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(C) 2014 Intel Corp., Rafael J. Wysocki <rafael.j.wysocki@intel.com> 6 71. Introduction 8 9Support for runtime power management (runtime PM) of I/O devices is provided 10at the power management core (PM core) level by means of: 11 12* The power management workqueue pm_wq in which bus types and device drivers can 13 put their PM-related work items. It is strongly recommended that pm_wq be 14 used for queuing all work items related to runtime PM, because this allows 15 them to be synchronized with system-wide power transitions (suspend to RAM, 16 hibernation and resume from system sleep states). pm_wq is declared in 17 include/linux/pm_runtime.h and defined in kernel/power/main.c. 18 19* A number of runtime PM fields in the 'power' member of 'struct device' (which 20 is of the type 'struct dev_pm_info', defined in include/linux/pm.h) that can 21 be used for synchronizing runtime PM operations with one another. 22 23* Three device runtime PM callbacks in 'struct dev_pm_ops' (defined in 24 include/linux/pm.h). 25 26* A set of helper functions defined in drivers/base/power/runtime.c that can be 27 used for carrying out runtime PM operations in such a way that the 28 synchronization between them is taken care of by the PM core. Bus types and 29 device drivers are encouraged to use these functions. 30 31The runtime PM callbacks present in 'struct dev_pm_ops', the device runtime PM 32fields of 'struct dev_pm_info' and the core helper functions provided for 33runtime PM are described below. 34 352. Device Runtime PM Callbacks 36 37There are three device runtime PM callbacks defined in 'struct dev_pm_ops': 38 39struct dev_pm_ops { 40 ... 41 int (*runtime_suspend)(struct device *dev); 42 int (*runtime_resume)(struct device *dev); 43 int (*runtime_idle)(struct device *dev); 44 ... 45}; 46 47The ->runtime_suspend(), ->runtime_resume() and ->runtime_idle() callbacks 48are executed by the PM core for the device's subsystem that may be either of 49the following: 50 51 1. PM domain of the device, if the device's PM domain object, dev->pm_domain, 52 is present. 53 54 2. Device type of the device, if both dev->type and dev->type->pm are present. 55 56 3. Device class of the device, if both dev->class and dev->class->pm are 57 present. 58 59 4. Bus type of the device, if both dev->bus and dev->bus->pm are present. 60 61If the subsystem chosen by applying the above rules doesn't provide the relevant 62callback, the PM core will invoke the corresponding driver callback stored in 63dev->driver->pm directly (if present). 64 65The PM core always checks which callback to use in the order given above, so the 66priority order of callbacks from high to low is: PM domain, device type, class 67and bus type. Moreover, the high-priority one will always take precedence over 68a low-priority one. The PM domain, bus type, device type and class callbacks 69are referred to as subsystem-level callbacks in what follows. 70 71By default, the callbacks are always invoked in process context with interrupts 72enabled. However, the pm_runtime_irq_safe() helper function can be used to tell 73the PM core that it is safe to run the ->runtime_suspend(), ->runtime_resume() 74and ->runtime_idle() callbacks for the given device in atomic context with 75interrupts disabled. This implies that the callback routines in question must 76not block or sleep, but it also means that the synchronous helper functions 77listed at the end of Section 4 may be used for that device within an interrupt 78handler or generally in an atomic context. 79 80The subsystem-level suspend callback, if present, is _entirely_ _responsible_ 81for handling the suspend of the device as appropriate, which may, but need not 82include executing the device driver's own ->runtime_suspend() callback (from the 83PM core's point of view it is not necessary to implement a ->runtime_suspend() 84callback in a device driver as long as the subsystem-level suspend callback 85knows what to do to handle the device). 86 87 * Once the subsystem-level suspend callback (or the driver suspend callback, 88 if invoked directly) has completed successfully for the given device, the PM 89 core regards the device as suspended, which need not mean that it has been 90 put into a low power state. It is supposed to mean, however, that the 91 device will not process data and will not communicate with the CPU(s) and 92 RAM until the appropriate resume callback is executed for it. The runtime 93 PM status of a device after successful execution of the suspend callback is 94 'suspended'. 95 96 * If the suspend callback returns -EBUSY or -EAGAIN, the device's runtime PM 97 status remains 'active', which means that the device _must_ be fully 98 operational afterwards. 99 100 * If the suspend callback returns an error code different from -EBUSY and 101 -EAGAIN, the PM core regards this as a fatal error and will refuse to run 102 the helper functions described in Section 4 for the device until its status 103 is directly set to either'active', or 'suspended' (the PM core provides 104 special helper functions for this purpose). 105 106In particular, if the driver requires remote wakeup capability (i.e. hardware 107mechanism allowing the device to request a change of its power state, such as 108PCI PME) for proper functioning and device_run_wake() returns 'false' for the 109device, then ->runtime_suspend() should return -EBUSY. On the other hand, if 110device_run_wake() returns 'true' for the device and the device is put into a 111low-power state during the execution of the suspend callback, it is expected 112that remote wakeup will be enabled for the device. Generally, remote wakeup 113should be enabled for all input devices put into low-power states at run time. 114 115The subsystem-level resume callback, if present, is _entirely_ _responsible_ for 116handling the resume of the device as appropriate, which may, but need not 117include executing the device driver's own ->runtime_resume() callback (from the 118PM core's point of view it is not necessary to implement a ->runtime_resume() 119callback in a device driver as long as the subsystem-level resume callback knows 120what to do to handle the device). 121 122 * Once the subsystem-level resume callback (or the driver resume callback, if 123 invoked directly) has completed successfully, the PM core regards the device 124 as fully operational, which means that the device _must_ be able to complete 125 I/O operations as needed. The runtime PM status of the device is then 126 'active'. 127 128 * If the resume callback returns an error code, the PM core regards this as a 129 fatal error and will refuse to run the helper functions described in Section 130 4 for the device, until its status is directly set to either 'active', or 131 'suspended' (by means of special helper functions provided by the PM core 132 for this purpose). 133 134The idle callback (a subsystem-level one, if present, or the driver one) is 135executed by the PM core whenever the device appears to be idle, which is 136indicated to the PM core by two counters, the device's usage counter and the 137counter of 'active' children of the device. 138 139 * If any of these counters is decreased using a helper function provided by 140 the PM core and it turns out to be equal to zero, the other counter is 141 checked. If that counter also is equal to zero, the PM core executes the 142 idle callback with the device as its argument. 143 144The action performed by the idle callback is totally dependent on the subsystem 145(or driver) in question, but the expected and recommended action is to check 146if the device can be suspended (i.e. if all of the conditions necessary for 147suspending the device are satisfied) and to queue up a suspend request for the 148device in that case. If there is no idle callback, or if the callback returns 1490, then the PM core will attempt to carry out a runtime suspend of the device, 150also respecting devices configured for autosuspend. In essence this means a 151call to pm_runtime_autosuspend() (do note that drivers needs to update the 152device last busy mark, pm_runtime_mark_last_busy(), to control the delay under 153this circumstance). To prevent this (for example, if the callback routine has 154started a delayed suspend), the routine must return a non-zero value. Negative 155error return codes are ignored by the PM core. 156 157The helper functions provided by the PM core, described in Section 4, guarantee 158that the following constraints are met with respect to runtime PM callbacks for 159one device: 160 161(1) The callbacks are mutually exclusive (e.g. it is forbidden to execute 162 ->runtime_suspend() in parallel with ->runtime_resume() or with another 163 instance of ->runtime_suspend() for the same device) with the exception that 164 ->runtime_suspend() or ->runtime_resume() can be executed in parallel with 165 ->runtime_idle() (although ->runtime_idle() will not be started while any 166 of the other callbacks is being executed for the same device). 167 168(2) ->runtime_idle() and ->runtime_suspend() can only be executed for 'active' 169 devices (i.e. the PM core will only execute ->runtime_idle() or 170 ->runtime_suspend() for the devices the runtime PM status of which is 171 'active'). 172 173(3) ->runtime_idle() and ->runtime_suspend() can only be executed for a device 174 the usage counter of which is equal to zero _and_ either the counter of 175 'active' children of which is equal to zero, or the 'power.ignore_children' 176 flag of which is set. 177 178(4) ->runtime_resume() can only be executed for 'suspended' devices (i.e. the 179 PM core will only execute ->runtime_resume() for the devices the runtime 180 PM status of which is 'suspended'). 181 182Additionally, the helper functions provided by the PM core obey the following 183rules: 184 185 * If ->runtime_suspend() is about to be executed or there's a pending request 186 to execute it, ->runtime_idle() will not be executed for the same device. 187 188 * A request to execute or to schedule the execution of ->runtime_suspend() 189 will cancel any pending requests to execute ->runtime_idle() for the same 190 device. 191 192 * If ->runtime_resume() is about to be executed or there's a pending request 193 to execute it, the other callbacks will not be executed for the same device. 194 195 * A request to execute ->runtime_resume() will cancel any pending or 196 scheduled requests to execute the other callbacks for the same device, 197 except for scheduled autosuspends. 198 1993. Runtime PM Device Fields 200 201The following device runtime PM fields are present in 'struct dev_pm_info', as 202defined in include/linux/pm.h: 203 204 struct timer_list suspend_timer; 205 - timer used for scheduling (delayed) suspend and autosuspend requests 206 207 unsigned long timer_expires; 208 - timer expiration time, in jiffies (if this is different from zero, the 209 timer is running and will expire at that time, otherwise the timer is not 210 running) 211 212 struct work_struct work; 213 - work structure used for queuing up requests (i.e. work items in pm_wq) 214 215 wait_queue_head_t wait_queue; 216 - wait queue used if any of the helper functions needs to wait for another 217 one to complete 218 219 spinlock_t lock; 220 - lock used for synchronisation 221 222 atomic_t usage_count; 223 - the usage counter of the device 224 225 atomic_t child_count; 226 - the count of 'active' children of the device 227 228 unsigned int ignore_children; 229 - if set, the value of child_count is ignored (but still updated) 230 231 unsigned int disable_depth; 232 - used for disabling the helper functions (they work normally if this is 233 equal to zero); the initial value of it is 1 (i.e. runtime PM is 234 initially disabled for all devices) 235 236 int runtime_error; 237 - if set, there was a fatal error (one of the callbacks returned error code 238 as described in Section 2), so the helper functions will not work until 239 this flag is cleared; this is the error code returned by the failing 240 callback 241 242 unsigned int idle_notification; 243 - if set, ->runtime_idle() is being executed 244 245 unsigned int request_pending; 246 - if set, there's a pending request (i.e. a work item queued up into pm_wq) 247 248 enum rpm_request request; 249 - type of request that's pending (valid if request_pending is set) 250 251 unsigned int deferred_resume; 252 - set if ->runtime_resume() is about to be run while ->runtime_suspend() is 253 being executed for that device and it is not practical to wait for the 254 suspend to complete; means "start a resume as soon as you've suspended" 255 256 unsigned int run_wake; 257 - set if the device is capable of generating runtime wake-up events 258 259 enum rpm_status runtime_status; 260 - the runtime PM status of the device; this field's initial value is 261 RPM_SUSPENDED, which means that each device is initially regarded by the 262 PM core as 'suspended', regardless of its real hardware status 263 264 unsigned int runtime_auto; 265 - if set, indicates that the user space has allowed the device driver to 266 power manage the device at run time via the /sys/devices/.../power/control 267 interface; it may only be modified with the help of the pm_runtime_allow() 268 and pm_runtime_forbid() helper functions 269 270 unsigned int no_callbacks; 271 - indicates that the device does not use the runtime PM callbacks (see 272 Section 8); it may be modified only by the pm_runtime_no_callbacks() 273 helper function 274 275 unsigned int irq_safe; 276 - indicates that the ->runtime_suspend() and ->runtime_resume() callbacks 277 will be invoked with the spinlock held and interrupts disabled 278 279 unsigned int use_autosuspend; 280 - indicates that the device's driver supports delayed autosuspend (see 281 Section 9); it may be modified only by the 282 pm_runtime{_dont}_use_autosuspend() helper functions 283 284 unsigned int timer_autosuspends; 285 - indicates that the PM core should attempt to carry out an autosuspend 286 when the timer expires rather than a normal suspend 287 288 int autosuspend_delay; 289 - the delay time (in milliseconds) to be used for autosuspend 290 291 unsigned long last_busy; 292 - the time (in jiffies) when the pm_runtime_mark_last_busy() helper 293 function was last called for this device; used in calculating inactivity 294 periods for autosuspend 295 296All of the above fields are members of the 'power' member of 'struct device'. 297 2984. Runtime PM Device Helper Functions 299 300The following runtime PM helper functions are defined in 301drivers/base/power/runtime.c and include/linux/pm_runtime.h: 302 303 void pm_runtime_init(struct device *dev); 304 - initialize the device runtime PM fields in 'struct dev_pm_info' 305 306 void pm_runtime_remove(struct device *dev); 307 - make sure that the runtime PM of the device will be disabled after 308 removing the device from device hierarchy 309 310 int pm_runtime_idle(struct device *dev); 311 - execute the subsystem-level idle callback for the device; returns an 312 error code on failure, where -EINPROGRESS means that ->runtime_idle() is 313 already being executed; if there is no callback or the callback returns 0 314 then run pm_runtime_autosuspend(dev) and return its result 315 316 int pm_runtime_suspend(struct device *dev); 317 - execute the subsystem-level suspend callback for the device; returns 0 on 318 success, 1 if the device's runtime PM status was already 'suspended', or 319 error code on failure, where -EAGAIN or -EBUSY means it is safe to attempt 320 to suspend the device again in future and -EACCES means that 321 'power.disable_depth' is different from 0 322 323 int pm_runtime_autosuspend(struct device *dev); 324 - same as pm_runtime_suspend() except that the autosuspend delay is taken 325 into account; if pm_runtime_autosuspend_expiration() says the delay has 326 not yet expired then an autosuspend is scheduled for the appropriate time 327 and 0 is returned 328 329 int pm_runtime_resume(struct device *dev); 330 - execute the subsystem-level resume callback for the device; returns 0 on 331 success, 1 if the device's runtime PM status was already 'active' or 332 error code on failure, where -EAGAIN means it may be safe to attempt to 333 resume the device again in future, but 'power.runtime_error' should be 334 checked additionally, and -EACCES means that 'power.disable_depth' is 335 different from 0 336 337 int pm_request_idle(struct device *dev); 338 - submit a request to execute the subsystem-level idle callback for the 339 device (the request is represented by a work item in pm_wq); returns 0 on 340 success or error code if the request has not been queued up 341 342 int pm_request_autosuspend(struct device *dev); 343 - schedule the execution of the subsystem-level suspend callback for the 344 device when the autosuspend delay has expired; if the delay has already 345 expired then the work item is queued up immediately 346 347 int pm_schedule_suspend(struct device *dev, unsigned int delay); 348 - schedule the execution of the subsystem-level suspend callback for the 349 device in future, where 'delay' is the time to wait before queuing up a 350 suspend work item in pm_wq, in milliseconds (if 'delay' is zero, the work 351 item is queued up immediately); returns 0 on success, 1 if the device's PM 352 runtime status was already 'suspended', or error code if the request 353 hasn't been scheduled (or queued up if 'delay' is 0); if the execution of 354 ->runtime_suspend() is already scheduled and not yet expired, the new 355 value of 'delay' will be used as the time to wait 356 357 int pm_request_resume(struct device *dev); 358 - submit a request to execute the subsystem-level resume callback for the 359 device (the request is represented by a work item in pm_wq); returns 0 on 360 success, 1 if the device's runtime PM status was already 'active', or 361 error code if the request hasn't been queued up 362 363 void pm_runtime_get_noresume(struct device *dev); 364 - increment the device's usage counter 365 366 int pm_runtime_get(struct device *dev); 367 - increment the device's usage counter, run pm_request_resume(dev) and 368 return its result 369 370 int pm_runtime_get_sync(struct device *dev); 371 - increment the device's usage counter, run pm_runtime_resume(dev) and 372 return its result 373 374 void pm_runtime_put_noidle(struct device *dev); 375 - decrement the device's usage counter 376 377 int pm_runtime_put(struct device *dev); 378 - decrement the device's usage counter; if the result is 0 then run 379 pm_request_idle(dev) and return its result 380 381 int pm_runtime_put_autosuspend(struct device *dev); 382 - decrement the device's usage counter; if the result is 0 then run 383 pm_request_autosuspend(dev) and return its result 384 385 int pm_runtime_put_sync(struct device *dev); 386 - decrement the device's usage counter; if the result is 0 then run 387 pm_runtime_idle(dev) and return its result 388 389 int pm_runtime_put_sync_suspend(struct device *dev); 390 - decrement the device's usage counter; if the result is 0 then run 391 pm_runtime_suspend(dev) and return its result 392 393 int pm_runtime_put_sync_autosuspend(struct device *dev); 394 - decrement the device's usage counter; if the result is 0 then run 395 pm_runtime_autosuspend(dev) and return its result 396 397 void pm_runtime_enable(struct device *dev); 398 - decrement the device's 'power.disable_depth' field; if that field is equal 399 to zero, the runtime PM helper functions can execute subsystem-level 400 callbacks described in Section 2 for the device 401 402 int pm_runtime_disable(struct device *dev); 403 - increment the device's 'power.disable_depth' field (if the value of that 404 field was previously zero, this prevents subsystem-level runtime PM 405 callbacks from being run for the device), make sure that all of the 406 pending runtime PM operations on the device are either completed or 407 canceled; returns 1 if there was a resume request pending and it was 408 necessary to execute the subsystem-level resume callback for the device 409 to satisfy that request, otherwise 0 is returned 410 411 int pm_runtime_barrier(struct device *dev); 412 - check if there's a resume request pending for the device and resume it 413 (synchronously) in that case, cancel any other pending runtime PM requests 414 regarding it and wait for all runtime PM operations on it in progress to 415 complete; returns 1 if there was a resume request pending and it was 416 necessary to execute the subsystem-level resume callback for the device to 417 satisfy that request, otherwise 0 is returned 418 419 void pm_suspend_ignore_children(struct device *dev, bool enable); 420 - set/unset the power.ignore_children flag of the device 421 422 int pm_runtime_set_active(struct device *dev); 423 - clear the device's 'power.runtime_error' flag, set the device's runtime 424 PM status to 'active' 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); it will fail and return error code if the device has a parent 428 which is not active and the 'power.ignore_children' flag of which is unset 429 430 void pm_runtime_set_suspended(struct device *dev); 431 - clear the device's 'power.runtime_error' flag, set the device's runtime 432 PM status to 'suspended' and update its parent's counter of 'active' 433 children as appropriate (it is only valid to use this function if 434 'power.runtime_error' is set or 'power.disable_depth' is greater than 435 zero) 436 437 bool pm_runtime_active(struct device *dev); 438 - return true if the device's runtime PM status is 'active' or its 439 'power.disable_depth' field is not equal to zero, or false otherwise 440 441 bool pm_runtime_suspended(struct device *dev); 442 - return true if the device's runtime PM status is 'suspended' and its 443 'power.disable_depth' field is equal to zero, or false otherwise 444 445 bool pm_runtime_status_suspended(struct device *dev); 446 - return true if the device's runtime PM status is 'suspended' 447 448 void pm_runtime_allow(struct device *dev); 449 - set the power.runtime_auto flag for the device and decrease its usage 450 counter (used by the /sys/devices/.../power/control interface to 451 effectively allow the device to be power managed at run time) 452 453 void pm_runtime_forbid(struct device *dev); 454 - unset the power.runtime_auto flag for the device and increase its usage 455 counter (used by the /sys/devices/.../power/control interface to 456 effectively prevent the device from being power managed at run time) 457 458 void pm_runtime_no_callbacks(struct device *dev); 459 - set the power.no_callbacks flag for the device and remove the runtime 460 PM attributes from /sys/devices/.../power (or prevent them from being 461 added when the device is registered) 462 463 void pm_runtime_irq_safe(struct device *dev); 464 - set the power.irq_safe flag for the device, causing the runtime-PM 465 callbacks to be invoked with interrupts off 466 467 bool pm_runtime_is_irq_safe(struct device *dev); 468 - return true if power.irq_safe flag was set for the device, causing 469 the runtime-PM callbacks to be invoked with interrupts off 470 471 void pm_runtime_mark_last_busy(struct device *dev); 472 - set the power.last_busy field to the current time 473 474 void pm_runtime_use_autosuspend(struct device *dev); 475 - set the power.use_autosuspend flag, enabling autosuspend delays 476 477 void pm_runtime_dont_use_autosuspend(struct device *dev); 478 - clear the power.use_autosuspend flag, disabling autosuspend delays 479 480 void pm_runtime_set_autosuspend_delay(struct device *dev, int delay); 481 - set the power.autosuspend_delay value to 'delay' (expressed in 482 milliseconds); if 'delay' is negative then runtime suspends are 483 prevented 484 485 unsigned long pm_runtime_autosuspend_expiration(struct device *dev); 486 - calculate the time when the current autosuspend delay period will expire, 487 based on power.last_busy and power.autosuspend_delay; if the delay time 488 is 1000 ms or larger then the expiration time is rounded up to the 489 nearest second; returns 0 if the delay period has already expired or 490 power.use_autosuspend isn't set, otherwise returns the expiration time 491 in jiffies 492 493It is safe to execute the following helper functions from interrupt context: 494 495pm_request_idle() 496pm_request_autosuspend() 497pm_schedule_suspend() 498pm_request_resume() 499pm_runtime_get_noresume() 500pm_runtime_get() 501pm_runtime_put_noidle() 502pm_runtime_put() 503pm_runtime_put_autosuspend() 504pm_runtime_enable() 505pm_suspend_ignore_children() 506pm_runtime_set_active() 507pm_runtime_set_suspended() 508pm_runtime_suspended() 509pm_runtime_mark_last_busy() 510pm_runtime_autosuspend_expiration() 511 512If pm_runtime_irq_safe() has been called for a device then the following helper 513functions may also be used in interrupt context: 514 515pm_runtime_idle() 516pm_runtime_suspend() 517pm_runtime_autosuspend() 518pm_runtime_resume() 519pm_runtime_get_sync() 520pm_runtime_put_sync() 521pm_runtime_put_sync_suspend() 522pm_runtime_put_sync_autosuspend() 523 5245. Runtime PM Initialization, Device Probing and Removal 525 526Initially, the runtime PM is disabled for all devices, which means that the 527majority of the runtime PM helper functions described in Section 4 will return 528-EAGAIN until pm_runtime_enable() is called for the device. 529 530In addition to that, the initial runtime PM status of all devices is 531'suspended', but it need not reflect the actual physical state of the device. 532Thus, if the device is initially active (i.e. it is able to process I/O), its 533runtime PM status must be changed to 'active', with the help of 534pm_runtime_set_active(), before pm_runtime_enable() is called for the device. 535 536However, if the device has a parent and the parent's runtime PM is enabled, 537calling pm_runtime_set_active() for the device will affect the parent, unless 538the parent's 'power.ignore_children' flag is set. Namely, in that case the 539parent won't be able to suspend at run time, using the PM core's helper 540functions, as long as the child's status is 'active', even if the child's 541runtime PM is still disabled (i.e. pm_runtime_enable() hasn't been called for 542the child yet or pm_runtime_disable() has been called for it). For this reason, 543once pm_runtime_set_active() has been called for the device, pm_runtime_enable() 544should be called for it too as soon as reasonably possible or its runtime PM 545status should be changed back to 'suspended' with the help of 546pm_runtime_set_suspended(). 547 548If the default initial runtime PM status of the device (i.e. 'suspended') 549reflects the actual state of the device, its bus type's or its driver's 550->probe() callback will likely need to wake it up using one of the PM core's 551helper functions described in Section 4. In that case, pm_runtime_resume() 552should be used. Of course, for this purpose the device's runtime PM has to be 553enabled earlier by calling pm_runtime_enable(). 554 555Note, if the device may execute pm_runtime calls during the probe (such as 556if it is registers with a subsystem that may call back in) then the 557pm_runtime_get_sync() call paired with a pm_runtime_put() call will be 558appropriate to ensure that the device is not put back to sleep during the 559probe. This can happen with systems such as the network device layer. 560 561It may be desirable to suspend the device once ->probe() has finished. 562Therefore the driver core uses the asyncronous pm_request_idle() to submit a 563request to execute the subsystem-level idle callback for the device at that 564time. A driver that makes use of the runtime autosuspend feature, may want to 565update the last busy mark before returning from ->probe(). 566 567Moreover, the driver core prevents runtime PM callbacks from racing with the bus 568notifier callback in __device_release_driver(), which is necessary, because the 569notifier is used by some subsystems to carry out operations affecting the 570runtime PM functionality. It does so by calling pm_runtime_get_sync() before 571driver_sysfs_remove() and the BUS_NOTIFY_UNBIND_DRIVER notifications. This 572resumes the device if it's in the suspended state and prevents it from 573being suspended again while those routines are being executed. 574 575To allow bus types and drivers to put devices into the suspended state by 576calling pm_runtime_suspend() from their ->remove() routines, the driver core 577executes pm_runtime_put_sync() after running the BUS_NOTIFY_UNBIND_DRIVER 578notifications in __device_release_driver(). This requires bus types and 579drivers to make their ->remove() callbacks avoid races with runtime PM directly, 580but also it allows of more flexibility in the handling of devices during the 581removal of their drivers. 582 583The user space can effectively disallow the driver of the device to power manage 584it at run time by changing the value of its /sys/devices/.../power/control 585attribute to "on", which causes pm_runtime_forbid() to be called. In principle, 586this mechanism may also be used by the driver to effectively turn off the 587runtime power management of the device until the user space turns it on. 588Namely, during the initialization the driver can make sure that the runtime PM 589status of the device is 'active' and call pm_runtime_forbid(). It should be 590noted, however, that if the user space has already intentionally changed the 591value of /sys/devices/.../power/control to "auto" to allow the driver to power 592manage the device at run time, the driver may confuse it by using 593pm_runtime_forbid() this way. 594 5956. Runtime PM and System Sleep 596 597Runtime PM and system sleep (i.e., system suspend and hibernation, also known 598as suspend-to-RAM and suspend-to-disk) interact with each other in a couple of 599ways. If a device is active when a system sleep starts, everything is 600straightforward. But what should happen if the device is already suspended? 601 602The device may have different wake-up settings for runtime PM and system sleep. 603For example, remote wake-up may be enabled for runtime suspend but disallowed 604for system sleep (device_may_wakeup(dev) returns 'false'). When this happens, 605the subsystem-level system suspend callback is responsible for changing the 606device's wake-up setting (it may leave that to the device driver's system 607suspend routine). It may be necessary to resume the device and suspend it again 608in order to do so. The same is true if the driver uses different power levels 609or other settings for runtime suspend and system sleep. 610 611During system resume, the simplest approach is to bring all devices back to full 612power, even if they had been suspended before the system suspend began. There 613are several reasons for this, including: 614 615 * The device might need to switch power levels, wake-up settings, etc. 616 617 * Remote wake-up events might have been lost by the firmware. 618 619 * The device's children may need the device to be at full power in order 620 to resume themselves. 621 622 * The driver's idea of the device state may not agree with the device's 623 physical state. This can happen during resume from hibernation. 624 625 * The device might need to be reset. 626 627 * Even though the device was suspended, if its usage counter was > 0 then most 628 likely it would need a runtime resume in the near future anyway. 629 630If the device had been suspended before the system suspend began and it's 631brought back to full power during resume, then its runtime PM status will have 632to be updated to reflect the actual post-system sleep status. The way to do 633this is: 634 635 pm_runtime_disable(dev); 636 pm_runtime_set_active(dev); 637 pm_runtime_enable(dev); 638 639The PM core always increments the runtime usage counter before calling the 640->suspend() callback and decrements it after calling the ->resume() callback. 641Hence disabling runtime PM temporarily like this will not cause any runtime 642suspend attempts to be permanently lost. If the usage count goes to zero 643following the return of the ->resume() callback, the ->runtime_idle() callback 644will be invoked as usual. 645 646On some systems, however, system sleep is not entered through a global firmware 647or hardware operation. Instead, all hardware components are put into low-power 648states directly by the kernel in a coordinated way. Then, the system sleep 649state effectively follows from the states the hardware components end up in 650and the system is woken up from that state by a hardware interrupt or a similar 651mechanism entirely under the kernel's control. As a result, the kernel never 652gives control away and the states of all devices during resume are precisely 653known to it. If that is the case and none of the situations listed above takes 654place (in particular, if the system is not waking up from hibernation), it may 655be more efficient to leave the devices that had been suspended before the system 656suspend began in the suspended state. 657 658To this end, the PM core provides a mechanism allowing some coordination between 659different levels of device hierarchy. Namely, if a system suspend .prepare() 660callback returns a positive number for a device, that indicates to the PM core 661that the device appears to be runtime-suspended and its state is fine, so it 662may be left in runtime suspend provided that all of its descendants are also 663left in runtime suspend. If that happens, the PM core will not execute any 664system suspend and resume callbacks for all of those devices, except for the 665complete callback, which is then entirely responsible for handling the device 666as appropriate. This only applies to system suspend transitions that are not 667related to hibernation (see Documentation/power/devices.txt for more 668information). 669 670The PM core does its best to reduce the probability of race conditions between 671the runtime PM and system suspend/resume (and hibernation) callbacks by carrying 672out the following operations: 673 674 * During system suspend pm_runtime_get_noresume() is called for every device 675 right before executing the subsystem-level .prepare() callback for it and 676 pm_runtime_barrier() is called for every device right before executing the 677 subsystem-level .suspend() callback for it. In addition to that the PM core 678 calls __pm_runtime_disable() with 'false' as the second argument for every 679 device right before executing the subsystem-level .suspend_late() callback 680 for it. 681 682 * During system resume pm_runtime_enable() and pm_runtime_put() are called for 683 every device right after executing the subsystem-level .resume_early() 684 callback and right after executing the subsystem-level .complete() callback 685 for it, respectively. 686 6877. Generic subsystem callbacks 688 689Subsystems may wish to conserve code space by using the set of generic power 690management callbacks provided by the PM core, defined in 691driver/base/power/generic_ops.c: 692 693 int pm_generic_runtime_suspend(struct device *dev); 694 - invoke the ->runtime_suspend() callback provided by the driver of this 695 device and return its result, or return 0 if not defined 696 697 int pm_generic_runtime_resume(struct device *dev); 698 - invoke the ->runtime_resume() callback provided by the driver of this 699 device and return its result, or return 0 if not defined 700 701 int pm_generic_suspend(struct device *dev); 702 - if the device has not been suspended at run time, invoke the ->suspend() 703 callback provided by its driver and return its result, or return 0 if not 704 defined 705 706 int pm_generic_suspend_noirq(struct device *dev); 707 - if pm_runtime_suspended(dev) returns "false", invoke the ->suspend_noirq() 708 callback provided by the device's driver and return its result, or return 709 0 if not defined 710 711 int pm_generic_resume(struct device *dev); 712 - invoke the ->resume() callback provided by the driver of this device and, 713 if successful, change the device's runtime PM status to 'active' 714 715 int pm_generic_resume_noirq(struct device *dev); 716 - invoke the ->resume_noirq() callback provided by the driver of this device 717 718 int pm_generic_freeze(struct device *dev); 719 - if the device has not been suspended at run time, invoke the ->freeze() 720 callback provided by its driver and return its result, or return 0 if not 721 defined 722 723 int pm_generic_freeze_noirq(struct device *dev); 724 - if pm_runtime_suspended(dev) returns "false", invoke the ->freeze_noirq() 725 callback provided by the device's driver and return its result, or return 726 0 if not defined 727 728 int pm_generic_thaw(struct device *dev); 729 - if the device has not been suspended at run time, invoke the ->thaw() 730 callback provided by its driver and return its result, or return 0 if not 731 defined 732 733 int pm_generic_thaw_noirq(struct device *dev); 734 - if pm_runtime_suspended(dev) returns "false", invoke the ->thaw_noirq() 735 callback provided by the device's driver and return its result, or return 736 0 if not defined 737 738 int pm_generic_poweroff(struct device *dev); 739 - if the device has not been suspended at run time, invoke the ->poweroff() 740 callback provided by its driver and return its result, or return 0 if not 741 defined 742 743 int pm_generic_poweroff_noirq(struct device *dev); 744 - if pm_runtime_suspended(dev) returns "false", run the ->poweroff_noirq() 745 callback provided by the device's driver and return its result, or return 746 0 if not defined 747 748 int pm_generic_restore(struct device *dev); 749 - invoke the ->restore() callback provided by the driver of this device and, 750 if successful, change the device's runtime PM status to 'active' 751 752 int pm_generic_restore_noirq(struct device *dev); 753 - invoke the ->restore_noirq() callback provided by the device's driver 754 755These functions are the defaults used by the PM core, if a subsystem doesn't 756provide its own callbacks for ->runtime_idle(), ->runtime_suspend(), 757->runtime_resume(), ->suspend(), ->suspend_noirq(), ->resume(), 758->resume_noirq(), ->freeze(), ->freeze_noirq(), ->thaw(), ->thaw_noirq(), 759->poweroff(), ->poweroff_noirq(), ->restore(), ->restore_noirq() in the 760subsystem-level dev_pm_ops structure. 761 762Device drivers that wish to use the same function as a system suspend, freeze, 763poweroff and runtime suspend callback, and similarly for system resume, thaw, 764restore, and runtime resume, can achieve this with the help of the 765UNIVERSAL_DEV_PM_OPS macro defined in include/linux/pm.h (possibly setting its 766last argument to NULL). 767 7688. "No-Callback" Devices 769 770Some "devices" are only logical sub-devices of their parent and cannot be 771power-managed on their own. (The prototype example is a USB interface. Entire 772USB devices can go into low-power mode or send wake-up requests, but neither is 773possible for individual interfaces.) The drivers for these devices have no 774need of runtime PM callbacks; if the callbacks did exist, ->runtime_suspend() 775and ->runtime_resume() would always return 0 without doing anything else and 776->runtime_idle() would always call pm_runtime_suspend(). 777 778Subsystems can tell the PM core about these devices by calling 779pm_runtime_no_callbacks(). This should be done after the device structure is 780initialized and before it is registered (although after device registration is 781also okay). The routine will set the device's power.no_callbacks flag and 782prevent the non-debugging runtime PM sysfs attributes from being created. 783 784When power.no_callbacks is set, the PM core will not invoke the 785->runtime_idle(), ->runtime_suspend(), or ->runtime_resume() callbacks. 786Instead it will assume that suspends and resumes always succeed and that idle 787devices should be suspended. 788 789As a consequence, the PM core will never directly inform the device's subsystem 790or driver about runtime power changes. Instead, the driver for the device's 791parent must take responsibility for telling the device's driver when the 792parent's power state changes. 793 7949. Autosuspend, or automatically-delayed suspends 795 796Changing a device's power state isn't free; it requires both time and energy. 797A device should be put in a low-power state only when there's some reason to 798think it will remain in that state for a substantial time. A common heuristic 799says that a device which hasn't been used for a while is liable to remain 800unused; following this advice, drivers should not allow devices to be suspended 801at runtime until they have been inactive for some minimum period. Even when 802the heuristic ends up being non-optimal, it will still prevent devices from 803"bouncing" too rapidly between low-power and full-power states. 804 805The term "autosuspend" is an historical remnant. It doesn't mean that the 806device is automatically suspended (the subsystem or driver still has to call 807the appropriate PM routines); rather it means that runtime suspends will 808automatically be delayed until the desired period of inactivity has elapsed. 809 810Inactivity is determined based on the power.last_busy field. Drivers should 811call pm_runtime_mark_last_busy() to update this field after carrying out I/O, 812typically just before calling pm_runtime_put_autosuspend(). The desired length 813of the inactivity period is a matter of policy. Subsystems can set this length 814initially by calling pm_runtime_set_autosuspend_delay(), but after device 815registration the length should be controlled by user space, using the 816/sys/devices/.../power/autosuspend_delay_ms attribute. 817 818In order to use autosuspend, subsystems or drivers must call 819pm_runtime_use_autosuspend() (preferably before registering the device), and 820thereafter they should use the various *_autosuspend() helper functions instead 821of the non-autosuspend counterparts: 822 823 Instead of: pm_runtime_suspend use: pm_runtime_autosuspend; 824 Instead of: pm_schedule_suspend use: pm_request_autosuspend; 825 Instead of: pm_runtime_put use: pm_runtime_put_autosuspend; 826 Instead of: pm_runtime_put_sync use: pm_runtime_put_sync_autosuspend. 827 828Drivers may also continue to use the non-autosuspend helper functions; they 829will behave normally, not taking the autosuspend delay into account. 830Similarly, if the power.use_autosuspend field isn't set then the autosuspend 831helper functions will behave just like the non-autosuspend counterparts. 832 833Under some circumstances a driver or subsystem may want to prevent a device 834from autosuspending immediately, even though the usage counter is zero and the 835autosuspend delay time has expired. If the ->runtime_suspend() callback 836returns -EAGAIN or -EBUSY, and if the next autosuspend delay expiration time is 837in the future (as it normally would be if the callback invoked 838pm_runtime_mark_last_busy()), the PM core will automatically reschedule the 839autosuspend. The ->runtime_suspend() callback can't do this rescheduling 840itself because no suspend requests of any kind are accepted while the device is 841suspending (i.e., while the callback is running). 842 843The implementation is well suited for asynchronous use in interrupt contexts. 844However such use inevitably involves races, because the PM core can't 845synchronize ->runtime_suspend() callbacks with the arrival of I/O requests. 846This synchronization must be handled by the driver, using its private lock. 847Here is a schematic pseudo-code example: 848 849 foo_read_or_write(struct foo_priv *foo, void *data) 850 { 851 lock(&foo->private_lock); 852 add_request_to_io_queue(foo, data); 853 if (foo->num_pending_requests++ == 0) 854 pm_runtime_get(&foo->dev); 855 if (!foo->is_suspended) 856 foo_process_next_request(foo); 857 unlock(&foo->private_lock); 858 } 859 860 foo_io_completion(struct foo_priv *foo, void *req) 861 { 862 lock(&foo->private_lock); 863 if (--foo->num_pending_requests == 0) { 864 pm_runtime_mark_last_busy(&foo->dev); 865 pm_runtime_put_autosuspend(&foo->dev); 866 } else { 867 foo_process_next_request(foo); 868 } 869 unlock(&foo->private_lock); 870 /* Send req result back to the user ... */ 871 } 872 873 int foo_runtime_suspend(struct device *dev) 874 { 875 struct foo_priv foo = container_of(dev, ...); 876 int ret = 0; 877 878 lock(&foo->private_lock); 879 if (foo->num_pending_requests > 0) { 880 ret = -EBUSY; 881 } else { 882 /* ... suspend the device ... */ 883 foo->is_suspended = 1; 884 } 885 unlock(&foo->private_lock); 886 return ret; 887 } 888 889 int foo_runtime_resume(struct device *dev) 890 { 891 struct foo_priv foo = container_of(dev, ...); 892 893 lock(&foo->private_lock); 894 /* ... resume the device ... */ 895 foo->is_suspended = 0; 896 pm_runtime_mark_last_busy(&foo->dev); 897 if (foo->num_pending_requests > 0) 898 foo_process_next_request(foo); 899 unlock(&foo->private_lock); 900 return 0; 901 } 902 903The important point is that after foo_io_completion() asks for an autosuspend, 904the foo_runtime_suspend() callback may race with foo_read_or_write(). 905Therefore foo_runtime_suspend() has to check whether there are any pending I/O 906requests (while holding the private lock) before allowing the suspend to 907proceed. 908 909In addition, the power.autosuspend_delay field can be changed by user space at 910any time. If a driver cares about this, it can call 911pm_runtime_autosuspend_expiration() from within the ->runtime_suspend() 912callback while holding its private lock. If the function returns a nonzero 913value then the delay has not yet expired and the callback should return 914-EAGAIN. 915