xref: /openbmc/linux/drivers/base/power/runtime.c (revision 95777591)
1 /*
2  * drivers/base/power/runtime.c - Helper functions for device runtime PM
3  *
4  * Copyright (c) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
5  * Copyright (C) 2010 Alan Stern <stern@rowland.harvard.edu>
6  *
7  * This file is released under the GPLv2.
8  */
9 
10 #include <linux/sched/mm.h>
11 #include <linux/ktime.h>
12 #include <linux/hrtimer.h>
13 #include <linux/export.h>
14 #include <linux/pm_runtime.h>
15 #include <linux/pm_wakeirq.h>
16 #include <trace/events/rpm.h>
17 
18 #include "../base.h"
19 #include "power.h"
20 
21 typedef int (*pm_callback_t)(struct device *);
22 
23 static pm_callback_t __rpm_get_callback(struct device *dev, size_t cb_offset)
24 {
25 	pm_callback_t cb;
26 	const struct dev_pm_ops *ops;
27 
28 	if (dev->pm_domain)
29 		ops = &dev->pm_domain->ops;
30 	else if (dev->type && dev->type->pm)
31 		ops = dev->type->pm;
32 	else if (dev->class && dev->class->pm)
33 		ops = dev->class->pm;
34 	else if (dev->bus && dev->bus->pm)
35 		ops = dev->bus->pm;
36 	else
37 		ops = NULL;
38 
39 	if (ops)
40 		cb = *(pm_callback_t *)((void *)ops + cb_offset);
41 	else
42 		cb = NULL;
43 
44 	if (!cb && dev->driver && dev->driver->pm)
45 		cb = *(pm_callback_t *)((void *)dev->driver->pm + cb_offset);
46 
47 	return cb;
48 }
49 
50 #define RPM_GET_CALLBACK(dev, callback) \
51 		__rpm_get_callback(dev, offsetof(struct dev_pm_ops, callback))
52 
53 static int rpm_resume(struct device *dev, int rpmflags);
54 static int rpm_suspend(struct device *dev, int rpmflags);
55 
56 /**
57  * update_pm_runtime_accounting - Update the time accounting of power states
58  * @dev: Device to update the accounting for
59  *
60  * In order to be able to have time accounting of the various power states
61  * (as used by programs such as PowerTOP to show the effectiveness of runtime
62  * PM), we need to track the time spent in each state.
63  * update_pm_runtime_accounting must be called each time before the
64  * runtime_status field is updated, to account the time in the old state
65  * correctly.
66  */
67 void update_pm_runtime_accounting(struct device *dev)
68 {
69 	u64 now, last, delta;
70 
71 	if (dev->power.disable_depth > 0)
72 		return;
73 
74 	last = dev->power.accounting_timestamp;
75 
76 	now = ktime_get_mono_fast_ns();
77 	dev->power.accounting_timestamp = now;
78 
79 	/*
80 	 * Because ktime_get_mono_fast_ns() is not monotonic during
81 	 * timekeeping updates, ensure that 'now' is after the last saved
82 	 * timesptamp.
83 	 */
84 	if (now < last)
85 		return;
86 
87 	delta = now - last;
88 
89 	if (dev->power.runtime_status == RPM_SUSPENDED)
90 		dev->power.suspended_time += delta;
91 	else
92 		dev->power.active_time += delta;
93 }
94 
95 static void __update_runtime_status(struct device *dev, enum rpm_status status)
96 {
97 	update_pm_runtime_accounting(dev);
98 	dev->power.runtime_status = status;
99 }
100 
101 u64 pm_runtime_suspended_time(struct device *dev)
102 {
103 	u64 time;
104 	unsigned long flags;
105 
106 	spin_lock_irqsave(&dev->power.lock, flags);
107 
108 	update_pm_runtime_accounting(dev);
109 	time = dev->power.suspended_time;
110 
111 	spin_unlock_irqrestore(&dev->power.lock, flags);
112 
113 	return time;
114 }
115 EXPORT_SYMBOL_GPL(pm_runtime_suspended_time);
116 
117 /**
118  * pm_runtime_deactivate_timer - Deactivate given device's suspend timer.
119  * @dev: Device to handle.
120  */
121 static void pm_runtime_deactivate_timer(struct device *dev)
122 {
123 	if (dev->power.timer_expires > 0) {
124 		hrtimer_try_to_cancel(&dev->power.suspend_timer);
125 		dev->power.timer_expires = 0;
126 	}
127 }
128 
129 /**
130  * pm_runtime_cancel_pending - Deactivate suspend timer and cancel requests.
131  * @dev: Device to handle.
132  */
133 static void pm_runtime_cancel_pending(struct device *dev)
134 {
135 	pm_runtime_deactivate_timer(dev);
136 	/*
137 	 * In case there's a request pending, make sure its work function will
138 	 * return without doing anything.
139 	 */
140 	dev->power.request = RPM_REQ_NONE;
141 }
142 
143 /*
144  * pm_runtime_autosuspend_expiration - Get a device's autosuspend-delay expiration time.
145  * @dev: Device to handle.
146  *
147  * Compute the autosuspend-delay expiration time based on the device's
148  * power.last_busy time.  If the delay has already expired or is disabled
149  * (negative) or the power.use_autosuspend flag isn't set, return 0.
150  * Otherwise return the expiration time in nanoseconds (adjusted to be nonzero).
151  *
152  * This function may be called either with or without dev->power.lock held.
153  * Either way it can be racy, since power.last_busy may be updated at any time.
154  */
155 u64 pm_runtime_autosuspend_expiration(struct device *dev)
156 {
157 	int autosuspend_delay;
158 	u64 expires;
159 
160 	if (!dev->power.use_autosuspend)
161 		return 0;
162 
163 	autosuspend_delay = READ_ONCE(dev->power.autosuspend_delay);
164 	if (autosuspend_delay < 0)
165 		return 0;
166 
167 	expires  = READ_ONCE(dev->power.last_busy);
168 	expires += (u64)autosuspend_delay * NSEC_PER_MSEC;
169 	if (expires > ktime_get_mono_fast_ns())
170 		return expires;	/* Expires in the future */
171 
172 	return 0;
173 }
174 EXPORT_SYMBOL_GPL(pm_runtime_autosuspend_expiration);
175 
176 static int dev_memalloc_noio(struct device *dev, void *data)
177 {
178 	return dev->power.memalloc_noio;
179 }
180 
181 /*
182  * pm_runtime_set_memalloc_noio - Set a device's memalloc_noio flag.
183  * @dev: Device to handle.
184  * @enable: True for setting the flag and False for clearing the flag.
185  *
186  * Set the flag for all devices in the path from the device to the
187  * root device in the device tree if @enable is true, otherwise clear
188  * the flag for devices in the path whose siblings don't set the flag.
189  *
190  * The function should only be called by block device, or network
191  * device driver for solving the deadlock problem during runtime
192  * resume/suspend:
193  *
194  *     If memory allocation with GFP_KERNEL is called inside runtime
195  *     resume/suspend callback of any one of its ancestors(or the
196  *     block device itself), the deadlock may be triggered inside the
197  *     memory allocation since it might not complete until the block
198  *     device becomes active and the involed page I/O finishes. The
199  *     situation is pointed out first by Alan Stern. Network device
200  *     are involved in iSCSI kind of situation.
201  *
202  * The lock of dev_hotplug_mutex is held in the function for handling
203  * hotplug race because pm_runtime_set_memalloc_noio() may be called
204  * in async probe().
205  *
206  * The function should be called between device_add() and device_del()
207  * on the affected device(block/network device).
208  */
209 void pm_runtime_set_memalloc_noio(struct device *dev, bool enable)
210 {
211 	static DEFINE_MUTEX(dev_hotplug_mutex);
212 
213 	mutex_lock(&dev_hotplug_mutex);
214 	for (;;) {
215 		bool enabled;
216 
217 		/* hold power lock since bitfield is not SMP-safe. */
218 		spin_lock_irq(&dev->power.lock);
219 		enabled = dev->power.memalloc_noio;
220 		dev->power.memalloc_noio = enable;
221 		spin_unlock_irq(&dev->power.lock);
222 
223 		/*
224 		 * not need to enable ancestors any more if the device
225 		 * has been enabled.
226 		 */
227 		if (enabled && enable)
228 			break;
229 
230 		dev = dev->parent;
231 
232 		/*
233 		 * clear flag of the parent device only if all the
234 		 * children don't set the flag because ancestor's
235 		 * flag was set by any one of the descendants.
236 		 */
237 		if (!dev || (!enable &&
238 			     device_for_each_child(dev, NULL,
239 						   dev_memalloc_noio)))
240 			break;
241 	}
242 	mutex_unlock(&dev_hotplug_mutex);
243 }
244 EXPORT_SYMBOL_GPL(pm_runtime_set_memalloc_noio);
245 
246 /**
247  * rpm_check_suspend_allowed - Test whether a device may be suspended.
248  * @dev: Device to test.
249  */
250 static int rpm_check_suspend_allowed(struct device *dev)
251 {
252 	int retval = 0;
253 
254 	if (dev->power.runtime_error)
255 		retval = -EINVAL;
256 	else if (dev->power.disable_depth > 0)
257 		retval = -EACCES;
258 	else if (atomic_read(&dev->power.usage_count) > 0)
259 		retval = -EAGAIN;
260 	else if (!dev->power.ignore_children &&
261 			atomic_read(&dev->power.child_count))
262 		retval = -EBUSY;
263 
264 	/* Pending resume requests take precedence over suspends. */
265 	else if ((dev->power.deferred_resume
266 			&& dev->power.runtime_status == RPM_SUSPENDING)
267 	    || (dev->power.request_pending
268 			&& dev->power.request == RPM_REQ_RESUME))
269 		retval = -EAGAIN;
270 	else if (__dev_pm_qos_read_value(dev) == 0)
271 		retval = -EPERM;
272 	else if (dev->power.runtime_status == RPM_SUSPENDED)
273 		retval = 1;
274 
275 	return retval;
276 }
277 
278 static int rpm_get_suppliers(struct device *dev)
279 {
280 	struct device_link *link;
281 
282 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node) {
283 		int retval;
284 
285 		if (!(link->flags & DL_FLAG_PM_RUNTIME) ||
286 		    READ_ONCE(link->status) == DL_STATE_SUPPLIER_UNBIND)
287 			continue;
288 
289 		retval = pm_runtime_get_sync(link->supplier);
290 		/* Ignore suppliers with disabled runtime PM. */
291 		if (retval < 0 && retval != -EACCES) {
292 			pm_runtime_put_noidle(link->supplier);
293 			return retval;
294 		}
295 		refcount_inc(&link->rpm_active);
296 	}
297 	return 0;
298 }
299 
300 static void rpm_put_suppliers(struct device *dev)
301 {
302 	struct device_link *link;
303 
304 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node) {
305 		if (READ_ONCE(link->status) == DL_STATE_SUPPLIER_UNBIND)
306 			continue;
307 
308 		while (refcount_dec_not_one(&link->rpm_active))
309 			pm_runtime_put(link->supplier);
310 	}
311 }
312 
313 /**
314  * __rpm_callback - Run a given runtime PM callback for a given device.
315  * @cb: Runtime PM callback to run.
316  * @dev: Device to run the callback for.
317  */
318 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
319 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
320 {
321 	int retval, idx;
322 	bool use_links = dev->power.links_count > 0;
323 
324 	if (dev->power.irq_safe) {
325 		spin_unlock(&dev->power.lock);
326 	} else {
327 		spin_unlock_irq(&dev->power.lock);
328 
329 		/*
330 		 * Resume suppliers if necessary.
331 		 *
332 		 * The device's runtime PM status cannot change until this
333 		 * routine returns, so it is safe to read the status outside of
334 		 * the lock.
335 		 */
336 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
337 			idx = device_links_read_lock();
338 
339 			retval = rpm_get_suppliers(dev);
340 			if (retval)
341 				goto fail;
342 
343 			device_links_read_unlock(idx);
344 		}
345 	}
346 
347 	retval = cb(dev);
348 
349 	if (dev->power.irq_safe) {
350 		spin_lock(&dev->power.lock);
351 	} else {
352 		/*
353 		 * If the device is suspending and the callback has returned
354 		 * success, drop the usage counters of the suppliers that have
355 		 * been reference counted on its resume.
356 		 *
357 		 * Do that if resume fails too.
358 		 */
359 		if (use_links
360 		    && ((dev->power.runtime_status == RPM_SUSPENDING && !retval)
361 		    || (dev->power.runtime_status == RPM_RESUMING && retval))) {
362 			idx = device_links_read_lock();
363 
364  fail:
365 			rpm_put_suppliers(dev);
366 
367 			device_links_read_unlock(idx);
368 		}
369 
370 		spin_lock_irq(&dev->power.lock);
371 	}
372 
373 	return retval;
374 }
375 
376 /**
377  * rpm_idle - Notify device bus type if the device can be suspended.
378  * @dev: Device to notify the bus type about.
379  * @rpmflags: Flag bits.
380  *
381  * Check if the device's runtime PM status allows it to be suspended.  If
382  * another idle notification has been started earlier, return immediately.  If
383  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
384  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
385  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
386  *
387  * This function must be called under dev->power.lock with interrupts disabled.
388  */
389 static int rpm_idle(struct device *dev, int rpmflags)
390 {
391 	int (*callback)(struct device *);
392 	int retval;
393 
394 	trace_rpm_idle_rcuidle(dev, rpmflags);
395 	retval = rpm_check_suspend_allowed(dev);
396 	if (retval < 0)
397 		;	/* Conditions are wrong. */
398 
399 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
400 	else if (dev->power.runtime_status != RPM_ACTIVE)
401 		retval = -EAGAIN;
402 
403 	/*
404 	 * Any pending request other than an idle notification takes
405 	 * precedence over us, except that the timer may be running.
406 	 */
407 	else if (dev->power.request_pending &&
408 	    dev->power.request > RPM_REQ_IDLE)
409 		retval = -EAGAIN;
410 
411 	/* Act as though RPM_NOWAIT is always set. */
412 	else if (dev->power.idle_notification)
413 		retval = -EINPROGRESS;
414 	if (retval)
415 		goto out;
416 
417 	/* Pending requests need to be canceled. */
418 	dev->power.request = RPM_REQ_NONE;
419 
420 	if (dev->power.no_callbacks)
421 		goto out;
422 
423 	/* Carry out an asynchronous or a synchronous idle notification. */
424 	if (rpmflags & RPM_ASYNC) {
425 		dev->power.request = RPM_REQ_IDLE;
426 		if (!dev->power.request_pending) {
427 			dev->power.request_pending = true;
428 			queue_work(pm_wq, &dev->power.work);
429 		}
430 		trace_rpm_return_int_rcuidle(dev, _THIS_IP_, 0);
431 		return 0;
432 	}
433 
434 	dev->power.idle_notification = true;
435 
436 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
437 
438 	if (callback)
439 		retval = __rpm_callback(callback, dev);
440 
441 	dev->power.idle_notification = false;
442 	wake_up_all(&dev->power.wait_queue);
443 
444  out:
445 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
446 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
447 }
448 
449 /**
450  * rpm_callback - Run a given runtime PM callback for a given device.
451  * @cb: Runtime PM callback to run.
452  * @dev: Device to run the callback for.
453  */
454 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
455 {
456 	int retval;
457 
458 	if (!cb)
459 		return -ENOSYS;
460 
461 	if (dev->power.memalloc_noio) {
462 		unsigned int noio_flag;
463 
464 		/*
465 		 * Deadlock might be caused if memory allocation with
466 		 * GFP_KERNEL happens inside runtime_suspend and
467 		 * runtime_resume callbacks of one block device's
468 		 * ancestor or the block device itself. Network
469 		 * device might be thought as part of iSCSI block
470 		 * device, so network device and its ancestor should
471 		 * be marked as memalloc_noio too.
472 		 */
473 		noio_flag = memalloc_noio_save();
474 		retval = __rpm_callback(cb, dev);
475 		memalloc_noio_restore(noio_flag);
476 	} else {
477 		retval = __rpm_callback(cb, dev);
478 	}
479 
480 	dev->power.runtime_error = retval;
481 	return retval != -EACCES ? retval : -EIO;
482 }
483 
484 /**
485  * rpm_suspend - Carry out runtime suspend of given device.
486  * @dev: Device to suspend.
487  * @rpmflags: Flag bits.
488  *
489  * Check if the device's runtime PM status allows it to be suspended.
490  * Cancel a pending idle notification, autosuspend or suspend. If
491  * another suspend has been started earlier, either return immediately
492  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
493  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
494  * otherwise run the ->runtime_suspend() callback directly. When
495  * ->runtime_suspend succeeded, if a deferred resume was requested while
496  * the callback was running then carry it out, otherwise send an idle
497  * notification for its parent (if the suspend succeeded and both
498  * ignore_children of parent->power and irq_safe of dev->power are not set).
499  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
500  * flag is set and the next autosuspend-delay expiration time is in the
501  * future, schedule another autosuspend attempt.
502  *
503  * This function must be called under dev->power.lock with interrupts disabled.
504  */
505 static int rpm_suspend(struct device *dev, int rpmflags)
506 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
507 {
508 	int (*callback)(struct device *);
509 	struct device *parent = NULL;
510 	int retval;
511 
512 	trace_rpm_suspend_rcuidle(dev, rpmflags);
513 
514  repeat:
515 	retval = rpm_check_suspend_allowed(dev);
516 
517 	if (retval < 0)
518 		;	/* Conditions are wrong. */
519 
520 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
521 	else if (dev->power.runtime_status == RPM_RESUMING &&
522 	    !(rpmflags & RPM_ASYNC))
523 		retval = -EAGAIN;
524 	if (retval)
525 		goto out;
526 
527 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
528 	if ((rpmflags & RPM_AUTO)
529 	    && dev->power.runtime_status != RPM_SUSPENDING) {
530 		u64 expires = pm_runtime_autosuspend_expiration(dev);
531 
532 		if (expires != 0) {
533 			/* Pending requests need to be canceled. */
534 			dev->power.request = RPM_REQ_NONE;
535 
536 			/*
537 			 * Optimization: If the timer is already running and is
538 			 * set to expire at or before the autosuspend delay,
539 			 * avoid the overhead of resetting it.  Just let it
540 			 * expire; pm_suspend_timer_fn() will take care of the
541 			 * rest.
542 			 */
543 			if (!(dev->power.timer_expires &&
544 					dev->power.timer_expires <= expires)) {
545 				/*
546 				 * We add a slack of 25% to gather wakeups
547 				 * without sacrificing the granularity.
548 				 */
549 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
550 						    (NSEC_PER_MSEC >> 2);
551 
552 				dev->power.timer_expires = expires;
553 				hrtimer_start_range_ns(&dev->power.suspend_timer,
554 						ns_to_ktime(expires),
555 						slack,
556 						HRTIMER_MODE_ABS);
557 			}
558 			dev->power.timer_autosuspends = 1;
559 			goto out;
560 		}
561 	}
562 
563 	/* Other scheduled or pending requests need to be canceled. */
564 	pm_runtime_cancel_pending(dev);
565 
566 	if (dev->power.runtime_status == RPM_SUSPENDING) {
567 		DEFINE_WAIT(wait);
568 
569 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
570 			retval = -EINPROGRESS;
571 			goto out;
572 		}
573 
574 		if (dev->power.irq_safe) {
575 			spin_unlock(&dev->power.lock);
576 
577 			cpu_relax();
578 
579 			spin_lock(&dev->power.lock);
580 			goto repeat;
581 		}
582 
583 		/* Wait for the other suspend running in parallel with us. */
584 		for (;;) {
585 			prepare_to_wait(&dev->power.wait_queue, &wait,
586 					TASK_UNINTERRUPTIBLE);
587 			if (dev->power.runtime_status != RPM_SUSPENDING)
588 				break;
589 
590 			spin_unlock_irq(&dev->power.lock);
591 
592 			schedule();
593 
594 			spin_lock_irq(&dev->power.lock);
595 		}
596 		finish_wait(&dev->power.wait_queue, &wait);
597 		goto repeat;
598 	}
599 
600 	if (dev->power.no_callbacks)
601 		goto no_callback;	/* Assume success. */
602 
603 	/* Carry out an asynchronous or a synchronous suspend. */
604 	if (rpmflags & RPM_ASYNC) {
605 		dev->power.request = (rpmflags & RPM_AUTO) ?
606 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
607 		if (!dev->power.request_pending) {
608 			dev->power.request_pending = true;
609 			queue_work(pm_wq, &dev->power.work);
610 		}
611 		goto out;
612 	}
613 
614 	__update_runtime_status(dev, RPM_SUSPENDING);
615 
616 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
617 
618 	dev_pm_enable_wake_irq_check(dev, true);
619 	retval = rpm_callback(callback, dev);
620 	if (retval)
621 		goto fail;
622 
623  no_callback:
624 	__update_runtime_status(dev, RPM_SUSPENDED);
625 	pm_runtime_deactivate_timer(dev);
626 
627 	if (dev->parent) {
628 		parent = dev->parent;
629 		atomic_add_unless(&parent->power.child_count, -1, 0);
630 	}
631 	wake_up_all(&dev->power.wait_queue);
632 
633 	if (dev->power.deferred_resume) {
634 		dev->power.deferred_resume = false;
635 		rpm_resume(dev, 0);
636 		retval = -EAGAIN;
637 		goto out;
638 	}
639 
640 	/* Maybe the parent is now able to suspend. */
641 	if (parent && !parent->power.ignore_children && !dev->power.irq_safe) {
642 		spin_unlock(&dev->power.lock);
643 
644 		spin_lock(&parent->power.lock);
645 		rpm_idle(parent, RPM_ASYNC);
646 		spin_unlock(&parent->power.lock);
647 
648 		spin_lock(&dev->power.lock);
649 	}
650 
651  out:
652 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
653 
654 	return retval;
655 
656  fail:
657 	dev_pm_disable_wake_irq_check(dev);
658 	__update_runtime_status(dev, RPM_ACTIVE);
659 	dev->power.deferred_resume = false;
660 	wake_up_all(&dev->power.wait_queue);
661 
662 	if (retval == -EAGAIN || retval == -EBUSY) {
663 		dev->power.runtime_error = 0;
664 
665 		/*
666 		 * If the callback routine failed an autosuspend, and
667 		 * if the last_busy time has been updated so that there
668 		 * is a new autosuspend expiration time, automatically
669 		 * reschedule another autosuspend.
670 		 */
671 		if ((rpmflags & RPM_AUTO) &&
672 		    pm_runtime_autosuspend_expiration(dev) != 0)
673 			goto repeat;
674 	} else {
675 		pm_runtime_cancel_pending(dev);
676 	}
677 	goto out;
678 }
679 
680 /**
681  * rpm_resume - Carry out runtime resume of given device.
682  * @dev: Device to resume.
683  * @rpmflags: Flag bits.
684  *
685  * Check if the device's runtime PM status allows it to be resumed.  Cancel
686  * any scheduled or pending requests.  If another resume has been started
687  * earlier, either return immediately or wait for it to finish, depending on the
688  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
689  * parallel with this function, either tell the other process to resume after
690  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
691  * flag is set then queue a resume request; otherwise run the
692  * ->runtime_resume() callback directly.  Queue an idle notification for the
693  * device if the resume succeeded.
694  *
695  * This function must be called under dev->power.lock with interrupts disabled.
696  */
697 static int rpm_resume(struct device *dev, int rpmflags)
698 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
699 {
700 	int (*callback)(struct device *);
701 	struct device *parent = NULL;
702 	int retval = 0;
703 
704 	trace_rpm_resume_rcuidle(dev, rpmflags);
705 
706  repeat:
707 	if (dev->power.runtime_error)
708 		retval = -EINVAL;
709 	else if (dev->power.disable_depth == 1 && dev->power.is_suspended
710 	    && dev->power.runtime_status == RPM_ACTIVE)
711 		retval = 1;
712 	else if (dev->power.disable_depth > 0)
713 		retval = -EACCES;
714 	if (retval)
715 		goto out;
716 
717 	/*
718 	 * Other scheduled or pending requests need to be canceled.  Small
719 	 * optimization: If an autosuspend timer is running, leave it running
720 	 * rather than cancelling it now only to restart it again in the near
721 	 * future.
722 	 */
723 	dev->power.request = RPM_REQ_NONE;
724 	if (!dev->power.timer_autosuspends)
725 		pm_runtime_deactivate_timer(dev);
726 
727 	if (dev->power.runtime_status == RPM_ACTIVE) {
728 		retval = 1;
729 		goto out;
730 	}
731 
732 	if (dev->power.runtime_status == RPM_RESUMING
733 	    || dev->power.runtime_status == RPM_SUSPENDING) {
734 		DEFINE_WAIT(wait);
735 
736 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
737 			if (dev->power.runtime_status == RPM_SUSPENDING)
738 				dev->power.deferred_resume = true;
739 			else
740 				retval = -EINPROGRESS;
741 			goto out;
742 		}
743 
744 		if (dev->power.irq_safe) {
745 			spin_unlock(&dev->power.lock);
746 
747 			cpu_relax();
748 
749 			spin_lock(&dev->power.lock);
750 			goto repeat;
751 		}
752 
753 		/* Wait for the operation carried out in parallel with us. */
754 		for (;;) {
755 			prepare_to_wait(&dev->power.wait_queue, &wait,
756 					TASK_UNINTERRUPTIBLE);
757 			if (dev->power.runtime_status != RPM_RESUMING
758 			    && dev->power.runtime_status != RPM_SUSPENDING)
759 				break;
760 
761 			spin_unlock_irq(&dev->power.lock);
762 
763 			schedule();
764 
765 			spin_lock_irq(&dev->power.lock);
766 		}
767 		finish_wait(&dev->power.wait_queue, &wait);
768 		goto repeat;
769 	}
770 
771 	/*
772 	 * See if we can skip waking up the parent.  This is safe only if
773 	 * power.no_callbacks is set, because otherwise we don't know whether
774 	 * the resume will actually succeed.
775 	 */
776 	if (dev->power.no_callbacks && !parent && dev->parent) {
777 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
778 		if (dev->parent->power.disable_depth > 0
779 		    || dev->parent->power.ignore_children
780 		    || dev->parent->power.runtime_status == RPM_ACTIVE) {
781 			atomic_inc(&dev->parent->power.child_count);
782 			spin_unlock(&dev->parent->power.lock);
783 			retval = 1;
784 			goto no_callback;	/* Assume success. */
785 		}
786 		spin_unlock(&dev->parent->power.lock);
787 	}
788 
789 	/* Carry out an asynchronous or a synchronous resume. */
790 	if (rpmflags & RPM_ASYNC) {
791 		dev->power.request = RPM_REQ_RESUME;
792 		if (!dev->power.request_pending) {
793 			dev->power.request_pending = true;
794 			queue_work(pm_wq, &dev->power.work);
795 		}
796 		retval = 0;
797 		goto out;
798 	}
799 
800 	if (!parent && dev->parent) {
801 		/*
802 		 * Increment the parent's usage counter and resume it if
803 		 * necessary.  Not needed if dev is irq-safe; then the
804 		 * parent is permanently resumed.
805 		 */
806 		parent = dev->parent;
807 		if (dev->power.irq_safe)
808 			goto skip_parent;
809 		spin_unlock(&dev->power.lock);
810 
811 		pm_runtime_get_noresume(parent);
812 
813 		spin_lock(&parent->power.lock);
814 		/*
815 		 * Resume the parent if it has runtime PM enabled and not been
816 		 * set to ignore its children.
817 		 */
818 		if (!parent->power.disable_depth
819 		    && !parent->power.ignore_children) {
820 			rpm_resume(parent, 0);
821 			if (parent->power.runtime_status != RPM_ACTIVE)
822 				retval = -EBUSY;
823 		}
824 		spin_unlock(&parent->power.lock);
825 
826 		spin_lock(&dev->power.lock);
827 		if (retval)
828 			goto out;
829 		goto repeat;
830 	}
831  skip_parent:
832 
833 	if (dev->power.no_callbacks)
834 		goto no_callback;	/* Assume success. */
835 
836 	__update_runtime_status(dev, RPM_RESUMING);
837 
838 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
839 
840 	dev_pm_disable_wake_irq_check(dev);
841 	retval = rpm_callback(callback, dev);
842 	if (retval) {
843 		__update_runtime_status(dev, RPM_SUSPENDED);
844 		pm_runtime_cancel_pending(dev);
845 		dev_pm_enable_wake_irq_check(dev, false);
846 	} else {
847  no_callback:
848 		__update_runtime_status(dev, RPM_ACTIVE);
849 		pm_runtime_mark_last_busy(dev);
850 		if (parent)
851 			atomic_inc(&parent->power.child_count);
852 	}
853 	wake_up_all(&dev->power.wait_queue);
854 
855 	if (retval >= 0)
856 		rpm_idle(dev, RPM_ASYNC);
857 
858  out:
859 	if (parent && !dev->power.irq_safe) {
860 		spin_unlock_irq(&dev->power.lock);
861 
862 		pm_runtime_put(parent);
863 
864 		spin_lock_irq(&dev->power.lock);
865 	}
866 
867 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
868 
869 	return retval;
870 }
871 
872 /**
873  * pm_runtime_work - Universal runtime PM work function.
874  * @work: Work structure used for scheduling the execution of this function.
875  *
876  * Use @work to get the device object the work is to be done for, determine what
877  * is to be done and execute the appropriate runtime PM function.
878  */
879 static void pm_runtime_work(struct work_struct *work)
880 {
881 	struct device *dev = container_of(work, struct device, power.work);
882 	enum rpm_request req;
883 
884 	spin_lock_irq(&dev->power.lock);
885 
886 	if (!dev->power.request_pending)
887 		goto out;
888 
889 	req = dev->power.request;
890 	dev->power.request = RPM_REQ_NONE;
891 	dev->power.request_pending = false;
892 
893 	switch (req) {
894 	case RPM_REQ_NONE:
895 		break;
896 	case RPM_REQ_IDLE:
897 		rpm_idle(dev, RPM_NOWAIT);
898 		break;
899 	case RPM_REQ_SUSPEND:
900 		rpm_suspend(dev, RPM_NOWAIT);
901 		break;
902 	case RPM_REQ_AUTOSUSPEND:
903 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
904 		break;
905 	case RPM_REQ_RESUME:
906 		rpm_resume(dev, RPM_NOWAIT);
907 		break;
908 	}
909 
910  out:
911 	spin_unlock_irq(&dev->power.lock);
912 }
913 
914 /**
915  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
916  * @data: Device pointer passed by pm_schedule_suspend().
917  *
918  * Check if the time is right and queue a suspend request.
919  */
920 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
921 {
922 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
923 	unsigned long flags;
924 	u64 expires;
925 
926 	spin_lock_irqsave(&dev->power.lock, flags);
927 
928 	expires = dev->power.timer_expires;
929 	/*
930 	 * If 'expires' is after the current time, we've been called
931 	 * too early.
932 	 */
933 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
934 		dev->power.timer_expires = 0;
935 		rpm_suspend(dev, dev->power.timer_autosuspends ?
936 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
937 	}
938 
939 	spin_unlock_irqrestore(&dev->power.lock, flags);
940 
941 	return HRTIMER_NORESTART;
942 }
943 
944 /**
945  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
946  * @dev: Device to suspend.
947  * @delay: Time to wait before submitting a suspend request, in milliseconds.
948  */
949 int pm_schedule_suspend(struct device *dev, unsigned int delay)
950 {
951 	unsigned long flags;
952 	u64 expires;
953 	int retval;
954 
955 	spin_lock_irqsave(&dev->power.lock, flags);
956 
957 	if (!delay) {
958 		retval = rpm_suspend(dev, RPM_ASYNC);
959 		goto out;
960 	}
961 
962 	retval = rpm_check_suspend_allowed(dev);
963 	if (retval)
964 		goto out;
965 
966 	/* Other scheduled or pending requests need to be canceled. */
967 	pm_runtime_cancel_pending(dev);
968 
969 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
970 	dev->power.timer_expires = expires;
971 	dev->power.timer_autosuspends = 0;
972 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
973 
974  out:
975 	spin_unlock_irqrestore(&dev->power.lock, flags);
976 
977 	return retval;
978 }
979 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
980 
981 /**
982  * __pm_runtime_idle - Entry point for runtime idle operations.
983  * @dev: Device to send idle notification for.
984  * @rpmflags: Flag bits.
985  *
986  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
987  * return immediately if it is larger than zero.  Then carry out an idle
988  * notification, either synchronous or asynchronous.
989  *
990  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
991  * or if pm_runtime_irq_safe() has been called.
992  */
993 int __pm_runtime_idle(struct device *dev, int rpmflags)
994 {
995 	unsigned long flags;
996 	int retval;
997 
998 	if (rpmflags & RPM_GET_PUT) {
999 		if (!atomic_dec_and_test(&dev->power.usage_count))
1000 			return 0;
1001 	}
1002 
1003 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1004 
1005 	spin_lock_irqsave(&dev->power.lock, flags);
1006 	retval = rpm_idle(dev, rpmflags);
1007 	spin_unlock_irqrestore(&dev->power.lock, flags);
1008 
1009 	return retval;
1010 }
1011 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1012 
1013 /**
1014  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1015  * @dev: Device to suspend.
1016  * @rpmflags: Flag bits.
1017  *
1018  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1019  * return immediately if it is larger than zero.  Then carry out a suspend,
1020  * either synchronous or asynchronous.
1021  *
1022  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1023  * or if pm_runtime_irq_safe() has been called.
1024  */
1025 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1026 {
1027 	unsigned long flags;
1028 	int retval;
1029 
1030 	if (rpmflags & RPM_GET_PUT) {
1031 		if (!atomic_dec_and_test(&dev->power.usage_count))
1032 			return 0;
1033 	}
1034 
1035 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1036 
1037 	spin_lock_irqsave(&dev->power.lock, flags);
1038 	retval = rpm_suspend(dev, rpmflags);
1039 	spin_unlock_irqrestore(&dev->power.lock, flags);
1040 
1041 	return retval;
1042 }
1043 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1044 
1045 /**
1046  * __pm_runtime_resume - Entry point for runtime resume operations.
1047  * @dev: Device to resume.
1048  * @rpmflags: Flag bits.
1049  *
1050  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1051  * carry out a resume, either synchronous or asynchronous.
1052  *
1053  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1054  * or if pm_runtime_irq_safe() has been called.
1055  */
1056 int __pm_runtime_resume(struct device *dev, int rpmflags)
1057 {
1058 	unsigned long flags;
1059 	int retval;
1060 
1061 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1062 			dev->power.runtime_status != RPM_ACTIVE);
1063 
1064 	if (rpmflags & RPM_GET_PUT)
1065 		atomic_inc(&dev->power.usage_count);
1066 
1067 	spin_lock_irqsave(&dev->power.lock, flags);
1068 	retval = rpm_resume(dev, rpmflags);
1069 	spin_unlock_irqrestore(&dev->power.lock, flags);
1070 
1071 	return retval;
1072 }
1073 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1074 
1075 /**
1076  * pm_runtime_get_if_in_use - Conditionally bump up the device's usage counter.
1077  * @dev: Device to handle.
1078  *
1079  * Return -EINVAL if runtime PM is disabled for the device.
1080  *
1081  * If that's not the case and if the device's runtime PM status is RPM_ACTIVE
1082  * and the runtime PM usage counter is nonzero, increment the counter and
1083  * return 1.  Otherwise return 0 without changing the counter.
1084  */
1085 int pm_runtime_get_if_in_use(struct device *dev)
1086 {
1087 	unsigned long flags;
1088 	int retval;
1089 
1090 	spin_lock_irqsave(&dev->power.lock, flags);
1091 	retval = dev->power.disable_depth > 0 ? -EINVAL :
1092 		dev->power.runtime_status == RPM_ACTIVE
1093 			&& atomic_inc_not_zero(&dev->power.usage_count);
1094 	spin_unlock_irqrestore(&dev->power.lock, flags);
1095 	return retval;
1096 }
1097 EXPORT_SYMBOL_GPL(pm_runtime_get_if_in_use);
1098 
1099 /**
1100  * __pm_runtime_set_status - Set runtime PM status of a device.
1101  * @dev: Device to handle.
1102  * @status: New runtime PM status of the device.
1103  *
1104  * If runtime PM of the device is disabled or its power.runtime_error field is
1105  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1106  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1107  * However, if the device has a parent and the parent is not active, and the
1108  * parent's power.ignore_children flag is unset, the device's status cannot be
1109  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1110  *
1111  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1112  * and the device parent's counter of unsuspended children is modified to
1113  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1114  * notification request for the parent is submitted.
1115  *
1116  * If @dev has any suppliers (as reflected by device links to them), and @status
1117  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1118  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1119  * of the @status value) and the suppliers will be deacticated on exit.  The
1120  * error returned by the failing supplier activation will be returned in that
1121  * case.
1122  */
1123 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1124 {
1125 	struct device *parent = dev->parent;
1126 	bool notify_parent = false;
1127 	int error = 0;
1128 
1129 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1130 		return -EINVAL;
1131 
1132 	spin_lock_irq(&dev->power.lock);
1133 
1134 	/*
1135 	 * Prevent PM-runtime from being enabled for the device or return an
1136 	 * error if it is enabled already and working.
1137 	 */
1138 	if (dev->power.runtime_error || dev->power.disable_depth)
1139 		dev->power.disable_depth++;
1140 	else
1141 		error = -EAGAIN;
1142 
1143 	spin_unlock_irq(&dev->power.lock);
1144 
1145 	if (error)
1146 		return error;
1147 
1148 	/*
1149 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1150 	 * upfront regardless of the current status, because next time
1151 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1152 	 * involved will be dropped down to one anyway.
1153 	 */
1154 	if (status == RPM_ACTIVE) {
1155 		int idx = device_links_read_lock();
1156 
1157 		error = rpm_get_suppliers(dev);
1158 		if (error)
1159 			status = RPM_SUSPENDED;
1160 
1161 		device_links_read_unlock(idx);
1162 	}
1163 
1164 	spin_lock_irq(&dev->power.lock);
1165 
1166 	if (dev->power.runtime_status == status || !parent)
1167 		goto out_set;
1168 
1169 	if (status == RPM_SUSPENDED) {
1170 		atomic_add_unless(&parent->power.child_count, -1, 0);
1171 		notify_parent = !parent->power.ignore_children;
1172 	} else {
1173 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1174 
1175 		/*
1176 		 * It is invalid to put an active child under a parent that is
1177 		 * not active, has runtime PM enabled and the
1178 		 * 'power.ignore_children' flag unset.
1179 		 */
1180 		if (!parent->power.disable_depth
1181 		    && !parent->power.ignore_children
1182 		    && parent->power.runtime_status != RPM_ACTIVE) {
1183 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1184 				dev_name(dev),
1185 				dev_name(parent));
1186 			error = -EBUSY;
1187 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1188 			atomic_inc(&parent->power.child_count);
1189 		}
1190 
1191 		spin_unlock(&parent->power.lock);
1192 
1193 		if (error) {
1194 			status = RPM_SUSPENDED;
1195 			goto out;
1196 		}
1197 	}
1198 
1199  out_set:
1200 	__update_runtime_status(dev, status);
1201 	if (!error)
1202 		dev->power.runtime_error = 0;
1203 
1204  out:
1205 	spin_unlock_irq(&dev->power.lock);
1206 
1207 	if (notify_parent)
1208 		pm_request_idle(parent);
1209 
1210 	if (status == RPM_SUSPENDED) {
1211 		int idx = device_links_read_lock();
1212 
1213 		rpm_put_suppliers(dev);
1214 
1215 		device_links_read_unlock(idx);
1216 	}
1217 
1218 	pm_runtime_enable(dev);
1219 
1220 	return error;
1221 }
1222 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1223 
1224 /**
1225  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1226  * @dev: Device to handle.
1227  *
1228  * Flush all pending requests for the device from pm_wq and wait for all
1229  * runtime PM operations involving the device in progress to complete.
1230  *
1231  * Should be called under dev->power.lock with interrupts disabled.
1232  */
1233 static void __pm_runtime_barrier(struct device *dev)
1234 {
1235 	pm_runtime_deactivate_timer(dev);
1236 
1237 	if (dev->power.request_pending) {
1238 		dev->power.request = RPM_REQ_NONE;
1239 		spin_unlock_irq(&dev->power.lock);
1240 
1241 		cancel_work_sync(&dev->power.work);
1242 
1243 		spin_lock_irq(&dev->power.lock);
1244 		dev->power.request_pending = false;
1245 	}
1246 
1247 	if (dev->power.runtime_status == RPM_SUSPENDING
1248 	    || dev->power.runtime_status == RPM_RESUMING
1249 	    || dev->power.idle_notification) {
1250 		DEFINE_WAIT(wait);
1251 
1252 		/* Suspend, wake-up or idle notification in progress. */
1253 		for (;;) {
1254 			prepare_to_wait(&dev->power.wait_queue, &wait,
1255 					TASK_UNINTERRUPTIBLE);
1256 			if (dev->power.runtime_status != RPM_SUSPENDING
1257 			    && dev->power.runtime_status != RPM_RESUMING
1258 			    && !dev->power.idle_notification)
1259 				break;
1260 			spin_unlock_irq(&dev->power.lock);
1261 
1262 			schedule();
1263 
1264 			spin_lock_irq(&dev->power.lock);
1265 		}
1266 		finish_wait(&dev->power.wait_queue, &wait);
1267 	}
1268 }
1269 
1270 /**
1271  * pm_runtime_barrier - Flush pending requests and wait for completions.
1272  * @dev: Device to handle.
1273  *
1274  * Prevent the device from being suspended by incrementing its usage counter and
1275  * if there's a pending resume request for the device, wake the device up.
1276  * Next, make sure that all pending requests for the device have been flushed
1277  * from pm_wq and wait for all runtime PM operations involving the device in
1278  * progress to complete.
1279  *
1280  * Return value:
1281  * 1, if there was a resume request pending and the device had to be woken up,
1282  * 0, otherwise
1283  */
1284 int pm_runtime_barrier(struct device *dev)
1285 {
1286 	int retval = 0;
1287 
1288 	pm_runtime_get_noresume(dev);
1289 	spin_lock_irq(&dev->power.lock);
1290 
1291 	if (dev->power.request_pending
1292 	    && dev->power.request == RPM_REQ_RESUME) {
1293 		rpm_resume(dev, 0);
1294 		retval = 1;
1295 	}
1296 
1297 	__pm_runtime_barrier(dev);
1298 
1299 	spin_unlock_irq(&dev->power.lock);
1300 	pm_runtime_put_noidle(dev);
1301 
1302 	return retval;
1303 }
1304 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1305 
1306 /**
1307  * __pm_runtime_disable - Disable runtime PM of a device.
1308  * @dev: Device to handle.
1309  * @check_resume: If set, check if there's a resume request for the device.
1310  *
1311  * Increment power.disable_depth for the device and if it was zero previously,
1312  * cancel all pending runtime PM requests for the device and wait for all
1313  * operations in progress to complete.  The device can be either active or
1314  * suspended after its runtime PM has been disabled.
1315  *
1316  * If @check_resume is set and there's a resume request pending when
1317  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1318  * function will wake up the device before disabling its runtime PM.
1319  */
1320 void __pm_runtime_disable(struct device *dev, bool check_resume)
1321 {
1322 	spin_lock_irq(&dev->power.lock);
1323 
1324 	if (dev->power.disable_depth > 0) {
1325 		dev->power.disable_depth++;
1326 		goto out;
1327 	}
1328 
1329 	/*
1330 	 * Wake up the device if there's a resume request pending, because that
1331 	 * means there probably is some I/O to process and disabling runtime PM
1332 	 * shouldn't prevent the device from processing the I/O.
1333 	 */
1334 	if (check_resume && dev->power.request_pending
1335 	    && dev->power.request == RPM_REQ_RESUME) {
1336 		/*
1337 		 * Prevent suspends and idle notifications from being carried
1338 		 * out after we have woken up the device.
1339 		 */
1340 		pm_runtime_get_noresume(dev);
1341 
1342 		rpm_resume(dev, 0);
1343 
1344 		pm_runtime_put_noidle(dev);
1345 	}
1346 
1347 	/* Update time accounting before disabling PM-runtime. */
1348 	update_pm_runtime_accounting(dev);
1349 
1350 	if (!dev->power.disable_depth++)
1351 		__pm_runtime_barrier(dev);
1352 
1353  out:
1354 	spin_unlock_irq(&dev->power.lock);
1355 }
1356 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1357 
1358 /**
1359  * pm_runtime_enable - Enable runtime PM of a device.
1360  * @dev: Device to handle.
1361  */
1362 void pm_runtime_enable(struct device *dev)
1363 {
1364 	unsigned long flags;
1365 
1366 	spin_lock_irqsave(&dev->power.lock, flags);
1367 
1368 	if (dev->power.disable_depth > 0) {
1369 		dev->power.disable_depth--;
1370 
1371 		/* About to enable runtime pm, set accounting_timestamp to now */
1372 		if (!dev->power.disable_depth)
1373 			dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1374 	} else {
1375 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1376 	}
1377 
1378 	WARN(!dev->power.disable_depth &&
1379 	     dev->power.runtime_status == RPM_SUSPENDED &&
1380 	     !dev->power.ignore_children &&
1381 	     atomic_read(&dev->power.child_count) > 0,
1382 	     "Enabling runtime PM for inactive device (%s) with active children\n",
1383 	     dev_name(dev));
1384 
1385 	spin_unlock_irqrestore(&dev->power.lock, flags);
1386 }
1387 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1388 
1389 /**
1390  * pm_runtime_forbid - Block runtime PM of a device.
1391  * @dev: Device to handle.
1392  *
1393  * Increase the device's usage count and clear its power.runtime_auto flag,
1394  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1395  * for it.
1396  */
1397 void pm_runtime_forbid(struct device *dev)
1398 {
1399 	spin_lock_irq(&dev->power.lock);
1400 	if (!dev->power.runtime_auto)
1401 		goto out;
1402 
1403 	dev->power.runtime_auto = false;
1404 	atomic_inc(&dev->power.usage_count);
1405 	rpm_resume(dev, 0);
1406 
1407  out:
1408 	spin_unlock_irq(&dev->power.lock);
1409 }
1410 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1411 
1412 /**
1413  * pm_runtime_allow - Unblock runtime PM of a device.
1414  * @dev: Device to handle.
1415  *
1416  * Decrease the device's usage count and set its power.runtime_auto flag.
1417  */
1418 void pm_runtime_allow(struct device *dev)
1419 {
1420 	spin_lock_irq(&dev->power.lock);
1421 	if (dev->power.runtime_auto)
1422 		goto out;
1423 
1424 	dev->power.runtime_auto = true;
1425 	if (atomic_dec_and_test(&dev->power.usage_count))
1426 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1427 
1428  out:
1429 	spin_unlock_irq(&dev->power.lock);
1430 }
1431 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1432 
1433 /**
1434  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1435  * @dev: Device to handle.
1436  *
1437  * Set the power.no_callbacks flag, which tells the PM core that this
1438  * device is power-managed through its parent and has no runtime PM
1439  * callbacks of its own.  The runtime sysfs attributes will be removed.
1440  */
1441 void pm_runtime_no_callbacks(struct device *dev)
1442 {
1443 	spin_lock_irq(&dev->power.lock);
1444 	dev->power.no_callbacks = 1;
1445 	spin_unlock_irq(&dev->power.lock);
1446 	if (device_is_registered(dev))
1447 		rpm_sysfs_remove(dev);
1448 }
1449 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1450 
1451 /**
1452  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1453  * @dev: Device to handle
1454  *
1455  * Set the power.irq_safe flag, which tells the PM core that the
1456  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1457  * always be invoked with the spinlock held and interrupts disabled.  It also
1458  * causes the parent's usage counter to be permanently incremented, preventing
1459  * the parent from runtime suspending -- otherwise an irq-safe child might have
1460  * to wait for a non-irq-safe parent.
1461  */
1462 void pm_runtime_irq_safe(struct device *dev)
1463 {
1464 	if (dev->parent)
1465 		pm_runtime_get_sync(dev->parent);
1466 	spin_lock_irq(&dev->power.lock);
1467 	dev->power.irq_safe = 1;
1468 	spin_unlock_irq(&dev->power.lock);
1469 }
1470 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1471 
1472 /**
1473  * update_autosuspend - Handle a change to a device's autosuspend settings.
1474  * @dev: Device to handle.
1475  * @old_delay: The former autosuspend_delay value.
1476  * @old_use: The former use_autosuspend value.
1477  *
1478  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1479  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1480  *
1481  * This function must be called under dev->power.lock with interrupts disabled.
1482  */
1483 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1484 {
1485 	int delay = dev->power.autosuspend_delay;
1486 
1487 	/* Should runtime suspend be prevented now? */
1488 	if (dev->power.use_autosuspend && delay < 0) {
1489 
1490 		/* If it used to be allowed then prevent it. */
1491 		if (!old_use || old_delay >= 0) {
1492 			atomic_inc(&dev->power.usage_count);
1493 			rpm_resume(dev, 0);
1494 		}
1495 	}
1496 
1497 	/* Runtime suspend should be allowed now. */
1498 	else {
1499 
1500 		/* If it used to be prevented then allow it. */
1501 		if (old_use && old_delay < 0)
1502 			atomic_dec(&dev->power.usage_count);
1503 
1504 		/* Maybe we can autosuspend now. */
1505 		rpm_idle(dev, RPM_AUTO);
1506 	}
1507 }
1508 
1509 /**
1510  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1511  * @dev: Device to handle.
1512  * @delay: Value of the new delay in milliseconds.
1513  *
1514  * Set the device's power.autosuspend_delay value.  If it changes to negative
1515  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1516  * changes the other way, allow runtime suspends.
1517  */
1518 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1519 {
1520 	int old_delay, old_use;
1521 
1522 	spin_lock_irq(&dev->power.lock);
1523 	old_delay = dev->power.autosuspend_delay;
1524 	old_use = dev->power.use_autosuspend;
1525 	dev->power.autosuspend_delay = delay;
1526 	update_autosuspend(dev, old_delay, old_use);
1527 	spin_unlock_irq(&dev->power.lock);
1528 }
1529 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1530 
1531 /**
1532  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1533  * @dev: Device to handle.
1534  * @use: New value for use_autosuspend.
1535  *
1536  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1537  * suspends as needed.
1538  */
1539 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1540 {
1541 	int old_delay, old_use;
1542 
1543 	spin_lock_irq(&dev->power.lock);
1544 	old_delay = dev->power.autosuspend_delay;
1545 	old_use = dev->power.use_autosuspend;
1546 	dev->power.use_autosuspend = use;
1547 	update_autosuspend(dev, old_delay, old_use);
1548 	spin_unlock_irq(&dev->power.lock);
1549 }
1550 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1551 
1552 /**
1553  * pm_runtime_init - Initialize runtime PM fields in given device object.
1554  * @dev: Device object to initialize.
1555  */
1556 void pm_runtime_init(struct device *dev)
1557 {
1558 	dev->power.runtime_status = RPM_SUSPENDED;
1559 	dev->power.idle_notification = false;
1560 
1561 	dev->power.disable_depth = 1;
1562 	atomic_set(&dev->power.usage_count, 0);
1563 
1564 	dev->power.runtime_error = 0;
1565 
1566 	atomic_set(&dev->power.child_count, 0);
1567 	pm_suspend_ignore_children(dev, false);
1568 	dev->power.runtime_auto = true;
1569 
1570 	dev->power.request_pending = false;
1571 	dev->power.request = RPM_REQ_NONE;
1572 	dev->power.deferred_resume = false;
1573 	INIT_WORK(&dev->power.work, pm_runtime_work);
1574 
1575 	dev->power.timer_expires = 0;
1576 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1577 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1578 
1579 	init_waitqueue_head(&dev->power.wait_queue);
1580 }
1581 
1582 /**
1583  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1584  * @dev: Device object to re-initialize.
1585  */
1586 void pm_runtime_reinit(struct device *dev)
1587 {
1588 	if (!pm_runtime_enabled(dev)) {
1589 		if (dev->power.runtime_status == RPM_ACTIVE)
1590 			pm_runtime_set_suspended(dev);
1591 		if (dev->power.irq_safe) {
1592 			spin_lock_irq(&dev->power.lock);
1593 			dev->power.irq_safe = 0;
1594 			spin_unlock_irq(&dev->power.lock);
1595 			if (dev->parent)
1596 				pm_runtime_put(dev->parent);
1597 		}
1598 	}
1599 }
1600 
1601 /**
1602  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1603  * @dev: Device object being removed from device hierarchy.
1604  */
1605 void pm_runtime_remove(struct device *dev)
1606 {
1607 	__pm_runtime_disable(dev, false);
1608 	pm_runtime_reinit(dev);
1609 }
1610 
1611 /**
1612  * pm_runtime_clean_up_links - Prepare links to consumers for driver removal.
1613  * @dev: Device whose driver is going to be removed.
1614  *
1615  * Check links from this device to any consumers and if any of them have active
1616  * runtime PM references to the device, drop the usage counter of the device
1617  * (as many times as needed).
1618  *
1619  * Links with the DL_FLAG_STATELESS flag set are ignored.
1620  *
1621  * Since the device is guaranteed to be runtime-active at the point this is
1622  * called, nothing else needs to be done here.
1623  *
1624  * Moreover, this is called after device_links_busy() has returned 'false', so
1625  * the status of each link is guaranteed to be DL_STATE_SUPPLIER_UNBIND and
1626  * therefore rpm_active can't be manipulated concurrently.
1627  */
1628 void pm_runtime_clean_up_links(struct device *dev)
1629 {
1630 	struct device_link *link;
1631 	int idx;
1632 
1633 	idx = device_links_read_lock();
1634 
1635 	list_for_each_entry_rcu(link, &dev->links.consumers, s_node) {
1636 		if (link->flags & DL_FLAG_STATELESS)
1637 			continue;
1638 
1639 		while (refcount_dec_not_one(&link->rpm_active))
1640 			pm_runtime_put_noidle(dev);
1641 	}
1642 
1643 	device_links_read_unlock(idx);
1644 }
1645 
1646 /**
1647  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1648  * @dev: Consumer device.
1649  */
1650 void pm_runtime_get_suppliers(struct device *dev)
1651 {
1652 	struct device_link *link;
1653 	int idx;
1654 
1655 	idx = device_links_read_lock();
1656 
1657 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node)
1658 		if (link->flags & DL_FLAG_PM_RUNTIME) {
1659 			link->supplier_preactivated = true;
1660 			refcount_inc(&link->rpm_active);
1661 			pm_runtime_get_sync(link->supplier);
1662 		}
1663 
1664 	device_links_read_unlock(idx);
1665 }
1666 
1667 /**
1668  * pm_runtime_put_suppliers - Drop references to supplier devices.
1669  * @dev: Consumer device.
1670  */
1671 void pm_runtime_put_suppliers(struct device *dev)
1672 {
1673 	struct device_link *link;
1674 	int idx;
1675 
1676 	idx = device_links_read_lock();
1677 
1678 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node)
1679 		if (link->supplier_preactivated) {
1680 			link->supplier_preactivated = false;
1681 			if (refcount_dec_not_one(&link->rpm_active))
1682 				pm_runtime_put(link->supplier);
1683 		}
1684 
1685 	device_links_read_unlock(idx);
1686 }
1687 
1688 void pm_runtime_new_link(struct device *dev)
1689 {
1690 	spin_lock_irq(&dev->power.lock);
1691 	dev->power.links_count++;
1692 	spin_unlock_irq(&dev->power.lock);
1693 }
1694 
1695 void pm_runtime_drop_link(struct device *dev)
1696 {
1697 	spin_lock_irq(&dev->power.lock);
1698 	WARN_ON(dev->power.links_count == 0);
1699 	dev->power.links_count--;
1700 	spin_unlock_irq(&dev->power.lock);
1701 }
1702 
1703 static bool pm_runtime_need_not_resume(struct device *dev)
1704 {
1705 	return atomic_read(&dev->power.usage_count) <= 1 &&
1706 		(atomic_read(&dev->power.child_count) == 0 ||
1707 		 dev->power.ignore_children);
1708 }
1709 
1710 /**
1711  * pm_runtime_force_suspend - Force a device into suspend state if needed.
1712  * @dev: Device to suspend.
1713  *
1714  * Disable runtime PM so we safely can check the device's runtime PM status and
1715  * if it is active, invoke its ->runtime_suspend callback to suspend it and
1716  * change its runtime PM status field to RPM_SUSPENDED.  Also, if the device's
1717  * usage and children counters don't indicate that the device was in use before
1718  * the system-wide transition under way, decrement its parent's children counter
1719  * (if there is a parent).  Keep runtime PM disabled to preserve the state
1720  * unless we encounter errors.
1721  *
1722  * Typically this function may be invoked from a system suspend callback to make
1723  * sure the device is put into low power state and it should only be used during
1724  * system-wide PM transitions to sleep states.  It assumes that the analogous
1725  * pm_runtime_force_resume() will be used to resume the device.
1726  */
1727 int pm_runtime_force_suspend(struct device *dev)
1728 {
1729 	int (*callback)(struct device *);
1730 	int ret;
1731 
1732 	pm_runtime_disable(dev);
1733 	if (pm_runtime_status_suspended(dev))
1734 		return 0;
1735 
1736 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1737 
1738 	ret = callback ? callback(dev) : 0;
1739 	if (ret)
1740 		goto err;
1741 
1742 	/*
1743 	 * If the device can stay in suspend after the system-wide transition
1744 	 * to the working state that will follow, drop the children counter of
1745 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1746 	 * function will be called again for it in the meantime.
1747 	 */
1748 	if (pm_runtime_need_not_resume(dev))
1749 		pm_runtime_set_suspended(dev);
1750 	else
1751 		__update_runtime_status(dev, RPM_SUSPENDED);
1752 
1753 	return 0;
1754 
1755 err:
1756 	pm_runtime_enable(dev);
1757 	return ret;
1758 }
1759 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1760 
1761 /**
1762  * pm_runtime_force_resume - Force a device into resume state if needed.
1763  * @dev: Device to resume.
1764  *
1765  * Prior invoking this function we expect the user to have brought the device
1766  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1767  * those actions and bring the device into full power, if it is expected to be
1768  * used on system resume.  In the other case, we defer the resume to be managed
1769  * via runtime PM.
1770  *
1771  * Typically this function may be invoked from a system resume callback.
1772  */
1773 int pm_runtime_force_resume(struct device *dev)
1774 {
1775 	int (*callback)(struct device *);
1776 	int ret = 0;
1777 
1778 	if (!pm_runtime_status_suspended(dev) || pm_runtime_need_not_resume(dev))
1779 		goto out;
1780 
1781 	/*
1782 	 * The value of the parent's children counter is correct already, so
1783 	 * just update the status of the device.
1784 	 */
1785 	__update_runtime_status(dev, RPM_ACTIVE);
1786 
1787 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1788 
1789 	ret = callback ? callback(dev) : 0;
1790 	if (ret) {
1791 		pm_runtime_set_suspended(dev);
1792 		goto out;
1793 	}
1794 
1795 	pm_runtime_mark_last_busy(dev);
1796 out:
1797 	pm_runtime_enable(dev);
1798 	return ret;
1799 }
1800 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1801