1 // SPDX-License-Identifier: GPL-2.0-only
3 * Generic entry points for the idle threads and
4 * implementation of the idle task scheduling class.
6 * (NOTE: these are not related to SCHED_IDLE batch scheduled
7 * tasks which are handled in sched/fair.c )
10 /* Linker adds these: start and end of __cpuidle functions */
11 extern char __cpuidle_text_start[], __cpuidle_text_end[];
14 * sched_idle_set_state - Record idle state for the current CPU.
15 * @idle_state: State to record.
17 void sched_idle_set_state(struct cpuidle_state *idle_state)
19 idle_set_state(this_rq(), idle_state);
22 static int __read_mostly cpu_idle_force_poll;
24 void cpu_idle_poll_ctrl(bool enable)
27 cpu_idle_force_poll++;
29 cpu_idle_force_poll--;
30 WARN_ON_ONCE(cpu_idle_force_poll < 0);
34 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
35 static int __init cpu_idle_poll_setup(char *__unused)
37 cpu_idle_force_poll = 1;
41 __setup("nohlt", cpu_idle_poll_setup);
43 static int __init cpu_idle_nopoll_setup(char *__unused)
45 cpu_idle_force_poll = 0;
49 __setup("hlt", cpu_idle_nopoll_setup);
52 static noinline int __cpuidle cpu_idle_poll(void)
54 trace_cpu_idle(0, smp_processor_id());
55 stop_critical_timings();
59 while (!tif_need_resched() &&
60 (cpu_idle_force_poll || tick_check_broadcast_expired()))
64 start_critical_timings();
65 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
70 /* Weak implementations for optional arch specific functions */
71 void __weak arch_cpu_idle_prepare(void) { }
72 void __weak arch_cpu_idle_enter(void) { }
73 void __weak arch_cpu_idle_exit(void) { }
74 void __weak arch_cpu_idle_dead(void) { }
75 void __weak arch_cpu_idle(void)
77 cpu_idle_force_poll = 1;
78 raw_local_irq_enable();
82 * default_idle_call - Default CPU idle routine.
84 * To use when the cpuidle framework cannot be used.
86 void __cpuidle default_idle_call(void)
88 if (current_clr_polling_and_test()) {
92 trace_cpu_idle(1, smp_processor_id());
93 stop_critical_timings();
96 * arch_cpu_idle() is supposed to enable IRQs, however
97 * we can't do that because of RCU and tracing.
99 * Trace IRQs enable here, then switch off RCU, and have
100 * arch_cpu_idle() use raw_local_irq_enable(). Note that
101 * ct_idle_enter() relies on lockdep IRQ state, so switch that
102 * last -- this is very similar to the entry code.
104 trace_hardirqs_on_prepare();
105 lockdep_hardirqs_on_prepare();
107 lockdep_hardirqs_on(_THIS_IP_);
112 * OK, so IRQs are enabled here, but RCU needs them disabled to
113 * turn itself back on.. funny thing is that disabling IRQs
114 * will cause tracing, which needs RCU. Jump through hoops to
117 raw_local_irq_disable();
118 lockdep_hardirqs_off(_THIS_IP_);
120 lockdep_hardirqs_on(_THIS_IP_);
121 raw_local_irq_enable();
123 start_critical_timings();
124 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
128 static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
129 struct cpuidle_device *dev)
131 if (current_clr_polling_and_test())
134 return cpuidle_enter_s2idle(drv, dev);
137 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
141 * The idle task must be scheduled, it is pointless to go to idle, just
142 * update no idle residency and return.
144 if (current_clr_polling_and_test()) {
145 dev->last_residency_ns = 0;
151 * Enter the idle state previously returned by the governor decision.
152 * This function will block until an interrupt occurs and will take
153 * care of re-enabling the local interrupts
155 return cpuidle_enter(drv, dev, next_state);
159 * cpuidle_idle_call - the main idle function
161 * NOTE: no locks or semaphores should be used here
163 * On architectures that support TIF_POLLING_NRFLAG, is called with polling
164 * set, and it returns with polling set. If it ever stops polling, it
165 * must clear the polling bit.
167 static void cpuidle_idle_call(void)
169 struct cpuidle_device *dev = cpuidle_get_device();
170 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
171 int next_state, entered_state;
174 * Check if the idle task must be rescheduled. If it is the
175 * case, exit the function after re-enabling the local irq.
177 if (need_resched()) {
183 * The RCU framework needs to be told that we are entering an idle
184 * section, so no more rcu read side critical sections and one more
185 * step to the grace period
188 if (cpuidle_not_available(drv, dev)) {
189 tick_nohz_idle_stop_tick();
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.
205 if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
208 if (idle_should_enter_s2idle()) {
210 entered_state = call_cpuidle_s2idle(drv, dev);
211 if (entered_state > 0)
214 max_latency_ns = U64_MAX;
216 max_latency_ns = dev->forced_idle_latency_limit_ns;
219 tick_nohz_idle_stop_tick();
221 next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
222 call_cpuidle(drv, dev, next_state);
224 bool stop_tick = true;
227 * Ask the cpuidle framework to choose a convenient idle state.
229 next_state = cpuidle_select(drv, dev, &stop_tick);
231 if (stop_tick || tick_nohz_tick_stopped())
232 tick_nohz_idle_stop_tick();
234 tick_nohz_idle_retain_tick();
236 entered_state = call_cpuidle(drv, dev, next_state);
238 * Give the governor an opportunity to reflect on the outcome
240 cpuidle_reflect(dev, entered_state);
244 __current_set_polling();
247 * It is up to the idle functions to reenable local interrupts
249 if (WARN_ON_ONCE(irqs_disabled()))
254 * Generic idle loop implementation
256 * Called with polling cleared.
258 static void do_idle(void)
260 int cpu = smp_processor_id();
263 * Check if we need to update blocked load
265 nohz_run_idle_balance(cpu);
268 * If the arch has a polling bit, we maintain an invariant:
270 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
271 * rq->idle). This means that, if rq->idle has the polling bit set,
272 * then setting need_resched is guaranteed to cause the CPU to
276 __current_set_polling();
277 tick_nohz_idle_enter();
279 while (!need_resched()) {
284 if (cpu_is_offline(cpu)) {
285 tick_nohz_idle_stop_tick();
286 cpuhp_report_idle_dead();
287 arch_cpu_idle_dead();
290 arch_cpu_idle_enter();
291 rcu_nocb_flush_deferred_wakeup();
294 * In poll mode we reenable interrupts and spin. Also if we
295 * detected in the wakeup from idle path that the tick
296 * broadcast device expired for us, we don't want to go deep
297 * idle as we know that the IPI is going to arrive right away.
299 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
300 tick_nohz_idle_restart_tick();
305 arch_cpu_idle_exit();
309 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
310 * be set, propagate it into PREEMPT_NEED_RESCHED.
312 * This is required because for polling idle loops we will not have had
313 * an IPI to fold the state for us.
315 preempt_set_need_resched();
316 tick_nohz_idle_exit();
317 __current_clr_polling();
320 * We promise to call sched_ttwu_pending() and reschedule if
321 * need_resched() is set while polling is set. That means that clearing
322 * polling needs to be visible before doing these things.
324 smp_mb__after_atomic();
327 * RCU relies on this call to be done outside of an RCU read-side
330 flush_smp_call_function_queue();
333 if (unlikely(klp_patch_pending(current)))
334 klp_update_patch_state(current);
337 bool cpu_in_idle(unsigned long pc)
339 return pc >= (unsigned long)__cpuidle_text_start &&
340 pc < (unsigned long)__cpuidle_text_end;
344 struct hrtimer timer;
348 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
350 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
352 WRITE_ONCE(it->done, 1);
353 set_tsk_need_resched(current);
355 return HRTIMER_NORESTART;
358 void play_idle_precise(u64 duration_ns, u64 latency_ns)
360 struct idle_timer it;
363 * Only FIFO tasks can disable the tick since they don't need the forced
366 WARN_ON_ONCE(current->policy != SCHED_FIFO);
367 WARN_ON_ONCE(current->nr_cpus_allowed != 1);
368 WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
369 WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
370 WARN_ON_ONCE(!duration_ns);
371 WARN_ON_ONCE(current->mm);
375 current->flags |= PF_IDLE;
376 cpuidle_use_deepest_state(latency_ns);
379 hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
380 it.timer.function = idle_inject_timer_fn;
381 hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
382 HRTIMER_MODE_REL_PINNED_HARD);
384 while (!READ_ONCE(it.done))
387 cpuidle_use_deepest_state(0);
388 current->flags &= ~PF_IDLE;
390 preempt_fold_need_resched();
393 EXPORT_SYMBOL_GPL(play_idle_precise);
395 void cpu_startup_entry(enum cpuhp_state state)
397 arch_cpu_idle_prepare();
398 cpuhp_online_idle(state);
404 * idle-task scheduling class.
409 select_task_rq_idle(struct task_struct *p, int cpu, int flags)
411 return task_cpu(p); /* IDLE tasks as never migrated */
415 balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
417 return WARN_ON_ONCE(1);
422 * Idle tasks are unconditionally rescheduled:
424 static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
429 static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
433 static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
435 update_idle_core(rq);
436 schedstat_inc(rq->sched_goidle);
440 static struct task_struct *pick_task_idle(struct rq *rq)
446 struct task_struct *pick_next_task_idle(struct rq *rq)
448 struct task_struct *next = rq->idle;
450 set_next_task_idle(rq, next, true);
456 * It is not legal to sleep in the idle task - print a warning
457 * message if some code attempts to do it:
460 dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
462 raw_spin_rq_unlock_irq(rq);
463 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
465 raw_spin_rq_lock_irq(rq);
469 * scheduler tick hitting a task of our scheduling class.
471 * NOTE: This function can be called remotely by the tick offload that
472 * goes along full dynticks. Therefore no local assumption can be made
473 * and everything must be accessed through the @rq and @curr passed in
476 static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
480 static void switched_to_idle(struct rq *rq, struct task_struct *p)
486 prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
491 static void update_curr_idle(struct rq *rq)
496 * Simple, special scheduling class for the per-CPU idle tasks:
498 DEFINE_SCHED_CLASS(idle) = {
500 /* no enqueue/yield_task for idle tasks */
502 /* dequeue is not valid, we print a debug message there: */
503 .dequeue_task = dequeue_task_idle,
505 .check_preempt_curr = check_preempt_curr_idle,
507 .pick_next_task = pick_next_task_idle,
508 .put_prev_task = put_prev_task_idle,
509 .set_next_task = set_next_task_idle,
512 .balance = balance_idle,
513 .pick_task = pick_task_idle,
514 .select_task_rq = select_task_rq_idle,
515 .set_cpus_allowed = set_cpus_allowed_common,
518 .task_tick = task_tick_idle,
520 .prio_changed = prio_changed_idle,
521 .switched_to = switched_to_idle,
522 .update_curr = update_curr_idle,