xref: /openbmc/linux/Documentation/power/runtime_pm.rst (revision 11dc486ed5d4626e6b92a23b67ed76cb6c48bfc9)
1==================================================
2Runtime Power Management Framework for I/O Devices
3==================================================
4
5(C) 2009-2011 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
6
7(C) 2010 Alan Stern <stern@rowland.harvard.edu>
8
9(C) 2014 Intel Corp., Rafael J. Wysocki <rafael.j.wysocki@intel.com>
10
111. Introduction
12===============
13
14Support for runtime power management (runtime PM) of I/O devices is provided
15at the power management core (PM core) level by means of:
16
17* The power management workqueue pm_wq in which bus types and device drivers can
18  put their PM-related work items.  It is strongly recommended that pm_wq be
19  used for queuing all work items related to runtime PM, because this allows
20  them to be synchronized with system-wide power transitions (suspend to RAM,
21  hibernation and resume from system sleep states).  pm_wq is declared in
22  include/linux/pm_runtime.h and defined in kernel/power/main.c.
23
24* A number of runtime PM fields in the 'power' member of 'struct device' (which
25  is of the type 'struct dev_pm_info', defined in include/linux/pm.h) that can
26  be used for synchronizing runtime PM operations with one another.
27
28* Three device runtime PM callbacks in 'struct dev_pm_ops' (defined in
29  include/linux/pm.h).
30
31* A set of helper functions defined in drivers/base/power/runtime.c that can be
32  used for carrying out runtime PM operations in such a way that the
33  synchronization between them is taken care of by the PM core.  Bus types and
34  device drivers are encouraged to use these functions.
35
36The runtime PM callbacks present in 'struct dev_pm_ops', the device runtime PM
37fields of 'struct dev_pm_info' and the core helper functions provided for
38runtime PM are described below.
39
402. Device Runtime PM Callbacks
41==============================
42
43There are three device runtime PM callbacks defined in 'struct dev_pm_ops'::
44
45  struct dev_pm_ops {
46	...
47	int (*runtime_suspend)(struct device *dev);
48	int (*runtime_resume)(struct device *dev);
49	int (*runtime_idle)(struct device *dev);
50	...
51  };
52
53The ->runtime_suspend(), ->runtime_resume() and ->runtime_idle() callbacks
54are executed by the PM core for the device's subsystem that may be either of
55the following:
56
57  1. PM domain of the device, if the device's PM domain object, dev->pm_domain,
58     is present.
59
60  2. Device type of the device, if both dev->type and dev->type->pm are present.
61
62  3. Device class of the device, if both dev->class and dev->class->pm are
63     present.
64
65  4. Bus type of the device, if both dev->bus and dev->bus->pm are present.
66
67If the subsystem chosen by applying the above rules doesn't provide the relevant
68callback, the PM core will invoke the corresponding driver callback stored in
69dev->driver->pm directly (if present).
70
71The PM core always checks which callback to use in the order given above, so the
72priority order of callbacks from high to low is: PM domain, device type, class
73and bus type.  Moreover, the high-priority one will always take precedence over
74a low-priority one.  The PM domain, bus type, device type and class callbacks
75are referred to as subsystem-level callbacks in what follows.
76
77By default, the callbacks are always invoked in process context with interrupts
78enabled.  However, the pm_runtime_irq_safe() helper function can be used to tell
79the PM core that it is safe to run the ->runtime_suspend(), ->runtime_resume()
80and ->runtime_idle() callbacks for the given device in atomic context with
81interrupts disabled.  This implies that the callback routines in question must
82not block or sleep, but it also means that the synchronous helper functions
83listed at the end of Section 4 may be used for that device within an interrupt
84handler or generally in an atomic context.
85
86The subsystem-level suspend callback, if present, is _entirely_ _responsible_
87for handling the suspend of the device as appropriate, which may, but need not
88include executing the device driver's own ->runtime_suspend() callback (from the
89PM core's point of view it is not necessary to implement a ->runtime_suspend()
90callback in a device driver as long as the subsystem-level suspend callback
91knows what to do to handle the device).
92
93  * Once the subsystem-level suspend callback (or the driver suspend callback,
94    if invoked directly) has completed successfully for the given device, the PM
95    core regards the device as suspended, which need not mean that it has been
96    put into a low power state.  It is supposed to mean, however, that the
97    device will not process data and will not communicate with the CPU(s) and
98    RAM until the appropriate resume callback is executed for it.  The runtime
99    PM status of a device after successful execution of the suspend callback is
100    'suspended'.
101
102  * If the suspend callback returns -EBUSY or -EAGAIN, the device's runtime PM
103    status remains 'active', which means that the device _must_ be fully
104    operational afterwards.
105
106  * If the suspend callback returns an error code different from -EBUSY and
107    -EAGAIN, the PM core regards this as a fatal error and will refuse to run
108    the helper functions described in Section 4 for the device until its status
109    is directly set to  either 'active', or 'suspended' (the PM core provides
110    special helper functions for this purpose).
111
112In particular, if the driver requires remote wakeup capability (i.e. hardware
113mechanism allowing the device to request a change of its power state, such as
114PCI PME) for proper functioning and device_can_wakeup() returns 'false' for the
115device, then ->runtime_suspend() should return -EBUSY.  On the other hand, if
116device_can_wakeup() returns 'true' for the device and the device is put into a
117low-power state during the execution of the suspend callback, it is expected
118that remote wakeup will be enabled for the device.  Generally, remote wakeup
119should be enabled for all input devices put into low-power states at run time.
120
121The subsystem-level resume callback, if present, is **entirely responsible** for
122handling the resume of the device as appropriate, which may, but need not
123include executing the device driver's own ->runtime_resume() callback (from the
124PM core's point of view it is not necessary to implement a ->runtime_resume()
125callback in a device driver as long as the subsystem-level resume callback knows
126what to do to handle the device).
127
128  * Once the subsystem-level resume callback (or the driver resume callback, if
129    invoked directly) has completed successfully, the PM core regards the device
130    as fully operational, which means that the device _must_ be able to complete
131    I/O operations as needed.  The runtime PM status of the device is then
132    'active'.
133
134  * If the resume callback returns an error code, the PM core regards this as a
135    fatal error and will refuse to run the helper functions described in Section
136    4 for the device, until its status is directly set to either 'active', or
137    'suspended' (by means of special helper functions provided by the PM core
138    for this purpose).
139
140The idle callback (a subsystem-level one, if present, or the driver one) is
141executed by the PM core whenever the device appears to be idle, which is
142indicated to the PM core by two counters, the device's usage counter and the
143counter of 'active' children of the device.
144
145  * If any of these counters is decreased using a helper function provided by
146    the PM core and it turns out to be equal to zero, the other counter is
147    checked.  If that counter also is equal to zero, the PM core executes the
148    idle callback with the device as its argument.
149
150The action performed by the idle callback is totally dependent on the subsystem
151(or driver) in question, but the expected and recommended action is to check
152if the device can be suspended (i.e. if all of the conditions necessary for
153suspending the device are satisfied) and to queue up a suspend request for the
154device in that case.  If there is no idle callback, or if the callback returns
1550, then the PM core will attempt to carry out a runtime suspend of the device,
156also respecting devices configured for autosuspend.  In essence this means a
157call to pm_runtime_autosuspend() (do note that drivers needs to update the
158device last busy mark, pm_runtime_mark_last_busy(), to control the delay under
159this circumstance).  To prevent this (for example, if the callback routine has
160started a delayed suspend), the routine must return a non-zero value.  Negative
161error return codes are ignored by the PM core.
162
163The helper functions provided by the PM core, described in Section 4, guarantee
164that the following constraints are met with respect to runtime PM callbacks for
165one device:
166
167(1) The callbacks are mutually exclusive (e.g. it is forbidden to execute
168    ->runtime_suspend() in parallel with ->runtime_resume() or with another
169    instance of ->runtime_suspend() for the same device) with the exception that
170    ->runtime_suspend() or ->runtime_resume() can be executed in parallel with
171    ->runtime_idle() (although ->runtime_idle() will not be started while any
172    of the other callbacks is being executed for the same device).
173
174(2) ->runtime_idle() and ->runtime_suspend() can only be executed for 'active'
175    devices (i.e. the PM core will only execute ->runtime_idle() or
176    ->runtime_suspend() for the devices the runtime PM status of which is
177    'active').
178
179(3) ->runtime_idle() and ->runtime_suspend() can only be executed for a device
180    the usage counter of which is equal to zero _and_ either the counter of
181    'active' children of which is equal to zero, or the 'power.ignore_children'
182    flag of which is set.
183
184(4) ->runtime_resume() can only be executed for 'suspended' devices  (i.e. the
185    PM core will only execute ->runtime_resume() for the devices the runtime
186    PM status of which is 'suspended').
187
188Additionally, the helper functions provided by the PM core obey the following
189rules:
190
191  * If ->runtime_suspend() is about to be executed or there's a pending request
192    to execute it, ->runtime_idle() will not be executed for the same device.
193
194  * A request to execute or to schedule the execution of ->runtime_suspend()
195    will cancel any pending requests to execute ->runtime_idle() for the same
196    device.
197
198  * If ->runtime_resume() is about to be executed or there's a pending request
199    to execute it, the other callbacks will not be executed for the same device.
200
201  * A request to execute ->runtime_resume() will cancel any pending or
202    scheduled requests to execute the other callbacks for the same device,
203    except for scheduled autosuspends.
204
2053. Runtime PM Device Fields
206===========================
207
208The following device runtime PM fields are present in 'struct dev_pm_info', as
209defined in include/linux/pm.h:
210
211  `struct timer_list suspend_timer;`
212    - timer used for scheduling (delayed) suspend and autosuspend requests
213
214  `unsigned long timer_expires;`
215    - timer expiration time, in jiffies (if this is different from zero, the
216      timer is running and will expire at that time, otherwise the timer is not
217      running)
218
219  `struct work_struct work;`
220    - work structure used for queuing up requests (i.e. work items in pm_wq)
221
222  `wait_queue_head_t wait_queue;`
223    - wait queue used if any of the helper functions needs to wait for another
224      one to complete
225
226  `spinlock_t lock;`
227    - lock used for synchronization
228
229  `atomic_t usage_count;`
230    - the usage counter of the device
231
232  `atomic_t child_count;`
233    - the count of 'active' children of the device
234
235  `unsigned int ignore_children;`
236    - if set, the value of child_count is ignored (but still updated)
237
238  `unsigned int disable_depth;`
239    - used for disabling the helper functions (they work normally if this is
240      equal to zero); the initial value of it is 1 (i.e. runtime PM is
241      initially disabled for all devices)
242
243  `int runtime_error;`
244    - if set, there was a fatal error (one of the callbacks returned error code
245      as described in Section 2), so the helper functions will not work until
246      this flag is cleared; this is the error code returned by the failing
247      callback
248
249  `unsigned int idle_notification;`
250    - if set, ->runtime_idle() is being executed
251
252  `unsigned int request_pending;`
253    - if set, there's a pending request (i.e. a work item queued up into pm_wq)
254
255  `enum rpm_request request;`
256    - type of request that's pending (valid if request_pending is set)
257
258  `unsigned int deferred_resume;`
259    - set if ->runtime_resume() is about to be run while ->runtime_suspend() is
260      being executed for that device and it is not practical to wait for the
261      suspend to complete; means "start a resume as soon as you've suspended"
262
263  `enum rpm_status runtime_status;`
264    - the runtime PM status of the device; this field's initial value is
265      RPM_SUSPENDED, which means that each device is initially regarded by the
266      PM core as 'suspended', regardless of its real hardware status
267
268  `enum rpm_status last_status;`
269    - the last runtime PM status of the device captured before disabling runtime
270      PM for it (invalid initially and when disable_depth is 0)
271
272  `unsigned int runtime_auto;`
273    - if set, indicates that the user space has allowed the device driver to
274      power manage the device at run time via the /sys/devices/.../power/control
275      `interface;` it may only be modified with the help of the
276      pm_runtime_allow() and pm_runtime_forbid() helper functions
277
278  `unsigned int no_callbacks;`
279    - indicates that the device does not use the runtime PM callbacks (see
280      Section 8); it may be modified only by the pm_runtime_no_callbacks()
281      helper function
282
283  `unsigned int irq_safe;`
284    - indicates that the ->runtime_suspend() and ->runtime_resume() callbacks
285      will be invoked with the spinlock held and interrupts disabled
286
287  `unsigned int use_autosuspend;`
288    - indicates that the device's driver supports delayed autosuspend (see
289      Section 9); it may be modified only by the
290      pm_runtime{_dont}_use_autosuspend() helper functions
291
292  `unsigned int timer_autosuspends;`
293    - indicates that the PM core should attempt to carry out an autosuspend
294      when the timer expires rather than a normal suspend
295
296  `int autosuspend_delay;`
297    - the delay time (in milliseconds) to be used for autosuspend
298
299  `unsigned long last_busy;`
300    - the time (in jiffies) when the pm_runtime_mark_last_busy() helper
301      function was last called for this device; used in calculating inactivity
302      periods for autosuspend
303
304All of the above fields are members of the 'power' member of 'struct device'.
305
3064. Runtime PM Device Helper Functions
307=====================================
308
309The following runtime PM helper functions are defined in
310drivers/base/power/runtime.c and include/linux/pm_runtime.h:
311
312  `void pm_runtime_init(struct device *dev);`
313    - initialize the device runtime PM fields in 'struct dev_pm_info'
314
315  `void pm_runtime_remove(struct device *dev);`
316    - make sure that the runtime PM of the device will be disabled after
317      removing the device from device hierarchy
318
319  `int pm_runtime_idle(struct device *dev);`
320    - execute the subsystem-level idle callback for the device; returns an
321      error code on failure, where -EINPROGRESS means that ->runtime_idle() is
322      already being executed; if there is no callback or the callback returns 0
323      then run pm_runtime_autosuspend(dev) and return its result
324
325  `int pm_runtime_suspend(struct device *dev);`
326    - execute the subsystem-level suspend callback for the device; returns 0 on
327      success, 1 if the device's runtime PM status was already 'suspended', or
328      error code on failure, where -EAGAIN or -EBUSY means it is safe to attempt
329      to suspend the device again in future and -EACCES means that
330      'power.disable_depth' is different from 0
331
332  `int pm_runtime_autosuspend(struct device *dev);`
333    - same as pm_runtime_suspend() except that the autosuspend delay is taken
334      `into account;` if pm_runtime_autosuspend_expiration() says the delay has
335      not yet expired then an autosuspend is scheduled for the appropriate time
336      and 0 is returned
337
338  `int pm_runtime_resume(struct device *dev);`
339    - execute the subsystem-level resume callback for the device; returns 0 on
340      success, 1 if the device's runtime PM status is already 'active' (also if
341      'power.disable_depth' is nonzero, but the status was 'active' when it was
342      changing from 0 to 1) or error code on failure, where -EAGAIN means it may
343      be safe to attempt to resume the device again in future, but
344      'power.runtime_error' should be checked additionally, and -EACCES means
345      that the callback could not be run, because 'power.disable_depth' was
346      different from 0
347
348  `int pm_runtime_resume_and_get(struct device *dev);`
349    - run pm_runtime_resume(dev) and if successful, increment the device's
350      usage counter; returns 0 on success (whether or not the device's
351      runtime PM status was already 'active') or the error code from
352      pm_runtime_resume() on failure.
353
354  `int pm_request_idle(struct device *dev);`
355    - submit a request to execute the subsystem-level idle callback for the
356      device (the request is represented by a work item in pm_wq); returns 0 on
357      success or error code if the request has not been queued up
358
359  `int pm_request_autosuspend(struct device *dev);`
360    - schedule the execution of the subsystem-level suspend callback for the
361      device when the autosuspend delay has expired; if the delay has already
362      expired then the work item is queued up immediately
363
364  `int pm_schedule_suspend(struct device *dev, unsigned int delay);`
365    - schedule the execution of the subsystem-level suspend callback for the
366      device in future, where 'delay' is the time to wait before queuing up a
367      suspend work item in pm_wq, in milliseconds (if 'delay' is zero, the work
368      item is queued up immediately); returns 0 on success, 1 if the device's PM
369      runtime status was already 'suspended', or error code if the request
370      hasn't been scheduled (or queued up if 'delay' is 0); if the execution of
371      ->runtime_suspend() is already scheduled and not yet expired, the new
372      value of 'delay' will be used as the time to wait
373
374  `int pm_request_resume(struct device *dev);`
375    - submit a request to execute the subsystem-level resume callback for the
376      device (the request is represented by a work item in pm_wq); returns 0 on
377      success, 1 if the device's runtime PM status was already 'active', or
378      error code if the request hasn't been queued up
379
380  `void pm_runtime_get_noresume(struct device *dev);`
381    - increment the device's usage counter
382
383  `int pm_runtime_get(struct device *dev);`
384    - increment the device's usage counter, run pm_request_resume(dev) and
385      return its result
386
387  `int pm_runtime_get_sync(struct device *dev);`
388    - increment the device's usage counter, run pm_runtime_resume(dev) and
389      return its result;
390      note that it does not drop the device's usage counter on errors, so
391      consider using pm_runtime_resume_and_get() instead of it, especially
392      if its return value is checked by the caller, as this is likely to
393      result in cleaner code.
394
395  `int pm_runtime_get_if_in_use(struct device *dev);`
396    - return -EINVAL if 'power.disable_depth' is nonzero; otherwise, if the
397      runtime PM status is RPM_ACTIVE and the runtime PM usage counter is
398      nonzero, increment the counter and return 1; otherwise return 0 without
399      changing the counter
400
401  `int pm_runtime_get_if_active(struct device *dev, bool ign_usage_count);`
402    - return -EINVAL if 'power.disable_depth' is nonzero; otherwise, if the
403      runtime PM status is RPM_ACTIVE, and either ign_usage_count is true
404      or the device's usage_count is non-zero, increment the counter and
405      return 1; otherwise return 0 without changing the counter
406
407  `void pm_runtime_put_noidle(struct device *dev);`
408    - decrement the device's usage counter
409
410  `int pm_runtime_put(struct device *dev);`
411    - decrement the device's usage counter; if the result is 0 then run
412      pm_request_idle(dev) and return its result
413
414  `int pm_runtime_put_autosuspend(struct device *dev);`
415    - decrement the device's usage counter; if the result is 0 then run
416      pm_request_autosuspend(dev) and return its result
417
418  `int pm_runtime_put_sync(struct device *dev);`
419    - decrement the device's usage counter; if the result is 0 then run
420      pm_runtime_idle(dev) and return its result
421
422  `int pm_runtime_put_sync_suspend(struct device *dev);`
423    - decrement the device's usage counter; if the result is 0 then run
424      pm_runtime_suspend(dev) and return its result
425
426  `int pm_runtime_put_sync_autosuspend(struct device *dev);`
427    - decrement the device's usage counter; if the result is 0 then run
428      pm_runtime_autosuspend(dev) and return its result
429
430  `void pm_runtime_enable(struct device *dev);`
431    - decrement the device's 'power.disable_depth' field; if that field is equal
432      to zero, the runtime PM helper functions can execute subsystem-level
433      callbacks described in Section 2 for the device
434
435  `int pm_runtime_disable(struct device *dev);`
436    - increment the device's 'power.disable_depth' field (if the value of that
437      field was previously zero, this prevents subsystem-level runtime PM
438      callbacks from being run for the device), make sure that all of the
439      pending runtime PM operations on the device are either completed or
440      canceled; returns 1 if there was a resume request pending and it was
441      necessary to execute the subsystem-level resume callback for the device
442      to satisfy that request, otherwise 0 is returned
443
444  `int pm_runtime_barrier(struct device *dev);`
445    - check if there's a resume request pending for the device and resume it
446      (synchronously) in that case, cancel any other pending runtime PM requests
447      regarding it and wait for all runtime PM operations on it in progress to
448      complete; returns 1 if there was a resume request pending and it was
449      necessary to execute the subsystem-level resume callback for the device to
450      satisfy that request, otherwise 0 is returned
451
452  `void pm_suspend_ignore_children(struct device *dev, bool enable);`
453    - set/unset the power.ignore_children flag of the device
454
455  `int pm_runtime_set_active(struct device *dev);`
456    - clear the device's 'power.runtime_error' flag, set the device's runtime
457      PM status to 'active' and update its parent's counter of 'active'
458      children as appropriate (it is only valid to use this function if
459      'power.runtime_error' is set or 'power.disable_depth' is greater than
460      zero); it will fail and return error code if the device has a parent
461      which is not active and the 'power.ignore_children' flag of which is unset
462
463  `void pm_runtime_set_suspended(struct device *dev);`
464    - clear the device's 'power.runtime_error' flag, set the device's runtime
465      PM status to 'suspended' and update its parent's counter of 'active'
466      children as appropriate (it is only valid to use this function if
467      'power.runtime_error' is set or 'power.disable_depth' is greater than
468      zero)
469
470  `bool pm_runtime_active(struct device *dev);`
471    - return true if the device's runtime PM status is 'active' or its
472      'power.disable_depth' field is not equal to zero, or false otherwise
473
474  `bool pm_runtime_suspended(struct device *dev);`
475    - return true if the device's runtime PM status is 'suspended' and its
476      'power.disable_depth' field is equal to zero, or false otherwise
477
478  `bool pm_runtime_status_suspended(struct device *dev);`
479    - return true if the device's runtime PM status is 'suspended'
480
481  `void pm_runtime_allow(struct device *dev);`
482    - set the power.runtime_auto flag for the device and decrease its usage
483      counter (used by the /sys/devices/.../power/control interface to
484      effectively allow the device to be power managed at run time)
485
486  `void pm_runtime_forbid(struct device *dev);`
487    - unset the power.runtime_auto flag for the device and increase its usage
488      counter (used by the /sys/devices/.../power/control interface to
489      effectively prevent the device from being power managed at run time)
490
491  `void pm_runtime_no_callbacks(struct device *dev);`
492    - set the power.no_callbacks flag for the device and remove the runtime
493      PM attributes from /sys/devices/.../power (or prevent them from being
494      added when the device is registered)
495
496  `void pm_runtime_irq_safe(struct device *dev);`
497    - set the power.irq_safe flag for the device, causing the runtime-PM
498      callbacks to be invoked with interrupts off
499
500  `bool pm_runtime_is_irq_safe(struct device *dev);`
501    - return true if power.irq_safe flag was set for the device, causing
502      the runtime-PM callbacks to be invoked with interrupts off
503
504  `void pm_runtime_mark_last_busy(struct device *dev);`
505    - set the power.last_busy field to the current time
506
507  `void pm_runtime_use_autosuspend(struct device *dev);`
508    - set the power.use_autosuspend flag, enabling autosuspend delays; call
509      pm_runtime_get_sync if the flag was previously cleared and
510      power.autosuspend_delay is negative
511
512  `void pm_runtime_dont_use_autosuspend(struct device *dev);`
513    - clear the power.use_autosuspend flag, disabling autosuspend delays;
514      decrement the device's usage counter if the flag was previously set and
515      power.autosuspend_delay is negative; call pm_runtime_idle
516
517  `void pm_runtime_set_autosuspend_delay(struct device *dev, int delay);`
518    - set the power.autosuspend_delay value to 'delay' (expressed in
519      milliseconds); if 'delay' is negative then runtime suspends are
520      prevented; if power.use_autosuspend is set, pm_runtime_get_sync may be
521      called or the device's usage counter may be decremented and
522      pm_runtime_idle called depending on if power.autosuspend_delay is
523      changed to or from a negative value; if power.use_autosuspend is clear,
524      pm_runtime_idle is called
525
526  `unsigned long pm_runtime_autosuspend_expiration(struct device *dev);`
527    - calculate the time when the current autosuspend delay period will expire,
528      based on power.last_busy and power.autosuspend_delay; if the delay time
529      is 1000 ms or larger then the expiration time is rounded up to the
530      nearest second; returns 0 if the delay period has already expired or
531      power.use_autosuspend isn't set, otherwise returns the expiration time
532      in jiffies
533
534It is safe to execute the following helper functions from interrupt context:
535
536- pm_request_idle()
537- pm_request_autosuspend()
538- pm_schedule_suspend()
539- pm_request_resume()
540- pm_runtime_get_noresume()
541- pm_runtime_get()
542- pm_runtime_put_noidle()
543- pm_runtime_put()
544- pm_runtime_put_autosuspend()
545- pm_runtime_enable()
546- pm_suspend_ignore_children()
547- pm_runtime_set_active()
548- pm_runtime_set_suspended()
549- pm_runtime_suspended()
550- pm_runtime_mark_last_busy()
551- pm_runtime_autosuspend_expiration()
552
553If pm_runtime_irq_safe() has been called for a device then the following helper
554functions may also be used in interrupt context:
555
556- pm_runtime_idle()
557- pm_runtime_suspend()
558- pm_runtime_autosuspend()
559- pm_runtime_resume()
560- pm_runtime_get_sync()
561- pm_runtime_put_sync()
562- pm_runtime_put_sync_suspend()
563- pm_runtime_put_sync_autosuspend()
564
5655. Runtime PM Initialization, Device Probing and Removal
566========================================================
567
568Initially, the runtime PM is disabled for all devices, which means that the
569majority of the runtime PM helper functions described in Section 4 will return
570-EAGAIN until pm_runtime_enable() is called for the device.
571
572In addition to that, the initial runtime PM status of all devices is
573'suspended', but it need not reflect the actual physical state of the device.
574Thus, if the device is initially active (i.e. it is able to process I/O), its
575runtime PM status must be changed to 'active', with the help of
576pm_runtime_set_active(), before pm_runtime_enable() is called for the device.
577
578However, if the device has a parent and the parent's runtime PM is enabled,
579calling pm_runtime_set_active() for the device will affect the parent, unless
580the parent's 'power.ignore_children' flag is set.  Namely, in that case the
581parent won't be able to suspend at run time, using the PM core's helper
582functions, as long as the child's status is 'active', even if the child's
583runtime PM is still disabled (i.e. pm_runtime_enable() hasn't been called for
584the child yet or pm_runtime_disable() has been called for it).  For this reason,
585once pm_runtime_set_active() has been called for the device, pm_runtime_enable()
586should be called for it too as soon as reasonably possible or its runtime PM
587status should be changed back to 'suspended' with the help of
588pm_runtime_set_suspended().
589
590If the default initial runtime PM status of the device (i.e. 'suspended')
591reflects the actual state of the device, its bus type's or its driver's
592->probe() callback will likely need to wake it up using one of the PM core's
593helper functions described in Section 4.  In that case, pm_runtime_resume()
594should be used.  Of course, for this purpose the device's runtime PM has to be
595enabled earlier by calling pm_runtime_enable().
596
597Note, if the device may execute pm_runtime calls during the probe (such as
598if it is registered with a subsystem that may call back in) then the
599pm_runtime_get_sync() call paired with a pm_runtime_put() call will be
600appropriate to ensure that the device is not put back to sleep during the
601probe. This can happen with systems such as the network device layer.
602
603It may be desirable to suspend the device once ->probe() has finished.
604Therefore the driver core uses the asynchronous pm_request_idle() to submit a
605request to execute the subsystem-level idle callback for the device at that
606time.  A driver that makes use of the runtime autosuspend feature may want to
607update the last busy mark before returning from ->probe().
608
609Moreover, the driver core prevents runtime PM callbacks from racing with the bus
610notifier callback in __device_release_driver(), which is necessary because the
611notifier is used by some subsystems to carry out operations affecting the
612runtime PM functionality.  It does so by calling pm_runtime_get_sync() before
613driver_sysfs_remove() and the BUS_NOTIFY_UNBIND_DRIVER notifications.  This
614resumes the device if it's in the suspended state and prevents it from
615being suspended again while those routines are being executed.
616
617To allow bus types and drivers to put devices into the suspended state by
618calling pm_runtime_suspend() from their ->remove() routines, the driver core
619executes pm_runtime_put_sync() after running the BUS_NOTIFY_UNBIND_DRIVER
620notifications in __device_release_driver().  This requires bus types and
621drivers to make their ->remove() callbacks avoid races with runtime PM directly,
622but it also allows more flexibility in the handling of devices during the
623removal of their drivers.
624
625Drivers in ->remove() callback should undo the runtime PM changes done
626in ->probe(). Usually this means calling pm_runtime_disable(),
627pm_runtime_dont_use_autosuspend() etc.
628
629The user space can effectively disallow the driver of the device to power manage
630it at run time by changing the value of its /sys/devices/.../power/control
631attribute to "on", which causes pm_runtime_forbid() to be called.  In principle,
632this mechanism may also be used by the driver to effectively turn off the
633runtime power management of the device until the user space turns it on.
634Namely, during the initialization the driver can make sure that the runtime PM
635status of the device is 'active' and call pm_runtime_forbid().  It should be
636noted, however, that if the user space has already intentionally changed the
637value of /sys/devices/.../power/control to "auto" to allow the driver to power
638manage the device at run time, the driver may confuse it by using
639pm_runtime_forbid() this way.
640
6416. Runtime PM and System Sleep
642==============================
643
644Runtime PM and system sleep (i.e., system suspend and hibernation, also known
645as suspend-to-RAM and suspend-to-disk) interact with each other in a couple of
646ways.  If a device is active when a system sleep starts, everything is
647straightforward.  But what should happen if the device is already suspended?
648
649The device may have different wake-up settings for runtime PM and system sleep.
650For example, remote wake-up may be enabled for runtime suspend but disallowed
651for system sleep (device_may_wakeup(dev) returns 'false').  When this happens,
652the subsystem-level system suspend callback is responsible for changing the
653device's wake-up setting (it may leave that to the device driver's system
654suspend routine).  It may be necessary to resume the device and suspend it again
655in order to do so.  The same is true if the driver uses different power levels
656or other settings for runtime suspend and system sleep.
657
658During system resume, the simplest approach is to bring all devices back to full
659power, even if they had been suspended before the system suspend began.  There
660are several reasons for this, including:
661
662  * The device might need to switch power levels, wake-up settings, etc.
663
664  * Remote wake-up events might have been lost by the firmware.
665
666  * The device's children may need the device to be at full power in order
667    to resume themselves.
668
669  * The driver's idea of the device state may not agree with the device's
670    physical state.  This can happen during resume from hibernation.
671
672  * The device might need to be reset.
673
674  * Even though the device was suspended, if its usage counter was > 0 then most
675    likely it would need a runtime resume in the near future anyway.
676
677If the device had been suspended before the system suspend began and it's
678brought back to full power during resume, then its runtime PM status will have
679to be updated to reflect the actual post-system sleep status.  The way to do
680this is:
681
682	 - pm_runtime_disable(dev);
683	 - pm_runtime_set_active(dev);
684	 - pm_runtime_enable(dev);
685
686The PM core always increments the runtime usage counter before calling the
687->suspend() callback and decrements it after calling the ->resume() callback.
688Hence disabling runtime PM temporarily like this will not cause any runtime
689suspend attempts to be permanently lost.  If the usage count goes to zero
690following the return of the ->resume() callback, the ->runtime_idle() callback
691will be invoked as usual.
692
693On some systems, however, system sleep is not entered through a global firmware
694or hardware operation.  Instead, all hardware components are put into low-power
695states directly by the kernel in a coordinated way.  Then, the system sleep
696state effectively follows from the states the hardware components end up in
697and the system is woken up from that state by a hardware interrupt or a similar
698mechanism entirely under the kernel's control.  As a result, the kernel never
699gives control away and the states of all devices during resume are precisely
700known to it.  If that is the case and none of the situations listed above takes
701place (in particular, if the system is not waking up from hibernation), it may
702be more efficient to leave the devices that had been suspended before the system
703suspend began in the suspended state.
704
705To this end, the PM core provides a mechanism allowing some coordination between
706different levels of device hierarchy.  Namely, if a system suspend .prepare()
707callback returns a positive number for a device, that indicates to the PM core
708that the device appears to be runtime-suspended and its state is fine, so it
709may be left in runtime suspend provided that all of its descendants are also
710left in runtime suspend.  If that happens, the PM core will not execute any
711system suspend and resume callbacks for all of those devices, except for the
712.complete() callback, which is then entirely responsible for handling the device
713as appropriate.  This only applies to system suspend transitions that are not
714related to hibernation (see Documentation/driver-api/pm/devices.rst for more
715information).
716
717The PM core does its best to reduce the probability of race conditions between
718the runtime PM and system suspend/resume (and hibernation) callbacks by carrying
719out the following operations:
720
721  * During system suspend pm_runtime_get_noresume() is called for every device
722    right before executing the subsystem-level .prepare() callback for it and
723    pm_runtime_barrier() is called for every device right before executing the
724    subsystem-level .suspend() callback for it.  In addition to that the PM core
725    calls __pm_runtime_disable() with 'false' as the second argument for every
726    device right before executing the subsystem-level .suspend_late() callback
727    for it.
728
729  * During system resume pm_runtime_enable() and pm_runtime_put() are called for
730    every device right after executing the subsystem-level .resume_early()
731    callback and right after executing the subsystem-level .complete() callback
732    for it, respectively.
733
7347. Generic subsystem callbacks
735
736Subsystems may wish to conserve code space by using the set of generic power
737management callbacks provided by the PM core, defined in
738driver/base/power/generic_ops.c:
739
740  `int pm_generic_runtime_suspend(struct device *dev);`
741    - invoke the ->runtime_suspend() callback provided by the driver of this
742      device and return its result, or return 0 if not defined
743
744  `int pm_generic_runtime_resume(struct device *dev);`
745    - invoke the ->runtime_resume() callback provided by the driver of this
746      device and return its result, or return 0 if not defined
747
748  `int pm_generic_suspend(struct device *dev);`
749    - if the device has not been suspended at run time, invoke the ->suspend()
750      callback provided by its driver and return its result, or return 0 if not
751      defined
752
753  `int pm_generic_suspend_noirq(struct device *dev);`
754    - if pm_runtime_suspended(dev) returns "false", invoke the ->suspend_noirq()
755      callback provided by the device's driver and return its result, or return
756      0 if not defined
757
758  `int pm_generic_resume(struct device *dev);`
759    - invoke the ->resume() callback provided by the driver of this device and,
760      if successful, change the device's runtime PM status to 'active'
761
762  `int pm_generic_resume_noirq(struct device *dev);`
763    - invoke the ->resume_noirq() callback provided by the driver of this device
764
765  `int pm_generic_freeze(struct device *dev);`
766    - if the device has not been suspended at run time, invoke the ->freeze()
767      callback provided by its driver and return its result, or return 0 if not
768      defined
769
770  `int pm_generic_freeze_noirq(struct device *dev);`
771    - if pm_runtime_suspended(dev) returns "false", invoke the ->freeze_noirq()
772      callback provided by the device's driver and return its result, or return
773      0 if not defined
774
775  `int pm_generic_thaw(struct device *dev);`
776    - if the device has not been suspended at run time, invoke the ->thaw()
777      callback provided by its driver and return its result, or return 0 if not
778      defined
779
780  `int pm_generic_thaw_noirq(struct device *dev);`
781    - if pm_runtime_suspended(dev) returns "false", invoke the ->thaw_noirq()
782      callback provided by the device's driver and return its result, or return
783      0 if not defined
784
785  `int pm_generic_poweroff(struct device *dev);`
786    - if the device has not been suspended at run time, invoke the ->poweroff()
787      callback provided by its driver and return its result, or return 0 if not
788      defined
789
790  `int pm_generic_poweroff_noirq(struct device *dev);`
791    - if pm_runtime_suspended(dev) returns "false", run the ->poweroff_noirq()
792      callback provided by the device's driver and return its result, or return
793      0 if not defined
794
795  `int pm_generic_restore(struct device *dev);`
796    - invoke the ->restore() callback provided by the driver of this device and,
797      if successful, change the device's runtime PM status to 'active'
798
799  `int pm_generic_restore_noirq(struct device *dev);`
800    - invoke the ->restore_noirq() callback provided by the device's driver
801
802These functions are the defaults used by the PM core if a subsystem doesn't
803provide its own callbacks for ->runtime_idle(), ->runtime_suspend(),
804->runtime_resume(), ->suspend(), ->suspend_noirq(), ->resume(),
805->resume_noirq(), ->freeze(), ->freeze_noirq(), ->thaw(), ->thaw_noirq(),
806->poweroff(), ->poweroff_noirq(), ->restore(), ->restore_noirq() in the
807subsystem-level dev_pm_ops structure.
808
809Device drivers that wish to use the same function as a system suspend, freeze,
810poweroff and runtime suspend callback, and similarly for system resume, thaw,
811restore, and runtime resume, can achieve this with the help of the
812UNIVERSAL_DEV_PM_OPS macro defined in include/linux/pm.h (possibly setting its
813last argument to NULL).
814
8158. "No-Callback" Devices
816========================
817
818Some "devices" are only logical sub-devices of their parent and cannot be
819power-managed on their own.  (The prototype example is a USB interface.  Entire
820USB devices can go into low-power mode or send wake-up requests, but neither is
821possible for individual interfaces.)  The drivers for these devices have no
822need of runtime PM callbacks; if the callbacks did exist, ->runtime_suspend()
823and ->runtime_resume() would always return 0 without doing anything else and
824->runtime_idle() would always call pm_runtime_suspend().
825
826Subsystems can tell the PM core about these devices by calling
827pm_runtime_no_callbacks().  This should be done after the device structure is
828initialized and before it is registered (although after device registration is
829also okay).  The routine will set the device's power.no_callbacks flag and
830prevent the non-debugging runtime PM sysfs attributes from being created.
831
832When power.no_callbacks is set, the PM core will not invoke the
833->runtime_idle(), ->runtime_suspend(), or ->runtime_resume() callbacks.
834Instead it will assume that suspends and resumes always succeed and that idle
835devices should be suspended.
836
837As a consequence, the PM core will never directly inform the device's subsystem
838or driver about runtime power changes.  Instead, the driver for the device's
839parent must take responsibility for telling the device's driver when the
840parent's power state changes.
841
842Note that, in some cases it may not be desirable for subsystems/drivers to call
843pm_runtime_no_callbacks() for their devices. This could be because a subset of
844the runtime PM callbacks needs to be implemented, a platform dependent PM
845domain could get attached to the device or that the device is power managed
846through a supplier device link. For these reasons and to avoid boilerplate code
847in subsystems/drivers, the PM core allows runtime PM callbacks to be
848unassigned. More precisely, if a callback pointer is NULL, the PM core will act
849as though there was a callback and it returned 0.
850
8519. Autosuspend, or automatically-delayed suspends
852=================================================
853
854Changing a device's power state isn't free; it requires both time and energy.
855A device should be put in a low-power state only when there's some reason to
856think it will remain in that state for a substantial time.  A common heuristic
857says that a device which hasn't been used for a while is liable to remain
858unused; following this advice, drivers should not allow devices to be suspended
859at runtime until they have been inactive for some minimum period.  Even when
860the heuristic ends up being non-optimal, it will still prevent devices from
861"bouncing" too rapidly between low-power and full-power states.
862
863The term "autosuspend" is an historical remnant.  It doesn't mean that the
864device is automatically suspended (the subsystem or driver still has to call
865the appropriate PM routines); rather it means that runtime suspends will
866automatically be delayed until the desired period of inactivity has elapsed.
867
868Inactivity is determined based on the power.last_busy field.  Drivers should
869call pm_runtime_mark_last_busy() to update this field after carrying out I/O,
870typically just before calling pm_runtime_put_autosuspend().  The desired length
871of the inactivity period is a matter of policy.  Subsystems can set this length
872initially by calling pm_runtime_set_autosuspend_delay(), but after device
873registration the length should be controlled by user space, using the
874/sys/devices/.../power/autosuspend_delay_ms attribute.
875
876In order to use autosuspend, subsystems or drivers must call
877pm_runtime_use_autosuspend() (preferably before registering the device), and
878thereafter they should use the various `*_autosuspend()` helper functions
879instead of the non-autosuspend counterparts::
880
881	Instead of: pm_runtime_suspend    use: pm_runtime_autosuspend;
882	Instead of: pm_schedule_suspend   use: pm_request_autosuspend;
883	Instead of: pm_runtime_put        use: pm_runtime_put_autosuspend;
884	Instead of: pm_runtime_put_sync   use: pm_runtime_put_sync_autosuspend.
885
886Drivers may also continue to use the non-autosuspend helper functions; they
887will behave normally, which means sometimes taking the autosuspend delay into
888account (see pm_runtime_idle).
889
890Under some circumstances a driver or subsystem may want to prevent a device
891from autosuspending immediately, even though the usage counter is zero and the
892autosuspend delay time has expired.  If the ->runtime_suspend() callback
893returns -EAGAIN or -EBUSY, and if the next autosuspend delay expiration time is
894in the future (as it normally would be if the callback invoked
895pm_runtime_mark_last_busy()), the PM core will automatically reschedule the
896autosuspend.  The ->runtime_suspend() callback can't do this rescheduling
897itself because no suspend requests of any kind are accepted while the device is
898suspending (i.e., while the callback is running).
899
900The implementation is well suited for asynchronous use in interrupt contexts.
901However such use inevitably involves races, because the PM core can't
902synchronize ->runtime_suspend() callbacks with the arrival of I/O requests.
903This synchronization must be handled by the driver, using its private lock.
904Here is a schematic pseudo-code example::
905
906	foo_read_or_write(struct foo_priv *foo, void *data)
907	{
908		lock(&foo->private_lock);
909		add_request_to_io_queue(foo, data);
910		if (foo->num_pending_requests++ == 0)
911			pm_runtime_get(&foo->dev);
912		if (!foo->is_suspended)
913			foo_process_next_request(foo);
914		unlock(&foo->private_lock);
915	}
916
917	foo_io_completion(struct foo_priv *foo, void *req)
918	{
919		lock(&foo->private_lock);
920		if (--foo->num_pending_requests == 0) {
921			pm_runtime_mark_last_busy(&foo->dev);
922			pm_runtime_put_autosuspend(&foo->dev);
923		} else {
924			foo_process_next_request(foo);
925		}
926		unlock(&foo->private_lock);
927		/* Send req result back to the user ... */
928	}
929
930	int foo_runtime_suspend(struct device *dev)
931	{
932		struct foo_priv foo = container_of(dev, ...);
933		int ret = 0;
934
935		lock(&foo->private_lock);
936		if (foo->num_pending_requests > 0) {
937			ret = -EBUSY;
938		} else {
939			/* ... suspend the device ... */
940			foo->is_suspended = 1;
941		}
942		unlock(&foo->private_lock);
943		return ret;
944	}
945
946	int foo_runtime_resume(struct device *dev)
947	{
948		struct foo_priv foo = container_of(dev, ...);
949
950		lock(&foo->private_lock);
951		/* ... resume the device ... */
952		foo->is_suspended = 0;
953		pm_runtime_mark_last_busy(&foo->dev);
954		if (foo->num_pending_requests > 0)
955			foo_process_next_request(foo);
956		unlock(&foo->private_lock);
957		return 0;
958	}
959
960The important point is that after foo_io_completion() asks for an autosuspend,
961the foo_runtime_suspend() callback may race with foo_read_or_write().
962Therefore foo_runtime_suspend() has to check whether there are any pending I/O
963requests (while holding the private lock) before allowing the suspend to
964proceed.
965
966In addition, the power.autosuspend_delay field can be changed by user space at
967any time.  If a driver cares about this, it can call
968pm_runtime_autosuspend_expiration() from within the ->runtime_suspend()
969callback while holding its private lock.  If the function returns a nonzero
970value then the delay has not yet expired and the callback should return
971-EAGAIN.
972