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
 10/* Linker adds these: start and end of __cpuidle functions */
 11extern char __cpuidle_text_start[], __cpuidle_text_end[];
 12
 13/**
 14 * sched_idle_set_state - Record idle state for the current CPU.
 15 * @idle_state: State to record.
 16 */
 17void sched_idle_set_state(struct cpuidle_state *idle_state)
 18{
 19	idle_set_state(this_rq(), idle_state);
 20}
 21
 22static int __read_mostly cpu_idle_force_poll;
 23
 24void cpu_idle_poll_ctrl(bool enable)
 25{
 26	if (enable) {
 27		cpu_idle_force_poll++;
 28	} else {
 29		cpu_idle_force_poll--;
 30		WARN_ON_ONCE(cpu_idle_force_poll < 0);
 31	}
 32}
 33
 34#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
 35static int __init cpu_idle_poll_setup(char *__unused)
 36{
 37	cpu_idle_force_poll = 1;
 38
 39	return 1;
 40}
 41__setup("nohlt", cpu_idle_poll_setup);
 42
 43static int __init cpu_idle_nopoll_setup(char *__unused)
 44{
 45	cpu_idle_force_poll = 0;
 46
 47	return 1;
 48}
 49__setup("hlt", cpu_idle_nopoll_setup);
 50#endif
 51
 52static noinline int __cpuidle cpu_idle_poll(void)
 53{
 54	trace_cpu_idle(0, smp_processor_id());
 55	stop_critical_timings();
 56	ct_idle_enter();
 57	local_irq_enable();
 58
 59	while (!tif_need_resched() &&
 60	       (cpu_idle_force_poll || tick_check_broadcast_expired()))
 61		cpu_relax();
 62
 63	ct_idle_exit();
 64	start_critical_timings();
 65	trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 66
 67	return 1;
 68}
 69
 70/* Weak implementations for optional arch specific functions */
 71void __weak arch_cpu_idle_prepare(void) { }
 72void __weak arch_cpu_idle_enter(void) { }
 73void __weak arch_cpu_idle_exit(void) { }
 74void __weak arch_cpu_idle_dead(void) { }
 75void __weak arch_cpu_idle(void)
 76{
 77	cpu_idle_force_poll = 1;
 78	raw_local_irq_enable();
 79}
 80
 81/**
 82 * default_idle_call - Default CPU idle routine.
 83 *
 84 * To use when the cpuidle framework cannot be used.
 85 */
 86void __cpuidle default_idle_call(void)
 87{
 88	if (current_clr_polling_and_test()) {
 89		local_irq_enable();
 90	} else {
 91
 92		trace_cpu_idle(1, smp_processor_id());
 93		stop_critical_timings();
 94
 95		/*
 96		 * arch_cpu_idle() is supposed to enable IRQs, however
 97		 * we can't do that because of RCU and tracing.
 98		 *
 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.
103		 */
104		trace_hardirqs_on_prepare();
105		lockdep_hardirqs_on_prepare();
106		ct_idle_enter();
107		lockdep_hardirqs_on(_THIS_IP_);
108
109		arch_cpu_idle();
110
111		/*
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
115		 * make it 'work'.
116		 */
117		raw_local_irq_disable();
118		lockdep_hardirqs_off(_THIS_IP_);
119		ct_idle_exit();
120		lockdep_hardirqs_on(_THIS_IP_);
121		raw_local_irq_enable();
122
123		start_critical_timings();
124		trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
125	}
126}
127
128static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
129			       struct cpuidle_device *dev)
130{
131	if (current_clr_polling_and_test())
132		return -EBUSY;
133
134	return cpuidle_enter_s2idle(drv, dev);
135}
136
137static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
138		      int next_state)
139{
140	/*
141	 * The idle task must be scheduled, it is pointless to go to idle, just
142	 * update no idle residency and return.
143	 */
144	if (current_clr_polling_and_test()) {
145		dev->last_residency_ns = 0;
146		local_irq_enable();
147		return -EBUSY;
148	}
149
150	/*
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
154	 */
155	return cpuidle_enter(drv, dev, next_state);
156}
157
158/**
159 * cpuidle_idle_call - the main idle function
160 *
161 * NOTE: no locks or semaphores should be used here
162 *
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.
166 */
167static void cpuidle_idle_call(void)
168{
169	struct cpuidle_device *dev = cpuidle_get_device();
170	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
171	int next_state, entered_state;
172
173	/*
174	 * Check if the idle task must be rescheduled. If it is the
175	 * case, exit the function after re-enabling the local irq.
176	 */
177	if (need_resched()) {
178		local_irq_enable();
179		return;
180	}
181
182	/*
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
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
210			entered_state = call_cpuidle_s2idle(drv, dev);
211			if (entered_state > 0)
212				goto exit_idle;
213
214			max_latency_ns = U64_MAX;
215		} else {
216			max_latency_ns = dev->forced_idle_latency_limit_ns;
217		}
218
219		tick_nohz_idle_stop_tick();
220
221		next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
222		call_cpuidle(drv, dev, next_state);
223	} else {
224		bool stop_tick = true;
225
226		/*
227		 * Ask the cpuidle framework to choose a convenient idle state.
228		 */
229		next_state = cpuidle_select(drv, dev, &stop_tick);
230
231		if (stop_tick || tick_nohz_tick_stopped())
232			tick_nohz_idle_stop_tick();
233		else
234			tick_nohz_idle_retain_tick();
235
236		entered_state = call_cpuidle(drv, dev, next_state);
237		/*
238		 * Give the governor an opportunity to reflect on the outcome
239		 */
240		cpuidle_reflect(dev, entered_state);
241	}
242
243exit_idle:
244	__current_set_polling();
245
246	/*
247	 * It is up to the idle functions to reenable local interrupts
248	 */
249	if (WARN_ON_ONCE(irqs_disabled()))
250		local_irq_enable();
251}
252
253/*
254 * Generic idle loop implementation
255 *
256 * Called with polling cleared.
257 */
258static void do_idle(void)
259{
260	int cpu = smp_processor_id();
261
262	/*
263	 * Check if we need to update blocked load
264	 */
265	nohz_run_idle_balance(cpu);
266
267	/*
268	 * If the arch has a polling bit, we maintain an invariant:
269	 *
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
273	 * reschedule.
274	 */
275
276	__current_set_polling();
277	tick_nohz_idle_enter();
278
279	while (!need_resched()) {
280		rmb();
281
282		local_irq_disable();
283
284		if (cpu_is_offline(cpu)) {
285			tick_nohz_idle_stop_tick();
286			cpuhp_report_idle_dead();
287			arch_cpu_idle_dead();
288		}
289
290		arch_cpu_idle_enter();
291		rcu_nocb_flush_deferred_wakeup();
292
293		/*
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.
298		 */
299		if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
300			tick_nohz_idle_restart_tick();
301			cpu_idle_poll();
302		} else {
303			cpuidle_idle_call();
304		}
305		arch_cpu_idle_exit();
306	}
307
308	/*
309	 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
310	 * be set, propagate it into PREEMPT_NEED_RESCHED.
311	 *
312	 * This is required because for polling idle loops we will not have had
313	 * an IPI to fold the state for us.
314	 */
315	preempt_set_need_resched();
316	tick_nohz_idle_exit();
317	__current_clr_polling();
318
319	/*
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.
323	 */
324	smp_mb__after_atomic();
325
326	/*
327	 * RCU relies on this call to be done outside of an RCU read-side
328	 * critical section.
329	 */
330	flush_smp_call_function_queue();
331	schedule_idle();
332
333	if (unlikely(klp_patch_pending(current)))
334		klp_update_patch_state(current);
335}
336
337bool cpu_in_idle(unsigned long pc)
338{
339	return pc >= (unsigned long)__cpuidle_text_start &&
340		pc < (unsigned long)__cpuidle_text_end;
341}
342
343struct idle_timer {
344	struct hrtimer timer;
345	int done;
346};
347
348static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
349{
350	struct idle_timer *it = container_of(timer, struct idle_timer, timer);
351
352	WRITE_ONCE(it->done, 1);
353	set_tsk_need_resched(current);
354
355	return HRTIMER_NORESTART;
356}
357
358void play_idle_precise(u64 duration_ns, u64 latency_ns)
359{
360	struct idle_timer it;
361
362	/*
363	 * Only FIFO tasks can disable the tick since they don't need the forced
364	 * preemption.
365	 */
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);
372
373	rcu_sleep_check();
374	preempt_disable();
375	current->flags |= PF_IDLE;
376	cpuidle_use_deepest_state(latency_ns);
377
378	it.done = 0;
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);
383
384	while (!READ_ONCE(it.done))
385		do_idle();
386
387	cpuidle_use_deepest_state(0);
388	current->flags &= ~PF_IDLE;
389
390	preempt_fold_need_resched();
391	preempt_enable();
392}
393EXPORT_SYMBOL_GPL(play_idle_precise);
394
395void cpu_startup_entry(enum cpuhp_state state)
396{
397	arch_cpu_idle_prepare();
398	cpuhp_online_idle(state);
399	while (1)
400		do_idle();
401}
402
403/*
404 * idle-task scheduling class.
405 */
406
407#ifdef CONFIG_SMP
408static int
409select_task_rq_idle(struct task_struct *p, int cpu, int flags)
410{
411	return task_cpu(p); /* IDLE tasks as never migrated */
412}
413
414static int
415balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
416{
417	return WARN_ON_ONCE(1);
418}
419#endif
420
421/*
422 * Idle tasks are unconditionally rescheduled:
423 */
424static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
425{
426	resched_curr(rq);
427}
428
429static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
430{
431}
432
433static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
434{
435	update_idle_core(rq);
436	schedstat_inc(rq->sched_goidle);
437}
438
439#ifdef CONFIG_SMP
440static struct task_struct *pick_task_idle(struct rq *rq)
441{
442	return rq->idle;
443}
444#endif
445
446struct task_struct *pick_next_task_idle(struct rq *rq)
447{
448	struct task_struct *next = rq->idle;
449
450	set_next_task_idle(rq, next, true);
451
452	return next;
453}
454
455/*
456 * It is not legal to sleep in the idle task - print a warning
457 * message if some code attempts to do it:
458 */
459static void
460dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
461{
462	raw_spin_rq_unlock_irq(rq);
463	printk(KERN_ERR "bad: scheduling from the idle thread!\n");
464	dump_stack();
465	raw_spin_rq_lock_irq(rq);
466}
467
468/*
469 * scheduler tick hitting a task of our scheduling class.
470 *
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
474 * parameters.
475 */
476static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
477{
478}
479
480static void switched_to_idle(struct rq *rq, struct task_struct *p)
481{
482	BUG();
483}
484
485static void
486prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
487{
488	BUG();
489}
490
491static void update_curr_idle(struct rq *rq)
492{
493}
494
495/*
496 * Simple, special scheduling class for the per-CPU idle tasks:
497 */
498DEFINE_SCHED_CLASS(idle) = {
499
500	/* no enqueue/yield_task for idle tasks */
501
502	/* dequeue is not valid, we print a debug message there: */
503	.dequeue_task		= dequeue_task_idle,
504
505	.check_preempt_curr	= check_preempt_curr_idle,
506
507	.pick_next_task		= pick_next_task_idle,
508	.put_prev_task		= put_prev_task_idle,
509	.set_next_task          = set_next_task_idle,
510
511#ifdef CONFIG_SMP
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,
516#endif
517
518	.task_tick		= task_tick_idle,
519
520	.prio_changed		= prio_changed_idle,
521	.switched_to		= switched_to_idle,
522	.update_curr		= update_curr_idle,
523};