xref: /openbmc/linux/drivers/base/power/runtime.c (revision 71de0a05)
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, dev_memalloc_noio)))
247 			break;
248 	}
249 	mutex_unlock(&dev_hotplug_mutex);
250 }
251 EXPORT_SYMBOL_GPL(pm_runtime_set_memalloc_noio);
252 
253 /**
254  * rpm_check_suspend_allowed - Test whether a device may be suspended.
255  * @dev: Device to test.
256  */
257 static int rpm_check_suspend_allowed(struct device *dev)
258 {
259 	int retval = 0;
260 
261 	if (dev->power.runtime_error)
262 		retval = -EINVAL;
263 	else if (dev->power.disable_depth > 0)
264 		retval = -EACCES;
265 	else if (atomic_read(&dev->power.usage_count))
266 		retval = -EAGAIN;
267 	else if (!dev->power.ignore_children && atomic_read(&dev->power.child_count))
268 		retval = -EBUSY;
269 
270 	/* Pending resume requests take precedence over suspends. */
271 	else if ((dev->power.deferred_resume &&
272 	    dev->power.runtime_status == RPM_SUSPENDING) ||
273 	    (dev->power.request_pending && dev->power.request == RPM_REQ_RESUME))
274 		retval = -EAGAIN;
275 	else if (__dev_pm_qos_resume_latency(dev) == 0)
276 		retval = -EPERM;
277 	else if (dev->power.runtime_status == RPM_SUSPENDED)
278 		retval = 1;
279 
280 	return retval;
281 }
282 
283 static int rpm_get_suppliers(struct device *dev)
284 {
285 	struct device_link *link;
286 
287 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
288 				device_links_read_lock_held()) {
289 		int retval;
290 
291 		if (!(link->flags & DL_FLAG_PM_RUNTIME))
292 			continue;
293 
294 		retval = pm_runtime_get_sync(link->supplier);
295 		/* Ignore suppliers with disabled runtime PM. */
296 		if (retval < 0 && retval != -EACCES) {
297 			pm_runtime_put_noidle(link->supplier);
298 			return retval;
299 		}
300 		refcount_inc(&link->rpm_active);
301 	}
302 	return 0;
303 }
304 
305 /**
306  * pm_runtime_release_supplier - Drop references to device link's supplier.
307  * @link: Target device link.
308  *
309  * Drop all runtime PM references associated with @link to its supplier device.
310  */
311 void pm_runtime_release_supplier(struct device_link *link)
312 {
313 	struct device *supplier = link->supplier;
314 
315 	/*
316 	 * The additional power.usage_count check is a safety net in case
317 	 * the rpm_active refcount becomes saturated, in which case
318 	 * refcount_dec_not_one() would return true forever, but it is not
319 	 * strictly necessary.
320 	 */
321 	while (refcount_dec_not_one(&link->rpm_active) &&
322 	       atomic_read(&supplier->power.usage_count) > 0)
323 		pm_runtime_put_noidle(supplier);
324 }
325 
326 static void __rpm_put_suppliers(struct device *dev, bool try_to_suspend)
327 {
328 	struct device_link *link;
329 
330 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
331 				device_links_read_lock_held()) {
332 		pm_runtime_release_supplier(link);
333 		if (try_to_suspend)
334 			pm_request_idle(link->supplier);
335 	}
336 }
337 
338 static void rpm_put_suppliers(struct device *dev)
339 {
340 	__rpm_put_suppliers(dev, true);
341 }
342 
343 static void rpm_suspend_suppliers(struct device *dev)
344 {
345 	struct device_link *link;
346 	int idx = device_links_read_lock();
347 
348 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
349 				device_links_read_lock_held())
350 		pm_request_idle(link->supplier);
351 
352 	device_links_read_unlock(idx);
353 }
354 
355 /**
356  * __rpm_callback - Run a given runtime PM callback for a given device.
357  * @cb: Runtime PM callback to run.
358  * @dev: Device to run the callback for.
359  */
360 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
361 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
362 {
363 	int retval = 0, idx;
364 	bool use_links = dev->power.links_count > 0;
365 
366 	if (dev->power.irq_safe) {
367 		spin_unlock(&dev->power.lock);
368 	} else {
369 		spin_unlock_irq(&dev->power.lock);
370 
371 		/*
372 		 * Resume suppliers if necessary.
373 		 *
374 		 * The device's runtime PM status cannot change until this
375 		 * routine returns, so it is safe to read the status outside of
376 		 * the lock.
377 		 */
378 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
379 			idx = device_links_read_lock();
380 
381 			retval = rpm_get_suppliers(dev);
382 			if (retval) {
383 				rpm_put_suppliers(dev);
384 				goto fail;
385 			}
386 
387 			device_links_read_unlock(idx);
388 		}
389 	}
390 
391 	if (cb)
392 		retval = cb(dev);
393 
394 	if (dev->power.irq_safe) {
395 		spin_lock(&dev->power.lock);
396 	} else {
397 		/*
398 		 * If the device is suspending and the callback has returned
399 		 * success, drop the usage counters of the suppliers that have
400 		 * been reference counted on its resume.
401 		 *
402 		 * Do that if resume fails too.
403 		 */
404 		if (use_links &&
405 		    ((dev->power.runtime_status == RPM_SUSPENDING && !retval) ||
406 		    (dev->power.runtime_status == RPM_RESUMING && retval))) {
407 			idx = device_links_read_lock();
408 
409 			__rpm_put_suppliers(dev, false);
410 
411 fail:
412 			device_links_read_unlock(idx);
413 		}
414 
415 		spin_lock_irq(&dev->power.lock);
416 	}
417 
418 	return retval;
419 }
420 
421 /**
422  * rpm_callback - Run a given runtime PM callback for a given device.
423  * @cb: Runtime PM callback to run.
424  * @dev: Device to run the callback for.
425  */
426 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
427 {
428 	int retval;
429 
430 	if (dev->power.memalloc_noio) {
431 		unsigned int noio_flag;
432 
433 		/*
434 		 * Deadlock might be caused if memory allocation with
435 		 * GFP_KERNEL happens inside runtime_suspend and
436 		 * runtime_resume callbacks of one block device's
437 		 * ancestor or the block device itself. Network
438 		 * device might be thought as part of iSCSI block
439 		 * device, so network device and its ancestor should
440 		 * be marked as memalloc_noio too.
441 		 */
442 		noio_flag = memalloc_noio_save();
443 		retval = __rpm_callback(cb, dev);
444 		memalloc_noio_restore(noio_flag);
445 	} else {
446 		retval = __rpm_callback(cb, dev);
447 	}
448 
449 	dev->power.runtime_error = retval;
450 	return retval != -EACCES ? retval : -EIO;
451 }
452 
453 /**
454  * rpm_idle - Notify device bus type if the device can be suspended.
455  * @dev: Device to notify the bus type about.
456  * @rpmflags: Flag bits.
457  *
458  * Check if the device's runtime PM status allows it to be suspended.  If
459  * another idle notification has been started earlier, return immediately.  If
460  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
461  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
462  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
463  *
464  * This function must be called under dev->power.lock with interrupts disabled.
465  */
466 static int rpm_idle(struct device *dev, int rpmflags)
467 {
468 	int (*callback)(struct device *);
469 	int retval;
470 
471 	trace_rpm_idle(dev, rpmflags);
472 	retval = rpm_check_suspend_allowed(dev);
473 	if (retval < 0)
474 		;	/* Conditions are wrong. */
475 
476 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
477 	else if (dev->power.runtime_status != RPM_ACTIVE)
478 		retval = -EAGAIN;
479 
480 	/*
481 	 * Any pending request other than an idle notification takes
482 	 * precedence over us, except that the timer may be running.
483 	 */
484 	else if (dev->power.request_pending &&
485 	    dev->power.request > RPM_REQ_IDLE)
486 		retval = -EAGAIN;
487 
488 	/* Act as though RPM_NOWAIT is always set. */
489 	else if (dev->power.idle_notification)
490 		retval = -EINPROGRESS;
491 
492 	if (retval)
493 		goto out;
494 
495 	/* Pending requests need to be canceled. */
496 	dev->power.request = RPM_REQ_NONE;
497 
498 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
499 
500 	/* If no callback assume success. */
501 	if (!callback || dev->power.no_callbacks)
502 		goto out;
503 
504 	/* Carry out an asynchronous or a synchronous idle notification. */
505 	if (rpmflags & RPM_ASYNC) {
506 		dev->power.request = RPM_REQ_IDLE;
507 		if (!dev->power.request_pending) {
508 			dev->power.request_pending = true;
509 			queue_work(pm_wq, &dev->power.work);
510 		}
511 		trace_rpm_return_int(dev, _THIS_IP_, 0);
512 		return 0;
513 	}
514 
515 	dev->power.idle_notification = true;
516 
517 	if (dev->power.irq_safe)
518 		spin_unlock(&dev->power.lock);
519 	else
520 		spin_unlock_irq(&dev->power.lock);
521 
522 	retval = callback(dev);
523 
524 	if (dev->power.irq_safe)
525 		spin_lock(&dev->power.lock);
526 	else
527 		spin_lock_irq(&dev->power.lock);
528 
529 	dev->power.idle_notification = false;
530 	wake_up_all(&dev->power.wait_queue);
531 
532  out:
533 	trace_rpm_return_int(dev, _THIS_IP_, retval);
534 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
535 }
536 
537 /**
538  * rpm_suspend - Carry out runtime suspend of given device.
539  * @dev: Device to suspend.
540  * @rpmflags: Flag bits.
541  *
542  * Check if the device's runtime PM status allows it to be suspended.
543  * Cancel a pending idle notification, autosuspend or suspend. If
544  * another suspend has been started earlier, either return immediately
545  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
546  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
547  * otherwise run the ->runtime_suspend() callback directly. When
548  * ->runtime_suspend succeeded, if a deferred resume was requested while
549  * the callback was running then carry it out, otherwise send an idle
550  * notification for its parent (if the suspend succeeded and both
551  * ignore_children of parent->power and irq_safe of dev->power are not set).
552  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
553  * flag is set and the next autosuspend-delay expiration time is in the
554  * future, schedule another autosuspend attempt.
555  *
556  * This function must be called under dev->power.lock with interrupts disabled.
557  */
558 static int rpm_suspend(struct device *dev, int rpmflags)
559 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
560 {
561 	int (*callback)(struct device *);
562 	struct device *parent = NULL;
563 	int retval;
564 
565 	trace_rpm_suspend(dev, rpmflags);
566 
567  repeat:
568 	retval = rpm_check_suspend_allowed(dev);
569 	if (retval < 0)
570 		goto out;	/* Conditions are wrong. */
571 
572 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
573 	if (dev->power.runtime_status == RPM_RESUMING && !(rpmflags & RPM_ASYNC))
574 		retval = -EAGAIN;
575 
576 	if (retval)
577 		goto out;
578 
579 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
580 	if ((rpmflags & RPM_AUTO) && dev->power.runtime_status != RPM_SUSPENDING) {
581 		u64 expires = pm_runtime_autosuspend_expiration(dev);
582 
583 		if (expires != 0) {
584 			/* Pending requests need to be canceled. */
585 			dev->power.request = RPM_REQ_NONE;
586 
587 			/*
588 			 * Optimization: If the timer is already running and is
589 			 * set to expire at or before the autosuspend delay,
590 			 * avoid the overhead of resetting it.  Just let it
591 			 * expire; pm_suspend_timer_fn() will take care of the
592 			 * rest.
593 			 */
594 			if (!(dev->power.timer_expires &&
595 			    dev->power.timer_expires <= expires)) {
596 				/*
597 				 * We add a slack of 25% to gather wakeups
598 				 * without sacrificing the granularity.
599 				 */
600 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
601 						    (NSEC_PER_MSEC >> 2);
602 
603 				dev->power.timer_expires = expires;
604 				hrtimer_start_range_ns(&dev->power.suspend_timer,
605 						       ns_to_ktime(expires),
606 						       slack,
607 						       HRTIMER_MODE_ABS);
608 			}
609 			dev->power.timer_autosuspends = 1;
610 			goto out;
611 		}
612 	}
613 
614 	/* Other scheduled or pending requests need to be canceled. */
615 	pm_runtime_cancel_pending(dev);
616 
617 	if (dev->power.runtime_status == RPM_SUSPENDING) {
618 		DEFINE_WAIT(wait);
619 
620 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
621 			retval = -EINPROGRESS;
622 			goto out;
623 		}
624 
625 		if (dev->power.irq_safe) {
626 			spin_unlock(&dev->power.lock);
627 
628 			cpu_relax();
629 
630 			spin_lock(&dev->power.lock);
631 			goto repeat;
632 		}
633 
634 		/* Wait for the other suspend running in parallel with us. */
635 		for (;;) {
636 			prepare_to_wait(&dev->power.wait_queue, &wait,
637 					TASK_UNINTERRUPTIBLE);
638 			if (dev->power.runtime_status != RPM_SUSPENDING)
639 				break;
640 
641 			spin_unlock_irq(&dev->power.lock);
642 
643 			schedule();
644 
645 			spin_lock_irq(&dev->power.lock);
646 		}
647 		finish_wait(&dev->power.wait_queue, &wait);
648 		goto repeat;
649 	}
650 
651 	if (dev->power.no_callbacks)
652 		goto no_callback;	/* Assume success. */
653 
654 	/* Carry out an asynchronous or a synchronous suspend. */
655 	if (rpmflags & RPM_ASYNC) {
656 		dev->power.request = (rpmflags & RPM_AUTO) ?
657 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
658 		if (!dev->power.request_pending) {
659 			dev->power.request_pending = true;
660 			queue_work(pm_wq, &dev->power.work);
661 		}
662 		goto out;
663 	}
664 
665 	__update_runtime_status(dev, RPM_SUSPENDING);
666 
667 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
668 
669 	dev_pm_enable_wake_irq_check(dev, true);
670 	retval = rpm_callback(callback, dev);
671 	if (retval)
672 		goto fail;
673 
674 	dev_pm_enable_wake_irq_complete(dev);
675 
676  no_callback:
677 	__update_runtime_status(dev, RPM_SUSPENDED);
678 	pm_runtime_deactivate_timer(dev);
679 
680 	if (dev->parent) {
681 		parent = dev->parent;
682 		atomic_add_unless(&parent->power.child_count, -1, 0);
683 	}
684 	wake_up_all(&dev->power.wait_queue);
685 
686 	if (dev->power.deferred_resume) {
687 		dev->power.deferred_resume = false;
688 		rpm_resume(dev, 0);
689 		retval = -EAGAIN;
690 		goto out;
691 	}
692 
693 	if (dev->power.irq_safe)
694 		goto out;
695 
696 	/* Maybe the parent is now able to suspend. */
697 	if (parent && !parent->power.ignore_children) {
698 		spin_unlock(&dev->power.lock);
699 
700 		spin_lock(&parent->power.lock);
701 		rpm_idle(parent, RPM_ASYNC);
702 		spin_unlock(&parent->power.lock);
703 
704 		spin_lock(&dev->power.lock);
705 	}
706 	/* Maybe the suppliers are now able to suspend. */
707 	if (dev->power.links_count > 0) {
708 		spin_unlock_irq(&dev->power.lock);
709 
710 		rpm_suspend_suppliers(dev);
711 
712 		spin_lock_irq(&dev->power.lock);
713 	}
714 
715  out:
716 	trace_rpm_return_int(dev, _THIS_IP_, retval);
717 
718 	return retval;
719 
720  fail:
721 	dev_pm_disable_wake_irq_check(dev, true);
722 	__update_runtime_status(dev, RPM_ACTIVE);
723 	dev->power.deferred_resume = false;
724 	wake_up_all(&dev->power.wait_queue);
725 
726 	if (retval == -EAGAIN || retval == -EBUSY) {
727 		dev->power.runtime_error = 0;
728 
729 		/*
730 		 * If the callback routine failed an autosuspend, and
731 		 * if the last_busy time has been updated so that there
732 		 * is a new autosuspend expiration time, automatically
733 		 * reschedule another autosuspend.
734 		 */
735 		if ((rpmflags & RPM_AUTO) &&
736 		    pm_runtime_autosuspend_expiration(dev) != 0)
737 			goto repeat;
738 	} else {
739 		pm_runtime_cancel_pending(dev);
740 	}
741 	goto out;
742 }
743 
744 /**
745  * rpm_resume - Carry out runtime resume of given device.
746  * @dev: Device to resume.
747  * @rpmflags: Flag bits.
748  *
749  * Check if the device's runtime PM status allows it to be resumed.  Cancel
750  * any scheduled or pending requests.  If another resume has been started
751  * earlier, either return immediately or wait for it to finish, depending on the
752  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
753  * parallel with this function, either tell the other process to resume after
754  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
755  * flag is set then queue a resume request; otherwise run the
756  * ->runtime_resume() callback directly.  Queue an idle notification for the
757  * device if the resume succeeded.
758  *
759  * This function must be called under dev->power.lock with interrupts disabled.
760  */
761 static int rpm_resume(struct device *dev, int rpmflags)
762 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
763 {
764 	int (*callback)(struct device *);
765 	struct device *parent = NULL;
766 	int retval = 0;
767 
768 	trace_rpm_resume(dev, rpmflags);
769 
770  repeat:
771 	if (dev->power.runtime_error) {
772 		retval = -EINVAL;
773 	} else if (dev->power.disable_depth > 0) {
774 		if (dev->power.runtime_status == RPM_ACTIVE &&
775 		    dev->power.last_status == RPM_ACTIVE)
776 			retval = 1;
777 		else
778 			retval = -EACCES;
779 	}
780 	if (retval)
781 		goto out;
782 
783 	/*
784 	 * Other scheduled or pending requests need to be canceled.  Small
785 	 * optimization: If an autosuspend timer is running, leave it running
786 	 * rather than cancelling it now only to restart it again in the near
787 	 * future.
788 	 */
789 	dev->power.request = RPM_REQ_NONE;
790 	if (!dev->power.timer_autosuspends)
791 		pm_runtime_deactivate_timer(dev);
792 
793 	if (dev->power.runtime_status == RPM_ACTIVE) {
794 		retval = 1;
795 		goto out;
796 	}
797 
798 	if (dev->power.runtime_status == RPM_RESUMING ||
799 	    dev->power.runtime_status == RPM_SUSPENDING) {
800 		DEFINE_WAIT(wait);
801 
802 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
803 			if (dev->power.runtime_status == RPM_SUSPENDING) {
804 				dev->power.deferred_resume = true;
805 				if (rpmflags & RPM_NOWAIT)
806 					retval = -EINPROGRESS;
807 			} else {
808 				retval = -EINPROGRESS;
809 			}
810 			goto out;
811 		}
812 
813 		if (dev->power.irq_safe) {
814 			spin_unlock(&dev->power.lock);
815 
816 			cpu_relax();
817 
818 			spin_lock(&dev->power.lock);
819 			goto repeat;
820 		}
821 
822 		/* Wait for the operation carried out in parallel with us. */
823 		for (;;) {
824 			prepare_to_wait(&dev->power.wait_queue, &wait,
825 					TASK_UNINTERRUPTIBLE);
826 			if (dev->power.runtime_status != RPM_RESUMING &&
827 			    dev->power.runtime_status != RPM_SUSPENDING)
828 				break;
829 
830 			spin_unlock_irq(&dev->power.lock);
831 
832 			schedule();
833 
834 			spin_lock_irq(&dev->power.lock);
835 		}
836 		finish_wait(&dev->power.wait_queue, &wait);
837 		goto repeat;
838 	}
839 
840 	/*
841 	 * See if we can skip waking up the parent.  This is safe only if
842 	 * power.no_callbacks is set, because otherwise we don't know whether
843 	 * the resume will actually succeed.
844 	 */
845 	if (dev->power.no_callbacks && !parent && dev->parent) {
846 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
847 		if (dev->parent->power.disable_depth > 0 ||
848 		    dev->parent->power.ignore_children ||
849 		    dev->parent->power.runtime_status == RPM_ACTIVE) {
850 			atomic_inc(&dev->parent->power.child_count);
851 			spin_unlock(&dev->parent->power.lock);
852 			retval = 1;
853 			goto no_callback;	/* Assume success. */
854 		}
855 		spin_unlock(&dev->parent->power.lock);
856 	}
857 
858 	/* Carry out an asynchronous or a synchronous resume. */
859 	if (rpmflags & RPM_ASYNC) {
860 		dev->power.request = RPM_REQ_RESUME;
861 		if (!dev->power.request_pending) {
862 			dev->power.request_pending = true;
863 			queue_work(pm_wq, &dev->power.work);
864 		}
865 		retval = 0;
866 		goto out;
867 	}
868 
869 	if (!parent && dev->parent) {
870 		/*
871 		 * Increment the parent's usage counter and resume it if
872 		 * necessary.  Not needed if dev is irq-safe; then the
873 		 * parent is permanently resumed.
874 		 */
875 		parent = dev->parent;
876 		if (dev->power.irq_safe)
877 			goto skip_parent;
878 
879 		spin_unlock(&dev->power.lock);
880 
881 		pm_runtime_get_noresume(parent);
882 
883 		spin_lock(&parent->power.lock);
884 		/*
885 		 * Resume the parent if it has runtime PM enabled and not been
886 		 * set to ignore its children.
887 		 */
888 		if (!parent->power.disable_depth &&
889 		    !parent->power.ignore_children) {
890 			rpm_resume(parent, 0);
891 			if (parent->power.runtime_status != RPM_ACTIVE)
892 				retval = -EBUSY;
893 		}
894 		spin_unlock(&parent->power.lock);
895 
896 		spin_lock(&dev->power.lock);
897 		if (retval)
898 			goto out;
899 
900 		goto repeat;
901 	}
902  skip_parent:
903 
904 	if (dev->power.no_callbacks)
905 		goto no_callback;	/* Assume success. */
906 
907 	__update_runtime_status(dev, RPM_RESUMING);
908 
909 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
910 
911 	dev_pm_disable_wake_irq_check(dev, false);
912 	retval = rpm_callback(callback, dev);
913 	if (retval) {
914 		__update_runtime_status(dev, RPM_SUSPENDED);
915 		pm_runtime_cancel_pending(dev);
916 		dev_pm_enable_wake_irq_check(dev, false);
917 	} else {
918  no_callback:
919 		__update_runtime_status(dev, RPM_ACTIVE);
920 		pm_runtime_mark_last_busy(dev);
921 		if (parent)
922 			atomic_inc(&parent->power.child_count);
923 	}
924 	wake_up_all(&dev->power.wait_queue);
925 
926 	if (retval >= 0)
927 		rpm_idle(dev, RPM_ASYNC);
928 
929  out:
930 	if (parent && !dev->power.irq_safe) {
931 		spin_unlock_irq(&dev->power.lock);
932 
933 		pm_runtime_put(parent);
934 
935 		spin_lock_irq(&dev->power.lock);
936 	}
937 
938 	trace_rpm_return_int(dev, _THIS_IP_, retval);
939 
940 	return retval;
941 }
942 
943 /**
944  * pm_runtime_work - Universal runtime PM work function.
945  * @work: Work structure used for scheduling the execution of this function.
946  *
947  * Use @work to get the device object the work is to be done for, determine what
948  * is to be done and execute the appropriate runtime PM function.
949  */
950 static void pm_runtime_work(struct work_struct *work)
951 {
952 	struct device *dev = container_of(work, struct device, power.work);
953 	enum rpm_request req;
954 
955 	spin_lock_irq(&dev->power.lock);
956 
957 	if (!dev->power.request_pending)
958 		goto out;
959 
960 	req = dev->power.request;
961 	dev->power.request = RPM_REQ_NONE;
962 	dev->power.request_pending = false;
963 
964 	switch (req) {
965 	case RPM_REQ_NONE:
966 		break;
967 	case RPM_REQ_IDLE:
968 		rpm_idle(dev, RPM_NOWAIT);
969 		break;
970 	case RPM_REQ_SUSPEND:
971 		rpm_suspend(dev, RPM_NOWAIT);
972 		break;
973 	case RPM_REQ_AUTOSUSPEND:
974 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
975 		break;
976 	case RPM_REQ_RESUME:
977 		rpm_resume(dev, RPM_NOWAIT);
978 		break;
979 	}
980 
981  out:
982 	spin_unlock_irq(&dev->power.lock);
983 }
984 
985 /**
986  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
987  * @timer: hrtimer used by pm_schedule_suspend().
988  *
989  * Check if the time is right and queue a suspend request.
990  */
991 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
992 {
993 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
994 	unsigned long flags;
995 	u64 expires;
996 
997 	spin_lock_irqsave(&dev->power.lock, flags);
998 
999 	expires = dev->power.timer_expires;
1000 	/*
1001 	 * If 'expires' is after the current time, we've been called
1002 	 * too early.
1003 	 */
1004 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
1005 		dev->power.timer_expires = 0;
1006 		rpm_suspend(dev, dev->power.timer_autosuspends ?
1007 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
1008 	}
1009 
1010 	spin_unlock_irqrestore(&dev->power.lock, flags);
1011 
1012 	return HRTIMER_NORESTART;
1013 }
1014 
1015 /**
1016  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
1017  * @dev: Device to suspend.
1018  * @delay: Time to wait before submitting a suspend request, in milliseconds.
1019  */
1020 int pm_schedule_suspend(struct device *dev, unsigned int delay)
1021 {
1022 	unsigned long flags;
1023 	u64 expires;
1024 	int retval;
1025 
1026 	spin_lock_irqsave(&dev->power.lock, flags);
1027 
1028 	if (!delay) {
1029 		retval = rpm_suspend(dev, RPM_ASYNC);
1030 		goto out;
1031 	}
1032 
1033 	retval = rpm_check_suspend_allowed(dev);
1034 	if (retval)
1035 		goto out;
1036 
1037 	/* Other scheduled or pending requests need to be canceled. */
1038 	pm_runtime_cancel_pending(dev);
1039 
1040 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
1041 	dev->power.timer_expires = expires;
1042 	dev->power.timer_autosuspends = 0;
1043 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
1044 
1045  out:
1046 	spin_unlock_irqrestore(&dev->power.lock, flags);
1047 
1048 	return retval;
1049 }
1050 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
1051 
1052 static int rpm_drop_usage_count(struct device *dev)
1053 {
1054 	int ret;
1055 
1056 	ret = atomic_sub_return(1, &dev->power.usage_count);
1057 	if (ret >= 0)
1058 		return ret;
1059 
1060 	/*
1061 	 * Because rpm_resume() does not check the usage counter, it will resume
1062 	 * the device even if the usage counter is 0 or negative, so it is
1063 	 * sufficient to increment the usage counter here to reverse the change
1064 	 * made above.
1065 	 */
1066 	atomic_inc(&dev->power.usage_count);
1067 	dev_warn(dev, "Runtime PM usage count underflow!\n");
1068 	return -EINVAL;
1069 }
1070 
1071 /**
1072  * __pm_runtime_idle - Entry point for runtime idle operations.
1073  * @dev: Device to send idle notification for.
1074  * @rpmflags: Flag bits.
1075  *
1076  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1077  * return immediately if it is larger than zero (if it becomes negative, log a
1078  * warning, increment it, and return an error).  Then carry out an idle
1079  * notification, either synchronous or asynchronous.
1080  *
1081  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1082  * or if pm_runtime_irq_safe() has been called.
1083  */
1084 int __pm_runtime_idle(struct device *dev, int rpmflags)
1085 {
1086 	unsigned long flags;
1087 	int retval;
1088 
1089 	if (rpmflags & RPM_GET_PUT) {
1090 		retval = rpm_drop_usage_count(dev);
1091 		if (retval < 0) {
1092 			return retval;
1093 		} else if (retval > 0) {
1094 			trace_rpm_usage(dev, rpmflags);
1095 			return 0;
1096 		}
1097 	}
1098 
1099 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1100 
1101 	spin_lock_irqsave(&dev->power.lock, flags);
1102 	retval = rpm_idle(dev, rpmflags);
1103 	spin_unlock_irqrestore(&dev->power.lock, flags);
1104 
1105 	return retval;
1106 }
1107 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1108 
1109 /**
1110  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1111  * @dev: Device to suspend.
1112  * @rpmflags: Flag bits.
1113  *
1114  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1115  * return immediately if it is larger than zero (if it becomes negative, log a
1116  * warning, increment it, and return an error).  Then carry out a suspend,
1117  * either synchronous or asynchronous.
1118  *
1119  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1120  * or if pm_runtime_irq_safe() has been called.
1121  */
1122 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1123 {
1124 	unsigned long flags;
1125 	int retval;
1126 
1127 	if (rpmflags & RPM_GET_PUT) {
1128 		retval = rpm_drop_usage_count(dev);
1129 		if (retval < 0) {
1130 			return retval;
1131 		} else if (retval > 0) {
1132 			trace_rpm_usage(dev, rpmflags);
1133 			return 0;
1134 		}
1135 	}
1136 
1137 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1138 
1139 	spin_lock_irqsave(&dev->power.lock, flags);
1140 	retval = rpm_suspend(dev, rpmflags);
1141 	spin_unlock_irqrestore(&dev->power.lock, flags);
1142 
1143 	return retval;
1144 }
1145 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1146 
1147 /**
1148  * __pm_runtime_resume - Entry point for runtime resume operations.
1149  * @dev: Device to resume.
1150  * @rpmflags: Flag bits.
1151  *
1152  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1153  * carry out a resume, either synchronous or asynchronous.
1154  *
1155  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1156  * or if pm_runtime_irq_safe() has been called.
1157  */
1158 int __pm_runtime_resume(struct device *dev, int rpmflags)
1159 {
1160 	unsigned long flags;
1161 	int retval;
1162 
1163 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1164 			dev->power.runtime_status != RPM_ACTIVE);
1165 
1166 	if (rpmflags & RPM_GET_PUT)
1167 		atomic_inc(&dev->power.usage_count);
1168 
1169 	spin_lock_irqsave(&dev->power.lock, flags);
1170 	retval = rpm_resume(dev, rpmflags);
1171 	spin_unlock_irqrestore(&dev->power.lock, flags);
1172 
1173 	return retval;
1174 }
1175 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1176 
1177 /**
1178  * pm_runtime_get_if_active - Conditionally bump up device usage counter.
1179  * @dev: Device to handle.
1180  * @ign_usage_count: Whether or not to look at the current usage counter value.
1181  *
1182  * Return -EINVAL if runtime PM is disabled for @dev.
1183  *
1184  * Otherwise, if the runtime PM status of @dev is %RPM_ACTIVE and either
1185  * @ign_usage_count is %true or the runtime PM usage counter of @dev is not
1186  * zero, increment the usage counter of @dev and return 1. Otherwise, return 0
1187  * without changing the usage counter.
1188  *
1189  * If @ign_usage_count is %true, this function can be used to prevent suspending
1190  * the device when its runtime PM status is %RPM_ACTIVE.
1191  *
1192  * If @ign_usage_count is %false, this function can be used to prevent
1193  * suspending the device when both its runtime PM status is %RPM_ACTIVE and its
1194  * runtime PM usage counter is not zero.
1195  *
1196  * The caller is responsible for decrementing the runtime PM usage counter of
1197  * @dev after this function has returned a positive value for it.
1198  */
1199 int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count)
1200 {
1201 	unsigned long flags;
1202 	int retval;
1203 
1204 	spin_lock_irqsave(&dev->power.lock, flags);
1205 	if (dev->power.disable_depth > 0) {
1206 		retval = -EINVAL;
1207 	} else if (dev->power.runtime_status != RPM_ACTIVE) {
1208 		retval = 0;
1209 	} else if (ign_usage_count) {
1210 		retval = 1;
1211 		atomic_inc(&dev->power.usage_count);
1212 	} else {
1213 		retval = atomic_inc_not_zero(&dev->power.usage_count);
1214 	}
1215 	trace_rpm_usage(dev, 0);
1216 	spin_unlock_irqrestore(&dev->power.lock, flags);
1217 
1218 	return retval;
1219 }
1220 EXPORT_SYMBOL_GPL(pm_runtime_get_if_active);
1221 
1222 /**
1223  * __pm_runtime_set_status - Set runtime PM status of a device.
1224  * @dev: Device to handle.
1225  * @status: New runtime PM status of the device.
1226  *
1227  * If runtime PM of the device is disabled or its power.runtime_error field is
1228  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1229  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1230  * However, if the device has a parent and the parent is not active, and the
1231  * parent's power.ignore_children flag is unset, the device's status cannot be
1232  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1233  *
1234  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1235  * and the device parent's counter of unsuspended children is modified to
1236  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1237  * notification request for the parent is submitted.
1238  *
1239  * If @dev has any suppliers (as reflected by device links to them), and @status
1240  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1241  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1242  * of the @status value) and the suppliers will be deacticated on exit.  The
1243  * error returned by the failing supplier activation will be returned in that
1244  * case.
1245  */
1246 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1247 {
1248 	struct device *parent = dev->parent;
1249 	bool notify_parent = false;
1250 	unsigned long flags;
1251 	int error = 0;
1252 
1253 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1254 		return -EINVAL;
1255 
1256 	spin_lock_irqsave(&dev->power.lock, flags);
1257 
1258 	/*
1259 	 * Prevent PM-runtime from being enabled for the device or return an
1260 	 * error if it is enabled already and working.
1261 	 */
1262 	if (dev->power.runtime_error || dev->power.disable_depth)
1263 		dev->power.disable_depth++;
1264 	else
1265 		error = -EAGAIN;
1266 
1267 	spin_unlock_irqrestore(&dev->power.lock, flags);
1268 
1269 	if (error)
1270 		return error;
1271 
1272 	/*
1273 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1274 	 * upfront regardless of the current status, because next time
1275 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1276 	 * involved will be dropped down to one anyway.
1277 	 */
1278 	if (status == RPM_ACTIVE) {
1279 		int idx = device_links_read_lock();
1280 
1281 		error = rpm_get_suppliers(dev);
1282 		if (error)
1283 			status = RPM_SUSPENDED;
1284 
1285 		device_links_read_unlock(idx);
1286 	}
1287 
1288 	spin_lock_irqsave(&dev->power.lock, flags);
1289 
1290 	if (dev->power.runtime_status == status || !parent)
1291 		goto out_set;
1292 
1293 	if (status == RPM_SUSPENDED) {
1294 		atomic_add_unless(&parent->power.child_count, -1, 0);
1295 		notify_parent = !parent->power.ignore_children;
1296 	} else {
1297 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1298 
1299 		/*
1300 		 * It is invalid to put an active child under a parent that is
1301 		 * not active, has runtime PM enabled and the
1302 		 * 'power.ignore_children' flag unset.
1303 		 */
1304 		if (!parent->power.disable_depth &&
1305 		    !parent->power.ignore_children &&
1306 		    parent->power.runtime_status != RPM_ACTIVE) {
1307 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1308 				dev_name(dev),
1309 				dev_name(parent));
1310 			error = -EBUSY;
1311 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1312 			atomic_inc(&parent->power.child_count);
1313 		}
1314 
1315 		spin_unlock(&parent->power.lock);
1316 
1317 		if (error) {
1318 			status = RPM_SUSPENDED;
1319 			goto out;
1320 		}
1321 	}
1322 
1323  out_set:
1324 	__update_runtime_status(dev, status);
1325 	if (!error)
1326 		dev->power.runtime_error = 0;
1327 
1328  out:
1329 	spin_unlock_irqrestore(&dev->power.lock, flags);
1330 
1331 	if (notify_parent)
1332 		pm_request_idle(parent);
1333 
1334 	if (status == RPM_SUSPENDED) {
1335 		int idx = device_links_read_lock();
1336 
1337 		rpm_put_suppliers(dev);
1338 
1339 		device_links_read_unlock(idx);
1340 	}
1341 
1342 	pm_runtime_enable(dev);
1343 
1344 	return error;
1345 }
1346 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1347 
1348 /**
1349  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1350  * @dev: Device to handle.
1351  *
1352  * Flush all pending requests for the device from pm_wq and wait for all
1353  * runtime PM operations involving the device in progress to complete.
1354  *
1355  * Should be called under dev->power.lock with interrupts disabled.
1356  */
1357 static void __pm_runtime_barrier(struct device *dev)
1358 {
1359 	pm_runtime_deactivate_timer(dev);
1360 
1361 	if (dev->power.request_pending) {
1362 		dev->power.request = RPM_REQ_NONE;
1363 		spin_unlock_irq(&dev->power.lock);
1364 
1365 		cancel_work_sync(&dev->power.work);
1366 
1367 		spin_lock_irq(&dev->power.lock);
1368 		dev->power.request_pending = false;
1369 	}
1370 
1371 	if (dev->power.runtime_status == RPM_SUSPENDING ||
1372 	    dev->power.runtime_status == RPM_RESUMING ||
1373 	    dev->power.idle_notification) {
1374 		DEFINE_WAIT(wait);
1375 
1376 		/* Suspend, wake-up or idle notification in progress. */
1377 		for (;;) {
1378 			prepare_to_wait(&dev->power.wait_queue, &wait,
1379 					TASK_UNINTERRUPTIBLE);
1380 			if (dev->power.runtime_status != RPM_SUSPENDING
1381 			    && dev->power.runtime_status != RPM_RESUMING
1382 			    && !dev->power.idle_notification)
1383 				break;
1384 			spin_unlock_irq(&dev->power.lock);
1385 
1386 			schedule();
1387 
1388 			spin_lock_irq(&dev->power.lock);
1389 		}
1390 		finish_wait(&dev->power.wait_queue, &wait);
1391 	}
1392 }
1393 
1394 /**
1395  * pm_runtime_barrier - Flush pending requests and wait for completions.
1396  * @dev: Device to handle.
1397  *
1398  * Prevent the device from being suspended by incrementing its usage counter and
1399  * if there's a pending resume request for the device, wake the device up.
1400  * Next, make sure that all pending requests for the device have been flushed
1401  * from pm_wq and wait for all runtime PM operations involving the device in
1402  * progress to complete.
1403  *
1404  * Return value:
1405  * 1, if there was a resume request pending and the device had to be woken up,
1406  * 0, otherwise
1407  */
1408 int pm_runtime_barrier(struct device *dev)
1409 {
1410 	int retval = 0;
1411 
1412 	pm_runtime_get_noresume(dev);
1413 	spin_lock_irq(&dev->power.lock);
1414 
1415 	if (dev->power.request_pending
1416 	    && dev->power.request == RPM_REQ_RESUME) {
1417 		rpm_resume(dev, 0);
1418 		retval = 1;
1419 	}
1420 
1421 	__pm_runtime_barrier(dev);
1422 
1423 	spin_unlock_irq(&dev->power.lock);
1424 	pm_runtime_put_noidle(dev);
1425 
1426 	return retval;
1427 }
1428 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1429 
1430 /**
1431  * __pm_runtime_disable - Disable runtime PM of a device.
1432  * @dev: Device to handle.
1433  * @check_resume: If set, check if there's a resume request for the device.
1434  *
1435  * Increment power.disable_depth for the device and if it was zero previously,
1436  * cancel all pending runtime PM requests for the device and wait for all
1437  * operations in progress to complete.  The device can be either active or
1438  * suspended after its runtime PM has been disabled.
1439  *
1440  * If @check_resume is set and there's a resume request pending when
1441  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1442  * function will wake up the device before disabling its runtime PM.
1443  */
1444 void __pm_runtime_disable(struct device *dev, bool check_resume)
1445 {
1446 	spin_lock_irq(&dev->power.lock);
1447 
1448 	if (dev->power.disable_depth > 0) {
1449 		dev->power.disable_depth++;
1450 		goto out;
1451 	}
1452 
1453 	/*
1454 	 * Wake up the device if there's a resume request pending, because that
1455 	 * means there probably is some I/O to process and disabling runtime PM
1456 	 * shouldn't prevent the device from processing the I/O.
1457 	 */
1458 	if (check_resume && dev->power.request_pending &&
1459 	    dev->power.request == RPM_REQ_RESUME) {
1460 		/*
1461 		 * Prevent suspends and idle notifications from being carried
1462 		 * out after we have woken up the device.
1463 		 */
1464 		pm_runtime_get_noresume(dev);
1465 
1466 		rpm_resume(dev, 0);
1467 
1468 		pm_runtime_put_noidle(dev);
1469 	}
1470 
1471 	/* Update time accounting before disabling PM-runtime. */
1472 	update_pm_runtime_accounting(dev);
1473 
1474 	if (!dev->power.disable_depth++) {
1475 		__pm_runtime_barrier(dev);
1476 		dev->power.last_status = dev->power.runtime_status;
1477 	}
1478 
1479  out:
1480 	spin_unlock_irq(&dev->power.lock);
1481 }
1482 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1483 
1484 /**
1485  * pm_runtime_enable - Enable runtime PM of a device.
1486  * @dev: Device to handle.
1487  */
1488 void pm_runtime_enable(struct device *dev)
1489 {
1490 	unsigned long flags;
1491 
1492 	spin_lock_irqsave(&dev->power.lock, flags);
1493 
1494 	if (!dev->power.disable_depth) {
1495 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1496 		goto out;
1497 	}
1498 
1499 	if (--dev->power.disable_depth > 0)
1500 		goto out;
1501 
1502 	dev->power.last_status = RPM_INVALID;
1503 	dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1504 
1505 	if (dev->power.runtime_status == RPM_SUSPENDED &&
1506 	    !dev->power.ignore_children &&
1507 	    atomic_read(&dev->power.child_count) > 0)
1508 		dev_warn(dev, "Enabling runtime PM for inactive device with active children\n");
1509 
1510 out:
1511 	spin_unlock_irqrestore(&dev->power.lock, flags);
1512 }
1513 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1514 
1515 static void pm_runtime_disable_action(void *data)
1516 {
1517 	pm_runtime_dont_use_autosuspend(data);
1518 	pm_runtime_disable(data);
1519 }
1520 
1521 /**
1522  * devm_pm_runtime_enable - devres-enabled version of pm_runtime_enable.
1523  *
1524  * NOTE: this will also handle calling pm_runtime_dont_use_autosuspend() for
1525  * you at driver exit time if needed.
1526  *
1527  * @dev: Device to handle.
1528  */
1529 int devm_pm_runtime_enable(struct device *dev)
1530 {
1531 	pm_runtime_enable(dev);
1532 
1533 	return devm_add_action_or_reset(dev, pm_runtime_disable_action, dev);
1534 }
1535 EXPORT_SYMBOL_GPL(devm_pm_runtime_enable);
1536 
1537 /**
1538  * pm_runtime_forbid - Block runtime PM of a device.
1539  * @dev: Device to handle.
1540  *
1541  * Increase the device's usage count and clear its power.runtime_auto flag,
1542  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1543  * for it.
1544  */
1545 void pm_runtime_forbid(struct device *dev)
1546 {
1547 	spin_lock_irq(&dev->power.lock);
1548 	if (!dev->power.runtime_auto)
1549 		goto out;
1550 
1551 	dev->power.runtime_auto = false;
1552 	atomic_inc(&dev->power.usage_count);
1553 	rpm_resume(dev, 0);
1554 
1555  out:
1556 	spin_unlock_irq(&dev->power.lock);
1557 }
1558 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1559 
1560 /**
1561  * pm_runtime_allow - Unblock runtime PM of a device.
1562  * @dev: Device to handle.
1563  *
1564  * Decrease the device's usage count and set its power.runtime_auto flag.
1565  */
1566 void pm_runtime_allow(struct device *dev)
1567 {
1568 	int ret;
1569 
1570 	spin_lock_irq(&dev->power.lock);
1571 	if (dev->power.runtime_auto)
1572 		goto out;
1573 
1574 	dev->power.runtime_auto = true;
1575 	ret = rpm_drop_usage_count(dev);
1576 	if (ret == 0)
1577 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1578 	else if (ret > 0)
1579 		trace_rpm_usage(dev, RPM_AUTO | RPM_ASYNC);
1580 
1581  out:
1582 	spin_unlock_irq(&dev->power.lock);
1583 }
1584 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1585 
1586 /**
1587  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1588  * @dev: Device to handle.
1589  *
1590  * Set the power.no_callbacks flag, which tells the PM core that this
1591  * device is power-managed through its parent and has no runtime PM
1592  * callbacks of its own.  The runtime sysfs attributes will be removed.
1593  */
1594 void pm_runtime_no_callbacks(struct device *dev)
1595 {
1596 	spin_lock_irq(&dev->power.lock);
1597 	dev->power.no_callbacks = 1;
1598 	spin_unlock_irq(&dev->power.lock);
1599 	if (device_is_registered(dev))
1600 		rpm_sysfs_remove(dev);
1601 }
1602 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1603 
1604 /**
1605  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1606  * @dev: Device to handle
1607  *
1608  * Set the power.irq_safe flag, which tells the PM core that the
1609  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1610  * always be invoked with the spinlock held and interrupts disabled.  It also
1611  * causes the parent's usage counter to be permanently incremented, preventing
1612  * the parent from runtime suspending -- otherwise an irq-safe child might have
1613  * to wait for a non-irq-safe parent.
1614  */
1615 void pm_runtime_irq_safe(struct device *dev)
1616 {
1617 	if (dev->parent)
1618 		pm_runtime_get_sync(dev->parent);
1619 
1620 	spin_lock_irq(&dev->power.lock);
1621 	dev->power.irq_safe = 1;
1622 	spin_unlock_irq(&dev->power.lock);
1623 }
1624 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1625 
1626 /**
1627  * update_autosuspend - Handle a change to a device's autosuspend settings.
1628  * @dev: Device to handle.
1629  * @old_delay: The former autosuspend_delay value.
1630  * @old_use: The former use_autosuspend value.
1631  *
1632  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1633  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1634  *
1635  * This function must be called under dev->power.lock with interrupts disabled.
1636  */
1637 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1638 {
1639 	int delay = dev->power.autosuspend_delay;
1640 
1641 	/* Should runtime suspend be prevented now? */
1642 	if (dev->power.use_autosuspend && delay < 0) {
1643 
1644 		/* If it used to be allowed then prevent it. */
1645 		if (!old_use || old_delay >= 0) {
1646 			atomic_inc(&dev->power.usage_count);
1647 			rpm_resume(dev, 0);
1648 		} else {
1649 			trace_rpm_usage(dev, 0);
1650 		}
1651 	}
1652 
1653 	/* Runtime suspend should be allowed now. */
1654 	else {
1655 
1656 		/* If it used to be prevented then allow it. */
1657 		if (old_use && old_delay < 0)
1658 			atomic_dec(&dev->power.usage_count);
1659 
1660 		/* Maybe we can autosuspend now. */
1661 		rpm_idle(dev, RPM_AUTO);
1662 	}
1663 }
1664 
1665 /**
1666  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1667  * @dev: Device to handle.
1668  * @delay: Value of the new delay in milliseconds.
1669  *
1670  * Set the device's power.autosuspend_delay value.  If it changes to negative
1671  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1672  * changes the other way, allow runtime suspends.
1673  */
1674 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1675 {
1676 	int old_delay, old_use;
1677 
1678 	spin_lock_irq(&dev->power.lock);
1679 	old_delay = dev->power.autosuspend_delay;
1680 	old_use = dev->power.use_autosuspend;
1681 	dev->power.autosuspend_delay = delay;
1682 	update_autosuspend(dev, old_delay, old_use);
1683 	spin_unlock_irq(&dev->power.lock);
1684 }
1685 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1686 
1687 /**
1688  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1689  * @dev: Device to handle.
1690  * @use: New value for use_autosuspend.
1691  *
1692  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1693  * suspends as needed.
1694  */
1695 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1696 {
1697 	int old_delay, old_use;
1698 
1699 	spin_lock_irq(&dev->power.lock);
1700 	old_delay = dev->power.autosuspend_delay;
1701 	old_use = dev->power.use_autosuspend;
1702 	dev->power.use_autosuspend = use;
1703 	update_autosuspend(dev, old_delay, old_use);
1704 	spin_unlock_irq(&dev->power.lock);
1705 }
1706 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1707 
1708 /**
1709  * pm_runtime_init - Initialize runtime PM fields in given device object.
1710  * @dev: Device object to initialize.
1711  */
1712 void pm_runtime_init(struct device *dev)
1713 {
1714 	dev->power.runtime_status = RPM_SUSPENDED;
1715 	dev->power.last_status = RPM_INVALID;
1716 	dev->power.idle_notification = false;
1717 
1718 	dev->power.disable_depth = 1;
1719 	atomic_set(&dev->power.usage_count, 0);
1720 
1721 	dev->power.runtime_error = 0;
1722 
1723 	atomic_set(&dev->power.child_count, 0);
1724 	pm_suspend_ignore_children(dev, false);
1725 	dev->power.runtime_auto = true;
1726 
1727 	dev->power.request_pending = false;
1728 	dev->power.request = RPM_REQ_NONE;
1729 	dev->power.deferred_resume = false;
1730 	dev->power.needs_force_resume = 0;
1731 	INIT_WORK(&dev->power.work, pm_runtime_work);
1732 
1733 	dev->power.timer_expires = 0;
1734 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1735 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1736 
1737 	init_waitqueue_head(&dev->power.wait_queue);
1738 }
1739 
1740 /**
1741  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1742  * @dev: Device object to re-initialize.
1743  */
1744 void pm_runtime_reinit(struct device *dev)
1745 {
1746 	if (!pm_runtime_enabled(dev)) {
1747 		if (dev->power.runtime_status == RPM_ACTIVE)
1748 			pm_runtime_set_suspended(dev);
1749 		if (dev->power.irq_safe) {
1750 			spin_lock_irq(&dev->power.lock);
1751 			dev->power.irq_safe = 0;
1752 			spin_unlock_irq(&dev->power.lock);
1753 			if (dev->parent)
1754 				pm_runtime_put(dev->parent);
1755 		}
1756 	}
1757 }
1758 
1759 /**
1760  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1761  * @dev: Device object being removed from device hierarchy.
1762  */
1763 void pm_runtime_remove(struct device *dev)
1764 {
1765 	__pm_runtime_disable(dev, false);
1766 	pm_runtime_reinit(dev);
1767 }
1768 
1769 /**
1770  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1771  * @dev: Consumer device.
1772  */
1773 void pm_runtime_get_suppliers(struct device *dev)
1774 {
1775 	struct device_link *link;
1776 	int idx;
1777 
1778 	idx = device_links_read_lock();
1779 
1780 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1781 				device_links_read_lock_held())
1782 		if (link->flags & DL_FLAG_PM_RUNTIME) {
1783 			link->supplier_preactivated = true;
1784 			pm_runtime_get_sync(link->supplier);
1785 		}
1786 
1787 	device_links_read_unlock(idx);
1788 }
1789 
1790 /**
1791  * pm_runtime_put_suppliers - Drop references to supplier devices.
1792  * @dev: Consumer device.
1793  */
1794 void pm_runtime_put_suppliers(struct device *dev)
1795 {
1796 	struct device_link *link;
1797 	int idx;
1798 
1799 	idx = device_links_read_lock();
1800 
1801 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1802 				device_links_read_lock_held())
1803 		if (link->supplier_preactivated) {
1804 			link->supplier_preactivated = false;
1805 			pm_runtime_put(link->supplier);
1806 		}
1807 
1808 	device_links_read_unlock(idx);
1809 }
1810 
1811 void pm_runtime_new_link(struct device *dev)
1812 {
1813 	spin_lock_irq(&dev->power.lock);
1814 	dev->power.links_count++;
1815 	spin_unlock_irq(&dev->power.lock);
1816 }
1817 
1818 static void pm_runtime_drop_link_count(struct device *dev)
1819 {
1820 	spin_lock_irq(&dev->power.lock);
1821 	WARN_ON(dev->power.links_count == 0);
1822 	dev->power.links_count--;
1823 	spin_unlock_irq(&dev->power.lock);
1824 }
1825 
1826 /**
1827  * pm_runtime_drop_link - Prepare for device link removal.
1828  * @link: Device link going away.
1829  *
1830  * Drop the link count of the consumer end of @link and decrement the supplier
1831  * device's runtime PM usage counter as many times as needed to drop all of the
1832  * PM runtime reference to it from the consumer.
1833  */
1834 void pm_runtime_drop_link(struct device_link *link)
1835 {
1836 	if (!(link->flags & DL_FLAG_PM_RUNTIME))
1837 		return;
1838 
1839 	pm_runtime_drop_link_count(link->consumer);
1840 	pm_runtime_release_supplier(link);
1841 	pm_request_idle(link->supplier);
1842 }
1843 
1844 static bool pm_runtime_need_not_resume(struct device *dev)
1845 {
1846 	return atomic_read(&dev->power.usage_count) <= 1 &&
1847 		(atomic_read(&dev->power.child_count) == 0 ||
1848 		 dev->power.ignore_children);
1849 }
1850 
1851 /**
1852  * pm_runtime_force_suspend - Force a device into suspend state if needed.
1853  * @dev: Device to suspend.
1854  *
1855  * Disable runtime PM so we safely can check the device's runtime PM status and
1856  * if it is active, invoke its ->runtime_suspend callback to suspend it and
1857  * change its runtime PM status field to RPM_SUSPENDED.  Also, if the device's
1858  * usage and children counters don't indicate that the device was in use before
1859  * the system-wide transition under way, decrement its parent's children counter
1860  * (if there is a parent).  Keep runtime PM disabled to preserve the state
1861  * unless we encounter errors.
1862  *
1863  * Typically this function may be invoked from a system suspend callback to make
1864  * sure the device is put into low power state and it should only be used during
1865  * system-wide PM transitions to sleep states.  It assumes that the analogous
1866  * pm_runtime_force_resume() will be used to resume the device.
1867  *
1868  * Do not use with DPM_FLAG_SMART_SUSPEND as this can lead to an inconsistent
1869  * state where this function has called the ->runtime_suspend callback but the
1870  * PM core marks the driver as runtime active.
1871  */
1872 int pm_runtime_force_suspend(struct device *dev)
1873 {
1874 	int (*callback)(struct device *);
1875 	int ret;
1876 
1877 	pm_runtime_disable(dev);
1878 	if (pm_runtime_status_suspended(dev))
1879 		return 0;
1880 
1881 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1882 
1883 	dev_pm_enable_wake_irq_check(dev, true);
1884 	ret = callback ? callback(dev) : 0;
1885 	if (ret)
1886 		goto err;
1887 
1888 	dev_pm_enable_wake_irq_complete(dev);
1889 
1890 	/*
1891 	 * If the device can stay in suspend after the system-wide transition
1892 	 * to the working state that will follow, drop the children counter of
1893 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1894 	 * function will be called again for it in the meantime.
1895 	 */
1896 	if (pm_runtime_need_not_resume(dev)) {
1897 		pm_runtime_set_suspended(dev);
1898 	} else {
1899 		__update_runtime_status(dev, RPM_SUSPENDED);
1900 		dev->power.needs_force_resume = 1;
1901 	}
1902 
1903 	return 0;
1904 
1905 err:
1906 	dev_pm_disable_wake_irq_check(dev, true);
1907 	pm_runtime_enable(dev);
1908 	return ret;
1909 }
1910 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1911 
1912 /**
1913  * pm_runtime_force_resume - Force a device into resume state if needed.
1914  * @dev: Device to resume.
1915  *
1916  * Prior invoking this function we expect the user to have brought the device
1917  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1918  * those actions and bring the device into full power, if it is expected to be
1919  * used on system resume.  In the other case, we defer the resume to be managed
1920  * via runtime PM.
1921  *
1922  * Typically this function may be invoked from a system resume callback.
1923  */
1924 int pm_runtime_force_resume(struct device *dev)
1925 {
1926 	int (*callback)(struct device *);
1927 	int ret = 0;
1928 
1929 	if (!pm_runtime_status_suspended(dev) || !dev->power.needs_force_resume)
1930 		goto out;
1931 
1932 	/*
1933 	 * The value of the parent's children counter is correct already, so
1934 	 * just update the status of the device.
1935 	 */
1936 	__update_runtime_status(dev, RPM_ACTIVE);
1937 
1938 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1939 
1940 	dev_pm_disable_wake_irq_check(dev, false);
1941 	ret = callback ? callback(dev) : 0;
1942 	if (ret) {
1943 		pm_runtime_set_suspended(dev);
1944 		dev_pm_enable_wake_irq_check(dev, false);
1945 		goto out;
1946 	}
1947 
1948 	pm_runtime_mark_last_busy(dev);
1949 out:
1950 	dev->power.needs_force_resume = 0;
1951 	pm_runtime_enable(dev);
1952 	return ret;
1953 }
1954 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1955