xref: /openbmc/linux/drivers/base/power/runtime.c (revision c46a0d5a)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * drivers/base/power/runtime.c - Helper functions for device runtime PM
4  *
5  * Copyright (c) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
6  * Copyright (C) 2010 Alan Stern <stern@rowland.harvard.edu>
7  */
8 #include <linux/sched/mm.h>
9 #include <linux/ktime.h>
10 #include <linux/hrtimer.h>
11 #include <linux/export.h>
12 #include <linux/pm_runtime.h>
13 #include <linux/pm_wakeirq.h>
14 #include <trace/events/rpm.h>
15 
16 #include "../base.h"
17 #include "power.h"
18 
19 typedef int (*pm_callback_t)(struct device *);
20 
21 static pm_callback_t __rpm_get_callback(struct device *dev, size_t cb_offset)
22 {
23 	pm_callback_t cb;
24 	const struct dev_pm_ops *ops;
25 
26 	if (dev->pm_domain)
27 		ops = &dev->pm_domain->ops;
28 	else if (dev->type && dev->type->pm)
29 		ops = dev->type->pm;
30 	else if (dev->class && dev->class->pm)
31 		ops = dev->class->pm;
32 	else if (dev->bus && dev->bus->pm)
33 		ops = dev->bus->pm;
34 	else
35 		ops = NULL;
36 
37 	if (ops)
38 		cb = *(pm_callback_t *)((void *)ops + cb_offset);
39 	else
40 		cb = NULL;
41 
42 	if (!cb && dev->driver && dev->driver->pm)
43 		cb = *(pm_callback_t *)((void *)dev->driver->pm + cb_offset);
44 
45 	return cb;
46 }
47 
48 #define RPM_GET_CALLBACK(dev, callback) \
49 		__rpm_get_callback(dev, offsetof(struct dev_pm_ops, callback))
50 
51 static int rpm_resume(struct device *dev, int rpmflags);
52 static int rpm_suspend(struct device *dev, int rpmflags);
53 
54 /**
55  * update_pm_runtime_accounting - Update the time accounting of power states
56  * @dev: Device to update the accounting for
57  *
58  * In order to be able to have time accounting of the various power states
59  * (as used by programs such as PowerTOP to show the effectiveness of runtime
60  * PM), we need to track the time spent in each state.
61  * update_pm_runtime_accounting must be called each time before the
62  * runtime_status field is updated, to account the time in the old state
63  * correctly.
64  */
65 static void update_pm_runtime_accounting(struct device *dev)
66 {
67 	u64 now, last, delta;
68 
69 	if (dev->power.disable_depth > 0)
70 		return;
71 
72 	last = dev->power.accounting_timestamp;
73 
74 	now = ktime_get_mono_fast_ns();
75 	dev->power.accounting_timestamp = now;
76 
77 	/*
78 	 * Because ktime_get_mono_fast_ns() is not monotonic during
79 	 * timekeeping updates, ensure that 'now' is after the last saved
80 	 * timesptamp.
81 	 */
82 	if (now < last)
83 		return;
84 
85 	delta = now - last;
86 
87 	if (dev->power.runtime_status == RPM_SUSPENDED)
88 		dev->power.suspended_time += delta;
89 	else
90 		dev->power.active_time += delta;
91 }
92 
93 static void __update_runtime_status(struct device *dev, enum rpm_status status)
94 {
95 	update_pm_runtime_accounting(dev);
96 	dev->power.runtime_status = status;
97 }
98 
99 static u64 rpm_get_accounted_time(struct device *dev, bool suspended)
100 {
101 	u64 time;
102 	unsigned long flags;
103 
104 	spin_lock_irqsave(&dev->power.lock, flags);
105 
106 	update_pm_runtime_accounting(dev);
107 	time = suspended ? dev->power.suspended_time : dev->power.active_time;
108 
109 	spin_unlock_irqrestore(&dev->power.lock, flags);
110 
111 	return time;
112 }
113 
114 u64 pm_runtime_active_time(struct device *dev)
115 {
116 	return rpm_get_accounted_time(dev, false);
117 }
118 
119 u64 pm_runtime_suspended_time(struct device *dev)
120 {
121 	return rpm_get_accounted_time(dev, true);
122 }
123 EXPORT_SYMBOL_GPL(pm_runtime_suspended_time);
124 
125 /**
126  * pm_runtime_deactivate_timer - Deactivate given device's suspend timer.
127  * @dev: Device to handle.
128  */
129 static void pm_runtime_deactivate_timer(struct device *dev)
130 {
131 	if (dev->power.timer_expires > 0) {
132 		hrtimer_try_to_cancel(&dev->power.suspend_timer);
133 		dev->power.timer_expires = 0;
134 	}
135 }
136 
137 /**
138  * pm_runtime_cancel_pending - Deactivate suspend timer and cancel requests.
139  * @dev: Device to handle.
140  */
141 static void pm_runtime_cancel_pending(struct device *dev)
142 {
143 	pm_runtime_deactivate_timer(dev);
144 	/*
145 	 * In case there's a request pending, make sure its work function will
146 	 * return without doing anything.
147 	 */
148 	dev->power.request = RPM_REQ_NONE;
149 }
150 
151 /*
152  * pm_runtime_autosuspend_expiration - Get a device's autosuspend-delay expiration time.
153  * @dev: Device to handle.
154  *
155  * Compute the autosuspend-delay expiration time based on the device's
156  * power.last_busy time.  If the delay has already expired or is disabled
157  * (negative) or the power.use_autosuspend flag isn't set, return 0.
158  * Otherwise return the expiration time in nanoseconds (adjusted to be nonzero).
159  *
160  * This function may be called either with or without dev->power.lock held.
161  * Either way it can be racy, since power.last_busy may be updated at any time.
162  */
163 u64 pm_runtime_autosuspend_expiration(struct device *dev)
164 {
165 	int autosuspend_delay;
166 	u64 expires;
167 
168 	if (!dev->power.use_autosuspend)
169 		return 0;
170 
171 	autosuspend_delay = READ_ONCE(dev->power.autosuspend_delay);
172 	if (autosuspend_delay < 0)
173 		return 0;
174 
175 	expires  = READ_ONCE(dev->power.last_busy);
176 	expires += (u64)autosuspend_delay * NSEC_PER_MSEC;
177 	if (expires > ktime_get_mono_fast_ns())
178 		return expires;	/* Expires in the future */
179 
180 	return 0;
181 }
182 EXPORT_SYMBOL_GPL(pm_runtime_autosuspend_expiration);
183 
184 static int dev_memalloc_noio(struct device *dev, void *data)
185 {
186 	return dev->power.memalloc_noio;
187 }
188 
189 /*
190  * pm_runtime_set_memalloc_noio - Set a device's memalloc_noio flag.
191  * @dev: Device to handle.
192  * @enable: True for setting the flag and False for clearing the flag.
193  *
194  * Set the flag for all devices in the path from the device to the
195  * root device in the device tree if @enable is true, otherwise clear
196  * the flag for devices in the path whose siblings don't set the flag.
197  *
198  * The function should only be called by block device, or network
199  * device driver for solving the deadlock problem during runtime
200  * resume/suspend:
201  *
202  *     If memory allocation with GFP_KERNEL is called inside runtime
203  *     resume/suspend callback of any one of its ancestors(or the
204  *     block device itself), the deadlock may be triggered inside the
205  *     memory allocation since it might not complete until the block
206  *     device becomes active and the involed page I/O finishes. The
207  *     situation is pointed out first by Alan Stern. Network device
208  *     are involved in iSCSI kind of situation.
209  *
210  * The lock of dev_hotplug_mutex is held in the function for handling
211  * hotplug race because pm_runtime_set_memalloc_noio() may be called
212  * in async probe().
213  *
214  * The function should be called between device_add() and device_del()
215  * on the affected device(block/network device).
216  */
217 void pm_runtime_set_memalloc_noio(struct device *dev, bool enable)
218 {
219 	static DEFINE_MUTEX(dev_hotplug_mutex);
220 
221 	mutex_lock(&dev_hotplug_mutex);
222 	for (;;) {
223 		bool enabled;
224 
225 		/* hold power lock since bitfield is not SMP-safe. */
226 		spin_lock_irq(&dev->power.lock);
227 		enabled = dev->power.memalloc_noio;
228 		dev->power.memalloc_noio = enable;
229 		spin_unlock_irq(&dev->power.lock);
230 
231 		/*
232 		 * not need to enable ancestors any more if the device
233 		 * has been enabled.
234 		 */
235 		if (enabled && enable)
236 			break;
237 
238 		dev = dev->parent;
239 
240 		/*
241 		 * clear flag of the parent device only if all the
242 		 * children don't set the flag because ancestor's
243 		 * flag was set by any one of the descendants.
244 		 */
245 		if (!dev || (!enable &&
246 			     device_for_each_child(dev, NULL,
247 						   dev_memalloc_noio)))
248 			break;
249 	}
250 	mutex_unlock(&dev_hotplug_mutex);
251 }
252 EXPORT_SYMBOL_GPL(pm_runtime_set_memalloc_noio);
253 
254 /**
255  * rpm_check_suspend_allowed - Test whether a device may be suspended.
256  * @dev: Device to test.
257  */
258 static int rpm_check_suspend_allowed(struct device *dev)
259 {
260 	int retval = 0;
261 
262 	if (dev->power.runtime_error)
263 		retval = -EINVAL;
264 	else if (dev->power.disable_depth > 0)
265 		retval = -EACCES;
266 	else if (atomic_read(&dev->power.usage_count))
267 		retval = -EAGAIN;
268 	else if (!dev->power.ignore_children &&
269 			atomic_read(&dev->power.child_count))
270 		retval = -EBUSY;
271 
272 	/* Pending resume requests take precedence over suspends. */
273 	else if ((dev->power.deferred_resume
274 			&& dev->power.runtime_status == RPM_SUSPENDING)
275 	    || (dev->power.request_pending
276 			&& dev->power.request == RPM_REQ_RESUME))
277 		retval = -EAGAIN;
278 	else if (__dev_pm_qos_resume_latency(dev) == 0)
279 		retval = -EPERM;
280 	else if (dev->power.runtime_status == RPM_SUSPENDED)
281 		retval = 1;
282 
283 	return retval;
284 }
285 
286 static int rpm_get_suppliers(struct device *dev)
287 {
288 	struct device_link *link;
289 
290 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
291 				device_links_read_lock_held()) {
292 		int retval;
293 
294 		if (!(link->flags & DL_FLAG_PM_RUNTIME))
295 			continue;
296 
297 		retval = pm_runtime_get_sync(link->supplier);
298 		/* Ignore suppliers with disabled runtime PM. */
299 		if (retval < 0 && retval != -EACCES) {
300 			pm_runtime_put_noidle(link->supplier);
301 			return retval;
302 		}
303 		refcount_inc(&link->rpm_active);
304 	}
305 	return 0;
306 }
307 
308 /**
309  * pm_runtime_release_supplier - Drop references to device link's supplier.
310  * @link: Target device link.
311  *
312  * Drop all runtime PM references associated with @link to its supplier device.
313  */
314 void pm_runtime_release_supplier(struct device_link *link)
315 {
316 	struct device *supplier = link->supplier;
317 
318 	/*
319 	 * The additional power.usage_count check is a safety net in case
320 	 * the rpm_active refcount becomes saturated, in which case
321 	 * refcount_dec_not_one() would return true forever, but it is not
322 	 * strictly necessary.
323 	 */
324 	while (refcount_dec_not_one(&link->rpm_active) &&
325 	       atomic_read(&supplier->power.usage_count) > 0)
326 		pm_runtime_put_noidle(supplier);
327 }
328 
329 static void __rpm_put_suppliers(struct device *dev, bool try_to_suspend)
330 {
331 	struct device_link *link;
332 
333 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
334 				device_links_read_lock_held()) {
335 		pm_runtime_release_supplier(link);
336 		if (try_to_suspend)
337 			pm_request_idle(link->supplier);
338 	}
339 }
340 
341 static void rpm_put_suppliers(struct device *dev)
342 {
343 	__rpm_put_suppliers(dev, true);
344 }
345 
346 static void rpm_suspend_suppliers(struct device *dev)
347 {
348 	struct device_link *link;
349 	int idx = device_links_read_lock();
350 
351 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
352 				device_links_read_lock_held())
353 		pm_request_idle(link->supplier);
354 
355 	device_links_read_unlock(idx);
356 }
357 
358 /**
359  * __rpm_callback - Run a given runtime PM callback for a given device.
360  * @cb: Runtime PM callback to run.
361  * @dev: Device to run the callback for.
362  */
363 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
364 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
365 {
366 	int retval = 0, idx;
367 	bool use_links = dev->power.links_count > 0;
368 
369 	if (dev->power.irq_safe) {
370 		spin_unlock(&dev->power.lock);
371 	} else {
372 		spin_unlock_irq(&dev->power.lock);
373 
374 		/*
375 		 * Resume suppliers if necessary.
376 		 *
377 		 * The device's runtime PM status cannot change until this
378 		 * routine returns, so it is safe to read the status outside of
379 		 * the lock.
380 		 */
381 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
382 			idx = device_links_read_lock();
383 
384 			retval = rpm_get_suppliers(dev);
385 			if (retval) {
386 				rpm_put_suppliers(dev);
387 				goto fail;
388 			}
389 
390 			device_links_read_unlock(idx);
391 		}
392 	}
393 
394 	if (cb)
395 		retval = cb(dev);
396 
397 	if (dev->power.irq_safe) {
398 		spin_lock(&dev->power.lock);
399 	} else {
400 		/*
401 		 * If the device is suspending and the callback has returned
402 		 * success, drop the usage counters of the suppliers that have
403 		 * been reference counted on its resume.
404 		 *
405 		 * Do that if resume fails too.
406 		 */
407 		if (use_links
408 		    && ((dev->power.runtime_status == RPM_SUSPENDING && !retval)
409 		    || (dev->power.runtime_status == RPM_RESUMING && retval))) {
410 			idx = device_links_read_lock();
411 
412 			__rpm_put_suppliers(dev, false);
413 
414 fail:
415 			device_links_read_unlock(idx);
416 		}
417 
418 		spin_lock_irq(&dev->power.lock);
419 	}
420 
421 	return retval;
422 }
423 
424 /**
425  * rpm_idle - Notify device bus type if the device can be suspended.
426  * @dev: Device to notify the bus type about.
427  * @rpmflags: Flag bits.
428  *
429  * Check if the device's runtime PM status allows it to be suspended.  If
430  * another idle notification has been started earlier, return immediately.  If
431  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
432  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
433  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
434  *
435  * This function must be called under dev->power.lock with interrupts disabled.
436  */
437 static int rpm_idle(struct device *dev, int rpmflags)
438 {
439 	int (*callback)(struct device *);
440 	int retval;
441 
442 	trace_rpm_idle_rcuidle(dev, rpmflags);
443 	retval = rpm_check_suspend_allowed(dev);
444 	if (retval < 0)
445 		;	/* Conditions are wrong. */
446 
447 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
448 	else if (dev->power.runtime_status != RPM_ACTIVE)
449 		retval = -EAGAIN;
450 
451 	/*
452 	 * Any pending request other than an idle notification takes
453 	 * precedence over us, except that the timer may be running.
454 	 */
455 	else if (dev->power.request_pending &&
456 	    dev->power.request > RPM_REQ_IDLE)
457 		retval = -EAGAIN;
458 
459 	/* Act as though RPM_NOWAIT is always set. */
460 	else if (dev->power.idle_notification)
461 		retval = -EINPROGRESS;
462 	if (retval)
463 		goto out;
464 
465 	/* Pending requests need to be canceled. */
466 	dev->power.request = RPM_REQ_NONE;
467 
468 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
469 
470 	/* If no callback assume success. */
471 	if (!callback || dev->power.no_callbacks)
472 		goto out;
473 
474 	/* Carry out an asynchronous or a synchronous idle notification. */
475 	if (rpmflags & RPM_ASYNC) {
476 		dev->power.request = RPM_REQ_IDLE;
477 		if (!dev->power.request_pending) {
478 			dev->power.request_pending = true;
479 			queue_work(pm_wq, &dev->power.work);
480 		}
481 		trace_rpm_return_int_rcuidle(dev, _THIS_IP_, 0);
482 		return 0;
483 	}
484 
485 	dev->power.idle_notification = true;
486 
487 	retval = __rpm_callback(callback, dev);
488 
489 	dev->power.idle_notification = false;
490 	wake_up_all(&dev->power.wait_queue);
491 
492  out:
493 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
494 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
495 }
496 
497 /**
498  * rpm_callback - Run a given runtime PM callback for a given device.
499  * @cb: Runtime PM callback to run.
500  * @dev: Device to run the callback for.
501  */
502 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
503 {
504 	int retval;
505 
506 	if (dev->power.memalloc_noio) {
507 		unsigned int noio_flag;
508 
509 		/*
510 		 * Deadlock might be caused if memory allocation with
511 		 * GFP_KERNEL happens inside runtime_suspend and
512 		 * runtime_resume callbacks of one block device's
513 		 * ancestor or the block device itself. Network
514 		 * device might be thought as part of iSCSI block
515 		 * device, so network device and its ancestor should
516 		 * be marked as memalloc_noio too.
517 		 */
518 		noio_flag = memalloc_noio_save();
519 		retval = __rpm_callback(cb, dev);
520 		memalloc_noio_restore(noio_flag);
521 	} else {
522 		retval = __rpm_callback(cb, dev);
523 	}
524 
525 	dev->power.runtime_error = retval;
526 	return retval != -EACCES ? retval : -EIO;
527 }
528 
529 /**
530  * rpm_suspend - Carry out runtime suspend of given device.
531  * @dev: Device to suspend.
532  * @rpmflags: Flag bits.
533  *
534  * Check if the device's runtime PM status allows it to be suspended.
535  * Cancel a pending idle notification, autosuspend or suspend. If
536  * another suspend has been started earlier, either return immediately
537  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
538  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
539  * otherwise run the ->runtime_suspend() callback directly. When
540  * ->runtime_suspend succeeded, if a deferred resume was requested while
541  * the callback was running then carry it out, otherwise send an idle
542  * notification for its parent (if the suspend succeeded and both
543  * ignore_children of parent->power and irq_safe of dev->power are not set).
544  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
545  * flag is set and the next autosuspend-delay expiration time is in the
546  * future, schedule another autosuspend attempt.
547  *
548  * This function must be called under dev->power.lock with interrupts disabled.
549  */
550 static int rpm_suspend(struct device *dev, int rpmflags)
551 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
552 {
553 	int (*callback)(struct device *);
554 	struct device *parent = NULL;
555 	int retval;
556 
557 	trace_rpm_suspend_rcuidle(dev, rpmflags);
558 
559  repeat:
560 	retval = rpm_check_suspend_allowed(dev);
561 	if (retval < 0)
562 		goto out;	/* Conditions are wrong. */
563 
564 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
565 	if (dev->power.runtime_status == RPM_RESUMING && !(rpmflags & RPM_ASYNC))
566 		retval = -EAGAIN;
567 	if (retval)
568 		goto out;
569 
570 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
571 	if ((rpmflags & RPM_AUTO)
572 	    && dev->power.runtime_status != RPM_SUSPENDING) {
573 		u64 expires = pm_runtime_autosuspend_expiration(dev);
574 
575 		if (expires != 0) {
576 			/* Pending requests need to be canceled. */
577 			dev->power.request = RPM_REQ_NONE;
578 
579 			/*
580 			 * Optimization: If the timer is already running and is
581 			 * set to expire at or before the autosuspend delay,
582 			 * avoid the overhead of resetting it.  Just let it
583 			 * expire; pm_suspend_timer_fn() will take care of the
584 			 * rest.
585 			 */
586 			if (!(dev->power.timer_expires &&
587 					dev->power.timer_expires <= expires)) {
588 				/*
589 				 * We add a slack of 25% to gather wakeups
590 				 * without sacrificing the granularity.
591 				 */
592 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
593 						    (NSEC_PER_MSEC >> 2);
594 
595 				dev->power.timer_expires = expires;
596 				hrtimer_start_range_ns(&dev->power.suspend_timer,
597 						ns_to_ktime(expires),
598 						slack,
599 						HRTIMER_MODE_ABS);
600 			}
601 			dev->power.timer_autosuspends = 1;
602 			goto out;
603 		}
604 	}
605 
606 	/* Other scheduled or pending requests need to be canceled. */
607 	pm_runtime_cancel_pending(dev);
608 
609 	if (dev->power.runtime_status == RPM_SUSPENDING) {
610 		DEFINE_WAIT(wait);
611 
612 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
613 			retval = -EINPROGRESS;
614 			goto out;
615 		}
616 
617 		if (dev->power.irq_safe) {
618 			spin_unlock(&dev->power.lock);
619 
620 			cpu_relax();
621 
622 			spin_lock(&dev->power.lock);
623 			goto repeat;
624 		}
625 
626 		/* Wait for the other suspend running in parallel with us. */
627 		for (;;) {
628 			prepare_to_wait(&dev->power.wait_queue, &wait,
629 					TASK_UNINTERRUPTIBLE);
630 			if (dev->power.runtime_status != RPM_SUSPENDING)
631 				break;
632 
633 			spin_unlock_irq(&dev->power.lock);
634 
635 			schedule();
636 
637 			spin_lock_irq(&dev->power.lock);
638 		}
639 		finish_wait(&dev->power.wait_queue, &wait);
640 		goto repeat;
641 	}
642 
643 	if (dev->power.no_callbacks)
644 		goto no_callback;	/* Assume success. */
645 
646 	/* Carry out an asynchronous or a synchronous suspend. */
647 	if (rpmflags & RPM_ASYNC) {
648 		dev->power.request = (rpmflags & RPM_AUTO) ?
649 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
650 		if (!dev->power.request_pending) {
651 			dev->power.request_pending = true;
652 			queue_work(pm_wq, &dev->power.work);
653 		}
654 		goto out;
655 	}
656 
657 	__update_runtime_status(dev, RPM_SUSPENDING);
658 
659 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
660 
661 	dev_pm_enable_wake_irq_check(dev, true);
662 	retval = rpm_callback(callback, dev);
663 	if (retval)
664 		goto fail;
665 
666 	dev_pm_enable_wake_irq_complete(dev);
667 
668  no_callback:
669 	__update_runtime_status(dev, RPM_SUSPENDED);
670 	pm_runtime_deactivate_timer(dev);
671 
672 	if (dev->parent) {
673 		parent = dev->parent;
674 		atomic_add_unless(&parent->power.child_count, -1, 0);
675 	}
676 	wake_up_all(&dev->power.wait_queue);
677 
678 	if (dev->power.deferred_resume) {
679 		dev->power.deferred_resume = false;
680 		rpm_resume(dev, 0);
681 		retval = -EAGAIN;
682 		goto out;
683 	}
684 
685 	if (dev->power.irq_safe)
686 		goto out;
687 
688 	/* Maybe the parent is now able to suspend. */
689 	if (parent && !parent->power.ignore_children) {
690 		spin_unlock(&dev->power.lock);
691 
692 		spin_lock(&parent->power.lock);
693 		rpm_idle(parent, RPM_ASYNC);
694 		spin_unlock(&parent->power.lock);
695 
696 		spin_lock(&dev->power.lock);
697 	}
698 	/* Maybe the suppliers are now able to suspend. */
699 	if (dev->power.links_count > 0) {
700 		spin_unlock_irq(&dev->power.lock);
701 
702 		rpm_suspend_suppliers(dev);
703 
704 		spin_lock_irq(&dev->power.lock);
705 	}
706 
707  out:
708 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
709 
710 	return retval;
711 
712  fail:
713 	dev_pm_disable_wake_irq_check(dev, true);
714 	__update_runtime_status(dev, RPM_ACTIVE);
715 	dev->power.deferred_resume = false;
716 	wake_up_all(&dev->power.wait_queue);
717 
718 	if (retval == -EAGAIN || retval == -EBUSY) {
719 		dev->power.runtime_error = 0;
720 
721 		/*
722 		 * If the callback routine failed an autosuspend, and
723 		 * if the last_busy time has been updated so that there
724 		 * is a new autosuspend expiration time, automatically
725 		 * reschedule another autosuspend.
726 		 */
727 		if ((rpmflags & RPM_AUTO) &&
728 		    pm_runtime_autosuspend_expiration(dev) != 0)
729 			goto repeat;
730 	} else {
731 		pm_runtime_cancel_pending(dev);
732 	}
733 	goto out;
734 }
735 
736 /**
737  * rpm_resume - Carry out runtime resume of given device.
738  * @dev: Device to resume.
739  * @rpmflags: Flag bits.
740  *
741  * Check if the device's runtime PM status allows it to be resumed.  Cancel
742  * any scheduled or pending requests.  If another resume has been started
743  * earlier, either return immediately or wait for it to finish, depending on the
744  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
745  * parallel with this function, either tell the other process to resume after
746  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
747  * flag is set then queue a resume request; otherwise run the
748  * ->runtime_resume() callback directly.  Queue an idle notification for the
749  * device if the resume succeeded.
750  *
751  * This function must be called under dev->power.lock with interrupts disabled.
752  */
753 static int rpm_resume(struct device *dev, int rpmflags)
754 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
755 {
756 	int (*callback)(struct device *);
757 	struct device *parent = NULL;
758 	int retval = 0;
759 
760 	trace_rpm_resume_rcuidle(dev, rpmflags);
761 
762  repeat:
763 	if (dev->power.runtime_error) {
764 		retval = -EINVAL;
765 	} else if (dev->power.disable_depth > 0) {
766 		if (dev->power.runtime_status == RPM_ACTIVE &&
767 		    dev->power.last_status == RPM_ACTIVE)
768 			retval = 1;
769 		else
770 			retval = -EACCES;
771 	}
772 	if (retval)
773 		goto out;
774 
775 	/*
776 	 * Other scheduled or pending requests need to be canceled.  Small
777 	 * optimization: If an autosuspend timer is running, leave it running
778 	 * rather than cancelling it now only to restart it again in the near
779 	 * future.
780 	 */
781 	dev->power.request = RPM_REQ_NONE;
782 	if (!dev->power.timer_autosuspends)
783 		pm_runtime_deactivate_timer(dev);
784 
785 	if (dev->power.runtime_status == RPM_ACTIVE) {
786 		retval = 1;
787 		goto out;
788 	}
789 
790 	if (dev->power.runtime_status == RPM_RESUMING
791 	    || dev->power.runtime_status == RPM_SUSPENDING) {
792 		DEFINE_WAIT(wait);
793 
794 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
795 			if (dev->power.runtime_status == RPM_SUSPENDING)
796 				dev->power.deferred_resume = true;
797 			else
798 				retval = -EINPROGRESS;
799 			goto out;
800 		}
801 
802 		if (dev->power.irq_safe) {
803 			spin_unlock(&dev->power.lock);
804 
805 			cpu_relax();
806 
807 			spin_lock(&dev->power.lock);
808 			goto repeat;
809 		}
810 
811 		/* Wait for the operation carried out in parallel with us. */
812 		for (;;) {
813 			prepare_to_wait(&dev->power.wait_queue, &wait,
814 					TASK_UNINTERRUPTIBLE);
815 			if (dev->power.runtime_status != RPM_RESUMING
816 			    && dev->power.runtime_status != RPM_SUSPENDING)
817 				break;
818 
819 			spin_unlock_irq(&dev->power.lock);
820 
821 			schedule();
822 
823 			spin_lock_irq(&dev->power.lock);
824 		}
825 		finish_wait(&dev->power.wait_queue, &wait);
826 		goto repeat;
827 	}
828 
829 	/*
830 	 * See if we can skip waking up the parent.  This is safe only if
831 	 * power.no_callbacks is set, because otherwise we don't know whether
832 	 * the resume will actually succeed.
833 	 */
834 	if (dev->power.no_callbacks && !parent && dev->parent) {
835 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
836 		if (dev->parent->power.disable_depth > 0
837 		    || dev->parent->power.ignore_children
838 		    || dev->parent->power.runtime_status == RPM_ACTIVE) {
839 			atomic_inc(&dev->parent->power.child_count);
840 			spin_unlock(&dev->parent->power.lock);
841 			retval = 1;
842 			goto no_callback;	/* Assume success. */
843 		}
844 		spin_unlock(&dev->parent->power.lock);
845 	}
846 
847 	/* Carry out an asynchronous or a synchronous resume. */
848 	if (rpmflags & RPM_ASYNC) {
849 		dev->power.request = RPM_REQ_RESUME;
850 		if (!dev->power.request_pending) {
851 			dev->power.request_pending = true;
852 			queue_work(pm_wq, &dev->power.work);
853 		}
854 		retval = 0;
855 		goto out;
856 	}
857 
858 	if (!parent && dev->parent) {
859 		/*
860 		 * Increment the parent's usage counter and resume it if
861 		 * necessary.  Not needed if dev is irq-safe; then the
862 		 * parent is permanently resumed.
863 		 */
864 		parent = dev->parent;
865 		if (dev->power.irq_safe)
866 			goto skip_parent;
867 		spin_unlock(&dev->power.lock);
868 
869 		pm_runtime_get_noresume(parent);
870 
871 		spin_lock(&parent->power.lock);
872 		/*
873 		 * Resume the parent if it has runtime PM enabled and not been
874 		 * set to ignore its children.
875 		 */
876 		if (!parent->power.disable_depth
877 		    && !parent->power.ignore_children) {
878 			rpm_resume(parent, 0);
879 			if (parent->power.runtime_status != RPM_ACTIVE)
880 				retval = -EBUSY;
881 		}
882 		spin_unlock(&parent->power.lock);
883 
884 		spin_lock(&dev->power.lock);
885 		if (retval)
886 			goto out;
887 		goto repeat;
888 	}
889  skip_parent:
890 
891 	if (dev->power.no_callbacks)
892 		goto no_callback;	/* Assume success. */
893 
894 	__update_runtime_status(dev, RPM_RESUMING);
895 
896 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
897 
898 	dev_pm_disable_wake_irq_check(dev, false);
899 	retval = rpm_callback(callback, dev);
900 	if (retval) {
901 		__update_runtime_status(dev, RPM_SUSPENDED);
902 		pm_runtime_cancel_pending(dev);
903 		dev_pm_enable_wake_irq_check(dev, false);
904 	} else {
905  no_callback:
906 		__update_runtime_status(dev, RPM_ACTIVE);
907 		pm_runtime_mark_last_busy(dev);
908 		if (parent)
909 			atomic_inc(&parent->power.child_count);
910 	}
911 	wake_up_all(&dev->power.wait_queue);
912 
913 	if (retval >= 0)
914 		rpm_idle(dev, RPM_ASYNC);
915 
916  out:
917 	if (parent && !dev->power.irq_safe) {
918 		spin_unlock_irq(&dev->power.lock);
919 
920 		pm_runtime_put(parent);
921 
922 		spin_lock_irq(&dev->power.lock);
923 	}
924 
925 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
926 
927 	return retval;
928 }
929 
930 /**
931  * pm_runtime_work - Universal runtime PM work function.
932  * @work: Work structure used for scheduling the execution of this function.
933  *
934  * Use @work to get the device object the work is to be done for, determine what
935  * is to be done and execute the appropriate runtime PM function.
936  */
937 static void pm_runtime_work(struct work_struct *work)
938 {
939 	struct device *dev = container_of(work, struct device, power.work);
940 	enum rpm_request req;
941 
942 	spin_lock_irq(&dev->power.lock);
943 
944 	if (!dev->power.request_pending)
945 		goto out;
946 
947 	req = dev->power.request;
948 	dev->power.request = RPM_REQ_NONE;
949 	dev->power.request_pending = false;
950 
951 	switch (req) {
952 	case RPM_REQ_NONE:
953 		break;
954 	case RPM_REQ_IDLE:
955 		rpm_idle(dev, RPM_NOWAIT);
956 		break;
957 	case RPM_REQ_SUSPEND:
958 		rpm_suspend(dev, RPM_NOWAIT);
959 		break;
960 	case RPM_REQ_AUTOSUSPEND:
961 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
962 		break;
963 	case RPM_REQ_RESUME:
964 		rpm_resume(dev, RPM_NOWAIT);
965 		break;
966 	}
967 
968  out:
969 	spin_unlock_irq(&dev->power.lock);
970 }
971 
972 /**
973  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
974  * @timer: hrtimer used by pm_schedule_suspend().
975  *
976  * Check if the time is right and queue a suspend request.
977  */
978 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
979 {
980 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
981 	unsigned long flags;
982 	u64 expires;
983 
984 	spin_lock_irqsave(&dev->power.lock, flags);
985 
986 	expires = dev->power.timer_expires;
987 	/*
988 	 * If 'expires' is after the current time, we've been called
989 	 * too early.
990 	 */
991 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
992 		dev->power.timer_expires = 0;
993 		rpm_suspend(dev, dev->power.timer_autosuspends ?
994 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
995 	}
996 
997 	spin_unlock_irqrestore(&dev->power.lock, flags);
998 
999 	return HRTIMER_NORESTART;
1000 }
1001 
1002 /**
1003  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
1004  * @dev: Device to suspend.
1005  * @delay: Time to wait before submitting a suspend request, in milliseconds.
1006  */
1007 int pm_schedule_suspend(struct device *dev, unsigned int delay)
1008 {
1009 	unsigned long flags;
1010 	u64 expires;
1011 	int retval;
1012 
1013 	spin_lock_irqsave(&dev->power.lock, flags);
1014 
1015 	if (!delay) {
1016 		retval = rpm_suspend(dev, RPM_ASYNC);
1017 		goto out;
1018 	}
1019 
1020 	retval = rpm_check_suspend_allowed(dev);
1021 	if (retval)
1022 		goto out;
1023 
1024 	/* Other scheduled or pending requests need to be canceled. */
1025 	pm_runtime_cancel_pending(dev);
1026 
1027 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
1028 	dev->power.timer_expires = expires;
1029 	dev->power.timer_autosuspends = 0;
1030 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
1031 
1032  out:
1033 	spin_unlock_irqrestore(&dev->power.lock, flags);
1034 
1035 	return retval;
1036 }
1037 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
1038 
1039 static int rpm_drop_usage_count(struct device *dev)
1040 {
1041 	int ret;
1042 
1043 	ret = atomic_sub_return(1, &dev->power.usage_count);
1044 	if (ret >= 0)
1045 		return ret;
1046 
1047 	/*
1048 	 * Because rpm_resume() does not check the usage counter, it will resume
1049 	 * the device even if the usage counter is 0 or negative, so it is
1050 	 * sufficient to increment the usage counter here to reverse the change
1051 	 * made above.
1052 	 */
1053 	atomic_inc(&dev->power.usage_count);
1054 	dev_warn(dev, "Runtime PM usage count underflow!\n");
1055 	return -EINVAL;
1056 }
1057 
1058 /**
1059  * __pm_runtime_idle - Entry point for runtime idle operations.
1060  * @dev: Device to send idle notification for.
1061  * @rpmflags: Flag bits.
1062  *
1063  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1064  * return immediately if it is larger than zero (if it becomes negative, log a
1065  * warning, increment it, and return an error).  Then carry out an idle
1066  * notification, either synchronous or asynchronous.
1067  *
1068  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1069  * or if pm_runtime_irq_safe() has been called.
1070  */
1071 int __pm_runtime_idle(struct device *dev, int rpmflags)
1072 {
1073 	unsigned long flags;
1074 	int retval;
1075 
1076 	if (rpmflags & RPM_GET_PUT) {
1077 		retval = rpm_drop_usage_count(dev);
1078 		if (retval < 0) {
1079 			return retval;
1080 		} else if (retval > 0) {
1081 			trace_rpm_usage_rcuidle(dev, rpmflags);
1082 			return 0;
1083 		}
1084 	}
1085 
1086 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1087 
1088 	spin_lock_irqsave(&dev->power.lock, flags);
1089 	retval = rpm_idle(dev, rpmflags);
1090 	spin_unlock_irqrestore(&dev->power.lock, flags);
1091 
1092 	return retval;
1093 }
1094 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1095 
1096 /**
1097  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1098  * @dev: Device to suspend.
1099  * @rpmflags: Flag bits.
1100  *
1101  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1102  * return immediately if it is larger than zero (if it becomes negative, log a
1103  * warning, increment it, and return an error).  Then carry out a suspend,
1104  * either synchronous or asynchronous.
1105  *
1106  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1107  * or if pm_runtime_irq_safe() has been called.
1108  */
1109 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1110 {
1111 	unsigned long flags;
1112 	int retval;
1113 
1114 	if (rpmflags & RPM_GET_PUT) {
1115 		retval = rpm_drop_usage_count(dev);
1116 		if (retval < 0) {
1117 			return retval;
1118 		} else if (retval > 0) {
1119 			trace_rpm_usage_rcuidle(dev, rpmflags);
1120 			return 0;
1121 		}
1122 	}
1123 
1124 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1125 
1126 	spin_lock_irqsave(&dev->power.lock, flags);
1127 	retval = rpm_suspend(dev, rpmflags);
1128 	spin_unlock_irqrestore(&dev->power.lock, flags);
1129 
1130 	return retval;
1131 }
1132 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1133 
1134 /**
1135  * __pm_runtime_resume - Entry point for runtime resume operations.
1136  * @dev: Device to resume.
1137  * @rpmflags: Flag bits.
1138  *
1139  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1140  * carry out a resume, either synchronous or asynchronous.
1141  *
1142  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1143  * or if pm_runtime_irq_safe() has been called.
1144  */
1145 int __pm_runtime_resume(struct device *dev, int rpmflags)
1146 {
1147 	unsigned long flags;
1148 	int retval;
1149 
1150 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1151 			dev->power.runtime_status != RPM_ACTIVE);
1152 
1153 	if (rpmflags & RPM_GET_PUT)
1154 		atomic_inc(&dev->power.usage_count);
1155 
1156 	spin_lock_irqsave(&dev->power.lock, flags);
1157 	retval = rpm_resume(dev, rpmflags);
1158 	spin_unlock_irqrestore(&dev->power.lock, flags);
1159 
1160 	return retval;
1161 }
1162 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1163 
1164 /**
1165  * pm_runtime_get_if_active - Conditionally bump up device usage counter.
1166  * @dev: Device to handle.
1167  * @ign_usage_count: Whether or not to look at the current usage counter value.
1168  *
1169  * Return -EINVAL if runtime PM is disabled for @dev.
1170  *
1171  * Otherwise, if the runtime PM status of @dev is %RPM_ACTIVE and either
1172  * @ign_usage_count is %true or the runtime PM usage counter of @dev is not
1173  * zero, increment the usage counter of @dev and return 1. Otherwise, return 0
1174  * without changing the usage counter.
1175  *
1176  * If @ign_usage_count is %true, this function can be used to prevent suspending
1177  * the device when its runtime PM status is %RPM_ACTIVE.
1178  *
1179  * If @ign_usage_count is %false, this function can be used to prevent
1180  * suspending the device when both its runtime PM status is %RPM_ACTIVE and its
1181  * runtime PM usage counter is not zero.
1182  *
1183  * The caller is responsible for decrementing the runtime PM usage counter of
1184  * @dev after this function has returned a positive value for it.
1185  */
1186 int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count)
1187 {
1188 	unsigned long flags;
1189 	int retval;
1190 
1191 	spin_lock_irqsave(&dev->power.lock, flags);
1192 	if (dev->power.disable_depth > 0) {
1193 		retval = -EINVAL;
1194 	} else if (dev->power.runtime_status != RPM_ACTIVE) {
1195 		retval = 0;
1196 	} else if (ign_usage_count) {
1197 		retval = 1;
1198 		atomic_inc(&dev->power.usage_count);
1199 	} else {
1200 		retval = atomic_inc_not_zero(&dev->power.usage_count);
1201 	}
1202 	trace_rpm_usage_rcuidle(dev, 0);
1203 	spin_unlock_irqrestore(&dev->power.lock, flags);
1204 
1205 	return retval;
1206 }
1207 EXPORT_SYMBOL_GPL(pm_runtime_get_if_active);
1208 
1209 /**
1210  * __pm_runtime_set_status - Set runtime PM status of a device.
1211  * @dev: Device to handle.
1212  * @status: New runtime PM status of the device.
1213  *
1214  * If runtime PM of the device is disabled or its power.runtime_error field is
1215  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1216  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1217  * However, if the device has a parent and the parent is not active, and the
1218  * parent's power.ignore_children flag is unset, the device's status cannot be
1219  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1220  *
1221  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1222  * and the device parent's counter of unsuspended children is modified to
1223  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1224  * notification request for the parent is submitted.
1225  *
1226  * If @dev has any suppliers (as reflected by device links to them), and @status
1227  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1228  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1229  * of the @status value) and the suppliers will be deacticated on exit.  The
1230  * error returned by the failing supplier activation will be returned in that
1231  * case.
1232  */
1233 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1234 {
1235 	struct device *parent = dev->parent;
1236 	bool notify_parent = false;
1237 	unsigned long flags;
1238 	int error = 0;
1239 
1240 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1241 		return -EINVAL;
1242 
1243 	spin_lock_irqsave(&dev->power.lock, flags);
1244 
1245 	/*
1246 	 * Prevent PM-runtime from being enabled for the device or return an
1247 	 * error if it is enabled already and working.
1248 	 */
1249 	if (dev->power.runtime_error || dev->power.disable_depth)
1250 		dev->power.disable_depth++;
1251 	else
1252 		error = -EAGAIN;
1253 
1254 	spin_unlock_irqrestore(&dev->power.lock, flags);
1255 
1256 	if (error)
1257 		return error;
1258 
1259 	/*
1260 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1261 	 * upfront regardless of the current status, because next time
1262 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1263 	 * involved will be dropped down to one anyway.
1264 	 */
1265 	if (status == RPM_ACTIVE) {
1266 		int idx = device_links_read_lock();
1267 
1268 		error = rpm_get_suppliers(dev);
1269 		if (error)
1270 			status = RPM_SUSPENDED;
1271 
1272 		device_links_read_unlock(idx);
1273 	}
1274 
1275 	spin_lock_irqsave(&dev->power.lock, flags);
1276 
1277 	if (dev->power.runtime_status == status || !parent)
1278 		goto out_set;
1279 
1280 	if (status == RPM_SUSPENDED) {
1281 		atomic_add_unless(&parent->power.child_count, -1, 0);
1282 		notify_parent = !parent->power.ignore_children;
1283 	} else {
1284 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1285 
1286 		/*
1287 		 * It is invalid to put an active child under a parent that is
1288 		 * not active, has runtime PM enabled and the
1289 		 * 'power.ignore_children' flag unset.
1290 		 */
1291 		if (!parent->power.disable_depth
1292 		    && !parent->power.ignore_children
1293 		    && parent->power.runtime_status != RPM_ACTIVE) {
1294 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1295 				dev_name(dev),
1296 				dev_name(parent));
1297 			error = -EBUSY;
1298 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1299 			atomic_inc(&parent->power.child_count);
1300 		}
1301 
1302 		spin_unlock(&parent->power.lock);
1303 
1304 		if (error) {
1305 			status = RPM_SUSPENDED;
1306 			goto out;
1307 		}
1308 	}
1309 
1310  out_set:
1311 	__update_runtime_status(dev, status);
1312 	if (!error)
1313 		dev->power.runtime_error = 0;
1314 
1315  out:
1316 	spin_unlock_irqrestore(&dev->power.lock, flags);
1317 
1318 	if (notify_parent)
1319 		pm_request_idle(parent);
1320 
1321 	if (status == RPM_SUSPENDED) {
1322 		int idx = device_links_read_lock();
1323 
1324 		rpm_put_suppliers(dev);
1325 
1326 		device_links_read_unlock(idx);
1327 	}
1328 
1329 	pm_runtime_enable(dev);
1330 
1331 	return error;
1332 }
1333 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1334 
1335 /**
1336  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1337  * @dev: Device to handle.
1338  *
1339  * Flush all pending requests for the device from pm_wq and wait for all
1340  * runtime PM operations involving the device in progress to complete.
1341  *
1342  * Should be called under dev->power.lock with interrupts disabled.
1343  */
1344 static void __pm_runtime_barrier(struct device *dev)
1345 {
1346 	pm_runtime_deactivate_timer(dev);
1347 
1348 	if (dev->power.request_pending) {
1349 		dev->power.request = RPM_REQ_NONE;
1350 		spin_unlock_irq(&dev->power.lock);
1351 
1352 		cancel_work_sync(&dev->power.work);
1353 
1354 		spin_lock_irq(&dev->power.lock);
1355 		dev->power.request_pending = false;
1356 	}
1357 
1358 	if (dev->power.runtime_status == RPM_SUSPENDING
1359 	    || dev->power.runtime_status == RPM_RESUMING
1360 	    || dev->power.idle_notification) {
1361 		DEFINE_WAIT(wait);
1362 
1363 		/* Suspend, wake-up or idle notification in progress. */
1364 		for (;;) {
1365 			prepare_to_wait(&dev->power.wait_queue, &wait,
1366 					TASK_UNINTERRUPTIBLE);
1367 			if (dev->power.runtime_status != RPM_SUSPENDING
1368 			    && dev->power.runtime_status != RPM_RESUMING
1369 			    && !dev->power.idle_notification)
1370 				break;
1371 			spin_unlock_irq(&dev->power.lock);
1372 
1373 			schedule();
1374 
1375 			spin_lock_irq(&dev->power.lock);
1376 		}
1377 		finish_wait(&dev->power.wait_queue, &wait);
1378 	}
1379 }
1380 
1381 /**
1382  * pm_runtime_barrier - Flush pending requests and wait for completions.
1383  * @dev: Device to handle.
1384  *
1385  * Prevent the device from being suspended by incrementing its usage counter and
1386  * if there's a pending resume request for the device, wake the device up.
1387  * Next, make sure that all pending requests for the device have been flushed
1388  * from pm_wq and wait for all runtime PM operations involving the device in
1389  * progress to complete.
1390  *
1391  * Return value:
1392  * 1, if there was a resume request pending and the device had to be woken up,
1393  * 0, otherwise
1394  */
1395 int pm_runtime_barrier(struct device *dev)
1396 {
1397 	int retval = 0;
1398 
1399 	pm_runtime_get_noresume(dev);
1400 	spin_lock_irq(&dev->power.lock);
1401 
1402 	if (dev->power.request_pending
1403 	    && dev->power.request == RPM_REQ_RESUME) {
1404 		rpm_resume(dev, 0);
1405 		retval = 1;
1406 	}
1407 
1408 	__pm_runtime_barrier(dev);
1409 
1410 	spin_unlock_irq(&dev->power.lock);
1411 	pm_runtime_put_noidle(dev);
1412 
1413 	return retval;
1414 }
1415 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1416 
1417 /**
1418  * __pm_runtime_disable - Disable runtime PM of a device.
1419  * @dev: Device to handle.
1420  * @check_resume: If set, check if there's a resume request for the device.
1421  *
1422  * Increment power.disable_depth for the device and if it was zero previously,
1423  * cancel all pending runtime PM requests for the device and wait for all
1424  * operations in progress to complete.  The device can be either active or
1425  * suspended after its runtime PM has been disabled.
1426  *
1427  * If @check_resume is set and there's a resume request pending when
1428  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1429  * function will wake up the device before disabling its runtime PM.
1430  */
1431 void __pm_runtime_disable(struct device *dev, bool check_resume)
1432 {
1433 	spin_lock_irq(&dev->power.lock);
1434 
1435 	if (dev->power.disable_depth > 0) {
1436 		dev->power.disable_depth++;
1437 		goto out;
1438 	}
1439 
1440 	/*
1441 	 * Wake up the device if there's a resume request pending, because that
1442 	 * means there probably is some I/O to process and disabling runtime PM
1443 	 * shouldn't prevent the device from processing the I/O.
1444 	 */
1445 	if (check_resume && dev->power.request_pending
1446 	    && dev->power.request == RPM_REQ_RESUME) {
1447 		/*
1448 		 * Prevent suspends and idle notifications from being carried
1449 		 * out after we have woken up the device.
1450 		 */
1451 		pm_runtime_get_noresume(dev);
1452 
1453 		rpm_resume(dev, 0);
1454 
1455 		pm_runtime_put_noidle(dev);
1456 	}
1457 
1458 	/* Update time accounting before disabling PM-runtime. */
1459 	update_pm_runtime_accounting(dev);
1460 
1461 	if (!dev->power.disable_depth++) {
1462 		__pm_runtime_barrier(dev);
1463 		dev->power.last_status = dev->power.runtime_status;
1464 	}
1465 
1466  out:
1467 	spin_unlock_irq(&dev->power.lock);
1468 }
1469 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1470 
1471 /**
1472  * pm_runtime_enable - Enable runtime PM of a device.
1473  * @dev: Device to handle.
1474  */
1475 void pm_runtime_enable(struct device *dev)
1476 {
1477 	unsigned long flags;
1478 
1479 	spin_lock_irqsave(&dev->power.lock, flags);
1480 
1481 	if (!dev->power.disable_depth) {
1482 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1483 		goto out;
1484 	}
1485 
1486 	if (--dev->power.disable_depth > 0)
1487 		goto out;
1488 
1489 	dev->power.last_status = RPM_INVALID;
1490 	dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1491 
1492 	if (dev->power.runtime_status == RPM_SUSPENDED &&
1493 	    !dev->power.ignore_children &&
1494 	    atomic_read(&dev->power.child_count) > 0)
1495 		dev_warn(dev, "Enabling runtime PM for inactive device with active children\n");
1496 
1497 out:
1498 	spin_unlock_irqrestore(&dev->power.lock, flags);
1499 }
1500 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1501 
1502 static void pm_runtime_disable_action(void *data)
1503 {
1504 	pm_runtime_dont_use_autosuspend(data);
1505 	pm_runtime_disable(data);
1506 }
1507 
1508 /**
1509  * devm_pm_runtime_enable - devres-enabled version of pm_runtime_enable.
1510  *
1511  * NOTE: this will also handle calling pm_runtime_dont_use_autosuspend() for
1512  * you at driver exit time if needed.
1513  *
1514  * @dev: Device to handle.
1515  */
1516 int devm_pm_runtime_enable(struct device *dev)
1517 {
1518 	pm_runtime_enable(dev);
1519 
1520 	return devm_add_action_or_reset(dev, pm_runtime_disable_action, dev);
1521 }
1522 EXPORT_SYMBOL_GPL(devm_pm_runtime_enable);
1523 
1524 /**
1525  * pm_runtime_forbid - Block runtime PM of a device.
1526  * @dev: Device to handle.
1527  *
1528  * Increase the device's usage count and clear its power.runtime_auto flag,
1529  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1530  * for it.
1531  */
1532 void pm_runtime_forbid(struct device *dev)
1533 {
1534 	spin_lock_irq(&dev->power.lock);
1535 	if (!dev->power.runtime_auto)
1536 		goto out;
1537 
1538 	dev->power.runtime_auto = false;
1539 	atomic_inc(&dev->power.usage_count);
1540 	rpm_resume(dev, 0);
1541 
1542  out:
1543 	spin_unlock_irq(&dev->power.lock);
1544 }
1545 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1546 
1547 /**
1548  * pm_runtime_allow - Unblock runtime PM of a device.
1549  * @dev: Device to handle.
1550  *
1551  * Decrease the device's usage count and set its power.runtime_auto flag.
1552  */
1553 void pm_runtime_allow(struct device *dev)
1554 {
1555 	int ret;
1556 
1557 	spin_lock_irq(&dev->power.lock);
1558 	if (dev->power.runtime_auto)
1559 		goto out;
1560 
1561 	dev->power.runtime_auto = true;
1562 	ret = rpm_drop_usage_count(dev);
1563 	if (ret == 0)
1564 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1565 	else if (ret > 0)
1566 		trace_rpm_usage_rcuidle(dev, RPM_AUTO | RPM_ASYNC);
1567 
1568  out:
1569 	spin_unlock_irq(&dev->power.lock);
1570 }
1571 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1572 
1573 /**
1574  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1575  * @dev: Device to handle.
1576  *
1577  * Set the power.no_callbacks flag, which tells the PM core that this
1578  * device is power-managed through its parent and has no runtime PM
1579  * callbacks of its own.  The runtime sysfs attributes will be removed.
1580  */
1581 void pm_runtime_no_callbacks(struct device *dev)
1582 {
1583 	spin_lock_irq(&dev->power.lock);
1584 	dev->power.no_callbacks = 1;
1585 	spin_unlock_irq(&dev->power.lock);
1586 	if (device_is_registered(dev))
1587 		rpm_sysfs_remove(dev);
1588 }
1589 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1590 
1591 /**
1592  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1593  * @dev: Device to handle
1594  *
1595  * Set the power.irq_safe flag, which tells the PM core that the
1596  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1597  * always be invoked with the spinlock held and interrupts disabled.  It also
1598  * causes the parent's usage counter to be permanently incremented, preventing
1599  * the parent from runtime suspending -- otherwise an irq-safe child might have
1600  * to wait for a non-irq-safe parent.
1601  */
1602 void pm_runtime_irq_safe(struct device *dev)
1603 {
1604 	if (dev->parent)
1605 		pm_runtime_get_sync(dev->parent);
1606 	spin_lock_irq(&dev->power.lock);
1607 	dev->power.irq_safe = 1;
1608 	spin_unlock_irq(&dev->power.lock);
1609 }
1610 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1611 
1612 /**
1613  * update_autosuspend - Handle a change to a device's autosuspend settings.
1614  * @dev: Device to handle.
1615  * @old_delay: The former autosuspend_delay value.
1616  * @old_use: The former use_autosuspend value.
1617  *
1618  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1619  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1620  *
1621  * This function must be called under dev->power.lock with interrupts disabled.
1622  */
1623 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1624 {
1625 	int delay = dev->power.autosuspend_delay;
1626 
1627 	/* Should runtime suspend be prevented now? */
1628 	if (dev->power.use_autosuspend && delay < 0) {
1629 
1630 		/* If it used to be allowed then prevent it. */
1631 		if (!old_use || old_delay >= 0) {
1632 			atomic_inc(&dev->power.usage_count);
1633 			rpm_resume(dev, 0);
1634 		} else {
1635 			trace_rpm_usage_rcuidle(dev, 0);
1636 		}
1637 	}
1638 
1639 	/* Runtime suspend should be allowed now. */
1640 	else {
1641 
1642 		/* If it used to be prevented then allow it. */
1643 		if (old_use && old_delay < 0)
1644 			atomic_dec(&dev->power.usage_count);
1645 
1646 		/* Maybe we can autosuspend now. */
1647 		rpm_idle(dev, RPM_AUTO);
1648 	}
1649 }
1650 
1651 /**
1652  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1653  * @dev: Device to handle.
1654  * @delay: Value of the new delay in milliseconds.
1655  *
1656  * Set the device's power.autosuspend_delay value.  If it changes to negative
1657  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1658  * changes the other way, allow runtime suspends.
1659  */
1660 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1661 {
1662 	int old_delay, old_use;
1663 
1664 	spin_lock_irq(&dev->power.lock);
1665 	old_delay = dev->power.autosuspend_delay;
1666 	old_use = dev->power.use_autosuspend;
1667 	dev->power.autosuspend_delay = delay;
1668 	update_autosuspend(dev, old_delay, old_use);
1669 	spin_unlock_irq(&dev->power.lock);
1670 }
1671 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1672 
1673 /**
1674  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1675  * @dev: Device to handle.
1676  * @use: New value for use_autosuspend.
1677  *
1678  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1679  * suspends as needed.
1680  */
1681 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1682 {
1683 	int old_delay, old_use;
1684 
1685 	spin_lock_irq(&dev->power.lock);
1686 	old_delay = dev->power.autosuspend_delay;
1687 	old_use = dev->power.use_autosuspend;
1688 	dev->power.use_autosuspend = use;
1689 	update_autosuspend(dev, old_delay, old_use);
1690 	spin_unlock_irq(&dev->power.lock);
1691 }
1692 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1693 
1694 /**
1695  * pm_runtime_init - Initialize runtime PM fields in given device object.
1696  * @dev: Device object to initialize.
1697  */
1698 void pm_runtime_init(struct device *dev)
1699 {
1700 	dev->power.runtime_status = RPM_SUSPENDED;
1701 	dev->power.last_status = RPM_INVALID;
1702 	dev->power.idle_notification = false;
1703 
1704 	dev->power.disable_depth = 1;
1705 	atomic_set(&dev->power.usage_count, 0);
1706 
1707 	dev->power.runtime_error = 0;
1708 
1709 	atomic_set(&dev->power.child_count, 0);
1710 	pm_suspend_ignore_children(dev, false);
1711 	dev->power.runtime_auto = true;
1712 
1713 	dev->power.request_pending = false;
1714 	dev->power.request = RPM_REQ_NONE;
1715 	dev->power.deferred_resume = false;
1716 	dev->power.needs_force_resume = 0;
1717 	INIT_WORK(&dev->power.work, pm_runtime_work);
1718 
1719 	dev->power.timer_expires = 0;
1720 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1721 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1722 
1723 	init_waitqueue_head(&dev->power.wait_queue);
1724 }
1725 
1726 /**
1727  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1728  * @dev: Device object to re-initialize.
1729  */
1730 void pm_runtime_reinit(struct device *dev)
1731 {
1732 	if (!pm_runtime_enabled(dev)) {
1733 		if (dev->power.runtime_status == RPM_ACTIVE)
1734 			pm_runtime_set_suspended(dev);
1735 		if (dev->power.irq_safe) {
1736 			spin_lock_irq(&dev->power.lock);
1737 			dev->power.irq_safe = 0;
1738 			spin_unlock_irq(&dev->power.lock);
1739 			if (dev->parent)
1740 				pm_runtime_put(dev->parent);
1741 		}
1742 	}
1743 }
1744 
1745 /**
1746  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1747  * @dev: Device object being removed from device hierarchy.
1748  */
1749 void pm_runtime_remove(struct device *dev)
1750 {
1751 	__pm_runtime_disable(dev, false);
1752 	pm_runtime_reinit(dev);
1753 }
1754 
1755 /**
1756  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1757  * @dev: Consumer device.
1758  */
1759 void pm_runtime_get_suppliers(struct device *dev)
1760 {
1761 	struct device_link *link;
1762 	int idx;
1763 
1764 	idx = device_links_read_lock();
1765 
1766 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1767 				device_links_read_lock_held())
1768 		if (link->flags & DL_FLAG_PM_RUNTIME) {
1769 			link->supplier_preactivated = true;
1770 			pm_runtime_get_sync(link->supplier);
1771 		}
1772 
1773 	device_links_read_unlock(idx);
1774 }
1775 
1776 /**
1777  * pm_runtime_put_suppliers - Drop references to supplier devices.
1778  * @dev: Consumer device.
1779  */
1780 void pm_runtime_put_suppliers(struct device *dev)
1781 {
1782 	struct device_link *link;
1783 	int idx;
1784 
1785 	idx = device_links_read_lock();
1786 
1787 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1788 				device_links_read_lock_held())
1789 		if (link->supplier_preactivated) {
1790 			link->supplier_preactivated = false;
1791 			pm_runtime_put(link->supplier);
1792 		}
1793 
1794 	device_links_read_unlock(idx);
1795 }
1796 
1797 void pm_runtime_new_link(struct device *dev)
1798 {
1799 	spin_lock_irq(&dev->power.lock);
1800 	dev->power.links_count++;
1801 	spin_unlock_irq(&dev->power.lock);
1802 }
1803 
1804 static void pm_runtime_drop_link_count(struct device *dev)
1805 {
1806 	spin_lock_irq(&dev->power.lock);
1807 	WARN_ON(dev->power.links_count == 0);
1808 	dev->power.links_count--;
1809 	spin_unlock_irq(&dev->power.lock);
1810 }
1811 
1812 /**
1813  * pm_runtime_drop_link - Prepare for device link removal.
1814  * @link: Device link going away.
1815  *
1816  * Drop the link count of the consumer end of @link and decrement the supplier
1817  * device's runtime PM usage counter as many times as needed to drop all of the
1818  * PM runtime reference to it from the consumer.
1819  */
1820 void pm_runtime_drop_link(struct device_link *link)
1821 {
1822 	if (!(link->flags & DL_FLAG_PM_RUNTIME))
1823 		return;
1824 
1825 	pm_runtime_drop_link_count(link->consumer);
1826 	pm_runtime_release_supplier(link);
1827 	pm_request_idle(link->supplier);
1828 }
1829 
1830 static bool pm_runtime_need_not_resume(struct device *dev)
1831 {
1832 	return atomic_read(&dev->power.usage_count) <= 1 &&
1833 		(atomic_read(&dev->power.child_count) == 0 ||
1834 		 dev->power.ignore_children);
1835 }
1836 
1837 /**
1838  * pm_runtime_force_suspend - Force a device into suspend state if needed.
1839  * @dev: Device to suspend.
1840  *
1841  * Disable runtime PM so we safely can check the device's runtime PM status and
1842  * if it is active, invoke its ->runtime_suspend callback to suspend it and
1843  * change its runtime PM status field to RPM_SUSPENDED.  Also, if the device's
1844  * usage and children counters don't indicate that the device was in use before
1845  * the system-wide transition under way, decrement its parent's children counter
1846  * (if there is a parent).  Keep runtime PM disabled to preserve the state
1847  * unless we encounter errors.
1848  *
1849  * Typically this function may be invoked from a system suspend callback to make
1850  * sure the device is put into low power state and it should only be used during
1851  * system-wide PM transitions to sleep states.  It assumes that the analogous
1852  * pm_runtime_force_resume() will be used to resume the device.
1853  */
1854 int pm_runtime_force_suspend(struct device *dev)
1855 {
1856 	int (*callback)(struct device *);
1857 	int ret;
1858 
1859 	pm_runtime_disable(dev);
1860 	if (pm_runtime_status_suspended(dev))
1861 		return 0;
1862 
1863 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1864 
1865 	dev_pm_enable_wake_irq_check(dev, true);
1866 	ret = callback ? callback(dev) : 0;
1867 	if (ret)
1868 		goto err;
1869 
1870 	dev_pm_enable_wake_irq_complete(dev);
1871 
1872 	/*
1873 	 * If the device can stay in suspend after the system-wide transition
1874 	 * to the working state that will follow, drop the children counter of
1875 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1876 	 * function will be called again for it in the meantime.
1877 	 */
1878 	if (pm_runtime_need_not_resume(dev)) {
1879 		pm_runtime_set_suspended(dev);
1880 	} else {
1881 		__update_runtime_status(dev, RPM_SUSPENDED);
1882 		dev->power.needs_force_resume = 1;
1883 	}
1884 
1885 	return 0;
1886 
1887 err:
1888 	dev_pm_disable_wake_irq_check(dev, true);
1889 	pm_runtime_enable(dev);
1890 	return ret;
1891 }
1892 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1893 
1894 /**
1895  * pm_runtime_force_resume - Force a device into resume state if needed.
1896  * @dev: Device to resume.
1897  *
1898  * Prior invoking this function we expect the user to have brought the device
1899  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1900  * those actions and bring the device into full power, if it is expected to be
1901  * used on system resume.  In the other case, we defer the resume to be managed
1902  * via runtime PM.
1903  *
1904  * Typically this function may be invoked from a system resume callback.
1905  */
1906 int pm_runtime_force_resume(struct device *dev)
1907 {
1908 	int (*callback)(struct device *);
1909 	int ret = 0;
1910 
1911 	if (!pm_runtime_status_suspended(dev) || !dev->power.needs_force_resume)
1912 		goto out;
1913 
1914 	/*
1915 	 * The value of the parent's children counter is correct already, so
1916 	 * just update the status of the device.
1917 	 */
1918 	__update_runtime_status(dev, RPM_ACTIVE);
1919 
1920 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1921 
1922 	dev_pm_disable_wake_irq_check(dev, false);
1923 	ret = callback ? callback(dev) : 0;
1924 	if (ret) {
1925 		pm_runtime_set_suspended(dev);
1926 		dev_pm_enable_wake_irq_check(dev, false);
1927 		goto out;
1928 	}
1929 
1930 	pm_runtime_mark_last_busy(dev);
1931 out:
1932 	dev->power.needs_force_resume = 0;
1933 	pm_runtime_enable(dev);
1934 	return ret;
1935 }
1936 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1937