1 /*
2  * cpuidle.c - core cpuidle infrastructure
3  *
4  * (C) 2006-2007 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5  *               Shaohua Li <shaohua.li@intel.com>
6  *               Adam Belay <abelay@novell.com>
7  *
8  * This code is licenced under the GPL.
9  */
10 
11 #include "linux/percpu-defs.h"
12 #include <linux/clockchips.h>
13 #include <linux/kernel.h>
14 #include <linux/mutex.h>
15 #include <linux/sched.h>
16 #include <linux/sched/clock.h>
17 #include <linux/notifier.h>
18 #include <linux/pm_qos.h>
19 #include <linux/cpu.h>
20 #include <linux/cpuidle.h>
21 #include <linux/ktime.h>
22 #include <linux/hrtimer.h>
23 #include <linux/module.h>
24 #include <linux/suspend.h>
25 #include <linux/tick.h>
26 #include <linux/mmu_context.h>
27 #include <linux/context_tracking.h>
28 #include <trace/events/power.h>
29 
30 #include "cpuidle.h"
31 
32 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
33 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
34 
35 DEFINE_MUTEX(cpuidle_lock);
36 LIST_HEAD(cpuidle_detected_devices);
37 
38 static int enabled_devices;
39 static int off __read_mostly;
40 static int initialized __read_mostly;
41 
cpuidle_disabled(void)42 int cpuidle_disabled(void)
43 {
44 	return off;
45 }
disable_cpuidle(void)46 void disable_cpuidle(void)
47 {
48 	off = 1;
49 }
50 
cpuidle_not_available(struct cpuidle_driver * drv,struct cpuidle_device * dev)51 bool cpuidle_not_available(struct cpuidle_driver *drv,
52 			   struct cpuidle_device *dev)
53 {
54 	return off || !initialized || !drv || !dev || !dev->enabled;
55 }
56 
57 /**
58  * cpuidle_play_dead - cpu off-lining
59  *
60  * Returns in case of an error or no driver
61  */
cpuidle_play_dead(void)62 int cpuidle_play_dead(void)
63 {
64 	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
65 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
66 	int i;
67 
68 	if (!drv)
69 		return -ENODEV;
70 
71 	/* Find lowest-power state that supports long-term idle */
72 	for (i = drv->state_count - 1; i >= 0; i--)
73 		if (drv->states[i].enter_dead)
74 			return drv->states[i].enter_dead(dev, i);
75 
76 	return -ENODEV;
77 }
78 
find_deepest_state(struct cpuidle_driver * drv,struct cpuidle_device * dev,u64 max_latency_ns,unsigned int forbidden_flags,bool s2idle)79 static int find_deepest_state(struct cpuidle_driver *drv,
80 			      struct cpuidle_device *dev,
81 			      u64 max_latency_ns,
82 			      unsigned int forbidden_flags,
83 			      bool s2idle)
84 {
85 	u64 latency_req = 0;
86 	int i, ret = 0;
87 
88 	for (i = 1; i < drv->state_count; i++) {
89 		struct cpuidle_state *s = &drv->states[i];
90 
91 		if (dev->states_usage[i].disable ||
92 		    s->exit_latency_ns <= latency_req ||
93 		    s->exit_latency_ns > max_latency_ns ||
94 		    (s->flags & forbidden_flags) ||
95 		    (s2idle && !s->enter_s2idle))
96 			continue;
97 
98 		latency_req = s->exit_latency_ns;
99 		ret = i;
100 	}
101 	return ret;
102 }
103 
104 /**
105  * cpuidle_use_deepest_state - Set/unset governor override mode.
106  * @latency_limit_ns: Idle state exit latency limit (or no override if 0).
107  *
108  * If @latency_limit_ns is nonzero, set the current CPU to use the deepest idle
109  * state with exit latency within @latency_limit_ns (override governors going
110  * forward), or do not override governors if it is zero.
111  */
cpuidle_use_deepest_state(u64 latency_limit_ns)112 void cpuidle_use_deepest_state(u64 latency_limit_ns)
113 {
114 	struct cpuidle_device *dev;
115 
116 	preempt_disable();
117 	dev = cpuidle_get_device();
118 	if (dev)
119 		dev->forced_idle_latency_limit_ns = latency_limit_ns;
120 	preempt_enable();
121 }
122 
123 /**
124  * cpuidle_find_deepest_state - Find the deepest available idle state.
125  * @drv: cpuidle driver for the given CPU.
126  * @dev: cpuidle device for the given CPU.
127  * @latency_limit_ns: Idle state exit latency limit
128  *
129  * Return: the index of the deepest available idle state.
130  */
cpuidle_find_deepest_state(struct cpuidle_driver * drv,struct cpuidle_device * dev,u64 latency_limit_ns)131 int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
132 			       struct cpuidle_device *dev,
133 			       u64 latency_limit_ns)
134 {
135 	return find_deepest_state(drv, dev, latency_limit_ns, 0, false);
136 }
137 
138 #ifdef CONFIG_SUSPEND
enter_s2idle_proper(struct cpuidle_driver * drv,struct cpuidle_device * dev,int index)139 static void enter_s2idle_proper(struct cpuidle_driver *drv,
140 				struct cpuidle_device *dev, int index)
141 {
142 	ktime_t time_start, time_end;
143 	struct cpuidle_state *target_state = &drv->states[index];
144 
145 	time_start = ns_to_ktime(local_clock());
146 
147 	tick_freeze();
148 	/*
149 	 * The state used here cannot be a "coupled" one, because the "coupled"
150 	 * cpuidle mechanism enables interrupts and doing that with timekeeping
151 	 * suspended is generally unsafe.
152 	 */
153 	stop_critical_timings();
154 	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
155 		ct_idle_enter();
156 	target_state->enter_s2idle(dev, drv, index);
157 	if (WARN_ON_ONCE(!irqs_disabled()))
158 		local_irq_disable();
159 	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
160 		ct_idle_exit();
161 	tick_unfreeze();
162 	start_critical_timings();
163 
164 	time_end = ns_to_ktime(local_clock());
165 
166 	dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
167 	dev->states_usage[index].s2idle_usage++;
168 }
169 
170 /**
171  * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
172  * @drv: cpuidle driver for the given CPU.
173  * @dev: cpuidle device for the given CPU.
174  *
175  * If there are states with the ->enter_s2idle callback, find the deepest of
176  * them and enter it with frozen tick.
177  */
cpuidle_enter_s2idle(struct cpuidle_driver * drv,struct cpuidle_device * dev)178 int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
179 {
180 	int index;
181 
182 	/*
183 	 * Find the deepest state with ->enter_s2idle present, which guarantees
184 	 * that interrupts won't be enabled when it exits and allows the tick to
185 	 * be frozen safely.
186 	 */
187 	index = find_deepest_state(drv, dev, U64_MAX, 0, true);
188 	if (index > 0) {
189 		enter_s2idle_proper(drv, dev, index);
190 		local_irq_enable();
191 	}
192 	return index;
193 }
194 #endif /* CONFIG_SUSPEND */
195 
196 /**
197  * cpuidle_enter_state - enter the state and update stats
198  * @dev: cpuidle device for this cpu
199  * @drv: cpuidle driver for this cpu
200  * @index: index into the states table in @drv of the state to enter
201  */
cpuidle_enter_state(struct cpuidle_device * dev,struct cpuidle_driver * drv,int index)202 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
203 			int index)
204 {
205 	int entered_state;
206 
207 	struct cpuidle_state *target_state = &drv->states[index];
208 	bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
209 	ktime_t time_start, time_end;
210 
211 	/*
212 	 * Tell the time framework to switch to a broadcast timer because our
213 	 * local timer will be shut down.  If a local timer is used from another
214 	 * CPU as a broadcast timer, this call may fail if it is not available.
215 	 */
216 	if (broadcast && tick_broadcast_enter()) {
217 		index = find_deepest_state(drv, dev, target_state->exit_latency_ns,
218 					   CPUIDLE_FLAG_TIMER_STOP, false);
219 		if (index < 0) {
220 			default_idle_call();
221 			return -EBUSY;
222 		}
223 		target_state = &drv->states[index];
224 		broadcast = false;
225 	}
226 
227 	if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
228 		leave_mm(dev->cpu);
229 
230 	/* Take note of the planned idle state. */
231 	sched_idle_set_state(target_state);
232 
233 	trace_cpu_idle(index, dev->cpu);
234 	time_start = ns_to_ktime(local_clock());
235 
236 	stop_critical_timings();
237 	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
238 		ct_idle_enter();
239 	entered_state = target_state->enter(dev, drv, index);
240 	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE))
241 		ct_idle_exit();
242 	start_critical_timings();
243 
244 	sched_clock_idle_wakeup_event();
245 	time_end = ns_to_ktime(local_clock());
246 	trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
247 
248 	/* The cpu is no longer idle or about to enter idle. */
249 	sched_idle_set_state(NULL);
250 
251 	if (broadcast) {
252 		if (WARN_ON_ONCE(!irqs_disabled()))
253 			local_irq_disable();
254 
255 		tick_broadcast_exit();
256 	}
257 
258 	if (!cpuidle_state_is_coupled(drv, index))
259 		local_irq_enable();
260 
261 	if (entered_state >= 0) {
262 		s64 diff, delay = drv->states[entered_state].exit_latency_ns;
263 		int i;
264 
265 		/*
266 		 * Update cpuidle counters
267 		 * This can be moved to within driver enter routine,
268 		 * but that results in multiple copies of same code.
269 		 */
270 		diff = ktime_sub(time_end, time_start);
271 
272 		dev->last_residency_ns = diff;
273 		dev->states_usage[entered_state].time_ns += diff;
274 		dev->states_usage[entered_state].usage++;
275 
276 		if (diff < drv->states[entered_state].target_residency_ns) {
277 			for (i = entered_state - 1; i >= 0; i--) {
278 				if (dev->states_usage[i].disable)
279 					continue;
280 
281 				/* Shallower states are enabled, so update. */
282 				dev->states_usage[entered_state].above++;
283 				trace_cpu_idle_miss(dev->cpu, entered_state, false);
284 				break;
285 			}
286 		} else if (diff > delay) {
287 			for (i = entered_state + 1; i < drv->state_count; i++) {
288 				if (dev->states_usage[i].disable)
289 					continue;
290 
291 				/*
292 				 * Update if a deeper state would have been a
293 				 * better match for the observed idle duration.
294 				 */
295 				if (diff - delay >= drv->states[i].target_residency_ns) {
296 					dev->states_usage[entered_state].below++;
297 					trace_cpu_idle_miss(dev->cpu, entered_state, true);
298 				}
299 
300 				break;
301 			}
302 		}
303 	} else {
304 		dev->last_residency_ns = 0;
305 		dev->states_usage[index].rejected++;
306 	}
307 
308 	return entered_state;
309 }
310 
311 /**
312  * cpuidle_select - ask the cpuidle framework to choose an idle state
313  *
314  * @drv: the cpuidle driver
315  * @dev: the cpuidle device
316  * @stop_tick: indication on whether or not to stop the tick
317  *
318  * Returns the index of the idle state.  The return value must not be negative.
319  *
320  * The memory location pointed to by @stop_tick is expected to be written the
321  * 'false' boolean value if the scheduler tick should not be stopped before
322  * entering the returned state.
323  */
cpuidle_select(struct cpuidle_driver * drv,struct cpuidle_device * dev,bool * stop_tick)324 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
325 		   bool *stop_tick)
326 {
327 	return cpuidle_curr_governor->select(drv, dev, stop_tick);
328 }
329 
330 /**
331  * cpuidle_enter - enter into the specified idle state
332  *
333  * @drv:   the cpuidle driver tied with the cpu
334  * @dev:   the cpuidle device
335  * @index: the index in the idle state table
336  *
337  * Returns the index in the idle state, < 0 in case of error.
338  * The error code depends on the backend driver
339  */
cpuidle_enter(struct cpuidle_driver * drv,struct cpuidle_device * dev,int index)340 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
341 		  int index)
342 {
343 	int ret = 0;
344 
345 	/*
346 	 * Store the next hrtimer, which becomes either next tick or the next
347 	 * timer event, whatever expires first. Additionally, to make this data
348 	 * useful for consumers outside cpuidle, we rely on that the governor's
349 	 * ->select() callback have decided, whether to stop the tick or not.
350 	 */
351 	WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
352 
353 	if (cpuidle_state_is_coupled(drv, index))
354 		ret = cpuidle_enter_state_coupled(dev, drv, index);
355 	else
356 		ret = cpuidle_enter_state(dev, drv, index);
357 
358 	WRITE_ONCE(dev->next_hrtimer, 0);
359 	return ret;
360 }
361 
362 /**
363  * cpuidle_reflect - tell the underlying governor what was the state
364  * we were in
365  *
366  * @dev  : the cpuidle device
367  * @index: the index in the idle state table
368  *
369  */
cpuidle_reflect(struct cpuidle_device * dev,int index)370 void cpuidle_reflect(struct cpuidle_device *dev, int index)
371 {
372 	if (cpuidle_curr_governor->reflect && index >= 0)
373 		cpuidle_curr_governor->reflect(dev, index);
374 }
375 
376 /*
377  * Min polling interval of 10usec is a guess. It is assuming that
378  * for most users, the time for a single ping-pong workload like
379  * perf bench pipe would generally complete within 10usec but
380  * this is hardware dependant. Actual time can be estimated with
381  *
382  * perf bench sched pipe -l 10000
383  *
384  * Run multiple times to avoid cpufreq effects.
385  */
386 #define CPUIDLE_POLL_MIN 10000
387 #define CPUIDLE_POLL_MAX (TICK_NSEC / 16)
388 
389 /**
390  * cpuidle_poll_time - return amount of time to poll for,
391  * governors can override dev->poll_limit_ns if necessary
392  *
393  * @drv:   the cpuidle driver tied with the cpu
394  * @dev:   the cpuidle device
395  *
396  */
cpuidle_poll_time(struct cpuidle_driver * drv,struct cpuidle_device * dev)397 u64 cpuidle_poll_time(struct cpuidle_driver *drv,
398 		      struct cpuidle_device *dev)
399 {
400 	int i;
401 	u64 limit_ns;
402 
403 	BUILD_BUG_ON(CPUIDLE_POLL_MIN > CPUIDLE_POLL_MAX);
404 
405 	if (dev->poll_limit_ns)
406 		return dev->poll_limit_ns;
407 
408 	limit_ns = CPUIDLE_POLL_MAX;
409 	for (i = 1; i < drv->state_count; i++) {
410 		u64 state_limit;
411 
412 		if (dev->states_usage[i].disable)
413 			continue;
414 
415 		state_limit = drv->states[i].target_residency_ns;
416 		if (state_limit < CPUIDLE_POLL_MIN)
417 			continue;
418 
419 		limit_ns = min_t(u64, state_limit, CPUIDLE_POLL_MAX);
420 		break;
421 	}
422 
423 	dev->poll_limit_ns = limit_ns;
424 
425 	return dev->poll_limit_ns;
426 }
427 
428 /**
429  * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
430  */
cpuidle_install_idle_handler(void)431 void cpuidle_install_idle_handler(void)
432 {
433 	if (enabled_devices) {
434 		/* Make sure all changes finished before we switch to new idle */
435 		smp_wmb();
436 		initialized = 1;
437 	}
438 }
439 
440 /**
441  * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
442  */
cpuidle_uninstall_idle_handler(void)443 void cpuidle_uninstall_idle_handler(void)
444 {
445 	if (enabled_devices) {
446 		initialized = 0;
447 		wake_up_all_idle_cpus();
448 	}
449 
450 	/*
451 	 * Make sure external observers (such as the scheduler)
452 	 * are done looking at pointed idle states.
453 	 */
454 	synchronize_rcu();
455 }
456 
457 /**
458  * cpuidle_pause_and_lock - temporarily disables CPUIDLE
459  */
cpuidle_pause_and_lock(void)460 void cpuidle_pause_and_lock(void)
461 {
462 	mutex_lock(&cpuidle_lock);
463 	cpuidle_uninstall_idle_handler();
464 }
465 
466 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
467 
468 /**
469  * cpuidle_resume_and_unlock - resumes CPUIDLE operation
470  */
cpuidle_resume_and_unlock(void)471 void cpuidle_resume_and_unlock(void)
472 {
473 	cpuidle_install_idle_handler();
474 	mutex_unlock(&cpuidle_lock);
475 }
476 
477 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
478 
479 /* Currently used in suspend/resume path to suspend cpuidle */
cpuidle_pause(void)480 void cpuidle_pause(void)
481 {
482 	mutex_lock(&cpuidle_lock);
483 	cpuidle_uninstall_idle_handler();
484 	mutex_unlock(&cpuidle_lock);
485 }
486 
487 /* Currently used in suspend/resume path to resume cpuidle */
cpuidle_resume(void)488 void cpuidle_resume(void)
489 {
490 	mutex_lock(&cpuidle_lock);
491 	cpuidle_install_idle_handler();
492 	mutex_unlock(&cpuidle_lock);
493 }
494 
495 /**
496  * cpuidle_enable_device - enables idle PM for a CPU
497  * @dev: the CPU
498  *
499  * This function must be called between cpuidle_pause_and_lock and
500  * cpuidle_resume_and_unlock when used externally.
501  */
cpuidle_enable_device(struct cpuidle_device * dev)502 int cpuidle_enable_device(struct cpuidle_device *dev)
503 {
504 	int ret;
505 	struct cpuidle_driver *drv;
506 
507 	if (!dev)
508 		return -EINVAL;
509 
510 	if (dev->enabled)
511 		return 0;
512 
513 	if (!cpuidle_curr_governor)
514 		return -EIO;
515 
516 	drv = cpuidle_get_cpu_driver(dev);
517 
518 	if (!drv)
519 		return -EIO;
520 
521 	if (!dev->registered)
522 		return -EINVAL;
523 
524 	ret = cpuidle_add_device_sysfs(dev);
525 	if (ret)
526 		return ret;
527 
528 	if (cpuidle_curr_governor->enable) {
529 		ret = cpuidle_curr_governor->enable(drv, dev);
530 		if (ret)
531 			goto fail_sysfs;
532 	}
533 
534 	smp_wmb();
535 
536 	dev->enabled = 1;
537 
538 	enabled_devices++;
539 	return 0;
540 
541 fail_sysfs:
542 	cpuidle_remove_device_sysfs(dev);
543 
544 	return ret;
545 }
546 
547 EXPORT_SYMBOL_GPL(cpuidle_enable_device);
548 
549 /**
550  * cpuidle_disable_device - disables idle PM for a CPU
551  * @dev: the CPU
552  *
553  * This function must be called between cpuidle_pause_and_lock and
554  * cpuidle_resume_and_unlock when used externally.
555  */
cpuidle_disable_device(struct cpuidle_device * dev)556 void cpuidle_disable_device(struct cpuidle_device *dev)
557 {
558 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
559 
560 	if (!dev || !dev->enabled)
561 		return;
562 
563 	if (!drv || !cpuidle_curr_governor)
564 		return;
565 
566 	dev->enabled = 0;
567 
568 	if (cpuidle_curr_governor->disable)
569 		cpuidle_curr_governor->disable(drv, dev);
570 
571 	cpuidle_remove_device_sysfs(dev);
572 	enabled_devices--;
573 }
574 
575 EXPORT_SYMBOL_GPL(cpuidle_disable_device);
576 
__cpuidle_unregister_device(struct cpuidle_device * dev)577 static void __cpuidle_unregister_device(struct cpuidle_device *dev)
578 {
579 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
580 
581 	list_del(&dev->device_list);
582 	per_cpu(cpuidle_devices, dev->cpu) = NULL;
583 	module_put(drv->owner);
584 
585 	dev->registered = 0;
586 }
587 
__cpuidle_device_init(struct cpuidle_device * dev)588 static void __cpuidle_device_init(struct cpuidle_device *dev)
589 {
590 	memset(dev->states_usage, 0, sizeof(dev->states_usage));
591 	dev->last_residency_ns = 0;
592 	dev->next_hrtimer = 0;
593 }
594 
595 /**
596  * __cpuidle_register_device - internal register function called before register
597  * and enable routines
598  * @dev: the cpu
599  *
600  * cpuidle_lock mutex must be held before this is called
601  */
__cpuidle_register_device(struct cpuidle_device * dev)602 static int __cpuidle_register_device(struct cpuidle_device *dev)
603 {
604 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
605 	int i, ret;
606 
607 	if (!try_module_get(drv->owner))
608 		return -EINVAL;
609 
610 	for (i = 0; i < drv->state_count; i++) {
611 		if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
612 			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
613 
614 		if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
615 			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
616 	}
617 
618 	per_cpu(cpuidle_devices, dev->cpu) = dev;
619 	list_add(&dev->device_list, &cpuidle_detected_devices);
620 
621 	ret = cpuidle_coupled_register_device(dev);
622 	if (ret)
623 		__cpuidle_unregister_device(dev);
624 	else
625 		dev->registered = 1;
626 
627 	return ret;
628 }
629 
630 /**
631  * cpuidle_register_device - registers a CPU's idle PM feature
632  * @dev: the cpu
633  */
cpuidle_register_device(struct cpuidle_device * dev)634 int cpuidle_register_device(struct cpuidle_device *dev)
635 {
636 	int ret = -EBUSY;
637 
638 	if (!dev)
639 		return -EINVAL;
640 
641 	mutex_lock(&cpuidle_lock);
642 
643 	if (dev->registered)
644 		goto out_unlock;
645 
646 	__cpuidle_device_init(dev);
647 
648 	ret = __cpuidle_register_device(dev);
649 	if (ret)
650 		goto out_unlock;
651 
652 	ret = cpuidle_add_sysfs(dev);
653 	if (ret)
654 		goto out_unregister;
655 
656 	ret = cpuidle_enable_device(dev);
657 	if (ret)
658 		goto out_sysfs;
659 
660 	cpuidle_install_idle_handler();
661 
662 out_unlock:
663 	mutex_unlock(&cpuidle_lock);
664 
665 	return ret;
666 
667 out_sysfs:
668 	cpuidle_remove_sysfs(dev);
669 out_unregister:
670 	__cpuidle_unregister_device(dev);
671 	goto out_unlock;
672 }
673 
674 EXPORT_SYMBOL_GPL(cpuidle_register_device);
675 
676 /**
677  * cpuidle_unregister_device - unregisters a CPU's idle PM feature
678  * @dev: the cpu
679  */
cpuidle_unregister_device(struct cpuidle_device * dev)680 void cpuidle_unregister_device(struct cpuidle_device *dev)
681 {
682 	if (!dev || dev->registered == 0)
683 		return;
684 
685 	cpuidle_pause_and_lock();
686 
687 	cpuidle_disable_device(dev);
688 
689 	cpuidle_remove_sysfs(dev);
690 
691 	__cpuidle_unregister_device(dev);
692 
693 	cpuidle_coupled_unregister_device(dev);
694 
695 	cpuidle_resume_and_unlock();
696 }
697 
698 EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
699 
700 /**
701  * cpuidle_unregister: unregister a driver and the devices. This function
702  * can be used only if the driver has been previously registered through
703  * the cpuidle_register function.
704  *
705  * @drv: a valid pointer to a struct cpuidle_driver
706  */
cpuidle_unregister(struct cpuidle_driver * drv)707 void cpuidle_unregister(struct cpuidle_driver *drv)
708 {
709 	int cpu;
710 	struct cpuidle_device *device;
711 
712 	for_each_cpu(cpu, drv->cpumask) {
713 		device = &per_cpu(cpuidle_dev, cpu);
714 		cpuidle_unregister_device(device);
715 	}
716 
717 	cpuidle_unregister_driver(drv);
718 }
719 EXPORT_SYMBOL_GPL(cpuidle_unregister);
720 
721 /**
722  * cpuidle_register: registers the driver and the cpu devices with the
723  * coupled_cpus passed as parameter. This function is used for all common
724  * initialization pattern there are in the arch specific drivers. The
725  * devices is globally defined in this file.
726  *
727  * @drv         : a valid pointer to a struct cpuidle_driver
728  * @coupled_cpus: a cpumask for the coupled states
729  *
730  * Returns 0 on success, < 0 otherwise
731  */
cpuidle_register(struct cpuidle_driver * drv,const struct cpumask * const coupled_cpus)732 int cpuidle_register(struct cpuidle_driver *drv,
733 		     const struct cpumask *const coupled_cpus)
734 {
735 	int ret, cpu;
736 	struct cpuidle_device *device;
737 
738 	ret = cpuidle_register_driver(drv);
739 	if (ret) {
740 		pr_err("failed to register cpuidle driver\n");
741 		return ret;
742 	}
743 
744 	for_each_cpu(cpu, drv->cpumask) {
745 		device = &per_cpu(cpuidle_dev, cpu);
746 		device->cpu = cpu;
747 
748 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
749 		/*
750 		 * On multiplatform for ARM, the coupled idle states could be
751 		 * enabled in the kernel even if the cpuidle driver does not
752 		 * use it. Note, coupled_cpus is a struct copy.
753 		 */
754 		if (coupled_cpus)
755 			device->coupled_cpus = *coupled_cpus;
756 #endif
757 		ret = cpuidle_register_device(device);
758 		if (!ret)
759 			continue;
760 
761 		pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
762 
763 		cpuidle_unregister(drv);
764 		break;
765 	}
766 
767 	return ret;
768 }
769 EXPORT_SYMBOL_GPL(cpuidle_register);
770 
771 /**
772  * cpuidle_init - core initializer
773  */
cpuidle_init(void)774 static int __init cpuidle_init(void)
775 {
776 	if (cpuidle_disabled())
777 		return -ENODEV;
778 
779 	return cpuidle_add_interface(cpu_subsys.dev_root);
780 }
781 
782 module_param(off, int, 0444);
783 module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
784 core_initcall(cpuidle_init);
785