xref: /openbmc/linux/drivers/base/power/runtime.c (revision 07d9a767)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * drivers/base/power/runtime.c - Helper functions for device runtime PM
4  *
5  * Copyright (c) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
6  * Copyright (C) 2010 Alan Stern <stern@rowland.harvard.edu>
7  */
8 #include <linux/sched/mm.h>
9 #include <linux/ktime.h>
10 #include <linux/hrtimer.h>
11 #include <linux/export.h>
12 #include <linux/pm_runtime.h>
13 #include <linux/pm_wakeirq.h>
14 #include <trace/events/rpm.h>
15 
16 #include "../base.h"
17 #include "power.h"
18 
19 typedef int (*pm_callback_t)(struct device *);
20 
21 static pm_callback_t __rpm_get_callback(struct device *dev, size_t cb_offset)
22 {
23 	pm_callback_t cb;
24 	const struct dev_pm_ops *ops;
25 
26 	if (dev->pm_domain)
27 		ops = &dev->pm_domain->ops;
28 	else if (dev->type && dev->type->pm)
29 		ops = dev->type->pm;
30 	else if (dev->class && dev->class->pm)
31 		ops = dev->class->pm;
32 	else if (dev->bus && dev->bus->pm)
33 		ops = dev->bus->pm;
34 	else
35 		ops = NULL;
36 
37 	if (ops)
38 		cb = *(pm_callback_t *)((void *)ops + cb_offset);
39 	else
40 		cb = NULL;
41 
42 	if (!cb && dev->driver && dev->driver->pm)
43 		cb = *(pm_callback_t *)((void *)dev->driver->pm + cb_offset);
44 
45 	return cb;
46 }
47 
48 #define RPM_GET_CALLBACK(dev, callback) \
49 		__rpm_get_callback(dev, offsetof(struct dev_pm_ops, callback))
50 
51 static int rpm_resume(struct device *dev, int rpmflags);
52 static int rpm_suspend(struct device *dev, int rpmflags);
53 
54 /**
55  * update_pm_runtime_accounting - Update the time accounting of power states
56  * @dev: Device to update the accounting for
57  *
58  * In order to be able to have time accounting of the various power states
59  * (as used by programs such as PowerTOP to show the effectiveness of runtime
60  * PM), we need to track the time spent in each state.
61  * update_pm_runtime_accounting must be called each time before the
62  * runtime_status field is updated, to account the time in the old state
63  * correctly.
64  */
65 static void update_pm_runtime_accounting(struct device *dev)
66 {
67 	u64 now, last, delta;
68 
69 	if (dev->power.disable_depth > 0)
70 		return;
71 
72 	last = dev->power.accounting_timestamp;
73 
74 	now = ktime_get_mono_fast_ns();
75 	dev->power.accounting_timestamp = now;
76 
77 	/*
78 	 * Because ktime_get_mono_fast_ns() is not monotonic during
79 	 * timekeeping updates, ensure that 'now' is after the last saved
80 	 * timesptamp.
81 	 */
82 	if (now < last)
83 		return;
84 
85 	delta = now - last;
86 
87 	if (dev->power.runtime_status == RPM_SUSPENDED)
88 		dev->power.suspended_time += delta;
89 	else
90 		dev->power.active_time += delta;
91 }
92 
93 static void __update_runtime_status(struct device *dev, enum rpm_status status)
94 {
95 	update_pm_runtime_accounting(dev);
96 	dev->power.runtime_status = status;
97 }
98 
99 static u64 rpm_get_accounted_time(struct device *dev, bool suspended)
100 {
101 	u64 time;
102 	unsigned long flags;
103 
104 	spin_lock_irqsave(&dev->power.lock, flags);
105 
106 	update_pm_runtime_accounting(dev);
107 	time = suspended ? dev->power.suspended_time : dev->power.active_time;
108 
109 	spin_unlock_irqrestore(&dev->power.lock, flags);
110 
111 	return time;
112 }
113 
114 u64 pm_runtime_active_time(struct device *dev)
115 {
116 	return rpm_get_accounted_time(dev, false);
117 }
118 
119 u64 pm_runtime_suspended_time(struct device *dev)
120 {
121 	return rpm_get_accounted_time(dev, true);
122 }
123 EXPORT_SYMBOL_GPL(pm_runtime_suspended_time);
124 
125 /**
126  * pm_runtime_deactivate_timer - Deactivate given device's suspend timer.
127  * @dev: Device to handle.
128  */
129 static void pm_runtime_deactivate_timer(struct device *dev)
130 {
131 	if (dev->power.timer_expires > 0) {
132 		hrtimer_try_to_cancel(&dev->power.suspend_timer);
133 		dev->power.timer_expires = 0;
134 	}
135 }
136 
137 /**
138  * pm_runtime_cancel_pending - Deactivate suspend timer and cancel requests.
139  * @dev: Device to handle.
140  */
141 static void pm_runtime_cancel_pending(struct device *dev)
142 {
143 	pm_runtime_deactivate_timer(dev);
144 	/*
145 	 * In case there's a request pending, make sure its work function will
146 	 * return without doing anything.
147 	 */
148 	dev->power.request = RPM_REQ_NONE;
149 }
150 
151 /*
152  * pm_runtime_autosuspend_expiration - Get a device's autosuspend-delay expiration time.
153  * @dev: Device to handle.
154  *
155  * Compute the autosuspend-delay expiration time based on the device's
156  * power.last_busy time.  If the delay has already expired or is disabled
157  * (negative) or the power.use_autosuspend flag isn't set, return 0.
158  * Otherwise return the expiration time in nanoseconds (adjusted to be nonzero).
159  *
160  * This function may be called either with or without dev->power.lock held.
161  * Either way it can be racy, since power.last_busy may be updated at any time.
162  */
163 u64 pm_runtime_autosuspend_expiration(struct device *dev)
164 {
165 	int autosuspend_delay;
166 	u64 expires;
167 
168 	if (!dev->power.use_autosuspend)
169 		return 0;
170 
171 	autosuspend_delay = READ_ONCE(dev->power.autosuspend_delay);
172 	if (autosuspend_delay < 0)
173 		return 0;
174 
175 	expires  = READ_ONCE(dev->power.last_busy);
176 	expires += (u64)autosuspend_delay * NSEC_PER_MSEC;
177 	if (expires > ktime_get_mono_fast_ns())
178 		return expires;	/* Expires in the future */
179 
180 	return 0;
181 }
182 EXPORT_SYMBOL_GPL(pm_runtime_autosuspend_expiration);
183 
184 static int dev_memalloc_noio(struct device *dev, void *data)
185 {
186 	return dev->power.memalloc_noio;
187 }
188 
189 /*
190  * pm_runtime_set_memalloc_noio - Set a device's memalloc_noio flag.
191  * @dev: Device to handle.
192  * @enable: True for setting the flag and False for clearing the flag.
193  *
194  * Set the flag for all devices in the path from the device to the
195  * root device in the device tree if @enable is true, otherwise clear
196  * the flag for devices in the path whose siblings don't set the flag.
197  *
198  * The function should only be called by block device, or network
199  * device driver for solving the deadlock problem during runtime
200  * resume/suspend:
201  *
202  *     If memory allocation with GFP_KERNEL is called inside runtime
203  *     resume/suspend callback of any one of its ancestors(or the
204  *     block device itself), the deadlock may be triggered inside the
205  *     memory allocation since it might not complete until the block
206  *     device becomes active and the involed page I/O finishes. The
207  *     situation is pointed out first by Alan Stern. Network device
208  *     are involved in iSCSI kind of situation.
209  *
210  * The lock of dev_hotplug_mutex is held in the function for handling
211  * hotplug race because pm_runtime_set_memalloc_noio() may be called
212  * in async probe().
213  *
214  * The function should be called between device_add() and device_del()
215  * on the affected device(block/network device).
216  */
217 void pm_runtime_set_memalloc_noio(struct device *dev, bool enable)
218 {
219 	static DEFINE_MUTEX(dev_hotplug_mutex);
220 
221 	mutex_lock(&dev_hotplug_mutex);
222 	for (;;) {
223 		bool enabled;
224 
225 		/* hold power lock since bitfield is not SMP-safe. */
226 		spin_lock_irq(&dev->power.lock);
227 		enabled = dev->power.memalloc_noio;
228 		dev->power.memalloc_noio = enable;
229 		spin_unlock_irq(&dev->power.lock);
230 
231 		/*
232 		 * not need to enable ancestors any more if the device
233 		 * has been enabled.
234 		 */
235 		if (enabled && enable)
236 			break;
237 
238 		dev = dev->parent;
239 
240 		/*
241 		 * clear flag of the parent device only if all the
242 		 * children don't set the flag because ancestor's
243 		 * flag was set by any one of the descendants.
244 		 */
245 		if (!dev || (!enable &&
246 			     device_for_each_child(dev, NULL,
247 						   dev_memalloc_noio)))
248 			break;
249 	}
250 	mutex_unlock(&dev_hotplug_mutex);
251 }
252 EXPORT_SYMBOL_GPL(pm_runtime_set_memalloc_noio);
253 
254 /**
255  * rpm_check_suspend_allowed - Test whether a device may be suspended.
256  * @dev: Device to test.
257  */
258 static int rpm_check_suspend_allowed(struct device *dev)
259 {
260 	int retval = 0;
261 
262 	if (dev->power.runtime_error)
263 		retval = -EINVAL;
264 	else if (dev->power.disable_depth > 0)
265 		retval = -EACCES;
266 	else if (atomic_read(&dev->power.usage_count) > 0)
267 		retval = -EAGAIN;
268 	else if (!dev->power.ignore_children &&
269 			atomic_read(&dev->power.child_count))
270 		retval = -EBUSY;
271 
272 	/* Pending resume requests take precedence over suspends. */
273 	else if ((dev->power.deferred_resume
274 			&& dev->power.runtime_status == RPM_SUSPENDING)
275 	    || (dev->power.request_pending
276 			&& dev->power.request == RPM_REQ_RESUME))
277 		retval = -EAGAIN;
278 	else if (__dev_pm_qos_resume_latency(dev) == 0)
279 		retval = -EPERM;
280 	else if (dev->power.runtime_status == RPM_SUSPENDED)
281 		retval = 1;
282 
283 	return retval;
284 }
285 
286 static int rpm_get_suppliers(struct device *dev)
287 {
288 	struct device_link *link;
289 
290 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
291 				device_links_read_lock_held()) {
292 		int retval;
293 
294 		if (!(link->flags & DL_FLAG_PM_RUNTIME))
295 			continue;
296 
297 		retval = pm_runtime_get_sync(link->supplier);
298 		/* Ignore suppliers with disabled runtime PM. */
299 		if (retval < 0 && retval != -EACCES) {
300 			pm_runtime_put_noidle(link->supplier);
301 			return retval;
302 		}
303 		refcount_inc(&link->rpm_active);
304 	}
305 	return 0;
306 }
307 
308 static void rpm_put_suppliers(struct device *dev)
309 {
310 	struct device_link *link;
311 
312 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
313 				device_links_read_lock_held()) {
314 
315 		while (refcount_dec_not_one(&link->rpm_active))
316 			pm_runtime_put(link->supplier);
317 	}
318 }
319 
320 /**
321  * __rpm_callback - Run a given runtime PM callback for a given device.
322  * @cb: Runtime PM callback to run.
323  * @dev: Device to run the callback for.
324  */
325 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
326 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
327 {
328 	int retval, idx;
329 	bool use_links = dev->power.links_count > 0;
330 
331 	if (dev->power.irq_safe) {
332 		spin_unlock(&dev->power.lock);
333 	} else {
334 		spin_unlock_irq(&dev->power.lock);
335 
336 		/*
337 		 * Resume suppliers if necessary.
338 		 *
339 		 * The device's runtime PM status cannot change until this
340 		 * routine returns, so it is safe to read the status outside of
341 		 * the lock.
342 		 */
343 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
344 			idx = device_links_read_lock();
345 
346 			retval = rpm_get_suppliers(dev);
347 			if (retval)
348 				goto fail;
349 
350 			device_links_read_unlock(idx);
351 		}
352 	}
353 
354 	retval = cb(dev);
355 
356 	if (dev->power.irq_safe) {
357 		spin_lock(&dev->power.lock);
358 	} else {
359 		/*
360 		 * If the device is suspending and the callback has returned
361 		 * success, drop the usage counters of the suppliers that have
362 		 * been reference counted on its resume.
363 		 *
364 		 * Do that if resume fails too.
365 		 */
366 		if (use_links
367 		    && ((dev->power.runtime_status == RPM_SUSPENDING && !retval)
368 		    || (dev->power.runtime_status == RPM_RESUMING && retval))) {
369 			idx = device_links_read_lock();
370 
371  fail:
372 			rpm_put_suppliers(dev);
373 
374 			device_links_read_unlock(idx);
375 		}
376 
377 		spin_lock_irq(&dev->power.lock);
378 	}
379 
380 	return retval;
381 }
382 
383 /**
384  * rpm_idle - Notify device bus type if the device can be suspended.
385  * @dev: Device to notify the bus type about.
386  * @rpmflags: Flag bits.
387  *
388  * Check if the device's runtime PM status allows it to be suspended.  If
389  * another idle notification has been started earlier, return immediately.  If
390  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
391  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
392  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
393  *
394  * This function must be called under dev->power.lock with interrupts disabled.
395  */
396 static int rpm_idle(struct device *dev, int rpmflags)
397 {
398 	int (*callback)(struct device *);
399 	int retval;
400 
401 	trace_rpm_idle_rcuidle(dev, rpmflags);
402 	retval = rpm_check_suspend_allowed(dev);
403 	if (retval < 0)
404 		;	/* Conditions are wrong. */
405 
406 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
407 	else if (dev->power.runtime_status != RPM_ACTIVE)
408 		retval = -EAGAIN;
409 
410 	/*
411 	 * Any pending request other than an idle notification takes
412 	 * precedence over us, except that the timer may be running.
413 	 */
414 	else if (dev->power.request_pending &&
415 	    dev->power.request > RPM_REQ_IDLE)
416 		retval = -EAGAIN;
417 
418 	/* Act as though RPM_NOWAIT is always set. */
419 	else if (dev->power.idle_notification)
420 		retval = -EINPROGRESS;
421 	if (retval)
422 		goto out;
423 
424 	/* Pending requests need to be canceled. */
425 	dev->power.request = RPM_REQ_NONE;
426 
427 	if (dev->power.no_callbacks)
428 		goto out;
429 
430 	/* Carry out an asynchronous or a synchronous idle notification. */
431 	if (rpmflags & RPM_ASYNC) {
432 		dev->power.request = RPM_REQ_IDLE;
433 		if (!dev->power.request_pending) {
434 			dev->power.request_pending = true;
435 			queue_work(pm_wq, &dev->power.work);
436 		}
437 		trace_rpm_return_int_rcuidle(dev, _THIS_IP_, 0);
438 		return 0;
439 	}
440 
441 	dev->power.idle_notification = true;
442 
443 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
444 
445 	if (callback)
446 		retval = __rpm_callback(callback, dev);
447 
448 	dev->power.idle_notification = false;
449 	wake_up_all(&dev->power.wait_queue);
450 
451  out:
452 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
453 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
454 }
455 
456 /**
457  * rpm_callback - Run a given runtime PM callback for a given device.
458  * @cb: Runtime PM callback to run.
459  * @dev: Device to run the callback for.
460  */
461 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
462 {
463 	int retval;
464 
465 	if (!cb)
466 		return -ENOSYS;
467 
468 	if (dev->power.memalloc_noio) {
469 		unsigned int noio_flag;
470 
471 		/*
472 		 * Deadlock might be caused if memory allocation with
473 		 * GFP_KERNEL happens inside runtime_suspend and
474 		 * runtime_resume callbacks of one block device's
475 		 * ancestor or the block device itself. Network
476 		 * device might be thought as part of iSCSI block
477 		 * device, so network device and its ancestor should
478 		 * be marked as memalloc_noio too.
479 		 */
480 		noio_flag = memalloc_noio_save();
481 		retval = __rpm_callback(cb, dev);
482 		memalloc_noio_restore(noio_flag);
483 	} else {
484 		retval = __rpm_callback(cb, dev);
485 	}
486 
487 	dev->power.runtime_error = retval;
488 	return retval != -EACCES ? retval : -EIO;
489 }
490 
491 /**
492  * rpm_suspend - Carry out runtime suspend of given device.
493  * @dev: Device to suspend.
494  * @rpmflags: Flag bits.
495  *
496  * Check if the device's runtime PM status allows it to be suspended.
497  * Cancel a pending idle notification, autosuspend or suspend. If
498  * another suspend has been started earlier, either return immediately
499  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
500  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
501  * otherwise run the ->runtime_suspend() callback directly. When
502  * ->runtime_suspend succeeded, if a deferred resume was requested while
503  * the callback was running then carry it out, otherwise send an idle
504  * notification for its parent (if the suspend succeeded and both
505  * ignore_children of parent->power and irq_safe of dev->power are not set).
506  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
507  * flag is set and the next autosuspend-delay expiration time is in the
508  * future, schedule another autosuspend attempt.
509  *
510  * This function must be called under dev->power.lock with interrupts disabled.
511  */
512 static int rpm_suspend(struct device *dev, int rpmflags)
513 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
514 {
515 	int (*callback)(struct device *);
516 	struct device *parent = NULL;
517 	int retval;
518 
519 	trace_rpm_suspend_rcuidle(dev, rpmflags);
520 
521  repeat:
522 	retval = rpm_check_suspend_allowed(dev);
523 	if (retval < 0)
524 		goto out;	/* Conditions are wrong. */
525 
526 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
527 	if (dev->power.runtime_status == RPM_RESUMING && !(rpmflags & RPM_ASYNC))
528 		retval = -EAGAIN;
529 	if (retval)
530 		goto out;
531 
532 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
533 	if ((rpmflags & RPM_AUTO)
534 	    && dev->power.runtime_status != RPM_SUSPENDING) {
535 		u64 expires = pm_runtime_autosuspend_expiration(dev);
536 
537 		if (expires != 0) {
538 			/* Pending requests need to be canceled. */
539 			dev->power.request = RPM_REQ_NONE;
540 
541 			/*
542 			 * Optimization: If the timer is already running and is
543 			 * set to expire at or before the autosuspend delay,
544 			 * avoid the overhead of resetting it.  Just let it
545 			 * expire; pm_suspend_timer_fn() will take care of the
546 			 * rest.
547 			 */
548 			if (!(dev->power.timer_expires &&
549 					dev->power.timer_expires <= expires)) {
550 				/*
551 				 * We add a slack of 25% to gather wakeups
552 				 * without sacrificing the granularity.
553 				 */
554 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
555 						    (NSEC_PER_MSEC >> 2);
556 
557 				dev->power.timer_expires = expires;
558 				hrtimer_start_range_ns(&dev->power.suspend_timer,
559 						ns_to_ktime(expires),
560 						slack,
561 						HRTIMER_MODE_ABS);
562 			}
563 			dev->power.timer_autosuspends = 1;
564 			goto out;
565 		}
566 	}
567 
568 	/* Other scheduled or pending requests need to be canceled. */
569 	pm_runtime_cancel_pending(dev);
570 
571 	if (dev->power.runtime_status == RPM_SUSPENDING) {
572 		DEFINE_WAIT(wait);
573 
574 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
575 			retval = -EINPROGRESS;
576 			goto out;
577 		}
578 
579 		if (dev->power.irq_safe) {
580 			spin_unlock(&dev->power.lock);
581 
582 			cpu_relax();
583 
584 			spin_lock(&dev->power.lock);
585 			goto repeat;
586 		}
587 
588 		/* Wait for the other suspend running in parallel with us. */
589 		for (;;) {
590 			prepare_to_wait(&dev->power.wait_queue, &wait,
591 					TASK_UNINTERRUPTIBLE);
592 			if (dev->power.runtime_status != RPM_SUSPENDING)
593 				break;
594 
595 			spin_unlock_irq(&dev->power.lock);
596 
597 			schedule();
598 
599 			spin_lock_irq(&dev->power.lock);
600 		}
601 		finish_wait(&dev->power.wait_queue, &wait);
602 		goto repeat;
603 	}
604 
605 	if (dev->power.no_callbacks)
606 		goto no_callback;	/* Assume success. */
607 
608 	/* Carry out an asynchronous or a synchronous suspend. */
609 	if (rpmflags & RPM_ASYNC) {
610 		dev->power.request = (rpmflags & RPM_AUTO) ?
611 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
612 		if (!dev->power.request_pending) {
613 			dev->power.request_pending = true;
614 			queue_work(pm_wq, &dev->power.work);
615 		}
616 		goto out;
617 	}
618 
619 	__update_runtime_status(dev, RPM_SUSPENDING);
620 
621 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
622 
623 	dev_pm_enable_wake_irq_check(dev, true);
624 	retval = rpm_callback(callback, dev);
625 	if (retval)
626 		goto fail;
627 
628  no_callback:
629 	__update_runtime_status(dev, RPM_SUSPENDED);
630 	pm_runtime_deactivate_timer(dev);
631 
632 	if (dev->parent) {
633 		parent = dev->parent;
634 		atomic_add_unless(&parent->power.child_count, -1, 0);
635 	}
636 	wake_up_all(&dev->power.wait_queue);
637 
638 	if (dev->power.deferred_resume) {
639 		dev->power.deferred_resume = false;
640 		rpm_resume(dev, 0);
641 		retval = -EAGAIN;
642 		goto out;
643 	}
644 
645 	/* Maybe the parent is now able to suspend. */
646 	if (parent && !parent->power.ignore_children && !dev->power.irq_safe) {
647 		spin_unlock(&dev->power.lock);
648 
649 		spin_lock(&parent->power.lock);
650 		rpm_idle(parent, RPM_ASYNC);
651 		spin_unlock(&parent->power.lock);
652 
653 		spin_lock(&dev->power.lock);
654 	}
655 
656  out:
657 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
658 
659 	return retval;
660 
661  fail:
662 	dev_pm_disable_wake_irq_check(dev);
663 	__update_runtime_status(dev, RPM_ACTIVE);
664 	dev->power.deferred_resume = false;
665 	wake_up_all(&dev->power.wait_queue);
666 
667 	if (retval == -EAGAIN || retval == -EBUSY) {
668 		dev->power.runtime_error = 0;
669 
670 		/*
671 		 * If the callback routine failed an autosuspend, and
672 		 * if the last_busy time has been updated so that there
673 		 * is a new autosuspend expiration time, automatically
674 		 * reschedule another autosuspend.
675 		 */
676 		if ((rpmflags & RPM_AUTO) &&
677 		    pm_runtime_autosuspend_expiration(dev) != 0)
678 			goto repeat;
679 	} else {
680 		pm_runtime_cancel_pending(dev);
681 	}
682 	goto out;
683 }
684 
685 /**
686  * rpm_resume - Carry out runtime resume of given device.
687  * @dev: Device to resume.
688  * @rpmflags: Flag bits.
689  *
690  * Check if the device's runtime PM status allows it to be resumed.  Cancel
691  * any scheduled or pending requests.  If another resume has been started
692  * earlier, either return immediately or wait for it to finish, depending on the
693  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
694  * parallel with this function, either tell the other process to resume after
695  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
696  * flag is set then queue a resume request; otherwise run the
697  * ->runtime_resume() callback directly.  Queue an idle notification for the
698  * device if the resume succeeded.
699  *
700  * This function must be called under dev->power.lock with interrupts disabled.
701  */
702 static int rpm_resume(struct device *dev, int rpmflags)
703 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
704 {
705 	int (*callback)(struct device *);
706 	struct device *parent = NULL;
707 	int retval = 0;
708 
709 	trace_rpm_resume_rcuidle(dev, rpmflags);
710 
711  repeat:
712 	if (dev->power.runtime_error)
713 		retval = -EINVAL;
714 	else if (dev->power.disable_depth == 1 && dev->power.is_suspended
715 	    && dev->power.runtime_status == RPM_ACTIVE)
716 		retval = 1;
717 	else if (dev->power.disable_depth > 0)
718 		retval = -EACCES;
719 	if (retval)
720 		goto out;
721 
722 	/*
723 	 * Other scheduled or pending requests need to be canceled.  Small
724 	 * optimization: If an autosuspend timer is running, leave it running
725 	 * rather than cancelling it now only to restart it again in the near
726 	 * future.
727 	 */
728 	dev->power.request = RPM_REQ_NONE;
729 	if (!dev->power.timer_autosuspends)
730 		pm_runtime_deactivate_timer(dev);
731 
732 	if (dev->power.runtime_status == RPM_ACTIVE) {
733 		retval = 1;
734 		goto out;
735 	}
736 
737 	if (dev->power.runtime_status == RPM_RESUMING
738 	    || dev->power.runtime_status == RPM_SUSPENDING) {
739 		DEFINE_WAIT(wait);
740 
741 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
742 			if (dev->power.runtime_status == RPM_SUSPENDING)
743 				dev->power.deferred_resume = true;
744 			else
745 				retval = -EINPROGRESS;
746 			goto out;
747 		}
748 
749 		if (dev->power.irq_safe) {
750 			spin_unlock(&dev->power.lock);
751 
752 			cpu_relax();
753 
754 			spin_lock(&dev->power.lock);
755 			goto repeat;
756 		}
757 
758 		/* Wait for the operation carried out in parallel with us. */
759 		for (;;) {
760 			prepare_to_wait(&dev->power.wait_queue, &wait,
761 					TASK_UNINTERRUPTIBLE);
762 			if (dev->power.runtime_status != RPM_RESUMING
763 			    && dev->power.runtime_status != RPM_SUSPENDING)
764 				break;
765 
766 			spin_unlock_irq(&dev->power.lock);
767 
768 			schedule();
769 
770 			spin_lock_irq(&dev->power.lock);
771 		}
772 		finish_wait(&dev->power.wait_queue, &wait);
773 		goto repeat;
774 	}
775 
776 	/*
777 	 * See if we can skip waking up the parent.  This is safe only if
778 	 * power.no_callbacks is set, because otherwise we don't know whether
779 	 * the resume will actually succeed.
780 	 */
781 	if (dev->power.no_callbacks && !parent && dev->parent) {
782 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
783 		if (dev->parent->power.disable_depth > 0
784 		    || dev->parent->power.ignore_children
785 		    || dev->parent->power.runtime_status == RPM_ACTIVE) {
786 			atomic_inc(&dev->parent->power.child_count);
787 			spin_unlock(&dev->parent->power.lock);
788 			retval = 1;
789 			goto no_callback;	/* Assume success. */
790 		}
791 		spin_unlock(&dev->parent->power.lock);
792 	}
793 
794 	/* Carry out an asynchronous or a synchronous resume. */
795 	if (rpmflags & RPM_ASYNC) {
796 		dev->power.request = RPM_REQ_RESUME;
797 		if (!dev->power.request_pending) {
798 			dev->power.request_pending = true;
799 			queue_work(pm_wq, &dev->power.work);
800 		}
801 		retval = 0;
802 		goto out;
803 	}
804 
805 	if (!parent && dev->parent) {
806 		/*
807 		 * Increment the parent's usage counter and resume it if
808 		 * necessary.  Not needed if dev is irq-safe; then the
809 		 * parent is permanently resumed.
810 		 */
811 		parent = dev->parent;
812 		if (dev->power.irq_safe)
813 			goto skip_parent;
814 		spin_unlock(&dev->power.lock);
815 
816 		pm_runtime_get_noresume(parent);
817 
818 		spin_lock(&parent->power.lock);
819 		/*
820 		 * Resume the parent if it has runtime PM enabled and not been
821 		 * set to ignore its children.
822 		 */
823 		if (!parent->power.disable_depth
824 		    && !parent->power.ignore_children) {
825 			rpm_resume(parent, 0);
826 			if (parent->power.runtime_status != RPM_ACTIVE)
827 				retval = -EBUSY;
828 		}
829 		spin_unlock(&parent->power.lock);
830 
831 		spin_lock(&dev->power.lock);
832 		if (retval)
833 			goto out;
834 		goto repeat;
835 	}
836  skip_parent:
837 
838 	if (dev->power.no_callbacks)
839 		goto no_callback;	/* Assume success. */
840 
841 	__update_runtime_status(dev, RPM_RESUMING);
842 
843 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
844 
845 	dev_pm_disable_wake_irq_check(dev);
846 	retval = rpm_callback(callback, dev);
847 	if (retval) {
848 		__update_runtime_status(dev, RPM_SUSPENDED);
849 		pm_runtime_cancel_pending(dev);
850 		dev_pm_enable_wake_irq_check(dev, false);
851 	} else {
852  no_callback:
853 		__update_runtime_status(dev, RPM_ACTIVE);
854 		pm_runtime_mark_last_busy(dev);
855 		if (parent)
856 			atomic_inc(&parent->power.child_count);
857 	}
858 	wake_up_all(&dev->power.wait_queue);
859 
860 	if (retval >= 0)
861 		rpm_idle(dev, RPM_ASYNC);
862 
863  out:
864 	if (parent && !dev->power.irq_safe) {
865 		spin_unlock_irq(&dev->power.lock);
866 
867 		pm_runtime_put(parent);
868 
869 		spin_lock_irq(&dev->power.lock);
870 	}
871 
872 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
873 
874 	return retval;
875 }
876 
877 /**
878  * pm_runtime_work - Universal runtime PM work function.
879  * @work: Work structure used for scheduling the execution of this function.
880  *
881  * Use @work to get the device object the work is to be done for, determine what
882  * is to be done and execute the appropriate runtime PM function.
883  */
884 static void pm_runtime_work(struct work_struct *work)
885 {
886 	struct device *dev = container_of(work, struct device, power.work);
887 	enum rpm_request req;
888 
889 	spin_lock_irq(&dev->power.lock);
890 
891 	if (!dev->power.request_pending)
892 		goto out;
893 
894 	req = dev->power.request;
895 	dev->power.request = RPM_REQ_NONE;
896 	dev->power.request_pending = false;
897 
898 	switch (req) {
899 	case RPM_REQ_NONE:
900 		break;
901 	case RPM_REQ_IDLE:
902 		rpm_idle(dev, RPM_NOWAIT);
903 		break;
904 	case RPM_REQ_SUSPEND:
905 		rpm_suspend(dev, RPM_NOWAIT);
906 		break;
907 	case RPM_REQ_AUTOSUSPEND:
908 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
909 		break;
910 	case RPM_REQ_RESUME:
911 		rpm_resume(dev, RPM_NOWAIT);
912 		break;
913 	}
914 
915  out:
916 	spin_unlock_irq(&dev->power.lock);
917 }
918 
919 /**
920  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
921  * @data: Device pointer passed by pm_schedule_suspend().
922  *
923  * Check if the time is right and queue a suspend request.
924  */
925 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
926 {
927 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
928 	unsigned long flags;
929 	u64 expires;
930 
931 	spin_lock_irqsave(&dev->power.lock, flags);
932 
933 	expires = dev->power.timer_expires;
934 	/*
935 	 * If 'expires' is after the current time, we've been called
936 	 * too early.
937 	 */
938 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
939 		dev->power.timer_expires = 0;
940 		rpm_suspend(dev, dev->power.timer_autosuspends ?
941 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
942 	}
943 
944 	spin_unlock_irqrestore(&dev->power.lock, flags);
945 
946 	return HRTIMER_NORESTART;
947 }
948 
949 /**
950  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
951  * @dev: Device to suspend.
952  * @delay: Time to wait before submitting a suspend request, in milliseconds.
953  */
954 int pm_schedule_suspend(struct device *dev, unsigned int delay)
955 {
956 	unsigned long flags;
957 	u64 expires;
958 	int retval;
959 
960 	spin_lock_irqsave(&dev->power.lock, flags);
961 
962 	if (!delay) {
963 		retval = rpm_suspend(dev, RPM_ASYNC);
964 		goto out;
965 	}
966 
967 	retval = rpm_check_suspend_allowed(dev);
968 	if (retval)
969 		goto out;
970 
971 	/* Other scheduled or pending requests need to be canceled. */
972 	pm_runtime_cancel_pending(dev);
973 
974 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
975 	dev->power.timer_expires = expires;
976 	dev->power.timer_autosuspends = 0;
977 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
978 
979  out:
980 	spin_unlock_irqrestore(&dev->power.lock, flags);
981 
982 	return retval;
983 }
984 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
985 
986 /**
987  * __pm_runtime_idle - Entry point for runtime idle operations.
988  * @dev: Device to send idle notification for.
989  * @rpmflags: Flag bits.
990  *
991  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
992  * return immediately if it is larger than zero.  Then carry out an idle
993  * notification, either synchronous or asynchronous.
994  *
995  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
996  * or if pm_runtime_irq_safe() has been called.
997  */
998 int __pm_runtime_idle(struct device *dev, int rpmflags)
999 {
1000 	unsigned long flags;
1001 	int retval;
1002 
1003 	if (rpmflags & RPM_GET_PUT) {
1004 		if (!atomic_dec_and_test(&dev->power.usage_count)) {
1005 			trace_rpm_usage_rcuidle(dev, rpmflags);
1006 			return 0;
1007 		}
1008 	}
1009 
1010 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1011 
1012 	spin_lock_irqsave(&dev->power.lock, flags);
1013 	retval = rpm_idle(dev, rpmflags);
1014 	spin_unlock_irqrestore(&dev->power.lock, flags);
1015 
1016 	return retval;
1017 }
1018 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1019 
1020 /**
1021  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1022  * @dev: Device to suspend.
1023  * @rpmflags: Flag bits.
1024  *
1025  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1026  * return immediately if it is larger than zero.  Then carry out a suspend,
1027  * either synchronous or asynchronous.
1028  *
1029  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1030  * or if pm_runtime_irq_safe() has been called.
1031  */
1032 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1033 {
1034 	unsigned long flags;
1035 	int retval;
1036 
1037 	if (rpmflags & RPM_GET_PUT) {
1038 		if (!atomic_dec_and_test(&dev->power.usage_count)) {
1039 			trace_rpm_usage_rcuidle(dev, rpmflags);
1040 			return 0;
1041 		}
1042 	}
1043 
1044 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1045 
1046 	spin_lock_irqsave(&dev->power.lock, flags);
1047 	retval = rpm_suspend(dev, rpmflags);
1048 	spin_unlock_irqrestore(&dev->power.lock, flags);
1049 
1050 	return retval;
1051 }
1052 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1053 
1054 /**
1055  * __pm_runtime_resume - Entry point for runtime resume operations.
1056  * @dev: Device to resume.
1057  * @rpmflags: Flag bits.
1058  *
1059  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1060  * carry out a resume, either synchronous or asynchronous.
1061  *
1062  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1063  * or if pm_runtime_irq_safe() has been called.
1064  */
1065 int __pm_runtime_resume(struct device *dev, int rpmflags)
1066 {
1067 	unsigned long flags;
1068 	int retval;
1069 
1070 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1071 			dev->power.runtime_status != RPM_ACTIVE);
1072 
1073 	if (rpmflags & RPM_GET_PUT)
1074 		atomic_inc(&dev->power.usage_count);
1075 
1076 	spin_lock_irqsave(&dev->power.lock, flags);
1077 	retval = rpm_resume(dev, rpmflags);
1078 	spin_unlock_irqrestore(&dev->power.lock, flags);
1079 
1080 	return retval;
1081 }
1082 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1083 
1084 /**
1085  * pm_runtime_get_if_active - Conditionally bump up device usage counter.
1086  * @dev: Device to handle.
1087  * @ign_usage_count: Whether or not to look at the current usage counter value.
1088  *
1089  * Return -EINVAL if runtime PM is disabled for @dev.
1090  *
1091  * Otherwise, if the runtime PM status of @dev is %RPM_ACTIVE and either
1092  * @ign_usage_count is %true or the runtime PM usage counter of @dev is not
1093  * zero, increment the usage counter of @dev and return 1. Otherwise, return 0
1094  * without changing the usage counter.
1095  *
1096  * If @ign_usage_count is %true, this function can be used to prevent suspending
1097  * the device when its runtime PM status is %RPM_ACTIVE.
1098  *
1099  * If @ign_usage_count is %false, this function can be used to prevent
1100  * suspending the device when both its runtime PM status is %RPM_ACTIVE and its
1101  * runtime PM usage counter is not zero.
1102  *
1103  * The caller is resposible for decrementing the runtime PM usage counter of
1104  * @dev after this function has returned a positive value for it.
1105  */
1106 int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count)
1107 {
1108 	unsigned long flags;
1109 	int retval;
1110 
1111 	spin_lock_irqsave(&dev->power.lock, flags);
1112 	if (dev->power.disable_depth > 0) {
1113 		retval = -EINVAL;
1114 	} else if (dev->power.runtime_status != RPM_ACTIVE) {
1115 		retval = 0;
1116 	} else if (ign_usage_count) {
1117 		retval = 1;
1118 		atomic_inc(&dev->power.usage_count);
1119 	} else {
1120 		retval = atomic_inc_not_zero(&dev->power.usage_count);
1121 	}
1122 	trace_rpm_usage_rcuidle(dev, 0);
1123 	spin_unlock_irqrestore(&dev->power.lock, flags);
1124 
1125 	return retval;
1126 }
1127 EXPORT_SYMBOL_GPL(pm_runtime_get_if_active);
1128 
1129 /**
1130  * __pm_runtime_set_status - Set runtime PM status of a device.
1131  * @dev: Device to handle.
1132  * @status: New runtime PM status of the device.
1133  *
1134  * If runtime PM of the device is disabled or its power.runtime_error field is
1135  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1136  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1137  * However, if the device has a parent and the parent is not active, and the
1138  * parent's power.ignore_children flag is unset, the device's status cannot be
1139  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1140  *
1141  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1142  * and the device parent's counter of unsuspended children is modified to
1143  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1144  * notification request for the parent is submitted.
1145  *
1146  * If @dev has any suppliers (as reflected by device links to them), and @status
1147  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1148  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1149  * of the @status value) and the suppliers will be deacticated on exit.  The
1150  * error returned by the failing supplier activation will be returned in that
1151  * case.
1152  */
1153 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1154 {
1155 	struct device *parent = dev->parent;
1156 	bool notify_parent = false;
1157 	int error = 0;
1158 
1159 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1160 		return -EINVAL;
1161 
1162 	spin_lock_irq(&dev->power.lock);
1163 
1164 	/*
1165 	 * Prevent PM-runtime from being enabled for the device or return an
1166 	 * error if it is enabled already and working.
1167 	 */
1168 	if (dev->power.runtime_error || dev->power.disable_depth)
1169 		dev->power.disable_depth++;
1170 	else
1171 		error = -EAGAIN;
1172 
1173 	spin_unlock_irq(&dev->power.lock);
1174 
1175 	if (error)
1176 		return error;
1177 
1178 	/*
1179 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1180 	 * upfront regardless of the current status, because next time
1181 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1182 	 * involved will be dropped down to one anyway.
1183 	 */
1184 	if (status == RPM_ACTIVE) {
1185 		int idx = device_links_read_lock();
1186 
1187 		error = rpm_get_suppliers(dev);
1188 		if (error)
1189 			status = RPM_SUSPENDED;
1190 
1191 		device_links_read_unlock(idx);
1192 	}
1193 
1194 	spin_lock_irq(&dev->power.lock);
1195 
1196 	if (dev->power.runtime_status == status || !parent)
1197 		goto out_set;
1198 
1199 	if (status == RPM_SUSPENDED) {
1200 		atomic_add_unless(&parent->power.child_count, -1, 0);
1201 		notify_parent = !parent->power.ignore_children;
1202 	} else {
1203 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1204 
1205 		/*
1206 		 * It is invalid to put an active child under a parent that is
1207 		 * not active, has runtime PM enabled and the
1208 		 * 'power.ignore_children' flag unset.
1209 		 */
1210 		if (!parent->power.disable_depth
1211 		    && !parent->power.ignore_children
1212 		    && parent->power.runtime_status != RPM_ACTIVE) {
1213 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1214 				dev_name(dev),
1215 				dev_name(parent));
1216 			error = -EBUSY;
1217 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1218 			atomic_inc(&parent->power.child_count);
1219 		}
1220 
1221 		spin_unlock(&parent->power.lock);
1222 
1223 		if (error) {
1224 			status = RPM_SUSPENDED;
1225 			goto out;
1226 		}
1227 	}
1228 
1229  out_set:
1230 	__update_runtime_status(dev, status);
1231 	if (!error)
1232 		dev->power.runtime_error = 0;
1233 
1234  out:
1235 	spin_unlock_irq(&dev->power.lock);
1236 
1237 	if (notify_parent)
1238 		pm_request_idle(parent);
1239 
1240 	if (status == RPM_SUSPENDED) {
1241 		int idx = device_links_read_lock();
1242 
1243 		rpm_put_suppliers(dev);
1244 
1245 		device_links_read_unlock(idx);
1246 	}
1247 
1248 	pm_runtime_enable(dev);
1249 
1250 	return error;
1251 }
1252 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1253 
1254 /**
1255  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1256  * @dev: Device to handle.
1257  *
1258  * Flush all pending requests for the device from pm_wq and wait for all
1259  * runtime PM operations involving the device in progress to complete.
1260  *
1261  * Should be called under dev->power.lock with interrupts disabled.
1262  */
1263 static void __pm_runtime_barrier(struct device *dev)
1264 {
1265 	pm_runtime_deactivate_timer(dev);
1266 
1267 	if (dev->power.request_pending) {
1268 		dev->power.request = RPM_REQ_NONE;
1269 		spin_unlock_irq(&dev->power.lock);
1270 
1271 		cancel_work_sync(&dev->power.work);
1272 
1273 		spin_lock_irq(&dev->power.lock);
1274 		dev->power.request_pending = false;
1275 	}
1276 
1277 	if (dev->power.runtime_status == RPM_SUSPENDING
1278 	    || dev->power.runtime_status == RPM_RESUMING
1279 	    || dev->power.idle_notification) {
1280 		DEFINE_WAIT(wait);
1281 
1282 		/* Suspend, wake-up or idle notification in progress. */
1283 		for (;;) {
1284 			prepare_to_wait(&dev->power.wait_queue, &wait,
1285 					TASK_UNINTERRUPTIBLE);
1286 			if (dev->power.runtime_status != RPM_SUSPENDING
1287 			    && dev->power.runtime_status != RPM_RESUMING
1288 			    && !dev->power.idle_notification)
1289 				break;
1290 			spin_unlock_irq(&dev->power.lock);
1291 
1292 			schedule();
1293 
1294 			spin_lock_irq(&dev->power.lock);
1295 		}
1296 		finish_wait(&dev->power.wait_queue, &wait);
1297 	}
1298 }
1299 
1300 /**
1301  * pm_runtime_barrier - Flush pending requests and wait for completions.
1302  * @dev: Device to handle.
1303  *
1304  * Prevent the device from being suspended by incrementing its usage counter and
1305  * if there's a pending resume request for the device, wake the device up.
1306  * Next, make sure that all pending requests for the device have been flushed
1307  * from pm_wq and wait for all runtime PM operations involving the device in
1308  * progress to complete.
1309  *
1310  * Return value:
1311  * 1, if there was a resume request pending and the device had to be woken up,
1312  * 0, otherwise
1313  */
1314 int pm_runtime_barrier(struct device *dev)
1315 {
1316 	int retval = 0;
1317 
1318 	pm_runtime_get_noresume(dev);
1319 	spin_lock_irq(&dev->power.lock);
1320 
1321 	if (dev->power.request_pending
1322 	    && dev->power.request == RPM_REQ_RESUME) {
1323 		rpm_resume(dev, 0);
1324 		retval = 1;
1325 	}
1326 
1327 	__pm_runtime_barrier(dev);
1328 
1329 	spin_unlock_irq(&dev->power.lock);
1330 	pm_runtime_put_noidle(dev);
1331 
1332 	return retval;
1333 }
1334 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1335 
1336 /**
1337  * __pm_runtime_disable - Disable runtime PM of a device.
1338  * @dev: Device to handle.
1339  * @check_resume: If set, check if there's a resume request for the device.
1340  *
1341  * Increment power.disable_depth for the device and if it was zero previously,
1342  * cancel all pending runtime PM requests for the device and wait for all
1343  * operations in progress to complete.  The device can be either active or
1344  * suspended after its runtime PM has been disabled.
1345  *
1346  * If @check_resume is set and there's a resume request pending when
1347  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1348  * function will wake up the device before disabling its runtime PM.
1349  */
1350 void __pm_runtime_disable(struct device *dev, bool check_resume)
1351 {
1352 	spin_lock_irq(&dev->power.lock);
1353 
1354 	if (dev->power.disable_depth > 0) {
1355 		dev->power.disable_depth++;
1356 		goto out;
1357 	}
1358 
1359 	/*
1360 	 * Wake up the device if there's a resume request pending, because that
1361 	 * means there probably is some I/O to process and disabling runtime PM
1362 	 * shouldn't prevent the device from processing the I/O.
1363 	 */
1364 	if (check_resume && dev->power.request_pending
1365 	    && dev->power.request == RPM_REQ_RESUME) {
1366 		/*
1367 		 * Prevent suspends and idle notifications from being carried
1368 		 * out after we have woken up the device.
1369 		 */
1370 		pm_runtime_get_noresume(dev);
1371 
1372 		rpm_resume(dev, 0);
1373 
1374 		pm_runtime_put_noidle(dev);
1375 	}
1376 
1377 	/* Update time accounting before disabling PM-runtime. */
1378 	update_pm_runtime_accounting(dev);
1379 
1380 	if (!dev->power.disable_depth++)
1381 		__pm_runtime_barrier(dev);
1382 
1383  out:
1384 	spin_unlock_irq(&dev->power.lock);
1385 }
1386 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1387 
1388 /**
1389  * pm_runtime_enable - Enable runtime PM of a device.
1390  * @dev: Device to handle.
1391  */
1392 void pm_runtime_enable(struct device *dev)
1393 {
1394 	unsigned long flags;
1395 
1396 	spin_lock_irqsave(&dev->power.lock, flags);
1397 
1398 	if (dev->power.disable_depth > 0) {
1399 		dev->power.disable_depth--;
1400 
1401 		/* About to enable runtime pm, set accounting_timestamp to now */
1402 		if (!dev->power.disable_depth)
1403 			dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1404 	} else {
1405 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1406 	}
1407 
1408 	WARN(!dev->power.disable_depth &&
1409 	     dev->power.runtime_status == RPM_SUSPENDED &&
1410 	     !dev->power.ignore_children &&
1411 	     atomic_read(&dev->power.child_count) > 0,
1412 	     "Enabling runtime PM for inactive device (%s) with active children\n",
1413 	     dev_name(dev));
1414 
1415 	spin_unlock_irqrestore(&dev->power.lock, flags);
1416 }
1417 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1418 
1419 /**
1420  * pm_runtime_forbid - Block runtime PM of a device.
1421  * @dev: Device to handle.
1422  *
1423  * Increase the device's usage count and clear its power.runtime_auto flag,
1424  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1425  * for it.
1426  */
1427 void pm_runtime_forbid(struct device *dev)
1428 {
1429 	spin_lock_irq(&dev->power.lock);
1430 	if (!dev->power.runtime_auto)
1431 		goto out;
1432 
1433 	dev->power.runtime_auto = false;
1434 	atomic_inc(&dev->power.usage_count);
1435 	rpm_resume(dev, 0);
1436 
1437  out:
1438 	spin_unlock_irq(&dev->power.lock);
1439 }
1440 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1441 
1442 /**
1443  * pm_runtime_allow - Unblock runtime PM of a device.
1444  * @dev: Device to handle.
1445  *
1446  * Decrease the device's usage count and set its power.runtime_auto flag.
1447  */
1448 void pm_runtime_allow(struct device *dev)
1449 {
1450 	spin_lock_irq(&dev->power.lock);
1451 	if (dev->power.runtime_auto)
1452 		goto out;
1453 
1454 	dev->power.runtime_auto = true;
1455 	if (atomic_dec_and_test(&dev->power.usage_count))
1456 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1457 	else
1458 		trace_rpm_usage_rcuidle(dev, RPM_AUTO | RPM_ASYNC);
1459 
1460  out:
1461 	spin_unlock_irq(&dev->power.lock);
1462 }
1463 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1464 
1465 /**
1466  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1467  * @dev: Device to handle.
1468  *
1469  * Set the power.no_callbacks flag, which tells the PM core that this
1470  * device is power-managed through its parent and has no runtime PM
1471  * callbacks of its own.  The runtime sysfs attributes will be removed.
1472  */
1473 void pm_runtime_no_callbacks(struct device *dev)
1474 {
1475 	spin_lock_irq(&dev->power.lock);
1476 	dev->power.no_callbacks = 1;
1477 	spin_unlock_irq(&dev->power.lock);
1478 	if (device_is_registered(dev))
1479 		rpm_sysfs_remove(dev);
1480 }
1481 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1482 
1483 /**
1484  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1485  * @dev: Device to handle
1486  *
1487  * Set the power.irq_safe flag, which tells the PM core that the
1488  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1489  * always be invoked with the spinlock held and interrupts disabled.  It also
1490  * causes the parent's usage counter to be permanently incremented, preventing
1491  * the parent from runtime suspending -- otherwise an irq-safe child might have
1492  * to wait for a non-irq-safe parent.
1493  */
1494 void pm_runtime_irq_safe(struct device *dev)
1495 {
1496 	if (dev->parent)
1497 		pm_runtime_get_sync(dev->parent);
1498 	spin_lock_irq(&dev->power.lock);
1499 	dev->power.irq_safe = 1;
1500 	spin_unlock_irq(&dev->power.lock);
1501 }
1502 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1503 
1504 /**
1505  * update_autosuspend - Handle a change to a device's autosuspend settings.
1506  * @dev: Device to handle.
1507  * @old_delay: The former autosuspend_delay value.
1508  * @old_use: The former use_autosuspend value.
1509  *
1510  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1511  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1512  *
1513  * This function must be called under dev->power.lock with interrupts disabled.
1514  */
1515 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1516 {
1517 	int delay = dev->power.autosuspend_delay;
1518 
1519 	/* Should runtime suspend be prevented now? */
1520 	if (dev->power.use_autosuspend && delay < 0) {
1521 
1522 		/* If it used to be allowed then prevent it. */
1523 		if (!old_use || old_delay >= 0) {
1524 			atomic_inc(&dev->power.usage_count);
1525 			rpm_resume(dev, 0);
1526 		} else {
1527 			trace_rpm_usage_rcuidle(dev, 0);
1528 		}
1529 	}
1530 
1531 	/* Runtime suspend should be allowed now. */
1532 	else {
1533 
1534 		/* If it used to be prevented then allow it. */
1535 		if (old_use && old_delay < 0)
1536 			atomic_dec(&dev->power.usage_count);
1537 
1538 		/* Maybe we can autosuspend now. */
1539 		rpm_idle(dev, RPM_AUTO);
1540 	}
1541 }
1542 
1543 /**
1544  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1545  * @dev: Device to handle.
1546  * @delay: Value of the new delay in milliseconds.
1547  *
1548  * Set the device's power.autosuspend_delay value.  If it changes to negative
1549  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1550  * changes the other way, allow runtime suspends.
1551  */
1552 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1553 {
1554 	int old_delay, old_use;
1555 
1556 	spin_lock_irq(&dev->power.lock);
1557 	old_delay = dev->power.autosuspend_delay;
1558 	old_use = dev->power.use_autosuspend;
1559 	dev->power.autosuspend_delay = delay;
1560 	update_autosuspend(dev, old_delay, old_use);
1561 	spin_unlock_irq(&dev->power.lock);
1562 }
1563 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1564 
1565 /**
1566  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1567  * @dev: Device to handle.
1568  * @use: New value for use_autosuspend.
1569  *
1570  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1571  * suspends as needed.
1572  */
1573 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1574 {
1575 	int old_delay, old_use;
1576 
1577 	spin_lock_irq(&dev->power.lock);
1578 	old_delay = dev->power.autosuspend_delay;
1579 	old_use = dev->power.use_autosuspend;
1580 	dev->power.use_autosuspend = use;
1581 	update_autosuspend(dev, old_delay, old_use);
1582 	spin_unlock_irq(&dev->power.lock);
1583 }
1584 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1585 
1586 /**
1587  * pm_runtime_init - Initialize runtime PM fields in given device object.
1588  * @dev: Device object to initialize.
1589  */
1590 void pm_runtime_init(struct device *dev)
1591 {
1592 	dev->power.runtime_status = RPM_SUSPENDED;
1593 	dev->power.idle_notification = false;
1594 
1595 	dev->power.disable_depth = 1;
1596 	atomic_set(&dev->power.usage_count, 0);
1597 
1598 	dev->power.runtime_error = 0;
1599 
1600 	atomic_set(&dev->power.child_count, 0);
1601 	pm_suspend_ignore_children(dev, false);
1602 	dev->power.runtime_auto = true;
1603 
1604 	dev->power.request_pending = false;
1605 	dev->power.request = RPM_REQ_NONE;
1606 	dev->power.deferred_resume = false;
1607 	INIT_WORK(&dev->power.work, pm_runtime_work);
1608 
1609 	dev->power.timer_expires = 0;
1610 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1611 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1612 
1613 	init_waitqueue_head(&dev->power.wait_queue);
1614 }
1615 
1616 /**
1617  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1618  * @dev: Device object to re-initialize.
1619  */
1620 void pm_runtime_reinit(struct device *dev)
1621 {
1622 	if (!pm_runtime_enabled(dev)) {
1623 		if (dev->power.runtime_status == RPM_ACTIVE)
1624 			pm_runtime_set_suspended(dev);
1625 		if (dev->power.irq_safe) {
1626 			spin_lock_irq(&dev->power.lock);
1627 			dev->power.irq_safe = 0;
1628 			spin_unlock_irq(&dev->power.lock);
1629 			if (dev->parent)
1630 				pm_runtime_put(dev->parent);
1631 		}
1632 	}
1633 }
1634 
1635 /**
1636  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1637  * @dev: Device object being removed from device hierarchy.
1638  */
1639 void pm_runtime_remove(struct device *dev)
1640 {
1641 	__pm_runtime_disable(dev, false);
1642 	pm_runtime_reinit(dev);
1643 }
1644 
1645 /**
1646  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1647  * @dev: Consumer device.
1648  */
1649 void pm_runtime_get_suppliers(struct device *dev)
1650 {
1651 	struct device_link *link;
1652 	int idx;
1653 
1654 	idx = device_links_read_lock();
1655 
1656 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1657 				device_links_read_lock_held())
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 				device_links_read_lock_held())
1680 		if (link->supplier_preactivated) {
1681 			link->supplier_preactivated = false;
1682 			if (refcount_dec_not_one(&link->rpm_active))
1683 				pm_runtime_put(link->supplier);
1684 		}
1685 
1686 	device_links_read_unlock(idx);
1687 }
1688 
1689 void pm_runtime_new_link(struct device *dev)
1690 {
1691 	spin_lock_irq(&dev->power.lock);
1692 	dev->power.links_count++;
1693 	spin_unlock_irq(&dev->power.lock);
1694 }
1695 
1696 static void pm_runtime_drop_link_count(struct device *dev)
1697 {
1698 	spin_lock_irq(&dev->power.lock);
1699 	WARN_ON(dev->power.links_count == 0);
1700 	dev->power.links_count--;
1701 	spin_unlock_irq(&dev->power.lock);
1702 }
1703 
1704 /**
1705  * pm_runtime_drop_link - Prepare for device link removal.
1706  * @link: Device link going away.
1707  *
1708  * Drop the link count of the consumer end of @link and decrement the supplier
1709  * device's runtime PM usage counter as many times as needed to drop all of the
1710  * PM runtime reference to it from the consumer.
1711  */
1712 void pm_runtime_drop_link(struct device_link *link)
1713 {
1714 	if (!(link->flags & DL_FLAG_PM_RUNTIME))
1715 		return;
1716 
1717 	pm_runtime_drop_link_count(link->consumer);
1718 
1719 	while (refcount_dec_not_one(&link->rpm_active))
1720 		pm_runtime_put(link->supplier);
1721 }
1722 
1723 static bool pm_runtime_need_not_resume(struct device *dev)
1724 {
1725 	return atomic_read(&dev->power.usage_count) <= 1 &&
1726 		(atomic_read(&dev->power.child_count) == 0 ||
1727 		 dev->power.ignore_children);
1728 }
1729 
1730 /**
1731  * pm_runtime_force_suspend - Force a device into suspend state if needed.
1732  * @dev: Device to suspend.
1733  *
1734  * Disable runtime PM so we safely can check the device's runtime PM status and
1735  * if it is active, invoke its ->runtime_suspend callback to suspend it and
1736  * change its runtime PM status field to RPM_SUSPENDED.  Also, if the device's
1737  * usage and children counters don't indicate that the device was in use before
1738  * the system-wide transition under way, decrement its parent's children counter
1739  * (if there is a parent).  Keep runtime PM disabled to preserve the state
1740  * unless we encounter errors.
1741  *
1742  * Typically this function may be invoked from a system suspend callback to make
1743  * sure the device is put into low power state and it should only be used during
1744  * system-wide PM transitions to sleep states.  It assumes that the analogous
1745  * pm_runtime_force_resume() will be used to resume the device.
1746  */
1747 int pm_runtime_force_suspend(struct device *dev)
1748 {
1749 	int (*callback)(struct device *);
1750 	int ret;
1751 
1752 	pm_runtime_disable(dev);
1753 	if (pm_runtime_status_suspended(dev))
1754 		return 0;
1755 
1756 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1757 
1758 	ret = callback ? callback(dev) : 0;
1759 	if (ret)
1760 		goto err;
1761 
1762 	/*
1763 	 * If the device can stay in suspend after the system-wide transition
1764 	 * to the working state that will follow, drop the children counter of
1765 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1766 	 * function will be called again for it in the meantime.
1767 	 */
1768 	if (pm_runtime_need_not_resume(dev))
1769 		pm_runtime_set_suspended(dev);
1770 	else
1771 		__update_runtime_status(dev, RPM_SUSPENDED);
1772 
1773 	return 0;
1774 
1775 err:
1776 	pm_runtime_enable(dev);
1777 	return ret;
1778 }
1779 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1780 
1781 /**
1782  * pm_runtime_force_resume - Force a device into resume state if needed.
1783  * @dev: Device to resume.
1784  *
1785  * Prior invoking this function we expect the user to have brought the device
1786  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1787  * those actions and bring the device into full power, if it is expected to be
1788  * used on system resume.  In the other case, we defer the resume to be managed
1789  * via runtime PM.
1790  *
1791  * Typically this function may be invoked from a system resume callback.
1792  */
1793 int pm_runtime_force_resume(struct device *dev)
1794 {
1795 	int (*callback)(struct device *);
1796 	int ret = 0;
1797 
1798 	if (!pm_runtime_status_suspended(dev) || pm_runtime_need_not_resume(dev))
1799 		goto out;
1800 
1801 	/*
1802 	 * The value of the parent's children counter is correct already, so
1803 	 * just update the status of the device.
1804 	 */
1805 	__update_runtime_status(dev, RPM_ACTIVE);
1806 
1807 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1808 
1809 	ret = callback ? callback(dev) : 0;
1810 	if (ret) {
1811 		pm_runtime_set_suspended(dev);
1812 		goto out;
1813 	}
1814 
1815 	pm_runtime_mark_last_busy(dev);
1816 out:
1817 	pm_runtime_enable(dev);
1818 	return ret;
1819 }
1820 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1821