Documentation/power/runtime_pm.txt

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