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 )
11 #include <trace/events/power.h>
13 /* Linker adds these: start and end of __cpuidle functions */
14 extern char __cpuidle_text_start[], __cpuidle_text_end[];
17 * sched_idle_set_state - Record idle state for the current CPU.
18 * @idle_state: State to record.
20 void sched_idle_set_state(struct cpuidle_state *idle_state)
22 idle_set_state(this_rq(), idle_state);
25 static int __read_mostly cpu_idle_force_poll;
27 void cpu_idle_poll_ctrl(bool enable)
30 cpu_idle_force_poll++;
32 cpu_idle_force_poll--;
33 WARN_ON_ONCE(cpu_idle_force_poll < 0);
37 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
38 static int __init cpu_idle_poll_setup(char *__unused)
40 cpu_idle_force_poll = 1;
44 __setup("nohlt", cpu_idle_poll_setup);
46 static int __init cpu_idle_nopoll_setup(char *__unused)
48 cpu_idle_force_poll = 0;
52 __setup("hlt", cpu_idle_nopoll_setup);
55 static noinline int __cpuidle cpu_idle_poll(void)
57 trace_cpu_idle(0, smp_processor_id());
58 stop_critical_timings();
62 while (!tif_need_resched() &&
63 (cpu_idle_force_poll || tick_check_broadcast_expired()))
67 start_critical_timings();
68 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
73 /* Weak implementations for optional arch specific functions */
74 void __weak arch_cpu_idle_prepare(void) { }
75 void __weak arch_cpu_idle_enter(void) { }
76 void __weak arch_cpu_idle_exit(void) { }
77 void __weak arch_cpu_idle_dead(void) { }
78 void __weak arch_cpu_idle(void)
80 cpu_idle_force_poll = 1;
81 raw_local_irq_enable();
85 * default_idle_call - Default CPU idle routine.
87 * To use when the cpuidle framework cannot be used.
89 void __cpuidle default_idle_call(void)
91 if (current_clr_polling_and_test()) {
95 trace_cpu_idle(1, smp_processor_id());
96 stop_critical_timings();
99 * arch_cpu_idle() is supposed to enable IRQs, however
100 * we can't do that because of RCU and tracing.
102 * Trace IRQs enable here, then switch off RCU, and have
103 * arch_cpu_idle() use raw_local_irq_enable(). Note that
104 * rcu_idle_enter() relies on lockdep IRQ state, so switch that
105 * last -- this is very similar to the entry code.
107 trace_hardirqs_on_prepare();
108 lockdep_hardirqs_on_prepare(_THIS_IP_);
110 lockdep_hardirqs_on(_THIS_IP_);
115 * OK, so IRQs are enabled here, but RCU needs them disabled to
116 * turn itself back on.. funny thing is that disabling IRQs
117 * will cause tracing, which needs RCU. Jump through hoops to
120 raw_local_irq_disable();
121 lockdep_hardirqs_off(_THIS_IP_);
123 lockdep_hardirqs_on(_THIS_IP_);
124 raw_local_irq_enable();
126 start_critical_timings();
127 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
131 static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
132 struct cpuidle_device *dev)
134 if (current_clr_polling_and_test())
137 return cpuidle_enter_s2idle(drv, dev);
140 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
144 * The idle task must be scheduled, it is pointless to go to idle, just
145 * update no idle residency and return.
147 if (current_clr_polling_and_test()) {
148 dev->last_residency_ns = 0;
154 * Enter the idle state previously returned by the governor decision.
155 * This function will block until an interrupt occurs and will take
156 * care of re-enabling the local interrupts
158 return cpuidle_enter(drv, dev, next_state);
162 * cpuidle_idle_call - the main idle function
164 * NOTE: no locks or semaphores should be used here
166 * On archs that support TIF_POLLING_NRFLAG, is called with polling
167 * set, and it returns with polling set. If it ever stops polling, it
168 * must clear the polling bit.
170 static void cpuidle_idle_call(void)
172 struct cpuidle_device *dev = cpuidle_get_device();
173 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
174 int next_state, entered_state;
177 * Check if the idle task must be rescheduled. If it is the
178 * case, exit the function after re-enabling the local irq.
180 if (need_resched()) {
186 * The RCU framework needs to be told that we are entering an idle
187 * section, so no more rcu read side critical sections and one more
188 * step to the grace period
191 if (cpuidle_not_available(drv, dev)) {
192 tick_nohz_idle_stop_tick();
199 * Suspend-to-idle ("s2idle") is a system state in which all user space
200 * has been frozen, all I/O devices have been suspended and the only
201 * activity happens here and in interrupts (if any). In that case bypass
202 * the cpuidle governor and go stratight for the deepest idle state
203 * available. Possibly also suspend the local tick and the entire
204 * timekeeping to prevent timer interrupts from kicking us out of idle
205 * until a proper wakeup interrupt happens.
208 if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
211 if (idle_should_enter_s2idle()) {
213 entered_state = call_cpuidle_s2idle(drv, dev);
214 if (entered_state > 0)
217 max_latency_ns = U64_MAX;
219 max_latency_ns = dev->forced_idle_latency_limit_ns;
222 tick_nohz_idle_stop_tick();
224 next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
225 call_cpuidle(drv, dev, next_state);
227 bool stop_tick = true;
230 * Ask the cpuidle framework to choose a convenient idle state.
232 next_state = cpuidle_select(drv, dev, &stop_tick);
234 if (stop_tick || tick_nohz_tick_stopped())
235 tick_nohz_idle_stop_tick();
237 tick_nohz_idle_retain_tick();
239 entered_state = call_cpuidle(drv, dev, next_state);
241 * Give the governor an opportunity to reflect on the outcome
243 cpuidle_reflect(dev, entered_state);
247 __current_set_polling();
250 * It is up to the idle functions to reenable local interrupts
252 if (WARN_ON_ONCE(irqs_disabled()))
257 * Generic idle loop implementation
259 * Called with polling cleared.
261 static void do_idle(void)
263 int cpu = smp_processor_id();
265 * If the arch has a polling bit, we maintain an invariant:
267 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
268 * rq->idle). This means that, if rq->idle has the polling bit set,
269 * then setting need_resched is guaranteed to cause the CPU to
273 __current_set_polling();
274 tick_nohz_idle_enter();
276 while (!need_resched()) {
281 if (cpu_is_offline(cpu)) {
282 tick_nohz_idle_stop_tick();
283 cpuhp_report_idle_dead();
284 arch_cpu_idle_dead();
287 arch_cpu_idle_enter();
290 * In poll mode we reenable interrupts and spin. Also if we
291 * detected in the wakeup from idle path that the tick
292 * broadcast device expired for us, we don't want to go deep
293 * idle as we know that the IPI is going to arrive right away.
295 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
296 tick_nohz_idle_restart_tick();
301 arch_cpu_idle_exit();
305 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
306 * be set, propagate it into PREEMPT_NEED_RESCHED.
308 * This is required because for polling idle loops we will not have had
309 * an IPI to fold the state for us.
311 preempt_set_need_resched();
312 tick_nohz_idle_exit();
313 __current_clr_polling();
316 * We promise to call sched_ttwu_pending() and reschedule if
317 * need_resched() is set while polling is set. That means that clearing
318 * polling needs to be visible before doing these things.
320 smp_mb__after_atomic();
323 * RCU relies on this call to be done outside of an RCU read-side
326 flush_smp_call_function_from_idle();
329 if (unlikely(klp_patch_pending(current)))
330 klp_update_patch_state(current);
333 bool cpu_in_idle(unsigned long pc)
335 return pc >= (unsigned long)__cpuidle_text_start &&
336 pc < (unsigned long)__cpuidle_text_end;
340 struct hrtimer timer;
344 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
346 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
348 WRITE_ONCE(it->done, 1);
349 set_tsk_need_resched(current);
351 return HRTIMER_NORESTART;
354 void play_idle_precise(u64 duration_ns, u64 latency_ns)
356 struct idle_timer it;
359 * Only FIFO tasks can disable the tick since they don't need the forced
362 WARN_ON_ONCE(current->policy != SCHED_FIFO);
363 WARN_ON_ONCE(current->nr_cpus_allowed != 1);
364 WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
365 WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
366 WARN_ON_ONCE(!duration_ns);
367 WARN_ON_ONCE(current->mm);
371 current->flags |= PF_IDLE;
372 cpuidle_use_deepest_state(latency_ns);
375 hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
376 it.timer.function = idle_inject_timer_fn;
377 hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
378 HRTIMER_MODE_REL_PINNED);
380 while (!READ_ONCE(it.done))
383 cpuidle_use_deepest_state(0);
384 current->flags &= ~PF_IDLE;
386 preempt_fold_need_resched();
389 EXPORT_SYMBOL_GPL(play_idle_precise);
391 void cpu_startup_entry(enum cpuhp_state state)
393 arch_cpu_idle_prepare();
394 cpuhp_online_idle(state);
400 * idle-task scheduling class.
405 select_task_rq_idle(struct task_struct *p, int cpu, int flags)
407 return task_cpu(p); /* IDLE tasks as never migrated */
411 balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
413 return WARN_ON_ONCE(1);
418 * Idle tasks are unconditionally rescheduled:
420 static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
425 static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
429 static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
431 update_idle_core(rq);
432 schedstat_inc(rq->sched_goidle);
435 struct task_struct *pick_next_task_idle(struct rq *rq)
437 struct task_struct *next = rq->idle;
439 set_next_task_idle(rq, next, true);
445 * It is not legal to sleep in the idle task - print a warning
446 * message if some code attempts to do it:
449 dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
451 raw_spin_unlock_irq(&rq->lock);
452 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
454 raw_spin_lock_irq(&rq->lock);
458 * scheduler tick hitting a task of our scheduling class.
460 * NOTE: This function can be called remotely by the tick offload that
461 * goes along full dynticks. Therefore no local assumption can be made
462 * and everything must be accessed through the @rq and @curr passed in
465 static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
469 static void switched_to_idle(struct rq *rq, struct task_struct *p)
475 prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
480 static void update_curr_idle(struct rq *rq)
485 * Simple, special scheduling class for the per-CPU idle tasks:
487 DEFINE_SCHED_CLASS(idle) = {
489 /* no enqueue/yield_task for idle tasks */
491 /* dequeue is not valid, we print a debug message there: */
492 .dequeue_task = dequeue_task_idle,
494 .check_preempt_curr = check_preempt_curr_idle,
496 .pick_next_task = pick_next_task_idle,
497 .put_prev_task = put_prev_task_idle,
498 .set_next_task = set_next_task_idle,
501 .balance = balance_idle,
502 .select_task_rq = select_task_rq_idle,
503 .set_cpus_allowed = set_cpus_allowed_common,
506 .task_tick = task_tick_idle,
508 .prio_changed = prio_changed_idle,
509 .switched_to = switched_to_idle,
510 .update_curr = update_curr_idle,