1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Generic entry points for the idle threads and
4 * implementation of the idle task scheduling class.
5 *
6 * (NOTE: these are not related to SCHED_IDLE batch scheduled
7 * tasks which are handled in sched/fair.c )
8 */
9#include <linux/cpuidle.h>
10#include <linux/suspend.h>
11#include <linux/livepatch.h>
12#include "sched.h"
13#include "smp.h"
14
15/* Linker adds these: start and end of __cpuidle functions */
16extern char __cpuidle_text_start[], __cpuidle_text_end[];
17
18/**
19 * sched_idle_set_state - Record idle state for the current CPU.
20 * @idle_state: State to record.
21 */
22void sched_idle_set_state(struct cpuidle_state *idle_state)
23{
24 idle_set_state(this_rq(), idle_state);
25}
26
27static int __read_mostly cpu_idle_force_poll;
28
29void cpu_idle_poll_ctrl(bool enable)
30{
31 if (enable) {
32 cpu_idle_force_poll++;
33 } else {
34 cpu_idle_force_poll--;
35 WARN_ON_ONCE(cpu_idle_force_poll < 0);
36 }
37}
38
39#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
40static int __init cpu_idle_poll_setup(char *__unused)
41{
42 cpu_idle_force_poll = 1;
43
44 return 1;
45}
46__setup("nohlt", cpu_idle_poll_setup);
47
48static int __init cpu_idle_nopoll_setup(char *__unused)
49{
50 cpu_idle_force_poll = 0;
51
52 return 1;
53}
54__setup("hlt", cpu_idle_nopoll_setup);
55#endif /* CONFIG_GENERIC_IDLE_POLL_SETUP */
56
57static noinline int __cpuidle cpu_idle_poll(void)
58{
59 instrumentation_begin();
60 trace_cpu_idle(state: 0, smp_processor_id());
61 stop_critical_timings();
62 ct_cpuidle_enter();
63
64 raw_local_irq_enable();
65 while (!tif_need_resched() &&
66 (cpu_idle_force_poll || tick_check_broadcast_expired()))
67 cpu_relax();
68 raw_local_irq_disable();
69
70 ct_cpuidle_exit();
71 start_critical_timings();
72 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
73 local_irq_enable();
74 instrumentation_end();
75
76 return 1;
77}
78
79/* Weak implementations for optional arch specific functions */
80void __weak arch_cpu_idle_prepare(void) { }
81void __weak arch_cpu_idle_enter(void) { }
82void __weak arch_cpu_idle_exit(void) { }
83void __weak __noreturn arch_cpu_idle_dead(void) { while (1); }
84void __weak arch_cpu_idle(void)
85{
86 cpu_idle_force_poll = 1;
87}
88
89#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE
90DEFINE_STATIC_KEY_FALSE(arch_needs_tick_broadcast);
91
92static inline void cond_tick_broadcast_enter(void)
93{
94 if (static_branch_unlikely(&arch_needs_tick_broadcast))
95 tick_broadcast_enter();
96}
97
98static inline void cond_tick_broadcast_exit(void)
99{
100 if (static_branch_unlikely(&arch_needs_tick_broadcast))
101 tick_broadcast_exit();
102}
103#else /* !CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE: */
104static inline void cond_tick_broadcast_enter(void) { }
105static inline void cond_tick_broadcast_exit(void) { }
106#endif /* !CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE */
107
108/**
109 * default_idle_call - Default CPU idle routine.
110 *
111 * To use when the cpuidle framework cannot be used.
112 */
113void __cpuidle default_idle_call(void)
114{
115 instrumentation_begin();
116 if (!current_clr_polling_and_test()) {
117 cond_tick_broadcast_enter();
118 trace_cpu_idle(state: 1, smp_processor_id());
119 stop_critical_timings();
120
121 ct_cpuidle_enter();
122 arch_cpu_idle();
123 ct_cpuidle_exit();
124
125 start_critical_timings();
126 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
127 cond_tick_broadcast_exit();
128 }
129 local_irq_enable();
130 instrumentation_end();
131}
132
133static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
134 struct cpuidle_device *dev,
135 u64 max_latency_ns)
136{
137 if (current_clr_polling_and_test())
138 return -EBUSY;
139
140 return cpuidle_enter_s2idle(drv, dev, latency_limit_ns: max_latency_ns);
141}
142
143static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
144 int next_state)
145{
146 /*
147 * The idle task must be scheduled, it is pointless to go to idle, just
148 * update no idle residency and return.
149 */
150 if (current_clr_polling_and_test()) {
151 dev->last_residency_ns = 0;
152 local_irq_enable();
153 return -EBUSY;
154 }
155
156 /*
157 * Enter the idle state previously returned by the governor decision.
158 * This function will block until an interrupt occurs and will take
159 * care of re-enabling the local interrupts
160 */
161 return cpuidle_enter(drv, dev, index: next_state);
162}
163
164/**
165 * cpuidle_idle_call - the main idle function
166 *
167 * NOTE: no locks or semaphores should be used here
168 *
169 * On architectures that support TIF_POLLING_NRFLAG, is called with polling
170 * set, and it returns with polling set. If it ever stops polling, it
171 * must clear the polling bit.
172 */
173static void cpuidle_idle_call(void)
174{
175 struct cpuidle_device *dev = cpuidle_get_device();
176 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
177 int next_state, entered_state;
178
179 /*
180 * Check if the idle task must be rescheduled. If it is the
181 * case, exit the function after re-enabling the local IRQ.
182 */
183 if (need_resched()) {
184 local_irq_enable();
185 return;
186 }
187
188 if (cpuidle_not_available(drv, dev)) {
189 tick_nohz_idle_stop_tick();
190
191 default_idle_call();
192 goto exit_idle;
193 }
194
195 /*
196 * Suspend-to-idle ("s2idle") is a system state in which all user space
197 * has been frozen, all I/O devices have been suspended and the only
198 * activity happens here and in interrupts (if any). In that case bypass
199 * the cpuidle governor and go straight for the deepest idle state
200 * available. Possibly also suspend the local tick and the entire
201 * timekeeping to prevent timer interrupts from kicking us out of idle
202 * until a proper wakeup interrupt happens.
203 */
204
205 if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
206 u64 max_latency_ns;
207
208 if (idle_should_enter_s2idle()) {
209 max_latency_ns = cpu_wakeup_latency_qos_limit() *
210 NSEC_PER_USEC;
211
212 entered_state = call_cpuidle_s2idle(drv, dev,
213 max_latency_ns);
214 if (entered_state > 0)
215 goto exit_idle;
216 } else {
217 max_latency_ns = dev->forced_idle_latency_limit_ns;
218 }
219
220 tick_nohz_idle_stop_tick();
221
222 next_state = cpuidle_find_deepest_state(drv, dev, latency_limit_ns: max_latency_ns);
223 call_cpuidle(drv, dev, next_state);
224 } else {
225 bool stop_tick = true;
226
227 /*
228 * Ask the cpuidle framework to choose a convenient idle state.
229 */
230 next_state = cpuidle_select(drv, dev, stop_tick: &stop_tick);
231
232 if (stop_tick || tick_nohz_tick_stopped())
233 tick_nohz_idle_stop_tick();
234 else
235 tick_nohz_idle_retain_tick();
236
237 entered_state = call_cpuidle(drv, dev, next_state);
238 /*
239 * Give the governor an opportunity to reflect on the outcome
240 */
241 cpuidle_reflect(dev, index: entered_state);
242 }
243
244exit_idle:
245 __current_set_polling();
246
247 /*
248 * It is up to the idle functions to re-enable local interrupts
249 */
250 if (WARN_ON_ONCE(irqs_disabled()))
251 local_irq_enable();
252}
253
254/*
255 * Generic idle loop implementation
256 *
257 * Called with polling cleared.
258 */
259static void do_idle(void)
260{
261 int cpu = smp_processor_id();
262
263 /*
264 * Check if we need to update blocked load
265 */
266 nohz_run_idle_balance(cpu);
267
268 /*
269 * If the arch has a polling bit, we maintain an invariant:
270 *
271 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
272 * rq->idle). This means that, if rq->idle has the polling bit set,
273 * then setting need_resched is guaranteed to cause the CPU to
274 * reschedule.
275 */
276
277 __current_set_polling();
278 tick_nohz_idle_enter();
279
280 while (!need_resched()) {
281
282 /*
283 * Interrupts shouldn't be re-enabled from that point on until
284 * the CPU sleeping instruction is reached. Otherwise an interrupt
285 * may fire and queue a timer that would be ignored until the CPU
286 * wakes from the sleeping instruction. And testing need_resched()
287 * doesn't tell about pending needed timer reprogram.
288 *
289 * Several cases to consider:
290 *
291 * - SLEEP-UNTIL-PENDING-INTERRUPT based instructions such as
292 * "wfi" or "mwait" are fine because they can be entered with
293 * interrupt disabled.
294 *
295 * - sti;mwait() couple is fine because the interrupts are
296 * re-enabled only upon the execution of mwait, leaving no gap
297 * in-between.
298 *
299 * - ROLLBACK based idle handlers with the sleeping instruction
300 * called with interrupts enabled are NOT fine. In this scheme
301 * when the interrupt detects it has interrupted an idle handler,
302 * it rolls back to its beginning which performs the
303 * need_resched() check before re-executing the sleeping
304 * instruction. This can leak a pending needed timer reprogram.
305 * If such a scheme is really mandatory due to the lack of an
306 * appropriate CPU sleeping instruction, then a FAST-FORWARD
307 * must instead be applied: when the interrupt detects it has
308 * interrupted an idle handler, it must resume to the end of
309 * this idle handler so that the generic idle loop is iterated
310 * again to reprogram the tick.
311 */
312 local_irq_disable();
313
314 if (cpu_is_offline(cpu)) {
315 cpuhp_report_idle_dead();
316 arch_cpu_idle_dead();
317 }
318
319 arch_cpu_idle_enter();
320 rcu_nocb_flush_deferred_wakeup();
321
322 /*
323 * In poll mode we re-enable interrupts and spin. Also if we
324 * detected in the wakeup from idle path that the tick
325 * broadcast device expired for us, we don't want to go deep
326 * idle as we know that the IPI is going to arrive right away.
327 */
328 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
329 tick_nohz_idle_restart_tick();
330 cpu_idle_poll();
331 } else {
332 cpuidle_idle_call();
333 }
334 arch_cpu_idle_exit();
335 }
336
337 /*
338 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
339 * be set, propagate it into PREEMPT_NEED_RESCHED.
340 *
341 * This is required because for polling idle loops we will not have had
342 * an IPI to fold the state for us.
343 */
344 preempt_set_need_resched();
345 tick_nohz_idle_exit();
346 __current_clr_polling();
347
348 /*
349 * We promise to call sched_ttwu_pending() and reschedule if
350 * need_resched() is set while polling is set. That means that clearing
351 * polling needs to be visible before doing these things.
352 */
353 smp_mb__after_atomic();
354
355 /*
356 * RCU relies on this call to be done outside of an RCU read-side
357 * critical section.
358 */
359 flush_smp_call_function_queue();
360 schedule_idle();
361
362 if (unlikely(klp_patch_pending(current)))
363 klp_update_patch_state(current);
364}
365
366bool cpu_in_idle(unsigned long pc)
367{
368 return pc >= (unsigned long)__cpuidle_text_start &&
369 pc < (unsigned long)__cpuidle_text_end;
370}
371
372struct idle_timer {
373 struct hrtimer timer;
374 int done;
375};
376
377static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
378{
379 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
380
381 WRITE_ONCE(it->done, 1);
382 set_tsk_need_resched(current);
383
384 return HRTIMER_NORESTART;
385}
386
387void play_idle_precise(u64 duration_ns, u64 latency_ns)
388{
389 struct idle_timer it;
390
391 /*
392 * Only FIFO tasks can disable the tick since they don't need the forced
393 * preemption.
394 */
395 WARN_ON_ONCE(current->policy != SCHED_FIFO);
396 WARN_ON_ONCE(current->nr_cpus_allowed != 1);
397 WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
398 WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
399 WARN_ON_ONCE(!duration_ns);
400 WARN_ON_ONCE(current->mm);
401
402 rcu_sleep_check();
403 preempt_disable();
404 current->flags |= PF_IDLE;
405 cpuidle_use_deepest_state(latency_limit_ns: latency_ns);
406
407 it.done = 0;
408 hrtimer_setup_on_stack(timer: &it.timer, function: idle_inject_timer_fn, CLOCK_MONOTONIC,
409 mode: HRTIMER_MODE_REL_HARD);
410 hrtimer_start(timer: &it.timer, tim: ns_to_ktime(ns: duration_ns),
411 mode: HRTIMER_MODE_REL_PINNED_HARD);
412
413 while (!READ_ONCE(it.done))
414 do_idle();
415
416 cpuidle_use_deepest_state(latency_limit_ns: 0);
417 current->flags &= ~PF_IDLE;
418
419 preempt_fold_need_resched();
420 preempt_enable();
421}
422EXPORT_SYMBOL_GPL(play_idle_precise);
423
424void cpu_startup_entry(enum cpuhp_state state)
425{
426 current->flags |= PF_IDLE;
427 arch_cpu_idle_prepare();
428 cpuhp_online_idle(state);
429 while (1)
430 do_idle();
431}
432
433/*
434 * idle-task scheduling class.
435 */
436
437static int
438select_task_rq_idle(struct task_struct *p, int cpu, int flags)
439{
440 return task_cpu(p); /* IDLE tasks as never migrated */
441}
442
443static int
444balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
445{
446 return WARN_ON_ONCE(1);
447}
448
449/*
450 * Idle tasks are unconditionally rescheduled:
451 */
452static void wakeup_preempt_idle(struct rq *rq, struct task_struct *p, int flags)
453{
454 resched_curr(rq);
455}
456
457static void update_curr_idle(struct rq *rq);
458
459static void put_prev_task_idle(struct rq *rq, struct task_struct *prev, struct task_struct *next)
460{
461 update_curr_idle(rq);
462 scx_update_idle(rq, idle: false, do_notify: true);
463}
464
465static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
466{
467 update_idle_core(rq);
468 scx_update_idle(rq, idle: true, do_notify: true);
469 schedstat_inc(rq->sched_goidle);
470 next->se.exec_start = rq_clock_task(rq);
471
472 /*
473 * rq is about to be idle, check if we need to update the
474 * lost_idle_time of clock_pelt
475 */
476 update_idle_rq_clock_pelt(rq);
477}
478
479struct task_struct *pick_task_idle(struct rq *rq, struct rq_flags *rf)
480{
481 scx_update_idle(rq, idle: true, do_notify: false);
482 return rq->idle;
483}
484
485/*
486 * It is not legal to sleep in the idle task - print a warning
487 * message if some code attempts to do it:
488 */
489static bool
490dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
491{
492 raw_spin_rq_unlock_irq(rq);
493 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
494 dump_stack();
495 raw_spin_rq_lock_irq(rq);
496 return true;
497}
498
499/*
500 * scheduler tick hitting a task of our scheduling class.
501 *
502 * NOTE: This function can be called remotely by the tick offload that
503 * goes along full dynticks. Therefore no local assumption can be made
504 * and everything must be accessed through the @rq and @curr passed in
505 * parameters.
506 */
507static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
508{
509 update_curr_idle(rq);
510}
511
512static void switching_to_idle(struct rq *rq, struct task_struct *p)
513{
514 BUG();
515}
516
517static void
518prio_changed_idle(struct rq *rq, struct task_struct *p, u64 oldprio)
519{
520 if (p->prio == oldprio)
521 return;
522
523 BUG();
524}
525
526static void update_curr_idle(struct rq *rq)
527{
528 struct sched_entity *se = &rq->idle->se;
529 u64 now = rq_clock_task(rq);
530 s64 delta_exec;
531
532 delta_exec = now - se->exec_start;
533 if (unlikely(delta_exec <= 0))
534 return;
535
536 se->exec_start = now;
537
538 dl_server_update_idle(dl_se: &rq->fair_server, delta_exec);
539}
540
541/*
542 * Simple, special scheduling class for the per-CPU idle tasks:
543 */
544DEFINE_SCHED_CLASS(idle) = {
545
546 .queue_mask = 0,
547
548 /* no enqueue/yield_task for idle tasks */
549
550 /* dequeue is not valid, we print a debug message there: */
551 .dequeue_task = dequeue_task_idle,
552
553 .wakeup_preempt = wakeup_preempt_idle,
554
555 .pick_task = pick_task_idle,
556 .put_prev_task = put_prev_task_idle,
557 .set_next_task = set_next_task_idle,
558
559 .balance = balance_idle,
560 .select_task_rq = select_task_rq_idle,
561 .set_cpus_allowed = set_cpus_allowed_common,
562
563 .task_tick = task_tick_idle,
564
565 .prio_changed = prio_changed_idle,
566 .switching_to = switching_to_idle,
567 .update_curr = update_curr_idle,
568};
569

source code of linux/kernel/sched/idle.c