xref: /openbmc/linux/drivers/base/power/runtime.c (revision c4a11bf4)
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) > 0)
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 static void __rpm_put_suppliers(struct device *dev, bool try_to_suspend)
309 {
310 	struct device_link *link;
311 
312 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
313 				device_links_read_lock_held()) {
314 
315 		while (refcount_dec_not_one(&link->rpm_active))
316 			pm_runtime_put_noidle(link->supplier);
317 
318 		if (try_to_suspend)
319 			pm_request_idle(link->supplier);
320 	}
321 }
322 
323 static void rpm_put_suppliers(struct device *dev)
324 {
325 	__rpm_put_suppliers(dev, true);
326 }
327 
328 static void rpm_suspend_suppliers(struct device *dev)
329 {
330 	struct device_link *link;
331 	int idx = device_links_read_lock();
332 
333 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
334 				device_links_read_lock_held())
335 		pm_request_idle(link->supplier);
336 
337 	device_links_read_unlock(idx);
338 }
339 
340 /**
341  * __rpm_callback - Run a given runtime PM callback for a given device.
342  * @cb: Runtime PM callback to run.
343  * @dev: Device to run the callback for.
344  */
345 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
346 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
347 {
348 	int retval = 0, idx;
349 	bool use_links = dev->power.links_count > 0;
350 
351 	if (dev->power.irq_safe) {
352 		spin_unlock(&dev->power.lock);
353 	} else {
354 		spin_unlock_irq(&dev->power.lock);
355 
356 		/*
357 		 * Resume suppliers if necessary.
358 		 *
359 		 * The device's runtime PM status cannot change until this
360 		 * routine returns, so it is safe to read the status outside of
361 		 * the lock.
362 		 */
363 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
364 			idx = device_links_read_lock();
365 
366 			retval = rpm_get_suppliers(dev);
367 			if (retval) {
368 				rpm_put_suppliers(dev);
369 				goto fail;
370 			}
371 
372 			device_links_read_unlock(idx);
373 		}
374 	}
375 
376 	if (cb)
377 		retval = cb(dev);
378 
379 	if (dev->power.irq_safe) {
380 		spin_lock(&dev->power.lock);
381 	} else {
382 		/*
383 		 * If the device is suspending and the callback has returned
384 		 * success, drop the usage counters of the suppliers that have
385 		 * been reference counted on its resume.
386 		 *
387 		 * Do that if resume fails too.
388 		 */
389 		if (use_links
390 		    && ((dev->power.runtime_status == RPM_SUSPENDING && !retval)
391 		    || (dev->power.runtime_status == RPM_RESUMING && retval))) {
392 			idx = device_links_read_lock();
393 
394 			__rpm_put_suppliers(dev, false);
395 
396 fail:
397 			device_links_read_unlock(idx);
398 		}
399 
400 		spin_lock_irq(&dev->power.lock);
401 	}
402 
403 	return retval;
404 }
405 
406 /**
407  * rpm_idle - Notify device bus type if the device can be suspended.
408  * @dev: Device to notify the bus type about.
409  * @rpmflags: Flag bits.
410  *
411  * Check if the device's runtime PM status allows it to be suspended.  If
412  * another idle notification has been started earlier, return immediately.  If
413  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
414  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
415  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
416  *
417  * This function must be called under dev->power.lock with interrupts disabled.
418  */
419 static int rpm_idle(struct device *dev, int rpmflags)
420 {
421 	int (*callback)(struct device *);
422 	int retval;
423 
424 	trace_rpm_idle_rcuidle(dev, rpmflags);
425 	retval = rpm_check_suspend_allowed(dev);
426 	if (retval < 0)
427 		;	/* Conditions are wrong. */
428 
429 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
430 	else if (dev->power.runtime_status != RPM_ACTIVE)
431 		retval = -EAGAIN;
432 
433 	/*
434 	 * Any pending request other than an idle notification takes
435 	 * precedence over us, except that the timer may be running.
436 	 */
437 	else if (dev->power.request_pending &&
438 	    dev->power.request > RPM_REQ_IDLE)
439 		retval = -EAGAIN;
440 
441 	/* Act as though RPM_NOWAIT is always set. */
442 	else if (dev->power.idle_notification)
443 		retval = -EINPROGRESS;
444 	if (retval)
445 		goto out;
446 
447 	/* Pending requests need to be canceled. */
448 	dev->power.request = RPM_REQ_NONE;
449 
450 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
451 
452 	/* If no callback assume success. */
453 	if (!callback || dev->power.no_callbacks)
454 		goto out;
455 
456 	/* Carry out an asynchronous or a synchronous idle notification. */
457 	if (rpmflags & RPM_ASYNC) {
458 		dev->power.request = RPM_REQ_IDLE;
459 		if (!dev->power.request_pending) {
460 			dev->power.request_pending = true;
461 			queue_work(pm_wq, &dev->power.work);
462 		}
463 		trace_rpm_return_int_rcuidle(dev, _THIS_IP_, 0);
464 		return 0;
465 	}
466 
467 	dev->power.idle_notification = true;
468 
469 	retval = __rpm_callback(callback, dev);
470 
471 	dev->power.idle_notification = false;
472 	wake_up_all(&dev->power.wait_queue);
473 
474  out:
475 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
476 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
477 }
478 
479 /**
480  * rpm_callback - Run a given runtime PM callback for a given device.
481  * @cb: Runtime PM callback to run.
482  * @dev: Device to run the callback for.
483  */
484 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
485 {
486 	int retval;
487 
488 	if (dev->power.memalloc_noio) {
489 		unsigned int noio_flag;
490 
491 		/*
492 		 * Deadlock might be caused if memory allocation with
493 		 * GFP_KERNEL happens inside runtime_suspend and
494 		 * runtime_resume callbacks of one block device's
495 		 * ancestor or the block device itself. Network
496 		 * device might be thought as part of iSCSI block
497 		 * device, so network device and its ancestor should
498 		 * be marked as memalloc_noio too.
499 		 */
500 		noio_flag = memalloc_noio_save();
501 		retval = __rpm_callback(cb, dev);
502 		memalloc_noio_restore(noio_flag);
503 	} else {
504 		retval = __rpm_callback(cb, dev);
505 	}
506 
507 	dev->power.runtime_error = retval;
508 	return retval != -EACCES ? retval : -EIO;
509 }
510 
511 /**
512  * rpm_suspend - Carry out runtime suspend of given device.
513  * @dev: Device to suspend.
514  * @rpmflags: Flag bits.
515  *
516  * Check if the device's runtime PM status allows it to be suspended.
517  * Cancel a pending idle notification, autosuspend or suspend. If
518  * another suspend has been started earlier, either return immediately
519  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
520  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
521  * otherwise run the ->runtime_suspend() callback directly. When
522  * ->runtime_suspend succeeded, if a deferred resume was requested while
523  * the callback was running then carry it out, otherwise send an idle
524  * notification for its parent (if the suspend succeeded and both
525  * ignore_children of parent->power and irq_safe of dev->power are not set).
526  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
527  * flag is set and the next autosuspend-delay expiration time is in the
528  * future, schedule another autosuspend attempt.
529  *
530  * This function must be called under dev->power.lock with interrupts disabled.
531  */
532 static int rpm_suspend(struct device *dev, int rpmflags)
533 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
534 {
535 	int (*callback)(struct device *);
536 	struct device *parent = NULL;
537 	int retval;
538 
539 	trace_rpm_suspend_rcuidle(dev, rpmflags);
540 
541  repeat:
542 	retval = rpm_check_suspend_allowed(dev);
543 	if (retval < 0)
544 		goto out;	/* Conditions are wrong. */
545 
546 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
547 	if (dev->power.runtime_status == RPM_RESUMING && !(rpmflags & RPM_ASYNC))
548 		retval = -EAGAIN;
549 	if (retval)
550 		goto out;
551 
552 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
553 	if ((rpmflags & RPM_AUTO)
554 	    && dev->power.runtime_status != RPM_SUSPENDING) {
555 		u64 expires = pm_runtime_autosuspend_expiration(dev);
556 
557 		if (expires != 0) {
558 			/* Pending requests need to be canceled. */
559 			dev->power.request = RPM_REQ_NONE;
560 
561 			/*
562 			 * Optimization: If the timer is already running and is
563 			 * set to expire at or before the autosuspend delay,
564 			 * avoid the overhead of resetting it.  Just let it
565 			 * expire; pm_suspend_timer_fn() will take care of the
566 			 * rest.
567 			 */
568 			if (!(dev->power.timer_expires &&
569 					dev->power.timer_expires <= expires)) {
570 				/*
571 				 * We add a slack of 25% to gather wakeups
572 				 * without sacrificing the granularity.
573 				 */
574 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
575 						    (NSEC_PER_MSEC >> 2);
576 
577 				dev->power.timer_expires = expires;
578 				hrtimer_start_range_ns(&dev->power.suspend_timer,
579 						ns_to_ktime(expires),
580 						slack,
581 						HRTIMER_MODE_ABS);
582 			}
583 			dev->power.timer_autosuspends = 1;
584 			goto out;
585 		}
586 	}
587 
588 	/* Other scheduled or pending requests need to be canceled. */
589 	pm_runtime_cancel_pending(dev);
590 
591 	if (dev->power.runtime_status == RPM_SUSPENDING) {
592 		DEFINE_WAIT(wait);
593 
594 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
595 			retval = -EINPROGRESS;
596 			goto out;
597 		}
598 
599 		if (dev->power.irq_safe) {
600 			spin_unlock(&dev->power.lock);
601 
602 			cpu_relax();
603 
604 			spin_lock(&dev->power.lock);
605 			goto repeat;
606 		}
607 
608 		/* Wait for the other suspend running in parallel with us. */
609 		for (;;) {
610 			prepare_to_wait(&dev->power.wait_queue, &wait,
611 					TASK_UNINTERRUPTIBLE);
612 			if (dev->power.runtime_status != RPM_SUSPENDING)
613 				break;
614 
615 			spin_unlock_irq(&dev->power.lock);
616 
617 			schedule();
618 
619 			spin_lock_irq(&dev->power.lock);
620 		}
621 		finish_wait(&dev->power.wait_queue, &wait);
622 		goto repeat;
623 	}
624 
625 	if (dev->power.no_callbacks)
626 		goto no_callback;	/* Assume success. */
627 
628 	/* Carry out an asynchronous or a synchronous suspend. */
629 	if (rpmflags & RPM_ASYNC) {
630 		dev->power.request = (rpmflags & RPM_AUTO) ?
631 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
632 		if (!dev->power.request_pending) {
633 			dev->power.request_pending = true;
634 			queue_work(pm_wq, &dev->power.work);
635 		}
636 		goto out;
637 	}
638 
639 	__update_runtime_status(dev, RPM_SUSPENDING);
640 
641 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
642 
643 	dev_pm_enable_wake_irq_check(dev, true);
644 	retval = rpm_callback(callback, dev);
645 	if (retval)
646 		goto fail;
647 
648 	dev_pm_enable_wake_irq_complete(dev);
649 
650  no_callback:
651 	__update_runtime_status(dev, RPM_SUSPENDED);
652 	pm_runtime_deactivate_timer(dev);
653 
654 	if (dev->parent) {
655 		parent = dev->parent;
656 		atomic_add_unless(&parent->power.child_count, -1, 0);
657 	}
658 	wake_up_all(&dev->power.wait_queue);
659 
660 	if (dev->power.deferred_resume) {
661 		dev->power.deferred_resume = false;
662 		rpm_resume(dev, 0);
663 		retval = -EAGAIN;
664 		goto out;
665 	}
666 
667 	if (dev->power.irq_safe)
668 		goto out;
669 
670 	/* Maybe the parent is now able to suspend. */
671 	if (parent && !parent->power.ignore_children) {
672 		spin_unlock(&dev->power.lock);
673 
674 		spin_lock(&parent->power.lock);
675 		rpm_idle(parent, RPM_ASYNC);
676 		spin_unlock(&parent->power.lock);
677 
678 		spin_lock(&dev->power.lock);
679 	}
680 	/* Maybe the suppliers are now able to suspend. */
681 	if (dev->power.links_count > 0) {
682 		spin_unlock_irq(&dev->power.lock);
683 
684 		rpm_suspend_suppliers(dev);
685 
686 		spin_lock_irq(&dev->power.lock);
687 	}
688 
689  out:
690 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
691 
692 	return retval;
693 
694  fail:
695 	dev_pm_disable_wake_irq_check(dev, true);
696 	__update_runtime_status(dev, RPM_ACTIVE);
697 	dev->power.deferred_resume = false;
698 	wake_up_all(&dev->power.wait_queue);
699 
700 	if (retval == -EAGAIN || retval == -EBUSY) {
701 		dev->power.runtime_error = 0;
702 
703 		/*
704 		 * If the callback routine failed an autosuspend, and
705 		 * if the last_busy time has been updated so that there
706 		 * is a new autosuspend expiration time, automatically
707 		 * reschedule another autosuspend.
708 		 */
709 		if ((rpmflags & RPM_AUTO) &&
710 		    pm_runtime_autosuspend_expiration(dev) != 0)
711 			goto repeat;
712 	} else {
713 		pm_runtime_cancel_pending(dev);
714 	}
715 	goto out;
716 }
717 
718 /**
719  * rpm_resume - Carry out runtime resume of given device.
720  * @dev: Device to resume.
721  * @rpmflags: Flag bits.
722  *
723  * Check if the device's runtime PM status allows it to be resumed.  Cancel
724  * any scheduled or pending requests.  If another resume has been started
725  * earlier, either return immediately or wait for it to finish, depending on the
726  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
727  * parallel with this function, either tell the other process to resume after
728  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
729  * flag is set then queue a resume request; otherwise run the
730  * ->runtime_resume() callback directly.  Queue an idle notification for the
731  * device if the resume succeeded.
732  *
733  * This function must be called under dev->power.lock with interrupts disabled.
734  */
735 static int rpm_resume(struct device *dev, int rpmflags)
736 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
737 {
738 	int (*callback)(struct device *);
739 	struct device *parent = NULL;
740 	int retval = 0;
741 
742 	trace_rpm_resume_rcuidle(dev, rpmflags);
743 
744  repeat:
745 	if (dev->power.runtime_error)
746 		retval = -EINVAL;
747 	else if (dev->power.disable_depth == 1 && dev->power.is_suspended
748 	    && dev->power.runtime_status == RPM_ACTIVE)
749 		retval = 1;
750 	else if (dev->power.disable_depth > 0)
751 		retval = -EACCES;
752 	if (retval)
753 		goto out;
754 
755 	/*
756 	 * Other scheduled or pending requests need to be canceled.  Small
757 	 * optimization: If an autosuspend timer is running, leave it running
758 	 * rather than cancelling it now only to restart it again in the near
759 	 * future.
760 	 */
761 	dev->power.request = RPM_REQ_NONE;
762 	if (!dev->power.timer_autosuspends)
763 		pm_runtime_deactivate_timer(dev);
764 
765 	if (dev->power.runtime_status == RPM_ACTIVE) {
766 		retval = 1;
767 		goto out;
768 	}
769 
770 	if (dev->power.runtime_status == RPM_RESUMING
771 	    || dev->power.runtime_status == RPM_SUSPENDING) {
772 		DEFINE_WAIT(wait);
773 
774 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
775 			if (dev->power.runtime_status == RPM_SUSPENDING)
776 				dev->power.deferred_resume = true;
777 			else
778 				retval = -EINPROGRESS;
779 			goto out;
780 		}
781 
782 		if (dev->power.irq_safe) {
783 			spin_unlock(&dev->power.lock);
784 
785 			cpu_relax();
786 
787 			spin_lock(&dev->power.lock);
788 			goto repeat;
789 		}
790 
791 		/* Wait for the operation carried out in parallel with us. */
792 		for (;;) {
793 			prepare_to_wait(&dev->power.wait_queue, &wait,
794 					TASK_UNINTERRUPTIBLE);
795 			if (dev->power.runtime_status != RPM_RESUMING
796 			    && dev->power.runtime_status != RPM_SUSPENDING)
797 				break;
798 
799 			spin_unlock_irq(&dev->power.lock);
800 
801 			schedule();
802 
803 			spin_lock_irq(&dev->power.lock);
804 		}
805 		finish_wait(&dev->power.wait_queue, &wait);
806 		goto repeat;
807 	}
808 
809 	/*
810 	 * See if we can skip waking up the parent.  This is safe only if
811 	 * power.no_callbacks is set, because otherwise we don't know whether
812 	 * the resume will actually succeed.
813 	 */
814 	if (dev->power.no_callbacks && !parent && dev->parent) {
815 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
816 		if (dev->parent->power.disable_depth > 0
817 		    || dev->parent->power.ignore_children
818 		    || dev->parent->power.runtime_status == RPM_ACTIVE) {
819 			atomic_inc(&dev->parent->power.child_count);
820 			spin_unlock(&dev->parent->power.lock);
821 			retval = 1;
822 			goto no_callback;	/* Assume success. */
823 		}
824 		spin_unlock(&dev->parent->power.lock);
825 	}
826 
827 	/* Carry out an asynchronous or a synchronous resume. */
828 	if (rpmflags & RPM_ASYNC) {
829 		dev->power.request = RPM_REQ_RESUME;
830 		if (!dev->power.request_pending) {
831 			dev->power.request_pending = true;
832 			queue_work(pm_wq, &dev->power.work);
833 		}
834 		retval = 0;
835 		goto out;
836 	}
837 
838 	if (!parent && dev->parent) {
839 		/*
840 		 * Increment the parent's usage counter and resume it if
841 		 * necessary.  Not needed if dev is irq-safe; then the
842 		 * parent is permanently resumed.
843 		 */
844 		parent = dev->parent;
845 		if (dev->power.irq_safe)
846 			goto skip_parent;
847 		spin_unlock(&dev->power.lock);
848 
849 		pm_runtime_get_noresume(parent);
850 
851 		spin_lock(&parent->power.lock);
852 		/*
853 		 * Resume the parent if it has runtime PM enabled and not been
854 		 * set to ignore its children.
855 		 */
856 		if (!parent->power.disable_depth
857 		    && !parent->power.ignore_children) {
858 			rpm_resume(parent, 0);
859 			if (parent->power.runtime_status != RPM_ACTIVE)
860 				retval = -EBUSY;
861 		}
862 		spin_unlock(&parent->power.lock);
863 
864 		spin_lock(&dev->power.lock);
865 		if (retval)
866 			goto out;
867 		goto repeat;
868 	}
869  skip_parent:
870 
871 	if (dev->power.no_callbacks)
872 		goto no_callback;	/* Assume success. */
873 
874 	__update_runtime_status(dev, RPM_RESUMING);
875 
876 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
877 
878 	dev_pm_disable_wake_irq_check(dev, false);
879 	retval = rpm_callback(callback, dev);
880 	if (retval) {
881 		__update_runtime_status(dev, RPM_SUSPENDED);
882 		pm_runtime_cancel_pending(dev);
883 		dev_pm_enable_wake_irq_check(dev, false);
884 	} else {
885  no_callback:
886 		__update_runtime_status(dev, RPM_ACTIVE);
887 		pm_runtime_mark_last_busy(dev);
888 		if (parent)
889 			atomic_inc(&parent->power.child_count);
890 	}
891 	wake_up_all(&dev->power.wait_queue);
892 
893 	if (retval >= 0)
894 		rpm_idle(dev, RPM_ASYNC);
895 
896  out:
897 	if (parent && !dev->power.irq_safe) {
898 		spin_unlock_irq(&dev->power.lock);
899 
900 		pm_runtime_put(parent);
901 
902 		spin_lock_irq(&dev->power.lock);
903 	}
904 
905 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
906 
907 	return retval;
908 }
909 
910 /**
911  * pm_runtime_work - Universal runtime PM work function.
912  * @work: Work structure used for scheduling the execution of this function.
913  *
914  * Use @work to get the device object the work is to be done for, determine what
915  * is to be done and execute the appropriate runtime PM function.
916  */
917 static void pm_runtime_work(struct work_struct *work)
918 {
919 	struct device *dev = container_of(work, struct device, power.work);
920 	enum rpm_request req;
921 
922 	spin_lock_irq(&dev->power.lock);
923 
924 	if (!dev->power.request_pending)
925 		goto out;
926 
927 	req = dev->power.request;
928 	dev->power.request = RPM_REQ_NONE;
929 	dev->power.request_pending = false;
930 
931 	switch (req) {
932 	case RPM_REQ_NONE:
933 		break;
934 	case RPM_REQ_IDLE:
935 		rpm_idle(dev, RPM_NOWAIT);
936 		break;
937 	case RPM_REQ_SUSPEND:
938 		rpm_suspend(dev, RPM_NOWAIT);
939 		break;
940 	case RPM_REQ_AUTOSUSPEND:
941 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
942 		break;
943 	case RPM_REQ_RESUME:
944 		rpm_resume(dev, RPM_NOWAIT);
945 		break;
946 	}
947 
948  out:
949 	spin_unlock_irq(&dev->power.lock);
950 }
951 
952 /**
953  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
954  * @timer: hrtimer used by pm_schedule_suspend().
955  *
956  * Check if the time is right and queue a suspend request.
957  */
958 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
959 {
960 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
961 	unsigned long flags;
962 	u64 expires;
963 
964 	spin_lock_irqsave(&dev->power.lock, flags);
965 
966 	expires = dev->power.timer_expires;
967 	/*
968 	 * If 'expires' is after the current time, we've been called
969 	 * too early.
970 	 */
971 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
972 		dev->power.timer_expires = 0;
973 		rpm_suspend(dev, dev->power.timer_autosuspends ?
974 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
975 	}
976 
977 	spin_unlock_irqrestore(&dev->power.lock, flags);
978 
979 	return HRTIMER_NORESTART;
980 }
981 
982 /**
983  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
984  * @dev: Device to suspend.
985  * @delay: Time to wait before submitting a suspend request, in milliseconds.
986  */
987 int pm_schedule_suspend(struct device *dev, unsigned int delay)
988 {
989 	unsigned long flags;
990 	u64 expires;
991 	int retval;
992 
993 	spin_lock_irqsave(&dev->power.lock, flags);
994 
995 	if (!delay) {
996 		retval = rpm_suspend(dev, RPM_ASYNC);
997 		goto out;
998 	}
999 
1000 	retval = rpm_check_suspend_allowed(dev);
1001 	if (retval)
1002 		goto out;
1003 
1004 	/* Other scheduled or pending requests need to be canceled. */
1005 	pm_runtime_cancel_pending(dev);
1006 
1007 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
1008 	dev->power.timer_expires = expires;
1009 	dev->power.timer_autosuspends = 0;
1010 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
1011 
1012  out:
1013 	spin_unlock_irqrestore(&dev->power.lock, flags);
1014 
1015 	return retval;
1016 }
1017 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
1018 
1019 /**
1020  * __pm_runtime_idle - Entry point for runtime idle operations.
1021  * @dev: Device to send idle notification for.
1022  * @rpmflags: Flag bits.
1023  *
1024  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1025  * return immediately if it is larger than zero.  Then carry out an idle
1026  * notification, either synchronous or asynchronous.
1027  *
1028  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1029  * or if pm_runtime_irq_safe() has been called.
1030  */
1031 int __pm_runtime_idle(struct device *dev, int rpmflags)
1032 {
1033 	unsigned long flags;
1034 	int retval;
1035 
1036 	if (rpmflags & RPM_GET_PUT) {
1037 		if (!atomic_dec_and_test(&dev->power.usage_count)) {
1038 			trace_rpm_usage_rcuidle(dev, rpmflags);
1039 			return 0;
1040 		}
1041 	}
1042 
1043 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1044 
1045 	spin_lock_irqsave(&dev->power.lock, flags);
1046 	retval = rpm_idle(dev, rpmflags);
1047 	spin_unlock_irqrestore(&dev->power.lock, flags);
1048 
1049 	return retval;
1050 }
1051 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1052 
1053 /**
1054  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1055  * @dev: Device to suspend.
1056  * @rpmflags: Flag bits.
1057  *
1058  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1059  * return immediately if it is larger than zero.  Then carry out a suspend,
1060  * either synchronous or asynchronous.
1061  *
1062  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1063  * or if pm_runtime_irq_safe() has been called.
1064  */
1065 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1066 {
1067 	unsigned long flags;
1068 	int retval;
1069 
1070 	if (rpmflags & RPM_GET_PUT) {
1071 		if (!atomic_dec_and_test(&dev->power.usage_count)) {
1072 			trace_rpm_usage_rcuidle(dev, rpmflags);
1073 			return 0;
1074 		}
1075 	}
1076 
1077 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1078 
1079 	spin_lock_irqsave(&dev->power.lock, flags);
1080 	retval = rpm_suspend(dev, rpmflags);
1081 	spin_unlock_irqrestore(&dev->power.lock, flags);
1082 
1083 	return retval;
1084 }
1085 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1086 
1087 /**
1088  * __pm_runtime_resume - Entry point for runtime resume operations.
1089  * @dev: Device to resume.
1090  * @rpmflags: Flag bits.
1091  *
1092  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1093  * carry out a resume, either synchronous or asynchronous.
1094  *
1095  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1096  * or if pm_runtime_irq_safe() has been called.
1097  */
1098 int __pm_runtime_resume(struct device *dev, int rpmflags)
1099 {
1100 	unsigned long flags;
1101 	int retval;
1102 
1103 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1104 			dev->power.runtime_status != RPM_ACTIVE);
1105 
1106 	if (rpmflags & RPM_GET_PUT)
1107 		atomic_inc(&dev->power.usage_count);
1108 
1109 	spin_lock_irqsave(&dev->power.lock, flags);
1110 	retval = rpm_resume(dev, rpmflags);
1111 	spin_unlock_irqrestore(&dev->power.lock, flags);
1112 
1113 	return retval;
1114 }
1115 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1116 
1117 /**
1118  * pm_runtime_get_if_active - Conditionally bump up device usage counter.
1119  * @dev: Device to handle.
1120  * @ign_usage_count: Whether or not to look at the current usage counter value.
1121  *
1122  * Return -EINVAL if runtime PM is disabled for @dev.
1123  *
1124  * Otherwise, if the runtime PM status of @dev is %RPM_ACTIVE and either
1125  * @ign_usage_count is %true or the runtime PM usage counter of @dev is not
1126  * zero, increment the usage counter of @dev and return 1. Otherwise, return 0
1127  * without changing the usage counter.
1128  *
1129  * If @ign_usage_count is %true, this function can be used to prevent suspending
1130  * the device when its runtime PM status is %RPM_ACTIVE.
1131  *
1132  * If @ign_usage_count is %false, this function can be used to prevent
1133  * suspending the device when both its runtime PM status is %RPM_ACTIVE and its
1134  * runtime PM usage counter is not zero.
1135  *
1136  * The caller is responsible for decrementing the runtime PM usage counter of
1137  * @dev after this function has returned a positive value for it.
1138  */
1139 int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count)
1140 {
1141 	unsigned long flags;
1142 	int retval;
1143 
1144 	spin_lock_irqsave(&dev->power.lock, flags);
1145 	if (dev->power.disable_depth > 0) {
1146 		retval = -EINVAL;
1147 	} else if (dev->power.runtime_status != RPM_ACTIVE) {
1148 		retval = 0;
1149 	} else if (ign_usage_count) {
1150 		retval = 1;
1151 		atomic_inc(&dev->power.usage_count);
1152 	} else {
1153 		retval = atomic_inc_not_zero(&dev->power.usage_count);
1154 	}
1155 	trace_rpm_usage_rcuidle(dev, 0);
1156 	spin_unlock_irqrestore(&dev->power.lock, flags);
1157 
1158 	return retval;
1159 }
1160 EXPORT_SYMBOL_GPL(pm_runtime_get_if_active);
1161 
1162 /**
1163  * __pm_runtime_set_status - Set runtime PM status of a device.
1164  * @dev: Device to handle.
1165  * @status: New runtime PM status of the device.
1166  *
1167  * If runtime PM of the device is disabled or its power.runtime_error field is
1168  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1169  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1170  * However, if the device has a parent and the parent is not active, and the
1171  * parent's power.ignore_children flag is unset, the device's status cannot be
1172  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1173  *
1174  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1175  * and the device parent's counter of unsuspended children is modified to
1176  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1177  * notification request for the parent is submitted.
1178  *
1179  * If @dev has any suppliers (as reflected by device links to them), and @status
1180  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1181  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1182  * of the @status value) and the suppliers will be deacticated on exit.  The
1183  * error returned by the failing supplier activation will be returned in that
1184  * case.
1185  */
1186 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1187 {
1188 	struct device *parent = dev->parent;
1189 	bool notify_parent = false;
1190 	int error = 0;
1191 
1192 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1193 		return -EINVAL;
1194 
1195 	spin_lock_irq(&dev->power.lock);
1196 
1197 	/*
1198 	 * Prevent PM-runtime from being enabled for the device or return an
1199 	 * error if it is enabled already and working.
1200 	 */
1201 	if (dev->power.runtime_error || dev->power.disable_depth)
1202 		dev->power.disable_depth++;
1203 	else
1204 		error = -EAGAIN;
1205 
1206 	spin_unlock_irq(&dev->power.lock);
1207 
1208 	if (error)
1209 		return error;
1210 
1211 	/*
1212 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1213 	 * upfront regardless of the current status, because next time
1214 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1215 	 * involved will be dropped down to one anyway.
1216 	 */
1217 	if (status == RPM_ACTIVE) {
1218 		int idx = device_links_read_lock();
1219 
1220 		error = rpm_get_suppliers(dev);
1221 		if (error)
1222 			status = RPM_SUSPENDED;
1223 
1224 		device_links_read_unlock(idx);
1225 	}
1226 
1227 	spin_lock_irq(&dev->power.lock);
1228 
1229 	if (dev->power.runtime_status == status || !parent)
1230 		goto out_set;
1231 
1232 	if (status == RPM_SUSPENDED) {
1233 		atomic_add_unless(&parent->power.child_count, -1, 0);
1234 		notify_parent = !parent->power.ignore_children;
1235 	} else {
1236 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1237 
1238 		/*
1239 		 * It is invalid to put an active child under a parent that is
1240 		 * not active, has runtime PM enabled and the
1241 		 * 'power.ignore_children' flag unset.
1242 		 */
1243 		if (!parent->power.disable_depth
1244 		    && !parent->power.ignore_children
1245 		    && parent->power.runtime_status != RPM_ACTIVE) {
1246 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1247 				dev_name(dev),
1248 				dev_name(parent));
1249 			error = -EBUSY;
1250 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1251 			atomic_inc(&parent->power.child_count);
1252 		}
1253 
1254 		spin_unlock(&parent->power.lock);
1255 
1256 		if (error) {
1257 			status = RPM_SUSPENDED;
1258 			goto out;
1259 		}
1260 	}
1261 
1262  out_set:
1263 	__update_runtime_status(dev, status);
1264 	if (!error)
1265 		dev->power.runtime_error = 0;
1266 
1267  out:
1268 	spin_unlock_irq(&dev->power.lock);
1269 
1270 	if (notify_parent)
1271 		pm_request_idle(parent);
1272 
1273 	if (status == RPM_SUSPENDED) {
1274 		int idx = device_links_read_lock();
1275 
1276 		rpm_put_suppliers(dev);
1277 
1278 		device_links_read_unlock(idx);
1279 	}
1280 
1281 	pm_runtime_enable(dev);
1282 
1283 	return error;
1284 }
1285 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1286 
1287 /**
1288  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1289  * @dev: Device to handle.
1290  *
1291  * Flush all pending requests for the device from pm_wq and wait for all
1292  * runtime PM operations involving the device in progress to complete.
1293  *
1294  * Should be called under dev->power.lock with interrupts disabled.
1295  */
1296 static void __pm_runtime_barrier(struct device *dev)
1297 {
1298 	pm_runtime_deactivate_timer(dev);
1299 
1300 	if (dev->power.request_pending) {
1301 		dev->power.request = RPM_REQ_NONE;
1302 		spin_unlock_irq(&dev->power.lock);
1303 
1304 		cancel_work_sync(&dev->power.work);
1305 
1306 		spin_lock_irq(&dev->power.lock);
1307 		dev->power.request_pending = false;
1308 	}
1309 
1310 	if (dev->power.runtime_status == RPM_SUSPENDING
1311 	    || dev->power.runtime_status == RPM_RESUMING
1312 	    || dev->power.idle_notification) {
1313 		DEFINE_WAIT(wait);
1314 
1315 		/* Suspend, wake-up or idle notification in progress. */
1316 		for (;;) {
1317 			prepare_to_wait(&dev->power.wait_queue, &wait,
1318 					TASK_UNINTERRUPTIBLE);
1319 			if (dev->power.runtime_status != RPM_SUSPENDING
1320 			    && dev->power.runtime_status != RPM_RESUMING
1321 			    && !dev->power.idle_notification)
1322 				break;
1323 			spin_unlock_irq(&dev->power.lock);
1324 
1325 			schedule();
1326 
1327 			spin_lock_irq(&dev->power.lock);
1328 		}
1329 		finish_wait(&dev->power.wait_queue, &wait);
1330 	}
1331 }
1332 
1333 /**
1334  * pm_runtime_barrier - Flush pending requests and wait for completions.
1335  * @dev: Device to handle.
1336  *
1337  * Prevent the device from being suspended by incrementing its usage counter and
1338  * if there's a pending resume request for the device, wake the device up.
1339  * Next, make sure that all pending requests for the device have been flushed
1340  * from pm_wq and wait for all runtime PM operations involving the device in
1341  * progress to complete.
1342  *
1343  * Return value:
1344  * 1, if there was a resume request pending and the device had to be woken up,
1345  * 0, otherwise
1346  */
1347 int pm_runtime_barrier(struct device *dev)
1348 {
1349 	int retval = 0;
1350 
1351 	pm_runtime_get_noresume(dev);
1352 	spin_lock_irq(&dev->power.lock);
1353 
1354 	if (dev->power.request_pending
1355 	    && dev->power.request == RPM_REQ_RESUME) {
1356 		rpm_resume(dev, 0);
1357 		retval = 1;
1358 	}
1359 
1360 	__pm_runtime_barrier(dev);
1361 
1362 	spin_unlock_irq(&dev->power.lock);
1363 	pm_runtime_put_noidle(dev);
1364 
1365 	return retval;
1366 }
1367 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1368 
1369 /**
1370  * __pm_runtime_disable - Disable runtime PM of a device.
1371  * @dev: Device to handle.
1372  * @check_resume: If set, check if there's a resume request for the device.
1373  *
1374  * Increment power.disable_depth for the device and if it was zero previously,
1375  * cancel all pending runtime PM requests for the device and wait for all
1376  * operations in progress to complete.  The device can be either active or
1377  * suspended after its runtime PM has been disabled.
1378  *
1379  * If @check_resume is set and there's a resume request pending when
1380  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1381  * function will wake up the device before disabling its runtime PM.
1382  */
1383 void __pm_runtime_disable(struct device *dev, bool check_resume)
1384 {
1385 	spin_lock_irq(&dev->power.lock);
1386 
1387 	if (dev->power.disable_depth > 0) {
1388 		dev->power.disable_depth++;
1389 		goto out;
1390 	}
1391 
1392 	/*
1393 	 * Wake up the device if there's a resume request pending, because that
1394 	 * means there probably is some I/O to process and disabling runtime PM
1395 	 * shouldn't prevent the device from processing the I/O.
1396 	 */
1397 	if (check_resume && dev->power.request_pending
1398 	    && dev->power.request == RPM_REQ_RESUME) {
1399 		/*
1400 		 * Prevent suspends and idle notifications from being carried
1401 		 * out after we have woken up the device.
1402 		 */
1403 		pm_runtime_get_noresume(dev);
1404 
1405 		rpm_resume(dev, 0);
1406 
1407 		pm_runtime_put_noidle(dev);
1408 	}
1409 
1410 	/* Update time accounting before disabling PM-runtime. */
1411 	update_pm_runtime_accounting(dev);
1412 
1413 	if (!dev->power.disable_depth++)
1414 		__pm_runtime_barrier(dev);
1415 
1416  out:
1417 	spin_unlock_irq(&dev->power.lock);
1418 }
1419 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1420 
1421 /**
1422  * pm_runtime_enable - Enable runtime PM of a device.
1423  * @dev: Device to handle.
1424  */
1425 void pm_runtime_enable(struct device *dev)
1426 {
1427 	unsigned long flags;
1428 
1429 	spin_lock_irqsave(&dev->power.lock, flags);
1430 
1431 	if (dev->power.disable_depth > 0) {
1432 		dev->power.disable_depth--;
1433 
1434 		/* About to enable runtime pm, set accounting_timestamp to now */
1435 		if (!dev->power.disable_depth)
1436 			dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1437 	} else {
1438 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1439 	}
1440 
1441 	WARN(!dev->power.disable_depth &&
1442 	     dev->power.runtime_status == RPM_SUSPENDED &&
1443 	     !dev->power.ignore_children &&
1444 	     atomic_read(&dev->power.child_count) > 0,
1445 	     "Enabling runtime PM for inactive device (%s) with active children\n",
1446 	     dev_name(dev));
1447 
1448 	spin_unlock_irqrestore(&dev->power.lock, flags);
1449 }
1450 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1451 
1452 static void pm_runtime_disable_action(void *data)
1453 {
1454 	pm_runtime_disable(data);
1455 }
1456 
1457 /**
1458  * devm_pm_runtime_enable - devres-enabled version of pm_runtime_enable.
1459  * @dev: Device to handle.
1460  */
1461 int devm_pm_runtime_enable(struct device *dev)
1462 {
1463 	pm_runtime_enable(dev);
1464 
1465 	return devm_add_action_or_reset(dev, pm_runtime_disable_action, dev);
1466 }
1467 EXPORT_SYMBOL_GPL(devm_pm_runtime_enable);
1468 
1469 /**
1470  * pm_runtime_forbid - Block runtime PM of a device.
1471  * @dev: Device to handle.
1472  *
1473  * Increase the device's usage count and clear its power.runtime_auto flag,
1474  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1475  * for it.
1476  */
1477 void pm_runtime_forbid(struct device *dev)
1478 {
1479 	spin_lock_irq(&dev->power.lock);
1480 	if (!dev->power.runtime_auto)
1481 		goto out;
1482 
1483 	dev->power.runtime_auto = false;
1484 	atomic_inc(&dev->power.usage_count);
1485 	rpm_resume(dev, 0);
1486 
1487  out:
1488 	spin_unlock_irq(&dev->power.lock);
1489 }
1490 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1491 
1492 /**
1493  * pm_runtime_allow - Unblock runtime PM of a device.
1494  * @dev: Device to handle.
1495  *
1496  * Decrease the device's usage count and set its power.runtime_auto flag.
1497  */
1498 void pm_runtime_allow(struct device *dev)
1499 {
1500 	spin_lock_irq(&dev->power.lock);
1501 	if (dev->power.runtime_auto)
1502 		goto out;
1503 
1504 	dev->power.runtime_auto = true;
1505 	if (atomic_dec_and_test(&dev->power.usage_count))
1506 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1507 	else
1508 		trace_rpm_usage_rcuidle(dev, RPM_AUTO | RPM_ASYNC);
1509 
1510  out:
1511 	spin_unlock_irq(&dev->power.lock);
1512 }
1513 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1514 
1515 /**
1516  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1517  * @dev: Device to handle.
1518  *
1519  * Set the power.no_callbacks flag, which tells the PM core that this
1520  * device is power-managed through its parent and has no runtime PM
1521  * callbacks of its own.  The runtime sysfs attributes will be removed.
1522  */
1523 void pm_runtime_no_callbacks(struct device *dev)
1524 {
1525 	spin_lock_irq(&dev->power.lock);
1526 	dev->power.no_callbacks = 1;
1527 	spin_unlock_irq(&dev->power.lock);
1528 	if (device_is_registered(dev))
1529 		rpm_sysfs_remove(dev);
1530 }
1531 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1532 
1533 /**
1534  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1535  * @dev: Device to handle
1536  *
1537  * Set the power.irq_safe flag, which tells the PM core that the
1538  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1539  * always be invoked with the spinlock held and interrupts disabled.  It also
1540  * causes the parent's usage counter to be permanently incremented, preventing
1541  * the parent from runtime suspending -- otherwise an irq-safe child might have
1542  * to wait for a non-irq-safe parent.
1543  */
1544 void pm_runtime_irq_safe(struct device *dev)
1545 {
1546 	if (dev->parent)
1547 		pm_runtime_get_sync(dev->parent);
1548 	spin_lock_irq(&dev->power.lock);
1549 	dev->power.irq_safe = 1;
1550 	spin_unlock_irq(&dev->power.lock);
1551 }
1552 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1553 
1554 /**
1555  * update_autosuspend - Handle a change to a device's autosuspend settings.
1556  * @dev: Device to handle.
1557  * @old_delay: The former autosuspend_delay value.
1558  * @old_use: The former use_autosuspend value.
1559  *
1560  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1561  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1562  *
1563  * This function must be called under dev->power.lock with interrupts disabled.
1564  */
1565 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1566 {
1567 	int delay = dev->power.autosuspend_delay;
1568 
1569 	/* Should runtime suspend be prevented now? */
1570 	if (dev->power.use_autosuspend && delay < 0) {
1571 
1572 		/* If it used to be allowed then prevent it. */
1573 		if (!old_use || old_delay >= 0) {
1574 			atomic_inc(&dev->power.usage_count);
1575 			rpm_resume(dev, 0);
1576 		} else {
1577 			trace_rpm_usage_rcuidle(dev, 0);
1578 		}
1579 	}
1580 
1581 	/* Runtime suspend should be allowed now. */
1582 	else {
1583 
1584 		/* If it used to be prevented then allow it. */
1585 		if (old_use && old_delay < 0)
1586 			atomic_dec(&dev->power.usage_count);
1587 
1588 		/* Maybe we can autosuspend now. */
1589 		rpm_idle(dev, RPM_AUTO);
1590 	}
1591 }
1592 
1593 /**
1594  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1595  * @dev: Device to handle.
1596  * @delay: Value of the new delay in milliseconds.
1597  *
1598  * Set the device's power.autosuspend_delay value.  If it changes to negative
1599  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1600  * changes the other way, allow runtime suspends.
1601  */
1602 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1603 {
1604 	int old_delay, old_use;
1605 
1606 	spin_lock_irq(&dev->power.lock);
1607 	old_delay = dev->power.autosuspend_delay;
1608 	old_use = dev->power.use_autosuspend;
1609 	dev->power.autosuspend_delay = delay;
1610 	update_autosuspend(dev, old_delay, old_use);
1611 	spin_unlock_irq(&dev->power.lock);
1612 }
1613 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1614 
1615 /**
1616  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1617  * @dev: Device to handle.
1618  * @use: New value for use_autosuspend.
1619  *
1620  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1621  * suspends as needed.
1622  */
1623 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1624 {
1625 	int old_delay, old_use;
1626 
1627 	spin_lock_irq(&dev->power.lock);
1628 	old_delay = dev->power.autosuspend_delay;
1629 	old_use = dev->power.use_autosuspend;
1630 	dev->power.use_autosuspend = use;
1631 	update_autosuspend(dev, old_delay, old_use);
1632 	spin_unlock_irq(&dev->power.lock);
1633 }
1634 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1635 
1636 /**
1637  * pm_runtime_init - Initialize runtime PM fields in given device object.
1638  * @dev: Device object to initialize.
1639  */
1640 void pm_runtime_init(struct device *dev)
1641 {
1642 	dev->power.runtime_status = RPM_SUSPENDED;
1643 	dev->power.idle_notification = false;
1644 
1645 	dev->power.disable_depth = 1;
1646 	atomic_set(&dev->power.usage_count, 0);
1647 
1648 	dev->power.runtime_error = 0;
1649 
1650 	atomic_set(&dev->power.child_count, 0);
1651 	pm_suspend_ignore_children(dev, false);
1652 	dev->power.runtime_auto = true;
1653 
1654 	dev->power.request_pending = false;
1655 	dev->power.request = RPM_REQ_NONE;
1656 	dev->power.deferred_resume = false;
1657 	dev->power.needs_force_resume = 0;
1658 	INIT_WORK(&dev->power.work, pm_runtime_work);
1659 
1660 	dev->power.timer_expires = 0;
1661 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1662 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1663 
1664 	init_waitqueue_head(&dev->power.wait_queue);
1665 }
1666 
1667 /**
1668  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1669  * @dev: Device object to re-initialize.
1670  */
1671 void pm_runtime_reinit(struct device *dev)
1672 {
1673 	if (!pm_runtime_enabled(dev)) {
1674 		if (dev->power.runtime_status == RPM_ACTIVE)
1675 			pm_runtime_set_suspended(dev);
1676 		if (dev->power.irq_safe) {
1677 			spin_lock_irq(&dev->power.lock);
1678 			dev->power.irq_safe = 0;
1679 			spin_unlock_irq(&dev->power.lock);
1680 			if (dev->parent)
1681 				pm_runtime_put(dev->parent);
1682 		}
1683 	}
1684 }
1685 
1686 /**
1687  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1688  * @dev: Device object being removed from device hierarchy.
1689  */
1690 void pm_runtime_remove(struct device *dev)
1691 {
1692 	__pm_runtime_disable(dev, false);
1693 	pm_runtime_reinit(dev);
1694 }
1695 
1696 /**
1697  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1698  * @dev: Consumer device.
1699  */
1700 void pm_runtime_get_suppliers(struct device *dev)
1701 {
1702 	struct device_link *link;
1703 	int idx;
1704 
1705 	idx = device_links_read_lock();
1706 
1707 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1708 				device_links_read_lock_held())
1709 		if (link->flags & DL_FLAG_PM_RUNTIME) {
1710 			link->supplier_preactivated = true;
1711 			pm_runtime_get_sync(link->supplier);
1712 			refcount_inc(&link->rpm_active);
1713 		}
1714 
1715 	device_links_read_unlock(idx);
1716 }
1717 
1718 /**
1719  * pm_runtime_put_suppliers - Drop references to supplier devices.
1720  * @dev: Consumer device.
1721  */
1722 void pm_runtime_put_suppliers(struct device *dev)
1723 {
1724 	struct device_link *link;
1725 	unsigned long flags;
1726 	bool put;
1727 	int idx;
1728 
1729 	idx = device_links_read_lock();
1730 
1731 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1732 				device_links_read_lock_held())
1733 		if (link->supplier_preactivated) {
1734 			link->supplier_preactivated = false;
1735 			spin_lock_irqsave(&dev->power.lock, flags);
1736 			put = pm_runtime_status_suspended(dev) &&
1737 			      refcount_dec_not_one(&link->rpm_active);
1738 			spin_unlock_irqrestore(&dev->power.lock, flags);
1739 			if (put)
1740 				pm_runtime_put(link->supplier);
1741 		}
1742 
1743 	device_links_read_unlock(idx);
1744 }
1745 
1746 void pm_runtime_new_link(struct device *dev)
1747 {
1748 	spin_lock_irq(&dev->power.lock);
1749 	dev->power.links_count++;
1750 	spin_unlock_irq(&dev->power.lock);
1751 }
1752 
1753 static void pm_runtime_drop_link_count(struct device *dev)
1754 {
1755 	spin_lock_irq(&dev->power.lock);
1756 	WARN_ON(dev->power.links_count == 0);
1757 	dev->power.links_count--;
1758 	spin_unlock_irq(&dev->power.lock);
1759 }
1760 
1761 /**
1762  * pm_runtime_drop_link - Prepare for device link removal.
1763  * @link: Device link going away.
1764  *
1765  * Drop the link count of the consumer end of @link and decrement the supplier
1766  * device's runtime PM usage counter as many times as needed to drop all of the
1767  * PM runtime reference to it from the consumer.
1768  */
1769 void pm_runtime_drop_link(struct device_link *link)
1770 {
1771 	if (!(link->flags & DL_FLAG_PM_RUNTIME))
1772 		return;
1773 
1774 	pm_runtime_drop_link_count(link->consumer);
1775 
1776 	while (refcount_dec_not_one(&link->rpm_active))
1777 		pm_runtime_put(link->supplier);
1778 }
1779 
1780 static bool pm_runtime_need_not_resume(struct device *dev)
1781 {
1782 	return atomic_read(&dev->power.usage_count) <= 1 &&
1783 		(atomic_read(&dev->power.child_count) == 0 ||
1784 		 dev->power.ignore_children);
1785 }
1786 
1787 /**
1788  * pm_runtime_force_suspend - Force a device into suspend state if needed.
1789  * @dev: Device to suspend.
1790  *
1791  * Disable runtime PM so we safely can check the device's runtime PM status and
1792  * if it is active, invoke its ->runtime_suspend callback to suspend it and
1793  * change its runtime PM status field to RPM_SUSPENDED.  Also, if the device's
1794  * usage and children counters don't indicate that the device was in use before
1795  * the system-wide transition under way, decrement its parent's children counter
1796  * (if there is a parent).  Keep runtime PM disabled to preserve the state
1797  * unless we encounter errors.
1798  *
1799  * Typically this function may be invoked from a system suspend callback to make
1800  * sure the device is put into low power state and it should only be used during
1801  * system-wide PM transitions to sleep states.  It assumes that the analogous
1802  * pm_runtime_force_resume() will be used to resume the device.
1803  */
1804 int pm_runtime_force_suspend(struct device *dev)
1805 {
1806 	int (*callback)(struct device *);
1807 	int ret;
1808 
1809 	pm_runtime_disable(dev);
1810 	if (pm_runtime_status_suspended(dev))
1811 		return 0;
1812 
1813 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1814 
1815 	ret = callback ? callback(dev) : 0;
1816 	if (ret)
1817 		goto err;
1818 
1819 	/*
1820 	 * If the device can stay in suspend after the system-wide transition
1821 	 * to the working state that will follow, drop the children counter of
1822 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1823 	 * function will be called again for it in the meantime.
1824 	 */
1825 	if (pm_runtime_need_not_resume(dev)) {
1826 		pm_runtime_set_suspended(dev);
1827 	} else {
1828 		__update_runtime_status(dev, RPM_SUSPENDED);
1829 		dev->power.needs_force_resume = 1;
1830 	}
1831 
1832 	return 0;
1833 
1834 err:
1835 	pm_runtime_enable(dev);
1836 	return ret;
1837 }
1838 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1839 
1840 /**
1841  * pm_runtime_force_resume - Force a device into resume state if needed.
1842  * @dev: Device to resume.
1843  *
1844  * Prior invoking this function we expect the user to have brought the device
1845  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1846  * those actions and bring the device into full power, if it is expected to be
1847  * used on system resume.  In the other case, we defer the resume to be managed
1848  * via runtime PM.
1849  *
1850  * Typically this function may be invoked from a system resume callback.
1851  */
1852 int pm_runtime_force_resume(struct device *dev)
1853 {
1854 	int (*callback)(struct device *);
1855 	int ret = 0;
1856 
1857 	if (!pm_runtime_status_suspended(dev) || !dev->power.needs_force_resume)
1858 		goto out;
1859 
1860 	/*
1861 	 * The value of the parent's children counter is correct already, so
1862 	 * just update the status of the device.
1863 	 */
1864 	__update_runtime_status(dev, RPM_ACTIVE);
1865 
1866 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1867 
1868 	ret = callback ? callback(dev) : 0;
1869 	if (ret) {
1870 		pm_runtime_set_suspended(dev);
1871 		goto out;
1872 	}
1873 
1874 	pm_runtime_mark_last_busy(dev);
1875 out:
1876 	dev->power.needs_force_resume = 0;
1877 	pm_runtime_enable(dev);
1878 	return ret;
1879 }
1880 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1881