runtime_pm.txt 43 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898
  1. Runtime Power Management Framework for I/O Devices
  2. (C) 2009-2011 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
  3. (C) 2010 Alan Stern <stern@rowland.harvard.edu>
  4. 1. Introduction
  5. Support for runtime power management (runtime PM) of I/O devices is provided
  6. at the power management core (PM core) level by means of:
  7. * The power management workqueue pm_wq in which bus types and device drivers can
  8. put their PM-related work items. It is strongly recommended that pm_wq be
  9. used for queuing all work items related to runtime PM, because this allows
  10. them to be synchronized with system-wide power transitions (suspend to RAM,
  11. hibernation and resume from system sleep states). pm_wq is declared in
  12. include/linux/pm_runtime.h and defined in kernel/power/main.c.
  13. * A number of runtime PM fields in the 'power' member of 'struct device' (which
  14. is of the type 'struct dev_pm_info', defined in include/linux/pm.h) that can
  15. be used for synchronizing runtime PM operations with one another.
  16. * Three device runtime PM callbacks in 'struct dev_pm_ops' (defined in
  17. include/linux/pm.h).
  18. * A set of helper functions defined in drivers/base/power/runtime.c that can be
  19. used for carrying out runtime PM operations in such a way that the
  20. synchronization between them is taken care of by the PM core. Bus types and
  21. device drivers are encouraged to use these functions.
  22. The runtime PM callbacks present in 'struct dev_pm_ops', the device runtime PM
  23. fields of 'struct dev_pm_info' and the core helper functions provided for
  24. runtime PM are described below.
  25. 2. Device Runtime PM Callbacks
  26. There are three device runtime PM callbacks defined in 'struct dev_pm_ops':
  27. struct dev_pm_ops {
  28. ...
  29. int (*runtime_suspend)(struct device *dev);
  30. int (*runtime_resume)(struct device *dev);
  31. int (*runtime_idle)(struct device *dev);
  32. ...
  33. };
  34. The ->runtime_suspend(), ->runtime_resume() and ->runtime_idle() callbacks
  35. are executed by the PM core for the device's subsystem that may be either of
  36. the following:
  37. 1. PM domain of the device, if the device's PM domain object, dev->pm_domain,
  38. is present.
  39. 2. Device type of the device, if both dev->type and dev->type->pm are present.
  40. 3. Device class of the device, if both dev->class and dev->class->pm are
  41. present.
  42. 4. Bus type of the device, if both dev->bus and dev->bus->pm are present.
  43. If the subsystem chosen by applying the above rules doesn't provide the relevant
  44. callback, the PM core will invoke the corresponding driver callback stored in
  45. dev->driver->pm directly (if present).
  46. The PM core always checks which callback to use in the order given above, so the
  47. priority order of callbacks from high to low is: PM domain, device type, class
  48. and bus type. Moreover, the high-priority one will always take precedence over
  49. a low-priority one. The PM domain, bus type, device type and class callbacks
  50. are referred to as subsystem-level callbacks in what follows.
  51. By default, the callbacks are always invoked in process context with interrupts
  52. enabled. However, the pm_runtime_irq_safe() helper function can be used to tell
  53. the PM core that it is safe to run the ->runtime_suspend(), ->runtime_resume()
  54. and ->runtime_idle() callbacks for the given device in atomic context with
  55. interrupts disabled. This implies that the callback routines in question must
  56. not block or sleep, but it also means that the synchronous helper functions
  57. listed at the end of Section 4 may be used for that device within an interrupt
  58. handler or generally in an atomic context.
  59. The subsystem-level suspend callback, if present, is _entirely_ _responsible_
  60. for handling the suspend of the device as appropriate, which may, but need not
  61. include executing the device driver's own ->runtime_suspend() callback (from the
  62. PM core's point of view it is not necessary to implement a ->runtime_suspend()
  63. callback in a device driver as long as the subsystem-level suspend callback
  64. knows what to do to handle the device).
  65. * Once the subsystem-level suspend callback (or the driver suspend callback,
  66. if invoked directly) has completed successfully for the given device, the PM
  67. core regards the device as suspended, which need not mean that it has been
  68. put into a low power state. It is supposed to mean, however, that the
  69. device will not process data and will not communicate with the CPU(s) and
  70. RAM until the appropriate resume callback is executed for it. The runtime
  71. PM status of a device after successful execution of the suspend callback is
  72. 'suspended'.
  73. * If the suspend callback returns -EBUSY or -EAGAIN, the device's runtime PM
  74. status remains 'active', which means that the device _must_ be fully
  75. operational afterwards.
  76. * If the suspend callback returns an error code different from -EBUSY and
  77. -EAGAIN, the PM core regards this as a fatal error and will refuse to run
  78. the helper functions described in Section 4 for the device until its status
  79. is directly set to either'active', or 'suspended' (the PM core provides
  80. special helper functions for this purpose).
  81. In particular, if the driver requires remote wakeup capability (i.e. hardware
  82. mechanism allowing the device to request a change of its power state, such as
  83. PCI PME) for proper functioning and device_run_wake() returns 'false' for the
  84. device, then ->runtime_suspend() should return -EBUSY. On the other hand, if
  85. device_run_wake() returns 'true' for the device and the device is put into a
  86. low-power state during the execution of the suspend callback, it is expected
  87. that remote wakeup will be enabled for the device. Generally, remote wakeup
  88. should be enabled for all input devices put into low-power states at run time.
  89. The subsystem-level resume callback, if present, is _entirely_ _responsible_ for
  90. handling the resume of the device as appropriate, which may, but need not
  91. include executing the device driver's own ->runtime_resume() callback (from the
  92. PM core's point of view it is not necessary to implement a ->runtime_resume()
  93. callback in a device driver as long as the subsystem-level resume callback knows
  94. what to do to handle the device).
  95. * Once the subsystem-level resume callback (or the driver resume callback, if
  96. invoked directly) has completed successfully, the PM core regards the device
  97. as fully operational, which means that the device _must_ be able to complete
  98. I/O operations as needed. The runtime PM status of the device is then
  99. 'active'.
  100. * If the resume callback returns an error code, the PM core regards this as a
  101. fatal error and will refuse to run the helper functions described in Section
  102. 4 for the device, until its status is directly set to either 'active', or
  103. 'suspended' (by means of special helper functions provided by the PM core
  104. for this purpose).
  105. The idle callback (a subsystem-level one, if present, or the driver one) is
  106. executed by the PM core whenever the device appears to be idle, which is
  107. indicated to the PM core by two counters, the device's usage counter and the
  108. counter of 'active' children of the device.
  109. * If any of these counters is decreased using a helper function provided by
  110. the PM core and it turns out to be equal to zero, the other counter is
  111. checked. If that counter also is equal to zero, the PM core executes the
  112. idle callback with the device as its argument.
  113. The action performed by the idle callback is totally dependent on the subsystem
  114. (or driver) in question, but the expected and recommended action is to check
  115. if the device can be suspended (i.e. if all of the conditions necessary for
  116. suspending the device are satisfied) and to queue up a suspend request for the
  117. device in that case. If there is no idle callback, or if the callback returns
  118. 0, then the PM core will attempt to carry out a runtime suspend of the device;
  119. in essence, it will call pm_runtime_suspend() directly. To prevent this (for
  120. example, if the callback routine has started a delayed suspend), the routine
  121. should return a non-zero value. Negative error return codes are ignored by the
  122. PM core.
  123. The helper functions provided by the PM core, described in Section 4, guarantee
  124. that the following constraints are met with respect to runtime PM callbacks for
  125. one device:
  126. (1) The callbacks are mutually exclusive (e.g. it is forbidden to execute
  127. ->runtime_suspend() in parallel with ->runtime_resume() or with another
  128. instance of ->runtime_suspend() for the same device) with the exception that
  129. ->runtime_suspend() or ->runtime_resume() can be executed in parallel with
  130. ->runtime_idle() (although ->runtime_idle() will not be started while any
  131. of the other callbacks is being executed for the same device).
  132. (2) ->runtime_idle() and ->runtime_suspend() can only be executed for 'active'
  133. devices (i.e. the PM core will only execute ->runtime_idle() or
  134. ->runtime_suspend() for the devices the runtime PM status of which is
  135. 'active').
  136. (3) ->runtime_idle() and ->runtime_suspend() can only be executed for a device
  137. the usage counter of which is equal to zero _and_ either the counter of
  138. 'active' children of which is equal to zero, or the 'power.ignore_children'
  139. flag of which is set.
  140. (4) ->runtime_resume() can only be executed for 'suspended' devices (i.e. the
  141. PM core will only execute ->runtime_resume() for the devices the runtime
  142. PM status of which is 'suspended').
  143. Additionally, the helper functions provided by the PM core obey the following
  144. rules:
  145. * If ->runtime_suspend() is about to be executed or there's a pending request
  146. to execute it, ->runtime_idle() will not be executed for the same device.
  147. * A request to execute or to schedule the execution of ->runtime_suspend()
  148. will cancel any pending requests to execute ->runtime_idle() for the same
  149. device.
  150. * If ->runtime_resume() is about to be executed or there's a pending request
  151. to execute it, the other callbacks will not be executed for the same device.
  152. * A request to execute ->runtime_resume() will cancel any pending or
  153. scheduled requests to execute the other callbacks for the same device,
  154. except for scheduled autosuspends.
  155. 3. Runtime PM Device Fields
  156. The following device runtime PM fields are present in 'struct dev_pm_info', as
  157. defined in include/linux/pm.h:
  158. struct timer_list suspend_timer;
  159. - timer used for scheduling (delayed) suspend and autosuspend requests
  160. unsigned long timer_expires;
  161. - timer expiration time, in jiffies (if this is different from zero, the
  162. timer is running and will expire at that time, otherwise the timer is not
  163. running)
  164. struct work_struct work;
  165. - work structure used for queuing up requests (i.e. work items in pm_wq)
  166. wait_queue_head_t wait_queue;
  167. - wait queue used if any of the helper functions needs to wait for another
  168. one to complete
  169. spinlock_t lock;
  170. - lock used for synchronisation
  171. atomic_t usage_count;
  172. - the usage counter of the device
  173. atomic_t child_count;
  174. - the count of 'active' children of the device
  175. unsigned int ignore_children;
  176. - if set, the value of child_count is ignored (but still updated)
  177. unsigned int disable_depth;
  178. - used for disabling the helper funcions (they work normally if this is
  179. equal to zero); the initial value of it is 1 (i.e. runtime PM is
  180. initially disabled for all devices)
  181. unsigned int runtime_error;
  182. - if set, there was a fatal error (one of the callbacks returned error code
  183. as described in Section 2), so the helper funtions will not work until
  184. this flag is cleared; this is the error code returned by the failing
  185. callback
  186. unsigned int idle_notification;
  187. - if set, ->runtime_idle() is being executed
  188. unsigned int request_pending;
  189. - if set, there's a pending request (i.e. a work item queued up into pm_wq)
  190. enum rpm_request request;
  191. - type of request that's pending (valid if request_pending is set)
  192. unsigned int deferred_resume;
  193. - set if ->runtime_resume() is about to be run while ->runtime_suspend() is
  194. being executed for that device and it is not practical to wait for the
  195. suspend to complete; means "start a resume as soon as you've suspended"
  196. unsigned int run_wake;
  197. - set if the device is capable of generating runtime wake-up events
  198. enum rpm_status runtime_status;
  199. - the runtime PM status of the device; this field's initial value is
  200. RPM_SUSPENDED, which means that each device is initially regarded by the
  201. PM core as 'suspended', regardless of its real hardware status
  202. unsigned int runtime_auto;
  203. - if set, indicates that the user space has allowed the device driver to
  204. power manage the device at run time via the /sys/devices/.../power/control
  205. interface; it may only be modified with the help of the pm_runtime_allow()
  206. and pm_runtime_forbid() helper functions
  207. unsigned int no_callbacks;
  208. - indicates that the device does not use the runtime PM callbacks (see
  209. Section 8); it may be modified only by the pm_runtime_no_callbacks()
  210. helper function
  211. unsigned int irq_safe;
  212. - indicates that the ->runtime_suspend() and ->runtime_resume() callbacks
  213. will be invoked with the spinlock held and interrupts disabled
  214. unsigned int use_autosuspend;
  215. - indicates that the device's driver supports delayed autosuspend (see
  216. Section 9); it may be modified only by the
  217. pm_runtime{_dont}_use_autosuspend() helper functions
  218. unsigned int timer_autosuspends;
  219. - indicates that the PM core should attempt to carry out an autosuspend
  220. when the timer expires rather than a normal suspend
  221. int autosuspend_delay;
  222. - the delay time (in milliseconds) to be used for autosuspend
  223. unsigned long last_busy;
  224. - the time (in jiffies) when the pm_runtime_mark_last_busy() helper
  225. function was last called for this device; used in calculating inactivity
  226. periods for autosuspend
  227. All of the above fields are members of the 'power' member of 'struct device'.
  228. 4. Runtime PM Device Helper Functions
  229. The following runtime PM helper functions are defined in
  230. drivers/base/power/runtime.c and include/linux/pm_runtime.h:
  231. void pm_runtime_init(struct device *dev);
  232. - initialize the device runtime PM fields in 'struct dev_pm_info'
  233. void pm_runtime_remove(struct device *dev);
  234. - make sure that the runtime PM of the device will be disabled after
  235. removing the device from device hierarchy
  236. int pm_runtime_idle(struct device *dev);
  237. - execute the subsystem-level idle callback for the device; returns an
  238. error code on failure, where -EINPROGRESS means that ->runtime_idle() is
  239. already being executed; if there is no callback or the callback returns 0
  240. then run pm_runtime_suspend(dev) and return its result
  241. int pm_runtime_suspend(struct device *dev);
  242. - execute the subsystem-level suspend callback for the device; returns 0 on
  243. success, 1 if the device's runtime PM status was already 'suspended', or
  244. error code on failure, where -EAGAIN or -EBUSY means it is safe to attempt
  245. to suspend the device again in future and -EACCES means that
  246. 'power.disable_depth' is different from 0
  247. int pm_runtime_autosuspend(struct device *dev);
  248. - same as pm_runtime_suspend() except that the autosuspend delay is taken
  249. into account; if pm_runtime_autosuspend_expiration() says the delay has
  250. not yet expired then an autosuspend is scheduled for the appropriate time
  251. and 0 is returned
  252. int pm_runtime_resume(struct device *dev);
  253. - execute the subsystem-level resume callback for the device; returns 0 on
  254. success, 1 if the device's runtime PM status was already 'active' or
  255. error code on failure, where -EAGAIN means it may be safe to attempt to
  256. resume the device again in future, but 'power.runtime_error' should be
  257. checked additionally, and -EACCES means that 'power.disable_depth' is
  258. different from 0
  259. int pm_request_idle(struct device *dev);
  260. - submit a request to execute the subsystem-level idle callback for the
  261. device (the request is represented by a work item in pm_wq); returns 0 on
  262. success or error code if the request has not been queued up
  263. int pm_request_autosuspend(struct device *dev);
  264. - schedule the execution of the subsystem-level suspend callback for the
  265. device when the autosuspend delay has expired; if the delay has already
  266. expired then the work item is queued up immediately
  267. int pm_schedule_suspend(struct device *dev, unsigned int delay);
  268. - schedule the execution of the subsystem-level suspend callback for the
  269. device in future, where 'delay' is the time to wait before queuing up a
  270. suspend work item in pm_wq, in milliseconds (if 'delay' is zero, the work
  271. item is queued up immediately); returns 0 on success, 1 if the device's PM
  272. runtime status was already 'suspended', or error code if the request
  273. hasn't been scheduled (or queued up if 'delay' is 0); if the execution of
  274. ->runtime_suspend() is already scheduled and not yet expired, the new
  275. value of 'delay' will be used as the time to wait
  276. int pm_request_resume(struct device *dev);
  277. - submit a request to execute the subsystem-level resume callback for the
  278. device (the request is represented by a work item in pm_wq); returns 0 on
  279. success, 1 if the device's runtime PM status was already 'active', or
  280. error code if the request hasn't been queued up
  281. void pm_runtime_get_noresume(struct device *dev);
  282. - increment the device's usage counter
  283. int pm_runtime_get(struct device *dev);
  284. - increment the device's usage counter, run pm_request_resume(dev) and
  285. return its result
  286. int pm_runtime_get_sync(struct device *dev);
  287. - increment the device's usage counter, run pm_runtime_resume(dev) and
  288. return its result
  289. void pm_runtime_put_noidle(struct device *dev);
  290. - decrement the device's usage counter
  291. int pm_runtime_put(struct device *dev);
  292. - decrement the device's usage counter; if the result is 0 then run
  293. pm_request_idle(dev) and return its result
  294. int pm_runtime_put_autosuspend(struct device *dev);
  295. - decrement the device's usage counter; if the result is 0 then run
  296. pm_request_autosuspend(dev) and return its result
  297. int pm_runtime_put_sync(struct device *dev);
  298. - decrement the device's usage counter; if the result is 0 then run
  299. pm_runtime_idle(dev) and return its result
  300. int pm_runtime_put_sync_suspend(struct device *dev);
  301. - decrement the device's usage counter; if the result is 0 then run
  302. pm_runtime_suspend(dev) and return its result
  303. int pm_runtime_put_sync_autosuspend(struct device *dev);
  304. - decrement the device's usage counter; if the result is 0 then run
  305. pm_runtime_autosuspend(dev) and return its result
  306. void pm_runtime_enable(struct device *dev);
  307. - decrement the device's 'power.disable_depth' field; if that field is equal
  308. to zero, the runtime PM helper functions can execute subsystem-level
  309. callbacks described in Section 2 for the device
  310. int pm_runtime_disable(struct device *dev);
  311. - increment the device's 'power.disable_depth' field (if the value of that
  312. field was previously zero, this prevents subsystem-level runtime PM
  313. callbacks from being run for the device), make sure that all of the pending
  314. runtime PM operations on the device are either completed or canceled;
  315. returns 1 if there was a resume request pending and it was necessary to
  316. execute the subsystem-level resume callback for the device to satisfy that
  317. request, otherwise 0 is returned
  318. int pm_runtime_barrier(struct device *dev);
  319. - check if there's a resume request pending for the device and resume it
  320. (synchronously) in that case, cancel any other pending runtime PM requests
  321. regarding it and wait for all runtime PM operations on it in progress to
  322. complete; returns 1 if there was a resume request pending and it was
  323. necessary to execute the subsystem-level resume callback for the device to
  324. satisfy that request, otherwise 0 is returned
  325. void pm_suspend_ignore_children(struct device *dev, bool enable);
  326. - set/unset the power.ignore_children flag of the device
  327. int pm_runtime_set_active(struct device *dev);
  328. - clear the device's 'power.runtime_error' flag, set the device's runtime
  329. PM status to 'active' and update its parent's counter of 'active'
  330. children as appropriate (it is only valid to use this function if
  331. 'power.runtime_error' is set or 'power.disable_depth' is greater than
  332. zero); it will fail and return error code if the device has a parent
  333. which is not active and the 'power.ignore_children' flag of which is unset
  334. void pm_runtime_set_suspended(struct device *dev);
  335. - clear the device's 'power.runtime_error' flag, set the device's runtime
  336. PM status to 'suspended' and update its parent's counter of 'active'
  337. children as appropriate (it is only valid to use this function if
  338. 'power.runtime_error' is set or 'power.disable_depth' is greater than
  339. zero)
  340. bool pm_runtime_active(struct device *dev);
  341. - return true if the device's runtime PM status is 'active' or its
  342. 'power.disable_depth' field is not equal to zero, or false otherwise
  343. bool pm_runtime_suspended(struct device *dev);
  344. - return true if the device's runtime PM status is 'suspended' and its
  345. 'power.disable_depth' field is equal to zero, or false otherwise
  346. bool pm_runtime_status_suspended(struct device *dev);
  347. - return true if the device's runtime PM status is 'suspended'
  348. void pm_runtime_allow(struct device *dev);
  349. - set the power.runtime_auto flag for the device and decrease its usage
  350. counter (used by the /sys/devices/.../power/control interface to
  351. effectively allow the device to be power managed at run time)
  352. void pm_runtime_forbid(struct device *dev);
  353. - unset the power.runtime_auto flag for the device and increase its usage
  354. counter (used by the /sys/devices/.../power/control interface to
  355. effectively prevent the device from being power managed at run time)
  356. void pm_runtime_no_callbacks(struct device *dev);
  357. - set the power.no_callbacks flag for the device and remove the runtime
  358. PM attributes from /sys/devices/.../power (or prevent them from being
  359. added when the device is registered)
  360. void pm_runtime_irq_safe(struct device *dev);
  361. - set the power.irq_safe flag for the device, causing the runtime-PM
  362. callbacks to be invoked with interrupts off
  363. void pm_runtime_mark_last_busy(struct device *dev);
  364. - set the power.last_busy field to the current time
  365. void pm_runtime_use_autosuspend(struct device *dev);
  366. - set the power.use_autosuspend flag, enabling autosuspend delays
  367. void pm_runtime_dont_use_autosuspend(struct device *dev);
  368. - clear the power.use_autosuspend flag, disabling autosuspend delays
  369. void pm_runtime_set_autosuspend_delay(struct device *dev, int delay);
  370. - set the power.autosuspend_delay value to 'delay' (expressed in
  371. milliseconds); if 'delay' is negative then runtime suspends are
  372. prevented
  373. unsigned long pm_runtime_autosuspend_expiration(struct device *dev);
  374. - calculate the time when the current autosuspend delay period will expire,
  375. based on power.last_busy and power.autosuspend_delay; if the delay time
  376. is 1000 ms or larger then the expiration time is rounded up to the
  377. nearest second; returns 0 if the delay period has already expired or
  378. power.use_autosuspend isn't set, otherwise returns the expiration time
  379. in jiffies
  380. It is safe to execute the following helper functions from interrupt context:
  381. pm_request_idle()
  382. pm_request_autosuspend()
  383. pm_schedule_suspend()
  384. pm_request_resume()
  385. pm_runtime_get_noresume()
  386. pm_runtime_get()
  387. pm_runtime_put_noidle()
  388. pm_runtime_put()
  389. pm_runtime_put_autosuspend()
  390. pm_runtime_enable()
  391. pm_suspend_ignore_children()
  392. pm_runtime_set_active()
  393. pm_runtime_set_suspended()
  394. pm_runtime_suspended()
  395. pm_runtime_mark_last_busy()
  396. pm_runtime_autosuspend_expiration()
  397. If pm_runtime_irq_safe() has been called for a device then the following helper
  398. functions may also be used in interrupt context:
  399. pm_runtime_idle()
  400. pm_runtime_suspend()
  401. pm_runtime_autosuspend()
  402. pm_runtime_resume()
  403. pm_runtime_get_sync()
  404. pm_runtime_put_sync()
  405. pm_runtime_put_sync_suspend()
  406. pm_runtime_put_sync_autosuspend()
  407. 5. Runtime PM Initialization, Device Probing and Removal
  408. Initially, the runtime PM is disabled for all devices, which means that the
  409. majority of the runtime PM helper funtions described in Section 4 will return
  410. -EAGAIN until pm_runtime_enable() is called for the device.
  411. In addition to that, the initial runtime PM status of all devices is
  412. 'suspended', but it need not reflect the actual physical state of the device.
  413. Thus, if the device is initially active (i.e. it is able to process I/O), its
  414. runtime PM status must be changed to 'active', with the help of
  415. pm_runtime_set_active(), before pm_runtime_enable() is called for the device.
  416. However, if the device has a parent and the parent's runtime PM is enabled,
  417. calling pm_runtime_set_active() for the device will affect the parent, unless
  418. the parent's 'power.ignore_children' flag is set. Namely, in that case the
  419. parent won't be able to suspend at run time, using the PM core's helper
  420. functions, as long as the child's status is 'active', even if the child's
  421. runtime PM is still disabled (i.e. pm_runtime_enable() hasn't been called for
  422. the child yet or pm_runtime_disable() has been called for it). For this reason,
  423. once pm_runtime_set_active() has been called for the device, pm_runtime_enable()
  424. should be called for it too as soon as reasonably possible or its runtime PM
  425. status should be changed back to 'suspended' with the help of
  426. pm_runtime_set_suspended().
  427. If the default initial runtime PM status of the device (i.e. 'suspended')
  428. reflects the actual state of the device, its bus type's or its driver's
  429. ->probe() callback will likely need to wake it up using one of the PM core's
  430. helper functions described in Section 4. In that case, pm_runtime_resume()
  431. should be used. Of course, for this purpose the device's runtime PM has to be
  432. enabled earlier by calling pm_runtime_enable().
  433. If the device bus type's or driver's ->probe() callback runs
  434. pm_runtime_suspend() or pm_runtime_idle() or their asynchronous counterparts,
  435. they will fail returning -EAGAIN, because the device's usage counter is
  436. incremented by the driver core before executing ->probe(). Still, it may be
  437. desirable to suspend the device as soon as ->probe() has finished, so the driver
  438. core uses pm_runtime_put_sync() to invoke the subsystem-level idle callback for
  439. the device at that time.
  440. Moreover, the driver core prevents runtime PM callbacks from racing with the bus
  441. notifier callback in __device_release_driver(), which is necessary, because the
  442. notifier is used by some subsystems to carry out operations affecting the
  443. runtime PM functionality. It does so by calling pm_runtime_get_sync() before
  444. driver_sysfs_remove() and the BUS_NOTIFY_UNBIND_DRIVER notifications. This
  445. resumes the device if it's in the suspended state and prevents it from
  446. being suspended again while those routines are being executed.
  447. To allow bus types and drivers to put devices into the suspended state by
  448. calling pm_runtime_suspend() from their ->remove() routines, the driver core
  449. executes pm_runtime_put_sync() after running the BUS_NOTIFY_UNBIND_DRIVER
  450. notifications in __device_release_driver(). This requires bus types and
  451. drivers to make their ->remove() callbacks avoid races with runtime PM directly,
  452. but also it allows of more flexibility in the handling of devices during the
  453. removal of their drivers.
  454. The user space can effectively disallow the driver of the device to power manage
  455. it at run time by changing the value of its /sys/devices/.../power/control
  456. attribute to "on", which causes pm_runtime_forbid() to be called. In principle,
  457. this mechanism may also be used by the driver to effectively turn off the
  458. runtime power management of the device until the user space turns it on.
  459. Namely, during the initialization the driver can make sure that the runtime PM
  460. status of the device is 'active' and call pm_runtime_forbid(). It should be
  461. noted, however, that if the user space has already intentionally changed the
  462. value of /sys/devices/.../power/control to "auto" to allow the driver to power
  463. manage the device at run time, the driver may confuse it by using
  464. pm_runtime_forbid() this way.
  465. 6. Runtime PM and System Sleep
  466. Runtime PM and system sleep (i.e., system suspend and hibernation, also known
  467. as suspend-to-RAM and suspend-to-disk) interact with each other in a couple of
  468. ways. If a device is active when a system sleep starts, everything is
  469. straightforward. But what should happen if the device is already suspended?
  470. The device may have different wake-up settings for runtime PM and system sleep.
  471. For example, remote wake-up may be enabled for runtime suspend but disallowed
  472. for system sleep (device_may_wakeup(dev) returns 'false'). When this happens,
  473. the subsystem-level system suspend callback is responsible for changing the
  474. device's wake-up setting (it may leave that to the device driver's system
  475. suspend routine). It may be necessary to resume the device and suspend it again
  476. in order to do so. The same is true if the driver uses different power levels
  477. or other settings for runtime suspend and system sleep.
  478. During system resume, the simplest approach is to bring all devices back to full
  479. power, even if they had been suspended before the system suspend began. There
  480. are several reasons for this, including:
  481. * The device might need to switch power levels, wake-up settings, etc.
  482. * Remote wake-up events might have been lost by the firmware.
  483. * The device's children may need the device to be at full power in order
  484. to resume themselves.
  485. * The driver's idea of the device state may not agree with the device's
  486. physical state. This can happen during resume from hibernation.
  487. * The device might need to be reset.
  488. * Even though the device was suspended, if its usage counter was > 0 then most
  489. likely it would need a runtime resume in the near future anyway.
  490. If the device had been suspended before the system suspend began and it's
  491. brought back to full power during resume, then its runtime PM status will have
  492. to be updated to reflect the actual post-system sleep status. The way to do
  493. this is:
  494. pm_runtime_disable(dev);
  495. pm_runtime_set_active(dev);
  496. pm_runtime_enable(dev);
  497. The PM core always increments the runtime usage counter before calling the
  498. ->suspend() callback and decrements it after calling the ->resume() callback.
  499. Hence disabling runtime PM temporarily like this will not cause any runtime
  500. suspend attempts to be permanently lost. If the usage count goes to zero
  501. following the return of the ->resume() callback, the ->runtime_idle() callback
  502. will be invoked as usual.
  503. On some systems, however, system sleep is not entered through a global firmware
  504. or hardware operation. Instead, all hardware components are put into low-power
  505. states directly by the kernel in a coordinated way. Then, the system sleep
  506. state effectively follows from the states the hardware components end up in
  507. and the system is woken up from that state by a hardware interrupt or a similar
  508. mechanism entirely under the kernel's control. As a result, the kernel never
  509. gives control away and the states of all devices during resume are precisely
  510. known to it. If that is the case and none of the situations listed above takes
  511. place (in particular, if the system is not waking up from hibernation), it may
  512. be more efficient to leave the devices that had been suspended before the system
  513. suspend began in the suspended state.
  514. The PM core does its best to reduce the probability of race conditions between
  515. the runtime PM and system suspend/resume (and hibernation) callbacks by carrying
  516. out the following operations:
  517. * During system suspend it calls pm_runtime_get_noresume() and
  518. pm_runtime_barrier() for every device right before executing the
  519. subsystem-level .suspend() callback for it. In addition to that it calls
  520. __pm_runtime_disable() with 'false' as the second argument for every device
  521. right before executing the subsystem-level .suspend_late() callback for it.
  522. * During system resume it calls pm_runtime_enable() and pm_runtime_put_sync()
  523. for every device right after executing the subsystem-level .resume_early()
  524. callback and right after executing the subsystem-level .resume() callback
  525. for it, respectively.
  526. 7. Generic subsystem callbacks
  527. Subsystems may wish to conserve code space by using the set of generic power
  528. management callbacks provided by the PM core, defined in
  529. driver/base/power/generic_ops.c:
  530. int pm_generic_runtime_idle(struct device *dev);
  531. - invoke the ->runtime_idle() callback provided by the driver of this
  532. device, if defined, and call pm_runtime_suspend() for this device if the
  533. return value is 0 or the callback is not defined
  534. int pm_generic_runtime_suspend(struct device *dev);
  535. - invoke the ->runtime_suspend() callback provided by the driver of this
  536. device and return its result, or return -EINVAL if not defined
  537. int pm_generic_runtime_resume(struct device *dev);
  538. - invoke the ->runtime_resume() callback provided by the driver of this
  539. device and return its result, or return -EINVAL if not defined
  540. int pm_generic_suspend(struct device *dev);
  541. - if the device has not been suspended at run time, invoke the ->suspend()
  542. callback provided by its driver and return its result, or return 0 if not
  543. defined
  544. int pm_generic_suspend_noirq(struct device *dev);
  545. - if pm_runtime_suspended(dev) returns "false", invoke the ->suspend_noirq()
  546. callback provided by the device's driver and return its result, or return
  547. 0 if not defined
  548. int pm_generic_resume(struct device *dev);
  549. - invoke the ->resume() callback provided by the driver of this device and,
  550. if successful, change the device's runtime PM status to 'active'
  551. int pm_generic_resume_noirq(struct device *dev);
  552. - invoke the ->resume_noirq() callback provided by the driver of this device
  553. int pm_generic_freeze(struct device *dev);
  554. - if the device has not been suspended at run time, invoke the ->freeze()
  555. callback provided by its driver and return its result, or return 0 if not
  556. defined
  557. int pm_generic_freeze_noirq(struct device *dev);
  558. - if pm_runtime_suspended(dev) returns "false", invoke the ->freeze_noirq()
  559. callback provided by the device's driver and return its result, or return
  560. 0 if not defined
  561. int pm_generic_thaw(struct device *dev);
  562. - if the device has not been suspended at run time, invoke the ->thaw()
  563. callback provided by its driver and return its result, or return 0 if not
  564. defined
  565. int pm_generic_thaw_noirq(struct device *dev);
  566. - if pm_runtime_suspended(dev) returns "false", invoke the ->thaw_noirq()
  567. callback provided by the device's driver and return its result, or return
  568. 0 if not defined
  569. int pm_generic_poweroff(struct device *dev);
  570. - if the device has not been suspended at run time, invoke the ->poweroff()
  571. callback provided by its driver and return its result, or return 0 if not
  572. defined
  573. int pm_generic_poweroff_noirq(struct device *dev);
  574. - if pm_runtime_suspended(dev) returns "false", run the ->poweroff_noirq()
  575. callback provided by the device's driver and return its result, or return
  576. 0 if not defined
  577. int pm_generic_restore(struct device *dev);
  578. - invoke the ->restore() callback provided by the driver of this device and,
  579. if successful, change the device's runtime PM status to 'active'
  580. int pm_generic_restore_noirq(struct device *dev);
  581. - invoke the ->restore_noirq() callback provided by the device's driver
  582. These functions can be assigned to the ->runtime_idle(), ->runtime_suspend(),
  583. ->runtime_resume(), ->suspend(), ->suspend_noirq(), ->resume(),
  584. ->resume_noirq(), ->freeze(), ->freeze_noirq(), ->thaw(), ->thaw_noirq(),
  585. ->poweroff(), ->poweroff_noirq(), ->restore(), ->restore_noirq() callback
  586. pointers in the subsystem-level dev_pm_ops structures.
  587. If a subsystem wishes to use all of them at the same time, it can simply assign
  588. the GENERIC_SUBSYS_PM_OPS macro, defined in include/linux/pm.h, to its
  589. dev_pm_ops structure pointer.
  590. Device drivers that wish to use the same function as a system suspend, freeze,
  591. poweroff and runtime suspend callback, and similarly for system resume, thaw,
  592. restore, and runtime resume, can achieve this with the help of the
  593. UNIVERSAL_DEV_PM_OPS macro defined in include/linux/pm.h (possibly setting its
  594. last argument to NULL).
  595. 8. "No-Callback" Devices
  596. Some "devices" are only logical sub-devices of their parent and cannot be
  597. power-managed on their own. (The prototype example is a USB interface. Entire
  598. USB devices can go into low-power mode or send wake-up requests, but neither is
  599. possible for individual interfaces.) The drivers for these devices have no
  600. need of runtime PM callbacks; if the callbacks did exist, ->runtime_suspend()
  601. and ->runtime_resume() would always return 0 without doing anything else and
  602. ->runtime_idle() would always call pm_runtime_suspend().
  603. Subsystems can tell the PM core about these devices by calling
  604. pm_runtime_no_callbacks(). This should be done after the device structure is
  605. initialized and before it is registered (although after device registration is
  606. also okay). The routine will set the device's power.no_callbacks flag and
  607. prevent the non-debugging runtime PM sysfs attributes from being created.
  608. When power.no_callbacks is set, the PM core will not invoke the
  609. ->runtime_idle(), ->runtime_suspend(), or ->runtime_resume() callbacks.
  610. Instead it will assume that suspends and resumes always succeed and that idle
  611. devices should be suspended.
  612. As a consequence, the PM core will never directly inform the device's subsystem
  613. or driver about runtime power changes. Instead, the driver for the device's
  614. parent must take responsibility for telling the device's driver when the
  615. parent's power state changes.
  616. 9. Autosuspend, or automatically-delayed suspends
  617. Changing a device's power state isn't free; it requires both time and energy.
  618. A device should be put in a low-power state only when there's some reason to
  619. think it will remain in that state for a substantial time. A common heuristic
  620. says that a device which hasn't been used for a while is liable to remain
  621. unused; following this advice, drivers should not allow devices to be suspended
  622. at runtime until they have been inactive for some minimum period. Even when
  623. the heuristic ends up being non-optimal, it will still prevent devices from
  624. "bouncing" too rapidly between low-power and full-power states.
  625. The term "autosuspend" is an historical remnant. It doesn't mean that the
  626. device is automatically suspended (the subsystem or driver still has to call
  627. the appropriate PM routines); rather it means that runtime suspends will
  628. automatically be delayed until the desired period of inactivity has elapsed.
  629. Inactivity is determined based on the power.last_busy field. Drivers should
  630. call pm_runtime_mark_last_busy() to update this field after carrying out I/O,
  631. typically just before calling pm_runtime_put_autosuspend(). The desired length
  632. of the inactivity period is a matter of policy. Subsystems can set this length
  633. initially by calling pm_runtime_set_autosuspend_delay(), but after device
  634. registration the length should be controlled by user space, using the
  635. /sys/devices/.../power/autosuspend_delay_ms attribute.
  636. In order to use autosuspend, subsystems or drivers must call
  637. pm_runtime_use_autosuspend() (preferably before registering the device), and
  638. thereafter they should use the various *_autosuspend() helper functions instead
  639. of the non-autosuspend counterparts:
  640. Instead of: pm_runtime_suspend use: pm_runtime_autosuspend;
  641. Instead of: pm_schedule_suspend use: pm_request_autosuspend;
  642. Instead of: pm_runtime_put use: pm_runtime_put_autosuspend;
  643. Instead of: pm_runtime_put_sync use: pm_runtime_put_sync_autosuspend.
  644. Drivers may also continue to use the non-autosuspend helper functions; they
  645. will behave normally, not taking the autosuspend delay into account.
  646. Similarly, if the power.use_autosuspend field isn't set then the autosuspend
  647. helper functions will behave just like the non-autosuspend counterparts.
  648. Under some circumstances a driver or subsystem may want to prevent a device
  649. from autosuspending immediately, even though the usage counter is zero and the
  650. autosuspend delay time has expired. If the ->runtime_suspend() callback
  651. returns -EAGAIN or -EBUSY, and if the next autosuspend delay expiration time is
  652. in the future (as it normally would be if the callback invoked
  653. pm_runtime_mark_last_busy()), the PM core will automatically reschedule the
  654. autosuspend. The ->runtime_suspend() callback can't do this rescheduling
  655. itself because no suspend requests of any kind are accepted while the device is
  656. suspending (i.e., while the callback is running).
  657. The implementation is well suited for asynchronous use in interrupt contexts.
  658. However such use inevitably involves races, because the PM core can't
  659. synchronize ->runtime_suspend() callbacks with the arrival of I/O requests.
  660. This synchronization must be handled by the driver, using its private lock.
  661. Here is a schematic pseudo-code example:
  662. foo_read_or_write(struct foo_priv *foo, void *data)
  663. {
  664. lock(&foo->private_lock);
  665. add_request_to_io_queue(foo, data);
  666. if (foo->num_pending_requests++ == 0)
  667. pm_runtime_get(&foo->dev);
  668. if (!foo->is_suspended)
  669. foo_process_next_request(foo);
  670. unlock(&foo->private_lock);
  671. }
  672. foo_io_completion(struct foo_priv *foo, void *req)
  673. {
  674. lock(&foo->private_lock);
  675. if (--foo->num_pending_requests == 0) {
  676. pm_runtime_mark_last_busy(&foo->dev);
  677. pm_runtime_put_autosuspend(&foo->dev);
  678. } else {
  679. foo_process_next_request(foo);
  680. }
  681. unlock(&foo->private_lock);
  682. /* Send req result back to the user ... */
  683. }
  684. int foo_runtime_suspend(struct device *dev)
  685. {
  686. struct foo_priv foo = container_of(dev, ...);
  687. int ret = 0;
  688. lock(&foo->private_lock);
  689. if (foo->num_pending_requests > 0) {
  690. ret = -EBUSY;
  691. } else {
  692. /* ... suspend the device ... */
  693. foo->is_suspended = 1;
  694. }
  695. unlock(&foo->private_lock);
  696. return ret;
  697. }
  698. int foo_runtime_resume(struct device *dev)
  699. {
  700. struct foo_priv foo = container_of(dev, ...);
  701. lock(&foo->private_lock);
  702. /* ... resume the device ... */
  703. foo->is_suspended = 0;
  704. pm_runtime_mark_last_busy(&foo->dev);
  705. if (foo->num_pending_requests > 0)
  706. foo_process_requests(foo);
  707. unlock(&foo->private_lock);
  708. return 0;
  709. }
  710. The important point is that after foo_io_completion() asks for an autosuspend,
  711. the foo_runtime_suspend() callback may race with foo_read_or_write().
  712. Therefore foo_runtime_suspend() has to check whether there are any pending I/O
  713. requests (while holding the private lock) before allowing the suspend to
  714. proceed.
  715. In addition, the power.autosuspend_delay field can be changed by user space at
  716. any time. If a driver cares about this, it can call
  717. pm_runtime_autosuspend_expiration() from within the ->runtime_suspend()
  718. callback while holding its private lock. If the function returns a nonzero
  719. value then the delay has not yet expired and the callback should return
  720. -EAGAIN.