1Runtime Power Management Framework for I/O Devices
3(C) 2009-2011 Rafael J. Wysocki <>, Novell Inc.
4(C) 2010 Alan Stern <>
61. Introduction
8Support for runtime power management (runtime PM) of I/O devices is provided
9at the power management core (PM core) level by means of:
11* The power management workqueue pm_wq in which bus types and device drivers can
12  put their PM-related work items. It is strongly recommended that pm_wq be
13  used for queuing all work items related to runtime PM, because this allows
14  them to be synchronized with system-wide power transitions (suspend to RAM,
15  hibernation and resume from system sleep states). pm_wq is declared in
16  include/linux/pm_runtime.h and defined in kernel/power/main.c.
18* A number of runtime PM fields in the 'power' member of 'struct device' (which
19  is of the type 'struct dev_pm_info', defined in include/linux/pm.h) that can
20  be used for synchronizing runtime PM operations with one another.
22* Three device runtime PM callbacks in 'struct dev_pm_ops' (defined in
23  include/linux/pm.h).
25* A set of helper functions defined in drivers/base/power/runtime.c that can be
26  used for carrying out runtime PM operations in such a way that the
27  synchronization between them is taken care of by the PM core. Bus types and
28  device drivers are encouraged to use these functions.
30The runtime PM callbacks present in 'struct dev_pm_ops', the device runtime PM
31fields of 'struct dev_pm_info' and the core helper functions provided for
32runtime PM are described below.
342. Device Runtime PM Callbacks
36There are three device runtime PM callbacks defined in 'struct dev_pm_ops':
38struct dev_pm_ops {
39    ...
40    int (*runtime_suspend)(struct device *dev);
41    int (*runtime_resume)(struct device *dev);
42    int (*runtime_idle)(struct device *dev);
43    ...
46The ->runtime_suspend(), ->runtime_resume() and ->runtime_idle() callbacks are
47executed by the PM core for either the device type, or the class (if the device
48type's struct dev_pm_ops object does not exist), or the bus type (if the
49device type's and class' struct dev_pm_ops objects do not exist) of the given
50device (this allows device types to override callbacks provided by bus types or
51classes if necessary). The bus type, device type and class callbacks are
52referred to as subsystem-level callbacks in what follows.
54By default, the callbacks are always invoked in process context with interrupts
55enabled. However, subsystems can use the pm_runtime_irq_safe() helper function
56to tell the PM core that a device's ->runtime_suspend() and ->runtime_resume()
57callbacks should be invoked in atomic context with interrupts disabled.
58This implies that these callback routines must not block or sleep, but it also
59means that the synchronous helper functions listed at the end of Section 4 can
60be used within an interrupt handler or in an atomic context.
62The subsystem-level suspend callback is _entirely_ _responsible_ for handling
63the suspend of the device as appropriate, which may, but need not include
64executing the device driver's own ->runtime_suspend() callback (from the
65PM core's point of view it is not necessary to implement a ->runtime_suspend()
66callback in a device driver as long as the subsystem-level suspend callback
67knows what to do to handle the device).
69  * Once the subsystem-level suspend callback has completed successfully
70    for given device, the PM core regards the device as suspended, which need
71    not mean that the device has been put into a low power state. It is
72    supposed to mean, however, that the device will not process data and will
73    not communicate with the CPU(s) and RAM until the subsystem-level resume
74    callback is executed for it. The runtime PM status of a device after
75    successful execution of the subsystem-level suspend callback is 'suspended'.
77  * If the subsystem-level suspend callback returns -EBUSY or -EAGAIN,
78    the device's runtime PM status is 'active', which means that the device
79    _must_ be fully operational afterwards.
81  * If the subsystem-level suspend callback returns an error code different
82    from -EBUSY or -EAGAIN, the PM core regards this as a fatal error and will
83    refuse to run the helper functions described in Section 4 for the device,
84    until the status of it is directly set either to 'active', or to 'suspended'
85    (the PM core provides special helper functions for this purpose).
87In particular, if the driver requires remote wake-up capability (i.e. hardware
88mechanism allowing the device to request a change of its power state, such as
89PCI PME) for proper functioning and device_run_wake() returns 'false' for the
90device, then ->runtime_suspend() should return -EBUSY. On the other hand, if
91device_run_wake() returns 'true' for the device and the device is put into a low
92power state during the execution of the subsystem-level suspend callback, it is
93expected that remote wake-up will be enabled for the device. Generally, remote
94wake-up should be enabled for all input devices put into a low power state at
95run time.
97The subsystem-level resume callback is _entirely_ _responsible_ for handling the
98resume of the device as appropriate, which may, but need not include executing
99the device driver's own ->runtime_resume() callback (from the PM core's point of
100view it is not necessary to implement a ->runtime_resume() callback in a device
101driver as long as the subsystem-level resume callback knows what to do to handle
102the device).
104  * Once the subsystem-level resume callback has completed successfully, the PM
105    core regards the device as fully operational, which means that the device
106    _must_ be able to complete I/O operations as needed. The runtime PM status
107    of the device is then 'active'.
109  * If the subsystem-level resume callback returns an error code, the PM core
110    regards this as a fatal error and will refuse to run the helper functions
111    described in Section 4 for the device, until its status is directly set
112    either to 'active' or to 'suspended' (the PM core provides special helper
113    functions for this purpose).
115The subsystem-level idle callback is executed by the PM core whenever the device
116appears to be idle, which is indicated to the PM core by two counters, the
117device's usage counter and the counter of 'active' children of the device.
119  * If any of these counters is decreased using a helper function provided by
120    the PM core and it turns out to be equal to zero, the other counter is
121    checked. If that counter also is equal to zero, the PM core executes the
122    subsystem-level idle callback with the device as an argument.
124The action performed by a subsystem-level idle callback is totally dependent on
125the subsystem in question, but the expected and recommended action is to check
126if the device can be suspended (i.e. if all of the conditions necessary for
127suspending the device are satisfied) and to queue up a suspend request for the
128device in that case. The value returned by this callback is ignored by the PM
131The helper functions provided by the PM core, described in Section 4, guarantee
132that the following constraints are met with respect to the bus type's runtime
133PM callbacks:
135(1) The callbacks are mutually exclusive (e.g. it is forbidden to execute
136    ->runtime_suspend() in parallel with ->runtime_resume() or with another
137    instance of ->runtime_suspend() for the same device) with the exception that
138    ->runtime_suspend() or ->runtime_resume() can be executed in parallel with
139    ->runtime_idle() (although ->runtime_idle() will not be started while any
140    of the other callbacks is being executed for the same device).
142(2) ->runtime_idle() and ->runtime_suspend() can only be executed for 'active'
143    devices (i.e. the PM core will only execute ->runtime_idle() or
144    ->runtime_suspend() for the devices the runtime PM status of which is
145    'active').
147(3) ->runtime_idle() and ->runtime_suspend() can only be executed for a device
148    the usage counter of which is equal to zero _and_ either the counter of
149    'active' children of which is equal to zero, or the 'power.ignore_children'
150    flag of which is set.
152(4) ->runtime_resume() can only be executed for 'suspended' devices (i.e. the
153    PM core will only execute ->runtime_resume() for the devices the runtime
154    PM status of which is 'suspended').
156Additionally, the helper functions provided by the PM core obey the following
159  * If ->runtime_suspend() is about to be executed or there's a pending request
160    to execute it, ->runtime_idle() will not be executed for the same device.
162  * A request to execute or to schedule the execution of ->runtime_suspend()
163    will cancel any pending requests to execute ->runtime_idle() for the same
164    device.
166  * If ->runtime_resume() is about to be executed or there's a pending request
167    to execute it, the other callbacks will not be executed for the same device.
169  * A request to execute ->runtime_resume() will cancel any pending or
170    scheduled requests to execute the other callbacks for the same device,
171    except for scheduled autosuspends.
1733. Runtime PM Device Fields
175The following device runtime PM fields are present in 'struct dev_pm_info', as
176defined in include/linux/pm.h:
178  struct timer_list suspend_timer;
179    - timer used for scheduling (delayed) suspend and autosuspend requests
181  unsigned long timer_expires;
182    - timer expiration time, in jiffies (if this is different from zero, the
183      timer is running and will expire at that time, otherwise the timer is not
184      running)
186  struct work_struct work;
187    - work structure used for queuing up requests (i.e. work items in pm_wq)
189  wait_queue_head_t wait_queue;
190    - wait queue used if any of the helper functions needs to wait for another
191      one to complete
193  spinlock_t lock;
194    - lock used for synchronisation
196  atomic_t usage_count;
197    - the usage counter of the device
199  atomic_t child_count;
200    - the count of 'active' children of the device
202  unsigned int ignore_children;
203    - if set, the value of child_count is ignored (but still updated)
205  unsigned int disable_depth;
206    - used for disabling the helper funcions (they work normally if this is
207      equal to zero); the initial value of it is 1 (i.e. runtime PM is
208      initially disabled for all devices)
210  unsigned int runtime_error;
211    - if set, there was a fatal error (one of the callbacks returned error code
212      as described in Section 2), so the helper funtions will not work until
213      this flag is cleared; this is the error code returned by the failing
214      callback
216  unsigned int idle_notification;
217    - if set, ->runtime_idle() is being executed
219  unsigned int request_pending;
220    - if set, there's a pending request (i.e. a work item queued up into pm_wq)
222  enum rpm_request request;
223    - type of request that's pending (valid if request_pending is set)
225  unsigned int deferred_resume;
226    - set if ->runtime_resume() is about to be run while ->runtime_suspend() is
227      being executed for that device and it is not practical to wait for the
228      suspend to complete; means "start a resume as soon as you've suspended"
230  unsigned int run_wake;
231    - set if the device is capable of generating runtime wake-up events
233  enum rpm_status runtime_status;
234    - the runtime PM status of the device; this field's initial value is
235      RPM_SUSPENDED, which means that each device is initially regarded by the
236      PM core as 'suspended', regardless of its real hardware status
238  unsigned int runtime_auto;
239    - if set, indicates that the user space has allowed the device driver to
240      power manage the device at run time via the /sys/devices/.../power/control
241      interface; it may only be modified with the help of the pm_runtime_allow()
242      and pm_runtime_forbid() helper functions
244  unsigned int no_callbacks;
245    - indicates that the device does not use the runtime PM callbacks (see
246      Section 8); it may be modified only by the pm_runtime_no_callbacks()
247      helper function
249  unsigned int irq_safe;
250    - indicates that the ->runtime_suspend() and ->runtime_resume() callbacks
251      will be invoked with the spinlock held and interrupts disabled
253  unsigned int use_autosuspend;
254    - indicates that the device's driver supports delayed autosuspend (see
255      Section 9); it may be modified only by the
256      pm_runtime{_dont}_use_autosuspend() helper functions
258  unsigned int timer_autosuspends;
259    - indicates that the PM core should attempt to carry out an autosuspend
260      when the timer expires rather than a normal suspend
262  int autosuspend_delay;
263    - the delay time (in milliseconds) to be used for autosuspend
265  unsigned long last_busy;
266    - the time (in jiffies) when the pm_runtime_mark_last_busy() helper
267      function was last called for this device; used in calculating inactivity
268      periods for autosuspend
270All of the above fields are members of the 'power' member of 'struct device'.
2724. Runtime PM Device Helper Functions
274The following runtime PM helper functions are defined in
275drivers/base/power/runtime.c and include/linux/pm_runtime.h:
277  void pm_runtime_init(struct device *dev);
278    - initialize the device runtime PM fields in 'struct dev_pm_info'
280  void pm_runtime_remove(struct device *dev);
281    - make sure that the runtime PM of the device will be disabled after
282      removing the device from device hierarchy
284  int pm_runtime_idle(struct device *dev);
285    - execute the subsystem-level idle callback for the device; returns 0 on
286      success or error code on failure, where -EINPROGRESS means that
287      ->runtime_idle() is already being executed
289  int pm_runtime_suspend(struct device *dev);
290    - execute the subsystem-level suspend callback for the device; returns 0 on
291      success, 1 if the device's runtime PM status was already 'suspended', or
292      error code on failure, where -EAGAIN or -EBUSY means it is safe to attempt
293      to suspend the device again in future and -EACCES means that
294      'power.disable_depth' is different from 0
296  int pm_runtime_autosuspend(struct device *dev);
297    - same as pm_runtime_suspend() except that the autosuspend delay is taken
298      into account; if pm_runtime_autosuspend_expiration() says the delay has
299      not yet expired then an autosuspend is scheduled for the appropriate time
300      and 0 is returned
302  int pm_runtime_resume(struct device *dev);
303    - execute the subsystem-level resume callback for the device; returns 0 on
304      success, 1 if the device's runtime PM status was already 'active' or
305      error code on failure, where -EAGAIN means it may be safe to attempt to
306      resume the device again in future, but 'power.runtime_error' should be
307      checked additionally, and -EACCES means that 'power.disable_depth' is
308      different from 0
310  int pm_request_idle(struct device *dev);
311    - submit a request to execute the subsystem-level idle callback for the
312      device (the request is represented by a work item in pm_wq); returns 0 on
313      success or error code if the request has not been queued up
315  int pm_request_autosuspend(struct device *dev);
316    - schedule the execution of the subsystem-level suspend callback for the
317      device when the autosuspend delay has expired; if the delay has already
318      expired then the work item is queued up immediately
320  int pm_schedule_suspend(struct device *dev, unsigned int delay);
321    - schedule the execution of the subsystem-level suspend callback for the
322      device in future, where 'delay' is the time to wait before queuing up a
323      suspend work item in pm_wq, in milliseconds (if 'delay' is zero, the work
324      item is queued up immediately); returns 0 on success, 1 if the device's PM
325      runtime status was already 'suspended', or error code if the request
326      hasn't been scheduled (or queued up if 'delay' is 0); if the execution of
327      ->runtime_suspend() is already scheduled and not yet expired, the new
328      value of 'delay' will be used as the time to wait
330  int pm_request_resume(struct device *dev);
331    - submit a request to execute the subsystem-level resume callback for the
332      device (the request is represented by a work item in pm_wq); returns 0 on
333      success, 1 if the device's runtime PM status was already 'active', or
334      error code if the request hasn't been queued up
336  void pm_runtime_get_noresume(struct device *dev);
337    - increment the device's usage counter
339  int pm_runtime_get(struct device *dev);
340    - increment the device's usage counter, run pm_request_resume(dev) and
341      return its result
343  int pm_runtime_get_sync(struct device *dev);
344    - increment the device's usage counter, run pm_runtime_resume(dev) and
345      return its result
347  void pm_runtime_put_noidle(struct device *dev);
348    - decrement the device's usage counter
350  int pm_runtime_put(struct device *dev);
351    - decrement the device's usage counter; if the result is 0 then run
352      pm_request_idle(dev) and return its result
354  int pm_runtime_put_autosuspend(struct device *dev);
355    - decrement the device's usage counter; if the result is 0 then run
356      pm_request_autosuspend(dev) and return its result
358  int pm_runtime_put_sync(struct device *dev);
359    - decrement the device's usage counter; if the result is 0 then run
360      pm_runtime_idle(dev) and return its result
362  int pm_runtime_put_sync_suspend(struct device *dev);
363    - decrement the device's usage counter; if the result is 0 then run
364      pm_runtime_suspend(dev) and return its result
366  int pm_runtime_put_sync_autosuspend(struct device *dev);
367    - decrement the device's usage counter; if the result is 0 then run
368      pm_runtime_autosuspend(dev) and return its result
370  void pm_runtime_enable(struct device *dev);
371    - decrement the device's 'power.disable_depth' field; if that field is equal
372      to zero, the runtime PM helper functions can execute subsystem-level
373      callbacks described in Section 2 for the device
375  int pm_runtime_disable(struct device *dev);
376    - increment the device's 'power.disable_depth' field (if the value of that
377      field was previously zero, this prevents subsystem-level runtime PM
378      callbacks from being run for the device), make sure that all of the pending
379      runtime PM operations on the device are either completed or canceled;
380      returns 1 if there was a resume request pending and it was necessary to
381      execute the subsystem-level resume callback for the device to satisfy that
382      request, otherwise 0 is returned
384  int pm_runtime_barrier(struct device *dev);
385    - check if there's a resume request pending for the device and resume it
386      (synchronously) in that case, cancel any other pending runtime PM requests
387      regarding it and wait for all runtime PM operations on it in progress to
388      complete; returns 1 if there was a resume request pending and it was
389      necessary to execute the subsystem-level resume callback for the device to
390      satisfy that request, otherwise 0 is returned
392  void pm_suspend_ignore_children(struct device *dev, bool enable);
393    - set/unset the power.ignore_children flag of the device
395  int pm_runtime_set_active(struct device *dev);
396    - clear the device's 'power.runtime_error' flag, set the device's runtime
397      PM status to 'active' and update its parent's counter of 'active'
398      children as appropriate (it is only valid to use this function if
399      'power.runtime_error' is set or 'power.disable_depth' is greater than
400      zero); it will fail and return error code if the device has a parent
401      which is not active and the 'power.ignore_children' flag of which is unset
403  void pm_runtime_set_suspended(struct device *dev);
404    - clear the device's 'power.runtime_error' flag, set the device's runtime
405      PM status to 'suspended' and update its parent's counter of 'active'
406      children as appropriate (it is only valid to use this function if
407      'power.runtime_error' is set or 'power.disable_depth' is greater than
408      zero)
410  bool pm_runtime_suspended(struct device *dev);
411    - return true if the device's runtime PM status is 'suspended' and its
412      'power.disable_depth' field is equal to zero, or false otherwise
414  bool pm_runtime_status_suspended(struct device *dev);
415    - return true if the device's runtime PM status is 'suspended'
417  void pm_runtime_allow(struct device *dev);
418    - set the power.runtime_auto flag for the device and decrease its usage
419      counter (used by the /sys/devices/.../power/control interface to
420      effectively allow the device to be power managed at run time)
422  void pm_runtime_forbid(struct device *dev);
423    - unset the power.runtime_auto flag for the device and increase its usage
424      counter (used by the /sys/devices/.../power/control interface to
425      effectively prevent the device from being power managed at run time)
427  void pm_runtime_no_callbacks(struct device *dev);
428    - set the power.no_callbacks flag for the device and remove the runtime
429      PM attributes from /sys/devices/.../power (or prevent them from being
430      added when the device is registered)
432  void pm_runtime_irq_safe(struct device *dev);
433    - set the power.irq_safe flag for the device, causing the runtime-PM
434      callbacks to be invoked with interrupts off
436  void pm_runtime_mark_last_busy(struct device *dev);
437    - set the power.last_busy field to the current time
439  void pm_runtime_use_autosuspend(struct device *dev);
440    - set the power.use_autosuspend flag, enabling autosuspend delays
442  void pm_runtime_dont_use_autosuspend(struct device *dev);
443    - clear the power.use_autosuspend flag, disabling autosuspend delays
445  void pm_runtime_set_autosuspend_delay(struct device *dev, int delay);
446    - set the power.autosuspend_delay value to 'delay' (expressed in
447      milliseconds); if 'delay' is negative then runtime suspends are
448      prevented
450  unsigned long pm_runtime_autosuspend_expiration(struct device *dev);
451    - calculate the time when the current autosuspend delay period will expire,
452      based on power.last_busy and power.autosuspend_delay; if the delay time
453      is 1000 ms or larger then the expiration time is rounded up to the
454      nearest second; returns 0 if the delay period has already expired or
455      power.use_autosuspend isn't set, otherwise returns the expiration time
456      in jiffies
458It is safe to execute the following helper functions from interrupt context:
477If pm_runtime_irq_safe() has been called for a device then the following helper
478functions may also be used in interrupt context:
4875. Runtime PM Initialization, Device Probing and Removal
489Initially, the runtime PM is disabled for all devices, which means that the
490majority of the runtime PM helper funtions described in Section 4 will return
491-EAGAIN until pm_runtime_enable() is called for the device.
493In addition to that, the initial runtime PM status of all devices is
494'suspended', but it need not reflect the actual physical state of the device.
495Thus, if the device is initially active (i.e. it is able to process I/O), its
496runtime PM status must be changed to 'active', with the help of
497pm_runtime_set_active(), before pm_runtime_enable() is called for the device.
499However, if the device has a parent and the parent's runtime PM is enabled,
500calling pm_runtime_set_active() for the device will affect the parent, unless
501the parent's 'power.ignore_children' flag is set. Namely, in that case the
502parent won't be able to suspend at run time, using the PM core's helper
503functions, as long as the child's status is 'active', even if the child's
504runtime PM is still disabled (i.e. pm_runtime_enable() hasn't been called for
505the child yet or pm_runtime_disable() has been called for it). For this reason,
506once pm_runtime_set_active() has been called for the device, pm_runtime_enable()
507should be called for it too as soon as reasonably possible or its runtime PM
508status should be changed back to 'suspended' with the help of
511If the default initial runtime PM status of the device (i.e. 'suspended')
512reflects the actual state of the device, its bus type's or its driver's
513->probe() callback will likely need to wake it up using one of the PM core's
514helper functions described in Section 4. In that case, pm_runtime_resume()
515should be used. Of course, for this purpose the device's runtime PM has to be
516enabled earlier by calling pm_runtime_enable().
518If the device bus type's or driver's ->probe() callback runs
519pm_runtime_suspend() or pm_runtime_idle() or their asynchronous counterparts,
520they will fail returning -EAGAIN, because the device's usage counter is
521incremented by the driver core before executing ->probe(). Still, it may be
522desirable to suspend the device as soon as ->probe() has finished, so the driver
523core uses pm_runtime_put_sync() to invoke the subsystem-level idle callback for
524the device at that time.
526Moreover, the driver core prevents runtime PM callbacks from racing with the bus
527notifier callback in __device_release_driver(), which is necessary, because the
528notifier is used by some subsystems to carry out operations affecting the
529runtime PM functionality. It does so by calling pm_runtime_get_sync() before
530driver_sysfs_remove() and the BUS_NOTIFY_UNBIND_DRIVER notifications. This
531resumes the device if it's in the suspended state and prevents it from
532being suspended again while those routines are being executed.
534To allow bus types and drivers to put devices into the suspended state by
535calling pm_runtime_suspend() from their ->remove() routines, the driver core
536executes pm_runtime_put_sync() after running the BUS_NOTIFY_UNBIND_DRIVER
537notifications in __device_release_driver(). This requires bus types and
538drivers to make their ->remove() callbacks avoid races with runtime PM directly,
539but also it allows of more flexibility in the handling of devices during the
540removal of their drivers.
542The user space can effectively disallow the driver of the device to power manage
543it at run time by changing the value of its /sys/devices/.../power/control
544attribute to "on", which causes pm_runtime_forbid() to be called. In principle,
545this mechanism may also be used by the driver to effectively turn off the
546runtime power management of the device until the user space turns it on.
547Namely, during the initialization the driver can make sure that the runtime PM
548status of the device is 'active' and call pm_runtime_forbid(). It should be
549noted, however, that if the user space has already intentionally changed the
550value of /sys/devices/.../power/control to "auto" to allow the driver to power
551manage the device at run time, the driver may confuse it by using
552pm_runtime_forbid() this way.
5546. Runtime PM and System Sleep
556Runtime PM and system sleep (i.e., system suspend and hibernation, also known
557as suspend-to-RAM and suspend-to-disk) interact with each other in a couple of
558ways. If a device is active when a system sleep starts, everything is
559straightforward. But what should happen if the device is already suspended?
561The device may have different wake-up settings for runtime PM and system sleep.
562For example, remote wake-up may be enabled for runtime suspend but disallowed
563for system sleep (device_may_wakeup(dev) returns 'false'). When this happens,
564the subsystem-level system suspend callback is responsible for changing the
565device's wake-up setting (it may leave that to the device driver's system
566suspend routine). It may be necessary to resume the device and suspend it again
567in order to do so. The same is true if the driver uses different power levels
568or other settings for runtime suspend and system sleep.
570During system resume, the simplest approach is to bring all devices back to full
571power, even if they had been suspended before the system suspend began. There
572are several reasons for this, including:
574  * The device might need to switch power levels, wake-up settings, etc.
576  * Remote wake-up events might have been lost by the firmware.
578  * The device's children may need the device to be at full power in order
579    to resume themselves.
581  * The driver's idea of the device state may not agree with the device's
582    physical state. This can happen during resume from hibernation.
584  * The device might need to be reset.
586  * Even though the device was suspended, if its usage counter was > 0 then most
587    likely it would need a runtime resume in the near future anyway.
589If the device had been suspended before the system suspend began and it's
590brought back to full power during resume, then its runtime PM status will have
591to be updated to reflect the actual post-system sleep status. The way to do
592this is:
594    pm_runtime_disable(dev);
595    pm_runtime_set_active(dev);
596    pm_runtime_enable(dev);
598The PM core always increments the runtime usage counter before calling the
599->suspend() callback and decrements it after calling the ->resume() callback.
600Hence disabling runtime PM temporarily like this will not cause any runtime
601suspend attempts to be permanently lost. If the usage count goes to zero
602following the return of the ->resume() callback, the ->runtime_idle() callback
603will be invoked as usual.
605On some systems, however, system sleep is not entered through a global firmware
606or hardware operation. Instead, all hardware components are put into low-power
607states directly by the kernel in a coordinated way. Then, the system sleep
608state effectively follows from the states the hardware components end up in
609and the system is woken up from that state by a hardware interrupt or a similar
610mechanism entirely under the kernel's control. As a result, the kernel never
611gives control away and the states of all devices during resume are precisely
612known to it. If that is the case and none of the situations listed above takes
613place (in particular, if the system is not waking up from hibernation), it may
614be more efficient to leave the devices that had been suspended before the system
615suspend began in the suspended state.
617The PM core does its best to reduce the probability of race conditions between
618the runtime PM and system suspend/resume (and hibernation) callbacks by carrying
619out the following operations:
621  * During system suspend it calls pm_runtime_get_noresume() and
622    pm_runtime_barrier() for every device right before executing the
623    subsystem-level .suspend() callback for it. In addition to that it calls
624    pm_runtime_disable() for every device right after executing the
625    subsystem-level .suspend() callback for it.
627  * During system resume it calls pm_runtime_enable() and pm_runtime_put_sync()
628    for every device right before and right after executing the subsystem-level
629    .resume() callback for it, respectively.
6317. Generic subsystem callbacks
633Subsystems may wish to conserve code space by using the set of generic power
634management callbacks provided by the PM core, defined in
637  int pm_generic_runtime_idle(struct device *dev);
638    - invoke the ->runtime_idle() callback provided by the driver of this
639      device, if defined, and call pm_runtime_suspend() for this device if the
640      return value is 0 or the callback is not defined
642  int pm_generic_runtime_suspend(struct device *dev);
643    - invoke the ->runtime_suspend() callback provided by the driver of this
644      device and return its result, or return -EINVAL if not defined
646  int pm_generic_runtime_resume(struct device *dev);
647    - invoke the ->runtime_resume() callback provided by the driver of this
648      device and return its result, or return -EINVAL if not defined
650  int pm_generic_suspend(struct device *dev);
651    - if the device has not been suspended at run time, invoke the ->suspend()
652      callback provided by its driver and return its result, or return 0 if not
653      defined
655  int pm_generic_suspend_noirq(struct device *dev);
656    - if pm_runtime_suspended(dev) returns "false", invoke the ->suspend_noirq()
657      callback provided by the device's driver and return its result, or return
658      0 if not defined
660  int pm_generic_resume(struct device *dev);
661    - invoke the ->resume() callback provided by the driver of this device and,
662      if successful, change the device's runtime PM status to 'active'
664  int pm_generic_resume_noirq(struct device *dev);
665    - invoke the ->resume_noirq() callback provided by the driver of this device
667  int pm_generic_freeze(struct device *dev);
668    - if the device has not been suspended at run time, invoke the ->freeze()
669      callback provided by its driver and return its result, or return 0 if not
670      defined
672  int pm_generic_freeze_noirq(struct device *dev);
673    - if pm_runtime_suspended(dev) returns "false", invoke the ->freeze_noirq()
674      callback provided by the device's driver and return its result, or return
675      0 if not defined
677  int pm_generic_thaw(struct device *dev);
678    - if the device has not been suspended at run time, invoke the ->thaw()
679      callback provided by its driver and return its result, or return 0 if not
680      defined
682  int pm_generic_thaw_noirq(struct device *dev);
683    - if pm_runtime_suspended(dev) returns "false", invoke the ->thaw_noirq()
684      callback provided by the device's driver and return its result, or return
685      0 if not defined
687  int pm_generic_poweroff(struct device *dev);
688    - if the device has not been suspended at run time, invoke the ->poweroff()
689      callback provided by its driver and return its result, or return 0 if not
690      defined
692  int pm_generic_poweroff_noirq(struct device *dev);
693    - if pm_runtime_suspended(dev) returns "false", run the ->poweroff_noirq()
694      callback provided by the device's driver and return its result, or return
695      0 if not defined
697  int pm_generic_restore(struct device *dev);
698    - invoke the ->restore() callback provided by the driver of this device and,
699      if successful, change the device's runtime PM status to 'active'
701  int pm_generic_restore_noirq(struct device *dev);
702    - invoke the ->restore_noirq() callback provided by the device's driver
704These functions can be assigned to the ->runtime_idle(), ->runtime_suspend(),
705->runtime_resume(), ->suspend(), ->suspend_noirq(), ->resume(),
706->resume_noirq(), ->freeze(), ->freeze_noirq(), ->thaw(), ->thaw_noirq(),
707->poweroff(), ->poweroff_noirq(), ->restore(), ->restore_noirq() callback
708pointers in the subsystem-level dev_pm_ops structures.
710If a subsystem wishes to use all of them at the same time, it can simply assign
711the GENERIC_SUBSYS_PM_OPS macro, defined in include/linux/pm.h, to its
712dev_pm_ops structure pointer.
714Device drivers that wish to use the same function as a system suspend, freeze,
715poweroff and runtime suspend callback, and similarly for system resume, thaw,
716restore, and runtime resume, can achieve this with the help of the
717UNIVERSAL_DEV_PM_OPS macro defined in include/linux/pm.h (possibly setting its
718last argument to NULL).
7208. "No-Callback" Devices
722Some "devices" are only logical sub-devices of their parent and cannot be
723power-managed on their own. (The prototype example is a USB interface. Entire
724USB devices can go into low-power mode or send wake-up requests, but neither is
725possible for individual interfaces.) The drivers for these devices have no
726need of runtime PM callbacks; if the callbacks did exist, ->runtime_suspend()
727and ->runtime_resume() would always return 0 without doing anything else and
728->runtime_idle() would always call pm_runtime_suspend().
730Subsystems can tell the PM core about these devices by calling
731pm_runtime_no_callbacks(). This should be done after the device structure is
732initialized and before it is registered (although after device registration is
733also okay). The routine will set the device's power.no_callbacks flag and
734prevent the non-debugging runtime PM sysfs attributes from being created.
736When power.no_callbacks is set, the PM core will not invoke the
737->runtime_idle(), ->runtime_suspend(), or ->runtime_resume() callbacks.
738Instead it will assume that suspends and resumes always succeed and that idle
739devices should be suspended.
741As a consequence, the PM core will never directly inform the device's subsystem
742or driver about runtime power changes. Instead, the driver for the device's
743parent must take responsibility for telling the device's driver when the
744parent's power state changes.
7469. Autosuspend, or automatically-delayed suspends
748Changing a device's power state isn't free; it requires both time and energy.
749A device should be put in a low-power state only when there's some reason to
750think it will remain in that state for a substantial time. A common heuristic
751says that a device which hasn't been used for a while is liable to remain
752unused; following this advice, drivers should not allow devices to be suspended
753at runtime until they have been inactive for some minimum period. Even when
754the heuristic ends up being non-optimal, it will still prevent devices from
755"bouncing" too rapidly between low-power and full-power states.
757The term "autosuspend" is an historical remnant. It doesn't mean that the
758device is automatically suspended (the subsystem or driver still has to call
759the appropriate PM routines); rather it means that runtime suspends will
760automatically be delayed until the desired period of inactivity has elapsed.
762Inactivity is determined based on the power.last_busy field. Drivers should
763call pm_runtime_mark_last_busy() to update this field after carrying out I/O,
764typically just before calling pm_runtime_put_autosuspend(). The desired length
765of the inactivity period is a matter of policy. Subsystems can set this length
766initially by calling pm_runtime_set_autosuspend_delay(), but after device
767registration the length should be controlled by user space, using the
768/sys/devices/.../power/autosuspend_delay_ms attribute.
770In order to use autosuspend, subsystems or drivers must call
771pm_runtime_use_autosuspend() (preferably before registering the device), and
772thereafter they should use the various *_autosuspend() helper functions instead
773of the non-autosuspend counterparts:
775    Instead of: pm_runtime_suspend use: pm_runtime_autosuspend;
776    Instead of: pm_schedule_suspend use: pm_request_autosuspend;
777    Instead of: pm_runtime_put use: pm_runtime_put_autosuspend;
778    Instead of: pm_runtime_put_sync use: pm_runtime_put_sync_autosuspend.
780Drivers may also continue to use the non-autosuspend helper functions; they
781will behave normally, not taking the autosuspend delay into account.
782Similarly, if the power.use_autosuspend field isn't set then the autosuspend
783helper functions will behave just like the non-autosuspend counterparts.
785The implementation is well suited for asynchronous use in interrupt contexts.
786However such use inevitably involves races, because the PM core can't
787synchronize ->runtime_suspend() callbacks with the arrival of I/O requests.
788This synchronization must be handled by the driver, using its private lock.
789Here is a schematic pseudo-code example:
791    foo_read_or_write(struct foo_priv *foo, void *data)
792    {
793        lock(&foo->private_lock);
794        add_request_to_io_queue(foo, data);
795        if (foo->num_pending_requests++ == 0)
796            pm_runtime_get(&foo->dev);
797        if (!foo->is_suspended)
798            foo_process_next_request(foo);
799        unlock(&foo->private_lock);
800    }
802    foo_io_completion(struct foo_priv *foo, void *req)
803    {
804        lock(&foo->private_lock);
805        if (--foo->num_pending_requests == 0) {
806            pm_runtime_mark_last_busy(&foo->dev);
807            pm_runtime_put_autosuspend(&foo->dev);
808        } else {
809            foo_process_next_request(foo);
810        }
811        unlock(&foo->private_lock);
812        /* Send req result back to the user ... */
813    }
815    int foo_runtime_suspend(struct device *dev)
816    {
817        struct foo_priv foo = container_of(dev, ...);
818        int ret = 0;
820        lock(&foo->private_lock);
821        if (foo->num_pending_requests > 0) {
822            ret = -EBUSY;
823        } else {
824            /* ... suspend the device ... */
825            foo->is_suspended = 1;
826        }
827        unlock(&foo->private_lock);
828        return ret;
829    }
831    int foo_runtime_resume(struct device *dev)
832    {
833        struct foo_priv foo = container_of(dev, ...);
835        lock(&foo->private_lock);
836        /* ... resume the device ... */
837        foo->is_suspended = 0;
838        pm_runtime_mark_last_busy(&foo->dev);
839        if (foo->num_pending_requests > 0)
840            foo_process_requests(foo);
841        unlock(&foo->private_lock);
842        return 0;
843    }
845The important point is that after foo_io_completion() asks for an autosuspend,
846the foo_runtime_suspend() callback may race with foo_read_or_write().
847Therefore foo_runtime_suspend() has to check whether there are any pending I/O
848requests (while holding the private lock) before allowing the suspend to
851In addition, the power.autosuspend_delay field can be changed by user space at
852any time. If a driver cares about this, it can call
853pm_runtime_autosuspend_expiration() from within the ->runtime_suspend()
854callback while holding its private lock. If the function returns a nonzero
855value then the delay has not yet expired and the callback should return

Archive Download this file