188 #endif
3189         __preempt_count_add(val);
3190 #ifdef CONFIG_DEBUG_PREEMPT
3191         /*
3192          * Spinlock count overflowing soon?
3193          */
3194         DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3195                                 PREEMPT_MASK - 10);
3196 #endif
3197         preempt_latency_start(val);
3198 }
3199 EXPORT_SYMBOL(preempt_count_add);
3200 NOKPROBE_SYMBOL(preempt_count_add);
3201 
3202 /*
3203  * If the value passed in equals to the current preempt count
3204  * then we just enabled preemption. Stop timing the latency.
3205  */
3206 static inline void preempt_latency_stop(int val)
3207 {
3208         if (preempt_count() == val)
3209                 trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
3210 }
3211 
3212 void preempt_count_sub(int val)
3213 {
3214 #ifdef CONFIG_DEBUG_PREEMPT
3215         /*
3216          * Underflow?
3217          */
3218         if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3219                 return;
3220         /*
3221          * Is the spinlock portion underflowing?
3222          */
3223         if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3224                         !(preempt_count() & PREEMPT_MASK)))
3225                 return;
3226 #endif
3227 
3228         preempt_latency_stop(val);
3229         __preempt_count_sub(val);
3230 }
3231 EXPORT_SYMBOL(preempt_count_sub);
3232 NOKPROBE_SYMBOL(preempt_count_sub);
3233 
3234 #else
3235 static inline void preempt_latency_start(int val) { }
3236 static inline void preempt_latency_stop(int val) { }
3237 #endif
3238 
3239 static inline unsigned long get_preempt_disable_ip(struct task_struct *p)
3240 {
3241 #ifdef CONFIG_DEBUG_PREEMPT
3242         return p->preempt_disable_ip;
3243 #else
3244         return 0;
3245 #endif
3246 }
3247 
3248 /*
3249  * Print scheduling while atomic bug:
3250  */
3251 static noinline void __schedule_bug(struct task_struct *prev)
3252 {
3253         /* Save this before calling printk(), since that will clobber it */
3254         unsigned long preempt_disable_ip = get_preempt_disable_ip(current);
3255 
3256         if (oops_in_progress)
3257                 return;
3258 
3259         printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3260                 prev->comm, prev->pid, preempt_count());
3261 
3262         debug_show_held_locks(prev);
3263         print_modules();
3264         if (irqs_disabled())
3265                 print_irqtrace_events(prev);
3266         if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
3267             && in_atomic_preempt_off()) {
3268                 pr_err("Preemption disabled at:");
3269                 print_ip_sym(preempt_disable_ip);
3270                 pr_cont("\n");
3271         }
3272         if (panic_on_warn)
3273                 panic("scheduling while atomic\n");
3274 
3275         dump_stack();
3276         add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
3277 }
3278 
3279 /*
3280  * Various schedule()-time debugging checks and statistics:
3281  */
3282 static inline void schedule_debug(struct task_struct *prev)
3283 {
3284 #ifdef CONFIG_SCHED_STACK_END_CHECK
3285         if (task_stack_end_corrupted(prev))
3286                 panic("corrupted stack end detected inside scheduler\n");
3287 #endif
3288 
3289         if (unlikely(in_atomic_preempt_off())) {
3290                 __schedule_bug(prev);
3291                 preempt_count_set(PREEMPT_DISABLED);
3292         }
3293         rcu_sleep_check();
3294 
3295         profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3296 
3297         schedstat_inc(this_rq()->sched_count);
3298 }
3299 
3300 /*
3301  * Pick up the highest-prio task:
3302  */
3303 static inline struct task_struct *
3304 pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
3305 {
3306         const struct sched_class *class;
3307         struct task_struct *p;
3308 
3309         /*
3310          * Optimization: we know that if all tasks are in the fair class we can
3311          * call that function directly, but only if the @prev task wasn't of a
3312          * higher scheduling class, because otherwise those loose the
3313          * opportunity to pull in more work from other CPUs.
3314          */
3315         if (likely((prev->sched_class == &idle_sched_class ||
3316                     prev->sched_class == &fair_sched_class) &&
3317                    rq->nr_running == rq->cfs.h_nr_running)) {
3318 
3319                 p = fair_sched_class.pick_next_task(rq, prev, rf);
3320                 if (unlikely(p == RETRY_TASK))
3321                         goto again;
3322 
3323                 /* Assumes fair_sched_class->next == idle_sched_class */
3324                 if (unlikely(!p))
3325                         p = idle_sched_class.pick_next_task(rq, prev, rf);
3326 
3327                 return p;
3328         }
3329 
3330 again:
3331         for_each_class(class) {
3332                 p = class->pick_next_task(rq, prev, rf);
3333                 if (p) {
3334                         if (unlikely(p == RETRY_TASK))
3335                                 goto again;
3336                         return p;
3337                 }
3338         }
3339 
3340         /* The idle class should always have a runnable task: */
3341         BUG();
3342 }
3343 
3344 /*
3345  * __schedule() is the main scheduler function.
3346  *
3347  * The main means of driving the scheduler and thus entering this function are:
3348  *
3349  *   1. Explicit blocking: mutex, semaphore, waitqueue, etc.
3350  *
3351  *   2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return
3352  *      paths. For example, see arch/x86/entry_64.S.
3353  *
3354  *      To drive preemption between tasks, the scheduler sets the flag in timer
3355  *      interrupt handler scheduler_tick().
3356  *
3357  *   3. Wakeups don't really cause entry into schedule(). They add a
3358  *      task to the run-queue and that's it.
3359  *
3360  *      Now, if the new task added to the run-queue preempts the current
3361  *      task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets
3362  *      called on the nearest possible occasion:
3363  *
3364  *       - If the kernel is preemptible (CONFIG_PREEMPT=y):
3365  *
3366  *         - in syscall or exception context, at the next outmost
3367  *           preempt_enable(). (this might be as soon as the wake_up()'s
3368  *           spin_unlock()!)
3369  *
3370  *         - in IRQ context, return from interrupt-handler to
3371  *           preemptible context
3372  *
3373  *       - If the kernel is not preemptible (CONFIG_PREEMPT is not set)
3374  *         then at the next:
3375  *
3376  *          - cond_resched() call
3377  *          - explicit schedule() call
3378  *          - return from syscall or exception to user-space
3379  *          - return from interrupt-handler to user-space
3380  *
3381  * WARNING: must be called with preemption disabled!
3382  */
3383 static void __sched notrace __schedule(bool preempt)
3384 {
3385         struct task_struct *prev, *next;
3386         unsigned long *switch_count;
3387         struct rq_flags rf;
3388         struct rq *rq;
3389         int cpu;
3390 
3391         cpu = smp_processor_id();
3392         rq = cpu_rq(cpu);
3393         prev = rq->curr;
3394 
3395         schedule_debug(prev);
3396 
3397         if (sched_feat(HRTICK))
3398                 hrtick_clear(rq);
3399 
3400         local_irq_disable();
3401         rcu_note_context_switch(preempt);
3402 
3403         /*
3404          * Make sure that signal_pending_state()->signal_pending() below
3405          * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE)
3406          * done by the caller to avoid the race with signal_wake_up().
3407          *
3408          * The membarrier system call requires a full memory barrier
3409          * after coming from user-space, before storing to rq->curr.
3410          */
3411         rq_lock(rq, &rf);
3412         smp_mb__after_spinlock();
3413 
3414         /* Promote REQ to ACT */
3415         rq->clock_update_flags <<= 1;
3416         update_rq_clock(rq);
3417 
3418         switch_count = &prev->nivcsw;
3419         if (!preempt && prev->state) {
3420                 if (unlikely(signal_pending_state(prev->state, prev))) {
3421                         prev->state = TASK_RUNNING;
3422                 } else {
3423                         deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);
3424                         prev->on_rq = 0;
3425 
3426                         if (prev->in_iowait) {
3427                                 atomic_inc(&rq->nr_iowait);
3428                                 delayacct_blkio_start();
3429                         }
3430 
3431                         /*
3432                          * If a worker went to sleep, notify and ask workqueue
3433                          * whether it wants to wake up a task to maintain
3434                          * concurrency.
3435                          */
3436                         if (prev->flags & PF_WQ_WORKER) {
3437                                 struct task_struct *to_wakeup;
3438 
3439                                 to_wakeup = wq_worker_sleeping(prev);
3440                                 if (to_wakeup)
3441                                         try_to_wake_up_local(to_wakeup, &rf);
3442                         }
3443                 }
3444                 switch_count = &prev->nvcsw;
3445         }
3446 
3447         next = pick_next_task(rq, prev, &rf);
3448         clear_tsk_need_resched(prev);
3449         clear_preempt_need_resched();
3450 
3451         if (likely(prev != next)) {
3452                 rq->nr_switches++;
3453                 rq->curr = next;
3454                 /*
3455                  * The membarrier system call requires each architecture
3456                  * to have a full memory barrier after updating
3457                  * rq->curr, before returning to user-space.
3458                  *
3459                  * Here are the schemes providing that barrier on the
3460                  * various architectures:
3461                  * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC.
3462                  *   switch_mm() rely on membarrier_arch_switch_mm() on PowerPC.
3463                  * - finish_lock_switch() for weakly-ordered
3464                  *   architectures where spin_unlock is a full barrier,
3465                  * - switch_to() for arm64 (weakly-ordered, spin_unlock
3466                  *   is a RELEASE barrier),
3467                  */
3468                 ++*switch_count;
3469 
3470                 trace_sched_switch(preempt, prev, next);
3471 
3472                 /* Also unlocks the rq: */
3473                 rq = context_switch(rq, prev, next, &rf);
3474         } else {
3475                 rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3476                 rq_unlock_irq(rq, &rf);
3477         }
3478 
3479         balance_callback(rq);
3480 }
3481 
3482 void __noreturn do_task_dead(void)
3483 {
3484         /* Causes final put_task_struct in finish_task_switch(): */
3485         set_special_state(TASK_DEAD);
3486 
3487         /* Tell freezer to ignore us: */
3488         current->flags |= PF_NOFREEZE;
3489 
3490         __schedule(false);
3491         BUG();
3492 
3493         /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */
3494         for (;;)
3495                 cpu_relax();
3496 }
3497 
3498 static inline void sched_submit_work(struct task_struct *tsk)
3499 {
3500         if (!tsk->state || tsk_is_pi_blocked(tsk))
3501                 return;
3502         /*
3503          * If we are going to sleep and we have plugged IO queued,
3504          * make sure to submit it to avoid deadlocks.
3505          */
3506         if (blk_needs_flush_plug(tsk))
3507                 blk_schedule_flush_plug(tsk);
3508 }
3509 
3510 asmlinkage __visible void __sched schedule(void)
3511 {
3512         struct task_struct *tsk = current;
3513 
3514         sched_submit_work(tsk);
3515         do {
3516                 preempt_disable();
3517                 __schedule(false);
3518                 sched_preempt_enable_no_resched();
3519         } while (need_resched());
3520 }
3521 EXPORT_SYMBOL(schedule);
3522 
3523 /*
3524  * synchronize_rcu_tasks() makes sure that no task is stuck in preempted
3525  * state (have scheduled out non-voluntarily) by making sure that all
3526  * tasks have either left the run queue or have gone into user space.
3527  * As idle tasks do not do either, they must not ever be preempted
3528  * (schedule out non-voluntarily).
3529  *
3530  * schedule_idle() is similar to schedule_preempt_disable() except that it
3531  * never enables preemption because it does not call sched_submit_work().
3532  */
3533 void __sched schedule_idle(void)
3534 {
3535         /*
3536          * As this skips calling sched_submit_work(), which the idle task does
3537          * regardless because that function is a nop when the task is in a
3538          * TASK_RUNNING state, make sure this isn't used someplace that the
3539          * current task can be in any other state. Note, idle is always in the
3540          * TASK_RUNNING state.
3541          */
3542         WARN_ON_ONCE(current->state);
3543         do {
3544                 __schedule(false);
3545         } while (need_resched());
3546 }
3547 
3548 #ifdef CONFIG_CONTEXT_TRACKING
3549 asmlinkage __visible void __sched schedule_user(void)
3550 {
3551         /*
3552          * If we come here after a random call to set_need_resched(),
3553          * or we have been woken up remotely but the IPI has not yet arrived,
3554          * we haven't yet exited the RCU idle mode. Do it here manually until
3555          * we find a better solution.
3556          *
3557          * NB: There are buggy callers of this function.  Ideally we
3558          * should warn if prev_state != CONTEXT_USER, but that will trigger
3559          * too frequently to make sense yet.
3560          */
3561         enum ctx_state prev_state = exception_enter();
3562         schedule();
3563         exception_exit(prev_state);
3564 }
3565 #endif
3566 
3567 /**
3568  * schedule_preempt_disabled - called with preemption disabled
3569  *
3570  * Returns with preemption disabled. Note: preempt_count must be 1
3571  */
3572 void __sched schedule_preempt_disabled(void)
3573 {
3574         sched_preempt_enable_no_resched();
3575         schedule();
3576         preempt_disable();
3577 }
3578 
3579 static void __sched notrace preempt_schedule_common(void)
3580 {
3581         do {
3582                 /*
3583                  * Because the function tracer can trace preempt_count_sub()
3584                  * and it also uses preempt_enable/disable_notrace(), if
3585                  * NEED_RESCHED is set, the preempt_enable_notrace() called
3586                  * by the function tracer will call this function again and
3587                  * cause infinite recursion.
3588                  *
3589                  * Preemption must be disabled here before the function
3590                  * tracer can trace. Break up preempt_disable() into two
3591                  * calls. One to disable preemption without fear of being
3592                  * traced. The other to still record the preemption latency,
3593                  * which can also be traced by the function tracer.
3594                  */
3595                 preempt_disable_notrace();
3596                 preempt_latency_start(1);
3597                 __schedule(true);
3598                 preempt_latency_stop(1);
3599                 preempt_enable_no_resched_notrace();
3600 
3601                 /*
3602                  * Check again in case we missed a preemption opportunity
3603                  * between schedule and now.
3604                  */
3605         } while (need_resched());
3606 }
3607 
3608 #ifdef CONFIG_PREEMPT
3609 /*
3610  * this is the entry point to schedule() from in-kernel preemption
3611  * off of preempt_enable. Kernel preemptions off return from interrupt
3612  * occur there and call schedule directly.
3613  */
3614 asmlinkage __visible void __sched notrace preempt_schedule(void)
3615 {
3616         /*
3617          * If there is a non-zero preempt_count or interrupts are disabled,
3618          * we do not want to preempt the current task. Just return..
3619          */
3620         if (likely(!preemptible()))
3621                 return;
3622 
3623         preempt_schedule_common();
3624 }
3625 NOKPROBE_SYMBOL(preempt_schedule);
3626 EXPORT_SYMBOL(preempt_schedule);
3627 
3628 /**
3629  * preempt_schedule_notrace - preempt_schedule called by tracing
3630  *
3631  * The tracing infrastructure uses preempt_enable_notrace to prevent
3632  * recursion and tracing preempt enabling caused by the tracing
3633  * infrastructure itself. But as tracing can happen in areas coming
3634  * from userspace or just about to enter userspace, a preempt enable
3635  * can occur before user_exit() is called. This will cause the scheduler
3636  * to be called when the system is still in usermode.
3637  *
3638  * To prevent this, the preempt_enable_notrace will use this function
3639  * instead of preempt_schedule() to exit user context if needed before
3640  * calling the scheduler.
3641  */
3642 asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3643 {
3644         enum ctx_state prev_ctx;
3645 
3646         if (likely(!preemptible()))
3647                 return;
3648 
3649         do {
3650                 /*
3651                  * Because the function tracer can trace preempt_count_sub()
3652                  * and it also uses preempt_enable/disable_notrace(), if
3653                  * NEED_RESCHED is set, the preempt_enable_notrace() called
3654                  * by the function tracer will call this function again and
3655                  * cause infinite recursion.
3656                  *
3657                  * Preemption must be disabled here before the function
3658                  * tracer can trace. Break up preempt_disable() into two
3659                  * calls. One to disable preemption without fear of being
3660                  * traced. The other to still record the preemption latency,
3661                  * which can also be traced by the function tracer.
3662                  */
3663                 preempt_disable_notrace();
3664                 preempt_latency_start(1);
3665                 /*
3666                  * Needs preempt disabled in case user_exit() is traced
3667                  * and the tracer calls preempt_enable_notrace() causing
3668                  * an infinite recursion.
3669                  */
3670                 prev_ctx = exception_enter();
3671                 __schedule(true);
3672                 exception_exit(prev_ctx);
3673 
3674                 preempt_latency_stop(1);
3675                 preempt_enable_no_resched_notrace();
3676         } while (need_resched());
3677 }
3678 EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3679 
3680 #endif /* CONFIG_PREEMPT */
3681 
3682 /*
3683  * this is the entry point to schedule() from kernel preemption
3684  * off of irq context.
3685  * Note, that this is called and return with irqs disabled. This will
3686  * protect us against recursive calling from irq.
3687  */
3688 asmlinkage __visible void __sched preempt_schedule_irq(void)
3689 {
3690         enum ctx_state prev_state;
3691 
3692         /* Catch callers which need to be fixed */
3693         BUG_ON(preempt_count() || !irqs_disabled());
3694 
3695         prev_state = exception_enter();
3696 
3697         do {
3698                 preempt_disable();
3699                 local_irq_enable();
3700                 __schedule(true);
3701                 local_irq_disable();
3702                 sched_preempt_enable_no_resched();
3703         } while (need_resched());
3704 
3705         exception_exit(prev_state);
3706 }
3707 
3708 int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags,
3709                           void *key)
3710 {
3711         return try_to_wake_up(curr->private, mode, wake_flags);
3712 }
3713 EXPORT_SYMBOL(default_wake_function);
3714 
3715 #ifdef CONFIG_RT_MUTEXES
3716 
3717 static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
3718 {
3719         if (pi_task)
3720                 prio = min(prio, pi_task->prio);
3721 
3722         return prio;
3723 }
3724 
3725 static inline int rt_effective_prio(struct task_struct *p, int prio)
3726 {
3727         struct task_struct *pi_task = rt_mutex_get_top_task(p);
3728 
3729         return __rt_effective_prio(pi_task, prio);
3730 }
3731 
3732 /*
3733  * rt_mutex_setprio - set the current priority of a task
3734  * @p: task to boost
3735  * @pi_task: donor task
3736  *
3737  * This function changes the 'effective' priority of a task. It does
3738  * not touch ->normal_prio like __setscheduler().
3739  *
3740  * Used by the rt_mutex code to implement priority inheritance
3741  * logic. Call site only calls if the priority of the task changed.
3742  */
3743 void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
3744 {
3745         int prio, oldprio, queued, running, queue_flag =
3746                 DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
3747         const struct sched_class *prev_class;
3748         struct rq_flags rf;
3749         struct rq *rq;
3750 
3751         /* XXX used to be waiter->prio, not waiter->task->prio */
3752         prio = __rt_effective_prio(pi_task, p->normal_prio);
3753 
3754         /*
3755          * If nothing changed; bail early.
3756          */
3757         if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio))
3758                 return;
3759 
3760         rq = __task_rq_lock(p, &rf);
3761         update_rq_clock(rq);
3762         /*
3763          * Set under pi_lock && rq->lock, such that the value can be used under
3764          * either lock.
3765          *
3766          * Note that there is loads of tricky to make this pointer cache work
3767          * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to
3768          * ensure a task is de-boosted (pi_task is set to NULL) before the
3769          * task is allowed to run again (and can exit). This ensures the pointer
3770          * points to a blocked task -- which guaratees the task is present.
3771          */
3772         p->pi_top_task = pi_task;
3773 
3774         /*
3775          * For FIFO/RR we only need to set prio, if that matches we're done.
3776          */
3777         if (prio == p->prio && !dl_prio(prio))
3778                 goto out_unlock;
3779 
3780         /*
3781          * Idle task boosting is a nono in general. There is one
3782          * exception, when PREEMPT_RT and NOHZ is active:
3783          *
3784          * The idle task calls get_next_timer_interrupt() and holds
3785          * the timer wheel base->lock on the CPU and another CPU wants
3786          * to access the timer (probably to cancel it). We can safely
3787          * ignore the boosting request, as the idle CPU runs this code
3788          * with interrupts disabled and will complete the lock
3789          * protected section without being interrupted. So there is no
3790          * real need to boost.
3791          */
3792         if (unlikely(p == rq->idle)) {
3793                 WARN_ON(p != rq->curr);
3794                 WARN_ON(p->pi_blocked_on);
3795                 goto out_unlock;
3796         }
3797 
3798         trace_sched_pi_setprio(p, pi_task);
3799         oldprio = p->prio;
3800 
3801         if (oldprio == prio)
3802                 queue_flag &= ~DEQUEUE_MOVE;
3803 
3804         prev_class = p->sched_class;
3805         queued = task_on_rq_queued(p);
3806         running = task_current(rq, p);
3807         if (queued)
3808                 dequeue_task(rq, p, queue_flag);
3809         if (running)
3810                 put_prev_task(rq, p);
3811 
3812         /*
3813          * Boosting condition are:
3814          * 1. -rt task is running and holds mutex A
3815          *      --> -dl task blocks on mutex A
3816          *
3817          * 2. -dl task is running and holds mutex A
3818          *      --> -dl task blocks on mutex A and could preempt the
3819          *          running task
3820          */
3821         if (dl_prio(prio)) {
3822                 if (!dl_prio(p->normal_prio) ||
3823                     (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3824                         p->dl.dl_boosted = 1;
3825                         queue_flag |= ENQUEUE_REPLENISH;
3826                 } else
3827                         p->dl.dl_boosted = 0;
3828                 p->sched_class = &dl_sched_class;
3829         } else if (rt_prio(prio)) {
3830                 if (dl_prio(oldprio))
3831                         p->dl.dl_boosted = 0;
3832                 if (oldprio < prio)
3833                         queue_flag |= ENQUEUE_HEAD;
3834                 p->sched_class = &rt_sched_class;
3835         } else {
3836                 if (dl_prio(oldprio))
3837                         p->dl.dl_boosted = 0;
3838                 if (rt_prio(oldprio))
3839                         p->rt.timeout = 0;
3840                 p->sched_class = &fair_sched_class;
3841         }
3842 
3843         p->prio = prio;
3844 
3845         if (queued)
3846                 enqueue_task(rq, p, queue_flag);
3847         if (running)
3848                 set_curr_task(rq, p);
3849 
3850         check_class_changed(rq, p, prev_class, oldprio);
3851 out_unlock:
3852         /* Avoid rq from going away on us: */
3853         preempt_disable();
3854         __task_rq_unlock(rq, &rf);
3855 
3856         balance_callback(rq);
3857         preempt_enable();
3858 }
3859 #else
3860 static inline int rt_effective_prio(struct task_struct *p, int prio)
3861 {
3862         return prio;
3863 }
3864 #endif
3865 
3866 void set_user_nice(struct task_struct *p, long nice)
3867 {
3868         bool queued, running;
3869         int old_prio, delta;
3870         struct rq_flags rf;
3871         struct rq *rq;
3872 
3873         if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
3874                 return;
3875         /*
3876          * We have to be careful, if called from sys_setpriority(),
3877          * the task might be in the middle of scheduling on another CPU.
3878          */
3879         rq = task_rq_lock(p, &rf);
3880         update_rq_clock(rq);
3881 
3882         /*
3883          * The RT priorities are set via sched_setscheduler(), but we still
3884          * allow the 'normal' nice value to be set - but as expected
3885          * it wont have any effect on scheduling until the task is
3886          * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
3887          */
3888         if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
3889                 p->static_prio = NICE_TO_PRIO(nice);
3890                 goto out_unlock;
3891         }
3892         queued = task_on_rq_queued(p);
3893         running = task_current(rq, p);
3894         if (queued)
3895                 dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
3896         if (running)
3897                 put_prev_task(rq, p);
3898 
3899         p->static_prio = NICE_TO_PRIO(nice);
3900         set_load_weight(p, true);
3901         old_prio = p->prio;
3902         p->prio = effective_prio(p);
3903         delta = p->prio - old_prio;
3904 
3905         if (queued) {
3906                 enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
3907                 /*
3908                  * If the task increased its priority or is running and
3909                  * lowered its priority, then reschedule its CPU:
3910                  */
3911                 if (delta < 0 || (delta > 0 && task_running(rq, p)))
3912                         resched_curr(rq);
3913         }
3914         if (running)
3915                 set_curr_task(rq, p);
3916 out_unlock:
3917         task_rq_unlock(rq, p, &rf);
3918 }
3919 EXPORT_SYMBOL(set_user_nice);
3920 
3921 /*
3922  * can_nice - check if a task can reduce its nice value
3923  * @p: task
3924  * @nice: nice value
3925  */
3926 int can_nice(const struct task_struct *p, const int nice)
3927 {
3928         /* Convert nice value [19,-20] to rlimit style value [1,40]: */
3929         int nice_rlim = nice_to_rlimit(nice);
3930 
3931         return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
3932                 capable(CAP_SYS_NICE));
3933 }
3934 
3935 #ifdef __ARCH_WANT_SYS_NICE
3936 
3937 /*
3938  * sys_nice - change the priority of the current process.
3939  * @increment: priority increment
3940  *
3941  * sys_setpriority is a more generic, but much slower function that
3942  * does similar things.
3943  */
3944 SYSCALL_DEFINE1(nice, int, increment)
3945 {
3946         long nice, retval;
3947         if (!ccs_capable(CCS_SYS_NICE))
3948                 return -EPERM;
3949 
3950         /*
3951          * Setpriority might change our priority at the same moment.
3952          * We don't have to worry. Conceptually one call occurs first
3953          * and we have a single winner.
3954          */
3955         increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3956         nice = task_nice(current) + increment;
3957 
3958         nice = clamp_val(nice, MIN_NICE, MAX_NICE);
3959         if (increment < 0 && !can_nice(current, nice))
3960                 return -EPERM;
3961 
3962         retval = security_task_setnice(current, nice);
3963         if (retval)
3964                 return retval;
3965 
3966         set_user_nice(current, nice);
3967         return 0;
3968 }
3969 
3970 #endif
3971 
3972 /**
3973  * task_prio - return the priority value of a given task.
3974  * @p: the task in question.
3975  *
3976  * Return: The priority value as seen by users in /proc.
3977  * RT tasks are offset by -200. Normal tasks are centered
3978  * around 0, value goes from -16 to +15.
3979  */
3980 int task_prio(const struct task_struct *p)
3981 {
3982         return p->prio - MAX_RT_PRIO;
3983 }
3984 
3985 /**
3986  * idle_cpu - is a given CPU idle currently?
3987  * @cpu: the processor in question.
3988  *
3989  * Return: 1 if the CPU is currently idle. 0 otherwise.
3990  */
3991 int idle_cpu(int cpu)
3992 {
3993         struct rq *rq = cpu_rq(cpu);
3994 
3995         if (rq->curr != rq->idle)
3996                 return 0;
3997 
3998         if (rq->nr_running)
3999                 return 0;
4000 
4001 #ifdef CONFIG_SMP
4002         if (!llist_empty(&rq->wake_list))
4003                 return 0;
4004 #endif
4005 
4006         return 1;
4007 }
4008 
4009 /**
4010  * available_idle_cpu - is a given CPU idle for enqueuing work.
4011  * @cpu: the CPU in question.
4012  *
4013  * Return: 1 if the CPU is currently idle. 0 otherwise.
4014  */
4015 int available_idle_cpu(int cpu)
4016 {
4017         if (!idle_cpu(cpu))
4018                 return 0;
4019 
4020         if (vcpu_is_preempted(cpu))
4021                 return 0;
4022 
4023         return 1;
4024 }
4025 
4026 /**
4027  * idle_task - return the idle task for a given CPU.
4028  * @cpu: the processor in question.
4029  *
4030  * Return: The idle task for the CPU @cpu.
4031  */
4032 struct task_struct *idle_task(int cpu)
4033 {
4034         return cpu_rq(cpu)->idle;
4035 }
4036 
4037 /**
4038  * find_process_by_pid - find a process with a matching PID value.
4039  * @pid: the pid in question.
4040  *
4041  * The task of @pid, if found. %NULL otherwise.
4042  */
4043 static struct task_struct *find_process_by_pid(pid_t pid)
4044 {
4045         return pid ? find_task_by_vpid(pid) : current;
4046 }
4047 
4048 /*
4049  * sched_setparam() passes in -1 for its policy, to let the functions
4050  * it calls know not to change it.
4051  */
4052 #define SETPARAM_POLICY -1
4053 
4054 static void __setscheduler_params(struct task_struct *p,
4055                 const struct sched_attr *attr)
4056 {
4057         int policy = attr->sched_policy;
4058 
4059         if (policy == SETPARAM_POLICY)
4060                 policy = p->policy;
4061 
4062         p->policy = policy;
4063 
4064         if (dl_policy(policy))
4065                 __setparam_dl(p, attr);
4066         else if (fair_policy(policy))
4067                 p->static_prio = NICE_TO_PRIO(attr->sched_nice);
4068 
4069         /*
4070          * __sched_setscheduler() ensures attr->sched_priority == 0 when
4071          * !rt_policy. Always setting this ensures that things like
4072          * getparam()/getattr() don't report silly values for !rt tasks.
4073          */
4074         p->rt_priority = attr->sched_priority;
4075         p->normal_prio = normal_prio(p);
4076         set_load_weight(p, true);
4077 }
4078 
4079 /* Actually do priority change: must hold pi & rq lock. */
4080 static void __setscheduler(struct rq *rq, struct task_struct *p,
4081                            const struct sched_attr *attr, bool keep_boost)
4082 {
4083         __setscheduler_params(p, attr);
4084 
4085         /*
4086          * Keep a potential priority boosting if called from
4087          * sched_setscheduler().
4088          */
4089         p->prio = normal_prio(p);
4090         if (keep_boost)
4091                 p->prio = rt_effective_prio(p, p->prio);
4092 
4093         if (dl_prio(p->prio))
4094                 p->sched_class = &dl_sched_class;
4095         else if (rt_prio(p->prio))
4096                 p->sched_class = &rt_sched_class;
4097         else
4098                 p->sched_class = &fair_sched_class;
4099 }
4100 
4101 /*
4102  * Check the target process has a UID that matches the current process's:
4103  */
4104 static bool check_same_owner(struct task_struct *p)
4105 {
4106         const struct cred *cred = current_cred(), *pcred;
4107         bool match;
4108 
4109         rcu_read_lock();
4110         pcred = __task_cred(p);
4111         match = (uid_eq(cred->euid, pcred->euid) ||
4112                  uid_eq(cred->euid, pcred->uid));
4113         rcu_read_unlock();
4114         return match;
4115 }
4116 
4117 static int __sched_setscheduler(struct task_struct *p,
4118                                 const struct sched_attr *attr,
4119                                 bool user, bool pi)
4120 {
4121         int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
4122                       MAX_RT_PRIO - 1 - attr->sched_priority;
4123         int retval, oldprio, oldpolicy = -1, queued, running;
4124         int new_effective_prio, policy = attr->sched_policy;
4125         const struct sched_class *prev_class;
4126         struct rq_flags rf;
4127         int reset_on_fork;
4128         int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4129         struct rq *rq;
4130 
4131         /* The pi code expects interrupts enabled */
4132         BUG_ON(pi && in_interrupt());
4133 recheck:
4134         /* Double check policy once rq lock held: */
4135         if (policy < 0) {
4136                 reset_on_fork = p->sched_reset_on_fork;
4137                 policy = oldpolicy = p->policy;
4138         } else {
4139                 reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4140 
4141                 if (!valid_policy(policy))
4142                         return -EINVAL;
4143         }
4144 
4145         if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
4146                 return -EINVAL;
4147 
4148         /*
4149          * Valid priorities for SCHED_FIFO and SCHED_RR are
4150          * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4151          * SCHED_BATCH and SCHED_IDLE is 0.
4152          */
4153         if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4154             (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
4155                 return -EINVAL;
4156         if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
4157             (rt_policy(policy) != (attr->sched_priority != 0)))
4158                 return -EINVAL;
4159 
4160         /*
4161          * Allow unprivileged RT tasks to decrease priority:
4162          */
4163         if (user && !capable(CAP_SYS_NICE)) {
4164                 if (fair_policy(policy)) {
4165                         if (attr->sched_nice < task_nice(p) &&
4166                             !can_nice(p, attr->sched_nice))
4167                                 return -EPERM;
4168                 }
4169 
4170                 if (rt_policy(policy)) {
4171                         unsigned long rlim_rtprio =
4172                                         task_rlimit(p, RLIMIT_RTPRIO);
4173 
4174                         /* Can't set/change the rt policy: */
4175                         if (policy != p->policy && !rlim_rtprio)
4176                                 return -EPERM;
4177 
4178                         /* Can't increase priority: */
4179                         if (attr->sched_priority > p->rt_priority &&
4180                             attr->sched_priority > rlim_rtprio)
4181                                 return -EPERM;
4182                 }
4183 
4184                  /*
4185                   * Can't set/change SCHED_DEADLINE policy at all for now
4186                   * (safest behavior); in the future we would like to allow
4187                   * unprivileged DL tasks to increase their relative deadline
4188                   * or reduce their runtime (both ways reducing utilization)
4189                   */
4190                 if (dl_policy(policy))
4191                         return -EPERM;
4192 
4193                 /*
4194                  * Treat SCHED_IDLE as nice 20. Only allow a switch to
4195                  * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
4196                  */
4197                 if (idle_policy(p->policy) && !idle_policy(policy)) {
4198                         if (!can_nice(p, task_nice(p)))
4199                                 return -EPERM;
4200                 }
4201 
4202                 /* Can't change other user's priorities: */
4203                 if (!check_same_owner(p))
4204                         return -EPERM;
4205 
4206                 /* Normal users shall not reset the sched_reset_on_fork flag: */
4207                 if (p->sched_reset_on_fork && !reset_on_fork)
4208                         return -EPERM;
4209         }
4210 
4211         if (user) {
4212                 if (attr->sched_flags & SCHED_FLAG_SUGOV)
4213                         return -EINVAL;
4214 
4215                 retval = security_task_setscheduler(p);
4216                 if (retval)
4217                         return retval;
4218         }
4219 
4220         /*
4221          * Make sure no PI-waiters arrive (or leave) while we are
4222          * changing the priority of the task:
4223          *
4224          * To be able to change p->policy safely, the appropriate
4225          * runqueue lock must be held.
4226          */
4227         rq = task_rq_lock(p, &rf);
4228         update_rq_clock(rq);
4229 
4230         /*
4231          * Changing the policy of the stop threads its a very bad idea:
4232          */
4233         if (p == rq->stop) {
4234                 task_rq_unlock(rq, p, &rf);
4235                 return -EINVAL;
4236         }
4237 
4238         /*
4239          * If not changing anything there's no need to proceed further,
4240          * but store a possible modification of reset_on_fork.
4241          */
4242         if (unlikely(policy == p->policy)) {
4243                 if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4244                         goto change;
4245                 if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
4246                         goto change;
4247                 if (dl_policy(policy) && dl_param_changed(p, attr))
4248                         goto change;
4249 
4250                 p->sched_reset_on_fork = reset_on_fork;
4251                 task_rq_unlock(rq, p, &rf);
4252                 return 0;
4253         }
4254 change:
4255 
4256         if (user) {
4257 #ifdef CONFIG_RT_GROUP_SCHED
4258                 /*
4259                  * Do not allow realtime tasks into groups that have no runtime
4260                  * assigned.
4261                  */
4262                 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4263                                 task_group(p)->rt_bandwidth.rt_runtime == 0 &&
4264                                 !task_group_is_autogroup(task_group(p))) {
4265                         task_rq_unlock(rq, p, &rf);
4266                         return -EPERM;
4267                 }
4268 #endif
4269 #ifdef CONFIG_SMP
4270                 if (dl_bandwidth_enabled() && dl_policy(policy) &&
4271                                 !(attr->sched_flags & SCHED_FLAG_SUGOV)) {
4272                         cpumask_t *span = rq->rd->span;
4273 
4274                         /*
4275                          * Don't allow tasks with an affinity mask smaller than
4276                          * the entire root_domain to become SCHED_DEADLINE. We
4277                          * will also fail if there's no bandwidth available.
4278                          */
4279                         if (!cpumask_subset(span, &p->cpus_allowed) ||
4280                             rq->rd->dl_bw.bw == 0) {
4281                                 task_rq_unlock(rq, p, &rf);
4282                                 return -EPERM;
4283                         }
4284                 }
4285 #endif
4286         }
4287 
4288         /* Re-check policy now with rq lock held: */
4289         if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4290                 policy = oldpolicy = -1;
4291                 task_rq_unlock(rq, p, &rf);
4292                 goto recheck;
4293         }
4294 
4295         /*
4296          * If setscheduling to SCHED_DEADLINE (or changing the parameters
4297          * of a SCHED_DEADLINE task) we need to check if enough bandwidth
4298          * is available.
4299          */
4300         if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
4301                 task_rq_unlock(rq, p, &rf);
4302                 return -EBUSY;
4303         }
4304 
4305         p->sched_reset_on_fork = reset_on_fork;
4306         oldprio = p->prio;
4307 
4308         if (pi) {
4309                 /*
4310                  * Take priority boosted tasks into account. If the new
4311                  * effective priority is unchanged, we just store the new
4312                  * normal parameters and do not touch the scheduler class and
4313                  * the runqueue. This will be done when the task deboost
4314                  * itself.
4315                  */
4316                 new_effective_prio = rt_effective_prio(p, newprio);
4317                 if (new_effective_prio == oldprio)
4318                         queue_flags &= ~DEQUEUE_MOVE;
4319         }
4320 
4321         queued = task_on_rq_queued(p);
4322         running = task_current(rq, p);
4323         if (queued)
4324                 dequeue_task(rq, p, queue_flags);
4325         if (running)
4326                 put_prev_task(rq, p);
4327 
4328         prev_class = p->sched_class;
4329         __setscheduler(rq, p, attr, pi);
4330 
4331         if (queued) {
4332                 /*
4333                  * We enqueue to tail when the priority of a task is
4334                  * increased (user space view).
4335                  */
4336                 if (oldprio < p->prio)
4337                         queue_flags |= ENQUEUE_HEAD;
4338 
4339                 enqueue_task(rq, p, queue_flags);
4340         }
4341         if (running)
4342                 set_curr_task(rq, p);
4343 
4344         check_class_changed(rq, p, prev_class, oldprio);
4345 
4346         /* Avoid rq from going away on us: */
4347         preempt_disable();
4348         task_rq_unlock(rq, p, &rf);
4349 
4350         if (pi)
4351                 rt_mutex_adjust_pi(p);
4352 
4353         /* Run balance callbacks after we've adjusted the PI chain: */
4354         balance_callback(rq);
4355         preempt_enable();
4356 
4357         return 0;
4358 }
4359 
4360 static int _sched_setscheduler(struct task_struct *p, int policy,
4361                                const struct sched_param *param, bool check)
4362 {
4363         struct sched_attr attr = {
4364                 .sched_policy   = policy,
4365                 .sched_priority = param->sched_priority,
4366                 .sched_nice     = PRIO_TO_NICE(p->static_prio),
4367         };
4368 
4369         /* Fixup the legacy SCHED_RESET_ON_FORK hack. */
4370         if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4371                 attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4372                 policy &= ~SCHED_RESET_ON_FORK;
4373                 attr.sched_policy = policy;
4374         }
4375 
4376         return __sched_setscheduler(p, &attr, check, true);
4377 }
4378 /**
4379  * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4380  * @p: the task in question.
4381  * @policy: new policy.
4382  * @param: structure containing the new RT priority.
4383  *
4384  * Return: 0 on success. An error code otherwise.
4385  *
4386  * NOTE that the task may be already dead.
4387  */
4388 int sched_setscheduler(struct task_struct *p, int policy,
4389                        const struct sched_param *param)
4390 {
4391         return _sched_setscheduler(p, policy, param, true);
4392 }
4393 EXPORT_SYMBOL_GPL(sched_setscheduler);
4394 
4395 int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
4396 {
4397         return __sched_setscheduler(p, attr, true, true);
4398 }
4399 EXPORT_SYMBOL_GPL(sched_setattr);
4400 
4401 int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
4402 {
4403         return __sched_setscheduler(p, attr, false, true);
4404 }
4405 
4406 /**
4407  * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4408  * @p: the task in question.
4409  * @policy: new policy.
4410  * @param: structure containing the new RT priority.
4411  *
4412  * Just like sched_setscheduler, only don't bother checking if the
4413  * current context has permission.  For example, this is needed in
4414  * stop_machine(): we create temporary high priority worker threads,
4415  * but our caller might not have that capability.
4416  *
4417  * Return: 0 on success. An error code otherwise.
4418  */
4419 int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4420                                const struct sched_param *param)
4421 {
4422         return _sched_setscheduler(p, policy, param, false);
4423 }
4424 EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4425 
4426 static int
4427 do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
4428 {
4429         struct sched_param lparam;
4430         struct task_struct *p;
4431         int retval;
4432 
4433         if (!param || pid < 0)
4434                 return -EINVAL;
4435         if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4436                 return -EFAULT;
4437 
4438         rcu_read_lock();
4439         retval = -ESRCH;
4440         p = find_process_by_pid(pid);
4441         if (p != NULL)
4442                 retval = sched_setscheduler(p, policy, &lparam);
4443         rcu_read_unlock();
4444 
4445         return retval;
4446 }
4447 
4448 /*
4449  * Mimics kernel/events/core.c perf_copy_attr().
4450  */
4451 static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
4452 {
4453         u32 size;
4454         int ret;
4455 
4456         if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0))
4457                 return -EFAULT;
4458 
4459         /* Zero the full structure, so that a short copy will be nice: */
4460         memset(attr, 0, sizeof(*attr));
4461 
4462         ret = get_user(size, &uattr->size);
4463         if (ret)
4464                 return ret;
4465 
4466         /* Bail out on silly large: */
4467         if (size > PAGE_SIZE)
4468                 goto err_size;
4469 
4470         /* ABI compatibility quirk: */
4471         if (!size)
4472                 size = SCHED_ATTR_SIZE_VER0;
4473 
4474         if (size < SCHED_ATTR_SIZE_VER0)
4475                 goto err_size;
4476 
4477         /*
4478          * If we're handed a bigger struct than we know of,
4479          * ensure all the unknown bits are 0 - i.e. new
4480          * user-space does not rely on any kernel feature
4481          * extensions we dont know about yet.
4482          */
4483         if (size > sizeof(*attr)) {
4484                 unsigned char __user *addr;
4485                 unsigned char __user *end;
4486                 unsigned char val;
4487 
4488                 addr = (void __user *)uattr + sizeof(*attr);
4489                 end  = (void __user *)uattr + size;
4490 
4491                 for (; addr < end; addr++) {
4492                         ret = get_user(val, addr);
4493                         if (ret)
4494                                 return ret;
4495                         if (val)
4496                                 goto err_size;
4497                 }
4498                 size = sizeof(*attr);
4499         }
4500 
4501         ret = copy_from_user(attr, uattr, size);
4502         if (ret)
4503                 return -EFAULT;
4504 
4505         /*
4506          * XXX: Do we want to be lenient like existing syscalls; or do we want
4507          * to be strict and return an error on out-of-bounds values?
4508          */
4509         attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4510 
4511         return 0;
4512 
4513 err_size:
4514         put_user(sizeof(*attr), &uattr->size);
4515         return -E2BIG;
4516 }
4517 
4518 /**
4519  * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4520  * @pid: the pid in question.
4521  * @policy: new policy.
4522  * @param: structure containing the new RT priority.
4523  *
4524  * Return: 0 on success. An error code otherwise.
4525  */
4526 SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
4527 {
4528         if (policy < 0)
4529                 return -EINVAL;
4530 
4531         return do_sched_setscheduler(pid, policy, param);
4532 }
4533 
4534 /**
4535  * sys_sched_setparam - set/change the RT priority of a thread
4536  * @pid: the pid in question.
4537  * @param: structure containing the new RT priority.
4538  *
4539  * Return: 0 on success. An error code otherwise.
4540  */
4541 SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
4542 {
4543         return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
4544 }
4545 
4546 /**
4547  * sys_sched_setattr - same as above, but with extended sched_attr
4548  * @pid: the pid in question.
4549  * @uattr: structure containing the extended parameters.
4550  * @flags: for future extension.
4551  */
4552 SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
4553                                unsigned int, flags)
4554 {
4555         struct sched_attr attr;
4556         struct task_struct *p;
4557         int retval;
4558 
4559         if (!uattr || pid < 0 || flags)
4560                 return -EINVAL;
4561 
4562         retval = sched_copy_attr(uattr, &attr);
4563         if (retval)
4564                 return retval;
4565 
4566         if ((int)attr.sched_policy < 0)
4567                 return -EINVAL;
4568 
4569         rcu_read_lock();
4570         retval = -ESRCH;
4571         p = find_process_by_pid(pid);
4572         if (p != NULL)
4573                 retval = sched_setattr(p, &attr);
4574         rcu_read_unlock();
4575 
4576         return retval;
4577 }
4578 
4579 /**
4580  * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4581  * @pid: the pid in question.
4582  *
4583  * Return: On success, the policy of the thread. Otherwise, a negative error
4584  * code.
4585  */
4586 SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
4587 {
4588         struct task_struct *p;
4589         int retval;
4590 
4591         if (pid < 0)
4592                 return -EINVAL;
4593 
4594         retval = -ESRCH;
4595         rcu_read_lock();
4596         p = find_process_by_pid(pid);
4597         if (p) {
4598                 retval = security_task_getscheduler(p);
4599                 if (!retval)
4600                         retval = p->policy
4601                                 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
4602         }
4603         rcu_read_unlock();
4604         return retval;
4605 }
4606 
4607 /**
4608  * sys_sched_getparam - get the RT priority of a thread
4609  * @pid: the pid in question.
4610  * @param: structure containing the RT priority.
4611  *
4612  * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
4613  * code.
4614  */
4615 SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
4616 {
4617         struct sched_param lp = { .sched_priority = 0 };
4618         struct task_struct *p;
4619         int retval;
4620 
4621         if (!param || pid < 0)
4622                 return -EINVAL;
4623 
4624         rcu_read_lock();
4625         p = find_process_by_pid(pid);
4626         retval = -ESRCH;
4627         if (!p)
4628                 goto out_unlock;
4629 
4630         retval = security_task_getscheduler(p);
4631         if (retval)
4632                 goto out_unlock;
4633 
4634         if (task_has_rt_policy(p))
4635                 lp.sched_priority = p->rt_priority;
4636         rcu_read_unlock();
4637 
4638         /*
4639          * This one might sleep, we cannot do it with a spinlock held ...
4640          */
4641         retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4642 
4643         return retval;
4644 
4645 out_unlock:
4646         rcu_read_unlock();
4647         return retval;
4648 }
4649 
4650 static int sched_read_attr(struct sched_attr __user *uattr,
4651                            struct sched_attr *attr,
4652                            unsigned int usize)
4653 {
4654         int ret;
4655 
4656         if (!access_ok(VERIFY_WRITE, uattr, usize))
4657                 return -EFAULT;
4658 
4659         /*
4660          * If we're handed a smaller struct than we know of,
4661          * ensure all the unknown bits are 0 - i.e. old
4662          * user-space does not get uncomplete information.
4663          */
4664         if (usize < sizeof(*attr)) {
4665                 unsigned char *addr;
4666                 unsigned char *end;
4667 
4668                 addr = (void *)attr + usize;
4669                 end  = (void *)attr + sizeof(*attr);
4670 
4671                 for (; addr < end; addr++) {
4672                         if (*addr)
4673                                 return -EFBIG;
4674                 }
4675 
4676                 attr->size = usize;
4677         }
4678 
4679         ret = copy_to_user(uattr, attr, attr->size);
4680         if (ret)
4681                 return -EFAULT;
4682 
4683         return 0;
4684 }
4685 
4686 /**
4687  * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4688  * @pid: the pid in question.
4689  * @uattr: structure containing the extended parameters.
4690  * @size: sizeof(attr) for fwd/bwd comp.
4691  * @flags: for future extension.
4692  */
4693 SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
4694                 unsigned int, size, unsigned int, flags)
4695 {
4696         struct sched_attr attr = {
4697                 .size = sizeof(struct sched_attr),
4698         };
4699         struct task_struct *p;
4700         int retval;
4701 
4702         if (!uattr || pid < 0 || size > PAGE_SIZE ||
4703             size < SCHED_ATTR_SIZE_VER0 || flags)
4704                 return -EINVAL;
4705 
4706         rcu_read_lock();
4707         p = find_process_by_pid(pid);
4708         retval = -ESRCH;
4709         if (!p)
4710                 goto out_unlock;
4711 
4712         retval = security_task_getscheduler(p);
4713         if (retval)
4714                 goto out_unlock;
4715 
4716         attr.sched_policy = p->policy;
4717         if (p->sched_reset_on_fork)
4718                 attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4719         if (task_has_dl_policy(p))
4720                 __getparam_dl(p, &attr);
4721         else if (task_has_rt_policy(p))
4722                 attr.sched_priority = p->rt_priority;
4723         else
4724                 attr.sched_nice = task_nice(p);
4725 
4726         rcu_read_unlock();
4727 
4728         retval = sched_read_attr(uattr, &attr, size);
4729         return retval;
4730 
4731 out_unlock:
4732         rcu_read_unlock();
4733         return retval;
4734 }
4735 
4736 long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
4737 {
4738         cpumask_var_t cpus_allowed, new_mask;
4739         struct task_struct *p;
4740         int retval;
4741 
4742         rcu_read_lock();
4743 
4744         p = find_process_by_pid(pid);
4745         if (!p) {
4746                 rcu_read_unlock();
4747                 return -ESRCH;
4748         }
4749 
4750         /* Prevent p going away */
4751         get_task_struct(p);
4752         rcu_read_unlock();
4753 
4754         if (p->flags & PF_NO_SETAFFINITY) {
4755                 retval = -EINVAL;
4756                 goto out_put_task;
4757         }
4758         if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
4759                 retval = -ENOMEM;
4760                 goto out_put_task;
4761         }
4762         if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
4763                 retval = -ENOMEM;
4764                 goto out_free_cpus_allowed;
4765         }
4766         retval = -EPERM;
4767         if (!check_same_owner(p)) {
4768                 rcu_read_lock();
4769                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
4770                         rcu_read_unlock();
4771                         goto out_free_new_mask;
4772                 }
4773                 rcu_read_unlock();
4774         }
4775 
4776         retval = security_task_setscheduler(p);
4777         if (retval)
4778                 goto out_free_new_mask;
4779 
4780 
4781         cpuset_cpus_allowed(p, cpus_allowed);
4782         cpumask_and(new_mask, in_mask, cpus_allowed);
4783 
4784         /*
4785          * Since bandwidth control happens on root_domain basis,
4786          * if admission test is enabled, we only admit -deadline
4787          * tasks allowed to run on all the CPUs in the task's
4788          * root_domain.
4789          */
4790 #ifdef CONFIG_SMP
4791         if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
4792                 rcu_read_lock();
4793                 if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4794                         retval = -EBUSY;
4795                         rcu_read_unlock();
4796                         goto out_free_new_mask;
4797                 }
4798                 rcu_read_unlock();
4799         }
4800 #endif
4801 again:
4802         retval = __set_cpus_allowed_ptr(p, new_mask, true);
4803 
4804         if (!retval) {
4805                 cpuset_cpus_allowed(p, cpus_allowed);
4806                 if (!cpumask_subset(new_mask, cpus_allowed)) {
4807                         /*
4808                          * We must have raced with a concurrent cpuset
4809                          * update. Just reset the cpus_allowed to the
4810                          * cpuset's cpus_allowed
4811                          */
4812                         cpumask_copy(new_mask, cpus_allowed);
4813                         goto again;
4814                 }
4815         }
4816 out_free_new_mask:
4817         free_cpumask_var(new_mask);
4818 out_free_cpus_allowed:
4819         free_cpumask_var(cpus_allowed);
4820 out_put_task:
4821         put_task_struct(p);
4822         return retval;
4823 }
4824 
4825 static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4826                              struct cpumask *new_mask)
4827 {
4828         if (len < cpumask_size())
4829                 cpumask_clear(new_mask);
4830         else if (len > cpumask_size())
4831                 len = cpumask_size();
4832 
4833         return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
4834 }
4835 
4836 /**
4837  * sys_sched_setaffinity - set the CPU affinity of a process
4838  * @pid: pid of the process
4839  * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4840  * @user_mask_ptr: user-space pointer to the new CPU mask
4841  *
4842  * Return: 0 on success. An error code otherwise.
4843  */
4844 SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
4845                 unsigned long __user *, user_mask_ptr)
4846 {
4847         cpumask_var_t new_mask;
4848         int retval;
4849 
4850         if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
4851                 return -ENOMEM;
4852 
4853         retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
4854         if (retval == 0)
4855                 retval = sched_setaffinity(pid, new_mask);
4856         free_cpumask_var(new_mask);
4857         return retval;
4858 }
4859 
4860 long sched_getaffinity(pid_t pid, struct cpumask *mask)
4861 {
4862         struct task_struct *p;
4863         unsigned long flags;
4864         int retval;
4865 
4866         rcu_read_lock();
4867 
4868         retval = -ESRCH;
4869         p = find_process_by_pid(pid);
4870         if (!p)
4871                 goto out_unlock;
4872 
4873         retval = security_task_getscheduler(p);
4874         if (retval)
4875                 goto out_unlock;
4876 
4877         raw_spin_lock_irqsave(&p->pi_lock, flags);
4878         cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4879         raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4880 
4881 out_unlock:
4882         rcu_read_unlock();
4883 
4884         return retval;
4885 }
4886 
4887 /**
4888  * sys_sched_getaffinity - get the CPU affinity of a process
4889  * @pid: pid of the process
4890  * @len: length in bytes of the bitmask pointed to by user_mask_ptr
4891  * @user_mask_ptr: user-space pointer to hold the current CPU mask
4892  *
4893  * Return: size of CPU mask copied to user_mask_ptr on success. An
4894  * error code otherwise.
4895  */
4896 SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
4897                 unsigned long __user *, user_mask_ptr)
4898 {
4899         int ret;
4900         cpumask_var_t mask;
4901 
4902         if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4903                 return -EINVAL;
4904         if (len & (sizeof(unsigned long)-1))
4905                 return -EINVAL;
4906 
4907         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
4908                 return -ENOMEM;
4909 
4910         ret = sched_getaffinity(pid, mask);
4911         if (ret == 0) {
4912                 unsigned int retlen = min(len, cpumask_size());
4913 
4914                 if (copy_to_user(user_mask_ptr, mask, retlen))
4915                         ret = -EFAULT;
4916                 else
4917                         ret = retlen;
4918         }
4919         free_cpumask_var(mask);
4920 
4921         return ret;
4922 }
4923 
4924 /**
4925  * sys_sched_yield - yield the current processor to other threads.
4926  *
4927  * This function yields the current CPU to other tasks. If there are no
4928  * other threads running on this CPU then this function will return.
4929  *
4930  * Return: 0.
4931  */
4932 static void do_sched_yield(void)
4933 {
4934         struct rq_flags rf;
4935         struct rq *rq;
4936 
4937         local_irq_disable();
4938         rq = this_rq();
4939         rq_lock(rq, &rf);
4940 
4941         schedstat_inc(rq->yld_count);
4942         current->sched_class->yield_task(rq);
4943 
4944         /*
4945          * Since we are going to call schedule() anyway, there's
4946          * no need to preempt or enable interrupts:
4947          */
4948         preempt_disable();
4949         rq_unlock(rq, &rf);
4950         sched_preempt_enable_no_resched();
4951 
4952         schedule();
4953 }
4954 
4955 SYSCALL_DEFINE0(sched_yield)
4956 {
4957         do_sched_yield();
4958         return 0;
4959 }
4960 
4961 #ifndef CONFIG_PREEMPT
4962 int __sched _cond_resched(void)
4963 {
4964         if (should_resched(0)) {
4965                 preempt_schedule_common();
4966                 return 1;
4967         }
4968         rcu_all_qs();
4969         return 0;
4970 }
4971 EXPORT_SYMBOL(_cond_resched);
4972 #endif
4973 
4974 /*
4975  * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
4976  * call schedule, and on return reacquire the lock.
4977  *
4978  * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
4979  * operations here to prevent schedule() from being called twice (once via
4980  * spin_unlock(), once by hand).
4981  */
4982 int __cond_resched_lock(spinlock_t *lock)
4983 {
4984         int resched = should_resched(PREEMPT_LOCK_OFFSET);
4985         int ret = 0;
4986 
4987         lockdep_assert_held(lock);
4988 
4989         if (spin_needbreak(lock) || resched) {
4990                 spin_unlock(lock);
4991                 if (resched)
4992                         preempt_schedule_common();
4993                 else
4994                         cpu_relax();
4995                 ret = 1;
4996                 spin_lock(lock);
4997         }
4998         return ret;
4999 }
5000 EXPORT_SYMBOL(__cond_resched_lock);
5001 
5002 /**
5003  * yield - yield the current processor to other threads.
5004  *
5005  * Do not ever use this function, there's a 99% chance you're doing it wrong.
5006  *
5007  * The scheduler is at all times free to pick the calling task as the most
5008  * eligible task to run, if removing the yield() call from your code breaks
5009  * it, its already broken.
5010  *
5011  * Typical broken usage is:
5012  *
5013  * while (!event)
5014  *      yield();
5015  *
5016  * where one assumes that yield() will let 'the other' process run that will
5017  * make event true. If the current task is a SCHED_FIFO task that will never
5018  * happen. Never use yield() as a progress guarantee!!
5019  *
5020  * If you want to use yield() to wait for something, use wait_event().
5021  * If you want to use yield() to be 'nice' for others, use cond_resched().
5022  * If you still want to use yield(), do not!
5023  */
5024 void __sched yield(void)
5025 {
5026         set_current_state(TASK_RUNNING);
5027         do_sched_yield();
5028 }
5029 EXPORT_SYMBOL(yield);
5030 
5031 /**
5032  * yield_to - yield the current processor to another thread in
5033  * your thread group, or accelerate that thread toward the
5034  * processor it's on.
5035  * @p: target task
5036  * @preempt: whether task preemption is allowed or not
5037  *
5038  * It's the caller's job to ensure that the target task struct
5039  * can't go away on us before we can do any checks.
5040  *
5041  * Return:
5042  *      true (>0) if we indeed boosted the target task.
5043  *      false (0) if we failed to boost the target.
5044  *      -ESRCH if there's no task to yield to.
5045  */
5046 int __sched yield_to(struct task_struct *p, bool preempt)
5047 {
5048         struct task_struct *curr = current;
5049         struct rq *rq, *p_rq;
5050         unsigned long flags;
5051         int yielded = 0;
5052 
5053         local_irq_save(flags);
5054         rq = this_rq();
5055 
5056 again:
5057         p_rq = task_rq(p);
5058         /*
5059          * If we're the only runnable task on the rq and target rq also
5060          * has only one task, there's absolutely no point in yielding.
5061          */
5062         if (rq->nr_running == 1 && p_rq->nr_running == 1) {
5063                 yielded = -ESRCH;
5064                 goto out_irq;
5065         }
5066 
5067         double_rq_lock(rq, p_rq);
5068         if (task_rq(p) != p_rq) {
5069                 double_rq_unlock(rq, p_rq);
5070                 goto again;
5071         }
5072 
5073         if (!curr->sched_class->yield_to_task)
5074                 goto out_unlock;
5075 
5076         if (curr->sched_class != p->sched_class)
5077                 goto out_unlock;
5078 
5079         if (task_running(p_rq, p) || p->state)
5080                 goto out_unlock;
5081 
5082         yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5083         if (yielded) {
5084                 schedstat_inc(rq->yld_count);
5085                 /*
5086                  * Make p's CPU reschedule; pick_next_entity takes care of
5087                  * fairness.
5088                  */
5089                 if (preempt && rq != p_rq)
5090                         resched_curr(p_rq);
5091         }
5092 
5093 out_unlock:
5094         double_rq_unlock(rq, p_rq);
5095 out_irq:
5096         local_irq_restore(flags);
5097 
5098         if (yielded > 0)
5099                 schedule();
5100 
5101         return yielded;
5102 }
5103 EXPORT_SYMBOL_GPL(yield_to);
5104 
5105 int io_schedule_prepare(void)
5106 {
5107         int old_iowait = current->in_iowait;
5108 
5109         current->in_iowait = 1;
5110         blk_schedule_flush_plug(current);
5111 
5112         return old_iowait;
5113 }
5114 
5115 void io_schedule_finish(int token)
5116 {
5117         current->in_iowait = token;
5118 }
5119 
5120 /*
5121  * This task is about to go to sleep on IO. Increment rq->nr_iowait so
5122  * that process accounting knows that this is a task in IO wait state.
5123  */
5124 long __sched io_schedule_timeout(long timeout)
5125 {
5126         int token;
5127         long ret;
5128 
5129         token = io_schedule_prepare();
5130         ret = schedule_timeout(timeout);
5131         io_schedule_finish(token);
5132 
5133         return ret;
5134 }
5135 EXPORT_SYMBOL(io_schedule_timeout);
5136 
5137 void io_schedule(void)
5138 {
5139         int token;
5140 
5141         token = io_schedule_prepare();
5142         schedule();
5143         io_schedule_finish(token);
5144 }
5145 EXPORT_SYMBOL(io_schedule);
5146 
5147 /**
5148  * sys_sched_get_priority_max - return maximum RT priority.
5149  * @policy: scheduling class.
5150  *
5151  * Return: On success, this syscall returns the maximum
5152  * rt_priority that can be used by a given scheduling class.
5153  * On failure, a negative error code is returned.
5154  */
5155 SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
5156 {
5157         int ret = -EINVAL;
5158 
5159         switch (policy) {
5160         case SCHED_FIFO:
5161         case SCHED_RR:
5162                 ret = MAX_USER_RT_PRIO-1;
5163                 break;
5164         case SCHED_DEADLINE:
5165         case SCHED_NORMAL:
5166         case SCHED_BATCH:
5167         case SCHED_IDLE:
5168                 ret = 0;
5169                 break;
5170         }
5171         return ret;
5172 }
5173 
5174 /**
5175  * sys_sched_get_priority_min - return minimum RT priority.
5176  * @policy: scheduling class.
5177  *
5178  * Return: On success, this syscall returns the minimum
5179  * rt_priority that can be used by a given scheduling class.
5180  * On failure, a negative error code is returned.
5181  */
5182 SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
5183 {
5184         int ret = -EINVAL;
5185 
5186         switch (policy) {
5187         case SCHED_FIFO:
5188         case SCHED_RR:
5189                 ret = 1;
5190                 break;
5191         case SCHED_DEADLINE:
5192         case SCHED_NORMAL:
5193         case SCHED_BATCH:
5194         case SCHED_IDLE:
5195                 ret = 0;
5196         }
5197         return ret;
5198 }
5199 
5200 static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
5201 {
5202         struct task_struct *p;
5203         unsigned int time_slice;
5204         struct rq_flags rf;
5205         struct rq *rq;
5206         int retval;
5207 
5208         if (pid < 0)
5209                 return -EINVAL;
5210 
5211         retval = -ESRCH;
5212         rcu_read_lock();
5213         p = find_process_by_pid(pid);
5214         if (!p)
5215                 goto out_unlock;
5216 
5217         retval = security_task_getscheduler(p);
5218         if (retval)
5219                 goto out_unlock;
5220 
5221         rq = task_rq_lock(p, &rf);
5222         time_slice = 0;
5223         if (p->sched_class->get_rr_interval)
5224                 time_slice = p->sched_class->get_rr_interval(rq, p);
5225         task_rq_unlock(rq, p, &rf);
5226 
5227         rcu_read_unlock();
5228         jiffies_to_timespec64(time_slice, t);
5229         return 0;
5230 
5231 out_unlock:
5232         rcu_read_unlock();
5233         return retval;
5234 }
5235 
5236 /**
5237  * sys_sched_rr_get_interval - return the default timeslice of a process.
5238  * @pid: pid of the process.
5239  * @interval: userspace pointer to the timeslice value.
5240  *
5241  * this syscall writes the default timeslice value of a given process
5242  * into the user-space timespec buffer. A value of '' means infinity.
5243  *
5244  * Return: On success, 0 and the timeslice is in @interval. Otherwise,
5245  * an error code.
5246  */
5247 SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5248                 struct timespec __user *, interval)
5249 {
5250         struct timespec64 t;
5251         int retval = sched_rr_get_interval(pid, &t);
5252 
5253         if (retval == 0)
5254                 retval = put_timespec64(&t, interval);
5255 
5256         return retval;
5257 }
5258 
5259 #ifdef CONFIG_COMPAT
5260 COMPAT_SYSCALL_DEFINE2(sched_rr_get_interval,
5261                        compat_pid_t, pid,
5262                        struct compat_timespec __user *, interval)
5263 {
5264         struct timespec64 t;
5265         int retval = sched_rr_get_interval(pid, &t);
5266 
5267         if (retval == 0)
5268                 retval = compat_put_timespec64(&t, interval);
5269         return retval;
5270 }
5271 #endif
5272 
5273 void sched_show_task(struct task_struct *p)
5274 {
5275         unsigned long free = 0;
5276         int ppid;
5277 
5278         if (!try_get_task_stack(p))
5279                 return;
5280 
5281         printk(KERN_INFO "%-15.15s %c", p->comm, task_state_to_char(p));
5282 
5283         if (p->state == TASK_RUNNING)
5284                 printk(KERN_CONT "  running task    ");
5285 #ifdef CONFIG_DEBUG_STACK_USAGE
5286         free = stack_not_used(p);
5287 #endif
5288         ppid = 0;
5289         rcu_read_lock();
5290         if (pid_alive(p))
5291                 ppid = task_pid_nr(rcu_dereference(p->real_parent));
5292         rcu_read_unlock();
5293         printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5294                 task_pid_nr(p), ppid,
5295                 (unsigned long)task_thread_info(p)->flags);
5296 
5297         print_worker_info(KERN_INFO, p);
5298         show_stack(p, NULL);
5299         put_task_stack(p);
5300 }
5301 EXPORT_SYMBOL_GPL(sched_show_task);
5302 
5303 static inline bool
5304 state_filter_match(unsigned long state_filter, struct task_struct *p)
5305 {
5306         /* no filter, everything matches */
5307         if (!state_filter)
5308                 return true;
5309 
5310         /* filter, but doesn't match */
5311         if (!(p->state & state_filter))
5312                 return false;
5313 
5314         /*
5315          * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows
5316          * TASK_KILLABLE).
5317          */
5318         if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE)
5319                 return false;
5320 
5321         return true;
5322 }
5323 
5324 
5325 void show_state_filter(unsigned long state_filter)
5326 {
5327         struct task_struct *g, *p;
5328 
5329 #if BITS_PER_LONG == 32
5330         printk(KERN_INFO
5331                 "  task                PC stack   pid father\n");
5332 #else
5333         printk(KERN_INFO
5334                 "  task                        PC stack   pid father\n");
5335 #endif
5336         rcu_read_lock();
5337         for_each_process_thread(g, p) {
5338                 /*
5339                  * reset the NMI-timeout, listing all files on a slow
5340                  * console might take a lot of time:
5341                  * Also, reset softlockup watchdogs on all CPUs, because
5342                  * another CPU might be blocked waiting for us to process
5343                  * an IPI.
5344                  */
5345                 touch_nmi_watchdog();
5346                 touch_all_softlockup_watchdogs();
5347                 if (state_filter_match(state_filter, p))
5348                         sched_show_task(p);
5349         }
5350 
5351 #ifdef CONFIG_SCHED_DEBUG
5352         if (!state_filter)
5353                 sysrq_sched_debug_show();
5354 #endif
5355         rcu_read_unlock();
5356         /*
5357          * Only show locks if all tasks are dumped:
5358          */
5359         if (!state_filter)
5360                 debug_show_all_locks();
5361 }
5362 
5363 /**
5364  * init_idle - set up an idle thread for a given CPU
5365  * @idle: task in question
5366  * @cpu: CPU the idle task belongs to
5367  *
5368  * NOTE: this function does not set the idle thread's NEED_RESCHED
5369  * flag, to make booting more robust.
5370  */
5371 void init_idle(struct task_struct *idle, int cpu)
5372 {
5373         struct rq *rq = cpu_rq(cpu);
5374         unsigned long flags;
5375 
5376         raw_spin_lock_irqsave(&idle->pi_lock, flags);
5377         raw_spin_lock(&rq->lock);
5378 
5379         __sched_fork(0, idle);
5380         idle->state = TASK_RUNNING;
5381         idle->se.exec_start = sched_clock();
5382         idle->flags |= PF_IDLE;
5383 
5384         kasan_unpoison_task_stack(idle);
5385 
5386 #ifdef CONFIG_SMP
5387         /*
5388          * Its possible that init_idle() gets called multiple times on a task,
5389          * in that case do_set_cpus_allowed() will not do the right thing.
5390          *
5391          * And since this is boot we can forgo the serialization.
5392          */
5393         set_cpus_allowed_common(idle, cpumask_of(cpu));
5394 #endif
5395         /*
5396          * We're having a chicken and egg problem, even though we are
5397          * holding rq->lock, the CPU isn't yet set to this CPU so the
5398          * lockdep check in task_group() will fail.
5399          *
5400          * Similar case to sched_fork(). / Alternatively we could
5401          * use task_rq_lock() here and obtain the other rq->lock.
5402          *
5403          * Silence PROVE_RCU
5404          */
5405         rcu_read_lock();
5406         __set_task_cpu(idle, cpu);
5407         rcu_read_unlock();
5408 
5409         rq->curr = rq->idle = idle;
5410         idle->on_rq = TASK_ON_RQ_QUEUED;
5411 #ifdef CONFIG_SMP
5412         idle->on_cpu = 1;
5413 #endif
5414         raw_spin_unlock(&rq->lock);
5415         raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
5416 
5417         /* Set the preempt count _outside_ the spinlocks! */
5418         init_idle_preempt_count(idle, cpu);
5419 
5420         /*
5421          * The idle tasks have their own, simple scheduling class:
5422          */
5423         idle->sched_class = &idle_sched_class;
5424         ftrace_graph_init_idle_task(idle, cpu);
5425         vtime_init_idle(idle, cpu);
5426 #ifdef CONFIG_SMP
5427         sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
5428 #endif
5429 }
5430 
5431 #ifdef CONFIG_SMP
5432 
5433 int cpuset_cpumask_can_shrink(const struct cpumask *cur,
5434                               const struct cpumask *trial)
5435 {
5436         int ret = 1;
5437 
5438         if (!cpumask_weight(cur))
5439                 return ret;
5440 
5441         ret = dl_cpuset_cpumask_can_shrink(cur, trial);
5442 
5443         return ret;
5444 }
5445 
5446 int task_can_attach(struct task_struct *p,
5447                     const struct cpumask *cs_cpus_allowed)
5448 {
5449         int ret = 0;
5450 
5451         /*
5452          * Kthreads which disallow setaffinity shouldn't be moved
5453          * to a new cpuset; we don't want to change their CPU
5454          * affinity and isolating such threads by their set of
5455          * allowed nodes is unnecessary.  Thus, cpusets are not
5456          * applicable for such threads.  This prevents checking for
5457          * success of set_cpus_allowed_ptr() on all attached tasks
5458          * before cpus_allowed may be changed.
5459          */
5460         if (p->flags & PF_NO_SETAFFINITY) {
5461                 ret = -EINVAL;
5462                 goto out;
5463         }
5464 
5465         if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span,
5466                                               cs_cpus_allowed))
5467                 ret = dl_task_can_attach(p, cs_cpus_allowed);
5468 
5469 out:
5470         return ret;
5471 }
5472 
5473 bool sched_smp_initialized __read_mostly;
5474 
5475 #ifdef CONFIG_NUMA_BALANCING
5476 /* Migrate current task p to target_cpu */
5477 int migrate_task_to(struct task_struct *p, int target_cpu)
5478 {
5479         struct migration_arg arg = { p, target_cpu };
5480         int curr_cpu = task_cpu(p);
5481 
5482         if (curr_cpu == target_cpu)
5483                 return 0;
5484 
5485         if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed))
5486                 return -EINVAL;
5487 
5488         /* TODO: This is not properly updating schedstats */
5489 
5490         trace_sched_move_numa(p, curr_cpu, target_cpu);
5491         return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
5492 }
5493 
5494 /*
5495  * Requeue a task on a given node and accurately track the number of NUMA
5496  * tasks on the runqueues
5497  */
5498 void sched_setnuma(struct task_struct *p, int nid)
5499 {
5500         bool queued, running;
5501         struct rq_flags rf;
5502         struct rq *rq;
5503 
5504         rq = task_rq_lock(p, &rf);
5505         queued = task_on_rq_queued(p);
5506         running = task_current(rq, p);
5507 
5508         if (queued)
5509                 dequeue_task(rq, p, DEQUEUE_SAVE);
5510         if (running)
5511                 put_prev_task(rq, p);
5512 
5513         p->numa_preferred_nid = nid;
5514 
5515         if (queued)
5516                 enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
5517         if (running)
5518                 set_curr_task(rq, p);
5519         task_rq_unlock(rq, p, &rf);
5520 }
5521 #endif /* CONFIG_NUMA_BALANCING */
5522 
5523 #ifdef CONFIG_HOTPLUG_CPU
5524 /*
5525  * Ensure that the idle task is using init_mm right before its CPU goes
5526  * offline.
5527  */
5528 void idle_task_exit(void)
5529 {
5530         struct mm_struct *mm = current->active_mm;
5531 
5532         BUG_ON(cpu_online(smp_processor_id()));
5533 
5534         if (mm != &init_mm) {
5535                 switch_mm(mm, &init_mm, current);
5536                 current->active_mm = &init_mm;
5537                 finish_arch_post_lock_switch();
5538         }
5539         mmdrop(mm);
5540 }
5541 
5542 /*
5543  * Since this CPU is going 'away' for a while, fold any nr_active delta
5544  * we might have. Assumes we're called after migrate_tasks() so that the
5545  * nr_active count is stable. We need to take the teardown thread which
5546  * is calling this into account, so we hand in adjust = 1 to the load
5547  * calculation.
5548  *
5549  * Also see the comment "Global load-average calculations".
5550  */
5551 static void calc_load_migrate(struct rq *rq)
5552 {
5553         long delta = calc_load_fold_active(rq, 1);
5554         if (delta)
5555                 atomic_long_add(delta, &calc_load_tasks);
5556 }
5557 
5558 static void put_prev_task_fake(struct rq *rq, struct task_struct *prev)
5559 {
5560 }
5561 
5562 static const struct sched_class fake_sched_class = {
5563         .put_prev_task = put_prev_task_fake,
5564 };
5565 
5566 static struct task_struct fake_task = {
5567         /*
5568          * Avoid pull_{rt,dl}_task()
5569          */
5570         .prio = MAX_PRIO + 1,
5571         .sched_class = &fake_sched_class,
5572 };
5573 
5574 /*
5575  * Migrate all tasks from the rq, sleeping tasks will be migrated by
5576  * try_to_wake_up()->select_task_rq().
5577  *
5578  * Called with rq->lock held even though we'er in stop_machine() and
5579  * there's no concurrency possible, we hold the required locks anyway
5580  * because of lock validation efforts.
5581  */
5582 static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
5583 {
5584         struct rq *rq = dead_rq;
5585         struct task_struct *next, *stop = rq->stop;
5586         struct rq_flags orf = *rf;
5587         int dest_cpu;
5588 
5589         /*
5590          * Fudge the rq selection such that the below task selection loop
5591          * doesn't get stuck on the currently eligible stop task.
5592          *
5593          * We're currently inside stop_machine() and the rq is either stuck
5594          * in the stop_machine_cpu_stop() loop, or we're executing this code,
5595          * either way we should never end up calling schedule() until we're
5596          * done here.
5597          */
5598         rq->stop = NULL;
5599 
5600         /*
5601          * put_prev_task() and pick_next_task() sched
5602          * class method both need to have an up-to-date
5603          * value of rq->clock[_task]
5604          */
5605         update_rq_clock(rq);
5606 
5607         for (;;) {
5608                 /*
5609                  * There's this thread running, bail when that's the only
5610                  * remaining thread:
5611                  */
5612                 if (rq->nr_running == 1)
5613                         break;
5614 
5615                 /*
5616                  * pick_next_task() assumes pinned rq->lock:
5617                  */
5618                 next = pick_next_task(rq, &fake_task, rf);
5619                 BUG_ON(!next);
5620                 put_prev_task(rq, next);
5621 
5622                 /*
5623                  * Rules for changing task_struct::cpus_allowed are holding
5624                  * both pi_lock and rq->lock, such that holding either
5625                  * stabilizes the mask.
5626                  *
5627                  * Drop rq->lock is not quite as disastrous as it usually is
5628                  * because !cpu_active at this point, which means load-balance
5629                  * will not interfere. Also, stop-machine.
5630                  */
5631                 rq_unlock(rq, rf);
5632                 raw_spin_lock(&next->pi_lock);
5633                 rq_relock(rq, rf);
5634 
5635                 /*
5636                  * Since we're inside stop-machine, _nothing_ should have
5637                  * changed the task, WARN if weird stuff happened, because in
5638                  * that case the above rq->lock drop is a fail too.
5639                  */
5640                 if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) {
5641                         raw_spin_unlock(&next->pi_lock);
5642                         continue;
5643                 }
5644 
5645                 /* Find suitable destination for @next, with force if needed. */
5646                 dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5647                 rq = __migrate_task(rq, rf, next, dest_cpu);
5648                 if (rq != dead_rq) {
5649                         rq_unlock(rq, rf);
5650                         rq = dead_rq;
5651                         *rf = orf;
5652                         rq_relock(rq, rf);
5653                 }
5654                 raw_spin_unlock(&next->pi_lock);
5655         }
5656 
5657         rq->stop = stop;
5658 }
5659 #endif /* CONFIG_HOTPLUG_CPU */
5660 
5661 void set_rq_online(struct rq *rq)
5662 {
5663         if (!rq->online) {
5664                 const struct sched_class *class;
5665 
5666                 cpumask_set_cpu(rq->cpu, rq->rd->online);
5667                 rq->online = 1;
5668 
5669                 for_each_class(class) {
5670                         if (class->rq_online)
5671                                 class->rq_online(rq);
5672                 }
5673         }
5674 }
5675 
5676 void set_rq_offline(struct rq *rq)
5677 {
5678         if (rq->online) {
5679                 const struct sched_class *class;
5680 
5681                 for_each_class(class) {
5682                         if (class->rq_offline)
5683                                 class->rq_offline(rq);
5684                 }
5685 
5686                 cpumask_clear_cpu(rq->cpu, rq->rd->online);
5687                 rq->online = 0;
5688         }
5689 }
5690 
5691 /*
5692  * used to mark begin/end of suspend/resume:
5693  */
5694 static int num_cpus_frozen;
5695 
5696 /*
5697  * Update cpusets according to cpu_active mask.  If cpusets are
5698  * disabled, cpuset_update_active_cpus() becomes a simple wrapper
5699  * around partition_sched_domains().
5700  *
5701  * If we come here as part of a suspend/resume, don't touch cpusets because we
5702  * want to restore it back to its original state upon resume anyway.
5703  */
5704 static void cpuset_cpu_active(void)
5705 {
5706         if (cpuhp_tasks_frozen) {
5707                 /*
5708                  * num_cpus_frozen tracks how many CPUs are involved in suspend
5709                  * resume sequence. As long as this is not the last online
5710                  * operation in the resume sequence, just build a single sched
5711                  * domain, ignoring cpusets.
5712                  */
5713                 partition_sched_domains(1, NULL, NULL);
5714                 if (--num_cpus_frozen)
5715                         return;
5716                 /*
5717                  * This is the last CPU online operation. So fall through and
5718                  * restore the original sched domains by considering the
5719                  * cpuset configurations.
5720                  */
5721                 cpuset_force_rebuild();
5722         }
5723         cpuset_update_active_cpus();
5724 }
5725 
5726 static int cpuset_cpu_inactive(unsigned int cpu)
5727 {
5728         if (!cpuhp_tasks_frozen) {
5729                 if (dl_cpu_busy(cpu))
5730                         return -EBUSY;
5731                 cpuset_update_active_cpus();
5732         } else {
5733                 num_cpus_frozen++;
5734                 partition_sched_domains(1, NULL, NULL);
5735         }
5736         return 0;
5737 }
5738 
5739 int sched_cpu_activate(unsigned int cpu)
5740 {
5741         struct rq *rq = cpu_rq(cpu);
5742         struct rq_flags rf;
5743 
5744 #ifdef CONFIG_SCHED_SMT
5745         /*
5746          * The sched_smt_present static key needs to be evaluated on every
5747          * hotplug event because at boot time SMT might be disabled when
5748          * the number of booted CPUs is limited.
5749          *
5750          * If then later a sibling gets hotplugged, then the key would stay
5751          * off and SMT scheduling would never be functional.
5752          */
5753         if (cpumask_weight(cpu_smt_mask(cpu)) > 1)
5754                 static_branch_enable_cpuslocked(&sched_smt_present);
5755 #endif
5756         set_cpu_active(cpu, true);
5757 
5758         if (sched_smp_initialized) {
5759                 sched_domains_numa_masks_set(cpu);
5760                 cpuset_cpu_active();
5761         }
5762 
5763         /*
5764          * Put the rq online, if not already. This happens:
5765          *
5766          * 1) In the early boot process, because we build the real domains
5767          *    after all CPUs have been brought up.
5768          *
5769          * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
5770          *    domains.
5771          */
5772         rq_lock_irqsave(rq, &rf);
5773         if (rq->rd) {
5774                 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5775                 set_rq_online(rq);
5776         }
5777         rq_unlock_irqrestore(rq, &rf);
5778 
5779         update_max_interval();
5780 
5781         return 0;
5782 }
5783 
5784 int sched_cpu_deactivate(unsigned int cpu)
5785 {
5786         int ret;
5787 
5788         set_cpu_active(cpu, false);
5789         /*
5790          * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU
5791          * users of this state to go away such that all new such users will
5792          * observe it.
5793          *
5794          * Do sync before park smpboot threads to take care the rcu boost case.
5795          */
5796         synchronize_rcu_mult(call_rcu, call_rcu_sched);
5797 
5798         if (!sched_smp_initialized)
5799                 return 0;
5800 
5801         ret = cpuset_cpu_inactive(cpu);
5802         if (ret) {
5803                 set_cpu_active(cpu, true);
5804                 return ret;
5805         }
5806         sched_domains_numa_masks_clear(cpu);
5807         return 0;
5808 }
5809 
5810 static void sched_rq_cpu_starting(unsigned int cpu)
5811 {
5812         struct rq *rq = cpu_rq(cpu);
5813 
5814         rq->calc_load_update = calc_load_update;
5815         update_max_interval();
5816 }
5817 
5818 int sched_cpu_starting(unsigned int cpu)
5819 {
5820         sched_rq_cpu_starting(cpu);
5821         sched_tick_start(cpu);
5822         return 0;
5823 }
5824 
5825 #ifdef CONFIG_HOTPLUG_CPU
5826 int sched_cpu_dying(unsigned int cpu)
5827 {
5828         struct rq *rq = cpu_rq(cpu);
5829         struct rq_flags rf;
5830 
5831         /* Handle pending wakeups and then migrate everything off */
5832         sched_ttwu_pending();
5833         sched_tick_stop(cpu);
5834 
5835         rq_lock_irqsave(rq, &rf);
5836         if (rq->rd) {
5837                 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5838                 set_rq_offline(rq);
5839         }
5840         migrate_tasks(rq, &rf);
5841         BUG_ON(rq->nr_running != 1);
5842         rq_unlock_irqrestore(rq, &rf);
5843 
5844         calc_load_migrate(rq);
5845         update_max_interval();
5846         nohz_balance_exit_idle(rq);
5847         hrtick_clear(rq);
5848         return 0;
5849 }
5850 #endif
5851 
5852 void __init sched_init_smp(void)
5853 {
5854         sched_init_numa();
5855 
5856         /*
5857          * There's no userspace yet to cause hotplug operations; hence all the
5858          * CPU masks are stable and all blatant races in the below code cannot
5859          * happen.
5860          */
5861         mutex_lock(&sched_domains_mutex);
5862         sched_init_domains(cpu_active_mask);
5863         mutex_unlock(&sched_domains_mutex);
5864 
5865         /* Move init over to a non-isolated CPU */
5866         if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0)
5867                 BUG();
5868         sched_init_granularity();
5869 
5870         init_sched_rt_class();
5871         init_sched_dl_class();
5872 
5873         sched_smp_initialized = true;
5874 }
5875 
5876 static int __init migration_init(void)
5877 {
5878         sched_rq_cpu_starting(smp_processor_id());
5879         return 0;
5880 }
5881 early_initcall(migration_init);
5882 
5883 #else
5884 void __init sched_init_smp(void)
5885 {
5886         sched_init_granularity();
5887 }
5888 #endif /* CONFIG_SMP */
5889 
5890 int in_sched_functions(unsigned long addr)
5891 {
5892         return in_lock_functions(addr) ||
5893                 (addr >= (unsigned long)__sched_text_start
5894                 && addr < (unsigned long)__sched_text_end);
5895 }
5896 
5897 #ifdef CONFIG_CGROUP_SCHED
5898 /*
5899  * Default task group.
5900  * Every task in system belongs to this group at bootup.
5901  */
5902 struct task_group root_task_group;
5903 LIST_HEAD(task_groups);
5904 
5905 /* Cacheline aligned slab cache for task_group */
5906 static struct kmem_cache *task_group_cache __read_mostly;
5907 #endif
5908 
5909 DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
5910 DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
5911 
5912 void __init sched_init(void)
5913 {
5914         int i, j;
5915         unsigned long alloc_size = 0, ptr;
5916 
5917         wait_bit_init();
5918 
5919 #ifdef CONFIG_FAIR_GROUP_SCHED
5920         alloc_size += 2 * nr_cpu_ids * sizeof(void **);
5921 #endif
5922 #ifdef CONFIG_RT_GROUP_SCHED
5923         alloc_size += 2 * nr_cpu_ids * sizeof(void **);
5924 #endif
5925         if (alloc_size) {
5926                 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
5927 
5928 #ifdef CONFIG_FAIR_GROUP_SCHED
5929                 root_task_group.se = (struct sched_entity **)ptr;
5930                 ptr += nr_cpu_ids * sizeof(void **);
5931 
5932                 root_task_group.cfs_rq = (struct cfs_rq **)ptr;
5933                 ptr += nr_cpu_ids * sizeof(void **);
5934 
5935 #endif /* CONFIG_FAIR_GROUP_SCHED */
5936 #ifdef CONFIG_RT_GROUP_SCHED
5937                 root_task_group.rt_se = (struct sched_rt_entity **)ptr;
5938                 ptr += nr_cpu_ids * sizeof(void **);
5939 
5940                 root_task_group.rt_rq = (struct rt_rq **)ptr;
5941                 ptr += nr_cpu_ids * sizeof(void **);
5942 
5943 #endif /* CONFIG_RT_GROUP_SCHED */
5944         }
5945 #ifdef CONFIG_CPUMASK_OFFSTACK
5946         for_each_possible_cpu(i) {
5947                 per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
5948                         cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5949                 per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
5950                         cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5951         }
5952 #endif /* CONFIG_CPUMASK_OFFSTACK */
5953 
5954         init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
5955         init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime());
5956 
5957 #ifdef CONFIG_SMP
5958         init_defrootdomain();
5959 #endif
5960 
5961 #ifdef CONFIG_RT_GROUP_SCHED
5962         init_rt_bandwidth(&root_task_group.rt_bandwidth,
5963                         global_rt_period(), global_rt_runtime());
5964 #endif /* CONFIG_RT_GROUP_SCHED */
5965 
5966 #ifdef CONFIG_CGROUP_SCHED
5967         task_group_cache = KMEM_CACHE(task_group, 0);
5968 
5969         list_add(&root_task_group.list, &task_groups);
5970         INIT_LIST_HEAD(&root_task_group.children);
5971         INIT_LIST_HEAD(&root_task_group.siblings);
5972         autogroup_init(&init_task);
5973 #endif /* CONFIG_CGROUP_SCHED */
5974 
5975         for_each_possible_cpu(i) {
5976                 struct rq *rq;
5977 
5978                 rq = cpu_rq(i);
5979                 raw_spin_lock_init(&rq->lock);
5980                 rq->nr_running = 0;
5981                 rq->calc_load_active = 0;
5982                 rq->calc_load_update = jiffies + LOAD_FREQ;
5983                 init_cfs_rq(&rq->cfs);
5984                 init_rt_rq(&rq->rt);
5985                 init_dl_rq(&rq->dl);
5986 #ifdef CONFIG_FAIR_GROUP_SCHED
5987                 root_task_group.shares = ROOT_TASK_GROUP_LOAD;
5988                 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
5989                 rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
5990                 /*
5991                  * How much CPU bandwidth does root_task_group get?
5992                  *
5993                  * In case of task-groups formed thr' the cgroup filesystem, it
5994                  * gets 100% of the CPU resources in the system. This overall
5995                  * system CPU resource is divided among the tasks of
5996                  * root_task_group and its child task-groups in a fair manner,
5997                  * based on each entity's (task or task-group's) weight
5998                  * (se->load.weight).
5999                  *
6000                  * In other words, if root_task_group has 10 tasks of weight
6001                  * 1024) and two child groups A0 and A1 (of weight 1024 each),
6002                  * then A0's share of the CPU resource is:
6003                  *
6004                  *      A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
6005                  *
6006                  * We achieve this by letting root_task_group's tasks sit
6007                  * directly in rq->cfs (i.e root_task_group->se[] = NULL).
6008                  */
6009                 init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
6010                 init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
6011 #endif /* CONFIG_FAIR_GROUP_SCHED */
6012 
6013                 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
6014 #ifdef CONFIG_RT_GROUP_SCHED
6015                 init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
6016 #endif
6017 
6018                 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
6019                         rq->cpu_load[j] = 0;
6020 
6021 #ifdef CONFIG_SMP
6022                 rq->sd = NULL;
6023                 rq->rd = NULL;
6024                 rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
6025                 rq->balance_callback = NULL;
6026                 rq->active_balance = 0;
6027                 rq->next_balance = jiffies;
6028                 rq->push_cpu = 0;
6029                 rq->cpu = i;
6030                 rq->online = 0;
6031                 rq->idle_stamp = 0;
6032                 rq->avg_idle = 2*sysctl_sched_migration_cost;
6033                 rq->max_idle_balance_cost = sysctl_sched_migration_cost;
6034 
6035                 INIT_LIST_HEAD(&rq->cfs_tasks);
6036 
6037                 rq_attach_root(rq, &def_root_domain);
6038 #ifdef CONFIG_NO_HZ_COMMON
6039                 rq->last_load_update_tick = jiffies;
6040                 rq->last_blocked_load_update_tick = jiffies;
6041                 atomic_set(&rq->nohz_flags, 0);
6042 #endif
6043 #endif /* CONFIG_SMP */
6044                 hrtick_rq_init(rq);
6045                 atomic_set(&rq->nr_iowait, 0);
6046         }
6047 
6048         set_load_weight(&init_task, false);
6049 
6050         /*
6051          * The boot idle thread does lazy MMU switching as well:
6052          */
6053         mmgrab(&init_mm);
6054         enter_lazy_tlb(&init_mm, current);
6055 
6056         /*
6057          * Make us the idle thread. Technically, schedule() should not be
6058          * called from this thread, however somewhere below it might be,
6059          * but because we are the idle thread, we just pick up running again
6060          * when this runqueue becomes "idle".
6061          */
6062         init_idle(current, smp_processor_id());
6063 
6064         calc_load_update = jiffies + LOAD_FREQ;
6065 
6066 #ifdef CONFIG_SMP
6067         idle_thread_set_boot_cpu();
6068 #endif
6069         init_sched_fair_class();
6070 
6071         init_schedstats();
6072 
6073         scheduler_running = 1;
6074 }
6075 
6076 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6077 static inline int preempt_count_equals(int preempt_offset)
6078 {
6079         int nested = preempt_count() + rcu_preempt_depth();
6080 
6081         return (nested == preempt_offset);
6082 }
6083 
6084 void __might_sleep(const char *file, int line, int preempt_offset)
6085 {
6086         /*
6087          * Blocking primitives will set (and therefore destroy) current->state,
6088          * since we will exit with TASK_RUNNING make sure we enter with it,
6089          * otherwise we will destroy state.
6090          */
6091         WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
6092                         "do not call blocking ops when !TASK_RUNNING; "
6093                         "state=%lx set at [<%p>] %pS\n",
6094                         current->state,
6095                         (void *)current->task_state_change,
6096                         (void *)current->task_state_change);
6097 
6098         ___might_sleep(file, line, preempt_offset);
6099 }
6100 EXPORT_SYMBOL(__might_sleep);
6101 
6102 void ___might_sleep(const char *file, int line, int preempt_offset)
6103 {
6104         /* Ratelimiting timestamp: */
6105         static unsigned long prev_jiffy;
6106 
6107         unsigned long preempt_disable_ip;
6108 
6109         /* WARN_ON_ONCE() by default, no rate limit required: */
6110         rcu_sleep_check();
6111 
6112         if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
6113              !is_idle_task(current)) ||
6114             system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING ||
6115             oops_in_progress)
6116                 return;
6117 
6118         if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
6119                 return;
6120         prev_jiffy = jiffies;
6121 
6122         /* Save this before calling printk(), since that will clobber it: */
6123         preempt_disable_ip = get_preempt_disable_ip(current);
6124 
6125         printk(KERN_ERR
6126                 "BUG: sleeping function called from invalid context at %s:%d\n",
6127                         file, line);
6128         printk(KERN_ERR
6129                 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
6130                         in_atomic(), irqs_disabled(),
6131                         current->pid, current->comm);
6132 
6133         if (task_stack_end_corrupted(current))
6134                 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
6135 
6136         debug_show_held_locks(current);
6137         if (irqs_disabled())
6138                 print_irqtrace_events(current);
6139         if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
6140             && !preempt_count_equals(preempt_offset)) {
6141                 pr_err("Preemption disabled at:");
6142                 print_ip_sym(preempt_disable_ip);
6143                 pr_cont("\n");
6144         }
6145         dump_stack();
6146         add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
6147 }
6148 EXPORT_SYMBOL(___might_sleep);
6149 #endif
6150 
6151 #ifdef CONFIG_MAGIC_SYSRQ
6152 void normalize_rt_tasks(void)
6153 {
6154         struct task_struct *g, *p;
6155         struct sched_attr attr = {
6156                 .sched_policy = SCHED_NORMAL,
6157         };
6158 
6159         read_lock(&tasklist_lock);
6160         for_each_process_thread(g, p) {
6161                 /*
6162                  * Only normalize user tasks:
6163                  */
6164                 if (p->flags & PF_KTHREAD)
6165                         continue;
6166 
6167                 p->se.exec_start = 0;
6168                 schedstat_set(p->se.statistics.wait_start,  0);
6169                 schedstat_set(p->se.statistics.sleep_start, 0);
6170                 schedstat_set(p->se.statistics.block_start, 0);
6171 
6172                 if (!dl_task(p) && !rt_task(p)) {
6173                         /*
6174                          * Renice negative nice level userspace
6175                          * tasks back to 0:
6176                          */
6177                         if (task_nice(p) < 0)
6178                                 set_user_nice(p, 0);
6179                         continue;
6180                 }
6181 
6182                 __sched_setscheduler(p, &attr, false, false);
6183         }
6184         read_unlock(&tasklist_lock);
6185 }
6186 
6187 #endif /* CONFIG_MAGIC_SYSRQ */
6188 
6189 #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6190 /*
6191  * These functions are only useful for the IA64 MCA handling, or kdb.
6192  *
6193  * They can only be called when the whole system has been
6194  * stopped - every CPU needs to be quiescent, and no scheduling
6195  * activity can take place. Using them for anything else would
6196  * be a serious bug, and as a result, they aren't even visible
6197  * under any other configuration.
6198  */
6199 
6200 /**
6201  * curr_task - return the current task for a given CPU.
6202  * @cpu: the processor in question.
6203  *
6204  * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6205  *
6206  * Return: The current task for @cpu.
6207  */
6208 struct task_struct *curr_task(int cpu)
6209 {
6210         return cpu_curr(cpu);
6211 }
6212 
6213 #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
6214 
6215 #ifdef CONFIG_IA64
6216 /**
6217  * set_curr_task - set the current task for a given CPU.
6218  * @cpu: the processor in question.
6219  * @p: the task pointer to set.
6220  *
6221  * Description: This function must only be used when non-maskable interrupts
6222  * are serviced on a separate stack. It allows the architecture to switch the
6223  * notion of the current task on a CPU in a non-blocking manner. This function
6224  * must be called with all CPU's synchronized, and interrupts disabled, the
6225  * and caller must save the original value of the current task (see
6226  * curr_task() above) and restore that value before reenabling interrupts and
6227  * re-starting the system.
6228  *
6229  * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6230  */
6231 void ia64_set_curr_task(int cpu, struct task_struct *p)
6232 {
6233         cpu_curr(cpu) = p;
6234 }
6235 
6236 #endif
6237 
6238 #ifdef CONFIG_CGROUP_SCHED
6239 /* task_group_lock serializes the addition/removal of task groups */
6240 static DEFINE_SPINLOCK(task_group_lock);
6241 
6242 static void sched_free_group(struct task_group *tg)
6243 {
6244         free_fair_sched_group(tg);
6245         free_rt_sched_group(tg);
6246         autogroup_free(tg);
6247         kmem_cache_free(task_group_cache, tg);
6248 }
6249 
6250 /* allocate runqueue etc for a new task group */
6251 struct task_group *sched_create_group(struct task_group *parent)
6252 {
6253         struct task_group *tg;
6254 
6255         tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
6256         if (!tg)
6257                 return ERR_PTR(-ENOMEM);
6258 
6259         if (!alloc_fair_sched_group(tg, parent))
6260                 goto err;
6261 
6262         if (!alloc_rt_sched_group(tg, parent))
6263                 goto err;
6264 
6265         return tg;
6266 
6267 err:
6268         sched_free_group(tg);
6269         return ERR_PTR(-ENOMEM);
6270 }
6271 
6272 void sched_online_group(struct task_group *tg, struct task_group *parent)
6273 {
6274         unsigned long flags;
6275 
6276         spin_lock_irqsave(&task_group_lock, flags);
6277         list_add_rcu(&tg->list, &task_groups);
6278 
6279         /* Root should already exist: */
6280         WARN_ON(!parent);
6281 
6282         tg->parent = parent;
6283         INIT_LIST_HEAD(&tg->children);
6284         list_add_rcu(&tg->siblings, &parent->children);
6285         spin_unlock_irqrestore(&task_group_lock, flags);
6286 
6287         online_fair_sched_group(tg);
6288 }
6289 
6290 /* rcu callback to free various structures associated with a task group */
6291 static void sched_free_group_rcu(struct rcu_head *rhp)
6292 {
6293         /* Now it should be safe to free those cfs_rqs: */
6294         sched_free_group(container_of(rhp, struct task_group, rcu));
6295 }
6296 
6297 void sched_destroy_group(struct task_group *tg)
6298 {
6299         /* Wait for possible concurrent references to cfs_rqs complete: */
6300         call_rcu(&tg->rcu, sched_free_group_rcu);
6301 }
6302 
6303 void sched_offline_group(struct task_group *tg)
6304 {
6305         unsigned long flags;
6306 
6307         /* End participation in shares distribution: */
6308         unregister_fair_sched_group(tg);
6309 
6310         spin_lock_irqsave(&task_group_lock, flags);
6311         list_del_rcu(&tg->list);
6312         list_del_rcu(&tg->siblings);
6313         spin_unlock_irqrestore(&task_group_lock, flags);
6314 }
6315 
6316 static void sched_change_group(struct task_struct *tsk, int type)
6317 {
6318         struct task_group *tg;
6319 
6320         /*
6321          * All callers are synchronized by task_rq_lock(); we do not use RCU
6322          * which is pointless here. Thus, we pass "true" to task_css_check()
6323          * to prevent lockdep warnings.
6324          */
6325         tg = container_of(task_css_check(tsk, cpu_cgrp_id, true),
6326                           struct task_group, css);
6327         tg = autogroup_task_group(tsk, tg);
6328         tsk->sched_task_group = tg;
6329 
6330 #ifdef CONFIG_FAIR_GROUP_SCHED
6331         if (tsk->sched_class->task_change_group)
6332                 tsk->sched_class->task_change_group(tsk, type);
6333         else
6334 #endif
6335                 set_task_rq(tsk, task_cpu(tsk));
6336 }
6337 
6338 /*
6339  * Change task's runqueue when it moves between groups.
6340  *
6341  * The caller of this function should have put the task in its new group by
6342  * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect
6343  * its new group.
6344  */
6345 void sched_move_task(struct task_struct *tsk)
6346 {
6347         int queued, running, queue_flags =
6348                 DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
6349         struct rq_flags rf;
6350         struct rq *rq;
6351 
6352         rq = task_rq_lock(tsk, &rf);
6353         update_rq_clock(rq);
6354 
6355         running = task_current(rq, tsk);
6356         queued = task_on_rq_queued(tsk);
6357 
6358         if (queued)
6359                 dequeue_task(rq, tsk, queue_flags);
6360         if (running)
6361                 put_prev_task(rq, tsk);
6362 
6363         sched_change_group(tsk, TASK_MOVE_GROUP);
6364 
6365         if (queued)
6366                 enqueue_task(rq, tsk, queue_flags);
6367         if (running)
6368                 set_curr_task(rq, tsk);
6369 
6370         task_rq_unlock(rq, tsk, &rf);
6371 }
6372 
6373 static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
6374 {
6375         return css ? container_of(css, struct task_group, css) : NULL;
6376 }
6377 
6378 static struct cgroup_subsys_state *
6379 cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
6380 {
6381         struct task_group *parent = css_tg(parent_css);
6382         struct task_group *tg;
6383 
6384         if (!parent) {
6385                 /* This is early initialization for the top cgroup */
6386                 return &root_task_group.css;
6387         }
6388 
6389         tg = sched_create_group(parent);
6390         if (IS_ERR(tg))
6391                 return ERR_PTR(-ENOMEM);
6392 
6393         return &tg->css;
6394 }
6395 
6396 /* Expose task group only after completing cgroup initialization */
6397 static int cpu_cgroup_css_online(struct cgroup_subsys_state *css)
6398 {
6399         struct task_group *tg = css_tg(css);
6400         struct task_group *parent = css_tg(css->parent);
6401 
6402         if (parent)
6403                 sched_online_group(tg, parent);
6404         return 0;
6405 }
6406 
6407 static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
6408 {
6409         struct task_group *tg = css_tg(css);
6410 
6411         sched_offline_group(tg);
6412 }
6413 
6414 static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
6415 {
6416         struct task_group *tg = css_tg(css);
6417 
6418         /*
6419          * Relies on the RCU grace period between css_released() and this.
6420          */
6421         sched_free_group(tg);
6422 }
6423 
6424 /*
6425  * This is called before wake_up_new_task(), therefore we really only
6426  * have to set its group bits, all the other stuff does not apply.
6427  */
6428 static void cpu_cgroup_fork(struct task_struct *task)
6429 {
6430         struct rq_flags rf;
6431         struct rq *rq;
6432 
6433         rq = task_rq_lock(task, &rf);
6434 
6435         update_rq_clock(rq);
6436         sched_change_group(task, TASK_SET_GROUP);
6437 
6438         task_rq_unlock(rq, task, &rf);
6439 }
6440 
6441 static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
6442 {
6443         struct task_struct *task;
6444         struct cgroup_subsys_state *css;
6445         int ret = 0;
6446 
6447         cgroup_taskset_for_each(task, css, tset) {
6448 #ifdef CONFIG_RT_GROUP_SCHED
6449                 if (!sched_rt_can_attach(css_tg(css), task))
6450                         return -EINVAL;
6451 #else
6452                 /* We don't support RT-tasks being in separate groups */
6453                 if (task->sched_class != &fair_sched_class)
6454                         return -EINVAL;
6455 #endif
6456                 /*
6457                  * Serialize against wake_up_new_task() such that if its
6458                  * running, we're sure to observe its full state.
6459                  */
6460                 raw_spin_lock_irq(&task->pi_lock);
6461                 /*
6462                  * Avoid calling sched_move_task() before wake_up_new_task()
6463                  * has happened. This would lead to problems with PELT, due to
6464                  * move wanting to detach+attach while we're not attached yet.
6465                  */
6466                 if (task->state == TASK_NEW)
6467                         ret = -EINVAL;
6468                 raw_spin_unlock_irq(&task->pi_lock);
6469 
6470                 if (ret)
6471                         break;
6472         }
6473         return ret;
6474 }
6475 
6476 static void cpu_cgroup_attach(struct cgroup_taskset *tset)
6477 {
6478         struct task_struct *task;
6479         struct cgroup_subsys_state *css;
6480 
6481         cgroup_taskset_for_each(task, css, tset)
6482                 sched_move_task(task);
6483 }
6484 
6485 #ifdef CONFIG_FAIR_GROUP_SCHED
6486 static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
6487                                 struct cftype *cftype, u64 shareval)
6488 {
6489         return sched_group_set_shares(css_tg(css), scale_load(shareval));
6490 }
6491 
6492 static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
6493                                struct cftype *cft)
6494 {
6495         struct task_group *tg = css_tg(css);
6496 
6497         return (u64) scale_load_down(tg->shares);
6498 }
6499 
6500 #ifdef CONFIG_CFS_BANDWIDTH
6501 static DEFINE_MUTEX(cfs_constraints_mutex);
6502 
6503 const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
6504 const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */
6505 
6506 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);
6507 
6508 static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
6509 {
6510         int i, ret = 0, runtime_enabled, runtime_was_enabled;
6511         struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6512 
6513         if (tg == &root_task_group)
6514                 return -EINVAL;
6515 
6516         /*
6517          * Ensure we have at some amount of bandwidth every period.  This is
6518          * to prevent reaching a state of large arrears when throttled via
6519          * entity_tick() resulting in prolonged exit starvation.
6520          */
6521         if (quota < min_cfs_quota_period || period < min_cfs_quota_period)
6522                 return -EINVAL;
6523 
6524         /*
6525          * Likewise, bound things on the otherside by preventing insane quota
6526          * periods.  This also allows us to normalize in computing quota
6527          * feasibility.
6528          */
6529         if (period > max_cfs_quota_period)
6530                 return -EINVAL;
6531 
6532         /*
6533          * Prevent race between setting of cfs_rq->runtime_enabled and
6534          * unthrottle_offline_cfs_rqs().
6535          */
6536         get_online_cpus();
6537         mutex_lock(&cfs_constraints_mutex);
6538         ret = __cfs_schedulable(tg, period, quota);
6539         if (ret)
6540                 goto out_unlock;
6541 
6542         runtime_enabled = quota != RUNTIME_INF;
6543         runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
6544         /*
6545          * If we need to toggle cfs_bandwidth_used, off->on must occur
6546          * before making related changes, and on->off must occur afterwards
6547          */
6548         if (runtime_enabled && !runtime_was_enabled)
6549                 cfs_bandwidth_usage_inc();
6550         raw_spin_lock_irq(&cfs_b->lock);
6551         cfs_b->period = ns_to_ktime(period);
6552         cfs_b->quota = quota;
6553 
6554         __refill_cfs_bandwidth_runtime(cfs_b);
6555 
6556         /* Restart the period timer (if active) to handle new period expiry: */
6557         if (runtime_enabled)
6558                 start_cfs_bandwidth(cfs_b);
6559 
6560         raw_spin_unlock_irq(&cfs_b->lock);
6561 
6562         for_each_online_cpu(i) {
6563                 struct cfs_rq *cfs_rq = tg->cfs_rq[i];
6564                 struct rq *rq = cfs_rq->rq;
6565                 struct rq_flags rf;
6566 
6567                 rq_lock_irq(rq, &rf);
6568                 cfs_rq->runtime_enabled = runtime_enabled;
6569                 cfs_rq->runtime_remaining = 0;
6570 
6571                 if (cfs_rq->throttled)
6572                         unthrottle_cfs_rq(cfs_rq);
6573                 rq_unlock_irq(rq, &rf);
6574         }
6575         if (runtime_was_enabled && !runtime_enabled)
6576                 cfs_bandwidth_usage_dec();
6577 out_unlock:
6578         mutex_unlock(&cfs_constraints_mutex);
6579         put_online_cpus();
6580 
6581         return ret;
6582 }
6583 
6584 int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
6585 {
6586         u64 quota, period;
6587 
6588         period = ktime_to_ns(tg->cfs_bandwidth.period);
6589         if (cfs_quota_us < 0)
6590                 quota = RUNTIME_INF;
6591         else
6592                 quota = (u64)cfs_quota_us * NSEC_PER_USEC;
6593 
6594         return tg_set_cfs_bandwidth(tg, period, quota);
6595 }
6596 
6597 long tg_get_cfs_quota(struct task_group *tg)
6598 {
6599         u64 quota_us;
6600 
6601         if (tg->cfs_bandwidth.quota == RUNTIME_INF)
6602                 return -1;
6603 
6604         quota_us = tg->cfs_bandwidth.quota;
6605         do_div(quota_us, NSEC_PER_USEC);
6606 
6607         return quota_us;
6608 }
6609 
6610 int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
6611 {
6612         u64 quota, period;
6613 
6614         period = (u64)cfs_period_us * NSEC_PER_USEC;
6615         quota = tg->cfs_bandwidth.quota;
6616 
6617         return tg_set_cfs_bandwidth(tg, period, quota);
6618 }
6619 
6620 long tg_get_cfs_period(struct task_group *tg)
6621 {
6622         u64 cfs_period_us;
6623 
6624         cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
6625         do_div(cfs_period_us, NSEC_PER_USEC);
6626 
6627         return cfs_period_us;
6628 }
6629 
6630 static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
6631                                   struct cftype *cft)
6632 {
6633         return tg_get_cfs_quota(css_tg(css));
6634 }
6635 
6636 static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
6637                                    struct cftype *cftype, s64 cfs_quota_us)
6638 {
6639         return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
6640 }
6641 
6642 static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
6643                                    struct cftype *cft)
6644 {
6645         return tg_get_cfs_period(css_tg(css));
6646 }
6647 
6648 static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
6649                                     struct cftype *cftype, u64 cfs_period_us)
6650 {
6651         return tg_set_cfs_period(css_tg(css), cfs_period_us);
6652 }
6653 
6654 struct cfs_schedulable_data {
6655         struct task_group *tg;
6656         u64 period, quota;
6657 };
6658 
6659 /*
6660  * normalize group quota/period to be quota/max_period
6661  * note: units are usecs
6662  */
6663 static u64 normalize_cfs_quota(struct task_group *tg,
6664                                struct cfs_schedulable_data *d)
6665 {
6666         u64 quota, period;
6667 
6668         if (tg == d->tg) {
6669                 period = d->period;
6670                 quota = d->quota;
6671         } else {
6672                 period = tg_get_cfs_period(tg);
6673                 quota = tg_get_cfs_quota(tg);
6674         }
6675 
6676         /* note: these should typically be equivalent */
6677         if (quota == RUNTIME_INF || quota == -1)
6678                 return RUNTIME_INF;
6679 
6680         return to_ratio(period, quota);
6681 }
6682 
6683 static int tg_cfs_schedulable_down(struct task_group *tg, void *data)
6684 {
6685         struct cfs_schedulable_data *d = data;
6686         struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6687         s64 quota = 0, parent_quota = -1;
6688 
6689         if (!tg->parent) {
6690                 quota = RUNTIME_INF;
6691         } else {
6692                 struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
6693 
6694                 quota = normalize_cfs_quota(tg, d);
6695                 parent_quota = parent_b->hierarchical_quota;
6696 
6697                 /*
6698                  * Ensure max(child_quota) <= parent_quota.  On cgroup2,
6699                  * always take the min.  On cgroup1, only inherit when no
6700                  * limit is set:
6701                  */
6702                 if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) {
6703                         quota = min(quota, parent_quota);
6704                 } else {
6705                         if (quota == RUNTIME_INF)
6706                                 quota = parent_quota;
6707                         else if (parent_quota != RUNTIME_INF && quota > parent_quota)
6708                                 return -EINVAL;
6709                 }
6710         }
6711         cfs_b->hierarchical_quota = quota;
6712 
6713         return 0;
6714 }
6715 
6716 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
6717 {
6718         int ret;
6719         struct cfs_schedulable_data data = {
6720                 .tg = tg,
6721                 .period = period,
6722                 .quota = quota,
6723         };
6724 
6725         if (quota != RUNTIME_INF) {
6726                 do_div(data.period, NSEC_PER_USEC);
6727                 do_div(data.quota, NSEC_PER_USEC);
6728         }
6729 
6730         rcu_read_lock();
6731         ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
6732         rcu_read_unlock();
6733 
6734         return ret;
6735 }
6736 
6737 static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
6738 {
6739         struct task_group *tg = css_tg(seq_css(sf));
6740         struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6741 
6742         seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods);
6743         seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled);
6744         seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time);
6745 
6746         if (schedstat_enabled() && tg != &root_task_group) {
6747                 u64 ws = 0;
6748                 int i;
6749 
6750                 for_each_possible_cpu(i)
6751                         ws += schedstat_val(tg->se[i]->statistics.wait_sum);
6752 
6753                 seq_printf(sf, "wait_sum %llu\n", ws);
6754         }
6755 
6756         return 0;
6757 }
6758 #endif /* CONFIG_CFS_BANDWIDTH */
6759 #endif /* CONFIG_FAIR_GROUP_SCHED */
6760 
6761 #ifdef CONFIG_RT_GROUP_SCHED
6762 static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
6763                                 struct cftype *cft, s64 val)
6764 {
6765         return sched_group_set_rt_runtime(css_tg(css), val);
6766 }
6767 
6768 static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
6769                                struct cftype *cft)
6770 {
6771         return sched_group_rt_runtime(css_tg(css));
6772 }
6773 
6774 static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
6775                                     struct cftype *cftype, u64 rt_period_us)
6776 {
6777         return sched_group_set_rt_period(css_tg(css), rt_period_us);
6778 }
6779 
6780 static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
6781                                    struct cftype *cft)
6782 {
6783         return sched_group_rt_period(css_tg(css));
6784 }
6785 #endif /* CONFIG_RT_GROUP_SCHED */
6786 
6787 static struct cftype cpu_legacy_files[] = {
6788 #ifdef CONFIG_FAIR_GROUP_SCHED
6789         {
6790                 .name = "shares",
6791                 .read_u64 = cpu_shares_read_u64,
6792                 .write_u64 = cpu_shares_write_u64,
6793         },
6794 #endif
6795 #ifdef CONFIG_CFS_BANDWIDTH
6796         {
6797                 .name = "cfs_quota_us",
6798                 .read_s64 = cpu_cfs_quota_read_s64,
6799                 .write_s64 = cpu_cfs_quota_write_s64,
6800         },
6801         {
6802                 .name = "cfs_period_us",
6803                 .read_u64 = cpu_cfs_period_read_u64,
6804                 .write_u64 = cpu_cfs_period_write_u64,
6805         },
6806         {
6807                 .name = "stat",
6808                 .seq_show = cpu_cfs_stat_show,
6809         },
6810 #endif
6811 #ifdef CONFIG_RT_GROUP_SCHED
6812         {
6813                 .name = "rt_runtime_us",
6814                 .read_s64 = cpu_rt_runtime_read,
6815                 .write_s64 = cpu_rt_runtime_write,
6816         },
6817         {
6818                 .name = "rt_period_us",
6819                 .read_u64 = cpu_rt_period_read_uint,
6820                 .write_u64 = cpu_rt_period_write_uint,
6821         },
6822 #endif
6823         { }     /* Terminate */
6824 };
6825 
6826 static int cpu_extra_stat_show(struct seq_file *sf,
6827                                struct cgroup_subsys_state *css)
6828 {
6829 #ifdef CONFIG_CFS_BANDWIDTH
6830         {
6831                 struct task_group *tg = css_tg(css);
6832                 struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6833                 u64 throttled_usec;
6834 
6835                 throttled_usec = cfs_b->throttled_time;
6836                 do_div(throttled_usec, NSEC_PER_USEC);
6837 
6838                 seq_printf(sf, "nr_periods %d\n"
6839                            "nr_throttled %d\n"
6840                            "throttled_usec %llu\n",
6841                            cfs_b->nr_periods, cfs_b->nr_throttled,
6842                            throttled_usec);
6843         }
6844 #endif
6845         return 0;
6846 }
6847 
6848 #ifdef CONFIG_FAIR_GROUP_SCHED
6849 static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css,
6850                                struct cftype *cft)
6851 {
6852         struct task_group *tg = css_tg(css);
6853         u64 weight = scale_load_down(tg->shares);
6854 
6855         return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024);
6856 }
6857 
6858 static int cpu_weight_write_u64(struct cgroup_subsys_state *css,
6859                                 struct cftype *cft, u64 weight)
6860 {
6861         /*
6862          * cgroup weight knobs should use the common MIN, DFL and MAX
6863          * values which are 1, 100 and 10000 respectively.  While it loses
6864          * a bit of range on both ends, it maps pretty well onto the shares
6865          * value used by scheduler and the round-trip conversions preserve
6866          * the original value over the entire range.
6867          */
6868         if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX)
6869                 return -ERANGE;
6870 
6871         weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL);
6872 
6873         return sched_group_set_shares(css_tg(css), scale_load(weight));
6874 }
6875 
6876 static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css,
6877                                     struct cftype *cft)
6878 {
6879         unsigned long weight = scale_load_down(css_tg(css)->shares);
6880         int last_delta = INT_MAX;
6881         int prio, delta;
6882 
6883         /* find the closest nice value to the current weight */
6884         for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) {
6885                 delta = abs(sched_prio_to_weight[prio] - weight);
6886                 if (delta >= last_delta)
6887                         break;
6888                 last_delta = delta;
6889         }
6890 
6891         return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO);
6892 }
6893 
6894 static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css,
6895                                      struct cftype *cft, s64 nice)
6896 {
6897         unsigned long weight;
6898         int idx;
6899 
6900         if (nice < MIN_NICE || nice > MAX_NICE)
6901                 return -ERANGE;
6902 
6903         idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO;
6904         idx = array_index_nospec(idx, 40);
6905         weight = sched_prio_to_weight[idx];
6906 
6907         return sched_group_set_shares(css_tg(css), scale_load(weight));
6908 }
6909 #endif
6910 
6911 static void __maybe_unused cpu_period_quota_print(struct seq_file *sf,
6912                                                   long period, long quota)
6913 {
6914         if (quota < 0)
6915                 seq_puts(sf, "max");
6916         else
6917                 seq_printf(sf, "%ld", quota);
6918 
6919         seq_printf(sf, " %ld\n", period);
6920 }
6921 
6922 /* caller should put the current value in *@periodp before calling */
6923 static int __maybe_unused cpu_period_quota_parse(char *buf,
6924                                                  u64 *periodp, u64 *quotap)
6925 {
6926         char tok[21];   /* U64_MAX */
6927 
6928         if (!sscanf(buf, "%s %llu", tok, periodp))
6929                 return -EINVAL;
6930 
6931         *periodp *= NSEC_PER_USEC;
6932 
6933         if (sscanf(tok, "%llu", quotap))
6934                 *quotap *= NSEC_PER_USEC;
6935         else if (!strcmp(tok, "max"))
6936                 *quotap = RUNTIME_INF;
6937         else
6938                 return -EINVAL;
6939 
6940         return 0;
6941 }
6942 
6943 #ifdef CONFIG_CFS_BANDWIDTH
6944 static int cpu_max_show(struct seq_file *sf, void *v)
6945 {
6946         struct task_group *tg = css_tg(seq_css(sf));
6947 
6948         cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg));
6949         return 0;
6950 }
6951 
6952 static ssize_t cpu_max_write(struct kernfs_open_file *of,
6953                              char *buf, size_t nbytes, loff_t off)
6954 {
6955         struct task_group *tg = css_tg(of_css(of));
6956         u64 period = tg_get_cfs_period(tg);
6957         u64 quota;
6958         int ret;
6959 
6960         ret = cpu_period_quota_parse(buf, &period, &quota);
6961         if (!ret)
6962                 ret = tg_set_cfs_bandwidth(tg, period, quota);
6963         return ret ?: nbytes;
6964 }
6965 #endif
6966 
6967 static struct cftype cpu_files[] = {
6968 #ifdef CONFIG_FAIR_GROUP_SCHED
6969         {
6970                 .name = "weight",
6971                 .flags = CFTYPE_NOT_ON_ROOT,
6972                 .read_u64 = cpu_weight_read_u64,
6973                 .write_u64 = cpu_weight_write_u64,
6974         },
6975         {
6976                 .name = "weight.nice",
6977                 .flags = CFTYPE_NOT_ON_ROOT,
6978                 .read_s64 = cpu_weight_nice_read_s64,
6979                 .write_s64 = cpu_weight_nice_write_s64,
6980         },
6981 #endif
6982 #ifdef CONFIG_CFS_BANDWIDTH
6983         {
6984                 .name = "max",
6985                 .flags = CFTYPE_NOT_ON_ROOT,
6986                 .seq_show = cpu_max_show,
6987                 .write = cpu_max_write,
6988         },
6989 #endif
6990         { }     /* terminate */
6991 };
6992 
6993 struct cgroup_subsys cpu_cgrp_subsys = {
6994         .css_alloc      = cpu_cgroup_css_alloc,
6995         .css_online     = cpu_cgroup_css_online,
6996         .css_released   = cpu_cgroup_css_released,
6997         .css_free       = cpu_cgroup_css_free,
6998         .css_extra_stat_show = cpu_extra_stat_show,
6999         .fork           = cpu_cgroup_fork,
7000         .can_attach     = cpu_cgroup_can_attach,
7001         .attach         = cpu_cgroup_attach,
7002         .legacy_cftypes = cpu_legacy_files,
7003         .dfl_cftypes    = cpu_files,
7004         .early_init     = true,
7005         .threaded       = true,
7006 };
7007 
7008 #endif  /* CONFIG_CGROUP_SCHED */
7009 
7010 void dump_cpu_task(int cpu)
7011 {
7012         pr_info("Task dump for CPU %d:\n", cpu);
7013         sched_show_task(cpu_curr(cpu));
7014 }
7015 
7016 /*
7017  * Nice levels are multiplicative, with a gentle 10% change for every
7018  * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
7019  * nice 1, it will get ~10% less CPU time than another CPU-bound task
7020  * that remained on nice 0.
7021  *
7022  * The "10% effect" is relative and cumulative: from _any_ nice level,
7023  * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
7024  * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
7025  * If a task goes up by ~10% and another task goes down by ~10% then
7026  * the relative distance between them is ~25%.)
7027  */
7028 const int sched_prio_to_weight[40] = {
7029  /* -20 */     88761,     71755,     56483,     46273,     36291,
7030  /* -15 */     29154,     23254,     18705,     14949,     11916,
7031  /* -10 */      9548,      7620,      6100,      4904,      3906,
7032  /*  -5 */      3121,      2501,      1991,      1586,      1277,
7033  /*   0 */      1024,       820,       655,       526,       423,
7034  /*   5 */       335,       272,       215,       172,       137,
7035  /*  10 */       110,        87,        70,        56,        45,
7036  /*  15 */        36,        29,        23,        18,        15,
7037 };
7038 
7039 /*
7040  * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated.
7041  *
7042  * In cases where the weight does not change often, we can use the
7043  * precalculated inverse to speed up arithmetics by turning divisions
7044  * into multiplications:
7045  */
7046 const u32 sched_prio_to_wmult[40] = {
7047  /* -20 */     48388,     59856,     76040,     92818,    118348,
7048  /* -15 */    147320,    184698,    229616,    287308,    360437,
7049  /* -10 */    449829,    563644,    704093,    875809,   1099582,
7050  /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
7051  /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
7052  /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
7053  /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
7054  /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
7055 };
7056 
7057 #undef CREATE_TRACE_POINTS
7058 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp