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 instrumentation_begin();
55 trace_cpu_idle(0, smp_processor_id());
56 stop_critical_timings();
59 raw_local_irq_enable();
60 while (!tif_need_resched() &&
61 (cpu_idle_force_poll || tick_check_broadcast_expired()))
63 raw_local_irq_disable();
66 start_critical_timings();
67 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
69 instrumentation_end();
74 /* Weak implementations for optional arch specific functions */
75 void __weak arch_cpu_idle_prepare(void) { }
76 void __weak arch_cpu_idle_enter(void) { }
77 void __weak arch_cpu_idle_exit(void) { }
78 void __weak __noreturn arch_cpu_idle_dead(void) { while (1); }
79 void __weak arch_cpu_idle(void)
81 cpu_idle_force_poll = 1;
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 instrumentation_begin();
92 if (!current_clr_polling_and_test()) {
93 trace_cpu_idle(1, smp_processor_id());
94 stop_critical_timings();
100 start_critical_timings();
101 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
104 instrumentation_end();
107 static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
108 struct cpuidle_device *dev)
110 if (current_clr_polling_and_test())
113 return cpuidle_enter_s2idle(drv, dev);
116 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
120 * The idle task must be scheduled, it is pointless to go to idle, just
121 * update no idle residency and return.
123 if (current_clr_polling_and_test()) {
124 dev->last_residency_ns = 0;
130 * Enter the idle state previously returned by the governor decision.
131 * This function will block until an interrupt occurs and will take
132 * care of re-enabling the local interrupts
134 return cpuidle_enter(drv, dev, next_state);
138 * cpuidle_idle_call - the main idle function
140 * NOTE: no locks or semaphores should be used here
142 * On architectures that support TIF_POLLING_NRFLAG, is called with polling
143 * set, and it returns with polling set. If it ever stops polling, it
144 * must clear the polling bit.
146 static void cpuidle_idle_call(void)
148 struct cpuidle_device *dev = cpuidle_get_device();
149 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
150 int next_state, entered_state;
153 * Check if the idle task must be rescheduled. If it is the
154 * case, exit the function after re-enabling the local irq.
156 if (need_resched()) {
162 * The RCU framework needs to be told that we are entering an idle
163 * section, so no more rcu read side critical sections and one more
164 * step to the grace period
167 if (cpuidle_not_available(drv, dev)) {
168 tick_nohz_idle_stop_tick();
175 * Suspend-to-idle ("s2idle") is a system state in which all user space
176 * has been frozen, all I/O devices have been suspended and the only
177 * activity happens here and in interrupts (if any). In that case bypass
178 * the cpuidle governor and go straight for the deepest idle state
179 * available. Possibly also suspend the local tick and the entire
180 * timekeeping to prevent timer interrupts from kicking us out of idle
181 * until a proper wakeup interrupt happens.
184 if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
187 if (idle_should_enter_s2idle()) {
189 entered_state = call_cpuidle_s2idle(drv, dev);
190 if (entered_state > 0)
193 max_latency_ns = U64_MAX;
195 max_latency_ns = dev->forced_idle_latency_limit_ns;
198 tick_nohz_idle_stop_tick();
200 next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
201 call_cpuidle(drv, dev, next_state);
203 bool stop_tick = true;
206 * Ask the cpuidle framework to choose a convenient idle state.
208 next_state = cpuidle_select(drv, dev, &stop_tick);
210 if (stop_tick || tick_nohz_tick_stopped())
211 tick_nohz_idle_stop_tick();
213 tick_nohz_idle_retain_tick();
215 entered_state = call_cpuidle(drv, dev, next_state);
217 * Give the governor an opportunity to reflect on the outcome
219 cpuidle_reflect(dev, entered_state);
223 __current_set_polling();
226 * It is up to the idle functions to reenable local interrupts
228 if (WARN_ON_ONCE(irqs_disabled()))
233 * Generic idle loop implementation
235 * Called with polling cleared.
237 static void do_idle(void)
239 int cpu = smp_processor_id();
242 * Check if we need to update blocked load
244 nohz_run_idle_balance(cpu);
247 * If the arch has a polling bit, we maintain an invariant:
249 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
250 * rq->idle). This means that, if rq->idle has the polling bit set,
251 * then setting need_resched is guaranteed to cause the CPU to
255 __current_set_polling();
256 tick_nohz_idle_enter();
258 while (!need_resched()) {
262 * Interrupts shouldn't be re-enabled from that point on until
263 * the CPU sleeping instruction is reached. Otherwise an interrupt
264 * may fire and queue a timer that would be ignored until the CPU
265 * wakes from the sleeping instruction. And testing need_resched()
266 * doesn't tell about pending needed timer reprogram.
268 * Several cases to consider:
270 * - SLEEP-UNTIL-PENDING-INTERRUPT based instructions such as
271 * "wfi" or "mwait" are fine because they can be entered with
272 * interrupt disabled.
274 * - sti;mwait() couple is fine because the interrupts are
275 * re-enabled only upon the execution of mwait, leaving no gap
278 * - ROLLBACK based idle handlers with the sleeping instruction
279 * called with interrupts enabled are NOT fine. In this scheme
280 * when the interrupt detects it has interrupted an idle handler,
281 * it rolls back to its beginning which performs the
282 * need_resched() check before re-executing the sleeping
283 * instruction. This can leak a pending needed timer reprogram.
284 * If such a scheme is really mandatory due to the lack of an
285 * appropriate CPU sleeping instruction, then a FAST-FORWARD
286 * must instead be applied: when the interrupt detects it has
287 * interrupted an idle handler, it must resume to the end of
288 * this idle handler so that the generic idle loop is iterated
289 * again to reprogram the tick.
293 if (cpu_is_offline(cpu)) {
294 tick_nohz_idle_stop_tick();
295 cpuhp_report_idle_dead();
296 arch_cpu_idle_dead();
299 arch_cpu_idle_enter();
300 rcu_nocb_flush_deferred_wakeup();
303 * In poll mode we reenable interrupts and spin. Also if we
304 * detected in the wakeup from idle path that the tick
305 * broadcast device expired for us, we don't want to go deep
306 * idle as we know that the IPI is going to arrive right away.
308 if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
309 tick_nohz_idle_restart_tick();
314 arch_cpu_idle_exit();
318 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
319 * be set, propagate it into PREEMPT_NEED_RESCHED.
321 * This is required because for polling idle loops we will not have had
322 * an IPI to fold the state for us.
324 preempt_set_need_resched();
325 tick_nohz_idle_exit();
326 __current_clr_polling();
329 * We promise to call sched_ttwu_pending() and reschedule if
330 * need_resched() is set while polling is set. That means that clearing
331 * polling needs to be visible before doing these things.
333 smp_mb__after_atomic();
336 * RCU relies on this call to be done outside of an RCU read-side
339 flush_smp_call_function_queue();
342 if (unlikely(klp_patch_pending(current)))
343 klp_update_patch_state(current);
346 bool cpu_in_idle(unsigned long pc)
348 return pc >= (unsigned long)__cpuidle_text_start &&
349 pc < (unsigned long)__cpuidle_text_end;
353 struct hrtimer timer;
357 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
359 struct idle_timer *it = container_of(timer, struct idle_timer, timer);
361 WRITE_ONCE(it->done, 1);
362 set_tsk_need_resched(current);
364 return HRTIMER_NORESTART;
367 void play_idle_precise(u64 duration_ns, u64 latency_ns)
369 struct idle_timer it;
372 * Only FIFO tasks can disable the tick since they don't need the forced
375 WARN_ON_ONCE(current->policy != SCHED_FIFO);
376 WARN_ON_ONCE(current->nr_cpus_allowed != 1);
377 WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
378 WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
379 WARN_ON_ONCE(!duration_ns);
380 WARN_ON_ONCE(current->mm);
384 current->flags |= PF_IDLE;
385 cpuidle_use_deepest_state(latency_ns);
388 hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
389 it.timer.function = idle_inject_timer_fn;
390 hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
391 HRTIMER_MODE_REL_PINNED_HARD);
393 while (!READ_ONCE(it.done))
396 cpuidle_use_deepest_state(0);
397 current->flags &= ~PF_IDLE;
399 preempt_fold_need_resched();
402 EXPORT_SYMBOL_GPL(play_idle_precise);
404 void cpu_startup_entry(enum cpuhp_state state)
406 current->flags |= PF_IDLE;
407 arch_cpu_idle_prepare();
408 cpuhp_online_idle(state);
414 * idle-task scheduling class.
419 select_task_rq_idle(struct task_struct *p, int cpu, int flags)
421 return task_cpu(p); /* IDLE tasks as never migrated */
425 balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
427 return WARN_ON_ONCE(1);
432 * Idle tasks are unconditionally rescheduled:
434 static void wakeup_preempt_idle(struct rq *rq, struct task_struct *p, int flags)
439 static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
443 static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
445 update_idle_core(rq);
446 schedstat_inc(rq->sched_goidle);
450 static struct task_struct *pick_task_idle(struct rq *rq)
456 struct task_struct *pick_next_task_idle(struct rq *rq)
458 struct task_struct *next = rq->idle;
460 set_next_task_idle(rq, next, true);
466 * It is not legal to sleep in the idle task - print a warning
467 * message if some code attempts to do it:
470 dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
472 raw_spin_rq_unlock_irq(rq);
473 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
475 raw_spin_rq_lock_irq(rq);
479 * scheduler tick hitting a task of our scheduling class.
481 * NOTE: This function can be called remotely by the tick offload that
482 * goes along full dynticks. Therefore no local assumption can be made
483 * and everything must be accessed through the @rq and @curr passed in
486 static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
490 static void switched_to_idle(struct rq *rq, struct task_struct *p)
496 prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
501 static void update_curr_idle(struct rq *rq)
506 * Simple, special scheduling class for the per-CPU idle tasks:
508 DEFINE_SCHED_CLASS(idle) = {
510 /* no enqueue/yield_task for idle tasks */
512 /* dequeue is not valid, we print a debug message there: */
513 .dequeue_task = dequeue_task_idle,
515 .wakeup_preempt = wakeup_preempt_idle,
517 .pick_next_task = pick_next_task_idle,
518 .put_prev_task = put_prev_task_idle,
519 .set_next_task = set_next_task_idle,
522 .balance = balance_idle,
523 .pick_task = pick_task_idle,
524 .select_task_rq = select_task_rq_idle,
525 .set_cpus_allowed = set_cpus_allowed_common,
528 .task_tick = task_tick_idle,
530 .prio_changed = prio_changed_idle,
531 .switched_to = switched_to_idle,
532 .update_curr = update_curr_idle,