xref: /linux/drivers/base/power/runtime.c (revision f728c17fc97aea7a33151d9ba64106291c62bb02)
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 <linux/rculist.h>
15 #include <trace/events/rpm.h>
16 
17 #include "../base.h"
18 #include "power.h"
19 
20 typedef int (*pm_callback_t)(struct device *);
21 
22 static pm_callback_t __rpm_get_callback(struct device *dev, size_t cb_offset)
23 {
24 	pm_callback_t cb;
25 	const struct dev_pm_ops *ops;
26 
27 	if (dev->pm_domain)
28 		ops = &dev->pm_domain->ops;
29 	else if (dev->type && dev->type->pm)
30 		ops = dev->type->pm;
31 	else if (dev->class && dev->class->pm)
32 		ops = dev->class->pm;
33 	else if (dev->bus && dev->bus->pm)
34 		ops = dev->bus->pm;
35 	else
36 		ops = NULL;
37 
38 	if (ops)
39 		cb = *(pm_callback_t *)((void *)ops + cb_offset);
40 	else
41 		cb = NULL;
42 
43 	if (!cb && dev->driver && dev->driver->pm)
44 		cb = *(pm_callback_t *)((void *)dev->driver->pm + cb_offset);
45 
46 	return cb;
47 }
48 
49 #define RPM_GET_CALLBACK(dev, callback) \
50 		__rpm_get_callback(dev, offsetof(struct dev_pm_ops, callback))
51 
52 static int rpm_resume(struct device *dev, int rpmflags);
53 static int rpm_suspend(struct device *dev, int rpmflags);
54 
55 /**
56  * update_pm_runtime_accounting - Update the time accounting of power states
57  * @dev: Device to update the accounting for
58  *
59  * In order to be able to have time accounting of the various power states
60  * (as used by programs such as PowerTOP to show the effectiveness of runtime
61  * PM), we need to track the time spent in each state.
62  * update_pm_runtime_accounting must be called each time before the
63  * runtime_status field is updated, to account the time in the old state
64  * correctly.
65  */
66 static void update_pm_runtime_accounting(struct device *dev)
67 {
68 	u64 now, last, delta;
69 
70 	if (dev->power.disable_depth > 0)
71 		return;
72 
73 	last = dev->power.accounting_timestamp;
74 
75 	now = ktime_get_mono_fast_ns();
76 	dev->power.accounting_timestamp = now;
77 
78 	/*
79 	 * Because ktime_get_mono_fast_ns() is not monotonic during
80 	 * timekeeping updates, ensure that 'now' is after the last saved
81 	 * timesptamp.
82 	 */
83 	if (now < last)
84 		return;
85 
86 	delta = now - last;
87 
88 	if (dev->power.runtime_status == RPM_SUSPENDED)
89 		dev->power.suspended_time += delta;
90 	else
91 		dev->power.active_time += delta;
92 }
93 
94 static void __update_runtime_status(struct device *dev, enum rpm_status status)
95 {
96 	update_pm_runtime_accounting(dev);
97 	dev->power.runtime_status = status;
98 }
99 
100 static u64 rpm_get_accounted_time(struct device *dev, bool suspended)
101 {
102 	u64 time;
103 	unsigned long flags;
104 
105 	spin_lock_irqsave(&dev->power.lock, flags);
106 
107 	update_pm_runtime_accounting(dev);
108 	time = suspended ? dev->power.suspended_time : dev->power.active_time;
109 
110 	spin_unlock_irqrestore(&dev->power.lock, flags);
111 
112 	return time;
113 }
114 
115 u64 pm_runtime_active_time(struct device *dev)
116 {
117 	return rpm_get_accounted_time(dev, false);
118 }
119 
120 u64 pm_runtime_suspended_time(struct device *dev)
121 {
122 	return rpm_get_accounted_time(dev, true);
123 }
124 EXPORT_SYMBOL_GPL(pm_runtime_suspended_time);
125 
126 /**
127  * pm_runtime_deactivate_timer - Deactivate given device's suspend timer.
128  * @dev: Device to handle.
129  */
130 static void pm_runtime_deactivate_timer(struct device *dev)
131 {
132 	if (dev->power.timer_expires > 0) {
133 		hrtimer_try_to_cancel(&dev->power.suspend_timer);
134 		dev->power.timer_expires = 0;
135 	}
136 }
137 
138 /**
139  * pm_runtime_cancel_pending - Deactivate suspend timer and cancel requests.
140  * @dev: Device to handle.
141  */
142 static void pm_runtime_cancel_pending(struct device *dev)
143 {
144 	pm_runtime_deactivate_timer(dev);
145 	/*
146 	 * In case there's a request pending, make sure its work function will
147 	 * return without doing anything.
148 	 */
149 	dev->power.request = RPM_REQ_NONE;
150 }
151 
152 /*
153  * pm_runtime_autosuspend_expiration - Get a device's autosuspend-delay expiration time.
154  * @dev: Device to handle.
155  *
156  * Compute the autosuspend-delay expiration time based on the device's
157  * power.last_busy time.  If the delay has already expired or is disabled
158  * (negative) or the power.use_autosuspend flag isn't set, return 0.
159  * Otherwise return the expiration time in nanoseconds (adjusted to be nonzero).
160  *
161  * This function may be called either with or without dev->power.lock held.
162  * Either way it can be racy, since power.last_busy may be updated at any time.
163  */
164 u64 pm_runtime_autosuspend_expiration(struct device *dev)
165 {
166 	int autosuspend_delay;
167 	u64 expires;
168 
169 	if (!dev->power.use_autosuspend)
170 		return 0;
171 
172 	autosuspend_delay = READ_ONCE(dev->power.autosuspend_delay);
173 	if (autosuspend_delay < 0)
174 		return 0;
175 
176 	expires  = READ_ONCE(dev->power.last_busy);
177 	expires += (u64)autosuspend_delay * NSEC_PER_MSEC;
178 	if (expires > ktime_get_mono_fast_ns())
179 		return expires;	/* Expires in the future */
180 
181 	return 0;
182 }
183 EXPORT_SYMBOL_GPL(pm_runtime_autosuspend_expiration);
184 
185 static int dev_memalloc_noio(struct device *dev, void *data)
186 {
187 	return dev->power.memalloc_noio;
188 }
189 
190 /*
191  * pm_runtime_set_memalloc_noio - Set a device's memalloc_noio flag.
192  * @dev: Device to handle.
193  * @enable: True for setting the flag and False for clearing the flag.
194  *
195  * Set the flag for all devices in the path from the device to the
196  * root device in the device tree if @enable is true, otherwise clear
197  * the flag for devices in the path whose siblings don't set the flag.
198  *
199  * The function should only be called by block device, or network
200  * device driver for solving the deadlock problem during runtime
201  * resume/suspend:
202  *
203  *     If memory allocation with GFP_KERNEL is called inside runtime
204  *     resume/suspend callback of any one of its ancestors(or the
205  *     block device itself), the deadlock may be triggered inside the
206  *     memory allocation since it might not complete until the block
207  *     device becomes active and the involed page I/O finishes. The
208  *     situation is pointed out first by Alan Stern. Network device
209  *     are involved in iSCSI kind of situation.
210  *
211  * The lock of dev_hotplug_mutex is held in the function for handling
212  * hotplug race because pm_runtime_set_memalloc_noio() may be called
213  * in async probe().
214  *
215  * The function should be called between device_add() and device_del()
216  * on the affected device(block/network device).
217  */
218 void pm_runtime_set_memalloc_noio(struct device *dev, bool enable)
219 {
220 	static DEFINE_MUTEX(dev_hotplug_mutex);
221 
222 	mutex_lock(&dev_hotplug_mutex);
223 	for (;;) {
224 		bool enabled;
225 
226 		/* hold power lock since bitfield is not SMP-safe. */
227 		spin_lock_irq(&dev->power.lock);
228 		enabled = dev->power.memalloc_noio;
229 		dev->power.memalloc_noio = enable;
230 		spin_unlock_irq(&dev->power.lock);
231 
232 		/*
233 		 * not need to enable ancestors any more if the device
234 		 * has been enabled.
235 		 */
236 		if (enabled && enable)
237 			break;
238 
239 		dev = dev->parent;
240 
241 		/*
242 		 * clear flag of the parent device only if all the
243 		 * children don't set the flag because ancestor's
244 		 * flag was set by any one of the descendants.
245 		 */
246 		if (!dev || (!enable &&
247 		    device_for_each_child(dev, NULL, 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 && atomic_read(&dev->power.child_count))
269 		retval = -EBUSY;
270 
271 	/* Pending resume requests take precedence over suspends. */
272 	else if ((dev->power.deferred_resume &&
273 	    dev->power.runtime_status == RPM_SUSPENDING) ||
274 	    (dev->power.request_pending && dev->power.request == RPM_REQ_RESUME))
275 		retval = -EAGAIN;
276 	else if (__dev_pm_qos_resume_latency(dev) == 0)
277 		retval = -EPERM;
278 	else if (dev->power.runtime_status == RPM_SUSPENDED)
279 		retval = 1;
280 
281 	return retval;
282 }
283 
284 static int rpm_get_suppliers(struct device *dev)
285 {
286 	struct device_link *link;
287 
288 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
289 				device_links_read_lock_held()) {
290 		int retval;
291 
292 		if (!(link->flags & DL_FLAG_PM_RUNTIME))
293 			continue;
294 
295 		retval = pm_runtime_get_sync(link->supplier);
296 		/* Ignore suppliers with disabled runtime PM. */
297 		if (retval < 0 && retval != -EACCES) {
298 			pm_runtime_put_noidle(link->supplier);
299 			return retval;
300 		}
301 		refcount_inc(&link->rpm_active);
302 	}
303 	return 0;
304 }
305 
306 /**
307  * pm_runtime_release_supplier - Drop references to device link's supplier.
308  * @link: Target device link.
309  *
310  * Drop all runtime PM references associated with @link to its supplier device.
311  */
312 void pm_runtime_release_supplier(struct device_link *link)
313 {
314 	struct device *supplier = link->supplier;
315 
316 	/*
317 	 * The additional power.usage_count check is a safety net in case
318 	 * the rpm_active refcount becomes saturated, in which case
319 	 * refcount_dec_not_one() would return true forever, but it is not
320 	 * strictly necessary.
321 	 */
322 	while (refcount_dec_not_one(&link->rpm_active) &&
323 	       atomic_read(&supplier->power.usage_count) > 0)
324 		pm_runtime_put_noidle(supplier);
325 }
326 
327 static void __rpm_put_suppliers(struct device *dev, bool try_to_suspend)
328 {
329 	struct device_link *link;
330 
331 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
332 				device_links_read_lock_held()) {
333 		pm_runtime_release_supplier(link);
334 		if (try_to_suspend)
335 			pm_request_idle(link->supplier);
336 	}
337 }
338 
339 static void rpm_put_suppliers(struct device *dev)
340 {
341 	__rpm_put_suppliers(dev, true);
342 }
343 
344 static void rpm_suspend_suppliers(struct device *dev)
345 {
346 	struct device_link *link;
347 	int idx = device_links_read_lock();
348 
349 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
350 				device_links_read_lock_held())
351 		pm_request_idle(link->supplier);
352 
353 	device_links_read_unlock(idx);
354 }
355 
356 /**
357  * __rpm_callback - Run a given runtime PM callback for a given device.
358  * @cb: Runtime PM callback to run.
359  * @dev: Device to run the callback for.
360  */
361 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
362 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
363 {
364 	int retval = 0, idx;
365 	bool use_links = dev->power.links_count > 0;
366 
367 	if (dev->power.irq_safe) {
368 		spin_unlock(&dev->power.lock);
369 	} else {
370 		spin_unlock_irq(&dev->power.lock);
371 
372 		/*
373 		 * Resume suppliers if necessary.
374 		 *
375 		 * The device's runtime PM status cannot change until this
376 		 * routine returns, so it is safe to read the status outside of
377 		 * the lock.
378 		 */
379 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
380 			idx = device_links_read_lock();
381 
382 			retval = rpm_get_suppliers(dev);
383 			if (retval) {
384 				rpm_put_suppliers(dev);
385 				goto fail;
386 			}
387 
388 			device_links_read_unlock(idx);
389 		}
390 	}
391 
392 	if (cb)
393 		retval = cb(dev);
394 
395 	if (dev->power.irq_safe) {
396 		spin_lock(&dev->power.lock);
397 	} else {
398 		/*
399 		 * If the device is suspending and the callback has returned
400 		 * success, drop the usage counters of the suppliers that have
401 		 * been reference counted on its resume.
402 		 *
403 		 * Do that if resume fails too.
404 		 */
405 		if (use_links &&
406 		    ((dev->power.runtime_status == RPM_SUSPENDING && !retval) ||
407 		    (dev->power.runtime_status == RPM_RESUMING && retval))) {
408 			idx = device_links_read_lock();
409 
410 			__rpm_put_suppliers(dev, false);
411 
412 fail:
413 			device_links_read_unlock(idx);
414 		}
415 
416 		spin_lock_irq(&dev->power.lock);
417 	}
418 
419 	return retval;
420 }
421 
422 /**
423  * rpm_callback - Run a given runtime PM callback for a given device.
424  * @cb: Runtime PM callback to run.
425  * @dev: Device to run the callback for.
426  */
427 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
428 {
429 	int retval;
430 
431 	if (dev->power.memalloc_noio) {
432 		unsigned int noio_flag;
433 
434 		/*
435 		 * Deadlock might be caused if memory allocation with
436 		 * GFP_KERNEL happens inside runtime_suspend and
437 		 * runtime_resume callbacks of one block device's
438 		 * ancestor or the block device itself. Network
439 		 * device might be thought as part of iSCSI block
440 		 * device, so network device and its ancestor should
441 		 * be marked as memalloc_noio too.
442 		 */
443 		noio_flag = memalloc_noio_save();
444 		retval = __rpm_callback(cb, dev);
445 		memalloc_noio_restore(noio_flag);
446 	} else {
447 		retval = __rpm_callback(cb, dev);
448 	}
449 
450 	dev->power.runtime_error = retval;
451 	return retval != -EACCES ? retval : -EIO;
452 }
453 
454 /**
455  * rpm_idle - Notify device bus type if the device can be suspended.
456  * @dev: Device to notify the bus type about.
457  * @rpmflags: Flag bits.
458  *
459  * Check if the device's runtime PM status allows it to be suspended.  If
460  * another idle notification has been started earlier, return immediately.  If
461  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
462  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
463  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
464  *
465  * This function must be called under dev->power.lock with interrupts disabled.
466  */
467 static int rpm_idle(struct device *dev, int rpmflags)
468 {
469 	int (*callback)(struct device *);
470 	int retval;
471 
472 	trace_rpm_idle(dev, rpmflags);
473 	retval = rpm_check_suspend_allowed(dev);
474 	if (retval < 0)
475 		;	/* Conditions are wrong. */
476 
477 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
478 	else if (dev->power.runtime_status != RPM_ACTIVE)
479 		retval = -EAGAIN;
480 
481 	/*
482 	 * Any pending request other than an idle notification takes
483 	 * precedence over us, except that the timer may be running.
484 	 */
485 	else if (dev->power.request_pending &&
486 	    dev->power.request > RPM_REQ_IDLE)
487 		retval = -EAGAIN;
488 
489 	/* Act as though RPM_NOWAIT is always set. */
490 	else if (dev->power.idle_notification)
491 		retval = -EINPROGRESS;
492 
493 	if (retval)
494 		goto out;
495 
496 	/* Pending requests need to be canceled. */
497 	dev->power.request = RPM_REQ_NONE;
498 
499 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
500 
501 	/* If no callback assume success. */
502 	if (!callback || dev->power.no_callbacks)
503 		goto out;
504 
505 	/* Carry out an asynchronous or a synchronous idle notification. */
506 	if (rpmflags & RPM_ASYNC) {
507 		dev->power.request = RPM_REQ_IDLE;
508 		if (!dev->power.request_pending) {
509 			dev->power.request_pending = true;
510 			queue_work(pm_wq, &dev->power.work);
511 		}
512 		trace_rpm_return_int(dev, _THIS_IP_, 0);
513 		return 0;
514 	}
515 
516 	dev->power.idle_notification = true;
517 
518 	if (dev->power.irq_safe)
519 		spin_unlock(&dev->power.lock);
520 	else
521 		spin_unlock_irq(&dev->power.lock);
522 
523 	retval = callback(dev);
524 
525 	if (dev->power.irq_safe)
526 		spin_lock(&dev->power.lock);
527 	else
528 		spin_lock_irq(&dev->power.lock);
529 
530 	dev->power.idle_notification = false;
531 	wake_up_all(&dev->power.wait_queue);
532 
533  out:
534 	trace_rpm_return_int(dev, _THIS_IP_, retval);
535 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
536 }
537 
538 /**
539  * rpm_suspend - Carry out runtime suspend of given device.
540  * @dev: Device to suspend.
541  * @rpmflags: Flag bits.
542  *
543  * Check if the device's runtime PM status allows it to be suspended.
544  * Cancel a pending idle notification, autosuspend or suspend. If
545  * another suspend has been started earlier, either return immediately
546  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
547  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
548  * otherwise run the ->runtime_suspend() callback directly. When
549  * ->runtime_suspend succeeded, if a deferred resume was requested while
550  * the callback was running then carry it out, otherwise send an idle
551  * notification for its parent (if the suspend succeeded and both
552  * ignore_children of parent->power and irq_safe of dev->power are not set).
553  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
554  * flag is set and the next autosuspend-delay expiration time is in the
555  * future, schedule another autosuspend attempt.
556  *
557  * This function must be called under dev->power.lock with interrupts disabled.
558  */
559 static int rpm_suspend(struct device *dev, int rpmflags)
560 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
561 {
562 	int (*callback)(struct device *);
563 	struct device *parent = NULL;
564 	int retval;
565 
566 	trace_rpm_suspend(dev, rpmflags);
567 
568  repeat:
569 	retval = rpm_check_suspend_allowed(dev);
570 	if (retval < 0)
571 		goto out;	/* Conditions are wrong. */
572 
573 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
574 	if (dev->power.runtime_status == RPM_RESUMING && !(rpmflags & RPM_ASYNC))
575 		retval = -EAGAIN;
576 
577 	if (retval)
578 		goto out;
579 
580 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
581 	if ((rpmflags & RPM_AUTO) && dev->power.runtime_status != RPM_SUSPENDING) {
582 		u64 expires = pm_runtime_autosuspend_expiration(dev);
583 
584 		if (expires != 0) {
585 			/* Pending requests need to be canceled. */
586 			dev->power.request = RPM_REQ_NONE;
587 
588 			/*
589 			 * Optimization: If the timer is already running and is
590 			 * set to expire at or before the autosuspend delay,
591 			 * avoid the overhead of resetting it.  Just let it
592 			 * expire; pm_suspend_timer_fn() will take care of the
593 			 * rest.
594 			 */
595 			if (!(dev->power.timer_expires &&
596 			    dev->power.timer_expires <= expires)) {
597 				/*
598 				 * We add a slack of 25% to gather wakeups
599 				 * without sacrificing the granularity.
600 				 */
601 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
602 						    (NSEC_PER_MSEC >> 2);
603 
604 				dev->power.timer_expires = expires;
605 				hrtimer_start_range_ns(&dev->power.suspend_timer,
606 						       ns_to_ktime(expires),
607 						       slack,
608 						       HRTIMER_MODE_ABS);
609 			}
610 			dev->power.timer_autosuspends = 1;
611 			goto out;
612 		}
613 	}
614 
615 	/* Other scheduled or pending requests need to be canceled. */
616 	pm_runtime_cancel_pending(dev);
617 
618 	if (dev->power.runtime_status == RPM_SUSPENDING) {
619 		DEFINE_WAIT(wait);
620 
621 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
622 			retval = -EINPROGRESS;
623 			goto out;
624 		}
625 
626 		if (dev->power.irq_safe) {
627 			spin_unlock(&dev->power.lock);
628 
629 			cpu_relax();
630 
631 			spin_lock(&dev->power.lock);
632 			goto repeat;
633 		}
634 
635 		/* Wait for the other suspend running in parallel with us. */
636 		for (;;) {
637 			prepare_to_wait(&dev->power.wait_queue, &wait,
638 					TASK_UNINTERRUPTIBLE);
639 			if (dev->power.runtime_status != RPM_SUSPENDING)
640 				break;
641 
642 			spin_unlock_irq(&dev->power.lock);
643 
644 			schedule();
645 
646 			spin_lock_irq(&dev->power.lock);
647 		}
648 		finish_wait(&dev->power.wait_queue, &wait);
649 		goto repeat;
650 	}
651 
652 	if (dev->power.no_callbacks)
653 		goto no_callback;	/* Assume success. */
654 
655 	/* Carry out an asynchronous or a synchronous suspend. */
656 	if (rpmflags & RPM_ASYNC) {
657 		dev->power.request = (rpmflags & RPM_AUTO) ?
658 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
659 		if (!dev->power.request_pending) {
660 			dev->power.request_pending = true;
661 			queue_work(pm_wq, &dev->power.work);
662 		}
663 		goto out;
664 	}
665 
666 	__update_runtime_status(dev, RPM_SUSPENDING);
667 
668 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
669 
670 	dev_pm_enable_wake_irq_check(dev, true);
671 	retval = rpm_callback(callback, dev);
672 	if (retval)
673 		goto fail;
674 
675 	dev_pm_enable_wake_irq_complete(dev);
676 
677  no_callback:
678 	__update_runtime_status(dev, RPM_SUSPENDED);
679 	pm_runtime_deactivate_timer(dev);
680 
681 	if (dev->parent) {
682 		parent = dev->parent;
683 		atomic_add_unless(&parent->power.child_count, -1, 0);
684 	}
685 	wake_up_all(&dev->power.wait_queue);
686 
687 	if (dev->power.deferred_resume) {
688 		dev->power.deferred_resume = false;
689 		rpm_resume(dev, 0);
690 		retval = -EAGAIN;
691 		goto out;
692 	}
693 
694 	if (dev->power.irq_safe)
695 		goto out;
696 
697 	/* Maybe the parent is now able to suspend. */
698 	if (parent && !parent->power.ignore_children) {
699 		spin_unlock(&dev->power.lock);
700 
701 		spin_lock(&parent->power.lock);
702 		rpm_idle(parent, RPM_ASYNC);
703 		spin_unlock(&parent->power.lock);
704 
705 		spin_lock(&dev->power.lock);
706 	}
707 	/* Maybe the suppliers are now able to suspend. */
708 	if (dev->power.links_count > 0) {
709 		spin_unlock_irq(&dev->power.lock);
710 
711 		rpm_suspend_suppliers(dev);
712 
713 		spin_lock_irq(&dev->power.lock);
714 	}
715 
716  out:
717 	trace_rpm_return_int(dev, _THIS_IP_, retval);
718 
719 	return retval;
720 
721  fail:
722 	dev_pm_disable_wake_irq_check(dev, true);
723 	__update_runtime_status(dev, RPM_ACTIVE);
724 	dev->power.deferred_resume = false;
725 	wake_up_all(&dev->power.wait_queue);
726 
727 	if (retval == -EAGAIN || retval == -EBUSY) {
728 		dev->power.runtime_error = 0;
729 
730 		/*
731 		 * If the callback routine failed an autosuspend, and
732 		 * if the last_busy time has been updated so that there
733 		 * is a new autosuspend expiration time, automatically
734 		 * reschedule another autosuspend.
735 		 */
736 		if ((rpmflags & RPM_AUTO) &&
737 		    pm_runtime_autosuspend_expiration(dev) != 0)
738 			goto repeat;
739 	} else {
740 		pm_runtime_cancel_pending(dev);
741 	}
742 	goto out;
743 }
744 
745 /**
746  * rpm_resume - Carry out runtime resume of given device.
747  * @dev: Device to resume.
748  * @rpmflags: Flag bits.
749  *
750  * Check if the device's runtime PM status allows it to be resumed.  Cancel
751  * any scheduled or pending requests.  If another resume has been started
752  * earlier, either return immediately or wait for it to finish, depending on the
753  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
754  * parallel with this function, either tell the other process to resume after
755  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
756  * flag is set then queue a resume request; otherwise run the
757  * ->runtime_resume() callback directly.  Queue an idle notification for the
758  * device if the resume succeeded.
759  *
760  * This function must be called under dev->power.lock with interrupts disabled.
761  */
762 static int rpm_resume(struct device *dev, int rpmflags)
763 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
764 {
765 	int (*callback)(struct device *);
766 	struct device *parent = NULL;
767 	int retval = 0;
768 
769 	trace_rpm_resume(dev, rpmflags);
770 
771  repeat:
772 	if (dev->power.runtime_error) {
773 		retval = -EINVAL;
774 	} else if (dev->power.disable_depth > 0) {
775 		if (dev->power.runtime_status == RPM_ACTIVE &&
776 		    dev->power.last_status == RPM_ACTIVE)
777 			retval = 1;
778 		else
779 			retval = -EACCES;
780 	}
781 	if (retval)
782 		goto out;
783 
784 	/*
785 	 * Other scheduled or pending requests need to be canceled.  Small
786 	 * optimization: If an autosuspend timer is running, leave it running
787 	 * rather than cancelling it now only to restart it again in the near
788 	 * future.
789 	 */
790 	dev->power.request = RPM_REQ_NONE;
791 	if (!dev->power.timer_autosuspends)
792 		pm_runtime_deactivate_timer(dev);
793 
794 	if (dev->power.runtime_status == RPM_ACTIVE) {
795 		retval = 1;
796 		goto out;
797 	}
798 
799 	if (dev->power.runtime_status == RPM_RESUMING ||
800 	    dev->power.runtime_status == RPM_SUSPENDING) {
801 		DEFINE_WAIT(wait);
802 
803 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
804 			if (dev->power.runtime_status == RPM_SUSPENDING) {
805 				dev->power.deferred_resume = true;
806 				if (rpmflags & RPM_NOWAIT)
807 					retval = -EINPROGRESS;
808 			} else {
809 				retval = -EINPROGRESS;
810 			}
811 			goto out;
812 		}
813 
814 		if (dev->power.irq_safe) {
815 			spin_unlock(&dev->power.lock);
816 
817 			cpu_relax();
818 
819 			spin_lock(&dev->power.lock);
820 			goto repeat;
821 		}
822 
823 		/* Wait for the operation carried out in parallel with us. */
824 		for (;;) {
825 			prepare_to_wait(&dev->power.wait_queue, &wait,
826 					TASK_UNINTERRUPTIBLE);
827 			if (dev->power.runtime_status != RPM_RESUMING &&
828 			    dev->power.runtime_status != RPM_SUSPENDING)
829 				break;
830 
831 			spin_unlock_irq(&dev->power.lock);
832 
833 			schedule();
834 
835 			spin_lock_irq(&dev->power.lock);
836 		}
837 		finish_wait(&dev->power.wait_queue, &wait);
838 		goto repeat;
839 	}
840 
841 	/*
842 	 * See if we can skip waking up the parent.  This is safe only if
843 	 * power.no_callbacks is set, because otherwise we don't know whether
844 	 * the resume will actually succeed.
845 	 */
846 	if (dev->power.no_callbacks && !parent && dev->parent) {
847 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
848 		if (dev->parent->power.disable_depth > 0 ||
849 		    dev->parent->power.ignore_children ||
850 		    dev->parent->power.runtime_status == RPM_ACTIVE) {
851 			atomic_inc(&dev->parent->power.child_count);
852 			spin_unlock(&dev->parent->power.lock);
853 			retval = 1;
854 			goto no_callback;	/* Assume success. */
855 		}
856 		spin_unlock(&dev->parent->power.lock);
857 	}
858 
859 	/* Carry out an asynchronous or a synchronous resume. */
860 	if (rpmflags & RPM_ASYNC) {
861 		dev->power.request = RPM_REQ_RESUME;
862 		if (!dev->power.request_pending) {
863 			dev->power.request_pending = true;
864 			queue_work(pm_wq, &dev->power.work);
865 		}
866 		retval = 0;
867 		goto out;
868 	}
869 
870 	if (!parent && dev->parent) {
871 		/*
872 		 * Increment the parent's usage counter and resume it if
873 		 * necessary.  Not needed if dev is irq-safe; then the
874 		 * parent is permanently resumed.
875 		 */
876 		parent = dev->parent;
877 		if (dev->power.irq_safe)
878 			goto skip_parent;
879 
880 		spin_unlock(&dev->power.lock);
881 
882 		pm_runtime_get_noresume(parent);
883 
884 		spin_lock(&parent->power.lock);
885 		/*
886 		 * Resume the parent if it has runtime PM enabled and not been
887 		 * set to ignore its children.
888 		 */
889 		if (!parent->power.disable_depth &&
890 		    !parent->power.ignore_children) {
891 			rpm_resume(parent, 0);
892 			if (parent->power.runtime_status != RPM_ACTIVE)
893 				retval = -EBUSY;
894 		}
895 		spin_unlock(&parent->power.lock);
896 
897 		spin_lock(&dev->power.lock);
898 		if (retval)
899 			goto out;
900 
901 		goto repeat;
902 	}
903  skip_parent:
904 
905 	if (dev->power.no_callbacks)
906 		goto no_callback;	/* Assume success. */
907 
908 	__update_runtime_status(dev, RPM_RESUMING);
909 
910 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
911 
912 	dev_pm_disable_wake_irq_check(dev, false);
913 	retval = rpm_callback(callback, dev);
914 	if (retval) {
915 		__update_runtime_status(dev, RPM_SUSPENDED);
916 		pm_runtime_cancel_pending(dev);
917 		dev_pm_enable_wake_irq_check(dev, false);
918 	} else {
919  no_callback:
920 		__update_runtime_status(dev, RPM_ACTIVE);
921 		pm_runtime_mark_last_busy(dev);
922 		if (parent)
923 			atomic_inc(&parent->power.child_count);
924 	}
925 	wake_up_all(&dev->power.wait_queue);
926 
927 	if (retval >= 0)
928 		rpm_idle(dev, RPM_ASYNC);
929 
930  out:
931 	if (parent && !dev->power.irq_safe) {
932 		spin_unlock_irq(&dev->power.lock);
933 
934 		pm_runtime_put(parent);
935 
936 		spin_lock_irq(&dev->power.lock);
937 	}
938 
939 	trace_rpm_return_int(dev, _THIS_IP_, retval);
940 
941 	return retval;
942 }
943 
944 /**
945  * pm_runtime_work - Universal runtime PM work function.
946  * @work: Work structure used for scheduling the execution of this function.
947  *
948  * Use @work to get the device object the work is to be done for, determine what
949  * is to be done and execute the appropriate runtime PM function.
950  */
951 static void pm_runtime_work(struct work_struct *work)
952 {
953 	struct device *dev = container_of(work, struct device, power.work);
954 	enum rpm_request req;
955 
956 	spin_lock_irq(&dev->power.lock);
957 
958 	if (!dev->power.request_pending)
959 		goto out;
960 
961 	req = dev->power.request;
962 	dev->power.request = RPM_REQ_NONE;
963 	dev->power.request_pending = false;
964 
965 	switch (req) {
966 	case RPM_REQ_NONE:
967 		break;
968 	case RPM_REQ_IDLE:
969 		rpm_idle(dev, RPM_NOWAIT);
970 		break;
971 	case RPM_REQ_SUSPEND:
972 		rpm_suspend(dev, RPM_NOWAIT);
973 		break;
974 	case RPM_REQ_AUTOSUSPEND:
975 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
976 		break;
977 	case RPM_REQ_RESUME:
978 		rpm_resume(dev, RPM_NOWAIT);
979 		break;
980 	}
981 
982  out:
983 	spin_unlock_irq(&dev->power.lock);
984 }
985 
986 /**
987  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
988  * @timer: hrtimer used by pm_schedule_suspend().
989  *
990  * Check if the time is right and queue a suspend request.
991  */
992 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
993 {
994 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
995 	unsigned long flags;
996 	u64 expires;
997 
998 	spin_lock_irqsave(&dev->power.lock, flags);
999 
1000 	expires = dev->power.timer_expires;
1001 	/*
1002 	 * If 'expires' is after the current time, we've been called
1003 	 * too early.
1004 	 */
1005 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
1006 		dev->power.timer_expires = 0;
1007 		rpm_suspend(dev, dev->power.timer_autosuspends ?
1008 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
1009 	}
1010 
1011 	spin_unlock_irqrestore(&dev->power.lock, flags);
1012 
1013 	return HRTIMER_NORESTART;
1014 }
1015 
1016 /**
1017  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
1018  * @dev: Device to suspend.
1019  * @delay: Time to wait before submitting a suspend request, in milliseconds.
1020  */
1021 int pm_schedule_suspend(struct device *dev, unsigned int delay)
1022 {
1023 	unsigned long flags;
1024 	u64 expires;
1025 	int retval;
1026 
1027 	spin_lock_irqsave(&dev->power.lock, flags);
1028 
1029 	if (!delay) {
1030 		retval = rpm_suspend(dev, RPM_ASYNC);
1031 		goto out;
1032 	}
1033 
1034 	retval = rpm_check_suspend_allowed(dev);
1035 	if (retval)
1036 		goto out;
1037 
1038 	/* Other scheduled or pending requests need to be canceled. */
1039 	pm_runtime_cancel_pending(dev);
1040 
1041 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
1042 	dev->power.timer_expires = expires;
1043 	dev->power.timer_autosuspends = 0;
1044 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
1045 
1046  out:
1047 	spin_unlock_irqrestore(&dev->power.lock, flags);
1048 
1049 	return retval;
1050 }
1051 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
1052 
1053 static int rpm_drop_usage_count(struct device *dev)
1054 {
1055 	int ret;
1056 
1057 	ret = atomic_sub_return(1, &dev->power.usage_count);
1058 	if (ret >= 0)
1059 		return ret;
1060 
1061 	/*
1062 	 * Because rpm_resume() does not check the usage counter, it will resume
1063 	 * the device even if the usage counter is 0 or negative, so it is
1064 	 * sufficient to increment the usage counter here to reverse the change
1065 	 * made above.
1066 	 */
1067 	atomic_inc(&dev->power.usage_count);
1068 	dev_warn(dev, "Runtime PM usage count underflow!\n");
1069 	return -EINVAL;
1070 }
1071 
1072 /**
1073  * __pm_runtime_idle - Entry point for runtime idle operations.
1074  * @dev: Device to send idle notification for.
1075  * @rpmflags: Flag bits.
1076  *
1077  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1078  * return immediately if it is larger than zero (if it becomes negative, log a
1079  * warning, increment it, and return an error).  Then carry out an idle
1080  * notification, either synchronous or asynchronous.
1081  *
1082  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1083  * or if pm_runtime_irq_safe() has been called.
1084  */
1085 int __pm_runtime_idle(struct device *dev, int rpmflags)
1086 {
1087 	unsigned long flags;
1088 	int retval;
1089 
1090 	if (rpmflags & RPM_GET_PUT) {
1091 		retval = rpm_drop_usage_count(dev);
1092 		if (retval < 0) {
1093 			return retval;
1094 		} else if (retval > 0) {
1095 			trace_rpm_usage(dev, rpmflags);
1096 			return 0;
1097 		}
1098 	}
1099 
1100 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1101 
1102 	spin_lock_irqsave(&dev->power.lock, flags);
1103 	retval = rpm_idle(dev, rpmflags);
1104 	spin_unlock_irqrestore(&dev->power.lock, flags);
1105 
1106 	return retval;
1107 }
1108 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1109 
1110 /**
1111  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1112  * @dev: Device to suspend.
1113  * @rpmflags: Flag bits.
1114  *
1115  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1116  * return immediately if it is larger than zero (if it becomes negative, log a
1117  * warning, increment it, and return an error).  Then carry out a suspend,
1118  * either synchronous or asynchronous.
1119  *
1120  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1121  * or if pm_runtime_irq_safe() has been called.
1122  */
1123 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1124 {
1125 	unsigned long flags;
1126 	int retval;
1127 
1128 	if (rpmflags & RPM_GET_PUT) {
1129 		retval = rpm_drop_usage_count(dev);
1130 		if (retval < 0) {
1131 			return retval;
1132 		} else if (retval > 0) {
1133 			trace_rpm_usage(dev, rpmflags);
1134 			return 0;
1135 		}
1136 	}
1137 
1138 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1139 
1140 	spin_lock_irqsave(&dev->power.lock, flags);
1141 	retval = rpm_suspend(dev, rpmflags);
1142 	spin_unlock_irqrestore(&dev->power.lock, flags);
1143 
1144 	return retval;
1145 }
1146 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1147 
1148 /**
1149  * __pm_runtime_resume - Entry point for runtime resume operations.
1150  * @dev: Device to resume.
1151  * @rpmflags: Flag bits.
1152  *
1153  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1154  * carry out a resume, either synchronous or asynchronous.
1155  *
1156  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1157  * or if pm_runtime_irq_safe() has been called.
1158  */
1159 int __pm_runtime_resume(struct device *dev, int rpmflags)
1160 {
1161 	unsigned long flags;
1162 	int retval;
1163 
1164 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1165 			dev->power.runtime_status != RPM_ACTIVE);
1166 
1167 	if (rpmflags & RPM_GET_PUT)
1168 		atomic_inc(&dev->power.usage_count);
1169 
1170 	spin_lock_irqsave(&dev->power.lock, flags);
1171 	retval = rpm_resume(dev, rpmflags);
1172 	spin_unlock_irqrestore(&dev->power.lock, flags);
1173 
1174 	return retval;
1175 }
1176 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1177 
1178 /**
1179  * pm_runtime_get_if_active - Conditionally bump up device usage counter.
1180  * @dev: Device to handle.
1181  * @ign_usage_count: Whether or not to look at the current usage counter value.
1182  *
1183  * Return -EINVAL if runtime PM is disabled for @dev.
1184  *
1185  * Otherwise, if the runtime PM status of @dev is %RPM_ACTIVE and either
1186  * @ign_usage_count is %true or the runtime PM usage counter of @dev is not
1187  * zero, increment the usage counter of @dev and return 1. Otherwise, return 0
1188  * without changing the usage counter.
1189  *
1190  * If @ign_usage_count is %true, this function can be used to prevent suspending
1191  * the device when its runtime PM status is %RPM_ACTIVE.
1192  *
1193  * If @ign_usage_count is %false, this function can be used to prevent
1194  * suspending the device when both its runtime PM status is %RPM_ACTIVE and its
1195  * runtime PM usage counter is not zero.
1196  *
1197  * The caller is responsible for decrementing the runtime PM usage counter of
1198  * @dev after this function has returned a positive value for it.
1199  */
1200 int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count)
1201 {
1202 	unsigned long flags;
1203 	int retval;
1204 
1205 	spin_lock_irqsave(&dev->power.lock, flags);
1206 	if (dev->power.disable_depth > 0) {
1207 		retval = -EINVAL;
1208 	} else if (dev->power.runtime_status != RPM_ACTIVE) {
1209 		retval = 0;
1210 	} else if (ign_usage_count) {
1211 		retval = 1;
1212 		atomic_inc(&dev->power.usage_count);
1213 	} else {
1214 		retval = atomic_inc_not_zero(&dev->power.usage_count);
1215 	}
1216 	trace_rpm_usage(dev, 0);
1217 	spin_unlock_irqrestore(&dev->power.lock, flags);
1218 
1219 	return retval;
1220 }
1221 EXPORT_SYMBOL_GPL(pm_runtime_get_if_active);
1222 
1223 /**
1224  * __pm_runtime_set_status - Set runtime PM status of a device.
1225  * @dev: Device to handle.
1226  * @status: New runtime PM status of the device.
1227  *
1228  * If runtime PM of the device is disabled or its power.runtime_error field is
1229  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1230  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1231  * However, if the device has a parent and the parent is not active, and the
1232  * parent's power.ignore_children flag is unset, the device's status cannot be
1233  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1234  *
1235  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1236  * and the device parent's counter of unsuspended children is modified to
1237  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1238  * notification request for the parent is submitted.
1239  *
1240  * If @dev has any suppliers (as reflected by device links to them), and @status
1241  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1242  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1243  * of the @status value) and the suppliers will be deacticated on exit.  The
1244  * error returned by the failing supplier activation will be returned in that
1245  * case.
1246  */
1247 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1248 {
1249 	struct device *parent = dev->parent;
1250 	bool notify_parent = false;
1251 	unsigned long flags;
1252 	int error = 0;
1253 
1254 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1255 		return -EINVAL;
1256 
1257 	spin_lock_irqsave(&dev->power.lock, flags);
1258 
1259 	/*
1260 	 * Prevent PM-runtime from being enabled for the device or return an
1261 	 * error if it is enabled already and working.
1262 	 */
1263 	if (dev->power.runtime_error || dev->power.disable_depth)
1264 		dev->power.disable_depth++;
1265 	else
1266 		error = -EAGAIN;
1267 
1268 	spin_unlock_irqrestore(&dev->power.lock, flags);
1269 
1270 	if (error)
1271 		return error;
1272 
1273 	/*
1274 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1275 	 * upfront regardless of the current status, because next time
1276 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1277 	 * involved will be dropped down to one anyway.
1278 	 */
1279 	if (status == RPM_ACTIVE) {
1280 		int idx = device_links_read_lock();
1281 
1282 		error = rpm_get_suppliers(dev);
1283 		if (error)
1284 			status = RPM_SUSPENDED;
1285 
1286 		device_links_read_unlock(idx);
1287 	}
1288 
1289 	spin_lock_irqsave(&dev->power.lock, flags);
1290 
1291 	if (dev->power.runtime_status == status || !parent)
1292 		goto out_set;
1293 
1294 	if (status == RPM_SUSPENDED) {
1295 		atomic_add_unless(&parent->power.child_count, -1, 0);
1296 		notify_parent = !parent->power.ignore_children;
1297 	} else {
1298 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1299 
1300 		/*
1301 		 * It is invalid to put an active child under a parent that is
1302 		 * not active, has runtime PM enabled and the
1303 		 * 'power.ignore_children' flag unset.
1304 		 */
1305 		if (!parent->power.disable_depth &&
1306 		    !parent->power.ignore_children &&
1307 		    parent->power.runtime_status != RPM_ACTIVE) {
1308 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1309 				dev_name(dev),
1310 				dev_name(parent));
1311 			error = -EBUSY;
1312 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1313 			atomic_inc(&parent->power.child_count);
1314 		}
1315 
1316 		spin_unlock(&parent->power.lock);
1317 
1318 		if (error) {
1319 			status = RPM_SUSPENDED;
1320 			goto out;
1321 		}
1322 	}
1323 
1324  out_set:
1325 	__update_runtime_status(dev, status);
1326 	if (!error)
1327 		dev->power.runtime_error = 0;
1328 
1329  out:
1330 	spin_unlock_irqrestore(&dev->power.lock, flags);
1331 
1332 	if (notify_parent)
1333 		pm_request_idle(parent);
1334 
1335 	if (status == RPM_SUSPENDED) {
1336 		int idx = device_links_read_lock();
1337 
1338 		rpm_put_suppliers(dev);
1339 
1340 		device_links_read_unlock(idx);
1341 	}
1342 
1343 	pm_runtime_enable(dev);
1344 
1345 	return error;
1346 }
1347 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1348 
1349 /**
1350  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1351  * @dev: Device to handle.
1352  *
1353  * Flush all pending requests for the device from pm_wq and wait for all
1354  * runtime PM operations involving the device in progress to complete.
1355  *
1356  * Should be called under dev->power.lock with interrupts disabled.
1357  */
1358 static void __pm_runtime_barrier(struct device *dev)
1359 {
1360 	pm_runtime_deactivate_timer(dev);
1361 
1362 	if (dev->power.request_pending) {
1363 		dev->power.request = RPM_REQ_NONE;
1364 		spin_unlock_irq(&dev->power.lock);
1365 
1366 		cancel_work_sync(&dev->power.work);
1367 
1368 		spin_lock_irq(&dev->power.lock);
1369 		dev->power.request_pending = false;
1370 	}
1371 
1372 	if (dev->power.runtime_status == RPM_SUSPENDING ||
1373 	    dev->power.runtime_status == RPM_RESUMING ||
1374 	    dev->power.idle_notification) {
1375 		DEFINE_WAIT(wait);
1376 
1377 		/* Suspend, wake-up or idle notification in progress. */
1378 		for (;;) {
1379 			prepare_to_wait(&dev->power.wait_queue, &wait,
1380 					TASK_UNINTERRUPTIBLE);
1381 			if (dev->power.runtime_status != RPM_SUSPENDING
1382 			    && dev->power.runtime_status != RPM_RESUMING
1383 			    && !dev->power.idle_notification)
1384 				break;
1385 			spin_unlock_irq(&dev->power.lock);
1386 
1387 			schedule();
1388 
1389 			spin_lock_irq(&dev->power.lock);
1390 		}
1391 		finish_wait(&dev->power.wait_queue, &wait);
1392 	}
1393 }
1394 
1395 /**
1396  * pm_runtime_barrier - Flush pending requests and wait for completions.
1397  * @dev: Device to handle.
1398  *
1399  * Prevent the device from being suspended by incrementing its usage counter and
1400  * if there's a pending resume request for the device, wake the device up.
1401  * Next, make sure that all pending requests for the device have been flushed
1402  * from pm_wq and wait for all runtime PM operations involving the device in
1403  * progress to complete.
1404  *
1405  * Return value:
1406  * 1, if there was a resume request pending and the device had to be woken up,
1407  * 0, otherwise
1408  */
1409 int pm_runtime_barrier(struct device *dev)
1410 {
1411 	int retval = 0;
1412 
1413 	pm_runtime_get_noresume(dev);
1414 	spin_lock_irq(&dev->power.lock);
1415 
1416 	if (dev->power.request_pending
1417 	    && dev->power.request == RPM_REQ_RESUME) {
1418 		rpm_resume(dev, 0);
1419 		retval = 1;
1420 	}
1421 
1422 	__pm_runtime_barrier(dev);
1423 
1424 	spin_unlock_irq(&dev->power.lock);
1425 	pm_runtime_put_noidle(dev);
1426 
1427 	return retval;
1428 }
1429 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1430 
1431 /**
1432  * __pm_runtime_disable - Disable runtime PM of a device.
1433  * @dev: Device to handle.
1434  * @check_resume: If set, check if there's a resume request for the device.
1435  *
1436  * Increment power.disable_depth for the device and if it was zero previously,
1437  * cancel all pending runtime PM requests for the device and wait for all
1438  * operations in progress to complete.  The device can be either active or
1439  * suspended after its runtime PM has been disabled.
1440  *
1441  * If @check_resume is set and there's a resume request pending when
1442  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1443  * function will wake up the device before disabling its runtime PM.
1444  */
1445 void __pm_runtime_disable(struct device *dev, bool check_resume)
1446 {
1447 	spin_lock_irq(&dev->power.lock);
1448 
1449 	if (dev->power.disable_depth > 0) {
1450 		dev->power.disable_depth++;
1451 		goto out;
1452 	}
1453 
1454 	/*
1455 	 * Wake up the device if there's a resume request pending, because that
1456 	 * means there probably is some I/O to process and disabling runtime PM
1457 	 * shouldn't prevent the device from processing the I/O.
1458 	 */
1459 	if (check_resume && dev->power.request_pending &&
1460 	    dev->power.request == RPM_REQ_RESUME) {
1461 		/*
1462 		 * Prevent suspends and idle notifications from being carried
1463 		 * out after we have woken up the device.
1464 		 */
1465 		pm_runtime_get_noresume(dev);
1466 
1467 		rpm_resume(dev, 0);
1468 
1469 		pm_runtime_put_noidle(dev);
1470 	}
1471 
1472 	/* Update time accounting before disabling PM-runtime. */
1473 	update_pm_runtime_accounting(dev);
1474 
1475 	if (!dev->power.disable_depth++) {
1476 		__pm_runtime_barrier(dev);
1477 		dev->power.last_status = dev->power.runtime_status;
1478 	}
1479 
1480  out:
1481 	spin_unlock_irq(&dev->power.lock);
1482 }
1483 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1484 
1485 /**
1486  * pm_runtime_enable - Enable runtime PM of a device.
1487  * @dev: Device to handle.
1488  */
1489 void pm_runtime_enable(struct device *dev)
1490 {
1491 	unsigned long flags;
1492 
1493 	spin_lock_irqsave(&dev->power.lock, flags);
1494 
1495 	if (!dev->power.disable_depth) {
1496 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1497 		goto out;
1498 	}
1499 
1500 	if (--dev->power.disable_depth > 0)
1501 		goto out;
1502 
1503 	dev->power.last_status = RPM_INVALID;
1504 	dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1505 
1506 	if (dev->power.runtime_status == RPM_SUSPENDED &&
1507 	    !dev->power.ignore_children &&
1508 	    atomic_read(&dev->power.child_count) > 0)
1509 		dev_warn(dev, "Enabling runtime PM for inactive device with active children\n");
1510 
1511 out:
1512 	spin_unlock_irqrestore(&dev->power.lock, flags);
1513 }
1514 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1515 
1516 static void pm_runtime_disable_action(void *data)
1517 {
1518 	pm_runtime_dont_use_autosuspend(data);
1519 	pm_runtime_disable(data);
1520 }
1521 
1522 /**
1523  * devm_pm_runtime_enable - devres-enabled version of pm_runtime_enable.
1524  *
1525  * NOTE: this will also handle calling pm_runtime_dont_use_autosuspend() for
1526  * you at driver exit time if needed.
1527  *
1528  * @dev: Device to handle.
1529  */
1530 int devm_pm_runtime_enable(struct device *dev)
1531 {
1532 	pm_runtime_enable(dev);
1533 
1534 	return devm_add_action_or_reset(dev, pm_runtime_disable_action, dev);
1535 }
1536 EXPORT_SYMBOL_GPL(devm_pm_runtime_enable);
1537 
1538 /**
1539  * pm_runtime_forbid - Block runtime PM of a device.
1540  * @dev: Device to handle.
1541  *
1542  * Increase the device's usage count and clear its power.runtime_auto flag,
1543  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1544  * for it.
1545  */
1546 void pm_runtime_forbid(struct device *dev)
1547 {
1548 	spin_lock_irq(&dev->power.lock);
1549 	if (!dev->power.runtime_auto)
1550 		goto out;
1551 
1552 	dev->power.runtime_auto = false;
1553 	atomic_inc(&dev->power.usage_count);
1554 	rpm_resume(dev, 0);
1555 
1556  out:
1557 	spin_unlock_irq(&dev->power.lock);
1558 }
1559 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1560 
1561 /**
1562  * pm_runtime_allow - Unblock runtime PM of a device.
1563  * @dev: Device to handle.
1564  *
1565  * Decrease the device's usage count and set its power.runtime_auto flag.
1566  */
1567 void pm_runtime_allow(struct device *dev)
1568 {
1569 	int ret;
1570 
1571 	spin_lock_irq(&dev->power.lock);
1572 	if (dev->power.runtime_auto)
1573 		goto out;
1574 
1575 	dev->power.runtime_auto = true;
1576 	ret = rpm_drop_usage_count(dev);
1577 	if (ret == 0)
1578 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1579 	else if (ret > 0)
1580 		trace_rpm_usage(dev, RPM_AUTO | RPM_ASYNC);
1581 
1582  out:
1583 	spin_unlock_irq(&dev->power.lock);
1584 }
1585 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1586 
1587 /**
1588  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1589  * @dev: Device to handle.
1590  *
1591  * Set the power.no_callbacks flag, which tells the PM core that this
1592  * device is power-managed through its parent and has no runtime PM
1593  * callbacks of its own.  The runtime sysfs attributes will be removed.
1594  */
1595 void pm_runtime_no_callbacks(struct device *dev)
1596 {
1597 	spin_lock_irq(&dev->power.lock);
1598 	dev->power.no_callbacks = 1;
1599 	spin_unlock_irq(&dev->power.lock);
1600 	if (device_is_registered(dev))
1601 		rpm_sysfs_remove(dev);
1602 }
1603 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1604 
1605 /**
1606  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1607  * @dev: Device to handle
1608  *
1609  * Set the power.irq_safe flag, which tells the PM core that the
1610  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1611  * always be invoked with the spinlock held and interrupts disabled.  It also
1612  * causes the parent's usage counter to be permanently incremented, preventing
1613  * the parent from runtime suspending -- otherwise an irq-safe child might have
1614  * to wait for a non-irq-safe parent.
1615  */
1616 void pm_runtime_irq_safe(struct device *dev)
1617 {
1618 	if (dev->parent)
1619 		pm_runtime_get_sync(dev->parent);
1620 
1621 	spin_lock_irq(&dev->power.lock);
1622 	dev->power.irq_safe = 1;
1623 	spin_unlock_irq(&dev->power.lock);
1624 }
1625 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1626 
1627 /**
1628  * update_autosuspend - Handle a change to a device's autosuspend settings.
1629  * @dev: Device to handle.
1630  * @old_delay: The former autosuspend_delay value.
1631  * @old_use: The former use_autosuspend value.
1632  *
1633  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1634  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1635  *
1636  * This function must be called under dev->power.lock with interrupts disabled.
1637  */
1638 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1639 {
1640 	int delay = dev->power.autosuspend_delay;
1641 
1642 	/* Should runtime suspend be prevented now? */
1643 	if (dev->power.use_autosuspend && delay < 0) {
1644 
1645 		/* If it used to be allowed then prevent it. */
1646 		if (!old_use || old_delay >= 0) {
1647 			atomic_inc(&dev->power.usage_count);
1648 			rpm_resume(dev, 0);
1649 		} else {
1650 			trace_rpm_usage(dev, 0);
1651 		}
1652 	}
1653 
1654 	/* Runtime suspend should be allowed now. */
1655 	else {
1656 
1657 		/* If it used to be prevented then allow it. */
1658 		if (old_use && old_delay < 0)
1659 			atomic_dec(&dev->power.usage_count);
1660 
1661 		/* Maybe we can autosuspend now. */
1662 		rpm_idle(dev, RPM_AUTO);
1663 	}
1664 }
1665 
1666 /**
1667  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1668  * @dev: Device to handle.
1669  * @delay: Value of the new delay in milliseconds.
1670  *
1671  * Set the device's power.autosuspend_delay value.  If it changes to negative
1672  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1673  * changes the other way, allow runtime suspends.
1674  */
1675 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1676 {
1677 	int old_delay, old_use;
1678 
1679 	spin_lock_irq(&dev->power.lock);
1680 	old_delay = dev->power.autosuspend_delay;
1681 	old_use = dev->power.use_autosuspend;
1682 	dev->power.autosuspend_delay = delay;
1683 	update_autosuspend(dev, old_delay, old_use);
1684 	spin_unlock_irq(&dev->power.lock);
1685 }
1686 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1687 
1688 /**
1689  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1690  * @dev: Device to handle.
1691  * @use: New value for use_autosuspend.
1692  *
1693  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1694  * suspends as needed.
1695  */
1696 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1697 {
1698 	int old_delay, old_use;
1699 
1700 	spin_lock_irq(&dev->power.lock);
1701 	old_delay = dev->power.autosuspend_delay;
1702 	old_use = dev->power.use_autosuspend;
1703 	dev->power.use_autosuspend = use;
1704 	update_autosuspend(dev, old_delay, old_use);
1705 	spin_unlock_irq(&dev->power.lock);
1706 }
1707 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1708 
1709 /**
1710  * pm_runtime_init - Initialize runtime PM fields in given device object.
1711  * @dev: Device object to initialize.
1712  */
1713 void pm_runtime_init(struct device *dev)
1714 {
1715 	dev->power.runtime_status = RPM_SUSPENDED;
1716 	dev->power.last_status = RPM_INVALID;
1717 	dev->power.idle_notification = false;
1718 
1719 	dev->power.disable_depth = 1;
1720 	atomic_set(&dev->power.usage_count, 0);
1721 
1722 	dev->power.runtime_error = 0;
1723 
1724 	atomic_set(&dev->power.child_count, 0);
1725 	pm_suspend_ignore_children(dev, false);
1726 	dev->power.runtime_auto = true;
1727 
1728 	dev->power.request_pending = false;
1729 	dev->power.request = RPM_REQ_NONE;
1730 	dev->power.deferred_resume = false;
1731 	dev->power.needs_force_resume = 0;
1732 	INIT_WORK(&dev->power.work, pm_runtime_work);
1733 
1734 	dev->power.timer_expires = 0;
1735 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1736 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1737 
1738 	init_waitqueue_head(&dev->power.wait_queue);
1739 }
1740 
1741 /**
1742  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1743  * @dev: Device object to re-initialize.
1744  */
1745 void pm_runtime_reinit(struct device *dev)
1746 {
1747 	if (!pm_runtime_enabled(dev)) {
1748 		if (dev->power.runtime_status == RPM_ACTIVE)
1749 			pm_runtime_set_suspended(dev);
1750 		if (dev->power.irq_safe) {
1751 			spin_lock_irq(&dev->power.lock);
1752 			dev->power.irq_safe = 0;
1753 			spin_unlock_irq(&dev->power.lock);
1754 			if (dev->parent)
1755 				pm_runtime_put(dev->parent);
1756 		}
1757 	}
1758 }
1759 
1760 /**
1761  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1762  * @dev: Device object being removed from device hierarchy.
1763  */
1764 void pm_runtime_remove(struct device *dev)
1765 {
1766 	__pm_runtime_disable(dev, false);
1767 	pm_runtime_reinit(dev);
1768 }
1769 
1770 /**
1771  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1772  * @dev: Consumer device.
1773  */
1774 void pm_runtime_get_suppliers(struct device *dev)
1775 {
1776 	struct device_link *link;
1777 	int idx;
1778 
1779 	idx = device_links_read_lock();
1780 
1781 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1782 				device_links_read_lock_held())
1783 		if (link->flags & DL_FLAG_PM_RUNTIME) {
1784 			link->supplier_preactivated = true;
1785 			pm_runtime_get_sync(link->supplier);
1786 		}
1787 
1788 	device_links_read_unlock(idx);
1789 }
1790 
1791 /**
1792  * pm_runtime_put_suppliers - Drop references to supplier devices.
1793  * @dev: Consumer device.
1794  */
1795 void pm_runtime_put_suppliers(struct device *dev)
1796 {
1797 	struct device_link *link;
1798 	int idx;
1799 
1800 	idx = device_links_read_lock();
1801 
1802 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1803 				device_links_read_lock_held())
1804 		if (link->supplier_preactivated) {
1805 			link->supplier_preactivated = false;
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);
1842 	pm_request_idle(link->supplier);
1843 }
1844 
1845 static bool pm_runtime_need_not_resume(struct device *dev)
1846 {
1847 	return atomic_read(&dev->power.usage_count) <= 1 &&
1848 		(atomic_read(&dev->power.child_count) == 0 ||
1849 		 dev->power.ignore_children);
1850 }
1851 
1852 /**
1853  * pm_runtime_force_suspend - Force a device into suspend state if needed.
1854  * @dev: Device to suspend.
1855  *
1856  * Disable runtime PM so we safely can check the device's runtime PM status and
1857  * if it is active, invoke its ->runtime_suspend callback to suspend it and
1858  * change its runtime PM status field to RPM_SUSPENDED.  Also, if the device's
1859  * usage and children counters don't indicate that the device was in use before
1860  * the system-wide transition under way, decrement its parent's children counter
1861  * (if there is a parent).  Keep runtime PM disabled to preserve the state
1862  * unless we encounter errors.
1863  *
1864  * Typically this function may be invoked from a system suspend callback to make
1865  * sure the device is put into low power state and it should only be used during
1866  * system-wide PM transitions to sleep states.  It assumes that the analogous
1867  * pm_runtime_force_resume() will be used to resume the device.
1868  *
1869  * Do not use with DPM_FLAG_SMART_SUSPEND as this can lead to an inconsistent
1870  * state where this function has called the ->runtime_suspend callback but the
1871  * PM core marks the driver as runtime active.
1872  */
1873 int pm_runtime_force_suspend(struct device *dev)
1874 {
1875 	int (*callback)(struct device *);
1876 	int ret;
1877 
1878 	pm_runtime_disable(dev);
1879 	if (pm_runtime_status_suspended(dev))
1880 		return 0;
1881 
1882 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1883 
1884 	dev_pm_enable_wake_irq_check(dev, true);
1885 	ret = callback ? callback(dev) : 0;
1886 	if (ret)
1887 		goto err;
1888 
1889 	dev_pm_enable_wake_irq_complete(dev);
1890 
1891 	/*
1892 	 * If the device can stay in suspend after the system-wide transition
1893 	 * to the working state that will follow, drop the children counter of
1894 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1895 	 * function will be called again for it in the meantime.
1896 	 */
1897 	if (pm_runtime_need_not_resume(dev)) {
1898 		pm_runtime_set_suspended(dev);
1899 	} else {
1900 		__update_runtime_status(dev, RPM_SUSPENDED);
1901 		dev->power.needs_force_resume = 1;
1902 	}
1903 
1904 	return 0;
1905 
1906 err:
1907 	dev_pm_disable_wake_irq_check(dev, true);
1908 	pm_runtime_enable(dev);
1909 	return ret;
1910 }
1911 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1912 
1913 /**
1914  * pm_runtime_force_resume - Force a device into resume state if needed.
1915  * @dev: Device to resume.
1916  *
1917  * Prior invoking this function we expect the user to have brought the device
1918  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1919  * those actions and bring the device into full power, if it is expected to be
1920  * used on system resume.  In the other case, we defer the resume to be managed
1921  * via runtime PM.
1922  *
1923  * Typically this function may be invoked from a system resume callback.
1924  */
1925 int pm_runtime_force_resume(struct device *dev)
1926 {
1927 	int (*callback)(struct device *);
1928 	int ret = 0;
1929 
1930 	if (!pm_runtime_status_suspended(dev) || !dev->power.needs_force_resume)
1931 		goto out;
1932 
1933 	/*
1934 	 * The value of the parent's children counter is correct already, so
1935 	 * just update the status of the device.
1936 	 */
1937 	__update_runtime_status(dev, RPM_ACTIVE);
1938 
1939 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1940 
1941 	dev_pm_disable_wake_irq_check(dev, false);
1942 	ret = callback ? callback(dev) : 0;
1943 	if (ret) {
1944 		pm_runtime_set_suspended(dev);
1945 		dev_pm_enable_wake_irq_check(dev, false);
1946 		goto out;
1947 	}
1948 
1949 	pm_runtime_mark_last_busy(dev);
1950 out:
1951 	dev->power.needs_force_resume = 0;
1952 	pm_runtime_enable(dev);
1953 	return ret;
1954 }
1955 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1956