xref: /openbmc/linux/drivers/base/power/runtime.c (revision c059ee9d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * drivers/base/power/runtime.c - Helper functions for device runtime PM
4  *
5  * Copyright (c) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
6  * Copyright (C) 2010 Alan Stern <stern@rowland.harvard.edu>
7  */
8 #include <linux/sched/mm.h>
9 #include <linux/ktime.h>
10 #include <linux/hrtimer.h>
11 #include <linux/export.h>
12 #include <linux/pm_runtime.h>
13 #include <linux/pm_wakeirq.h>
14 #include <trace/events/rpm.h>
15 
16 #include "../base.h"
17 #include "power.h"
18 
19 typedef int (*pm_callback_t)(struct device *);
20 
21 static pm_callback_t __rpm_get_callback(struct device *dev, size_t cb_offset)
22 {
23 	pm_callback_t cb;
24 	const struct dev_pm_ops *ops;
25 
26 	if (dev->pm_domain)
27 		ops = &dev->pm_domain->ops;
28 	else if (dev->type && dev->type->pm)
29 		ops = dev->type->pm;
30 	else if (dev->class && dev->class->pm)
31 		ops = dev->class->pm;
32 	else if (dev->bus && dev->bus->pm)
33 		ops = dev->bus->pm;
34 	else
35 		ops = NULL;
36 
37 	if (ops)
38 		cb = *(pm_callback_t *)((void *)ops + cb_offset);
39 	else
40 		cb = NULL;
41 
42 	if (!cb && dev->driver && dev->driver->pm)
43 		cb = *(pm_callback_t *)((void *)dev->driver->pm + cb_offset);
44 
45 	return cb;
46 }
47 
48 #define RPM_GET_CALLBACK(dev, callback) \
49 		__rpm_get_callback(dev, offsetof(struct dev_pm_ops, callback))
50 
51 static int rpm_resume(struct device *dev, int rpmflags);
52 static int rpm_suspend(struct device *dev, int rpmflags);
53 
54 /**
55  * update_pm_runtime_accounting - Update the time accounting of power states
56  * @dev: Device to update the accounting for
57  *
58  * In order to be able to have time accounting of the various power states
59  * (as used by programs such as PowerTOP to show the effectiveness of runtime
60  * PM), we need to track the time spent in each state.
61  * update_pm_runtime_accounting must be called each time before the
62  * runtime_status field is updated, to account the time in the old state
63  * correctly.
64  */
65 static void update_pm_runtime_accounting(struct device *dev)
66 {
67 	u64 now, last, delta;
68 
69 	if (dev->power.disable_depth > 0)
70 		return;
71 
72 	last = dev->power.accounting_timestamp;
73 
74 	now = ktime_get_mono_fast_ns();
75 	dev->power.accounting_timestamp = now;
76 
77 	/*
78 	 * Because ktime_get_mono_fast_ns() is not monotonic during
79 	 * timekeeping updates, ensure that 'now' is after the last saved
80 	 * timesptamp.
81 	 */
82 	if (now < last)
83 		return;
84 
85 	delta = now - last;
86 
87 	if (dev->power.runtime_status == RPM_SUSPENDED)
88 		dev->power.suspended_time += delta;
89 	else
90 		dev->power.active_time += delta;
91 }
92 
93 static void __update_runtime_status(struct device *dev, enum rpm_status status)
94 {
95 	update_pm_runtime_accounting(dev);
96 	dev->power.runtime_status = status;
97 }
98 
99 static u64 rpm_get_accounted_time(struct device *dev, bool suspended)
100 {
101 	u64 time;
102 	unsigned long flags;
103 
104 	spin_lock_irqsave(&dev->power.lock, flags);
105 
106 	update_pm_runtime_accounting(dev);
107 	time = suspended ? dev->power.suspended_time : dev->power.active_time;
108 
109 	spin_unlock_irqrestore(&dev->power.lock, flags);
110 
111 	return time;
112 }
113 
114 u64 pm_runtime_active_time(struct device *dev)
115 {
116 	return rpm_get_accounted_time(dev, false);
117 }
118 
119 u64 pm_runtime_suspended_time(struct device *dev)
120 {
121 	return rpm_get_accounted_time(dev, true);
122 }
123 EXPORT_SYMBOL_GPL(pm_runtime_suspended_time);
124 
125 /**
126  * pm_runtime_deactivate_timer - Deactivate given device's suspend timer.
127  * @dev: Device to handle.
128  */
129 static void pm_runtime_deactivate_timer(struct device *dev)
130 {
131 	if (dev->power.timer_expires > 0) {
132 		hrtimer_try_to_cancel(&dev->power.suspend_timer);
133 		dev->power.timer_expires = 0;
134 	}
135 }
136 
137 /**
138  * pm_runtime_cancel_pending - Deactivate suspend timer and cancel requests.
139  * @dev: Device to handle.
140  */
141 static void pm_runtime_cancel_pending(struct device *dev)
142 {
143 	pm_runtime_deactivate_timer(dev);
144 	/*
145 	 * In case there's a request pending, make sure its work function will
146 	 * return without doing anything.
147 	 */
148 	dev->power.request = RPM_REQ_NONE;
149 }
150 
151 /*
152  * pm_runtime_autosuspend_expiration - Get a device's autosuspend-delay expiration time.
153  * @dev: Device to handle.
154  *
155  * Compute the autosuspend-delay expiration time based on the device's
156  * power.last_busy time.  If the delay has already expired or is disabled
157  * (negative) or the power.use_autosuspend flag isn't set, return 0.
158  * Otherwise return the expiration time in nanoseconds (adjusted to be nonzero).
159  *
160  * This function may be called either with or without dev->power.lock held.
161  * Either way it can be racy, since power.last_busy may be updated at any time.
162  */
163 u64 pm_runtime_autosuspend_expiration(struct device *dev)
164 {
165 	int autosuspend_delay;
166 	u64 expires;
167 
168 	if (!dev->power.use_autosuspend)
169 		return 0;
170 
171 	autosuspend_delay = READ_ONCE(dev->power.autosuspend_delay);
172 	if (autosuspend_delay < 0)
173 		return 0;
174 
175 	expires  = READ_ONCE(dev->power.last_busy);
176 	expires += (u64)autosuspend_delay * NSEC_PER_MSEC;
177 	if (expires > ktime_get_mono_fast_ns())
178 		return expires;	/* Expires in the future */
179 
180 	return 0;
181 }
182 EXPORT_SYMBOL_GPL(pm_runtime_autosuspend_expiration);
183 
184 static int dev_memalloc_noio(struct device *dev, void *data)
185 {
186 	return dev->power.memalloc_noio;
187 }
188 
189 /*
190  * pm_runtime_set_memalloc_noio - Set a device's memalloc_noio flag.
191  * @dev: Device to handle.
192  * @enable: True for setting the flag and False for clearing the flag.
193  *
194  * Set the flag for all devices in the path from the device to the
195  * root device in the device tree if @enable is true, otherwise clear
196  * the flag for devices in the path whose siblings don't set the flag.
197  *
198  * The function should only be called by block device, or network
199  * device driver for solving the deadlock problem during runtime
200  * resume/suspend:
201  *
202  *     If memory allocation with GFP_KERNEL is called inside runtime
203  *     resume/suspend callback of any one of its ancestors(or the
204  *     block device itself), the deadlock may be triggered inside the
205  *     memory allocation since it might not complete until the block
206  *     device becomes active and the involed page I/O finishes. The
207  *     situation is pointed out first by Alan Stern. Network device
208  *     are involved in iSCSI kind of situation.
209  *
210  * The lock of dev_hotplug_mutex is held in the function for handling
211  * hotplug race because pm_runtime_set_memalloc_noio() may be called
212  * in async probe().
213  *
214  * The function should be called between device_add() and device_del()
215  * on the affected device(block/network device).
216  */
217 void pm_runtime_set_memalloc_noio(struct device *dev, bool enable)
218 {
219 	static DEFINE_MUTEX(dev_hotplug_mutex);
220 
221 	mutex_lock(&dev_hotplug_mutex);
222 	for (;;) {
223 		bool enabled;
224 
225 		/* hold power lock since bitfield is not SMP-safe. */
226 		spin_lock_irq(&dev->power.lock);
227 		enabled = dev->power.memalloc_noio;
228 		dev->power.memalloc_noio = enable;
229 		spin_unlock_irq(&dev->power.lock);
230 
231 		/*
232 		 * not need to enable ancestors any more if the device
233 		 * has been enabled.
234 		 */
235 		if (enabled && enable)
236 			break;
237 
238 		dev = dev->parent;
239 
240 		/*
241 		 * clear flag of the parent device only if all the
242 		 * children don't set the flag because ancestor's
243 		 * flag was set by any one of the descendants.
244 		 */
245 		if (!dev || (!enable &&
246 			     device_for_each_child(dev, NULL,
247 						   dev_memalloc_noio)))
248 			break;
249 	}
250 	mutex_unlock(&dev_hotplug_mutex);
251 }
252 EXPORT_SYMBOL_GPL(pm_runtime_set_memalloc_noio);
253 
254 /**
255  * rpm_check_suspend_allowed - Test whether a device may be suspended.
256  * @dev: Device to test.
257  */
258 static int rpm_check_suspend_allowed(struct device *dev)
259 {
260 	int retval = 0;
261 
262 	if (dev->power.runtime_error)
263 		retval = -EINVAL;
264 	else if (dev->power.disable_depth > 0)
265 		retval = -EACCES;
266 	else if (atomic_read(&dev->power.usage_count))
267 		retval = -EAGAIN;
268 	else if (!dev->power.ignore_children &&
269 			atomic_read(&dev->power.child_count))
270 		retval = -EBUSY;
271 
272 	/* Pending resume requests take precedence over suspends. */
273 	else if ((dev->power.deferred_resume
274 			&& dev->power.runtime_status == RPM_SUSPENDING)
275 	    || (dev->power.request_pending
276 			&& dev->power.request == RPM_REQ_RESUME))
277 		retval = -EAGAIN;
278 	else if (__dev_pm_qos_resume_latency(dev) == 0)
279 		retval = -EPERM;
280 	else if (dev->power.runtime_status == RPM_SUSPENDED)
281 		retval = 1;
282 
283 	return retval;
284 }
285 
286 static int rpm_get_suppliers(struct device *dev)
287 {
288 	struct device_link *link;
289 
290 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
291 				device_links_read_lock_held()) {
292 		int retval;
293 
294 		if (!(link->flags & DL_FLAG_PM_RUNTIME))
295 			continue;
296 
297 		retval = pm_runtime_get_sync(link->supplier);
298 		/* Ignore suppliers with disabled runtime PM. */
299 		if (retval < 0 && retval != -EACCES) {
300 			pm_runtime_put_noidle(link->supplier);
301 			return retval;
302 		}
303 		refcount_inc(&link->rpm_active);
304 	}
305 	return 0;
306 }
307 
308 /**
309  * pm_runtime_release_supplier - Drop references to device link's supplier.
310  * @link: Target device link.
311  * @check_idle: Whether or not to check if the supplier device is idle.
312  *
313  * Drop all runtime PM references associated with @link to its supplier device
314  * and if @check_idle is set, check if that device is idle (and so it can be
315  * suspended).
316  */
317 void pm_runtime_release_supplier(struct device_link *link, bool check_idle)
318 {
319 	struct device *supplier = link->supplier;
320 
321 	/*
322 	 * The additional power.usage_count check is a safety net in case
323 	 * the rpm_active refcount becomes saturated, in which case
324 	 * refcount_dec_not_one() would return true forever, but it is not
325 	 * strictly necessary.
326 	 */
327 	while (refcount_dec_not_one(&link->rpm_active) &&
328 	       atomic_read(&supplier->power.usage_count) > 0)
329 		pm_runtime_put_noidle(supplier);
330 
331 	if (check_idle)
332 		pm_request_idle(supplier);
333 }
334 
335 static void __rpm_put_suppliers(struct device *dev, bool try_to_suspend)
336 {
337 	struct device_link *link;
338 
339 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
340 				device_links_read_lock_held())
341 		pm_runtime_release_supplier(link, try_to_suspend);
342 }
343 
344 static void rpm_put_suppliers(struct device *dev)
345 {
346 	__rpm_put_suppliers(dev, true);
347 }
348 
349 static void rpm_suspend_suppliers(struct device *dev)
350 {
351 	struct device_link *link;
352 	int idx = device_links_read_lock();
353 
354 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
355 				device_links_read_lock_held())
356 		pm_request_idle(link->supplier);
357 
358 	device_links_read_unlock(idx);
359 }
360 
361 /**
362  * __rpm_callback - Run a given runtime PM callback for a given device.
363  * @cb: Runtime PM callback to run.
364  * @dev: Device to run the callback for.
365  */
366 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
367 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
368 {
369 	int retval = 0, idx;
370 	bool use_links = dev->power.links_count > 0;
371 
372 	if (dev->power.irq_safe) {
373 		spin_unlock(&dev->power.lock);
374 	} else {
375 		spin_unlock_irq(&dev->power.lock);
376 
377 		/*
378 		 * Resume suppliers if necessary.
379 		 *
380 		 * The device's runtime PM status cannot change until this
381 		 * routine returns, so it is safe to read the status outside of
382 		 * the lock.
383 		 */
384 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
385 			idx = device_links_read_lock();
386 
387 			retval = rpm_get_suppliers(dev);
388 			if (retval) {
389 				rpm_put_suppliers(dev);
390 				goto fail;
391 			}
392 
393 			device_links_read_unlock(idx);
394 		}
395 	}
396 
397 	if (cb)
398 		retval = cb(dev);
399 
400 	if (dev->power.irq_safe) {
401 		spin_lock(&dev->power.lock);
402 	} else {
403 		/*
404 		 * If the device is suspending and the callback has returned
405 		 * success, drop the usage counters of the suppliers that have
406 		 * been reference counted on its resume.
407 		 *
408 		 * Do that if resume fails too.
409 		 */
410 		if (use_links
411 		    && ((dev->power.runtime_status == RPM_SUSPENDING && !retval)
412 		    || (dev->power.runtime_status == RPM_RESUMING && retval))) {
413 			idx = device_links_read_lock();
414 
415 			__rpm_put_suppliers(dev, false);
416 
417 fail:
418 			device_links_read_unlock(idx);
419 		}
420 
421 		spin_lock_irq(&dev->power.lock);
422 	}
423 
424 	return retval;
425 }
426 
427 /**
428  * rpm_idle - Notify device bus type if the device can be suspended.
429  * @dev: Device to notify the bus type about.
430  * @rpmflags: Flag bits.
431  *
432  * Check if the device's runtime PM status allows it to be suspended.  If
433  * another idle notification has been started earlier, return immediately.  If
434  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
435  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
436  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
437  *
438  * This function must be called under dev->power.lock with interrupts disabled.
439  */
440 static int rpm_idle(struct device *dev, int rpmflags)
441 {
442 	int (*callback)(struct device *);
443 	int retval;
444 
445 	trace_rpm_idle_rcuidle(dev, rpmflags);
446 	retval = rpm_check_suspend_allowed(dev);
447 	if (retval < 0)
448 		;	/* Conditions are wrong. */
449 
450 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
451 	else if (dev->power.runtime_status != RPM_ACTIVE)
452 		retval = -EAGAIN;
453 
454 	/*
455 	 * Any pending request other than an idle notification takes
456 	 * precedence over us, except that the timer may be running.
457 	 */
458 	else if (dev->power.request_pending &&
459 	    dev->power.request > RPM_REQ_IDLE)
460 		retval = -EAGAIN;
461 
462 	/* Act as though RPM_NOWAIT is always set. */
463 	else if (dev->power.idle_notification)
464 		retval = -EINPROGRESS;
465 	if (retval)
466 		goto out;
467 
468 	/* Pending requests need to be canceled. */
469 	dev->power.request = RPM_REQ_NONE;
470 
471 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
472 
473 	/* If no callback assume success. */
474 	if (!callback || dev->power.no_callbacks)
475 		goto out;
476 
477 	/* Carry out an asynchronous or a synchronous idle notification. */
478 	if (rpmflags & RPM_ASYNC) {
479 		dev->power.request = RPM_REQ_IDLE;
480 		if (!dev->power.request_pending) {
481 			dev->power.request_pending = true;
482 			queue_work(pm_wq, &dev->power.work);
483 		}
484 		trace_rpm_return_int_rcuidle(dev, _THIS_IP_, 0);
485 		return 0;
486 	}
487 
488 	dev->power.idle_notification = true;
489 
490 	retval = __rpm_callback(callback, dev);
491 
492 	dev->power.idle_notification = false;
493 	wake_up_all(&dev->power.wait_queue);
494 
495  out:
496 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
497 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
498 }
499 
500 /**
501  * rpm_callback - Run a given runtime PM callback for a given device.
502  * @cb: Runtime PM callback to run.
503  * @dev: Device to run the callback for.
504  */
505 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
506 {
507 	int retval;
508 
509 	if (dev->power.memalloc_noio) {
510 		unsigned int noio_flag;
511 
512 		/*
513 		 * Deadlock might be caused if memory allocation with
514 		 * GFP_KERNEL happens inside runtime_suspend and
515 		 * runtime_resume callbacks of one block device's
516 		 * ancestor or the block device itself. Network
517 		 * device might be thought as part of iSCSI block
518 		 * device, so network device and its ancestor should
519 		 * be marked as memalloc_noio too.
520 		 */
521 		noio_flag = memalloc_noio_save();
522 		retval = __rpm_callback(cb, dev);
523 		memalloc_noio_restore(noio_flag);
524 	} else {
525 		retval = __rpm_callback(cb, dev);
526 	}
527 
528 	dev->power.runtime_error = retval;
529 	return retval != -EACCES ? retval : -EIO;
530 }
531 
532 /**
533  * rpm_suspend - Carry out runtime suspend of given device.
534  * @dev: Device to suspend.
535  * @rpmflags: Flag bits.
536  *
537  * Check if the device's runtime PM status allows it to be suspended.
538  * Cancel a pending idle notification, autosuspend or suspend. If
539  * another suspend has been started earlier, either return immediately
540  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
541  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
542  * otherwise run the ->runtime_suspend() callback directly. When
543  * ->runtime_suspend succeeded, if a deferred resume was requested while
544  * the callback was running then carry it out, otherwise send an idle
545  * notification for its parent (if the suspend succeeded and both
546  * ignore_children of parent->power and irq_safe of dev->power are not set).
547  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
548  * flag is set and the next autosuspend-delay expiration time is in the
549  * future, schedule another autosuspend attempt.
550  *
551  * This function must be called under dev->power.lock with interrupts disabled.
552  */
553 static int rpm_suspend(struct device *dev, int rpmflags)
554 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
555 {
556 	int (*callback)(struct device *);
557 	struct device *parent = NULL;
558 	int retval;
559 
560 	trace_rpm_suspend_rcuidle(dev, rpmflags);
561 
562  repeat:
563 	retval = rpm_check_suspend_allowed(dev);
564 	if (retval < 0)
565 		goto out;	/* Conditions are wrong. */
566 
567 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
568 	if (dev->power.runtime_status == RPM_RESUMING && !(rpmflags & RPM_ASYNC))
569 		retval = -EAGAIN;
570 	if (retval)
571 		goto out;
572 
573 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
574 	if ((rpmflags & RPM_AUTO)
575 	    && dev->power.runtime_status != RPM_SUSPENDING) {
576 		u64 expires = pm_runtime_autosuspend_expiration(dev);
577 
578 		if (expires != 0) {
579 			/* Pending requests need to be canceled. */
580 			dev->power.request = RPM_REQ_NONE;
581 
582 			/*
583 			 * Optimization: If the timer is already running and is
584 			 * set to expire at or before the autosuspend delay,
585 			 * avoid the overhead of resetting it.  Just let it
586 			 * expire; pm_suspend_timer_fn() will take care of the
587 			 * rest.
588 			 */
589 			if (!(dev->power.timer_expires &&
590 					dev->power.timer_expires <= expires)) {
591 				/*
592 				 * We add a slack of 25% to gather wakeups
593 				 * without sacrificing the granularity.
594 				 */
595 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
596 						    (NSEC_PER_MSEC >> 2);
597 
598 				dev->power.timer_expires = expires;
599 				hrtimer_start_range_ns(&dev->power.suspend_timer,
600 						ns_to_ktime(expires),
601 						slack,
602 						HRTIMER_MODE_ABS);
603 			}
604 			dev->power.timer_autosuspends = 1;
605 			goto out;
606 		}
607 	}
608 
609 	/* Other scheduled or pending requests need to be canceled. */
610 	pm_runtime_cancel_pending(dev);
611 
612 	if (dev->power.runtime_status == RPM_SUSPENDING) {
613 		DEFINE_WAIT(wait);
614 
615 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
616 			retval = -EINPROGRESS;
617 			goto out;
618 		}
619 
620 		if (dev->power.irq_safe) {
621 			spin_unlock(&dev->power.lock);
622 
623 			cpu_relax();
624 
625 			spin_lock(&dev->power.lock);
626 			goto repeat;
627 		}
628 
629 		/* Wait for the other suspend running in parallel with us. */
630 		for (;;) {
631 			prepare_to_wait(&dev->power.wait_queue, &wait,
632 					TASK_UNINTERRUPTIBLE);
633 			if (dev->power.runtime_status != RPM_SUSPENDING)
634 				break;
635 
636 			spin_unlock_irq(&dev->power.lock);
637 
638 			schedule();
639 
640 			spin_lock_irq(&dev->power.lock);
641 		}
642 		finish_wait(&dev->power.wait_queue, &wait);
643 		goto repeat;
644 	}
645 
646 	if (dev->power.no_callbacks)
647 		goto no_callback;	/* Assume success. */
648 
649 	/* Carry out an asynchronous or a synchronous suspend. */
650 	if (rpmflags & RPM_ASYNC) {
651 		dev->power.request = (rpmflags & RPM_AUTO) ?
652 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
653 		if (!dev->power.request_pending) {
654 			dev->power.request_pending = true;
655 			queue_work(pm_wq, &dev->power.work);
656 		}
657 		goto out;
658 	}
659 
660 	__update_runtime_status(dev, RPM_SUSPENDING);
661 
662 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
663 
664 	dev_pm_enable_wake_irq_check(dev, true);
665 	retval = rpm_callback(callback, dev);
666 	if (retval)
667 		goto fail;
668 
669 	dev_pm_enable_wake_irq_complete(dev);
670 
671  no_callback:
672 	__update_runtime_status(dev, RPM_SUSPENDED);
673 	pm_runtime_deactivate_timer(dev);
674 
675 	if (dev->parent) {
676 		parent = dev->parent;
677 		atomic_add_unless(&parent->power.child_count, -1, 0);
678 	}
679 	wake_up_all(&dev->power.wait_queue);
680 
681 	if (dev->power.deferred_resume) {
682 		dev->power.deferred_resume = false;
683 		rpm_resume(dev, 0);
684 		retval = -EAGAIN;
685 		goto out;
686 	}
687 
688 	if (dev->power.irq_safe)
689 		goto out;
690 
691 	/* Maybe the parent is now able to suspend. */
692 	if (parent && !parent->power.ignore_children) {
693 		spin_unlock(&dev->power.lock);
694 
695 		spin_lock(&parent->power.lock);
696 		rpm_idle(parent, RPM_ASYNC);
697 		spin_unlock(&parent->power.lock);
698 
699 		spin_lock(&dev->power.lock);
700 	}
701 	/* Maybe the suppliers are now able to suspend. */
702 	if (dev->power.links_count > 0) {
703 		spin_unlock_irq(&dev->power.lock);
704 
705 		rpm_suspend_suppliers(dev);
706 
707 		spin_lock_irq(&dev->power.lock);
708 	}
709 
710  out:
711 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
712 
713 	return retval;
714 
715  fail:
716 	dev_pm_disable_wake_irq_check(dev, true);
717 	__update_runtime_status(dev, RPM_ACTIVE);
718 	dev->power.deferred_resume = false;
719 	wake_up_all(&dev->power.wait_queue);
720 
721 	if (retval == -EAGAIN || retval == -EBUSY) {
722 		dev->power.runtime_error = 0;
723 
724 		/*
725 		 * If the callback routine failed an autosuspend, and
726 		 * if the last_busy time has been updated so that there
727 		 * is a new autosuspend expiration time, automatically
728 		 * reschedule another autosuspend.
729 		 */
730 		if ((rpmflags & RPM_AUTO) &&
731 		    pm_runtime_autosuspend_expiration(dev) != 0)
732 			goto repeat;
733 	} else {
734 		pm_runtime_cancel_pending(dev);
735 	}
736 	goto out;
737 }
738 
739 /**
740  * rpm_resume - Carry out runtime resume of given device.
741  * @dev: Device to resume.
742  * @rpmflags: Flag bits.
743  *
744  * Check if the device's runtime PM status allows it to be resumed.  Cancel
745  * any scheduled or pending requests.  If another resume has been started
746  * earlier, either return immediately or wait for it to finish, depending on the
747  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
748  * parallel with this function, either tell the other process to resume after
749  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
750  * flag is set then queue a resume request; otherwise run the
751  * ->runtime_resume() callback directly.  Queue an idle notification for the
752  * device if the resume succeeded.
753  *
754  * This function must be called under dev->power.lock with interrupts disabled.
755  */
756 static int rpm_resume(struct device *dev, int rpmflags)
757 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
758 {
759 	int (*callback)(struct device *);
760 	struct device *parent = NULL;
761 	int retval = 0;
762 
763 	trace_rpm_resume_rcuidle(dev, rpmflags);
764 
765  repeat:
766 	if (dev->power.runtime_error) {
767 		retval = -EINVAL;
768 	} else if (dev->power.disable_depth > 0) {
769 		if (dev->power.runtime_status == RPM_ACTIVE &&
770 		    dev->power.last_status == RPM_ACTIVE)
771 			retval = 1;
772 		else
773 			retval = -EACCES;
774 	}
775 	if (retval)
776 		goto out;
777 
778 	/*
779 	 * Other scheduled or pending requests need to be canceled.  Small
780 	 * optimization: If an autosuspend timer is running, leave it running
781 	 * rather than cancelling it now only to restart it again in the near
782 	 * future.
783 	 */
784 	dev->power.request = RPM_REQ_NONE;
785 	if (!dev->power.timer_autosuspends)
786 		pm_runtime_deactivate_timer(dev);
787 
788 	if (dev->power.runtime_status == RPM_ACTIVE) {
789 		retval = 1;
790 		goto out;
791 	}
792 
793 	if (dev->power.runtime_status == RPM_RESUMING
794 	    || dev->power.runtime_status == RPM_SUSPENDING) {
795 		DEFINE_WAIT(wait);
796 
797 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
798 			if (dev->power.runtime_status == RPM_SUSPENDING)
799 				dev->power.deferred_resume = true;
800 			else
801 				retval = -EINPROGRESS;
802 			goto out;
803 		}
804 
805 		if (dev->power.irq_safe) {
806 			spin_unlock(&dev->power.lock);
807 
808 			cpu_relax();
809 
810 			spin_lock(&dev->power.lock);
811 			goto repeat;
812 		}
813 
814 		/* Wait for the operation carried out in parallel with us. */
815 		for (;;) {
816 			prepare_to_wait(&dev->power.wait_queue, &wait,
817 					TASK_UNINTERRUPTIBLE);
818 			if (dev->power.runtime_status != RPM_RESUMING
819 			    && dev->power.runtime_status != RPM_SUSPENDING)
820 				break;
821 
822 			spin_unlock_irq(&dev->power.lock);
823 
824 			schedule();
825 
826 			spin_lock_irq(&dev->power.lock);
827 		}
828 		finish_wait(&dev->power.wait_queue, &wait);
829 		goto repeat;
830 	}
831 
832 	/*
833 	 * See if we can skip waking up the parent.  This is safe only if
834 	 * power.no_callbacks is set, because otherwise we don't know whether
835 	 * the resume will actually succeed.
836 	 */
837 	if (dev->power.no_callbacks && !parent && dev->parent) {
838 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
839 		if (dev->parent->power.disable_depth > 0
840 		    || dev->parent->power.ignore_children
841 		    || dev->parent->power.runtime_status == RPM_ACTIVE) {
842 			atomic_inc(&dev->parent->power.child_count);
843 			spin_unlock(&dev->parent->power.lock);
844 			retval = 1;
845 			goto no_callback;	/* Assume success. */
846 		}
847 		spin_unlock(&dev->parent->power.lock);
848 	}
849 
850 	/* Carry out an asynchronous or a synchronous resume. */
851 	if (rpmflags & RPM_ASYNC) {
852 		dev->power.request = RPM_REQ_RESUME;
853 		if (!dev->power.request_pending) {
854 			dev->power.request_pending = true;
855 			queue_work(pm_wq, &dev->power.work);
856 		}
857 		retval = 0;
858 		goto out;
859 	}
860 
861 	if (!parent && dev->parent) {
862 		/*
863 		 * Increment the parent's usage counter and resume it if
864 		 * necessary.  Not needed if dev is irq-safe; then the
865 		 * parent is permanently resumed.
866 		 */
867 		parent = dev->parent;
868 		if (dev->power.irq_safe)
869 			goto skip_parent;
870 		spin_unlock(&dev->power.lock);
871 
872 		pm_runtime_get_noresume(parent);
873 
874 		spin_lock(&parent->power.lock);
875 		/*
876 		 * Resume the parent if it has runtime PM enabled and not been
877 		 * set to ignore its children.
878 		 */
879 		if (!parent->power.disable_depth
880 		    && !parent->power.ignore_children) {
881 			rpm_resume(parent, 0);
882 			if (parent->power.runtime_status != RPM_ACTIVE)
883 				retval = -EBUSY;
884 		}
885 		spin_unlock(&parent->power.lock);
886 
887 		spin_lock(&dev->power.lock);
888 		if (retval)
889 			goto out;
890 		goto repeat;
891 	}
892  skip_parent:
893 
894 	if (dev->power.no_callbacks)
895 		goto no_callback;	/* Assume success. */
896 
897 	__update_runtime_status(dev, RPM_RESUMING);
898 
899 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
900 
901 	dev_pm_disable_wake_irq_check(dev, false);
902 	retval = rpm_callback(callback, dev);
903 	if (retval) {
904 		__update_runtime_status(dev, RPM_SUSPENDED);
905 		pm_runtime_cancel_pending(dev);
906 		dev_pm_enable_wake_irq_check(dev, false);
907 	} else {
908  no_callback:
909 		__update_runtime_status(dev, RPM_ACTIVE);
910 		pm_runtime_mark_last_busy(dev);
911 		if (parent)
912 			atomic_inc(&parent->power.child_count);
913 	}
914 	wake_up_all(&dev->power.wait_queue);
915 
916 	if (retval >= 0)
917 		rpm_idle(dev, RPM_ASYNC);
918 
919  out:
920 	if (parent && !dev->power.irq_safe) {
921 		spin_unlock_irq(&dev->power.lock);
922 
923 		pm_runtime_put(parent);
924 
925 		spin_lock_irq(&dev->power.lock);
926 	}
927 
928 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
929 
930 	return retval;
931 }
932 
933 /**
934  * pm_runtime_work - Universal runtime PM work function.
935  * @work: Work structure used for scheduling the execution of this function.
936  *
937  * Use @work to get the device object the work is to be done for, determine what
938  * is to be done and execute the appropriate runtime PM function.
939  */
940 static void pm_runtime_work(struct work_struct *work)
941 {
942 	struct device *dev = container_of(work, struct device, power.work);
943 	enum rpm_request req;
944 
945 	spin_lock_irq(&dev->power.lock);
946 
947 	if (!dev->power.request_pending)
948 		goto out;
949 
950 	req = dev->power.request;
951 	dev->power.request = RPM_REQ_NONE;
952 	dev->power.request_pending = false;
953 
954 	switch (req) {
955 	case RPM_REQ_NONE:
956 		break;
957 	case RPM_REQ_IDLE:
958 		rpm_idle(dev, RPM_NOWAIT);
959 		break;
960 	case RPM_REQ_SUSPEND:
961 		rpm_suspend(dev, RPM_NOWAIT);
962 		break;
963 	case RPM_REQ_AUTOSUSPEND:
964 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
965 		break;
966 	case RPM_REQ_RESUME:
967 		rpm_resume(dev, RPM_NOWAIT);
968 		break;
969 	}
970 
971  out:
972 	spin_unlock_irq(&dev->power.lock);
973 }
974 
975 /**
976  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
977  * @timer: hrtimer used by pm_schedule_suspend().
978  *
979  * Check if the time is right and queue a suspend request.
980  */
981 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
982 {
983 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
984 	unsigned long flags;
985 	u64 expires;
986 
987 	spin_lock_irqsave(&dev->power.lock, flags);
988 
989 	expires = dev->power.timer_expires;
990 	/*
991 	 * If 'expires' is after the current time, we've been called
992 	 * too early.
993 	 */
994 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
995 		dev->power.timer_expires = 0;
996 		rpm_suspend(dev, dev->power.timer_autosuspends ?
997 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
998 	}
999 
1000 	spin_unlock_irqrestore(&dev->power.lock, flags);
1001 
1002 	return HRTIMER_NORESTART;
1003 }
1004 
1005 /**
1006  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
1007  * @dev: Device to suspend.
1008  * @delay: Time to wait before submitting a suspend request, in milliseconds.
1009  */
1010 int pm_schedule_suspend(struct device *dev, unsigned int delay)
1011 {
1012 	unsigned long flags;
1013 	u64 expires;
1014 	int retval;
1015 
1016 	spin_lock_irqsave(&dev->power.lock, flags);
1017 
1018 	if (!delay) {
1019 		retval = rpm_suspend(dev, RPM_ASYNC);
1020 		goto out;
1021 	}
1022 
1023 	retval = rpm_check_suspend_allowed(dev);
1024 	if (retval)
1025 		goto out;
1026 
1027 	/* Other scheduled or pending requests need to be canceled. */
1028 	pm_runtime_cancel_pending(dev);
1029 
1030 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
1031 	dev->power.timer_expires = expires;
1032 	dev->power.timer_autosuspends = 0;
1033 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
1034 
1035  out:
1036 	spin_unlock_irqrestore(&dev->power.lock, flags);
1037 
1038 	return retval;
1039 }
1040 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
1041 
1042 static int rpm_drop_usage_count(struct device *dev)
1043 {
1044 	int ret;
1045 
1046 	ret = atomic_sub_return(1, &dev->power.usage_count);
1047 	if (ret >= 0)
1048 		return ret;
1049 
1050 	/*
1051 	 * Because rpm_resume() does not check the usage counter, it will resume
1052 	 * the device even if the usage counter is 0 or negative, so it is
1053 	 * sufficient to increment the usage counter here to reverse the change
1054 	 * made above.
1055 	 */
1056 	atomic_inc(&dev->power.usage_count);
1057 	dev_warn(dev, "Runtime PM usage count underflow!\n");
1058 	return -EINVAL;
1059 }
1060 
1061 /**
1062  * __pm_runtime_idle - Entry point for runtime idle operations.
1063  * @dev: Device to send idle notification for.
1064  * @rpmflags: Flag bits.
1065  *
1066  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1067  * return immediately if it is larger than zero (if it becomes negative, log a
1068  * warning, increment it, and return an error).  Then carry out an idle
1069  * notification, either synchronous or asynchronous.
1070  *
1071  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1072  * or if pm_runtime_irq_safe() has been called.
1073  */
1074 int __pm_runtime_idle(struct device *dev, int rpmflags)
1075 {
1076 	unsigned long flags;
1077 	int retval;
1078 
1079 	if (rpmflags & RPM_GET_PUT) {
1080 		retval = rpm_drop_usage_count(dev);
1081 		if (retval < 0) {
1082 			return retval;
1083 		} else if (retval > 0) {
1084 			trace_rpm_usage_rcuidle(dev, rpmflags);
1085 			return 0;
1086 		}
1087 	}
1088 
1089 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1090 
1091 	spin_lock_irqsave(&dev->power.lock, flags);
1092 	retval = rpm_idle(dev, rpmflags);
1093 	spin_unlock_irqrestore(&dev->power.lock, flags);
1094 
1095 	return retval;
1096 }
1097 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1098 
1099 /**
1100  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1101  * @dev: Device to suspend.
1102  * @rpmflags: Flag bits.
1103  *
1104  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1105  * return immediately if it is larger than zero (if it becomes negative, log a
1106  * warning, increment it, and return an error).  Then carry out a suspend,
1107  * either synchronous or asynchronous.
1108  *
1109  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1110  * or if pm_runtime_irq_safe() has been called.
1111  */
1112 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1113 {
1114 	unsigned long flags;
1115 	int retval;
1116 
1117 	if (rpmflags & RPM_GET_PUT) {
1118 		retval = rpm_drop_usage_count(dev);
1119 		if (retval < 0) {
1120 			return retval;
1121 		} else if (retval > 0) {
1122 			trace_rpm_usage_rcuidle(dev, rpmflags);
1123 			return 0;
1124 		}
1125 	}
1126 
1127 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1128 
1129 	spin_lock_irqsave(&dev->power.lock, flags);
1130 	retval = rpm_suspend(dev, rpmflags);
1131 	spin_unlock_irqrestore(&dev->power.lock, flags);
1132 
1133 	return retval;
1134 }
1135 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1136 
1137 /**
1138  * __pm_runtime_resume - Entry point for runtime resume operations.
1139  * @dev: Device to resume.
1140  * @rpmflags: Flag bits.
1141  *
1142  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1143  * carry out a resume, either synchronous or asynchronous.
1144  *
1145  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1146  * or if pm_runtime_irq_safe() has been called.
1147  */
1148 int __pm_runtime_resume(struct device *dev, int rpmflags)
1149 {
1150 	unsigned long flags;
1151 	int retval;
1152 
1153 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1154 			dev->power.runtime_status != RPM_ACTIVE);
1155 
1156 	if (rpmflags & RPM_GET_PUT)
1157 		atomic_inc(&dev->power.usage_count);
1158 
1159 	spin_lock_irqsave(&dev->power.lock, flags);
1160 	retval = rpm_resume(dev, rpmflags);
1161 	spin_unlock_irqrestore(&dev->power.lock, flags);
1162 
1163 	return retval;
1164 }
1165 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1166 
1167 /**
1168  * pm_runtime_get_if_active - Conditionally bump up device usage counter.
1169  * @dev: Device to handle.
1170  * @ign_usage_count: Whether or not to look at the current usage counter value.
1171  *
1172  * Return -EINVAL if runtime PM is disabled for @dev.
1173  *
1174  * Otherwise, if the runtime PM status of @dev is %RPM_ACTIVE and either
1175  * @ign_usage_count is %true or the runtime PM usage counter of @dev is not
1176  * zero, increment the usage counter of @dev and return 1. Otherwise, return 0
1177  * without changing the usage counter.
1178  *
1179  * If @ign_usage_count is %true, this function can be used to prevent suspending
1180  * the device when its runtime PM status is %RPM_ACTIVE.
1181  *
1182  * If @ign_usage_count is %false, this function can be used to prevent
1183  * suspending the device when both its runtime PM status is %RPM_ACTIVE and its
1184  * runtime PM usage counter is not zero.
1185  *
1186  * The caller is responsible for decrementing the runtime PM usage counter of
1187  * @dev after this function has returned a positive value for it.
1188  */
1189 int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count)
1190 {
1191 	unsigned long flags;
1192 	int retval;
1193 
1194 	spin_lock_irqsave(&dev->power.lock, flags);
1195 	if (dev->power.disable_depth > 0) {
1196 		retval = -EINVAL;
1197 	} else if (dev->power.runtime_status != RPM_ACTIVE) {
1198 		retval = 0;
1199 	} else if (ign_usage_count) {
1200 		retval = 1;
1201 		atomic_inc(&dev->power.usage_count);
1202 	} else {
1203 		retval = atomic_inc_not_zero(&dev->power.usage_count);
1204 	}
1205 	trace_rpm_usage_rcuidle(dev, 0);
1206 	spin_unlock_irqrestore(&dev->power.lock, flags);
1207 
1208 	return retval;
1209 }
1210 EXPORT_SYMBOL_GPL(pm_runtime_get_if_active);
1211 
1212 /**
1213  * __pm_runtime_set_status - Set runtime PM status of a device.
1214  * @dev: Device to handle.
1215  * @status: New runtime PM status of the device.
1216  *
1217  * If runtime PM of the device is disabled or its power.runtime_error field is
1218  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1219  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1220  * However, if the device has a parent and the parent is not active, and the
1221  * parent's power.ignore_children flag is unset, the device's status cannot be
1222  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1223  *
1224  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1225  * and the device parent's counter of unsuspended children is modified to
1226  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1227  * notification request for the parent is submitted.
1228  *
1229  * If @dev has any suppliers (as reflected by device links to them), and @status
1230  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1231  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1232  * of the @status value) and the suppliers will be deacticated on exit.  The
1233  * error returned by the failing supplier activation will be returned in that
1234  * case.
1235  */
1236 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1237 {
1238 	struct device *parent = dev->parent;
1239 	bool notify_parent = false;
1240 	unsigned long flags;
1241 	int error = 0;
1242 
1243 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1244 		return -EINVAL;
1245 
1246 	spin_lock_irqsave(&dev->power.lock, flags);
1247 
1248 	/*
1249 	 * Prevent PM-runtime from being enabled for the device or return an
1250 	 * error if it is enabled already and working.
1251 	 */
1252 	if (dev->power.runtime_error || dev->power.disable_depth)
1253 		dev->power.disable_depth++;
1254 	else
1255 		error = -EAGAIN;
1256 
1257 	spin_unlock_irqrestore(&dev->power.lock, flags);
1258 
1259 	if (error)
1260 		return error;
1261 
1262 	/*
1263 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1264 	 * upfront regardless of the current status, because next time
1265 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1266 	 * involved will be dropped down to one anyway.
1267 	 */
1268 	if (status == RPM_ACTIVE) {
1269 		int idx = device_links_read_lock();
1270 
1271 		error = rpm_get_suppliers(dev);
1272 		if (error)
1273 			status = RPM_SUSPENDED;
1274 
1275 		device_links_read_unlock(idx);
1276 	}
1277 
1278 	spin_lock_irqsave(&dev->power.lock, flags);
1279 
1280 	if (dev->power.runtime_status == status || !parent)
1281 		goto out_set;
1282 
1283 	if (status == RPM_SUSPENDED) {
1284 		atomic_add_unless(&parent->power.child_count, -1, 0);
1285 		notify_parent = !parent->power.ignore_children;
1286 	} else {
1287 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1288 
1289 		/*
1290 		 * It is invalid to put an active child under a parent that is
1291 		 * not active, has runtime PM enabled and the
1292 		 * 'power.ignore_children' flag unset.
1293 		 */
1294 		if (!parent->power.disable_depth
1295 		    && !parent->power.ignore_children
1296 		    && parent->power.runtime_status != RPM_ACTIVE) {
1297 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1298 				dev_name(dev),
1299 				dev_name(parent));
1300 			error = -EBUSY;
1301 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1302 			atomic_inc(&parent->power.child_count);
1303 		}
1304 
1305 		spin_unlock(&parent->power.lock);
1306 
1307 		if (error) {
1308 			status = RPM_SUSPENDED;
1309 			goto out;
1310 		}
1311 	}
1312 
1313  out_set:
1314 	__update_runtime_status(dev, status);
1315 	if (!error)
1316 		dev->power.runtime_error = 0;
1317 
1318  out:
1319 	spin_unlock_irqrestore(&dev->power.lock, flags);
1320 
1321 	if (notify_parent)
1322 		pm_request_idle(parent);
1323 
1324 	if (status == RPM_SUSPENDED) {
1325 		int idx = device_links_read_lock();
1326 
1327 		rpm_put_suppliers(dev);
1328 
1329 		device_links_read_unlock(idx);
1330 	}
1331 
1332 	pm_runtime_enable(dev);
1333 
1334 	return error;
1335 }
1336 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1337 
1338 /**
1339  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1340  * @dev: Device to handle.
1341  *
1342  * Flush all pending requests for the device from pm_wq and wait for all
1343  * runtime PM operations involving the device in progress to complete.
1344  *
1345  * Should be called under dev->power.lock with interrupts disabled.
1346  */
1347 static void __pm_runtime_barrier(struct device *dev)
1348 {
1349 	pm_runtime_deactivate_timer(dev);
1350 
1351 	if (dev->power.request_pending) {
1352 		dev->power.request = RPM_REQ_NONE;
1353 		spin_unlock_irq(&dev->power.lock);
1354 
1355 		cancel_work_sync(&dev->power.work);
1356 
1357 		spin_lock_irq(&dev->power.lock);
1358 		dev->power.request_pending = false;
1359 	}
1360 
1361 	if (dev->power.runtime_status == RPM_SUSPENDING
1362 	    || dev->power.runtime_status == RPM_RESUMING
1363 	    || dev->power.idle_notification) {
1364 		DEFINE_WAIT(wait);
1365 
1366 		/* Suspend, wake-up or idle notification in progress. */
1367 		for (;;) {
1368 			prepare_to_wait(&dev->power.wait_queue, &wait,
1369 					TASK_UNINTERRUPTIBLE);
1370 			if (dev->power.runtime_status != RPM_SUSPENDING
1371 			    && dev->power.runtime_status != RPM_RESUMING
1372 			    && !dev->power.idle_notification)
1373 				break;
1374 			spin_unlock_irq(&dev->power.lock);
1375 
1376 			schedule();
1377 
1378 			spin_lock_irq(&dev->power.lock);
1379 		}
1380 		finish_wait(&dev->power.wait_queue, &wait);
1381 	}
1382 }
1383 
1384 /**
1385  * pm_runtime_barrier - Flush pending requests and wait for completions.
1386  * @dev: Device to handle.
1387  *
1388  * Prevent the device from being suspended by incrementing its usage counter and
1389  * if there's a pending resume request for the device, wake the device up.
1390  * Next, make sure that all pending requests for the device have been flushed
1391  * from pm_wq and wait for all runtime PM operations involving the device in
1392  * progress to complete.
1393  *
1394  * Return value:
1395  * 1, if there was a resume request pending and the device had to be woken up,
1396  * 0, otherwise
1397  */
1398 int pm_runtime_barrier(struct device *dev)
1399 {
1400 	int retval = 0;
1401 
1402 	pm_runtime_get_noresume(dev);
1403 	spin_lock_irq(&dev->power.lock);
1404 
1405 	if (dev->power.request_pending
1406 	    && dev->power.request == RPM_REQ_RESUME) {
1407 		rpm_resume(dev, 0);
1408 		retval = 1;
1409 	}
1410 
1411 	__pm_runtime_barrier(dev);
1412 
1413 	spin_unlock_irq(&dev->power.lock);
1414 	pm_runtime_put_noidle(dev);
1415 
1416 	return retval;
1417 }
1418 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1419 
1420 /**
1421  * __pm_runtime_disable - Disable runtime PM of a device.
1422  * @dev: Device to handle.
1423  * @check_resume: If set, check if there's a resume request for the device.
1424  *
1425  * Increment power.disable_depth for the device and if it was zero previously,
1426  * cancel all pending runtime PM requests for the device and wait for all
1427  * operations in progress to complete.  The device can be either active or
1428  * suspended after its runtime PM has been disabled.
1429  *
1430  * If @check_resume is set and there's a resume request pending when
1431  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1432  * function will wake up the device before disabling its runtime PM.
1433  */
1434 void __pm_runtime_disable(struct device *dev, bool check_resume)
1435 {
1436 	spin_lock_irq(&dev->power.lock);
1437 
1438 	if (dev->power.disable_depth > 0) {
1439 		dev->power.disable_depth++;
1440 		goto out;
1441 	}
1442 
1443 	/*
1444 	 * Wake up the device if there's a resume request pending, because that
1445 	 * means there probably is some I/O to process and disabling runtime PM
1446 	 * shouldn't prevent the device from processing the I/O.
1447 	 */
1448 	if (check_resume && dev->power.request_pending
1449 	    && dev->power.request == RPM_REQ_RESUME) {
1450 		/*
1451 		 * Prevent suspends and idle notifications from being carried
1452 		 * out after we have woken up the device.
1453 		 */
1454 		pm_runtime_get_noresume(dev);
1455 
1456 		rpm_resume(dev, 0);
1457 
1458 		pm_runtime_put_noidle(dev);
1459 	}
1460 
1461 	/* Update time accounting before disabling PM-runtime. */
1462 	update_pm_runtime_accounting(dev);
1463 
1464 	if (!dev->power.disable_depth++) {
1465 		__pm_runtime_barrier(dev);
1466 		dev->power.last_status = dev->power.runtime_status;
1467 	}
1468 
1469  out:
1470 	spin_unlock_irq(&dev->power.lock);
1471 }
1472 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1473 
1474 /**
1475  * pm_runtime_enable - Enable runtime PM of a device.
1476  * @dev: Device to handle.
1477  */
1478 void pm_runtime_enable(struct device *dev)
1479 {
1480 	unsigned long flags;
1481 
1482 	spin_lock_irqsave(&dev->power.lock, flags);
1483 
1484 	if (!dev->power.disable_depth) {
1485 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1486 		goto out;
1487 	}
1488 
1489 	if (--dev->power.disable_depth > 0)
1490 		goto out;
1491 
1492 	dev->power.last_status = RPM_INVALID;
1493 	dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1494 
1495 	if (dev->power.runtime_status == RPM_SUSPENDED &&
1496 	    !dev->power.ignore_children &&
1497 	    atomic_read(&dev->power.child_count) > 0)
1498 		dev_warn(dev, "Enabling runtime PM for inactive device with active children\n");
1499 
1500 out:
1501 	spin_unlock_irqrestore(&dev->power.lock, flags);
1502 }
1503 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1504 
1505 static void pm_runtime_disable_action(void *data)
1506 {
1507 	pm_runtime_dont_use_autosuspend(data);
1508 	pm_runtime_disable(data);
1509 }
1510 
1511 /**
1512  * devm_pm_runtime_enable - devres-enabled version of pm_runtime_enable.
1513  *
1514  * NOTE: this will also handle calling pm_runtime_dont_use_autosuspend() for
1515  * you at driver exit time if needed.
1516  *
1517  * @dev: Device to handle.
1518  */
1519 int devm_pm_runtime_enable(struct device *dev)
1520 {
1521 	pm_runtime_enable(dev);
1522 
1523 	return devm_add_action_or_reset(dev, pm_runtime_disable_action, dev);
1524 }
1525 EXPORT_SYMBOL_GPL(devm_pm_runtime_enable);
1526 
1527 /**
1528  * pm_runtime_forbid - Block runtime PM of a device.
1529  * @dev: Device to handle.
1530  *
1531  * Increase the device's usage count and clear its power.runtime_auto flag,
1532  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1533  * for it.
1534  */
1535 void pm_runtime_forbid(struct device *dev)
1536 {
1537 	spin_lock_irq(&dev->power.lock);
1538 	if (!dev->power.runtime_auto)
1539 		goto out;
1540 
1541 	dev->power.runtime_auto = false;
1542 	atomic_inc(&dev->power.usage_count);
1543 	rpm_resume(dev, 0);
1544 
1545  out:
1546 	spin_unlock_irq(&dev->power.lock);
1547 }
1548 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1549 
1550 /**
1551  * pm_runtime_allow - Unblock runtime PM of a device.
1552  * @dev: Device to handle.
1553  *
1554  * Decrease the device's usage count and set its power.runtime_auto flag.
1555  */
1556 void pm_runtime_allow(struct device *dev)
1557 {
1558 	int ret;
1559 
1560 	spin_lock_irq(&dev->power.lock);
1561 	if (dev->power.runtime_auto)
1562 		goto out;
1563 
1564 	dev->power.runtime_auto = true;
1565 	ret = rpm_drop_usage_count(dev);
1566 	if (ret == 0)
1567 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1568 	else if (ret > 0)
1569 		trace_rpm_usage_rcuidle(dev, RPM_AUTO | RPM_ASYNC);
1570 
1571  out:
1572 	spin_unlock_irq(&dev->power.lock);
1573 }
1574 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1575 
1576 /**
1577  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1578  * @dev: Device to handle.
1579  *
1580  * Set the power.no_callbacks flag, which tells the PM core that this
1581  * device is power-managed through its parent and has no runtime PM
1582  * callbacks of its own.  The runtime sysfs attributes will be removed.
1583  */
1584 void pm_runtime_no_callbacks(struct device *dev)
1585 {
1586 	spin_lock_irq(&dev->power.lock);
1587 	dev->power.no_callbacks = 1;
1588 	spin_unlock_irq(&dev->power.lock);
1589 	if (device_is_registered(dev))
1590 		rpm_sysfs_remove(dev);
1591 }
1592 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1593 
1594 /**
1595  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1596  * @dev: Device to handle
1597  *
1598  * Set the power.irq_safe flag, which tells the PM core that the
1599  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1600  * always be invoked with the spinlock held and interrupts disabled.  It also
1601  * causes the parent's usage counter to be permanently incremented, preventing
1602  * the parent from runtime suspending -- otherwise an irq-safe child might have
1603  * to wait for a non-irq-safe parent.
1604  */
1605 void pm_runtime_irq_safe(struct device *dev)
1606 {
1607 	if (dev->parent)
1608 		pm_runtime_get_sync(dev->parent);
1609 	spin_lock_irq(&dev->power.lock);
1610 	dev->power.irq_safe = 1;
1611 	spin_unlock_irq(&dev->power.lock);
1612 }
1613 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1614 
1615 /**
1616  * update_autosuspend - Handle a change to a device's autosuspend settings.
1617  * @dev: Device to handle.
1618  * @old_delay: The former autosuspend_delay value.
1619  * @old_use: The former use_autosuspend value.
1620  *
1621  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1622  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1623  *
1624  * This function must be called under dev->power.lock with interrupts disabled.
1625  */
1626 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1627 {
1628 	int delay = dev->power.autosuspend_delay;
1629 
1630 	/* Should runtime suspend be prevented now? */
1631 	if (dev->power.use_autosuspend && delay < 0) {
1632 
1633 		/* If it used to be allowed then prevent it. */
1634 		if (!old_use || old_delay >= 0) {
1635 			atomic_inc(&dev->power.usage_count);
1636 			rpm_resume(dev, 0);
1637 		} else {
1638 			trace_rpm_usage_rcuidle(dev, 0);
1639 		}
1640 	}
1641 
1642 	/* Runtime suspend should be allowed now. */
1643 	else {
1644 
1645 		/* If it used to be prevented then allow it. */
1646 		if (old_use && old_delay < 0)
1647 			atomic_dec(&dev->power.usage_count);
1648 
1649 		/* Maybe we can autosuspend now. */
1650 		rpm_idle(dev, RPM_AUTO);
1651 	}
1652 }
1653 
1654 /**
1655  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1656  * @dev: Device to handle.
1657  * @delay: Value of the new delay in milliseconds.
1658  *
1659  * Set the device's power.autosuspend_delay value.  If it changes to negative
1660  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1661  * changes the other way, allow runtime suspends.
1662  */
1663 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1664 {
1665 	int old_delay, old_use;
1666 
1667 	spin_lock_irq(&dev->power.lock);
1668 	old_delay = dev->power.autosuspend_delay;
1669 	old_use = dev->power.use_autosuspend;
1670 	dev->power.autosuspend_delay = delay;
1671 	update_autosuspend(dev, old_delay, old_use);
1672 	spin_unlock_irq(&dev->power.lock);
1673 }
1674 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1675 
1676 /**
1677  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1678  * @dev: Device to handle.
1679  * @use: New value for use_autosuspend.
1680  *
1681  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1682  * suspends as needed.
1683  */
1684 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1685 {
1686 	int old_delay, old_use;
1687 
1688 	spin_lock_irq(&dev->power.lock);
1689 	old_delay = dev->power.autosuspend_delay;
1690 	old_use = dev->power.use_autosuspend;
1691 	dev->power.use_autosuspend = use;
1692 	update_autosuspend(dev, old_delay, old_use);
1693 	spin_unlock_irq(&dev->power.lock);
1694 }
1695 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1696 
1697 /**
1698  * pm_runtime_init - Initialize runtime PM fields in given device object.
1699  * @dev: Device object to initialize.
1700  */
1701 void pm_runtime_init(struct device *dev)
1702 {
1703 	dev->power.runtime_status = RPM_SUSPENDED;
1704 	dev->power.last_status = RPM_INVALID;
1705 	dev->power.idle_notification = false;
1706 
1707 	dev->power.disable_depth = 1;
1708 	atomic_set(&dev->power.usage_count, 0);
1709 
1710 	dev->power.runtime_error = 0;
1711 
1712 	atomic_set(&dev->power.child_count, 0);
1713 	pm_suspend_ignore_children(dev, false);
1714 	dev->power.runtime_auto = true;
1715 
1716 	dev->power.request_pending = false;
1717 	dev->power.request = RPM_REQ_NONE;
1718 	dev->power.deferred_resume = false;
1719 	dev->power.needs_force_resume = 0;
1720 	INIT_WORK(&dev->power.work, pm_runtime_work);
1721 
1722 	dev->power.timer_expires = 0;
1723 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1724 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1725 
1726 	init_waitqueue_head(&dev->power.wait_queue);
1727 }
1728 
1729 /**
1730  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1731  * @dev: Device object to re-initialize.
1732  */
1733 void pm_runtime_reinit(struct device *dev)
1734 {
1735 	if (!pm_runtime_enabled(dev)) {
1736 		if (dev->power.runtime_status == RPM_ACTIVE)
1737 			pm_runtime_set_suspended(dev);
1738 		if (dev->power.irq_safe) {
1739 			spin_lock_irq(&dev->power.lock);
1740 			dev->power.irq_safe = 0;
1741 			spin_unlock_irq(&dev->power.lock);
1742 			if (dev->parent)
1743 				pm_runtime_put(dev->parent);
1744 		}
1745 	}
1746 }
1747 
1748 /**
1749  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1750  * @dev: Device object being removed from device hierarchy.
1751  */
1752 void pm_runtime_remove(struct device *dev)
1753 {
1754 	__pm_runtime_disable(dev, false);
1755 	pm_runtime_reinit(dev);
1756 }
1757 
1758 /**
1759  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1760  * @dev: Consumer device.
1761  */
1762 void pm_runtime_get_suppliers(struct device *dev)
1763 {
1764 	struct device_link *link;
1765 	int idx;
1766 
1767 	idx = device_links_read_lock();
1768 
1769 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1770 				device_links_read_lock_held())
1771 		if (link->flags & DL_FLAG_PM_RUNTIME) {
1772 			link->supplier_preactivated = true;
1773 			pm_runtime_get_sync(link->supplier);
1774 			refcount_inc(&link->rpm_active);
1775 		}
1776 
1777 	device_links_read_unlock(idx);
1778 }
1779 
1780 /**
1781  * pm_runtime_put_suppliers - Drop references to supplier devices.
1782  * @dev: Consumer device.
1783  */
1784 void pm_runtime_put_suppliers(struct device *dev)
1785 {
1786 	struct device_link *link;
1787 	int idx;
1788 
1789 	idx = device_links_read_lock();
1790 
1791 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1792 				device_links_read_lock_held())
1793 		if (link->supplier_preactivated) {
1794 			bool put;
1795 
1796 			link->supplier_preactivated = false;
1797 
1798 			spin_lock_irq(&dev->power.lock);
1799 
1800 			put = pm_runtime_status_suspended(dev) &&
1801 			      refcount_dec_not_one(&link->rpm_active);
1802 
1803 			spin_unlock_irq(&dev->power.lock);
1804 
1805 			if (put)
1806 				pm_runtime_put(link->supplier);
1807 		}
1808 
1809 	device_links_read_unlock(idx);
1810 }
1811 
1812 void pm_runtime_new_link(struct device *dev)
1813 {
1814 	spin_lock_irq(&dev->power.lock);
1815 	dev->power.links_count++;
1816 	spin_unlock_irq(&dev->power.lock);
1817 }
1818 
1819 static void pm_runtime_drop_link_count(struct device *dev)
1820 {
1821 	spin_lock_irq(&dev->power.lock);
1822 	WARN_ON(dev->power.links_count == 0);
1823 	dev->power.links_count--;
1824 	spin_unlock_irq(&dev->power.lock);
1825 }
1826 
1827 /**
1828  * pm_runtime_drop_link - Prepare for device link removal.
1829  * @link: Device link going away.
1830  *
1831  * Drop the link count of the consumer end of @link and decrement the supplier
1832  * device's runtime PM usage counter as many times as needed to drop all of the
1833  * PM runtime reference to it from the consumer.
1834  */
1835 void pm_runtime_drop_link(struct device_link *link)
1836 {
1837 	if (!(link->flags & DL_FLAG_PM_RUNTIME))
1838 		return;
1839 
1840 	pm_runtime_drop_link_count(link->consumer);
1841 	pm_runtime_release_supplier(link, true);
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 int pm_runtime_force_suspend(struct device *dev)
1869 {
1870 	int (*callback)(struct device *);
1871 	int ret;
1872 
1873 	pm_runtime_disable(dev);
1874 	if (pm_runtime_status_suspended(dev))
1875 		return 0;
1876 
1877 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1878 
1879 	ret = callback ? callback(dev) : 0;
1880 	if (ret)
1881 		goto err;
1882 
1883 	/*
1884 	 * If the device can stay in suspend after the system-wide transition
1885 	 * to the working state that will follow, drop the children counter of
1886 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1887 	 * function will be called again for it in the meantime.
1888 	 */
1889 	if (pm_runtime_need_not_resume(dev)) {
1890 		pm_runtime_set_suspended(dev);
1891 	} else {
1892 		__update_runtime_status(dev, RPM_SUSPENDED);
1893 		dev->power.needs_force_resume = 1;
1894 	}
1895 
1896 	return 0;
1897 
1898 err:
1899 	pm_runtime_enable(dev);
1900 	return ret;
1901 }
1902 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1903 
1904 /**
1905  * pm_runtime_force_resume - Force a device into resume state if needed.
1906  * @dev: Device to resume.
1907  *
1908  * Prior invoking this function we expect the user to have brought the device
1909  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1910  * those actions and bring the device into full power, if it is expected to be
1911  * used on system resume.  In the other case, we defer the resume to be managed
1912  * via runtime PM.
1913  *
1914  * Typically this function may be invoked from a system resume callback.
1915  */
1916 int pm_runtime_force_resume(struct device *dev)
1917 {
1918 	int (*callback)(struct device *);
1919 	int ret = 0;
1920 
1921 	if (!pm_runtime_status_suspended(dev) || !dev->power.needs_force_resume)
1922 		goto out;
1923 
1924 	/*
1925 	 * The value of the parent's children counter is correct already, so
1926 	 * just update the status of the device.
1927 	 */
1928 	__update_runtime_status(dev, RPM_ACTIVE);
1929 
1930 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1931 
1932 	ret = callback ? callback(dev) : 0;
1933 	if (ret) {
1934 		pm_runtime_set_suspended(dev);
1935 		goto out;
1936 	}
1937 
1938 	pm_runtime_mark_last_busy(dev);
1939 out:
1940 	dev->power.needs_force_resume = 0;
1941 	pm_runtime_enable(dev);
1942 	return ret;
1943 }
1944 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1945