xref: /openbmc/linux/drivers/base/power/runtime.c (revision 26721b02)
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 		    READ_ONCE(link->status) == DL_STATE_SUPPLIER_UNBIND)
296 			continue;
297 
298 		retval = pm_runtime_get_sync(link->supplier);
299 		/* Ignore suppliers with disabled runtime PM. */
300 		if (retval < 0 && retval != -EACCES) {
301 			pm_runtime_put_noidle(link->supplier);
302 			return retval;
303 		}
304 		refcount_inc(&link->rpm_active);
305 	}
306 	return 0;
307 }
308 
309 static void rpm_put_suppliers(struct device *dev)
310 {
311 	struct device_link *link;
312 
313 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
314 				device_links_read_lock_held()) {
315 		if (READ_ONCE(link->status) == DL_STATE_SUPPLIER_UNBIND)
316 			continue;
317 
318 		while (refcount_dec_not_one(&link->rpm_active))
319 			pm_runtime_put(link->supplier);
320 	}
321 }
322 
323 /**
324  * __rpm_callback - Run a given runtime PM callback for a given device.
325  * @cb: Runtime PM callback to run.
326  * @dev: Device to run the callback for.
327  */
328 static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
329 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
330 {
331 	int retval, idx;
332 	bool use_links = dev->power.links_count > 0;
333 
334 	if (dev->power.irq_safe) {
335 		spin_unlock(&dev->power.lock);
336 	} else {
337 		spin_unlock_irq(&dev->power.lock);
338 
339 		/*
340 		 * Resume suppliers if necessary.
341 		 *
342 		 * The device's runtime PM status cannot change until this
343 		 * routine returns, so it is safe to read the status outside of
344 		 * the lock.
345 		 */
346 		if (use_links && dev->power.runtime_status == RPM_RESUMING) {
347 			idx = device_links_read_lock();
348 
349 			retval = rpm_get_suppliers(dev);
350 			if (retval)
351 				goto fail;
352 
353 			device_links_read_unlock(idx);
354 		}
355 	}
356 
357 	retval = cb(dev);
358 
359 	if (dev->power.irq_safe) {
360 		spin_lock(&dev->power.lock);
361 	} else {
362 		/*
363 		 * If the device is suspending and the callback has returned
364 		 * success, drop the usage counters of the suppliers that have
365 		 * been reference counted on its resume.
366 		 *
367 		 * Do that if resume fails too.
368 		 */
369 		if (use_links
370 		    && ((dev->power.runtime_status == RPM_SUSPENDING && !retval)
371 		    || (dev->power.runtime_status == RPM_RESUMING && retval))) {
372 			idx = device_links_read_lock();
373 
374  fail:
375 			rpm_put_suppliers(dev);
376 
377 			device_links_read_unlock(idx);
378 		}
379 
380 		spin_lock_irq(&dev->power.lock);
381 	}
382 
383 	return retval;
384 }
385 
386 /**
387  * rpm_idle - Notify device bus type if the device can be suspended.
388  * @dev: Device to notify the bus type about.
389  * @rpmflags: Flag bits.
390  *
391  * Check if the device's runtime PM status allows it to be suspended.  If
392  * another idle notification has been started earlier, return immediately.  If
393  * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
394  * run the ->runtime_idle() callback directly. If the ->runtime_idle callback
395  * doesn't exist or if it returns 0, call rpm_suspend with the RPM_AUTO flag.
396  *
397  * This function must be called under dev->power.lock with interrupts disabled.
398  */
399 static int rpm_idle(struct device *dev, int rpmflags)
400 {
401 	int (*callback)(struct device *);
402 	int retval;
403 
404 	trace_rpm_idle_rcuidle(dev, rpmflags);
405 	retval = rpm_check_suspend_allowed(dev);
406 	if (retval < 0)
407 		;	/* Conditions are wrong. */
408 
409 	/* Idle notifications are allowed only in the RPM_ACTIVE state. */
410 	else if (dev->power.runtime_status != RPM_ACTIVE)
411 		retval = -EAGAIN;
412 
413 	/*
414 	 * Any pending request other than an idle notification takes
415 	 * precedence over us, except that the timer may be running.
416 	 */
417 	else if (dev->power.request_pending &&
418 	    dev->power.request > RPM_REQ_IDLE)
419 		retval = -EAGAIN;
420 
421 	/* Act as though RPM_NOWAIT is always set. */
422 	else if (dev->power.idle_notification)
423 		retval = -EINPROGRESS;
424 	if (retval)
425 		goto out;
426 
427 	/* Pending requests need to be canceled. */
428 	dev->power.request = RPM_REQ_NONE;
429 
430 	if (dev->power.no_callbacks)
431 		goto out;
432 
433 	/* Carry out an asynchronous or a synchronous idle notification. */
434 	if (rpmflags & RPM_ASYNC) {
435 		dev->power.request = RPM_REQ_IDLE;
436 		if (!dev->power.request_pending) {
437 			dev->power.request_pending = true;
438 			queue_work(pm_wq, &dev->power.work);
439 		}
440 		trace_rpm_return_int_rcuidle(dev, _THIS_IP_, 0);
441 		return 0;
442 	}
443 
444 	dev->power.idle_notification = true;
445 
446 	callback = RPM_GET_CALLBACK(dev, runtime_idle);
447 
448 	if (callback)
449 		retval = __rpm_callback(callback, dev);
450 
451 	dev->power.idle_notification = false;
452 	wake_up_all(&dev->power.wait_queue);
453 
454  out:
455 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
456 	return retval ? retval : rpm_suspend(dev, rpmflags | RPM_AUTO);
457 }
458 
459 /**
460  * rpm_callback - Run a given runtime PM callback for a given device.
461  * @cb: Runtime PM callback to run.
462  * @dev: Device to run the callback for.
463  */
464 static int rpm_callback(int (*cb)(struct device *), struct device *dev)
465 {
466 	int retval;
467 
468 	if (!cb)
469 		return -ENOSYS;
470 
471 	if (dev->power.memalloc_noio) {
472 		unsigned int noio_flag;
473 
474 		/*
475 		 * Deadlock might be caused if memory allocation with
476 		 * GFP_KERNEL happens inside runtime_suspend and
477 		 * runtime_resume callbacks of one block device's
478 		 * ancestor or the block device itself. Network
479 		 * device might be thought as part of iSCSI block
480 		 * device, so network device and its ancestor should
481 		 * be marked as memalloc_noio too.
482 		 */
483 		noio_flag = memalloc_noio_save();
484 		retval = __rpm_callback(cb, dev);
485 		memalloc_noio_restore(noio_flag);
486 	} else {
487 		retval = __rpm_callback(cb, dev);
488 	}
489 
490 	dev->power.runtime_error = retval;
491 	return retval != -EACCES ? retval : -EIO;
492 }
493 
494 /**
495  * rpm_suspend - Carry out runtime suspend of given device.
496  * @dev: Device to suspend.
497  * @rpmflags: Flag bits.
498  *
499  * Check if the device's runtime PM status allows it to be suspended.
500  * Cancel a pending idle notification, autosuspend or suspend. If
501  * another suspend has been started earlier, either return immediately
502  * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
503  * flags. If the RPM_ASYNC flag is set then queue a suspend request;
504  * otherwise run the ->runtime_suspend() callback directly. When
505  * ->runtime_suspend succeeded, if a deferred resume was requested while
506  * the callback was running then carry it out, otherwise send an idle
507  * notification for its parent (if the suspend succeeded and both
508  * ignore_children of parent->power and irq_safe of dev->power are not set).
509  * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
510  * flag is set and the next autosuspend-delay expiration time is in the
511  * future, schedule another autosuspend attempt.
512  *
513  * This function must be called under dev->power.lock with interrupts disabled.
514  */
515 static int rpm_suspend(struct device *dev, int rpmflags)
516 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
517 {
518 	int (*callback)(struct device *);
519 	struct device *parent = NULL;
520 	int retval;
521 
522 	trace_rpm_suspend_rcuidle(dev, rpmflags);
523 
524  repeat:
525 	retval = rpm_check_suspend_allowed(dev);
526 	if (retval < 0)
527 		goto out;	/* Conditions are wrong. */
528 
529 	/* Synchronous suspends are not allowed in the RPM_RESUMING state. */
530 	if (dev->power.runtime_status == RPM_RESUMING && !(rpmflags & RPM_ASYNC))
531 		retval = -EAGAIN;
532 	if (retval)
533 		goto out;
534 
535 	/* If the autosuspend_delay time hasn't expired yet, reschedule. */
536 	if ((rpmflags & RPM_AUTO)
537 	    && dev->power.runtime_status != RPM_SUSPENDING) {
538 		u64 expires = pm_runtime_autosuspend_expiration(dev);
539 
540 		if (expires != 0) {
541 			/* Pending requests need to be canceled. */
542 			dev->power.request = RPM_REQ_NONE;
543 
544 			/*
545 			 * Optimization: If the timer is already running and is
546 			 * set to expire at or before the autosuspend delay,
547 			 * avoid the overhead of resetting it.  Just let it
548 			 * expire; pm_suspend_timer_fn() will take care of the
549 			 * rest.
550 			 */
551 			if (!(dev->power.timer_expires &&
552 					dev->power.timer_expires <= expires)) {
553 				/*
554 				 * We add a slack of 25% to gather wakeups
555 				 * without sacrificing the granularity.
556 				 */
557 				u64 slack = (u64)READ_ONCE(dev->power.autosuspend_delay) *
558 						    (NSEC_PER_MSEC >> 2);
559 
560 				dev->power.timer_expires = expires;
561 				hrtimer_start_range_ns(&dev->power.suspend_timer,
562 						ns_to_ktime(expires),
563 						slack,
564 						HRTIMER_MODE_ABS);
565 			}
566 			dev->power.timer_autosuspends = 1;
567 			goto out;
568 		}
569 	}
570 
571 	/* Other scheduled or pending requests need to be canceled. */
572 	pm_runtime_cancel_pending(dev);
573 
574 	if (dev->power.runtime_status == RPM_SUSPENDING) {
575 		DEFINE_WAIT(wait);
576 
577 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
578 			retval = -EINPROGRESS;
579 			goto out;
580 		}
581 
582 		if (dev->power.irq_safe) {
583 			spin_unlock(&dev->power.lock);
584 
585 			cpu_relax();
586 
587 			spin_lock(&dev->power.lock);
588 			goto repeat;
589 		}
590 
591 		/* Wait for the other suspend running in parallel with us. */
592 		for (;;) {
593 			prepare_to_wait(&dev->power.wait_queue, &wait,
594 					TASK_UNINTERRUPTIBLE);
595 			if (dev->power.runtime_status != RPM_SUSPENDING)
596 				break;
597 
598 			spin_unlock_irq(&dev->power.lock);
599 
600 			schedule();
601 
602 			spin_lock_irq(&dev->power.lock);
603 		}
604 		finish_wait(&dev->power.wait_queue, &wait);
605 		goto repeat;
606 	}
607 
608 	if (dev->power.no_callbacks)
609 		goto no_callback;	/* Assume success. */
610 
611 	/* Carry out an asynchronous or a synchronous suspend. */
612 	if (rpmflags & RPM_ASYNC) {
613 		dev->power.request = (rpmflags & RPM_AUTO) ?
614 		    RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
615 		if (!dev->power.request_pending) {
616 			dev->power.request_pending = true;
617 			queue_work(pm_wq, &dev->power.work);
618 		}
619 		goto out;
620 	}
621 
622 	__update_runtime_status(dev, RPM_SUSPENDING);
623 
624 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
625 
626 	dev_pm_enable_wake_irq_check(dev, true);
627 	retval = rpm_callback(callback, dev);
628 	if (retval)
629 		goto fail;
630 
631  no_callback:
632 	__update_runtime_status(dev, RPM_SUSPENDED);
633 	pm_runtime_deactivate_timer(dev);
634 
635 	if (dev->parent) {
636 		parent = dev->parent;
637 		atomic_add_unless(&parent->power.child_count, -1, 0);
638 	}
639 	wake_up_all(&dev->power.wait_queue);
640 
641 	if (dev->power.deferred_resume) {
642 		dev->power.deferred_resume = false;
643 		rpm_resume(dev, 0);
644 		retval = -EAGAIN;
645 		goto out;
646 	}
647 
648 	/* Maybe the parent is now able to suspend. */
649 	if (parent && !parent->power.ignore_children && !dev->power.irq_safe) {
650 		spin_unlock(&dev->power.lock);
651 
652 		spin_lock(&parent->power.lock);
653 		rpm_idle(parent, RPM_ASYNC);
654 		spin_unlock(&parent->power.lock);
655 
656 		spin_lock(&dev->power.lock);
657 	}
658 
659  out:
660 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
661 
662 	return retval;
663 
664  fail:
665 	dev_pm_disable_wake_irq_check(dev);
666 	__update_runtime_status(dev, RPM_ACTIVE);
667 	dev->power.deferred_resume = false;
668 	wake_up_all(&dev->power.wait_queue);
669 
670 	if (retval == -EAGAIN || retval == -EBUSY) {
671 		dev->power.runtime_error = 0;
672 
673 		/*
674 		 * If the callback routine failed an autosuspend, and
675 		 * if the last_busy time has been updated so that there
676 		 * is a new autosuspend expiration time, automatically
677 		 * reschedule another autosuspend.
678 		 */
679 		if ((rpmflags & RPM_AUTO) &&
680 		    pm_runtime_autosuspend_expiration(dev) != 0)
681 			goto repeat;
682 	} else {
683 		pm_runtime_cancel_pending(dev);
684 	}
685 	goto out;
686 }
687 
688 /**
689  * rpm_resume - Carry out runtime resume of given device.
690  * @dev: Device to resume.
691  * @rpmflags: Flag bits.
692  *
693  * Check if the device's runtime PM status allows it to be resumed.  Cancel
694  * any scheduled or pending requests.  If another resume has been started
695  * earlier, either return immediately or wait for it to finish, depending on the
696  * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
697  * parallel with this function, either tell the other process to resume after
698  * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
699  * flag is set then queue a resume request; otherwise run the
700  * ->runtime_resume() callback directly.  Queue an idle notification for the
701  * device if the resume succeeded.
702  *
703  * This function must be called under dev->power.lock with interrupts disabled.
704  */
705 static int rpm_resume(struct device *dev, int rpmflags)
706 	__releases(&dev->power.lock) __acquires(&dev->power.lock)
707 {
708 	int (*callback)(struct device *);
709 	struct device *parent = NULL;
710 	int retval = 0;
711 
712 	trace_rpm_resume_rcuidle(dev, rpmflags);
713 
714  repeat:
715 	if (dev->power.runtime_error)
716 		retval = -EINVAL;
717 	else if (dev->power.disable_depth == 1 && dev->power.is_suspended
718 	    && dev->power.runtime_status == RPM_ACTIVE)
719 		retval = 1;
720 	else if (dev->power.disable_depth > 0)
721 		retval = -EACCES;
722 	if (retval)
723 		goto out;
724 
725 	/*
726 	 * Other scheduled or pending requests need to be canceled.  Small
727 	 * optimization: If an autosuspend timer is running, leave it running
728 	 * rather than cancelling it now only to restart it again in the near
729 	 * future.
730 	 */
731 	dev->power.request = RPM_REQ_NONE;
732 	if (!dev->power.timer_autosuspends)
733 		pm_runtime_deactivate_timer(dev);
734 
735 	if (dev->power.runtime_status == RPM_ACTIVE) {
736 		retval = 1;
737 		goto out;
738 	}
739 
740 	if (dev->power.runtime_status == RPM_RESUMING
741 	    || dev->power.runtime_status == RPM_SUSPENDING) {
742 		DEFINE_WAIT(wait);
743 
744 		if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
745 			if (dev->power.runtime_status == RPM_SUSPENDING)
746 				dev->power.deferred_resume = true;
747 			else
748 				retval = -EINPROGRESS;
749 			goto out;
750 		}
751 
752 		if (dev->power.irq_safe) {
753 			spin_unlock(&dev->power.lock);
754 
755 			cpu_relax();
756 
757 			spin_lock(&dev->power.lock);
758 			goto repeat;
759 		}
760 
761 		/* Wait for the operation carried out in parallel with us. */
762 		for (;;) {
763 			prepare_to_wait(&dev->power.wait_queue, &wait,
764 					TASK_UNINTERRUPTIBLE);
765 			if (dev->power.runtime_status != RPM_RESUMING
766 			    && dev->power.runtime_status != RPM_SUSPENDING)
767 				break;
768 
769 			spin_unlock_irq(&dev->power.lock);
770 
771 			schedule();
772 
773 			spin_lock_irq(&dev->power.lock);
774 		}
775 		finish_wait(&dev->power.wait_queue, &wait);
776 		goto repeat;
777 	}
778 
779 	/*
780 	 * See if we can skip waking up the parent.  This is safe only if
781 	 * power.no_callbacks is set, because otherwise we don't know whether
782 	 * the resume will actually succeed.
783 	 */
784 	if (dev->power.no_callbacks && !parent && dev->parent) {
785 		spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
786 		if (dev->parent->power.disable_depth > 0
787 		    || dev->parent->power.ignore_children
788 		    || dev->parent->power.runtime_status == RPM_ACTIVE) {
789 			atomic_inc(&dev->parent->power.child_count);
790 			spin_unlock(&dev->parent->power.lock);
791 			retval = 1;
792 			goto no_callback;	/* Assume success. */
793 		}
794 		spin_unlock(&dev->parent->power.lock);
795 	}
796 
797 	/* Carry out an asynchronous or a synchronous resume. */
798 	if (rpmflags & RPM_ASYNC) {
799 		dev->power.request = RPM_REQ_RESUME;
800 		if (!dev->power.request_pending) {
801 			dev->power.request_pending = true;
802 			queue_work(pm_wq, &dev->power.work);
803 		}
804 		retval = 0;
805 		goto out;
806 	}
807 
808 	if (!parent && dev->parent) {
809 		/*
810 		 * Increment the parent's usage counter and resume it if
811 		 * necessary.  Not needed if dev is irq-safe; then the
812 		 * parent is permanently resumed.
813 		 */
814 		parent = dev->parent;
815 		if (dev->power.irq_safe)
816 			goto skip_parent;
817 		spin_unlock(&dev->power.lock);
818 
819 		pm_runtime_get_noresume(parent);
820 
821 		spin_lock(&parent->power.lock);
822 		/*
823 		 * Resume the parent if it has runtime PM enabled and not been
824 		 * set to ignore its children.
825 		 */
826 		if (!parent->power.disable_depth
827 		    && !parent->power.ignore_children) {
828 			rpm_resume(parent, 0);
829 			if (parent->power.runtime_status != RPM_ACTIVE)
830 				retval = -EBUSY;
831 		}
832 		spin_unlock(&parent->power.lock);
833 
834 		spin_lock(&dev->power.lock);
835 		if (retval)
836 			goto out;
837 		goto repeat;
838 	}
839  skip_parent:
840 
841 	if (dev->power.no_callbacks)
842 		goto no_callback;	/* Assume success. */
843 
844 	__update_runtime_status(dev, RPM_RESUMING);
845 
846 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
847 
848 	dev_pm_disable_wake_irq_check(dev);
849 	retval = rpm_callback(callback, dev);
850 	if (retval) {
851 		__update_runtime_status(dev, RPM_SUSPENDED);
852 		pm_runtime_cancel_pending(dev);
853 		dev_pm_enable_wake_irq_check(dev, false);
854 	} else {
855  no_callback:
856 		__update_runtime_status(dev, RPM_ACTIVE);
857 		pm_runtime_mark_last_busy(dev);
858 		if (parent)
859 			atomic_inc(&parent->power.child_count);
860 	}
861 	wake_up_all(&dev->power.wait_queue);
862 
863 	if (retval >= 0)
864 		rpm_idle(dev, RPM_ASYNC);
865 
866  out:
867 	if (parent && !dev->power.irq_safe) {
868 		spin_unlock_irq(&dev->power.lock);
869 
870 		pm_runtime_put(parent);
871 
872 		spin_lock_irq(&dev->power.lock);
873 	}
874 
875 	trace_rpm_return_int_rcuidle(dev, _THIS_IP_, retval);
876 
877 	return retval;
878 }
879 
880 /**
881  * pm_runtime_work - Universal runtime PM work function.
882  * @work: Work structure used for scheduling the execution of this function.
883  *
884  * Use @work to get the device object the work is to be done for, determine what
885  * is to be done and execute the appropriate runtime PM function.
886  */
887 static void pm_runtime_work(struct work_struct *work)
888 {
889 	struct device *dev = container_of(work, struct device, power.work);
890 	enum rpm_request req;
891 
892 	spin_lock_irq(&dev->power.lock);
893 
894 	if (!dev->power.request_pending)
895 		goto out;
896 
897 	req = dev->power.request;
898 	dev->power.request = RPM_REQ_NONE;
899 	dev->power.request_pending = false;
900 
901 	switch (req) {
902 	case RPM_REQ_NONE:
903 		break;
904 	case RPM_REQ_IDLE:
905 		rpm_idle(dev, RPM_NOWAIT);
906 		break;
907 	case RPM_REQ_SUSPEND:
908 		rpm_suspend(dev, RPM_NOWAIT);
909 		break;
910 	case RPM_REQ_AUTOSUSPEND:
911 		rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
912 		break;
913 	case RPM_REQ_RESUME:
914 		rpm_resume(dev, RPM_NOWAIT);
915 		break;
916 	}
917 
918  out:
919 	spin_unlock_irq(&dev->power.lock);
920 }
921 
922 /**
923  * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
924  * @data: Device pointer passed by pm_schedule_suspend().
925  *
926  * Check if the time is right and queue a suspend request.
927  */
928 static enum hrtimer_restart  pm_suspend_timer_fn(struct hrtimer *timer)
929 {
930 	struct device *dev = container_of(timer, struct device, power.suspend_timer);
931 	unsigned long flags;
932 	u64 expires;
933 
934 	spin_lock_irqsave(&dev->power.lock, flags);
935 
936 	expires = dev->power.timer_expires;
937 	/*
938 	 * If 'expires' is after the current time, we've been called
939 	 * too early.
940 	 */
941 	if (expires > 0 && expires < ktime_get_mono_fast_ns()) {
942 		dev->power.timer_expires = 0;
943 		rpm_suspend(dev, dev->power.timer_autosuspends ?
944 		    (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
945 	}
946 
947 	spin_unlock_irqrestore(&dev->power.lock, flags);
948 
949 	return HRTIMER_NORESTART;
950 }
951 
952 /**
953  * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
954  * @dev: Device to suspend.
955  * @delay: Time to wait before submitting a suspend request, in milliseconds.
956  */
957 int pm_schedule_suspend(struct device *dev, unsigned int delay)
958 {
959 	unsigned long flags;
960 	u64 expires;
961 	int retval;
962 
963 	spin_lock_irqsave(&dev->power.lock, flags);
964 
965 	if (!delay) {
966 		retval = rpm_suspend(dev, RPM_ASYNC);
967 		goto out;
968 	}
969 
970 	retval = rpm_check_suspend_allowed(dev);
971 	if (retval)
972 		goto out;
973 
974 	/* Other scheduled or pending requests need to be canceled. */
975 	pm_runtime_cancel_pending(dev);
976 
977 	expires = ktime_get_mono_fast_ns() + (u64)delay * NSEC_PER_MSEC;
978 	dev->power.timer_expires = expires;
979 	dev->power.timer_autosuspends = 0;
980 	hrtimer_start(&dev->power.suspend_timer, expires, HRTIMER_MODE_ABS);
981 
982  out:
983 	spin_unlock_irqrestore(&dev->power.lock, flags);
984 
985 	return retval;
986 }
987 EXPORT_SYMBOL_GPL(pm_schedule_suspend);
988 
989 /**
990  * __pm_runtime_idle - Entry point for runtime idle operations.
991  * @dev: Device to send idle notification for.
992  * @rpmflags: Flag bits.
993  *
994  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
995  * return immediately if it is larger than zero.  Then carry out an idle
996  * notification, either synchronous or asynchronous.
997  *
998  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
999  * or if pm_runtime_irq_safe() has been called.
1000  */
1001 int __pm_runtime_idle(struct device *dev, int rpmflags)
1002 {
1003 	unsigned long flags;
1004 	int retval;
1005 
1006 	if (rpmflags & RPM_GET_PUT) {
1007 		if (!atomic_dec_and_test(&dev->power.usage_count)) {
1008 			trace_rpm_usage_rcuidle(dev, rpmflags);
1009 			return 0;
1010 		}
1011 	}
1012 
1013 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1014 
1015 	spin_lock_irqsave(&dev->power.lock, flags);
1016 	retval = rpm_idle(dev, rpmflags);
1017 	spin_unlock_irqrestore(&dev->power.lock, flags);
1018 
1019 	return retval;
1020 }
1021 EXPORT_SYMBOL_GPL(__pm_runtime_idle);
1022 
1023 /**
1024  * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
1025  * @dev: Device to suspend.
1026  * @rpmflags: Flag bits.
1027  *
1028  * If the RPM_GET_PUT flag is set, decrement the device's usage count and
1029  * return immediately if it is larger than zero.  Then carry out a suspend,
1030  * either synchronous or asynchronous.
1031  *
1032  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1033  * or if pm_runtime_irq_safe() has been called.
1034  */
1035 int __pm_runtime_suspend(struct device *dev, int rpmflags)
1036 {
1037 	unsigned long flags;
1038 	int retval;
1039 
1040 	if (rpmflags & RPM_GET_PUT) {
1041 		if (!atomic_dec_and_test(&dev->power.usage_count)) {
1042 			trace_rpm_usage_rcuidle(dev, rpmflags);
1043 			return 0;
1044 		}
1045 	}
1046 
1047 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
1048 
1049 	spin_lock_irqsave(&dev->power.lock, flags);
1050 	retval = rpm_suspend(dev, rpmflags);
1051 	spin_unlock_irqrestore(&dev->power.lock, flags);
1052 
1053 	return retval;
1054 }
1055 EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
1056 
1057 /**
1058  * __pm_runtime_resume - Entry point for runtime resume operations.
1059  * @dev: Device to resume.
1060  * @rpmflags: Flag bits.
1061  *
1062  * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
1063  * carry out a resume, either synchronous or asynchronous.
1064  *
1065  * This routine may be called in atomic context if the RPM_ASYNC flag is set,
1066  * or if pm_runtime_irq_safe() has been called.
1067  */
1068 int __pm_runtime_resume(struct device *dev, int rpmflags)
1069 {
1070 	unsigned long flags;
1071 	int retval;
1072 
1073 	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
1074 			dev->power.runtime_status != RPM_ACTIVE);
1075 
1076 	if (rpmflags & RPM_GET_PUT)
1077 		atomic_inc(&dev->power.usage_count);
1078 
1079 	spin_lock_irqsave(&dev->power.lock, flags);
1080 	retval = rpm_resume(dev, rpmflags);
1081 	spin_unlock_irqrestore(&dev->power.lock, flags);
1082 
1083 	return retval;
1084 }
1085 EXPORT_SYMBOL_GPL(__pm_runtime_resume);
1086 
1087 /**
1088  * pm_runtime_get_if_active - Conditionally bump up device usage counter.
1089  * @dev: Device to handle.
1090  * @ign_usage_count: Whether or not to look at the current usage counter value.
1091  *
1092  * Return -EINVAL if runtime PM is disabled for @dev.
1093  *
1094  * Otherwise, if the runtime PM status of @dev is %RPM_ACTIVE and either
1095  * @ign_usage_count is %true or the runtime PM usage counter of @dev is not
1096  * zero, increment the usage counter of @dev and return 1. Otherwise, return 0
1097  * without changing the usage counter.
1098  *
1099  * If @ign_usage_count is %true, this function can be used to prevent suspending
1100  * the device when its runtime PM status is %RPM_ACTIVE.
1101  *
1102  * If @ign_usage_count is %false, this function can be used to prevent
1103  * suspending the device when both its runtime PM status is %RPM_ACTIVE and its
1104  * runtime PM usage counter is not zero.
1105  *
1106  * The caller is resposible for decrementing the runtime PM usage counter of
1107  * @dev after this function has returned a positive value for it.
1108  */
1109 int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count)
1110 {
1111 	unsigned long flags;
1112 	int retval;
1113 
1114 	spin_lock_irqsave(&dev->power.lock, flags);
1115 	if (dev->power.disable_depth > 0) {
1116 		retval = -EINVAL;
1117 	} else if (dev->power.runtime_status != RPM_ACTIVE) {
1118 		retval = 0;
1119 	} else if (ign_usage_count) {
1120 		retval = 1;
1121 		atomic_inc(&dev->power.usage_count);
1122 	} else {
1123 		retval = atomic_inc_not_zero(&dev->power.usage_count);
1124 	}
1125 	trace_rpm_usage_rcuidle(dev, 0);
1126 	spin_unlock_irqrestore(&dev->power.lock, flags);
1127 
1128 	return retval;
1129 }
1130 EXPORT_SYMBOL_GPL(pm_runtime_get_if_active);
1131 
1132 /**
1133  * __pm_runtime_set_status - Set runtime PM status of a device.
1134  * @dev: Device to handle.
1135  * @status: New runtime PM status of the device.
1136  *
1137  * If runtime PM of the device is disabled or its power.runtime_error field is
1138  * different from zero, the status may be changed either to RPM_ACTIVE, or to
1139  * RPM_SUSPENDED, as long as that reflects the actual state of the device.
1140  * However, if the device has a parent and the parent is not active, and the
1141  * parent's power.ignore_children flag is unset, the device's status cannot be
1142  * set to RPM_ACTIVE, so -EBUSY is returned in that case.
1143  *
1144  * If successful, __pm_runtime_set_status() clears the power.runtime_error field
1145  * and the device parent's counter of unsuspended children is modified to
1146  * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
1147  * notification request for the parent is submitted.
1148  *
1149  * If @dev has any suppliers (as reflected by device links to them), and @status
1150  * is RPM_ACTIVE, they will be activated upfront and if the activation of one
1151  * of them fails, the status of @dev will be changed to RPM_SUSPENDED (instead
1152  * of the @status value) and the suppliers will be deacticated on exit.  The
1153  * error returned by the failing supplier activation will be returned in that
1154  * case.
1155  */
1156 int __pm_runtime_set_status(struct device *dev, unsigned int status)
1157 {
1158 	struct device *parent = dev->parent;
1159 	bool notify_parent = false;
1160 	int error = 0;
1161 
1162 	if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
1163 		return -EINVAL;
1164 
1165 	spin_lock_irq(&dev->power.lock);
1166 
1167 	/*
1168 	 * Prevent PM-runtime from being enabled for the device or return an
1169 	 * error if it is enabled already and working.
1170 	 */
1171 	if (dev->power.runtime_error || dev->power.disable_depth)
1172 		dev->power.disable_depth++;
1173 	else
1174 		error = -EAGAIN;
1175 
1176 	spin_unlock_irq(&dev->power.lock);
1177 
1178 	if (error)
1179 		return error;
1180 
1181 	/*
1182 	 * If the new status is RPM_ACTIVE, the suppliers can be activated
1183 	 * upfront regardless of the current status, because next time
1184 	 * rpm_put_suppliers() runs, the rpm_active refcounts of the links
1185 	 * involved will be dropped down to one anyway.
1186 	 */
1187 	if (status == RPM_ACTIVE) {
1188 		int idx = device_links_read_lock();
1189 
1190 		error = rpm_get_suppliers(dev);
1191 		if (error)
1192 			status = RPM_SUSPENDED;
1193 
1194 		device_links_read_unlock(idx);
1195 	}
1196 
1197 	spin_lock_irq(&dev->power.lock);
1198 
1199 	if (dev->power.runtime_status == status || !parent)
1200 		goto out_set;
1201 
1202 	if (status == RPM_SUSPENDED) {
1203 		atomic_add_unless(&parent->power.child_count, -1, 0);
1204 		notify_parent = !parent->power.ignore_children;
1205 	} else {
1206 		spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
1207 
1208 		/*
1209 		 * It is invalid to put an active child under a parent that is
1210 		 * not active, has runtime PM enabled and the
1211 		 * 'power.ignore_children' flag unset.
1212 		 */
1213 		if (!parent->power.disable_depth
1214 		    && !parent->power.ignore_children
1215 		    && parent->power.runtime_status != RPM_ACTIVE) {
1216 			dev_err(dev, "runtime PM trying to activate child device %s but parent (%s) is not active\n",
1217 				dev_name(dev),
1218 				dev_name(parent));
1219 			error = -EBUSY;
1220 		} else if (dev->power.runtime_status == RPM_SUSPENDED) {
1221 			atomic_inc(&parent->power.child_count);
1222 		}
1223 
1224 		spin_unlock(&parent->power.lock);
1225 
1226 		if (error) {
1227 			status = RPM_SUSPENDED;
1228 			goto out;
1229 		}
1230 	}
1231 
1232  out_set:
1233 	__update_runtime_status(dev, status);
1234 	if (!error)
1235 		dev->power.runtime_error = 0;
1236 
1237  out:
1238 	spin_unlock_irq(&dev->power.lock);
1239 
1240 	if (notify_parent)
1241 		pm_request_idle(parent);
1242 
1243 	if (status == RPM_SUSPENDED) {
1244 		int idx = device_links_read_lock();
1245 
1246 		rpm_put_suppliers(dev);
1247 
1248 		device_links_read_unlock(idx);
1249 	}
1250 
1251 	pm_runtime_enable(dev);
1252 
1253 	return error;
1254 }
1255 EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
1256 
1257 /**
1258  * __pm_runtime_barrier - Cancel pending requests and wait for completions.
1259  * @dev: Device to handle.
1260  *
1261  * Flush all pending requests for the device from pm_wq and wait for all
1262  * runtime PM operations involving the device in progress to complete.
1263  *
1264  * Should be called under dev->power.lock with interrupts disabled.
1265  */
1266 static void __pm_runtime_barrier(struct device *dev)
1267 {
1268 	pm_runtime_deactivate_timer(dev);
1269 
1270 	if (dev->power.request_pending) {
1271 		dev->power.request = RPM_REQ_NONE;
1272 		spin_unlock_irq(&dev->power.lock);
1273 
1274 		cancel_work_sync(&dev->power.work);
1275 
1276 		spin_lock_irq(&dev->power.lock);
1277 		dev->power.request_pending = false;
1278 	}
1279 
1280 	if (dev->power.runtime_status == RPM_SUSPENDING
1281 	    || dev->power.runtime_status == RPM_RESUMING
1282 	    || dev->power.idle_notification) {
1283 		DEFINE_WAIT(wait);
1284 
1285 		/* Suspend, wake-up or idle notification in progress. */
1286 		for (;;) {
1287 			prepare_to_wait(&dev->power.wait_queue, &wait,
1288 					TASK_UNINTERRUPTIBLE);
1289 			if (dev->power.runtime_status != RPM_SUSPENDING
1290 			    && dev->power.runtime_status != RPM_RESUMING
1291 			    && !dev->power.idle_notification)
1292 				break;
1293 			spin_unlock_irq(&dev->power.lock);
1294 
1295 			schedule();
1296 
1297 			spin_lock_irq(&dev->power.lock);
1298 		}
1299 		finish_wait(&dev->power.wait_queue, &wait);
1300 	}
1301 }
1302 
1303 /**
1304  * pm_runtime_barrier - Flush pending requests and wait for completions.
1305  * @dev: Device to handle.
1306  *
1307  * Prevent the device from being suspended by incrementing its usage counter and
1308  * if there's a pending resume request for the device, wake the device up.
1309  * Next, make sure that all pending requests for the device have been flushed
1310  * from pm_wq and wait for all runtime PM operations involving the device in
1311  * progress to complete.
1312  *
1313  * Return value:
1314  * 1, if there was a resume request pending and the device had to be woken up,
1315  * 0, otherwise
1316  */
1317 int pm_runtime_barrier(struct device *dev)
1318 {
1319 	int retval = 0;
1320 
1321 	pm_runtime_get_noresume(dev);
1322 	spin_lock_irq(&dev->power.lock);
1323 
1324 	if (dev->power.request_pending
1325 	    && dev->power.request == RPM_REQ_RESUME) {
1326 		rpm_resume(dev, 0);
1327 		retval = 1;
1328 	}
1329 
1330 	__pm_runtime_barrier(dev);
1331 
1332 	spin_unlock_irq(&dev->power.lock);
1333 	pm_runtime_put_noidle(dev);
1334 
1335 	return retval;
1336 }
1337 EXPORT_SYMBOL_GPL(pm_runtime_barrier);
1338 
1339 /**
1340  * __pm_runtime_disable - Disable runtime PM of a device.
1341  * @dev: Device to handle.
1342  * @check_resume: If set, check if there's a resume request for the device.
1343  *
1344  * Increment power.disable_depth for the device and if it was zero previously,
1345  * cancel all pending runtime PM requests for the device and wait for all
1346  * operations in progress to complete.  The device can be either active or
1347  * suspended after its runtime PM has been disabled.
1348  *
1349  * If @check_resume is set and there's a resume request pending when
1350  * __pm_runtime_disable() is called and power.disable_depth is zero, the
1351  * function will wake up the device before disabling its runtime PM.
1352  */
1353 void __pm_runtime_disable(struct device *dev, bool check_resume)
1354 {
1355 	spin_lock_irq(&dev->power.lock);
1356 
1357 	if (dev->power.disable_depth > 0) {
1358 		dev->power.disable_depth++;
1359 		goto out;
1360 	}
1361 
1362 	/*
1363 	 * Wake up the device if there's a resume request pending, because that
1364 	 * means there probably is some I/O to process and disabling runtime PM
1365 	 * shouldn't prevent the device from processing the I/O.
1366 	 */
1367 	if (check_resume && dev->power.request_pending
1368 	    && dev->power.request == RPM_REQ_RESUME) {
1369 		/*
1370 		 * Prevent suspends and idle notifications from being carried
1371 		 * out after we have woken up the device.
1372 		 */
1373 		pm_runtime_get_noresume(dev);
1374 
1375 		rpm_resume(dev, 0);
1376 
1377 		pm_runtime_put_noidle(dev);
1378 	}
1379 
1380 	/* Update time accounting before disabling PM-runtime. */
1381 	update_pm_runtime_accounting(dev);
1382 
1383 	if (!dev->power.disable_depth++)
1384 		__pm_runtime_barrier(dev);
1385 
1386  out:
1387 	spin_unlock_irq(&dev->power.lock);
1388 }
1389 EXPORT_SYMBOL_GPL(__pm_runtime_disable);
1390 
1391 /**
1392  * pm_runtime_enable - Enable runtime PM of a device.
1393  * @dev: Device to handle.
1394  */
1395 void pm_runtime_enable(struct device *dev)
1396 {
1397 	unsigned long flags;
1398 
1399 	spin_lock_irqsave(&dev->power.lock, flags);
1400 
1401 	if (dev->power.disable_depth > 0) {
1402 		dev->power.disable_depth--;
1403 
1404 		/* About to enable runtime pm, set accounting_timestamp to now */
1405 		if (!dev->power.disable_depth)
1406 			dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
1407 	} else {
1408 		dev_warn(dev, "Unbalanced %s!\n", __func__);
1409 	}
1410 
1411 	WARN(!dev->power.disable_depth &&
1412 	     dev->power.runtime_status == RPM_SUSPENDED &&
1413 	     !dev->power.ignore_children &&
1414 	     atomic_read(&dev->power.child_count) > 0,
1415 	     "Enabling runtime PM for inactive device (%s) with active children\n",
1416 	     dev_name(dev));
1417 
1418 	spin_unlock_irqrestore(&dev->power.lock, flags);
1419 }
1420 EXPORT_SYMBOL_GPL(pm_runtime_enable);
1421 
1422 /**
1423  * pm_runtime_forbid - Block runtime PM of a device.
1424  * @dev: Device to handle.
1425  *
1426  * Increase the device's usage count and clear its power.runtime_auto flag,
1427  * so that it cannot be suspended at run time until pm_runtime_allow() is called
1428  * for it.
1429  */
1430 void pm_runtime_forbid(struct device *dev)
1431 {
1432 	spin_lock_irq(&dev->power.lock);
1433 	if (!dev->power.runtime_auto)
1434 		goto out;
1435 
1436 	dev->power.runtime_auto = false;
1437 	atomic_inc(&dev->power.usage_count);
1438 	rpm_resume(dev, 0);
1439 
1440  out:
1441 	spin_unlock_irq(&dev->power.lock);
1442 }
1443 EXPORT_SYMBOL_GPL(pm_runtime_forbid);
1444 
1445 /**
1446  * pm_runtime_allow - Unblock runtime PM of a device.
1447  * @dev: Device to handle.
1448  *
1449  * Decrease the device's usage count and set its power.runtime_auto flag.
1450  */
1451 void pm_runtime_allow(struct device *dev)
1452 {
1453 	spin_lock_irq(&dev->power.lock);
1454 	if (dev->power.runtime_auto)
1455 		goto out;
1456 
1457 	dev->power.runtime_auto = true;
1458 	if (atomic_dec_and_test(&dev->power.usage_count))
1459 		rpm_idle(dev, RPM_AUTO | RPM_ASYNC);
1460 	else
1461 		trace_rpm_usage_rcuidle(dev, RPM_AUTO | RPM_ASYNC);
1462 
1463  out:
1464 	spin_unlock_irq(&dev->power.lock);
1465 }
1466 EXPORT_SYMBOL_GPL(pm_runtime_allow);
1467 
1468 /**
1469  * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
1470  * @dev: Device to handle.
1471  *
1472  * Set the power.no_callbacks flag, which tells the PM core that this
1473  * device is power-managed through its parent and has no runtime PM
1474  * callbacks of its own.  The runtime sysfs attributes will be removed.
1475  */
1476 void pm_runtime_no_callbacks(struct device *dev)
1477 {
1478 	spin_lock_irq(&dev->power.lock);
1479 	dev->power.no_callbacks = 1;
1480 	spin_unlock_irq(&dev->power.lock);
1481 	if (device_is_registered(dev))
1482 		rpm_sysfs_remove(dev);
1483 }
1484 EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
1485 
1486 /**
1487  * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
1488  * @dev: Device to handle
1489  *
1490  * Set the power.irq_safe flag, which tells the PM core that the
1491  * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
1492  * always be invoked with the spinlock held and interrupts disabled.  It also
1493  * causes the parent's usage counter to be permanently incremented, preventing
1494  * the parent from runtime suspending -- otherwise an irq-safe child might have
1495  * to wait for a non-irq-safe parent.
1496  */
1497 void pm_runtime_irq_safe(struct device *dev)
1498 {
1499 	if (dev->parent)
1500 		pm_runtime_get_sync(dev->parent);
1501 	spin_lock_irq(&dev->power.lock);
1502 	dev->power.irq_safe = 1;
1503 	spin_unlock_irq(&dev->power.lock);
1504 }
1505 EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
1506 
1507 /**
1508  * update_autosuspend - Handle a change to a device's autosuspend settings.
1509  * @dev: Device to handle.
1510  * @old_delay: The former autosuspend_delay value.
1511  * @old_use: The former use_autosuspend value.
1512  *
1513  * Prevent runtime suspend if the new delay is negative and use_autosuspend is
1514  * set; otherwise allow it.  Send an idle notification if suspends are allowed.
1515  *
1516  * This function must be called under dev->power.lock with interrupts disabled.
1517  */
1518 static void update_autosuspend(struct device *dev, int old_delay, int old_use)
1519 {
1520 	int delay = dev->power.autosuspend_delay;
1521 
1522 	/* Should runtime suspend be prevented now? */
1523 	if (dev->power.use_autosuspend && delay < 0) {
1524 
1525 		/* If it used to be allowed then prevent it. */
1526 		if (!old_use || old_delay >= 0) {
1527 			atomic_inc(&dev->power.usage_count);
1528 			rpm_resume(dev, 0);
1529 		} else {
1530 			trace_rpm_usage_rcuidle(dev, 0);
1531 		}
1532 	}
1533 
1534 	/* Runtime suspend should be allowed now. */
1535 	else {
1536 
1537 		/* If it used to be prevented then allow it. */
1538 		if (old_use && old_delay < 0)
1539 			atomic_dec(&dev->power.usage_count);
1540 
1541 		/* Maybe we can autosuspend now. */
1542 		rpm_idle(dev, RPM_AUTO);
1543 	}
1544 }
1545 
1546 /**
1547  * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
1548  * @dev: Device to handle.
1549  * @delay: Value of the new delay in milliseconds.
1550  *
1551  * Set the device's power.autosuspend_delay value.  If it changes to negative
1552  * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
1553  * changes the other way, allow runtime suspends.
1554  */
1555 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
1556 {
1557 	int old_delay, old_use;
1558 
1559 	spin_lock_irq(&dev->power.lock);
1560 	old_delay = dev->power.autosuspend_delay;
1561 	old_use = dev->power.use_autosuspend;
1562 	dev->power.autosuspend_delay = delay;
1563 	update_autosuspend(dev, old_delay, old_use);
1564 	spin_unlock_irq(&dev->power.lock);
1565 }
1566 EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
1567 
1568 /**
1569  * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
1570  * @dev: Device to handle.
1571  * @use: New value for use_autosuspend.
1572  *
1573  * Set the device's power.use_autosuspend flag, and allow or prevent runtime
1574  * suspends as needed.
1575  */
1576 void __pm_runtime_use_autosuspend(struct device *dev, bool use)
1577 {
1578 	int old_delay, old_use;
1579 
1580 	spin_lock_irq(&dev->power.lock);
1581 	old_delay = dev->power.autosuspend_delay;
1582 	old_use = dev->power.use_autosuspend;
1583 	dev->power.use_autosuspend = use;
1584 	update_autosuspend(dev, old_delay, old_use);
1585 	spin_unlock_irq(&dev->power.lock);
1586 }
1587 EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
1588 
1589 /**
1590  * pm_runtime_init - Initialize runtime PM fields in given device object.
1591  * @dev: Device object to initialize.
1592  */
1593 void pm_runtime_init(struct device *dev)
1594 {
1595 	dev->power.runtime_status = RPM_SUSPENDED;
1596 	dev->power.idle_notification = false;
1597 
1598 	dev->power.disable_depth = 1;
1599 	atomic_set(&dev->power.usage_count, 0);
1600 
1601 	dev->power.runtime_error = 0;
1602 
1603 	atomic_set(&dev->power.child_count, 0);
1604 	pm_suspend_ignore_children(dev, false);
1605 	dev->power.runtime_auto = true;
1606 
1607 	dev->power.request_pending = false;
1608 	dev->power.request = RPM_REQ_NONE;
1609 	dev->power.deferred_resume = false;
1610 	INIT_WORK(&dev->power.work, pm_runtime_work);
1611 
1612 	dev->power.timer_expires = 0;
1613 	hrtimer_init(&dev->power.suspend_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1614 	dev->power.suspend_timer.function = pm_suspend_timer_fn;
1615 
1616 	init_waitqueue_head(&dev->power.wait_queue);
1617 }
1618 
1619 /**
1620  * pm_runtime_reinit - Re-initialize runtime PM fields in given device object.
1621  * @dev: Device object to re-initialize.
1622  */
1623 void pm_runtime_reinit(struct device *dev)
1624 {
1625 	if (!pm_runtime_enabled(dev)) {
1626 		if (dev->power.runtime_status == RPM_ACTIVE)
1627 			pm_runtime_set_suspended(dev);
1628 		if (dev->power.irq_safe) {
1629 			spin_lock_irq(&dev->power.lock);
1630 			dev->power.irq_safe = 0;
1631 			spin_unlock_irq(&dev->power.lock);
1632 			if (dev->parent)
1633 				pm_runtime_put(dev->parent);
1634 		}
1635 	}
1636 }
1637 
1638 /**
1639  * pm_runtime_remove - Prepare for removing a device from device hierarchy.
1640  * @dev: Device object being removed from device hierarchy.
1641  */
1642 void pm_runtime_remove(struct device *dev)
1643 {
1644 	__pm_runtime_disable(dev, false);
1645 	pm_runtime_reinit(dev);
1646 }
1647 
1648 /**
1649  * pm_runtime_clean_up_links - Prepare links to consumers for driver removal.
1650  * @dev: Device whose driver is going to be removed.
1651  *
1652  * Check links from this device to any consumers and if any of them have active
1653  * runtime PM references to the device, drop the usage counter of the device
1654  * (as many times as needed).
1655  *
1656  * Links with the DL_FLAG_MANAGED flag unset are ignored.
1657  *
1658  * Since the device is guaranteed to be runtime-active at the point this is
1659  * called, nothing else needs to be done here.
1660  *
1661  * Moreover, this is called after device_links_busy() has returned 'false', so
1662  * the status of each link is guaranteed to be DL_STATE_SUPPLIER_UNBIND and
1663  * therefore rpm_active can't be manipulated concurrently.
1664  */
1665 void pm_runtime_clean_up_links(struct device *dev)
1666 {
1667 	struct device_link *link;
1668 	int idx;
1669 
1670 	idx = device_links_read_lock();
1671 
1672 	list_for_each_entry_rcu(link, &dev->links.consumers, s_node,
1673 				device_links_read_lock_held()) {
1674 		if (!(link->flags & DL_FLAG_MANAGED))
1675 			continue;
1676 
1677 		while (refcount_dec_not_one(&link->rpm_active))
1678 			pm_runtime_put_noidle(dev);
1679 	}
1680 
1681 	device_links_read_unlock(idx);
1682 }
1683 
1684 /**
1685  * pm_runtime_get_suppliers - Resume and reference-count supplier devices.
1686  * @dev: Consumer device.
1687  */
1688 void pm_runtime_get_suppliers(struct device *dev)
1689 {
1690 	struct device_link *link;
1691 	int idx;
1692 
1693 	idx = device_links_read_lock();
1694 
1695 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1696 				device_links_read_lock_held())
1697 		if (link->flags & DL_FLAG_PM_RUNTIME) {
1698 			link->supplier_preactivated = true;
1699 			refcount_inc(&link->rpm_active);
1700 			pm_runtime_get_sync(link->supplier);
1701 		}
1702 
1703 	device_links_read_unlock(idx);
1704 }
1705 
1706 /**
1707  * pm_runtime_put_suppliers - Drop references to supplier devices.
1708  * @dev: Consumer device.
1709  */
1710 void pm_runtime_put_suppliers(struct device *dev)
1711 {
1712 	struct device_link *link;
1713 	int idx;
1714 
1715 	idx = device_links_read_lock();
1716 
1717 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
1718 				device_links_read_lock_held())
1719 		if (link->supplier_preactivated) {
1720 			link->supplier_preactivated = false;
1721 			if (refcount_dec_not_one(&link->rpm_active))
1722 				pm_runtime_put(link->supplier);
1723 		}
1724 
1725 	device_links_read_unlock(idx);
1726 }
1727 
1728 void pm_runtime_new_link(struct device *dev)
1729 {
1730 	spin_lock_irq(&dev->power.lock);
1731 	dev->power.links_count++;
1732 	spin_unlock_irq(&dev->power.lock);
1733 }
1734 
1735 void pm_runtime_drop_link(struct device *dev)
1736 {
1737 	spin_lock_irq(&dev->power.lock);
1738 	WARN_ON(dev->power.links_count == 0);
1739 	dev->power.links_count--;
1740 	spin_unlock_irq(&dev->power.lock);
1741 }
1742 
1743 static bool pm_runtime_need_not_resume(struct device *dev)
1744 {
1745 	return atomic_read(&dev->power.usage_count) <= 1 &&
1746 		(atomic_read(&dev->power.child_count) == 0 ||
1747 		 dev->power.ignore_children);
1748 }
1749 
1750 /**
1751  * pm_runtime_force_suspend - Force a device into suspend state if needed.
1752  * @dev: Device to suspend.
1753  *
1754  * Disable runtime PM so we safely can check the device's runtime PM status and
1755  * if it is active, invoke its ->runtime_suspend callback to suspend it and
1756  * change its runtime PM status field to RPM_SUSPENDED.  Also, if the device's
1757  * usage and children counters don't indicate that the device was in use before
1758  * the system-wide transition under way, decrement its parent's children counter
1759  * (if there is a parent).  Keep runtime PM disabled to preserve the state
1760  * unless we encounter errors.
1761  *
1762  * Typically this function may be invoked from a system suspend callback to make
1763  * sure the device is put into low power state and it should only be used during
1764  * system-wide PM transitions to sleep states.  It assumes that the analogous
1765  * pm_runtime_force_resume() will be used to resume the device.
1766  */
1767 int pm_runtime_force_suspend(struct device *dev)
1768 {
1769 	int (*callback)(struct device *);
1770 	int ret;
1771 
1772 	pm_runtime_disable(dev);
1773 	if (pm_runtime_status_suspended(dev))
1774 		return 0;
1775 
1776 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
1777 
1778 	ret = callback ? callback(dev) : 0;
1779 	if (ret)
1780 		goto err;
1781 
1782 	/*
1783 	 * If the device can stay in suspend after the system-wide transition
1784 	 * to the working state that will follow, drop the children counter of
1785 	 * its parent, but set its status to RPM_SUSPENDED anyway in case this
1786 	 * function will be called again for it in the meantime.
1787 	 */
1788 	if (pm_runtime_need_not_resume(dev))
1789 		pm_runtime_set_suspended(dev);
1790 	else
1791 		__update_runtime_status(dev, RPM_SUSPENDED);
1792 
1793 	return 0;
1794 
1795 err:
1796 	pm_runtime_enable(dev);
1797 	return ret;
1798 }
1799 EXPORT_SYMBOL_GPL(pm_runtime_force_suspend);
1800 
1801 /**
1802  * pm_runtime_force_resume - Force a device into resume state if needed.
1803  * @dev: Device to resume.
1804  *
1805  * Prior invoking this function we expect the user to have brought the device
1806  * into low power state by a call to pm_runtime_force_suspend(). Here we reverse
1807  * those actions and bring the device into full power, if it is expected to be
1808  * used on system resume.  In the other case, we defer the resume to be managed
1809  * via runtime PM.
1810  *
1811  * Typically this function may be invoked from a system resume callback.
1812  */
1813 int pm_runtime_force_resume(struct device *dev)
1814 {
1815 	int (*callback)(struct device *);
1816 	int ret = 0;
1817 
1818 	if (!pm_runtime_status_suspended(dev) || pm_runtime_need_not_resume(dev))
1819 		goto out;
1820 
1821 	/*
1822 	 * The value of the parent's children counter is correct already, so
1823 	 * just update the status of the device.
1824 	 */
1825 	__update_runtime_status(dev, RPM_ACTIVE);
1826 
1827 	callback = RPM_GET_CALLBACK(dev, runtime_resume);
1828 
1829 	ret = callback ? callback(dev) : 0;
1830 	if (ret) {
1831 		pm_runtime_set_suspended(dev);
1832 		goto out;
1833 	}
1834 
1835 	pm_runtime_mark_last_busy(dev);
1836 out:
1837 	pm_runtime_enable(dev);
1838 	return ret;
1839 }
1840 EXPORT_SYMBOL_GPL(pm_runtime_force_resume);
1841