Linux Audio

Check our new training course

Loading...
v6.8
  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	/* Find lowest-power state that supports long-term idle */
 73	for (i = drv->state_count - 1; i >= 0; i--)
 74		if (drv->states[i].enter_dead)
 75			return drv->states[i].enter_dead(dev, i);
 76
 77	return -ENODEV;
 78}
 79
 80static int find_deepest_state(struct cpuidle_driver *drv,
 81			      struct cpuidle_device *dev,
 82			      u64 max_latency_ns,
 83			      unsigned int forbidden_flags,
 84			      bool s2idle)
 85{
 86	u64 latency_req = 0;
 87	int i, ret = 0;
 88
 89	for (i = 1; i < drv->state_count; i++) {
 90		struct cpuidle_state *s = &drv->states[i];
 91
 92		if (dev->states_usage[i].disable ||
 93		    s->exit_latency_ns <= latency_req ||
 94		    s->exit_latency_ns > max_latency_ns ||
 95		    (s->flags & forbidden_flags) ||
 96		    (s2idle && !s->enter_s2idle))
 97			continue;
 98
 99		latency_req = s->exit_latency_ns;
100		ret = i;
101	}
102	return ret;
103}
104
105/**
106 * cpuidle_use_deepest_state - Set/unset governor override mode.
107 * @latency_limit_ns: Idle state exit latency limit (or no override if 0).
108 *
109 * If @latency_limit_ns is nonzero, set the current CPU to use the deepest idle
110 * state with exit latency within @latency_limit_ns (override governors going
111 * forward), or do not override governors if it is zero.
112 */
113void cpuidle_use_deepest_state(u64 latency_limit_ns)
114{
115	struct cpuidle_device *dev;
116
117	preempt_disable();
118	dev = cpuidle_get_device();
119	if (dev)
120		dev->forced_idle_latency_limit_ns = latency_limit_ns;
121	preempt_enable();
122}
123
124/**
125 * cpuidle_find_deepest_state - Find the deepest available idle state.
126 * @drv: cpuidle driver for the given CPU.
127 * @dev: cpuidle device for the given CPU.
128 * @latency_limit_ns: Idle state exit latency limit
129 *
130 * Return: the index of the deepest available idle state.
131 */
132int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
133			       struct cpuidle_device *dev,
134			       u64 latency_limit_ns)
135{
136	return find_deepest_state(drv, dev, latency_limit_ns, 0, false);
137}
138
139#ifdef CONFIG_SUSPEND
140static noinstr void enter_s2idle_proper(struct cpuidle_driver *drv,
141					 struct cpuidle_device *dev, int index)
142{
143	struct cpuidle_state *target_state = &drv->states[index];
144	ktime_t time_start, time_end;
145
146	instrumentation_begin();
147
148	time_start = ns_to_ktime(local_clock_noinstr());
149
150	tick_freeze();
151	/*
152	 * The state used here cannot be a "coupled" one, because the "coupled"
153	 * cpuidle mechanism enables interrupts and doing that with timekeeping
154	 * suspended is generally unsafe.
155	 */
156	stop_critical_timings();
157	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
158		ct_cpuidle_enter();
159		/* Annotate away the indirect call */
160		instrumentation_begin();
161	}
162	target_state->enter_s2idle(dev, drv, index);
163	if (WARN_ON_ONCE(!irqs_disabled()))
164		raw_local_irq_disable();
165	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
166		instrumentation_end();
167		ct_cpuidle_exit();
168	}
169	tick_unfreeze();
170	start_critical_timings();
171
172	time_end = ns_to_ktime(local_clock_noinstr());
173
174	dev->states_usage[index].s2idle_time += ktime_us_delta(time_end, time_start);
175	dev->states_usage[index].s2idle_usage++;
176	instrumentation_end();
177}
178
179/**
180 * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle.
181 * @drv: cpuidle driver for the given CPU.
182 * @dev: cpuidle device for the given CPU.
183 *
184 * If there are states with the ->enter_s2idle callback, find the deepest of
185 * them and enter it with frozen tick.
186 */
187int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev)
188{
189	int index;
190
191	/*
192	 * Find the deepest state with ->enter_s2idle present, which guarantees
193	 * that interrupts won't be enabled when it exits and allows the tick to
194	 * be frozen safely.
195	 */
196	index = find_deepest_state(drv, dev, U64_MAX, 0, true);
197	if (index > 0) {
198		enter_s2idle_proper(drv, dev, index);
199		local_irq_enable();
200	}
201	return index;
202}
203#endif /* CONFIG_SUSPEND */
204
205/**
206 * cpuidle_enter_state - enter the state and update stats
207 * @dev: cpuidle device for this cpu
208 * @drv: cpuidle driver for this cpu
209 * @index: index into the states table in @drv of the state to enter
210 */
211noinstr int cpuidle_enter_state(struct cpuidle_device *dev,
212				 struct cpuidle_driver *drv,
213				 int index)
214{
215	int entered_state;
216
217	struct cpuidle_state *target_state = &drv->states[index];
218	bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
219	ktime_t time_start, time_end;
220
221	instrumentation_begin();
 
 
222
 
 
223	/*
224	 * Tell the time framework to switch to a broadcast timer because our
225	 * local timer will be shut down.  If a local timer is used from another
226	 * CPU as a broadcast timer, this call may fail if it is not available.
227	 */
228	if (broadcast && tick_broadcast_enter()) {
229		index = find_deepest_state(drv, dev, target_state->exit_latency_ns,
230					   CPUIDLE_FLAG_TIMER_STOP, false);
231		if (index < 0) {
232			default_idle_call();
233			return -EBUSY;
234		}
235		target_state = &drv->states[index];
236		broadcast = false;
237	}
238
239	if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
240		leave_mm(dev->cpu);
241
242	/* Take note of the planned idle state. */
243	sched_idle_set_state(target_state);
244
245	trace_cpu_idle(index, dev->cpu);
246	time_start = ns_to_ktime(local_clock_noinstr());
247
248	stop_critical_timings();
249	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
250		ct_cpuidle_enter();
251		/* Annotate away the indirect call */
252		instrumentation_begin();
253	}
254
255	/*
256	 * NOTE!!
257	 *
258	 * For cpuidle_state::enter() methods that do *NOT* set
259	 * CPUIDLE_FLAG_RCU_IDLE RCU will be disabled here and these functions
260	 * must be marked either noinstr or __cpuidle.
261	 *
262	 * For cpuidle_state::enter() methods that *DO* set
263	 * CPUIDLE_FLAG_RCU_IDLE this isn't required, but they must mark the
264	 * function calling ct_cpuidle_enter() as noinstr/__cpuidle and all
265	 * functions called within the RCU-idle region.
266	 */
267	entered_state = target_state->enter(dev, drv, index);
268
269	if (WARN_ONCE(!irqs_disabled(), "%ps leaked IRQ state", target_state->enter))
270		raw_local_irq_disable();
271
272	if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) {
273		instrumentation_end();
274		ct_cpuidle_exit();
275	}
276	start_critical_timings();
277
278	sched_clock_idle_wakeup_event();
279	time_end = ns_to_ktime(local_clock_noinstr());
280	trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
281
282	/* The cpu is no longer idle or about to enter idle. */
283	sched_idle_set_state(NULL);
284
285	if (broadcast)
286		tick_broadcast_exit();
287
288	if (!cpuidle_state_is_coupled(drv, index))
289		local_irq_enable();
290
291	if (entered_state >= 0) {
292		s64 diff, delay = drv->states[entered_state].exit_latency_ns;
293		int i;
294
295		/*
296		 * Update cpuidle counters
297		 * This can be moved to within driver enter routine,
298		 * but that results in multiple copies of same code.
299		 */
300		diff = ktime_sub(time_end, time_start);
301
302		dev->last_residency_ns = diff;
303		dev->states_usage[entered_state].time_ns += diff;
304		dev->states_usage[entered_state].usage++;
305
306		if (diff < drv->states[entered_state].target_residency_ns) {
307			for (i = entered_state - 1; i >= 0; i--) {
308				if (dev->states_usage[i].disable)
309					continue;
310
311				/* Shallower states are enabled, so update. */
312				dev->states_usage[entered_state].above++;
313				trace_cpu_idle_miss(dev->cpu, entered_state, false);
314				break;
315			}
316		} else if (diff > delay) {
317			for (i = entered_state + 1; i < drv->state_count; i++) {
318				if (dev->states_usage[i].disable)
319					continue;
320
321				/*
322				 * Update if a deeper state would have been a
323				 * better match for the observed idle duration.
324				 */
325				if (diff - delay >= drv->states[i].target_residency_ns) {
326					dev->states_usage[entered_state].below++;
327					trace_cpu_idle_miss(dev->cpu, entered_state, true);
328				}
329
330				break;
331			}
332		}
333	} else {
334		dev->last_residency_ns = 0;
335		dev->states_usage[index].rejected++;
336	}
337
338	instrumentation_end();
339
340	return entered_state;
341}
342
343/**
344 * cpuidle_select - ask the cpuidle framework to choose an idle state
345 *
346 * @drv: the cpuidle driver
347 * @dev: the cpuidle device
348 * @stop_tick: indication on whether or not to stop the tick
349 *
350 * Returns the index of the idle state.  The return value must not be negative.
351 *
352 * The memory location pointed to by @stop_tick is expected to be written the
353 * 'false' boolean value if the scheduler tick should not be stopped before
354 * entering the returned state.
355 */
356int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
357		   bool *stop_tick)
358{
359	return cpuidle_curr_governor->select(drv, dev, stop_tick);
360}
361
362/**
363 * cpuidle_enter - enter into the specified idle state
364 *
365 * @drv:   the cpuidle driver tied with the cpu
366 * @dev:   the cpuidle device
367 * @index: the index in the idle state table
368 *
369 * Returns the index in the idle state, < 0 in case of error.
370 * The error code depends on the backend driver
371 */
372int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
373		  int index)
374{
375	int ret = 0;
376
377	/*
378	 * Store the next hrtimer, which becomes either next tick or the next
379	 * timer event, whatever expires first. Additionally, to make this data
380	 * useful for consumers outside cpuidle, we rely on that the governor's
381	 * ->select() callback have decided, whether to stop the tick or not.
382	 */
383	WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer());
384
385	if (cpuidle_state_is_coupled(drv, index))
386		ret = cpuidle_enter_state_coupled(dev, drv, index);
387	else
388		ret = cpuidle_enter_state(dev, drv, index);
389
390	WRITE_ONCE(dev->next_hrtimer, 0);
391	return ret;
392}
393
394/**
395 * cpuidle_reflect - tell the underlying governor what was the state
396 * we were in
397 *
398 * @dev  : the cpuidle device
399 * @index: the index in the idle state table
400 *
401 */
402void cpuidle_reflect(struct cpuidle_device *dev, int index)
403{
404	if (cpuidle_curr_governor->reflect && index >= 0)
405		cpuidle_curr_governor->reflect(dev, index);
406}
407
408/*
409 * Min polling interval of 10usec is a guess. It is assuming that
410 * for most users, the time for a single ping-pong workload like
411 * perf bench pipe would generally complete within 10usec but
412 * this is hardware dependant. Actual time can be estimated with
413 *
414 * perf bench sched pipe -l 10000
415 *
416 * Run multiple times to avoid cpufreq effects.
417 */
418#define CPUIDLE_POLL_MIN 10000
419#define CPUIDLE_POLL_MAX (TICK_NSEC / 16)
420
421/**
422 * cpuidle_poll_time - return amount of time to poll for,
423 * governors can override dev->poll_limit_ns if necessary
424 *
425 * @drv:   the cpuidle driver tied with the cpu
426 * @dev:   the cpuidle device
427 *
428 */
429__cpuidle u64 cpuidle_poll_time(struct cpuidle_driver *drv,
430		      struct cpuidle_device *dev)
431{
432	int i;
433	u64 limit_ns;
434
435	BUILD_BUG_ON(CPUIDLE_POLL_MIN > CPUIDLE_POLL_MAX);
 
436
437	if (dev->poll_limit_ns)
438		return dev->poll_limit_ns;
439
440	limit_ns = CPUIDLE_POLL_MAX;
441	for (i = 1; i < drv->state_count; i++) {
442		u64 state_limit;
443
444		if (dev->states_usage[i].disable)
445			continue;
446
447		state_limit = drv->states[i].target_residency_ns;
448		if (state_limit < CPUIDLE_POLL_MIN)
449			continue;
450
451		limit_ns = min_t(u64, state_limit, CPUIDLE_POLL_MAX);
452		break;
453	}
454
455	dev->poll_limit_ns = limit_ns;
 
 
456
457	return dev->poll_limit_ns;
458}
459
460/**
461 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
462 */
463void cpuidle_install_idle_handler(void)
464{
465	if (enabled_devices) {
466		/* Make sure all changes finished before we switch to new idle */
467		smp_wmb();
468		initialized = 1;
469	}
470}
471
472/**
473 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
474 */
475void cpuidle_uninstall_idle_handler(void)
476{
477	if (enabled_devices) {
478		initialized = 0;
479		wake_up_all_idle_cpus();
480	}
481
482	/*
483	 * Make sure external observers (such as the scheduler)
484	 * are done looking at pointed idle states.
485	 */
486	synchronize_rcu();
487}
488
489/**
490 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
491 */
492void cpuidle_pause_and_lock(void)
493{
494	mutex_lock(&cpuidle_lock);
495	cpuidle_uninstall_idle_handler();
496}
497
498EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
499
500/**
501 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
502 */
503void cpuidle_resume_and_unlock(void)
504{
505	cpuidle_install_idle_handler();
506	mutex_unlock(&cpuidle_lock);
507}
508
509EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
510
511/* Currently used in suspend/resume path to suspend cpuidle */
512void cpuidle_pause(void)
513{
514	mutex_lock(&cpuidle_lock);
515	cpuidle_uninstall_idle_handler();
516	mutex_unlock(&cpuidle_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
517}
518
519/* Currently used in suspend/resume path to resume cpuidle */
520void cpuidle_resume(void)
521{
522	mutex_lock(&cpuidle_lock);
523	cpuidle_install_idle_handler();
524	mutex_unlock(&cpuidle_lock);
525}
 
 
 
 
 
 
 
 
 
 
 
526
527/**
528 * cpuidle_enable_device - enables idle PM for a CPU
529 * @dev: the CPU
530 *
531 * This function must be called between cpuidle_pause_and_lock and
532 * cpuidle_resume_and_unlock when used externally.
533 */
534int cpuidle_enable_device(struct cpuidle_device *dev)
535{
536	int ret;
537	struct cpuidle_driver *drv;
538
539	if (!dev)
540		return -EINVAL;
541
542	if (dev->enabled)
543		return 0;
544
545	if (!cpuidle_curr_governor)
546		return -EIO;
 
 
547
548	drv = cpuidle_get_cpu_driver(dev);
549
550	if (!drv)
551		return -EIO;
 
552
553	if (!dev->registered)
554		return -EINVAL;
555
556	ret = cpuidle_add_device_sysfs(dev);
557	if (ret)
558		return ret;
559
560	if (cpuidle_curr_governor->enable) {
561		ret = cpuidle_curr_governor->enable(drv, dev);
562		if (ret)
563			goto fail_sysfs;
 
 
 
564	}
 
 
565
566	smp_wmb();
567
568	dev->enabled = 1;
569
570	enabled_devices++;
571	return 0;
572
573fail_sysfs:
574	cpuidle_remove_device_sysfs(dev);
575
576	return ret;
577}
578
579EXPORT_SYMBOL_GPL(cpuidle_enable_device);
580
581/**
582 * cpuidle_disable_device - disables idle PM for a CPU
583 * @dev: the CPU
584 *
585 * This function must be called between cpuidle_pause_and_lock and
586 * cpuidle_resume_and_unlock when used externally.
587 */
588void cpuidle_disable_device(struct cpuidle_device *dev)
589{
590	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
591
592	if (!dev || !dev->enabled)
593		return;
594
595	if (!drv || !cpuidle_curr_governor)
596		return;
597
598	dev->enabled = 0;
599
600	if (cpuidle_curr_governor->disable)
601		cpuidle_curr_governor->disable(drv, dev);
602
603	cpuidle_remove_device_sysfs(dev);
604	enabled_devices--;
605}
606
607EXPORT_SYMBOL_GPL(cpuidle_disable_device);
608
609static void __cpuidle_unregister_device(struct cpuidle_device *dev)
610{
611	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
612
613	list_del(&dev->device_list);
614	per_cpu(cpuidle_devices, dev->cpu) = NULL;
615	module_put(drv->owner);
616
617	dev->registered = 0;
618}
619
620static void __cpuidle_device_init(struct cpuidle_device *dev)
621{
622	memset(dev->states_usage, 0, sizeof(dev->states_usage));
623	dev->last_residency_ns = 0;
624	dev->next_hrtimer = 0;
625}
626
627/**
628 * __cpuidle_register_device - internal register function called before register
629 * and enable routines
630 * @dev: the cpu
631 *
632 * cpuidle_lock mutex must be held before this is called
633 */
634static int __cpuidle_register_device(struct cpuidle_device *dev)
635{
636	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
637	int i, ret;
 
638
639	if (!try_module_get(drv->owner))
 
 
640		return -EINVAL;
641
642	for (i = 0; i < drv->state_count; i++) {
643		if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE)
644			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER;
645
646		if (drv->states[i].flags & CPUIDLE_FLAG_OFF)
647			dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
648	}
649
650	per_cpu(cpuidle_devices, dev->cpu) = dev;
651	list_add(&dev->device_list, &cpuidle_detected_devices);
 
 
 
 
652
653	ret = cpuidle_coupled_register_device(dev);
654	if (ret)
655		__cpuidle_unregister_device(dev);
656	else
657		dev->registered = 1;
658
659	return ret;
660}
661
662/**
663 * cpuidle_register_device - registers a CPU's idle PM feature
664 * @dev: the cpu
665 */
666int cpuidle_register_device(struct cpuidle_device *dev)
667{
668	int ret = -EBUSY;
669
670	if (!dev)
671		return -EINVAL;
672
673	mutex_lock(&cpuidle_lock);
674
675	if (dev->registered)
676		goto out_unlock;
677
678	__cpuidle_device_init(dev);
679
680	ret = __cpuidle_register_device(dev);
681	if (ret)
682		goto out_unlock;
683
684	ret = cpuidle_add_sysfs(dev);
685	if (ret)
686		goto out_unregister;
687
688	ret = cpuidle_enable_device(dev);
689	if (ret)
690		goto out_sysfs;
691
 
692	cpuidle_install_idle_handler();
693
694out_unlock:
695	mutex_unlock(&cpuidle_lock);
696
697	return ret;
698
699out_sysfs:
700	cpuidle_remove_sysfs(dev);
701out_unregister:
702	__cpuidle_unregister_device(dev);
703	goto out_unlock;
704}
705
706EXPORT_SYMBOL_GPL(cpuidle_register_device);
707
708/**
709 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
710 * @dev: the cpu
711 */
712void cpuidle_unregister_device(struct cpuidle_device *dev)
713{
714	if (!dev || dev->registered == 0)
 
 
 
715		return;
716
717	cpuidle_pause_and_lock();
718
719	cpuidle_disable_device(dev);
720
721	cpuidle_remove_sysfs(dev);
722
723	__cpuidle_unregister_device(dev);
724
725	cpuidle_coupled_unregister_device(dev);
726
727	cpuidle_resume_and_unlock();
 
 
728}
729
730EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
731
732/**
733 * cpuidle_unregister: unregister a driver and the devices. This function
734 * can be used only if the driver has been previously registered through
735 * the cpuidle_register function.
736 *
737 * @drv: a valid pointer to a struct cpuidle_driver
738 */
739void cpuidle_unregister(struct cpuidle_driver *drv)
740{
741	int cpu;
742	struct cpuidle_device *device;
743
744	for_each_cpu(cpu, drv->cpumask) {
745		device = &per_cpu(cpuidle_dev, cpu);
746		cpuidle_unregister_device(device);
747	}
748
749	cpuidle_unregister_driver(drv);
750}
751EXPORT_SYMBOL_GPL(cpuidle_unregister);
752
753/**
754 * cpuidle_register: registers the driver and the cpu devices with the
755 * coupled_cpus passed as parameter. This function is used for all common
756 * initialization pattern there are in the arch specific drivers. The
757 * devices is globally defined in this file.
758 *
759 * @drv         : a valid pointer to a struct cpuidle_driver
760 * @coupled_cpus: a cpumask for the coupled states
761 *
762 * Returns 0 on success, < 0 otherwise
763 */
764int cpuidle_register(struct cpuidle_driver *drv,
765		     const struct cpumask *const coupled_cpus)
766{
767	int ret, cpu;
768	struct cpuidle_device *device;
 
769
770	ret = cpuidle_register_driver(drv);
771	if (ret) {
772		pr_err("failed to register cpuidle driver\n");
773		return ret;
774	}
775
776	for_each_cpu(cpu, drv->cpumask) {
777		device = &per_cpu(cpuidle_dev, cpu);
778		device->cpu = cpu;
779
780#ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
781		/*
782		 * On multiplatform for ARM, the coupled idle states could be
783		 * enabled in the kernel even if the cpuidle driver does not
784		 * use it. Note, coupled_cpus is a struct copy.
785		 */
786		if (coupled_cpus)
787			device->coupled_cpus = *coupled_cpus;
788#endif
789		ret = cpuidle_register_device(device);
790		if (!ret)
791			continue;
792
793		pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
794
795		cpuidle_unregister(drv);
796		break;
797	}
798
799	return ret;
800}
801EXPORT_SYMBOL_GPL(cpuidle_register);
802
803/**
804 * cpuidle_init - core initializer
805 */
806static int __init cpuidle_init(void)
807{
 
 
808	if (cpuidle_disabled())
809		return -ENODEV;
810
811	return cpuidle_add_interface();
 
 
 
 
 
 
812}
813
814module_param(off, int, 0444);
815module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444);
816core_initcall(cpuidle_init);
v3.1
  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/kernel.h>
 12#include <linux/mutex.h>
 13#include <linux/sched.h>
 
 
 14#include <linux/notifier.h>
 15#include <linux/pm_qos_params.h>
 16#include <linux/cpu.h>
 17#include <linux/cpuidle.h>
 18#include <linux/ktime.h>
 19#include <linux/hrtimer.h>
 
 
 
 
 
 20#include <trace/events/power.h>
 21
 22#include "cpuidle.h"
 23
 24DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
 
 25
 26DEFINE_MUTEX(cpuidle_lock);
 27LIST_HEAD(cpuidle_detected_devices);
 28
 29static int enabled_devices;
 30static int off __read_mostly;
 31static int initialized __read_mostly;
 32
 33int cpuidle_disabled(void)
 34{
 35	return off;
 36}
 37void disable_cpuidle(void)
 38{
 39	off = 1;
 40}
 41
 42#if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT)
 43static void cpuidle_kick_cpus(void)
 44{
 45	cpu_idle_wait();
 46}
 47#elif defined(CONFIG_SMP)
 48# error "Arch needs cpu_idle_wait() equivalent here"
 49#else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */
 50static void cpuidle_kick_cpus(void) {}
 51#endif
 52
 53static int __cpuidle_register_device(struct cpuidle_device *dev);
 54
 55/**
 56 * cpuidle_idle_call - the main idle loop
 57 *
 58 * NOTE: no locks or semaphores should be used here
 59 * return non-zero on failure
 60 */
 61int cpuidle_idle_call(void)
 62{
 63	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
 64	struct cpuidle_state *target_state;
 65	int next_state;
 66
 67	if (off)
 68		return -ENODEV;
 69
 70	if (!initialized)
 71		return -ENODEV;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 72
 73	/* check if the device is ready */
 74	if (!dev || !dev->enabled)
 75		return -EBUSY;
 76
 77#if 0
 78	/* shows regressions, re-enable for 2.6.29 */
 79	/*
 80	 * run any timers that can be run now, at this point
 81	 * before calculating the idle duration etc.
 
 82	 */
 83	hrtimer_peek_ahead_timers();
 84#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 85
 86	/*
 87	 * Call the device's prepare function before calling the
 88	 * governor's select function.  ->prepare gives the device's
 89	 * cpuidle driver a chance to update any dynamic information
 90	 * of its cpuidle states for the current idle period, e.g.
 91	 * state availability, latencies, residencies, etc.
 
 
 
 
 
 92	 */
 93	if (dev->prepare)
 94		dev->prepare(dev);
 
 
 
 
 
 
 
 
 95
 96	/* ask the governor for the next state */
 97	next_state = cpuidle_curr_governor->select(dev);
 98	if (need_resched()) {
 
 
 
 
 
 
 
 
 99		local_irq_enable();
100		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
101	}
102
103	target_state = &dev->states[next_state];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
104
105	/* enter the state and update stats */
106	dev->last_state = target_state;
107
108	trace_power_start(POWER_CSTATE, next_state, dev->cpu);
109	trace_cpu_idle(next_state, dev->cpu);
110
111	dev->last_residency = target_state->enter(dev, target_state);
 
 
112
113	trace_power_end(dev->cpu);
114	trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);
115
116	if (dev->last_state)
117		target_state = dev->last_state;
 
118
119	target_state->time += (unsigned long long)dev->last_residency;
120	target_state->usage++;
 
121
122	/* give the governor an opportunity to reflect on the outcome */
123	if (cpuidle_curr_governor->reflect)
124		cpuidle_curr_governor->reflect(dev);
125
126	return 0;
127}
128
129/**
130 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
131 */
132void cpuidle_install_idle_handler(void)
133{
134	if (enabled_devices) {
135		/* Make sure all changes finished before we switch to new idle */
136		smp_wmb();
137		initialized = 1;
138	}
139}
140
141/**
142 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
143 */
144void cpuidle_uninstall_idle_handler(void)
145{
146	if (enabled_devices) {
147		initialized = 0;
148		cpuidle_kick_cpus();
149	}
 
 
 
 
 
 
150}
151
152/**
153 * cpuidle_pause_and_lock - temporarily disables CPUIDLE
154 */
155void cpuidle_pause_and_lock(void)
156{
157	mutex_lock(&cpuidle_lock);
158	cpuidle_uninstall_idle_handler();
159}
160
161EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
162
163/**
164 * cpuidle_resume_and_unlock - resumes CPUIDLE operation
165 */
166void cpuidle_resume_and_unlock(void)
167{
168	cpuidle_install_idle_handler();
169	mutex_unlock(&cpuidle_lock);
170}
171
172EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
173
174#ifdef CONFIG_ARCH_HAS_CPU_RELAX
175static int poll_idle(struct cpuidle_device *dev, struct cpuidle_state *st)
176{
177	ktime_t	t1, t2;
178	s64 diff;
179	int ret;
180
181	t1 = ktime_get();
182	local_irq_enable();
183	while (!need_resched())
184		cpu_relax();
185
186	t2 = ktime_get();
187	diff = ktime_to_us(ktime_sub(t2, t1));
188	if (diff > INT_MAX)
189		diff = INT_MAX;
190
191	ret = (int) diff;
192	return ret;
193}
194
195static void poll_idle_init(struct cpuidle_device *dev)
 
196{
197	struct cpuidle_state *state = &dev->states[0];
198
199	cpuidle_set_statedata(state, NULL);
200
201	snprintf(state->name, CPUIDLE_NAME_LEN, "POLL");
202	snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE");
203	state->exit_latency = 0;
204	state->target_residency = 0;
205	state->power_usage = -1;
206	state->flags = 0;
207	state->enter = poll_idle;
208}
209#else
210static void poll_idle_init(struct cpuidle_device *dev) {}
211#endif /* CONFIG_ARCH_HAS_CPU_RELAX */
212
213/**
214 * cpuidle_enable_device - enables idle PM for a CPU
215 * @dev: the CPU
216 *
217 * This function must be called between cpuidle_pause_and_lock and
218 * cpuidle_resume_and_unlock when used externally.
219 */
220int cpuidle_enable_device(struct cpuidle_device *dev)
221{
222	int ret, i;
 
 
 
 
223
224	if (dev->enabled)
225		return 0;
226	if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 
227		return -EIO;
228	if (!dev->state_count)
229		return -EINVAL;
230
231	if (dev->registered == 0) {
232		ret = __cpuidle_register_device(dev);
233		if (ret)
234			return ret;
235	}
236
237	poll_idle_init(dev);
 
238
239	if ((ret = cpuidle_add_state_sysfs(dev)))
 
240		return ret;
241
242	if (cpuidle_curr_governor->enable &&
243	    (ret = cpuidle_curr_governor->enable(dev)))
244		goto fail_sysfs;
245
246	for (i = 0; i < dev->state_count; i++) {
247		dev->states[i].usage = 0;
248		dev->states[i].time = 0;
249	}
250	dev->last_residency = 0;
251	dev->last_state = NULL;
252
253	smp_wmb();
254
255	dev->enabled = 1;
256
257	enabled_devices++;
258	return 0;
259
260fail_sysfs:
261	cpuidle_remove_state_sysfs(dev);
262
263	return ret;
264}
265
266EXPORT_SYMBOL_GPL(cpuidle_enable_device);
267
268/**
269 * cpuidle_disable_device - disables idle PM for a CPU
270 * @dev: the CPU
271 *
272 * This function must be called between cpuidle_pause_and_lock and
273 * cpuidle_resume_and_unlock when used externally.
274 */
275void cpuidle_disable_device(struct cpuidle_device *dev)
276{
277	if (!dev->enabled)
 
 
278		return;
279	if (!cpuidle_get_driver() || !cpuidle_curr_governor)
 
280		return;
281
282	dev->enabled = 0;
283
284	if (cpuidle_curr_governor->disable)
285		cpuidle_curr_governor->disable(dev);
286
287	cpuidle_remove_state_sysfs(dev);
288	enabled_devices--;
289}
290
291EXPORT_SYMBOL_GPL(cpuidle_disable_device);
292
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
293/**
294 * __cpuidle_register_device - internal register function called before register
295 * and enable routines
296 * @dev: the cpu
297 *
298 * cpuidle_lock mutex must be held before this is called
299 */
300static int __cpuidle_register_device(struct cpuidle_device *dev)
301{
302	int ret;
303	struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
304	struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
305
306	if (!sys_dev)
307		return -EINVAL;
308	if (!try_module_get(cpuidle_driver->owner))
309		return -EINVAL;
310
311	init_completion(&dev->kobj_unregister);
 
 
312
313	/*
314	 * cpuidle driver should set the dev->power_specified bit
315	 * before registering the device if the driver provides
316	 * power_usage numbers.
317	 *
318	 * For those devices whose ->power_specified is not set,
319	 * we fill in power_usage with decreasing values as the
320	 * cpuidle code has an implicit assumption that state Cn
321	 * uses less power than C(n-1).
322	 *
323	 * With CONFIG_ARCH_HAS_CPU_RELAX, C0 is already assigned
324	 * an power value of -1.  So we use -2, -3, etc, for other
325	 * c-states.
326	 */
327	if (!dev->power_specified) {
328		int i;
329		for (i = CPUIDLE_DRIVER_STATE_START; i < dev->state_count; i++)
330			dev->states[i].power_usage = -1 - i;
331	}
332
333	per_cpu(cpuidle_devices, dev->cpu) = dev;
334	list_add(&dev->device_list, &cpuidle_detected_devices);
335	if ((ret = cpuidle_add_sysfs(sys_dev))) {
336		module_put(cpuidle_driver->owner);
337		return ret;
338	}
339
340	dev->registered = 1;
341	return 0;
 
 
 
 
 
342}
343
344/**
345 * cpuidle_register_device - registers a CPU's idle PM feature
346 * @dev: the cpu
347 */
348int cpuidle_register_device(struct cpuidle_device *dev)
349{
350	int ret;
 
 
 
351
352	mutex_lock(&cpuidle_lock);
353
354	if ((ret = __cpuidle_register_device(dev))) {
355		mutex_unlock(&cpuidle_lock);
356		return ret;
357	}
 
 
 
 
 
 
 
 
 
 
 
 
358
359	cpuidle_enable_device(dev);
360	cpuidle_install_idle_handler();
361
 
362	mutex_unlock(&cpuidle_lock);
363
364	return 0;
365
 
 
 
 
 
366}
367
368EXPORT_SYMBOL_GPL(cpuidle_register_device);
369
370/**
371 * cpuidle_unregister_device - unregisters a CPU's idle PM feature
372 * @dev: the cpu
373 */
374void cpuidle_unregister_device(struct cpuidle_device *dev)
375{
376	struct sys_device *sys_dev = get_cpu_sysdev((unsigned long)dev->cpu);
377	struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver();
378
379	if (dev->registered == 0)
380		return;
381
382	cpuidle_pause_and_lock();
383
384	cpuidle_disable_device(dev);
385
386	cpuidle_remove_sysfs(sys_dev);
387	list_del(&dev->device_list);
388	wait_for_completion(&dev->kobj_unregister);
389	per_cpu(cpuidle_devices, dev->cpu) = NULL;
 
390
391	cpuidle_resume_and_unlock();
392
393	module_put(cpuidle_driver->owner);
394}
395
396EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
397
398#ifdef CONFIG_SMP
 
 
 
 
 
 
 
 
 
 
399
400static void smp_callback(void *v)
401{
402	/* we already woke the CPU up, nothing more to do */
 
 
 
403}
 
404
405/*
406 * This function gets called when a part of the kernel has a new latency
407 * requirement.  This means we need to get all processors out of their C-state,
408 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
409 * wakes them all right up.
 
 
 
 
 
410 */
411static int cpuidle_latency_notify(struct notifier_block *b,
412		unsigned long l, void *v)
413{
414	smp_call_function(smp_callback, NULL, 1);
415	return NOTIFY_OK;
416}
417
418static struct notifier_block cpuidle_latency_notifier = {
419	.notifier_call = cpuidle_latency_notify,
420};
 
 
421
422static inline void latency_notifier_init(struct notifier_block *n)
423{
424	pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
425}
 
 
 
 
 
 
 
 
 
 
 
 
426
427#else /* CONFIG_SMP */
428
429#define latency_notifier_init(x) do { } while (0)
 
 
430
431#endif /* CONFIG_SMP */
 
 
432
433/**
434 * cpuidle_init - core initializer
435 */
436static int __init cpuidle_init(void)
437{
438	int ret;
439
440	if (cpuidle_disabled())
441		return -ENODEV;
442
443	ret = cpuidle_add_class_sysfs(&cpu_sysdev_class);
444	if (ret)
445		return ret;
446
447	latency_notifier_init(&cpuidle_latency_notifier);
448
449	return 0;
450}
451
452module_param(off, int, 0444);
 
453core_initcall(cpuidle_init);