linux/Documentation/power/runtime_pm.txt
<<
>>
Prefs
   1Runtime Power Management Framework for I/O Devices
   2
   3(C) 2009-2011 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
   4(C) 2010 Alan Stern <stern@rowland.harvard.edu>
   5
   61. Introduction
   7
   8Support for runtime power management (runtime PM) of I/O devices is provided
   9at the power management core (PM core) level by means of:
  10
  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.
  17
  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.
  21
  22* Three device runtime PM callbacks in 'struct dev_pm_ops' (defined in
  23  include/linux/pm.h).
  24
  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.
  29
  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.
  33
  342. Device Runtime PM Callbacks
  35
  36There are three device runtime PM callbacks defined in 'struct dev_pm_ops':
  37
  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        ...
  44};
  45
  46The ->runtime_suspend(), ->runtime_resume() and ->runtime_idle() callbacks
  47are executed by the PM core for the device's subsystem that may be either of
  48the following:
  49
  50  1. PM domain of the device, if the device's PM domain object, dev->pm_domain,
  51     is present.
  52
  53  2. Device type of the device, if both dev->type and dev->type->pm are present.
  54
  55  3. Device class of the device, if both dev->class and dev->class->pm are
  56     present.
  57
  58  4. Bus type of the device, if both dev->bus and dev->bus->pm are present.
  59
  60If the subsystem chosen by applying the above rules doesn't provide the relevant
  61callback, the PM core will invoke the corresponding driver callback stored in
  62dev->driver->pm directly (if present).
  63
  64The PM core always checks which callback to use in the order given above, so the
  65priority order of callbacks from high to low is: PM domain, device type, class
  66and bus type.  Moreover, the high-priority one will always take precedence over
  67a low-priority one.  The PM domain, bus type, device type and class callbacks
  68are referred to as subsystem-level callbacks in what follows.
  69
  70By default, the callbacks are always invoked in process context with interrupts
  71enabled.  However, the pm_runtime_irq_safe() helper function can be used to tell
  72the PM core that it is safe to run the ->runtime_suspend(), ->runtime_resume()
  73and ->runtime_idle() callbacks for the given device in atomic context with
  74interrupts disabled.  This implies that the callback routines in question must
  75not block or sleep, but it also means that the synchronous helper functions
  76listed at the end of Section 4 may be used for that device within an interrupt
  77handler or generally in an atomic context.
  78
  79The subsystem-level suspend callback, if present, is _entirely_ _responsible_
  80for handling the suspend of the device as appropriate, which may, but need not
  81include executing the device driver's own ->runtime_suspend() callback (from the
  82PM core's point of view it is not necessary to implement a ->runtime_suspend()
  83callback in a device driver as long as the subsystem-level suspend callback
  84knows what to do to handle the device).
  85
  86  * Once the subsystem-level suspend callback (or the driver suspend callback,
  87    if invoked directly) has completed successfully for the given device, the PM
  88    core regards the device as suspended, which need not mean that it has been
  89    put into a low power state.  It is supposed to mean, however, that the
  90    device will not process data and will not communicate with the CPU(s) and
  91    RAM until the appropriate resume callback is executed for it.  The runtime
  92    PM status of a device after successful execution of the suspend callback is
  93    'suspended'.
  94
  95  * If the suspend callback returns -EBUSY or -EAGAIN, the device's runtime PM
  96    status remains 'active', which means that the device _must_ be fully
  97    operational afterwards.
  98
  99  * If the suspend callback returns an error code different from -EBUSY and
 100    -EAGAIN, the PM core regards this as a fatal error and will refuse to run
 101    the helper functions described in Section 4 for the device until its status
 102    is directly set to  either'active', or 'suspended' (the PM core provides
 103    special helper functions for this purpose).
 104
 105In particular, if the driver requires remote wakeup capability (i.e. hardware
 106mechanism allowing the device to request a change of its power state, such as
 107PCI PME) for proper functioning and device_run_wake() returns 'false' for the
 108device, then ->runtime_suspend() should return -EBUSY.  On the other hand, if
 109device_run_wake() returns 'true' for the device and the device is put into a
 110low-power state during the execution of the suspend callback, it is expected
 111that remote wakeup will be enabled for the device.  Generally, remote wakeup
 112should be enabled for all input devices put into low-power states at run time.
 113
 114The subsystem-level resume callback, if present, is _entirely_ _responsible_ for
 115handling the resume of the device as appropriate, which may, but need not
 116include executing the device driver's own ->runtime_resume() callback (from the
 117PM core's point of view it is not necessary to implement a ->runtime_resume()
 118callback in a device driver as long as the subsystem-level resume callback knows
 119what to do to handle the device).
 120
 121  * Once the subsystem-level resume callback (or the driver resume callback, if
 122    invoked directly) has completed successfully, the PM core regards the device
 123    as fully operational, which means that the device _must_ be able to complete
 124    I/O operations as needed.  The runtime PM status of the device is then
 125    'active'.
 126
 127  * If the resume callback returns an error code, the PM core regards this as a
 128    fatal error and will refuse to run the helper functions described in Section
 129    4 for the device, until its status is directly set to either 'active', or
 130    'suspended' (by means of special helper functions provided by the PM core
 131    for this purpose).
 132
 133The idle callback (a subsystem-level one, if present, or the driver one) is
 134executed by the PM core whenever the device appears to be idle, which is
 135indicated to the PM core by two counters, the device's usage counter and the
 136counter of 'active' children of the device.
 137
 138  * If any of these counters is decreased using a helper function provided by
 139    the PM core and it turns out to be equal to zero, the other counter is
 140    checked.  If that counter also is equal to zero, the PM core executes the
 141    idle callback with the device as its argument.
 142
 143The action performed by the idle callback is totally dependent on the subsystem
 144(or driver) in question, but the expected and recommended action is to check
 145if the device can be suspended (i.e. if all of the conditions necessary for
 146suspending the device are satisfied) and to queue up a suspend request for the
 147device in that case.  The value returned by this callback is ignored by the PM
 148core.
 149
 150The helper functions provided by the PM core, described in Section 4, guarantee
 151that the following constraints are met with respect to runtime PM callbacks for
 152one device:
 153
 154(1) The callbacks are mutually exclusive (e.g. it is forbidden to execute
 155    ->runtime_suspend() in parallel with ->runtime_resume() or with another
 156    instance of ->runtime_suspend() for the same device) with the exception that
 157    ->runtime_suspend() or ->runtime_resume() can be executed in parallel with
 158    ->runtime_idle() (although ->runtime_idle() will not be started while any
 159    of the other callbacks is being executed for the same device).
 160
 161(2) ->runtime_idle() and ->runtime_suspend() can only be executed for 'active'
 162    devices (i.e. the PM core will only execute ->runtime_idle() or
 163    ->runtime_suspend() for the devices the runtime PM status of which is
 164    'active').
 165
 166(3) ->runtime_idle() and ->runtime_suspend() can only be executed for a device
 167    the usage counter of which is equal to zero _and_ either the counter of
 168    'active' children of which is equal to zero, or the 'power.ignore_children'
 169    flag of which is set.
 170
 171(4) ->runtime_resume() can only be executed for 'suspended' devices  (i.e. the
 172    PM core will only execute ->runtime_resume() for the devices the runtime
 173    PM status of which is 'suspended').
 174
 175Additionally, the helper functions provided by the PM core obey the following
 176rules:
 177
 178  * If ->runtime_suspend() is about to be executed or there's a pending request
 179    to execute it, ->runtime_idle() will not be executed for the same device.
 180
 181  * A request to execute or to schedule the execution of ->runtime_suspend()
 182    will cancel any pending requests to execute ->runtime_idle() for the same
 183    device.
 184
 185  * If ->runtime_resume() is about to be executed or there's a pending request
 186    to execute it, the other callbacks will not be executed for the same device.
 187
 188  * A request to execute ->runtime_resume() will cancel any pending or
 189    scheduled requests to execute the other callbacks for the same device,
 190    except for scheduled autosuspends.
 191
 1923. Runtime PM Device Fields
 193
 194The following device runtime PM fields are present in 'struct dev_pm_info', as
 195defined in include/linux/pm.h:
 196
 197  struct timer_list suspend_timer;
 198    - timer used for scheduling (delayed) suspend and autosuspend requests
 199
 200  unsigned long timer_expires;
 201    - timer expiration time, in jiffies (if this is different from zero, the
 202      timer is running and will expire at that time, otherwise the timer is not
 203      running)
 204
 205  struct work_struct work;
 206    - work structure used for queuing up requests (i.e. work items in pm_wq)
 207
 208  wait_queue_head_t wait_queue;
 209    - wait queue used if any of the helper functions needs to wait for another
 210      one to complete
 211
 212  spinlock_t lock;
 213    - lock used for synchronisation
 214
 215  atomic_t usage_count;
 216    - the usage counter of the device
 217
 218  atomic_t child_count;
 219    - the count of 'active' children of the device
 220
 221  unsigned int ignore_children;
 222    - if set, the value of child_count is ignored (but still updated)
 223
 224  unsigned int disable_depth;
 225    - used for disabling the helper funcions (they work normally if this is
 226      equal to zero); the initial value of it is 1 (i.e. runtime PM is
 227      initially disabled for all devices)
 228
 229  unsigned int runtime_error;
 230    - if set, there was a fatal error (one of the callbacks returned error code
 231      as described in Section 2), so the helper funtions will not work until
 232      this flag is cleared; this is the error code returned by the failing
 233      callback
 234
 235  unsigned int idle_notification;
 236    - if set, ->runtime_idle() is being executed
 237
 238  unsigned int request_pending;
 239    - if set, there's a pending request (i.e. a work item queued up into pm_wq)
 240
 241  enum rpm_request request;
 242    - type of request that's pending (valid if request_pending is set)
 243
 244  unsigned int deferred_resume;
 245    - set if ->runtime_resume() is about to be run while ->runtime_suspend() is
 246      being executed for that device and it is not practical to wait for the
 247      suspend to complete; means "start a resume as soon as you've suspended"
 248
 249  unsigned int run_wake;
 250    - set if the device is capable of generating runtime wake-up events
 251
 252  enum rpm_status runtime_status;
 253    - the runtime PM status of the device; this field's initial value is
 254      RPM_SUSPENDED, which means that each device is initially regarded by the
 255      PM core as 'suspended', regardless of its real hardware status
 256
 257  unsigned int runtime_auto;
 258    - if set, indicates that the user space has allowed the device driver to
 259      power manage the device at run time via the /sys/devices/.../power/control
 260      interface; it may only be modified with the help of the pm_runtime_allow()
 261      and pm_runtime_forbid() helper functions
 262
 263  unsigned int no_callbacks;
 264    - indicates that the device does not use the runtime PM callbacks (see
 265      Section 8); it may be modified only by the pm_runtime_no_callbacks()
 266      helper function
 267
 268  unsigned int irq_safe;
 269    - indicates that the ->runtime_suspend() and ->runtime_resume() callbacks
 270      will be invoked with the spinlock held and interrupts disabled
 271
 272  unsigned int use_autosuspend;
 273    - indicates that the device's driver supports delayed autosuspend (see
 274      Section 9); it may be modified only by the
 275      pm_runtime{_dont}_use_autosuspend() helper functions
 276
 277  unsigned int timer_autosuspends;
 278    - indicates that the PM core should attempt to carry out an autosuspend
 279      when the timer expires rather than a normal suspend
 280
 281  int autosuspend_delay;
 282    - the delay time (in milliseconds) to be used for autosuspend
 283
 284  unsigned long last_busy;
 285    - the time (in jiffies) when the pm_runtime_mark_last_busy() helper
 286      function was last called for this device; used in calculating inactivity
 287      periods for autosuspend
 288
 289All of the above fields are members of the 'power' member of 'struct device'.
 290
 2914. Runtime PM Device Helper Functions
 292
 293The following runtime PM helper functions are defined in
 294drivers/base/power/runtime.c and include/linux/pm_runtime.h:
 295
 296  void pm_runtime_init(struct device *dev);
 297    - initialize the device runtime PM fields in 'struct dev_pm_info'
 298
 299  void pm_runtime_remove(struct device *dev);
 300    - make sure that the runtime PM of the device will be disabled after
 301      removing the device from device hierarchy
 302
 303  int pm_runtime_idle(struct device *dev);
 304    - execute the subsystem-level idle callback for the device; returns 0 on
 305      success or error code on failure, where -EINPROGRESS means that
 306      ->runtime_idle() is already being executed
 307
 308  int pm_runtime_suspend(struct device *dev);
 309    - execute the subsystem-level suspend callback for the device; returns 0 on
 310      success, 1 if the device's runtime PM status was already 'suspended', or
 311      error code on failure, where -EAGAIN or -EBUSY means it is safe to attempt
 312      to suspend the device again in future and -EACCES means that
 313      'power.disable_depth' is different from 0
 314
 315  int pm_runtime_autosuspend(struct device *dev);
 316    - same as pm_runtime_suspend() except that the autosuspend delay is taken
 317      into account; if pm_runtime_autosuspend_expiration() says the delay has
 318      not yet expired then an autosuspend is scheduled for the appropriate time
 319      and 0 is returned
 320
 321  int pm_runtime_resume(struct device *dev);
 322    - execute the subsystem-level resume callback for the device; returns 0 on
 323      success, 1 if the device's runtime PM status was already 'active' or
 324      error code on failure, where -EAGAIN means it may be safe to attempt to
 325      resume the device again in future, but 'power.runtime_error' should be
 326      checked additionally, and -EACCES means that 'power.disable_depth' is
 327      different from 0
 328
 329  int pm_request_idle(struct device *dev);
 330    - submit a request to execute the subsystem-level idle callback for the
 331      device (the request is represented by a work item in pm_wq); returns 0 on
 332      success or error code if the request has not been queued up
 333
 334  int pm_request_autosuspend(struct device *dev);
 335    - schedule the execution of the subsystem-level suspend callback for the
 336      device when the autosuspend delay has expired; if the delay has already
 337      expired then the work item is queued up immediately
 338
 339  int pm_schedule_suspend(struct device *dev, unsigned int delay);
 340    - schedule the execution of the subsystem-level suspend callback for the
 341      device in future, where 'delay' is the time to wait before queuing up a
 342      suspend work item in pm_wq, in milliseconds (if 'delay' is zero, the work
 343      item is queued up immediately); returns 0 on success, 1 if the device's PM
 344      runtime status was already 'suspended', or error code if the request
 345      hasn't been scheduled (or queued up if 'delay' is 0); if the execution of
 346      ->runtime_suspend() is already scheduled and not yet expired, the new
 347      value of 'delay' will be used as the time to wait
 348
 349  int pm_request_resume(struct device *dev);
 350    - submit a request to execute the subsystem-level resume callback for the
 351      device (the request is represented by a work item in pm_wq); returns 0 on
 352      success, 1 if the device's runtime PM status was already 'active', or
 353      error code if the request hasn't been queued up
 354
 355  void pm_runtime_get_noresume(struct device *dev);
 356    - increment the device's usage counter
 357
 358  int pm_runtime_get(struct device *dev);
 359    - increment the device's usage counter, run pm_request_resume(dev) and
 360      return its result
 361
 362  int pm_runtime_get_sync(struct device *dev);
 363    - increment the device's usage counter, run pm_runtime_resume(dev) and
 364      return its result
 365
 366  void pm_runtime_put_noidle(struct device *dev);
 367    - decrement the device's usage counter
 368
 369  int pm_runtime_put(struct device *dev);
 370    - decrement the device's usage counter; if the result is 0 then run
 371      pm_request_idle(dev) and return its result
 372
 373  int pm_runtime_put_autosuspend(struct device *dev);
 374    - decrement the device's usage counter; if the result is 0 then run
 375      pm_request_autosuspend(dev) and return its result
 376
 377  int pm_runtime_put_sync(struct device *dev);
 378    - decrement the device's usage counter; if the result is 0 then run
 379      pm_runtime_idle(dev) and return its result
 380
 381  int pm_runtime_put_sync_suspend(struct device *dev);
 382    - decrement the device's usage counter; if the result is 0 then run
 383      pm_runtime_suspend(dev) and return its result
 384
 385  int pm_runtime_put_sync_autosuspend(struct device *dev);
 386    - decrement the device's usage counter; if the result is 0 then run
 387      pm_runtime_autosuspend(dev) and return its result
 388
 389  void pm_runtime_enable(struct device *dev);
 390    - decrement the device's 'power.disable_depth' field; if that field is equal
 391      to zero, the runtime PM helper functions can execute subsystem-level
 392      callbacks described in Section 2 for the device
 393
 394  int pm_runtime_disable(struct device *dev);
 395    - increment the device's 'power.disable_depth' field (if the value of that
 396      field was previously zero, this prevents subsystem-level runtime PM
 397      callbacks from being run for the device), make sure that all of the pending
 398      runtime PM operations on the device are either completed or canceled;
 399      returns 1 if there was a resume request pending and it was necessary to
 400      execute the subsystem-level resume callback for the device to satisfy that
 401      request, otherwise 0 is returned
 402
 403  int pm_runtime_barrier(struct device *dev);
 404    - check if there's a resume request pending for the device and resume it
 405      (synchronously) in that case, cancel any other pending runtime PM requests
 406      regarding it and wait for all runtime PM operations on it in progress to
 407      complete; returns 1 if there was a resume request pending and it was
 408      necessary to execute the subsystem-level resume callback for the device to
 409      satisfy that request, otherwise 0 is returned
 410
 411  void pm_suspend_ignore_children(struct device *dev, bool enable);
 412    - set/unset the power.ignore_children flag of the device
 413
 414  int pm_runtime_set_active(struct device *dev);
 415    - clear the device's 'power.runtime_error' flag, set the device's runtime
 416      PM status to 'active' and update its parent's counter of 'active'
 417      children as appropriate (it is only valid to use this function if
 418      'power.runtime_error' is set or 'power.disable_depth' is greater than
 419      zero); it will fail and return error code if the device has a parent
 420      which is not active and the 'power.ignore_children' flag of which is unset
 421
 422  void pm_runtime_set_suspended(struct device *dev);
 423    - clear the device's 'power.runtime_error' flag, set the device's runtime
 424      PM status to 'suspended' and update its parent's counter of 'active'
 425      children as appropriate (it is only valid to use this function if
 426      'power.runtime_error' is set or 'power.disable_depth' is greater than
 427      zero)
 428
 429  bool pm_runtime_suspended(struct device *dev);
 430    - return true if the device's runtime PM status is 'suspended' and its
 431      'power.disable_depth' field is equal to zero, or false otherwise
 432
 433  bool pm_runtime_status_suspended(struct device *dev);
 434    - return true if the device's runtime PM status is 'suspended'
 435
 436  void pm_runtime_allow(struct device *dev);
 437    - set the power.runtime_auto flag for the device and decrease its usage
 438      counter (used by the /sys/devices/.../power/control interface to
 439      effectively allow the device to be power managed at run time)
 440
 441  void pm_runtime_forbid(struct device *dev);
 442    - unset the power.runtime_auto flag for the device and increase its usage
 443      counter (used by the /sys/devices/.../power/control interface to
 444      effectively prevent the device from being power managed at run time)
 445
 446  void pm_runtime_no_callbacks(struct device *dev);
 447    - set the power.no_callbacks flag for the device and remove the runtime
 448      PM attributes from /sys/devices/.../power (or prevent them from being
 449      added when the device is registered)
 450
 451  void pm_runtime_irq_safe(struct device *dev);
 452    - set the power.irq_safe flag for the device, causing the runtime-PM
 453      callbacks to be invoked with interrupts off
 454
 455  void pm_runtime_mark_last_busy(struct device *dev);
 456    - set the power.last_busy field to the current time
 457
 458  void pm_runtime_use_autosuspend(struct device *dev);
 459    - set the power.use_autosuspend flag, enabling autosuspend delays
 460
 461  void pm_runtime_dont_use_autosuspend(struct device *dev);
 462    - clear the power.use_autosuspend flag, disabling autosuspend delays
 463
 464  void pm_runtime_set_autosuspend_delay(struct device *dev, int delay);
 465    - set the power.autosuspend_delay value to 'delay' (expressed in
 466      milliseconds); if 'delay' is negative then runtime suspends are
 467      prevented
 468
 469  unsigned long pm_runtime_autosuspend_expiration(struct device *dev);
 470    - calculate the time when the current autosuspend delay period will expire,
 471      based on power.last_busy and power.autosuspend_delay; if the delay time
 472      is 1000 ms or larger then the expiration time is rounded up to the
 473      nearest second; returns 0 if the delay period has already expired or
 474      power.use_autosuspend isn't set, otherwise returns the expiration time
 475      in jiffies
 476
 477It is safe to execute the following helper functions from interrupt context:
 478
 479pm_request_idle()
 480pm_request_autosuspend()
 481pm_schedule_suspend()
 482pm_request_resume()
 483pm_runtime_get_noresume()
 484pm_runtime_get()
 485pm_runtime_put_noidle()
 486pm_runtime_put()
 487pm_runtime_put_autosuspend()
 488pm_runtime_enable()
 489pm_suspend_ignore_children()
 490pm_runtime_set_active()
 491pm_runtime_set_suspended()
 492pm_runtime_suspended()
 493pm_runtime_mark_last_busy()
 494pm_runtime_autosuspend_expiration()
 495
 496If pm_runtime_irq_safe() has been called for a device then the following helper
 497functions may also be used in interrupt context:
 498
 499pm_runtime_idle()
 500pm_runtime_suspend()
 501pm_runtime_autosuspend()
 502pm_runtime_resume()
 503pm_runtime_get_sync()
 504pm_runtime_put_sync()
 505pm_runtime_put_sync_suspend()
 506pm_runtime_put_sync_autosuspend()
 507
 5085. Runtime PM Initialization, Device Probing and Removal
 509
 510Initially, the runtime PM is disabled for all devices, which means that the
 511majority of the runtime PM helper funtions described in Section 4 will return
 512-EAGAIN until pm_runtime_enable() is called for the device.
 513
 514In addition to that, the initial runtime PM status of all devices is
 515'suspended', but it need not reflect the actual physical state of the device.
 516Thus, if the device is initially active (i.e. it is able to process I/O), its
 517runtime PM status must be changed to 'active', with the help of
 518pm_runtime_set_active(), before pm_runtime_enable() is called for the device.
 519
 520However, if the device has a parent and the parent's runtime PM is enabled,
 521calling pm_runtime_set_active() for the device will affect the parent, unless
 522the parent's 'power.ignore_children' flag is set.  Namely, in that case the
 523parent won't be able to suspend at run time, using the PM core's helper
 524functions, as long as the child's status is 'active', even if the child's
 525runtime PM is still disabled (i.e. pm_runtime_enable() hasn't been called for
 526the child yet or pm_runtime_disable() has been called for it).  For this reason,
 527once pm_runtime_set_active() has been called for the device, pm_runtime_enable()
 528should be called for it too as soon as reasonably possible or its runtime PM
 529status should be changed back to 'suspended' with the help of
 530pm_runtime_set_suspended().
 531
 532If the default initial runtime PM status of the device (i.e. 'suspended')
 533reflects the actual state of the device, its bus type's or its driver's
 534->probe() callback will likely need to wake it up using one of the PM core's
 535helper functions described in Section 4.  In that case, pm_runtime_resume()
 536should be used.  Of course, for this purpose the device's runtime PM has to be
 537enabled earlier by calling pm_runtime_enable().
 538
 539If the device bus type's or driver's ->probe() callback runs
 540pm_runtime_suspend() or pm_runtime_idle() or their asynchronous counterparts,
 541they will fail returning -EAGAIN, because the device's usage counter is
 542incremented by the driver core before executing ->probe().  Still, it may be
 543desirable to suspend the device as soon as ->probe() has finished, so the driver
 544core uses pm_runtime_put_sync() to invoke the subsystem-level idle callback for
 545the device at that time.
 546
 547Moreover, the driver core prevents runtime PM callbacks from racing with the bus
 548notifier callback in __device_release_driver(), which is necessary, because the
 549notifier is used by some subsystems to carry out operations affecting the
 550runtime PM functionality.  It does so by calling pm_runtime_get_sync() before
 551driver_sysfs_remove() and the BUS_NOTIFY_UNBIND_DRIVER notifications.  This
 552resumes the device if it's in the suspended state and prevents it from
 553being suspended again while those routines are being executed.
 554
 555To allow bus types and drivers to put devices into the suspended state by
 556calling pm_runtime_suspend() from their ->remove() routines, the driver core
 557executes pm_runtime_put_sync() after running the BUS_NOTIFY_UNBIND_DRIVER
 558notifications in __device_release_driver().  This requires bus types and
 559drivers to make their ->remove() callbacks avoid races with runtime PM directly,
 560but also it allows of more flexibility in the handling of devices during the
 561removal of their drivers.
 562
 563The user space can effectively disallow the driver of the device to power manage
 564it at run time by changing the value of its /sys/devices/.../power/control
 565attribute to "on", which causes pm_runtime_forbid() to be called.  In principle,
 566this mechanism may also be used by the driver to effectively turn off the
 567runtime power management of the device until the user space turns it on.
 568Namely, during the initialization the driver can make sure that the runtime PM
 569status of the device is 'active' and call pm_runtime_forbid().  It should be
 570noted, however, that if the user space has already intentionally changed the
 571value of /sys/devices/.../power/control to "auto" to allow the driver to power
 572manage the device at run time, the driver may confuse it by using
 573pm_runtime_forbid() this way.
 574
 5756. Runtime PM and System Sleep
 576
 577Runtime PM and system sleep (i.e., system suspend and hibernation, also known
 578as suspend-to-RAM and suspend-to-disk) interact with each other in a couple of
 579ways.  If a device is active when a system sleep starts, everything is
 580straightforward.  But what should happen if the device is already suspended?
 581
 582The device may have different wake-up settings for runtime PM and system sleep.
 583For example, remote wake-up may be enabled for runtime suspend but disallowed
 584for system sleep (device_may_wakeup(dev) returns 'false').  When this happens,
 585the subsystem-level system suspend callback is responsible for changing the
 586device's wake-up setting (it may leave that to the device driver's system
 587suspend routine).  It may be necessary to resume the device and suspend it again
 588in order to do so.  The same is true if the driver uses different power levels
 589or other settings for runtime suspend and system sleep.
 590
 591During system resume, the simplest approach is to bring all devices back to full
 592power, even if they had been suspended before the system suspend began.  There
 593are several reasons for this, including:
 594
 595  * The device might need to switch power levels, wake-up settings, etc.
 596
 597  * Remote wake-up events might have been lost by the firmware.
 598
 599  * The device's children may need the device to be at full power in order
 600    to resume themselves.
 601
 602  * The driver's idea of the device state may not agree with the device's
 603    physical state.  This can happen during resume from hibernation.
 604
 605  * The device might need to be reset.
 606
 607  * Even though the device was suspended, if its usage counter was > 0 then most
 608    likely it would need a runtime resume in the near future anyway.
 609
 610If the device had been suspended before the system suspend began and it's
 611brought back to full power during resume, then its runtime PM status will have
 612to be updated to reflect the actual post-system sleep status.  The way to do
 613this is:
 614
 615        pm_runtime_disable(dev);
 616        pm_runtime_set_active(dev);
 617        pm_runtime_enable(dev);
 618
 619The PM core always increments the runtime usage counter before calling the
 620->suspend() callback and decrements it after calling the ->resume() callback.
 621Hence disabling runtime PM temporarily like this will not cause any runtime
 622suspend attempts to be permanently lost.  If the usage count goes to zero
 623following the return of the ->resume() callback, the ->runtime_idle() callback
 624will be invoked as usual.
 625
 626On some systems, however, system sleep is not entered through a global firmware
 627or hardware operation.  Instead, all hardware components are put into low-power
 628states directly by the kernel in a coordinated way.  Then, the system sleep
 629state effectively follows from the states the hardware components end up in
 630and the system is woken up from that state by a hardware interrupt or a similar
 631mechanism entirely under the kernel's control.  As a result, the kernel never
 632gives control away and the states of all devices during resume are precisely
 633known to it.  If that is the case and none of the situations listed above takes
 634place (in particular, if the system is not waking up from hibernation), it may
 635be more efficient to leave the devices that had been suspended before the system
 636suspend began in the suspended state.
 637
 638The PM core does its best to reduce the probability of race conditions between
 639the runtime PM and system suspend/resume (and hibernation) callbacks by carrying
 640out the following operations:
 641
 642  * During system suspend it calls pm_runtime_get_noresume() and
 643    pm_runtime_barrier() for every device right before executing the
 644    subsystem-level .suspend() callback for it.  In addition to that it calls
 645    pm_runtime_disable() for every device right after executing the
 646    subsystem-level .suspend() callback for it.
 647
 648  * During system resume it calls pm_runtime_enable() and pm_runtime_put_sync()
 649    for every device right before and right after executing the subsystem-level
 650    .resume() callback for it, respectively.
 651
 6527. Generic subsystem callbacks
 653
 654Subsystems may wish to conserve code space by using the set of generic power
 655management callbacks provided by the PM core, defined in
 656driver/base/power/generic_ops.c:
 657
 658  int pm_generic_runtime_idle(struct device *dev);
 659    - invoke the ->runtime_idle() callback provided by the driver of this
 660      device, if defined, and call pm_runtime_suspend() for this device if the
 661      return value is 0 or the callback is not defined
 662
 663  int pm_generic_runtime_suspend(struct device *dev);
 664    - invoke the ->runtime_suspend() callback provided by the driver of this
 665      device and return its result, or return -EINVAL if not defined
 666
 667  int pm_generic_runtime_resume(struct device *dev);
 668    - invoke the ->runtime_resume() callback provided by the driver of this
 669      device and return its result, or return -EINVAL if not defined
 670
 671  int pm_generic_suspend(struct device *dev);
 672    - if the device has not been suspended at run time, invoke the ->suspend()
 673      callback provided by its driver and return its result, or return 0 if not
 674      defined
 675
 676  int pm_generic_suspend_noirq(struct device *dev);
 677    - if pm_runtime_suspended(dev) returns "false", invoke the ->suspend_noirq()
 678      callback provided by the device's driver and return its result, or return
 679      0 if not defined
 680
 681  int pm_generic_resume(struct device *dev);
 682    - invoke the ->resume() callback provided by the driver of this device and,
 683      if successful, change the device's runtime PM status to 'active'
 684
 685  int pm_generic_resume_noirq(struct device *dev);
 686    - invoke the ->resume_noirq() callback provided by the driver of this device
 687
 688  int pm_generic_freeze(struct device *dev);
 689    - if the device has not been suspended at run time, invoke the ->freeze()
 690      callback provided by its driver and return its result, or return 0 if not
 691      defined
 692
 693  int pm_generic_freeze_noirq(struct device *dev);
 694    - if pm_runtime_suspended(dev) returns "false", invoke the ->freeze_noirq()
 695      callback provided by the device's driver and return its result, or return
 696      0 if not defined
 697
 698  int pm_generic_thaw(struct device *dev);
 699    - if the device has not been suspended at run time, invoke the ->thaw()
 700      callback provided by its driver and return its result, or return 0 if not
 701      defined
 702
 703  int pm_generic_thaw_noirq(struct device *dev);
 704    - if pm_runtime_suspended(dev) returns "false", invoke the ->thaw_noirq()
 705      callback provided by the device's driver and return its result, or return
 706      0 if not defined
 707
 708  int pm_generic_poweroff(struct device *dev);
 709    - if the device has not been suspended at run time, invoke the ->poweroff()
 710      callback provided by its driver and return its result, or return 0 if not
 711      defined
 712
 713  int pm_generic_poweroff_noirq(struct device *dev);
 714    - if pm_runtime_suspended(dev) returns "false", run the ->poweroff_noirq()
 715      callback provided by the device's driver and return its result, or return
 716      0 if not defined
 717
 718  int pm_generic_restore(struct device *dev);
 719    - invoke the ->restore() callback provided by the driver of this device and,
 720      if successful, change the device's runtime PM status to 'active'
 721
 722  int pm_generic_restore_noirq(struct device *dev);
 723    - invoke the ->restore_noirq() callback provided by the device's driver
 724
 725These functions can be assigned to the ->runtime_idle(), ->runtime_suspend(),
 726->runtime_resume(), ->suspend(), ->suspend_noirq(), ->resume(),
 727->resume_noirq(), ->freeze(), ->freeze_noirq(), ->thaw(), ->thaw_noirq(),
 728->poweroff(), ->poweroff_noirq(), ->restore(), ->restore_noirq() callback
 729pointers in the subsystem-level dev_pm_ops structures.
 730
 731If a subsystem wishes to use all of them at the same time, it can simply assign
 732the GENERIC_SUBSYS_PM_OPS macro, defined in include/linux/pm.h, to its
 733dev_pm_ops structure pointer.
 734
 735Device drivers that wish to use the same function as a system suspend, freeze,
 736poweroff and runtime suspend callback, and similarly for system resume, thaw,
 737restore, and runtime resume, can achieve this with the help of the
 738UNIVERSAL_DEV_PM_OPS macro defined in include/linux/pm.h (possibly setting its
 739last argument to NULL).
 740
 7418. "No-Callback" Devices
 742
 743Some "devices" are only logical sub-devices of their parent and cannot be
 744power-managed on their own.  (The prototype example is a USB interface.  Entire
 745USB devices can go into low-power mode or send wake-up requests, but neither is
 746possible for individual interfaces.)  The drivers for these devices have no
 747need of runtime PM callbacks; if the callbacks did exist, ->runtime_suspend()
 748and ->runtime_resume() would always return 0 without doing anything else and
 749->runtime_idle() would always call pm_runtime_suspend().
 750
 751Subsystems can tell the PM core about these devices by calling
 752pm_runtime_no_callbacks().  This should be done after the device structure is
 753initialized and before it is registered (although after device registration is
 754also okay).  The routine will set the device's power.no_callbacks flag and
 755prevent the non-debugging runtime PM sysfs attributes from being created.
 756
 757When power.no_callbacks is set, the PM core will not invoke the
 758->runtime_idle(), ->runtime_suspend(), or ->runtime_resume() callbacks.
 759Instead it will assume that suspends and resumes always succeed and that idle
 760devices should be suspended.
 761
 762As a consequence, the PM core will never directly inform the device's subsystem
 763or driver about runtime power changes.  Instead, the driver for the device's
 764parent must take responsibility for telling the device's driver when the
 765parent's power state changes.
 766
 7679. Autosuspend, or automatically-delayed suspends
 768
 769Changing a device's power state isn't free; it requires both time and energy.
 770A device should be put in a low-power state only when there's some reason to
 771think it will remain in that state for a substantial time.  A common heuristic
 772says that a device which hasn't been used for a while is liable to remain
 773unused; following this advice, drivers should not allow devices to be suspended
 774at runtime until they have been inactive for some minimum period.  Even when
 775the heuristic ends up being non-optimal, it will still prevent devices from
 776"bouncing" too rapidly between low-power and full-power states.
 777
 778The term "autosuspend" is an historical remnant.  It doesn't mean that the
 779device is automatically suspended (the subsystem or driver still has to call
 780the appropriate PM routines); rather it means that runtime suspends will
 781automatically be delayed until the desired period of inactivity has elapsed.
 782
 783Inactivity is determined based on the power.last_busy field.  Drivers should
 784call pm_runtime_mark_last_busy() to update this field after carrying out I/O,
 785typically just before calling pm_runtime_put_autosuspend().  The desired length
 786of the inactivity period is a matter of policy.  Subsystems can set this length
 787initially by calling pm_runtime_set_autosuspend_delay(), but after device
 788registration the length should be controlled by user space, using the
 789/sys/devices/.../power/autosuspend_delay_ms attribute.
 790
 791In order to use autosuspend, subsystems or drivers must call
 792pm_runtime_use_autosuspend() (preferably before registering the device), and
 793thereafter they should use the various *_autosuspend() helper functions instead
 794of the non-autosuspend counterparts:
 795
 796        Instead of: pm_runtime_suspend    use: pm_runtime_autosuspend;
 797        Instead of: pm_schedule_suspend   use: pm_request_autosuspend;
 798        Instead of: pm_runtime_put        use: pm_runtime_put_autosuspend;
 799        Instead of: pm_runtime_put_sync   use: pm_runtime_put_sync_autosuspend.
 800
 801Drivers may also continue to use the non-autosuspend helper functions; they
 802will behave normally, not taking the autosuspend delay into account.
 803Similarly, if the power.use_autosuspend field isn't set then the autosuspend
 804helper functions will behave just like the non-autosuspend counterparts.
 805
 806Under some circumstances a driver or subsystem may want to prevent a device
 807from autosuspending immediately, even though the usage counter is zero and the
 808autosuspend delay time has expired.  If the ->runtime_suspend() callback
 809returns -EAGAIN or -EBUSY, and if the next autosuspend delay expiration time is
 810in the future (as it normally would be if the callback invoked
 811pm_runtime_mark_last_busy()), the PM core will automatically reschedule the
 812autosuspend.  The ->runtime_suspend() callback can't do this rescheduling
 813itself because no suspend requests of any kind are accepted while the device is
 814suspending (i.e., while the callback is running).
 815
 816The implementation is well suited for asynchronous use in interrupt contexts.
 817However such use inevitably involves races, because the PM core can't
 818synchronize ->runtime_suspend() callbacks with the arrival of I/O requests.
 819This synchronization must be handled by the driver, using its private lock.
 820Here is a schematic pseudo-code example:
 821
 822        foo_read_or_write(struct foo_priv *foo, void *data)
 823        {
 824                lock(&foo->private_lock);
 825                add_request_to_io_queue(foo, data);
 826                if (foo->num_pending_requests++ == 0)
 827                        pm_runtime_get(&foo->dev);
 828                if (!foo->is_suspended)
 829                        foo_process_next_request(foo);
 830                unlock(&foo->private_lock);
 831        }
 832
 833        foo_io_completion(struct foo_priv *foo, void *req)
 834        {
 835                lock(&foo->private_lock);
 836                if (--foo->num_pending_requests == 0) {
 837                        pm_runtime_mark_last_busy(&foo->dev);
 838                        pm_runtime_put_autosuspend(&foo->dev);
 839                } else {
 840                        foo_process_next_request(foo);
 841                }
 842                unlock(&foo->private_lock);
 843                /* Send req result back to the user ... */
 844        }
 845
 846        int foo_runtime_suspend(struct device *dev)
 847        {
 848                struct foo_priv foo = container_of(dev, ...);
 849                int ret = 0;
 850
 851                lock(&foo->private_lock);
 852                if (foo->num_pending_requests > 0) {
 853                        ret = -EBUSY;
 854                } else {
 855                        /* ... suspend the device ... */
 856                        foo->is_suspended = 1;
 857                }
 858                unlock(&foo->private_lock);
 859                return ret;
 860        }
 861
 862        int foo_runtime_resume(struct device *dev)
 863        {
 864                struct foo_priv foo = container_of(dev, ...);
 865
 866                lock(&foo->private_lock);
 867                /* ... resume the device ... */
 868                foo->is_suspended = 0;
 869                pm_runtime_mark_last_busy(&foo->dev);
 870                if (foo->num_pending_requests > 0)
 871                        foo_process_requests(foo);
 872                unlock(&foo->private_lock);
 873                return 0;
 874        }
 875
 876The important point is that after foo_io_completion() asks for an autosuspend,
 877the foo_runtime_suspend() callback may race with foo_read_or_write().
 878Therefore foo_runtime_suspend() has to check whether there are any pending I/O
 879requests (while holding the private lock) before allowing the suspend to
 880proceed.
 881
 882In addition, the power.autosuspend_delay field can be changed by user space at
 883any time.  If a driver cares about this, it can call
 884pm_runtime_autosuspend_expiration() from within the ->runtime_suspend()
 885callback while holding its private lock.  If the function returns a nonzero
 886value then the delay has not yet expired and the callback should return
 887-EAGAIN.
 888
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.