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)
58 trace_cpu_idle_rcuidle(0, smp_processor_id());
60 stop_critical_timings();
62 while (!tif_need_resched() &&
63 (cpu_idle_force_poll || tick_check_broadcast_expired()))
65 start_critical_timings();
66 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
72 /* Weak implementations for optional arch specific functions */
73 void __weak arch_cpu_idle_prepare(void) { }
74 void __weak arch_cpu_idle_enter(void) { }
75 void __weak arch_cpu_idle_exit(void) { }
76 void __weak arch_cpu_idle_dead(void) { }
77 void __weak arch_cpu_idle(void)
79 cpu_idle_force_poll = 1;
84 * default_idle_call - Default CPU idle routine.
86 * To use when the cpuidle framework cannot be used.
88 void __cpuidle default_idle_call(void)
90 if (current_clr_polling_and_test()) {
93 stop_critical_timings();
95 start_critical_timings();
99 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
103 * The idle task must be scheduled, it is pointless to go to idle, just
104 * update no idle residency and return.
106 if (current_clr_polling_and_test()) {
107 dev->last_residency = 0;
113 * Enter the idle state previously returned by the governor decision.
114 * This function will block until an interrupt occurs and will take
115 * care of re-enabling the local interrupts
117 return cpuidle_enter(drv, dev, next_state);
121 * cpuidle_idle_call - the main idle function
123 * NOTE: no locks or semaphores should be used here
125 * On archs that support TIF_POLLING_NRFLAG, is called with polling
126 * set, and it returns with polling set. If it ever stops polling, it
127 * must clear the polling bit.
129 static void cpuidle_idle_call(void)
131 struct cpuidle_device *dev = cpuidle_get_device();
132 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
133 int next_state, entered_state;
136 * Check if the idle task must be rescheduled. If it is the
137 * case, exit the function after re-enabling the local irq.
139 if (need_resched()) {
145 * The RCU framework needs to be told that we are entering an idle
146 * section, so no more rcu read side critical sections and one more
147 * step to the grace period
150 if (cpuidle_not_available(drv, dev)) {
151 tick_nohz_idle_stop_tick();
159 * Suspend-to-idle ("s2idle") is a system state in which all user space
160 * has been frozen, all I/O devices have been suspended and the only
161 * activity happens here and in iterrupts (if any). In that case bypass
162 * the cpuidle governor and go stratight for the deepest idle state
163 * available. Possibly also suspend the local tick and the entire
164 * timekeeping to prevent timer interrupts from kicking us out of idle
165 * until a proper wakeup interrupt happens.
168 if (idle_should_enter_s2idle() || dev->use_deepest_state) {
169 if (idle_should_enter_s2idle()) {
172 entered_state = cpuidle_enter_s2idle(drv, dev);
173 if (entered_state > 0) {
181 tick_nohz_idle_stop_tick();
184 next_state = cpuidle_find_deepest_state(drv, dev);
185 call_cpuidle(drv, dev, next_state);
187 bool stop_tick = true;
190 * Ask the cpuidle framework to choose a convenient idle state.
192 next_state = cpuidle_select(drv, dev, &stop_tick);
194 if (stop_tick || tick_nohz_tick_stopped())
195 tick_nohz_idle_stop_tick();
197 tick_nohz_idle_retain_tick();
201 entered_state = call_cpuidle(drv, dev, next_state);
203 * Give the governor an opportunity to reflect on the outcome
205 cpuidle_reflect(dev, entered_state);
209 __current_set_polling();
212 * It is up to the idle functions to reenable local interrupts
214 if (WARN_ON_ONCE(irqs_disabled()))
221 * Generic idle loop implementation
223 * Called with polling cleared.
225 static void do_idle(void)
227 int cpu = smp_processor_id();
229 * If the arch has a polling bit, we maintain an invariant:
231 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
232 * rq->idle). This means that, if rq->idle has the polling bit set,
233 * then setting need_resched is guaranteed to cause the CPU to
237 __current_set_polling();
238 tick_nohz_idle_enter();
240 while (!need_resched()) {
244 if (cpu_is_offline(cpu)) {
245 tick_nohz_idle_stop_tick_protected();
246 cpuhp_report_idle_dead();
247 arch_cpu_idle_dead();
251 arch_cpu_idle_enter();
254 * In poll mode we reenable interrupts and spin. Also if we
255 * detected in the wakeup from idle path that the tick
256 * broadcast device expired for us, we don't want to go deep
257 * idle as we know that the IPI is going to arrive right away.
259 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
260 tick_nohz_idle_restart_tick();
265 arch_cpu_idle_exit();
269 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
270 * be set, propagate it into PREEMPT_NEED_RESCHED.
272 * This is required because for polling idle loops we will not have had
273 * an IPI to fold the state for us.
275 preempt_set_need_resched();
276 tick_nohz_idle_exit();
277 __current_clr_polling();
280 * We promise to call sched_ttwu_pending() and reschedule if
281 * need_resched() is set while polling is set. That means that clearing
282 * polling needs to be visible before doing these things.
284 smp_mb__after_atomic();
286 sched_ttwu_pending();
289 if (unlikely(klp_patch_pending(current)))
290 klp_update_patch_state(current);
293 bool cpu_in_idle(unsigned long pc)
295 return pc >= (unsigned long)__cpuidle_text_start &&
296 pc < (unsigned long)__cpuidle_text_end;
300 struct hrtimer timer;
304 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
306 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
308 WRITE_ONCE(it->done, 1);
309 set_tsk_need_resched(current);
311 return HRTIMER_NORESTART;
314 void play_idle(unsigned long duration_ms)
316 struct idle_timer it;
319 * Only FIFO tasks can disable the tick since they don't need the forced
322 WARN_ON_ONCE(current->policy != SCHED_FIFO);
323 WARN_ON_ONCE(current->nr_cpus_allowed != 1);
324 WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
325 WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
326 WARN_ON_ONCE(!duration_ms);
330 current->flags |= PF_IDLE;
331 cpuidle_use_deepest_state(true);
334 hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
335 it.timer.function = idle_inject_timer_fn;
336 hrtimer_start(&it.timer, ms_to_ktime(duration_ms), HRTIMER_MODE_REL_PINNED);
338 while (!READ_ONCE(it.done))
341 cpuidle_use_deepest_state(false);
342 current->flags &= ~PF_IDLE;
344 preempt_fold_need_resched();
347 EXPORT_SYMBOL_GPL(play_idle);
349 void cpu_startup_entry(enum cpuhp_state state)
351 arch_cpu_idle_prepare();
352 cpuhp_online_idle(state);
358 * idle-task scheduling class.
363 select_task_rq_idle(struct task_struct *p, int cpu, int sd_flag, int flags)
365 return task_cpu(p); /* IDLE tasks as never migrated */
370 * Idle tasks are unconditionally rescheduled:
372 static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
377 static struct task_struct *
378 pick_next_task_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
380 put_prev_task(rq, prev);
381 update_idle_core(rq);
382 schedstat_inc(rq->sched_goidle);
388 * It is not legal to sleep in the idle task - print a warning
389 * message if some code attempts to do it:
392 dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
394 raw_spin_unlock_irq(&rq->lock);
395 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
397 raw_spin_lock_irq(&rq->lock);
400 static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
405 * scheduler tick hitting a task of our scheduling class.
407 * NOTE: This function can be called remotely by the tick offload that
408 * goes along full dynticks. Therefore no local assumption can be made
409 * and everything must be accessed through the @rq and @curr passed in
412 static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
416 static void set_curr_task_idle(struct rq *rq)
420 static void switched_to_idle(struct rq *rq, struct task_struct *p)
426 prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
431 static unsigned int get_rr_interval_idle(struct rq *rq, struct task_struct *task)
436 static void update_curr_idle(struct rq *rq)
441 * Simple, special scheduling class for the per-CPU idle tasks:
443 const struct sched_class idle_sched_class = {
445 /* no enqueue/yield_task for idle tasks */
447 /* dequeue is not valid, we print a debug message there: */
448 .dequeue_task = dequeue_task_idle,
450 .check_preempt_curr = check_preempt_curr_idle,
452 .pick_next_task = pick_next_task_idle,
453 .put_prev_task = put_prev_task_idle,
456 .select_task_rq = select_task_rq_idle,
457 .set_cpus_allowed = set_cpus_allowed_common,
460 .set_curr_task = set_curr_task_idle,
461 .task_tick = task_tick_idle,
463 .get_rr_interval = get_rr_interval_idle,
465 .prio_changed = prio_changed_idle,
466 .switched_to = switched_to_idle,
467 .update_curr = update_curr_idle,