Linux Audio

Check our new training course

Loading...
v4.17
 
  1/*
  2 * Generic entry points for the idle threads and
  3 * implementation of the idle task scheduling class.
  4 *
  5 * (NOTE: these are not related to SCHED_IDLE batch scheduled
  6 *        tasks which are handled in sched/fair.c )
  7 */
  8#include "sched.h"
  9
 10#include <trace/events/power.h>
 11
 12/* Linker adds these: start and end of __cpuidle functions */
 13extern char __cpuidle_text_start[], __cpuidle_text_end[];
 14
 15/**
 16 * sched_idle_set_state - Record idle state for the current CPU.
 17 * @idle_state: State to record.
 18 */
 19void sched_idle_set_state(struct cpuidle_state *idle_state)
 20{
 21	idle_set_state(this_rq(), idle_state);
 22}
 23
 24static int __read_mostly cpu_idle_force_poll;
 25
 26void cpu_idle_poll_ctrl(bool enable)
 27{
 28	if (enable) {
 29		cpu_idle_force_poll++;
 30	} else {
 31		cpu_idle_force_poll--;
 32		WARN_ON_ONCE(cpu_idle_force_poll < 0);
 33	}
 34}
 35
 36#ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
 37static int __init cpu_idle_poll_setup(char *__unused)
 38{
 39	cpu_idle_force_poll = 1;
 40
 41	return 1;
 42}
 43__setup("nohlt", cpu_idle_poll_setup);
 44
 45static int __init cpu_idle_nopoll_setup(char *__unused)
 46{
 47	cpu_idle_force_poll = 0;
 48
 49	return 1;
 50}
 51__setup("hlt", cpu_idle_nopoll_setup);
 52#endif
 53
 54static noinline int __cpuidle cpu_idle_poll(void)
 55{
 56	rcu_idle_enter();
 57	trace_cpu_idle_rcuidle(0, smp_processor_id());
 58	local_irq_enable();
 59	stop_critical_timings();
 
 60
 
 61	while (!tif_need_resched() &&
 62		(cpu_idle_force_poll || tick_check_broadcast_expired()))
 63		cpu_relax();
 
 
 
 64	start_critical_timings();
 65	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
 66	rcu_idle_exit();
 
 67
 68	return 1;
 69}
 70
 71/* Weak implementations for optional arch specific functions */
 72void __weak arch_cpu_idle_prepare(void) { }
 73void __weak arch_cpu_idle_enter(void) { }
 74void __weak arch_cpu_idle_exit(void) { }
 75void __weak arch_cpu_idle_dead(void) { }
 76void __weak arch_cpu_idle(void)
 77{
 78	cpu_idle_force_poll = 1;
 79	local_irq_enable();
 80}
 81
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 82/**
 83 * default_idle_call - Default CPU idle routine.
 84 *
 85 * To use when the cpuidle framework cannot be used.
 86 */
 87void __cpuidle default_idle_call(void)
 88{
 89	if (current_clr_polling_and_test()) {
 90		local_irq_enable();
 91	} else {
 
 92		stop_critical_timings();
 
 
 93		arch_cpu_idle();
 
 
 94		start_critical_timings();
 
 
 95	}
 
 
 
 
 
 
 
 
 
 
 
 96}
 97
 98static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 99		      int next_state)
100{
101	/*
102	 * The idle task must be scheduled, it is pointless to go to idle, just
103	 * update no idle residency and return.
104	 */
105	if (current_clr_polling_and_test()) {
106		dev->last_residency = 0;
107		local_irq_enable();
108		return -EBUSY;
109	}
110
111	/*
112	 * Enter the idle state previously returned by the governor decision.
113	 * This function will block until an interrupt occurs and will take
114	 * care of re-enabling the local interrupts
115	 */
116	return cpuidle_enter(drv, dev, next_state);
117}
118
119/**
120 * cpuidle_idle_call - the main idle function
121 *
122 * NOTE: no locks or semaphores should be used here
123 *
124 * On archs that support TIF_POLLING_NRFLAG, is called with polling
125 * set, and it returns with polling set.  If it ever stops polling, it
126 * must clear the polling bit.
127 */
128static void cpuidle_idle_call(void)
129{
130	struct cpuidle_device *dev = cpuidle_get_device();
131	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
132	int next_state, entered_state;
133
134	/*
135	 * Check if the idle task must be rescheduled. If it is the
136	 * case, exit the function after re-enabling the local irq.
137	 */
138	if (need_resched()) {
139		local_irq_enable();
140		return;
141	}
142
143	/*
144	 * The RCU framework needs to be told that we are entering an idle
145	 * section, so no more rcu read side critical sections and one more
146	 * step to the grace period
147	 */
148
149	if (cpuidle_not_available(drv, dev)) {
150		tick_nohz_idle_stop_tick();
151		rcu_idle_enter();
152
153		default_idle_call();
154		goto exit_idle;
155	}
156
157	/*
158	 * Suspend-to-idle ("s2idle") is a system state in which all user space
159	 * has been frozen, all I/O devices have been suspended and the only
160	 * activity happens here and in iterrupts (if any).  In that case bypass
161	 * the cpuidle governor and go stratight for the deepest idle state
162	 * available.  Possibly also suspend the local tick and the entire
163	 * timekeeping to prevent timer interrupts from kicking us out of idle
164	 * until a proper wakeup interrupt happens.
165	 */
166
167	if (idle_should_enter_s2idle() || dev->use_deepest_state) {
 
 
168		if (idle_should_enter_s2idle()) {
169			rcu_idle_enter();
170
171			entered_state = cpuidle_enter_s2idle(drv, dev);
172			if (entered_state > 0) {
173				local_irq_enable();
174				goto exit_idle;
175			}
176
177			rcu_idle_exit();
 
 
178		}
179
180		tick_nohz_idle_stop_tick();
181		rcu_idle_enter();
182
183		next_state = cpuidle_find_deepest_state(drv, dev);
184		call_cpuidle(drv, dev, next_state);
185	} else {
186		bool stop_tick = true;
187
188		/*
189		 * Ask the cpuidle framework to choose a convenient idle state.
190		 */
191		next_state = cpuidle_select(drv, dev, &stop_tick);
192
193		if (stop_tick)
194			tick_nohz_idle_stop_tick();
195		else
196			tick_nohz_idle_retain_tick();
197
198		rcu_idle_enter();
199
200		entered_state = call_cpuidle(drv, dev, next_state);
201		/*
202		 * Give the governor an opportunity to reflect on the outcome
203		 */
204		cpuidle_reflect(dev, entered_state);
205	}
206
207exit_idle:
208	__current_set_polling();
209
210	/*
211	 * It is up to the idle functions to reenable local interrupts
212	 */
213	if (WARN_ON_ONCE(irqs_disabled()))
214		local_irq_enable();
215
216	rcu_idle_exit();
217}
218
219/*
220 * Generic idle loop implementation
221 *
222 * Called with polling cleared.
223 */
224static void do_idle(void)
225{
226	int cpu = smp_processor_id();
 
 
 
 
 
 
227	/*
228	 * If the arch has a polling bit, we maintain an invariant:
229	 *
230	 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
231	 * rq->idle). This means that, if rq->idle has the polling bit set,
232	 * then setting need_resched is guaranteed to cause the CPU to
233	 * reschedule.
234	 */
235
236	__current_set_polling();
237	tick_nohz_idle_enter();
238
239	while (!need_resched()) {
240		check_pgt_cache();
241		rmb();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
242
243		if (cpu_is_offline(cpu)) {
244			tick_nohz_idle_stop_tick_protected();
245			cpuhp_report_idle_dead();
246			arch_cpu_idle_dead();
247		}
248
249		local_irq_disable();
250		arch_cpu_idle_enter();
 
251
252		/*
253		 * In poll mode we reenable interrupts and spin. Also if we
254		 * detected in the wakeup from idle path that the tick
255		 * broadcast device expired for us, we don't want to go deep
256		 * idle as we know that the IPI is going to arrive right away.
257		 */
258		if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
259			tick_nohz_idle_restart_tick();
260			cpu_idle_poll();
261		} else {
262			cpuidle_idle_call();
263		}
264		arch_cpu_idle_exit();
265	}
266
267	/*
268	 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
269	 * be set, propagate it into PREEMPT_NEED_RESCHED.
270	 *
271	 * This is required because for polling idle loops we will not have had
272	 * an IPI to fold the state for us.
273	 */
274	preempt_set_need_resched();
275	tick_nohz_idle_exit();
276	__current_clr_polling();
277
278	/*
279	 * We promise to call sched_ttwu_pending() and reschedule if
280	 * need_resched() is set while polling is set. That means that clearing
281	 * polling needs to be visible before doing these things.
282	 */
283	smp_mb__after_atomic();
284
285	sched_ttwu_pending();
 
 
 
 
286	schedule_idle();
287
288	if (unlikely(klp_patch_pending(current)))
289		klp_update_patch_state(current);
290}
291
292bool cpu_in_idle(unsigned long pc)
293{
294	return pc >= (unsigned long)__cpuidle_text_start &&
295		pc < (unsigned long)__cpuidle_text_end;
296}
297
298struct idle_timer {
299	struct hrtimer timer;
300	int done;
301};
302
303static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
304{
305	struct idle_timer *it = container_of(timer, struct idle_timer, timer);
306
307	WRITE_ONCE(it->done, 1);
308	set_tsk_need_resched(current);
309
310	return HRTIMER_NORESTART;
311}
312
313void play_idle(unsigned long duration_ms)
314{
315	struct idle_timer it;
316
317	/*
318	 * Only FIFO tasks can disable the tick since they don't need the forced
319	 * preemption.
320	 */
321	WARN_ON_ONCE(current->policy != SCHED_FIFO);
322	WARN_ON_ONCE(current->nr_cpus_allowed != 1);
323	WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
324	WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
325	WARN_ON_ONCE(!duration_ms);
 
326
327	rcu_sleep_check();
328	preempt_disable();
329	current->flags |= PF_IDLE;
330	cpuidle_use_deepest_state(true);
331
332	it.done = 0;
333	hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
334	it.timer.function = idle_inject_timer_fn;
335	hrtimer_start(&it.timer, ms_to_ktime(duration_ms), HRTIMER_MODE_REL_PINNED);
 
336
337	while (!READ_ONCE(it.done))
338		do_idle();
339
340	cpuidle_use_deepest_state(false);
341	current->flags &= ~PF_IDLE;
342
343	preempt_fold_need_resched();
344	preempt_enable();
345}
346EXPORT_SYMBOL_GPL(play_idle);
347
348void cpu_startup_entry(enum cpuhp_state state)
349{
350	/*
351	 * This #ifdef needs to die, but it's too late in the cycle to
352	 * make this generic (ARM and SH have never invoked the canary
353	 * init for the non boot CPUs!). Will be fixed in 3.11
354	 */
355#ifdef CONFIG_X86
356	/*
357	 * If we're the non-boot CPU, nothing set the stack canary up
358	 * for us. The boot CPU already has it initialized but no harm
359	 * in doing it again. This is a good place for updating it, as
360	 * we wont ever return from this function (so the invalid
361	 * canaries already on the stack wont ever trigger).
362	 */
363	boot_init_stack_canary();
364#endif
365	arch_cpu_idle_prepare();
366	cpuhp_online_idle(state);
367	while (1)
368		do_idle();
369}
370
371/*
372 * idle-task scheduling class.
373 */
374
375#ifdef CONFIG_SMP
376static int
377select_task_rq_idle(struct task_struct *p, int cpu, int sd_flag, int flags)
378{
379	return task_cpu(p); /* IDLE tasks as never migrated */
380}
 
 
 
 
 
 
381#endif
382
383/*
384 * Idle tasks are unconditionally rescheduled:
385 */
386static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int flags)
387{
388	resched_curr(rq);
389}
390
391static struct task_struct *
392pick_next_task_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 
 
 
 
 
393{
394	put_prev_task(rq, prev);
395	update_idle_core(rq);
 
396	schedstat_inc(rq->sched_goidle);
 
 
397
 
 
 
398	return rq->idle;
399}
400
401/*
402 * It is not legal to sleep in the idle task - print a warning
403 * message if some code attempts to do it:
404 */
405static void
406dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
407{
408	raw_spin_unlock_irq(&rq->lock);
409	printk(KERN_ERR "bad: scheduling from the idle thread!\n");
410	dump_stack();
411	raw_spin_lock_irq(&rq->lock);
412}
413
414static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
415{
416}
417
418/*
419 * scheduler tick hitting a task of our scheduling class.
420 *
421 * NOTE: This function can be called remotely by the tick offload that
422 * goes along full dynticks. Therefore no local assumption can be made
423 * and everything must be accessed through the @rq and @curr passed in
424 * parameters.
425 */
426static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
427{
428}
429
430static void set_curr_task_idle(struct rq *rq)
431{
432}
433
434static void switched_to_idle(struct rq *rq, struct task_struct *p)
435{
436	BUG();
437}
438
439static void
440prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
441{
442	BUG();
443}
444
445static unsigned int get_rr_interval_idle(struct rq *rq, struct task_struct *task)
446{
447	return 0;
448}
449
450static void update_curr_idle(struct rq *rq)
451{
452}
453
454/*
455 * Simple, special scheduling class for the per-CPU idle tasks:
456 */
457const struct sched_class idle_sched_class = {
458	/* .next is NULL */
459	/* no enqueue/yield_task for idle tasks */
460
461	/* dequeue is not valid, we print a debug message there: */
462	.dequeue_task		= dequeue_task_idle,
463
464	.check_preempt_curr	= check_preempt_curr_idle,
465
466	.pick_next_task		= pick_next_task_idle,
467	.put_prev_task		= put_prev_task_idle,
 
468
469#ifdef CONFIG_SMP
 
470	.select_task_rq		= select_task_rq_idle,
471	.set_cpus_allowed	= set_cpus_allowed_common,
472#endif
473
474	.set_curr_task          = set_curr_task_idle,
475	.task_tick		= task_tick_idle,
476
477	.get_rr_interval	= get_rr_interval_idle,
478
479	.prio_changed		= prio_changed_idle,
480	.switched_to		= switched_to_idle,
481	.update_curr		= update_curr_idle,
482};
v6.13.7
  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	instrumentation_begin();
 55	trace_cpu_idle(0, smp_processor_id());
 
 56	stop_critical_timings();
 57	ct_cpuidle_enter();
 58
 59	raw_local_irq_enable();
 60	while (!tif_need_resched() &&
 61	       (cpu_idle_force_poll || tick_check_broadcast_expired()))
 62		cpu_relax();
 63	raw_local_irq_disable();
 64
 65	ct_cpuidle_exit();
 66	start_critical_timings();
 67	trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 68	local_irq_enable();
 69	instrumentation_end();
 70
 71	return 1;
 72}
 73
 74/* Weak implementations for optional arch specific functions */
 75void __weak arch_cpu_idle_prepare(void) { }
 76void __weak arch_cpu_idle_enter(void) { }
 77void __weak arch_cpu_idle_exit(void) { }
 78void __weak __noreturn arch_cpu_idle_dead(void) { while (1); }
 79void __weak arch_cpu_idle(void)
 80{
 81	cpu_idle_force_poll = 1;
 
 82}
 83
 84#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST_IDLE
 85DEFINE_STATIC_KEY_FALSE(arch_needs_tick_broadcast);
 86
 87static inline void cond_tick_broadcast_enter(void)
 88{
 89	if (static_branch_unlikely(&arch_needs_tick_broadcast))
 90		tick_broadcast_enter();
 91}
 92
 93static inline void cond_tick_broadcast_exit(void)
 94{
 95	if (static_branch_unlikely(&arch_needs_tick_broadcast))
 96		tick_broadcast_exit();
 97}
 98#else
 99static inline void cond_tick_broadcast_enter(void) { }
100static inline void cond_tick_broadcast_exit(void) { }
101#endif
102
103/**
104 * default_idle_call - Default CPU idle routine.
105 *
106 * To use when the cpuidle framework cannot be used.
107 */
108void __cpuidle default_idle_call(void)
109{
110	instrumentation_begin();
111	if (!current_clr_polling_and_test()) {
112		cond_tick_broadcast_enter();
113		trace_cpu_idle(1, smp_processor_id());
114		stop_critical_timings();
115
116		ct_cpuidle_enter();
117		arch_cpu_idle();
118		ct_cpuidle_exit();
119
120		start_critical_timings();
121		trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
122		cond_tick_broadcast_exit();
123	}
124	local_irq_enable();
125	instrumentation_end();
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	if (cpuidle_not_available(drv, dev)) {
183		tick_nohz_idle_stop_tick();
 
184
185		default_idle_call();
186		goto exit_idle;
187	}
188
189	/*
190	 * Suspend-to-idle ("s2idle") is a system state in which all user space
191	 * has been frozen, all I/O devices have been suspended and the only
192	 * activity happens here and in interrupts (if any). In that case bypass
193	 * the cpuidle governor and go straight for the deepest idle state
194	 * available.  Possibly also suspend the local tick and the entire
195	 * timekeeping to prevent timer interrupts from kicking us out of idle
196	 * until a proper wakeup interrupt happens.
197	 */
198
199	if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
200		u64 max_latency_ns;
201
202		if (idle_should_enter_s2idle()) {
 
203
204			entered_state = call_cpuidle_s2idle(drv, dev);
205			if (entered_state > 0)
 
206				goto exit_idle;
 
207
208			max_latency_ns = U64_MAX;
209		} else {
210			max_latency_ns = dev->forced_idle_latency_limit_ns;
211		}
212
213		tick_nohz_idle_stop_tick();
 
214
215		next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
216		call_cpuidle(drv, dev, next_state);
217	} else {
218		bool stop_tick = true;
219
220		/*
221		 * Ask the cpuidle framework to choose a convenient idle state.
222		 */
223		next_state = cpuidle_select(drv, dev, &stop_tick);
224
225		if (stop_tick || tick_nohz_tick_stopped())
226			tick_nohz_idle_stop_tick();
227		else
228			tick_nohz_idle_retain_tick();
229
 
 
230		entered_state = call_cpuidle(drv, dev, next_state);
231		/*
232		 * Give the governor an opportunity to reflect on the outcome
233		 */
234		cpuidle_reflect(dev, entered_state);
235	}
236
237exit_idle:
238	__current_set_polling();
239
240	/*
241	 * It is up to the idle functions to re-enable local interrupts
242	 */
243	if (WARN_ON_ONCE(irqs_disabled()))
244		local_irq_enable();
 
 
245}
246
247/*
248 * Generic idle loop implementation
249 *
250 * Called with polling cleared.
251 */
252static void do_idle(void)
253{
254	int cpu = smp_processor_id();
255
256	/*
257	 * Check if we need to update blocked load
258	 */
259	nohz_run_idle_balance(cpu);
260
261	/*
262	 * If the arch has a polling bit, we maintain an invariant:
263	 *
264	 * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
265	 * rq->idle). This means that, if rq->idle has the polling bit set,
266	 * then setting need_resched is guaranteed to cause the CPU to
267	 * reschedule.
268	 */
269
270	__current_set_polling();
271	tick_nohz_idle_enter();
272
273	while (!need_resched()) {
274
275		/*
276		 * Interrupts shouldn't be re-enabled from that point on until
277		 * the CPU sleeping instruction is reached. Otherwise an interrupt
278		 * may fire and queue a timer that would be ignored until the CPU
279		 * wakes from the sleeping instruction. And testing need_resched()
280		 * doesn't tell about pending needed timer reprogram.
281		 *
282		 * Several cases to consider:
283		 *
284		 * - SLEEP-UNTIL-PENDING-INTERRUPT based instructions such as
285		 *   "wfi" or "mwait" are fine because they can be entered with
286		 *   interrupt disabled.
287		 *
288		 * - sti;mwait() couple is fine because the interrupts are
289		 *   re-enabled only upon the execution of mwait, leaving no gap
290		 *   in-between.
291		 *
292		 * - ROLLBACK based idle handlers with the sleeping instruction
293		 *   called with interrupts enabled are NOT fine. In this scheme
294		 *   when the interrupt detects it has interrupted an idle handler,
295		 *   it rolls back to its beginning which performs the
296		 *   need_resched() check before re-executing the sleeping
297		 *   instruction. This can leak a pending needed timer reprogram.
298		 *   If such a scheme is really mandatory due to the lack of an
299		 *   appropriate CPU sleeping instruction, then a FAST-FORWARD
300		 *   must instead be applied: when the interrupt detects it has
301		 *   interrupted an idle handler, it must resume to the end of
302		 *   this idle handler so that the generic idle loop is iterated
303		 *   again to reprogram the tick.
304		 */
305		local_irq_disable();
306
307		if (cpu_is_offline(cpu)) {
 
308			cpuhp_report_idle_dead();
309			arch_cpu_idle_dead();
310		}
311
 
312		arch_cpu_idle_enter();
313		rcu_nocb_flush_deferred_wakeup();
314
315		/*
316		 * In poll mode we re-enable interrupts and spin. Also if we
317		 * detected in the wakeup from idle path that the tick
318		 * broadcast device expired for us, we don't want to go deep
319		 * idle as we know that the IPI is going to arrive right away.
320		 */
321		if (cpu_idle_force_poll || tick_check_broadcast_expired()) {
322			tick_nohz_idle_restart_tick();
323			cpu_idle_poll();
324		} else {
325			cpuidle_idle_call();
326		}
327		arch_cpu_idle_exit();
328	}
329
330	/*
331	 * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
332	 * be set, propagate it into PREEMPT_NEED_RESCHED.
333	 *
334	 * This is required because for polling idle loops we will not have had
335	 * an IPI to fold the state for us.
336	 */
337	preempt_set_need_resched();
338	tick_nohz_idle_exit();
339	__current_clr_polling();
340
341	/*
342	 * We promise to call sched_ttwu_pending() and reschedule if
343	 * need_resched() is set while polling is set. That means that clearing
344	 * polling needs to be visible before doing these things.
345	 */
346	smp_mb__after_atomic();
347
348	/*
349	 * RCU relies on this call to be done outside of an RCU read-side
350	 * critical section.
351	 */
352	flush_smp_call_function_queue();
353	schedule_idle();
354
355	if (unlikely(klp_patch_pending(current)))
356		klp_update_patch_state(current);
357}
358
359bool cpu_in_idle(unsigned long pc)
360{
361	return pc >= (unsigned long)__cpuidle_text_start &&
362		pc < (unsigned long)__cpuidle_text_end;
363}
364
365struct idle_timer {
366	struct hrtimer timer;
367	int done;
368};
369
370static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
371{
372	struct idle_timer *it = container_of(timer, struct idle_timer, timer);
373
374	WRITE_ONCE(it->done, 1);
375	set_tsk_need_resched(current);
376
377	return HRTIMER_NORESTART;
378}
379
380void play_idle_precise(u64 duration_ns, u64 latency_ns)
381{
382	struct idle_timer it;
383
384	/*
385	 * Only FIFO tasks can disable the tick since they don't need the forced
386	 * preemption.
387	 */
388	WARN_ON_ONCE(current->policy != SCHED_FIFO);
389	WARN_ON_ONCE(current->nr_cpus_allowed != 1);
390	WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
391	WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
392	WARN_ON_ONCE(!duration_ns);
393	WARN_ON_ONCE(current->mm);
394
395	rcu_sleep_check();
396	preempt_disable();
397	current->flags |= PF_IDLE;
398	cpuidle_use_deepest_state(latency_ns);
399
400	it.done = 0;
401	hrtimer_setup_on_stack(&it.timer, idle_inject_timer_fn, CLOCK_MONOTONIC,
402			       HRTIMER_MODE_REL_HARD);
403	hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
404		      HRTIMER_MODE_REL_PINNED_HARD);
405
406	while (!READ_ONCE(it.done))
407		do_idle();
408
409	cpuidle_use_deepest_state(0);
410	current->flags &= ~PF_IDLE;
411
412	preempt_fold_need_resched();
413	preempt_enable();
414}
415EXPORT_SYMBOL_GPL(play_idle_precise);
416
417void cpu_startup_entry(enum cpuhp_state state)
418{
419	current->flags |= PF_IDLE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
420	arch_cpu_idle_prepare();
421	cpuhp_online_idle(state);
422	while (1)
423		do_idle();
424}
425
426/*
427 * idle-task scheduling class.
428 */
429
430#ifdef CONFIG_SMP
431static int
432select_task_rq_idle(struct task_struct *p, int cpu, int flags)
433{
434	return task_cpu(p); /* IDLE tasks as never migrated */
435}
436
437static int
438balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
439{
440	return WARN_ON_ONCE(1);
441}
442#endif
443
444/*
445 * Idle tasks are unconditionally rescheduled:
446 */
447static void wakeup_preempt_idle(struct rq *rq, struct task_struct *p, int flags)
448{
449	resched_curr(rq);
450}
451
452static void put_prev_task_idle(struct rq *rq, struct task_struct *prev, struct task_struct *next)
453{
454	dl_server_update_idle_time(rq, prev);
455	scx_update_idle(rq, false, true);
456}
457
458static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
459{
 
460	update_idle_core(rq);
461	scx_update_idle(rq, true, true);
462	schedstat_inc(rq->sched_goidle);
463	next->se.exec_start = rq_clock_task(rq);
464}
465
466struct task_struct *pick_task_idle(struct rq *rq)
467{
468	scx_update_idle(rq, true, false);
469	return rq->idle;
470}
471
472/*
473 * It is not legal to sleep in the idle task - print a warning
474 * message if some code attempts to do it:
475 */
476static bool
477dequeue_task_idle(struct rq *rq, struct task_struct *p, int flags)
478{
479	raw_spin_rq_unlock_irq(rq);
480	printk(KERN_ERR "bad: scheduling from the idle thread!\n");
481	dump_stack();
482	raw_spin_rq_lock_irq(rq);
483	return true;
 
 
 
484}
485
486/*
487 * scheduler tick hitting a task of our scheduling class.
488 *
489 * NOTE: This function can be called remotely by the tick offload that
490 * goes along full dynticks. Therefore no local assumption can be made
491 * and everything must be accessed through the @rq and @curr passed in
492 * parameters.
493 */
494static void task_tick_idle(struct rq *rq, struct task_struct *curr, int queued)
495{
496}
497
 
 
 
 
498static void switched_to_idle(struct rq *rq, struct task_struct *p)
499{
500	BUG();
501}
502
503static void
504prio_changed_idle(struct rq *rq, struct task_struct *p, int oldprio)
505{
506	BUG();
507}
508
 
 
 
 
 
509static void update_curr_idle(struct rq *rq)
510{
511}
512
513/*
514 * Simple, special scheduling class for the per-CPU idle tasks:
515 */
516DEFINE_SCHED_CLASS(idle) = {
517
518	/* no enqueue/yield_task for idle tasks */
519
520	/* dequeue is not valid, we print a debug message there: */
521	.dequeue_task		= dequeue_task_idle,
522
523	.wakeup_preempt		= wakeup_preempt_idle,
524
525	.pick_task		= pick_task_idle,
526	.put_prev_task		= put_prev_task_idle,
527	.set_next_task          = set_next_task_idle,
528
529#ifdef CONFIG_SMP
530	.balance		= balance_idle,
531	.select_task_rq		= select_task_rq_idle,
532	.set_cpus_allowed	= set_cpus_allowed_common,
533#endif
534
 
535	.task_tick		= task_tick_idle,
 
 
536
537	.prio_changed		= prio_changed_idle,
538	.switched_to		= switched_to_idle,
539	.update_curr		= update_curr_idle,
540};