cpuidle.c

  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/sched/idle.h>
 18#include <linux/notifier.h>
 19#include <linux/pm_qos.h>
 20#include <linux/cpu.h>
 21#include <linux/cpuidle.h>
 22#include <linux/ktime.h>
 23#include <linux/hrtimer.h>
 24#include <linux/module.h>
 25#include <linux/suspend.h>
 26#include <linux/tick.h>
 27#include <linux/mmu_context.h>
 28#include <linux/context_tracking.h>
 29#include <trace/events/power.h>
 30
 31#include "cpuidle.h"
 32
 33DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
 34DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
 35
 36DEFINE_MUTEX(cpuidle_lock);
 37LIST_HEAD(cpuidle_detected_devices);
 38
 39static int enabled_devices;
 40static int off __read_mostly;
 41static int initialized __read_mostly;
 42
 43int cpuidle_disabled(void)
 44{
 45	return off;
 46}
 47void disable_cpuidle(void)
 48{
 49	off = 1;
 50}
 51
 52bool cpuidle_not_available(struct cpuidle_driver *drv,
 53			   struct cpuidle_device *dev)
 54{
 55	return off || !initialized || !drv || !dev || !dev->enabled;
 56}
 57
 58/**
 59 * cpuidle_play_dead - cpu off-lining
 60 *
 61 * Returns in case of an error or no driver
 62 */
 63int cpuidle_play_dead(void)
 64{
 65	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
 66	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 67	int i;
 68
 69	if (!drv)
 70		return -ENODEV;
 71
 72	for (i = drv->state_count - 1; i >= 0; i--) {
 73		if (drv->states[i].enter_dead)
 74			drv->states[i].enter_dead(dev, i);
 75	}
 76
 77	/*
 78	 * If :enter_dead() is successful, it will never return, so reaching
 79	 * here means that all of them failed above or were not present.
 80	 */
 81	return -ENODEV;
 82}
 83
 84static int find_deepest_state(struct cpuidle_driver *drv,
 85			      struct cpuidle_device *dev,
 86			      u64 max_latency_ns,
 87			      unsigned int forbidden_flags,
 88			      bool s2idle)
 89{
 90	u64 latency_req = 0;
 91	int i, ret = 0;
 92
 93	for (i = 1; i < drv->state_count; i++) {
 94		struct cpuidle_state *s = &drv->states[i];
 95
 96		if (dev->states_usage[i].disable ||
 97		    s->exit_latency_ns <= latency_req ||
 98		    s->exit_latency_ns > max_latency_ns ||
 99		    (s->flags & forbidden_flags) ||
100		    (s2idle && !s->enter_s2idle))
101			continue;
102
103		latency_req = s->exit_latency_ns;
104		ret = i;
105	}
106	return ret;
107}
108
109/**
110 * cpuidle_use_deepest_state - Set/unset governor override mode.
111 * @latency_limit_ns: Idle state exit latency limit (or no override if 0).
112 *
113 * If @latency_limit_ns is nonzero, set the current CPU to use the deepest idle
114 * state with exit latency within @latency_limit_ns (override governors going
115 * forward), or do not override governors if it is zero.
116 */
117void cpuidle_use_deepest_state(u64 latency_limit_ns)
118{
119	struct cpuidle_device *dev;
120
121	preempt_disable();
122	dev = cpuidle_get_device();
123	if (dev)
124		dev->forced_idle_latency_limit_ns = latency_limit_ns;
125	preempt_enable();
126}
127
128/**
129 * cpuidle_find_deepest_state - Find the deepest available idle state.
130 * @drv: cpuidle driver for the given CPU.
131 * @dev: cpuidle device for the given CPU.
132 * @latency_limit_ns: Idle state exit latency limit
133 *
134 * Return: the index of the deepest available idle state.
135 */
136int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
137			       struct cpuidle_device *dev,
138			       u64 latency_limit_ns)
139{
140	return find_deepest_state(drv, dev, latency_limit_ns, 0, false);
141}
142
143#ifdef CONFIG_SUSPEND
144static noinstr void enter_s2idle_proper(struct cpuidle_driver *drv,
145					 struct cpuidle_device *dev, int index)
146{
147	struct cpuidle_state *target_state = &drv->states[index];
148	ktime_t time_start, time_end;
149
150	instrumentation_begin();
151
152	time_start = ns_to_ktime(local_clock_noinstr());
153
154	tick_freeze();
155	/*
156	 * The state used here cannot be a "coupled" one, because the "coupled"
157	 * cpuidle mechanism enables interrupts and doing that with timekeeping
158	 * suspended is generally unsafe.
159	 */
160	stop_critical_timings();
161	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
162		ct_cpuidle_enter();
163		/* Annotate away the indirect call */
164		instrumentation_begin();
165	}
166	target_state->enter_s2idle(dev, drv, index);
167	if (WARN_ON_ONCE(!irqs_disabled()))
168		raw_local_irq_disable();
169	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
170		instrumentation_end();
171		ct_cpuidle_exit();
172	}
173	tick_unfreeze();
174	start_critical_timings();
175
176	time_end = ns_to_ktime(local_clock_noinstr());
177
178	dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
179	dev->states_usage[index].s2idle_usage++;
180	instrumentation_end();
181}
182
183/**
184 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
185 * @drv: cpuidle driver for the given CPU.
186 * @dev: cpuidle device for the given CPU.
187 * @latency_limit_ns: Idle state exit latency limit
188 *
189 * If there are states with the ->enter_s2idle callback, find the deepest of
190 * them and enter it with frozen tick.
191 */
192int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
193			 u64 latency_limit_ns)
194{
195	int index;
196
197	/*
198	 * Find the deepest state with ->enter_s2idle present that meets the
199	 * specified latency limit, which guarantees that interrupts won't be
200	 * enabled when it exits and allows the tick to be frozen safely.
201	 */
202	index = find_deepest_state(drv, dev, latency_limit_ns, 0, true);
203	if (index > 0) {
204		enter_s2idle_proper(drv, dev, index);
205		local_irq_enable();
206	}
207	return index;
208}
209#endif /* CONFIG_SUSPEND */
210
211/**
212 * cpuidle_enter_state - enter the state and update stats
213 * @dev: cpuidle device for this cpu
214 * @drv: cpuidle driver for this cpu
215 * @index: index into the states table in @drv of the state to enter
216 */
217noinstr int cpuidle_enter_state(struct cpuidle_device *dev,
218				 struct cpuidle_driver *drv,
219				 int index)
220{
221	int entered_state;
222
223	struct cpuidle_state *target_state = &drv->states[index];
224	bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
225	ktime_t time_start, time_end;
226
227	instrumentation_begin();
228
229	/*
230	 * Tell the time framework to switch to a broadcast timer because our
231	 * local timer will be shut down.  If a local timer is used from another
232	 * CPU as a broadcast timer, this call may fail if it is not available.
233	 */
234	if (broadcast && tick_broadcast_enter()) {
235		index = find_deepest_state(drv, dev, target_state->exit_latency_ns,
236					   CPUIDLE_FLAG_TIMER_STOP, false);
237
238		target_state = &drv->states[index];
239		broadcast = false;
240	}
241
242	if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
243		leave_mm();
244
245	/* Take note of the planned idle state. */
246	sched_idle_set_state(target_state);
247
248	trace_cpu_idle(index, dev->cpu);
249	time_start = ns_to_ktime(local_clock_noinstr());
250
251	stop_critical_timings();
252	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
253		ct_cpuidle_enter();
254		/* Annotate away the indirect call */
255		instrumentation_begin();
256	}
257
258	/*
259	 * NOTE!!
260	 *
261	 * For cpuidle_state::enter() methods that do *NOT* set
262	 * CPUIDLE_FLAG_RCU_IDLE RCU will be disabled here and these functions
263	 * must be marked either noinstr or __cpuidle.
264	 *
265	 * For cpuidle_state::enter() methods that *DO* set
266	 * CPUIDLE_FLAG_RCU_IDLE this isn't required, but they must mark the
267	 * function calling ct_cpuidle_enter() as noinstr/__cpuidle and all
268	 * functions called within the RCU-idle region.
269	 */
270	entered_state = target_state->enter(dev, drv, index);
271
272	if (WARN_ONCE(!irqs_disabled(), "%ps leaked IRQ state", target_state->enter))
273		raw_local_irq_disable();
274
275	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
276		instrumentation_end();
277		ct_cpuidle_exit();
278	}
279	start_critical_timings();
280
281	sched_clock_idle_wakeup_event();
282	time_end = ns_to_ktime(local_clock_noinstr());
283	trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
284
285	/* The cpu is no longer idle or about to enter idle. */
286	sched_idle_set_state(NULL);
287
288	if (broadcast)
289		tick_broadcast_exit();
290
291	if (!cpuidle_state_is_coupled(drv, index))
292		local_irq_enable();
293
294	if (entered_state >= 0) {
295		s64 diff, delay = drv->states[entered_state].exit_latency_ns;
296		int i;
297
298		/*
299		 * Update cpuidle counters
300		 * This can be moved to within driver enter routine,
301		 * but that results in multiple copies of same code.
302		 */
303		diff = ktime_sub(time_end, time_start);
304
305		dev->last_residency_ns = diff;
306		dev->states_usage[entered_state].time_ns += diff;
307		dev->states_usage[entered_state].usage++;
308
309		if (diff < drv->states[entered_state].target_residency_ns) {
310			for (i = entered_state - 1; i >= 0; i--) {
311				if (dev->states_usage[i].disable)
312					continue;
313
314				/* Shallower states are enabled, so update. */
315				dev->states_usage[entered_state].above++;
316				trace_cpu_idle_miss(dev->cpu, entered_state, false);
317				break;
318			}
319		} else if (diff > delay) {
320			for (i = entered_state + 1; i < drv->state_count; i++) {
321				if (dev->states_usage[i].disable)
322					continue;
323
324				/*
325				 * Update if a deeper state would have been a
326				 * better match for the observed idle duration.
327				 */
328				if (diff - delay >= drv->states[i].target_residency_ns) {
329					dev->states_usage[entered_state].below++;
330					trace_cpu_idle_miss(dev->cpu, entered_state, true);
331				}
332
333				break;
334			}
335		}
336	} else {
337		dev->last_residency_ns = 0;
338		dev->states_usage[index].rejected++;
339	}
340
341	instrumentation_end();
342
343	return entered_state;
344}
345
346/**
347 * cpuidle_select - ask the cpuidle framework to choose an idle state
348 *
349 * @drv: the cpuidle driver
350 * @dev: the cpuidle device
351 * @stop_tick: indication on whether or not to stop the tick
352 *
353 * Returns the index of the idle state.  The return value must not be negative.
354 *
355 * The memory location pointed to by @stop_tick is expected to be written the
356 * 'false' boolean value if the scheduler tick should not be stopped before
357 * entering the returned state.
358 */
359int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
360		   bool *stop_tick)
361{
362	return cpuidle_curr_governor->select(drv, dev, stop_tick);
363}
364
365/**
366 * cpuidle_enter - enter into the specified idle state
367 *
368 * @drv:   the cpuidle driver tied with the cpu
369 * @dev:   the cpuidle device
370 * @index: the index in the idle state table
371 *
372 * Returns the index in the idle state, < 0 in case of error.
373 * The error code depends on the backend driver
374 */
375int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
376		  int index)
377{
378	int ret = 0;
379
380	/*
381	 * Store the next hrtimer, which becomes either next tick or the next
382	 * timer event, whatever expires first. Additionally, to make this data
383	 * useful for consumers outside cpuidle, we rely on that the governor's
384	 * ->select() callback have decided, whether to stop the tick or not.
385	 */
386	WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
387
388	if (cpuidle_state_is_coupled(drv, index))
389		ret = cpuidle_enter_state_coupled(dev, drv, index);
390	else
391		ret = cpuidle_enter_state(dev, drv, index);
392
393	WRITE_ONCE(dev->next_hrtimer, 0);
394	return ret;
395}
396
397/**
398 * cpuidle_reflect - tell the underlying governor what was the state
399 * we were in
400 *
401 * @dev  : the cpuidle device
402 * @index: the index in the idle state table
403 *
404 */
405void cpuidle_reflect(struct cpuidle_device *dev, int index)
406{
407	if (cpuidle_curr_governor->reflect && index >= 0)
408		cpuidle_curr_governor->reflect(dev, index);
409}
410
411/*
412 * Min polling interval of 10usec is a guess. It is assuming that
413 * for most users, the time for a single ping-pong workload like
414 * perf bench pipe would generally complete within 10usec but
415 * this is hardware dependent. Actual time can be estimated with
416 *
417 * perf bench sched pipe -l 10000
418 *
419 * Run multiple times to avoid cpufreq effects.
420 */
421#define CPUIDLE_POLL_MIN 10000
422#define CPUIDLE_POLL_MAX (TICK_NSEC / 16)
423
424/**
425 * cpuidle_poll_time - return amount of time to poll for,
426 * governors can override dev->poll_limit_ns if necessary
427 *
428 * @drv:   the cpuidle driver tied with the cpu
429 * @dev:   the cpuidle device
430 *
431 */
432__cpuidle u64 cpuidle_poll_time(struct cpuidle_driver *drv,
433		      struct cpuidle_device *dev)
434{
435	int i;
436	u64 limit_ns;
437
438	BUILD_BUG_ON(CPUIDLE_POLL_MIN > CPUIDLE_POLL_MAX);
439
440	if (dev->poll_limit_ns)
441		return dev->poll_limit_ns;
442
443	limit_ns = CPUIDLE_POLL_MAX;
444	for (i = 1; i < drv->state_count; i++) {
445		u64 state_limit;
446
447		if (dev->states_usage[i].disable)
448			continue;
449
450		state_limit = drv->states[i].target_residency_ns;
451		if (state_limit < CPUIDLE_POLL_MIN)
452			continue;
453
454		limit_ns = min_t(u64, state_limit, CPUIDLE_POLL_MAX);
455		break;
456	}
457
458	dev->poll_limit_ns = limit_ns;
459
460	return dev->poll_limit_ns;
461}
462
463/**
464 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
465 */
466void cpuidle_install_idle_handler(void)
467{
468	if (enabled_devices) {
469		/* Make sure all changes finished before we switch to new idle */
470		smp_wmb();
471		initialized = 1;
472	}
473}
474
475/**
476 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
477 */
478void cpuidle_uninstall_idle_handler(void)
479{
480	if (enabled_devices) {
481		initialized = 0;
482		wake_up_all_idle_cpus();
483	}
484
485	/*
486	 * Make sure external observers (such as the scheduler)
487	 * are done looking at pointed idle states.
488	 */
489	synchronize_rcu();
490}
491
492/**
493 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
494 */
495void cpuidle_pause_and_lock(void)
496{
497	mutex_lock(&cpuidle_lock);
498	cpuidle_uninstall_idle_handler();
499}
500
501EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
502
503/**
504 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
505 */
506void cpuidle_resume_and_unlock(void)
507{
508	cpuidle_install_idle_handler();
509	mutex_unlock(&cpuidle_lock);
510}
511
512EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
513
514/* Currently used in suspend/resume path to suspend cpuidle */
515void cpuidle_pause(void)
516{
517	mutex_lock(&cpuidle_lock);
518	cpuidle_uninstall_idle_handler();
519	mutex_unlock(&cpuidle_lock);
520}
521
522/* Currently used in suspend/resume path to resume cpuidle */
523void cpuidle_resume(void)
524{
525	mutex_lock(&cpuidle_lock);
526	cpuidle_install_idle_handler();
527	mutex_unlock(&cpuidle_lock);
528}
529
530/**
531 * cpuidle_enable_device - enables idle PM for a CPU
532 * @dev: the CPU
533 *
534 * This function must be called between cpuidle_pause_and_lock and
535 * cpuidle_resume_and_unlock when used externally.
536 */
537int cpuidle_enable_device(struct cpuidle_device *dev)
538{
539	int ret;
540	struct cpuidle_driver *drv;
541
542	if (!dev)
543		return -EINVAL;
544
545	if (dev->enabled)
546		return 0;
547
548	if (!cpuidle_curr_governor)
549		return -EIO;
550
551	drv = cpuidle_get_cpu_driver(dev);
552
553	if (!drv)
554		return -EIO;
555
556	if (!dev->registered)
557		return -EINVAL;
558
559	ret = cpuidle_add_device_sysfs(dev);
560	if (ret)
561		return ret;
562
563	if (cpuidle_curr_governor->enable) {
564		ret = cpuidle_curr_governor->enable(drv, dev);
565		if (ret)
566			goto fail_sysfs;
567	}
568
569	smp_wmb();
570
571	dev->enabled = 1;
572
573	enabled_devices++;
574	return 0;
575
576fail_sysfs:
577	cpuidle_remove_device_sysfs(dev);
578
579	return ret;
580}
581
582EXPORT_SYMBOL_GPL(cpuidle_enable_device);
583
584/**
585 * cpuidle_disable_device - disables idle PM for a CPU
586 * @dev: the CPU
587 *
588 * This function must be called between cpuidle_pause_and_lock and
589 * cpuidle_resume_and_unlock when used externally.
590 */
591void cpuidle_disable_device(struct cpuidle_device *dev)
592{
593	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
594
595	if (!dev || !dev->enabled)
596		return;
597
598	if (!drv || !cpuidle_curr_governor)
599		return;
600
601	dev->enabled = 0;
602
603	if (cpuidle_curr_governor->disable)
604		cpuidle_curr_governor->disable(drv, dev);
605
606	cpuidle_remove_device_sysfs(dev);
607	enabled_devices--;
608}
609
610EXPORT_SYMBOL_GPL(cpuidle_disable_device);
611
612static void __cpuidle_unregister_device(struct cpuidle_device *dev)
613{
614	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
615
616	list_del(&dev->device_list);
617	per_cpu(cpuidle_devices, dev->cpu) = NULL;
618	module_put(drv->owner);
619
620	dev->registered = 0;
621}
622
623static void __cpuidle_device_init(struct cpuidle_device *dev)
624{
625	memset(dev->states_usage, 0, sizeof(dev->states_usage));
626	dev->last_residency_ns = 0;
627	dev->next_hrtimer = 0;
628}
629
630/**
631 * __cpuidle_register_device - internal register function called before register
632 * and enable routines
633 * @dev: the cpu
634 *
635 * cpuidle_lock mutex must be held before this is called
636 */
637static int __cpuidle_register_device(struct cpuidle_device *dev)
638{
639	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
640	unsigned int cpu = dev->cpu;
641	int i, ret;
642
643	if (per_cpu(cpuidle_devices, cpu)) {
644		pr_info("CPU%d: cpuidle device already registered\n", cpu);
645		return -EEXIST;
646	}
647
648	if (!try_module_get(drv->owner))
649		return -EINVAL;
650
651	for (i = 0; i < drv->state_count; i++) {
652		if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
653			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
654
655		if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
656			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
657	}
658
659	per_cpu(cpuidle_devices, cpu) = dev;
660	list_add(&dev->device_list, &cpuidle_detected_devices);
661
662	ret = cpuidle_coupled_register_device(dev);
663	if (ret)
664		__cpuidle_unregister_device(dev);
665	else
666		dev->registered = 1;
667
668	return ret;
669}
670
671/**
672 * cpuidle_register_device - registers a CPU's idle PM feature
673 * @dev: the cpu
674 */
675int cpuidle_register_device(struct cpuidle_device *dev)
676{
677	int ret = -EBUSY;
678
679	if (!dev)
680		return -EINVAL;
681
682	mutex_lock(&cpuidle_lock);
683
684	if (dev->registered)
685		goto out_unlock;
686
687	__cpuidle_device_init(dev);
688
689	ret = __cpuidle_register_device(dev);
690	if (ret)
691		goto out_unlock;
692
693	ret = cpuidle_add_sysfs(dev);
694	if (ret)
695		goto out_unregister;
696
697	ret = cpuidle_enable_device(dev);
698	if (ret)
699		goto out_sysfs;
700
701	cpuidle_install_idle_handler();
702
703out_unlock:
704	mutex_unlock(&cpuidle_lock);
705
706	return ret;
707
708out_sysfs:
709	cpuidle_remove_sysfs(dev);
710out_unregister:
711	__cpuidle_unregister_device(dev);
712	goto out_unlock;
713}
714
715EXPORT_SYMBOL_GPL(cpuidle_register_device);
716
717/**
718 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
719 * @dev: the cpu
720 */
721void cpuidle_unregister_device(struct cpuidle_device *dev)
722{
723	if (!dev || dev->registered == 0)
724		return;
725
726	cpuidle_pause_and_lock();
727
728	cpuidle_disable_device(dev);
729
730	cpuidle_remove_sysfs(dev);
731
732	__cpuidle_unregister_device(dev);
733
734	cpuidle_coupled_unregister_device(dev);
735
736	cpuidle_resume_and_unlock();
737}
738
739EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
740
741/**
742 * cpuidle_unregister: unregister a driver and the devices. This function
743 * can be used only if the driver has been previously registered through
744 * the cpuidle_register function.
745 *
746 * @drv: a valid pointer to a struct cpuidle_driver
747 */
748void cpuidle_unregister(struct cpuidle_driver *drv)
749{
750	int cpu;
751	struct cpuidle_device *device;
752
753	for_each_cpu(cpu, drv->cpumask) {
754		device = &per_cpu(cpuidle_dev, cpu);
755		cpuidle_unregister_device(device);
756	}
757
758	cpuidle_unregister_driver(drv);
759}
760EXPORT_SYMBOL_GPL(cpuidle_unregister);
761
762/**
763 * cpuidle_register: registers the driver and the cpu devices with the
764 * coupled_cpus passed as parameter. This function is used for all common
765 * initialization pattern there are in the arch specific drivers. The
766 * devices is globally defined in this file.
767 *
768 * @drv         : a valid pointer to a struct cpuidle_driver
769 * @coupled_cpus: a cpumask for the coupled states
770 *
771 * Returns 0 on success, < 0 otherwise
772 */
773int cpuidle_register(struct cpuidle_driver *drv,
774		     const struct cpumask *const coupled_cpus)
775{
776	int ret, cpu;
777	struct cpuidle_device *device;
778
779	ret = cpuidle_register_driver(drv);
780	if (ret) {
781		pr_err("failed to register cpuidle driver\n");
782		return ret;
783	}
784
785	for_each_cpu(cpu, drv->cpumask) {
786		device = &per_cpu(cpuidle_dev, cpu);
787		device->cpu = cpu;
788
789#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
790		/*
791		 * On multiplatform for ARM, the coupled idle states could be
792		 * enabled in the kernel even if the cpuidle driver does not
793		 * use it. Note, coupled_cpus is a struct copy.
794		 */
795		if (coupled_cpus)
796			device->coupled_cpus = *coupled_cpus;
797#endif
798		ret = cpuidle_register_device(device);
799		if (!ret)
800			continue;
801
802		pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
803
804		cpuidle_unregister(drv);
805		break;
806	}
807
808	return ret;
809}
810EXPORT_SYMBOL_GPL(cpuidle_register);
811
812/**
813 * cpuidle_init - core initializer
814 */
815static int __init cpuidle_init(void)
816{
817	if (cpuidle_disabled())
818		return -ENODEV;
819
820	return cpuidle_add_interface();
821}
822
823module_param(off, int, 0444);
824module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
825core_initcall(cpuidle_init);