1Run-time Power Management Framework for I/O Devices 2 3(C) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc. 4 51. Introduction 6 7Support for run-time power management (run-time PM) of I/O devices is provided 8at the power management core (PM core) level by means of: 9 10* The power management workqueue pm_wq in which bus types and device drivers can 11 put their PM-related work items. It is strongly recommended that pm_wq be 12 used for queuing all work items related to run-time PM, because this allows 13 them to be synchronized with system-wide power transitions (suspend to RAM, 14 hibernation and resume from system sleep states). pm_wq is declared in 15 include/linux/pm_runtime.h and defined in kernel/power/main.c. 16 17* A number of run-time PM fields in the 'power' member of 'struct device' (which 18 is of the type 'struct dev_pm_info', defined in include/linux/pm.h) that can 19 be used for synchronizing run-time PM operations with one another. 20 21* Three device run-time PM callbacks in 'struct dev_pm_ops' (defined in 22 include/linux/pm.h). 23 24* A set of helper functions defined in drivers/base/power/runtime.c that can be 25 used for carrying out run-time PM operations in such a way that the 26 synchronization between them is taken care of by the PM core. Bus types and 27 device drivers are encouraged to use these functions. 28 29The run-time PM callbacks present in 'struct dev_pm_ops', the device run-time PM 30fields of 'struct dev_pm_info' and the core helper functions provided for 31run-time PM are described below. 32 332. Device Run-time PM Callbacks 34 35There are three device run-time PM callbacks defined in 'struct dev_pm_ops': 36 37struct dev_pm_ops { 38 ... 39 int (*runtime_suspend)(struct device *dev); 40 int (*runtime_resume)(struct device *dev); 41 int (*runtime_idle)(struct device *dev); 42 ... 43}; 44 45The ->runtime_suspend(), ->runtime_resume() and ->runtime_idle() callbacks are 46executed by the PM core for either the bus type, or device type (if the bus 47type's callback is not defined), or device class (if the bus type's and device 48type's callbacks are not defined) of given device. The bus type, device type 49and device class callbacks are referred to as subsystem-level callbacks in what 50follows. 51 52The subsystem-level suspend callback is _entirely_ _responsible_ for handling 53the suspend of the device as appropriate, which may, but need not include 54executing the device driver's own ->runtime_suspend() callback (from the 55PM core's point of view it is not necessary to implement a ->runtime_suspend() 56callback in a device driver as long as the subsystem-level suspend callback 57knows what to do to handle the device). 58 59 * Once the subsystem-level suspend callback has completed successfully 60 for given device, the PM core regards the device as suspended, which need 61 not mean that the device has been put into a low power state. It is 62 supposed to mean, however, that the device will not process data and will 63 not communicate with the CPU(s) and RAM until the subsystem-level resume 64 callback is executed for it. The run-time PM status of a device after 65 successful execution of the subsystem-level suspend callback is 'suspended'. 66 67 * If the subsystem-level suspend callback returns -EBUSY or -EAGAIN, 68 the device's run-time PM status is 'active', which means that the device 69 _must_ be fully operational afterwards. 70 71 * If the subsystem-level suspend callback returns an error code different 72 from -EBUSY or -EAGAIN, the PM core regards this as a fatal error and will 73 refuse to run the helper functions described in Section 4 for the device, 74 until the status of it is directly set either to 'active', or to 'suspended' 75 (the PM core provides special helper functions for this purpose). 76 77In particular, if the driver requires remote wake-up capability (i.e. hardware 78mechanism allowing the device to request a change of its power state, such as 79PCI PME) for proper functioning and device_run_wake() returns 'false' for the 80device, then ->runtime_suspend() should return -EBUSY. On the other hand, if 81device_run_wake() returns 'true' for the device and the device is put into a low 82power state during the execution of the subsystem-level suspend callback, it is 83expected that remote wake-up will be enabled for the device. Generally, remote 84wake-up should be enabled for all input devices put into a low power state at 85run time. 86 87The subsystem-level resume callback is _entirely_ _responsible_ for handling the 88resume of the device as appropriate, which may, but need not include executing 89the device driver's own ->runtime_resume() callback (from the PM core's point of 90view it is not necessary to implement a ->runtime_resume() callback in a device 91driver as long as the subsystem-level resume callback knows what to do to handle 92the device). 93 94 * Once the subsystem-level resume callback has completed successfully, the PM 95 core regards the device as fully operational, which means that the device 96 _must_ be able to complete I/O operations as needed. The run-time PM status 97 of the device is then 'active'. 98 99 * If the subsystem-level resume callback returns an error code, the PM core 100 regards this as a fatal error and will refuse to run the helper functions 101 described in Section 4 for the device, until its status is directly set 102 either to 'active' or to 'suspended' (the PM core provides special helper 103 functions for this purpose). 104 105The subsystem-level idle callback is executed by the PM core whenever the device 106appears to be idle, which is indicated to the PM core by two counters, the 107device's usage counter and the counter of 'active' children of the device. 108 109 * 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 subsystem-level idle callback with the device as an argument. 113 114The action performed by a subsystem-level idle callback is totally dependent on 115the subsystem in question, but the expected and recommended action is to check 116if the device can be suspended (i.e. if all of the conditions necessary for 117suspending the device are satisfied) and to queue up a suspend request for the 118device in that case. The value returned by this callback is ignored by the PM 119core. 120 121The helper functions provided by the PM core, described in Section 4, guarantee 122that the following constraints are met with respect to the bus type's run-time 123PM callbacks: 124 125(1) The callbacks are mutually exclusive (e.g. it is forbidden to execute 126 ->runtime_suspend() in parallel with ->runtime_resume() or with another 127 instance of ->runtime_suspend() for the same device) with the exception that 128 ->runtime_suspend() or ->runtime_resume() can be executed in parallel with 129 ->runtime_idle() (although ->runtime_idle() will not be started while any 130 of the other callbacks is being executed for the same device). 131 132(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 run-time PM status of which is 135 'active'). 136 137(3) ->runtime_idle() and ->runtime_suspend() can only be executed for a device 138 the usage counter of which is equal to zero _and_ either the counter of 139 'active' children of which is equal to zero, or the 'power.ignore_children' 140 flag of which is set. 141 142(4) ->runtime_resume() can only be executed for 'suspended' devices (i.e. the 143 PM core will only execute ->runtime_resume() for the devices the run-time 144 PM status of which is 'suspended'). 145 146Additionally, the helper functions provided by the PM core obey the following 147rules: 148 149 * If ->runtime_suspend() is about to be executed or there's a pending request 150 to execute it, ->runtime_idle() will not be executed for the same device. 151 152 * A request to execute or to schedule the execution of ->runtime_suspend() 153 will cancel any pending requests to execute ->runtime_idle() for the same 154 device. 155 156 * If ->runtime_resume() is about to be executed or there's a pending request 157 to execute it, the other callbacks will not be executed for the same device. 158 159 * A request to execute ->runtime_resume() will cancel any pending or 160 scheduled requests to execute the other callbacks for the same device. 161 1623. Run-time PM Device Fields 163 164The following device run-time PM fields are present in 'struct dev_pm_info', as 165defined in include/linux/pm.h: 166 167 struct timer_list suspend_timer; 168 - timer used for scheduling (delayed) suspend request 169 170 unsigned long timer_expires; 171 - timer expiration time, in jiffies (if this is different from zero, the 172 timer is running and will expire at that time, otherwise the timer is not 173 running) 174 175 struct work_struct work; 176 - work structure used for queuing up requests (i.e. work items in pm_wq) 177 178 wait_queue_head_t wait_queue; 179 - wait queue used if any of the helper functions needs to wait for another 180 one to complete 181 182 spinlock_t lock; 183 - lock used for synchronisation 184 185 atomic_t usage_count; 186 - the usage counter of the device 187 188 atomic_t child_count; 189 - the count of 'active' children of the device 190 191 unsigned int ignore_children; 192 - if set, the value of child_count is ignored (but still updated) 193 194 unsigned int disable_depth; 195 - used for disabling the helper funcions (they work normally if this is 196 equal to zero); the initial value of it is 1 (i.e. run-time PM is 197 initially disabled for all devices) 198 199 unsigned int runtime_error; 200 - if set, there was a fatal error (one of the callbacks returned error code 201 as described in Section 2), so the helper funtions will not work until 202 this flag is cleared; this is the error code returned by the failing 203 callback 204 205 unsigned int idle_notification; 206 - if set, ->runtime_idle() is being executed 207 208 unsigned int request_pending; 209 - if set, there's a pending request (i.e. a work item queued up into pm_wq) 210 211 enum rpm_request request; 212 - type of request that's pending (valid if request_pending is set) 213 214 unsigned int deferred_resume; 215 - set if ->runtime_resume() is about to be run while ->runtime_suspend() is 216 being executed for that device and it is not practical to wait for the 217 suspend to complete; means "start a resume as soon as you've suspended" 218 219 unsigned int run_wake; 220 - set if the device is capable of generating run-time wake-up events 221 222 enum rpm_status runtime_status; 223 - the run-time PM status of the device; this field's initial value is 224 RPM_SUSPENDED, which means that each device is initially regarded by the 225 PM core as 'suspended', regardless of its real hardware status 226 227 unsigned int runtime_auto; 228 - if set, indicates that the user space has allowed the device driver to 229 power manage the device at run time via the /sys/devices/.../power/control 230 interface; it may only be modified with the help of the pm_runtime_allow() 231 and pm_runtime_forbid() helper functions 232 233All of the above fields are members of the 'power' member of 'struct device'. 234 2354. Run-time PM Device Helper Functions 236 237The following run-time PM helper functions are defined in 238drivers/base/power/runtime.c and include/linux/pm_runtime.h: 239 240 void pm_runtime_init(struct device *dev); 241 - initialize the device run-time PM fields in 'struct dev_pm_info' 242 243 void pm_runtime_remove(struct device *dev); 244 - make sure that the run-time PM of the device will be disabled after 245 removing the device from device hierarchy 246 247 int pm_runtime_idle(struct device *dev); 248 - execute the subsystem-level idle callback for the device; returns 0 on 249 success or error code on failure, where -EINPROGRESS means that 250 ->runtime_idle() is already being executed 251 252 int pm_runtime_suspend(struct device *dev); 253 - execute the subsystem-level suspend callback for the device; returns 0 on 254 success, 1 if the device's run-time PM status was already 'suspended', or 255 error code on failure, where -EAGAIN or -EBUSY means it is safe to attempt 256 to suspend the device again in future 257 258 int pm_runtime_resume(struct device *dev); 259 - execute the subsystem-level resume callback for the device; returns 0 on 260 success, 1 if the device's run-time PM status was already 'active' or 261 error code on failure, where -EAGAIN means it may be safe to attempt to 262 resume the device again in future, but 'power.runtime_error' should be 263 checked additionally 264 265 int pm_request_idle(struct device *dev); 266 - submit a request to execute the subsystem-level idle callback for the 267 device (the request is represented by a work item in pm_wq); returns 0 on 268 success or error code if the request has not been queued up 269 270 int pm_schedule_suspend(struct device *dev, unsigned int delay); 271 - schedule the execution of the subsystem-level suspend callback for the 272 device in future, where 'delay' is the time to wait before queuing up a 273 suspend work item in pm_wq, in milliseconds (if 'delay' is zero, the work 274 item is queued up immediately); returns 0 on success, 1 if the device's PM 275 run-time status was already 'suspended', or error code if the request 276 hasn't been scheduled (or queued up if 'delay' is 0); if the execution of 277 ->runtime_suspend() is already scheduled and not yet expired, the new 278 value of 'delay' will be used as the time to wait 279 280 int pm_request_resume(struct device *dev); 281 - submit a request to execute the subsystem-level resume callback for the 282 device (the request is represented by a work item in pm_wq); returns 0 on 283 success, 1 if the device's run-time PM status was already 'active', or 284 error code if the request hasn't been queued up 285 286 void pm_runtime_get_noresume(struct device *dev); 287 - increment the device's usage counter 288 289 int pm_runtime_get(struct device *dev); 290 - increment the device's usage counter, run pm_request_resume(dev) and 291 return its result 292 293 int pm_runtime_get_sync(struct device *dev); 294 - increment the device's usage counter, run pm_runtime_resume(dev) and 295 return its result 296 297 void pm_runtime_put_noidle(struct device *dev); 298 - decrement the device's usage counter 299 300 int pm_runtime_put(struct device *dev); 301 - decrement the device's usage counter, run pm_request_idle(dev) and return 302 its result 303 304 int pm_runtime_put_sync(struct device *dev); 305 - decrement the device's usage counter, run pm_runtime_idle(dev) and return 306 its result 307 308 void pm_runtime_enable(struct device *dev); 309 - enable the run-time PM helper functions to run the device bus type's 310 run-time PM callbacks described in Section 2 311 312 int pm_runtime_disable(struct device *dev); 313 - prevent the run-time PM helper functions from running subsystem-level 314 run-time PM callbacks for the device, make sure that all of the pending 315 run-time PM operations on the device are either completed or canceled; 316 returns 1 if there was a resume request pending and it was necessary to 317 execute the subsystem-level resume callback for the device to satisfy that 318 request, otherwise 0 is returned 319 320 void pm_suspend_ignore_children(struct device *dev, bool enable); 321 - set/unset the power.ignore_children flag of the device 322 323 int pm_runtime_set_active(struct device *dev); 324 - clear the device's 'power.runtime_error' flag, set the device's run-time 325 PM status to 'active' and update its parent's counter of 'active' 326 children as appropriate (it is only valid to use this function if 327 'power.runtime_error' is set or 'power.disable_depth' is greater than 328 zero); it will fail and return error code if the device has a parent 329 which is not active and the 'power.ignore_children' flag of which is unset 330 331 void pm_runtime_set_suspended(struct device *dev); 332 - clear the device's 'power.runtime_error' flag, set the device's run-time 333 PM status to 'suspended' and update its parent's counter of 'active' 334 children as appropriate (it is only valid to use this function if 335 'power.runtime_error' is set or 'power.disable_depth' is greater than 336 zero) 337 338 bool pm_runtime_suspended(struct device *dev); 339 - return true if the device's runtime PM status is 'suspended' and its 340 'power.disable_depth' field is equal to zero, or false otherwise 341 342 void pm_runtime_allow(struct device *dev); 343 - set the power.runtime_auto flag for the device and decrease its usage 344 counter (used by the /sys/devices/.../power/control interface to 345 effectively allow the device to be power managed at run time) 346 347 void pm_runtime_forbid(struct device *dev); 348 - unset the power.runtime_auto flag for the device and increase its usage 349 counter (used by the /sys/devices/.../power/control interface to 350 effectively prevent the device from being power managed at run time) 351 352It is safe to execute the following helper functions from interrupt context: 353 354pm_request_idle() 355pm_schedule_suspend() 356pm_request_resume() 357pm_runtime_get_noresume() 358pm_runtime_get() 359pm_runtime_put_noidle() 360pm_runtime_put() 361pm_suspend_ignore_children() 362pm_runtime_set_active() 363pm_runtime_set_suspended() 364pm_runtime_enable() 365 3665. Run-time PM Initialization, Device Probing and Removal 367 368Initially, the run-time PM is disabled for all devices, which means that the 369majority of the run-time PM helper funtions described in Section 4 will return 370-EAGAIN until pm_runtime_enable() is called for the device. 371 372In addition to that, the initial run-time PM status of all devices is 373'suspended', but it need not reflect the actual physical state of the device. 374Thus, if the device is initially active (i.e. it is able to process I/O), its 375run-time PM status must be changed to 'active', with the help of 376pm_runtime_set_active(), before pm_runtime_enable() is called for the device. 377 378However, if the device has a parent and the parent's run-time PM is enabled, 379calling pm_runtime_set_active() for the device will affect the parent, unless 380the parent's 'power.ignore_children' flag is set. Namely, in that case the 381parent won't be able to suspend at run time, using the PM core's helper 382functions, as long as the child's status is 'active', even if the child's 383run-time PM is still disabled (i.e. pm_runtime_enable() hasn't been called for 384the child yet or pm_runtime_disable() has been called for it). For this reason, 385once pm_runtime_set_active() has been called for the device, pm_runtime_enable() 386should be called for it too as soon as reasonably possible or its run-time PM 387status should be changed back to 'suspended' with the help of 388pm_runtime_set_suspended(). 389 390If the default initial run-time PM status of the device (i.e. 'suspended') 391reflects the actual state of the device, its bus type's or its driver's 392->probe() callback will likely need to wake it up using one of the PM core's 393helper functions described in Section 4. In that case, pm_runtime_resume() 394should be used. Of course, for this purpose the device's run-time PM has to be 395enabled earlier by calling pm_runtime_enable(). 396 397If the device bus type's or driver's ->probe() or ->remove() callback runs 398pm_runtime_suspend() or pm_runtime_idle() or their asynchronous counterparts, 399they will fail returning -EAGAIN, because the device's usage counter is 400incremented by the core before executing ->probe() and ->remove(). Still, it 401may be desirable to suspend the device as soon as ->probe() or ->remove() has 402finished, so the PM core uses pm_runtime_idle_sync() to invoke the 403subsystem-level idle callback for the device at that time. 404 405The user space can effectively disallow the driver of the device to power manage 406it at run time by changing the value of its /sys/devices/.../power/control 407attribute to "on", which causes pm_runtime_forbid() to be called. In principle, 408this mechanism may also be used by the driver to effectively turn off the 409run-time power management of the device until the user space turns it on. 410Namely, during the initialization the driver can make sure that the run-time PM 411status of the device is 'active' and call pm_runtime_forbid(). It should be 412noted, however, that if the user space has already intentionally changed the 413value of /sys/devices/.../power/control to "auto" to allow the driver to power 414manage the device at run time, the driver may confuse it by using 415pm_runtime_forbid() this way. 416 4176. Run-time PM and System Sleep 418 419Run-time PM and system sleep (i.e., system suspend and hibernation, also known 420as suspend-to-RAM and suspend-to-disk) interact with each other in a couple of 421ways. If a device is active when a system sleep starts, everything is 422straightforward. But what should happen if the device is already suspended? 423 424The device may have different wake-up settings for run-time PM and system sleep. 425For example, remote wake-up may be enabled for run-time suspend but disallowed 426for system sleep (device_may_wakeup(dev) returns 'false'). When this happens, 427the subsystem-level system suspend callback is responsible for changing the 428device's wake-up setting (it may leave that to the device driver's system 429suspend routine). It may be necessary to resume the device and suspend it again 430in order to do so. The same is true if the driver uses different power levels 431or other settings for run-time suspend and system sleep. 432 433During system resume, devices generally should be brought back to full power, 434even if they were suspended before the system sleep began. There are several 435reasons for this, including: 436 437 * The device might need to switch power levels, wake-up settings, etc. 438 439 * Remote wake-up events might have been lost by the firmware. 440 441 * The device's children may need the device to be at full power in order 442 to resume themselves. 443 444 * The driver's idea of the device state may not agree with the device's 445 physical state. This can happen during resume from hibernation. 446 447 * The device might need to be reset. 448 449 * Even though the device was suspended, if its usage counter was > 0 then most 450 likely it would need a run-time resume in the near future anyway. 451 452 * Always going back to full power is simplest. 453 454If the device was suspended before the sleep began, then its run-time PM status 455will have to be updated to reflect the actual post-system sleep status. The way 456to do this is: 457 458 pm_runtime_disable(dev); 459 pm_runtime_set_active(dev); 460 pm_runtime_enable(dev); 461 462The PM core always increments the run-time usage counter before calling the 463->prepare() callback and decrements it after calling the ->complete() callback. 464Hence disabling run-time PM temporarily like this will not cause any run-time 465suspend callbacks to be lost. 466 4677. Generic subsystem callbacks 468 469Subsystems may wish to conserve code space by using the set of generic power 470management callbacks provided by the PM core, defined in 471driver/base/power/generic_ops.c: 472 473 int pm_generic_runtime_idle(struct device *dev); 474 - invoke the ->runtime_idle() callback provided by the driver of this 475 device, if defined, and call pm_runtime_suspend() for this device if the 476 return value is 0 or the callback is not defined 477 478 int pm_generic_runtime_suspend(struct device *dev); 479 - invoke the ->runtime_suspend() callback provided by the driver of this 480 device and return its result, or return -EINVAL if not defined 481 482 int pm_generic_runtime_resume(struct device *dev); 483 - invoke the ->runtime_resume() callback provided by the driver of this 484 device and return its result, or return -EINVAL if not defined 485 486 int pm_generic_suspend(struct device *dev); 487 - if the device has not been suspended at run time, invoke the ->suspend() 488 callback provided by its driver and return its result, or return 0 if not 489 defined 490 491 int pm_generic_resume(struct device *dev); 492 - invoke the ->resume() callback provided by the driver of this device and, 493 if successful, change the device's runtime PM status to 'active' 494 495 int pm_generic_freeze(struct device *dev); 496 - if the device has not been suspended at run time, invoke the ->freeze() 497 callback provided by its driver and return its result, or return 0 if not 498 defined 499 500 int pm_generic_thaw(struct device *dev); 501 - if the device has not been suspended at run time, invoke the ->thaw() 502 callback provided by its driver and return its result, or return 0 if not 503 defined 504 505 int pm_generic_poweroff(struct device *dev); 506 - if the device has not been suspended at run time, invoke the ->poweroff() 507 callback provided by its driver and return its result, or return 0 if not 508 defined 509 510 int pm_generic_restore(struct device *dev); 511 - invoke the ->restore() callback provided by the driver of this device and, 512 if successful, change the device's runtime PM status to 'active' 513 514These functions can be assigned to the ->runtime_idle(), ->runtime_suspend(), 515->runtime_resume(), ->suspend(), ->resume(), ->freeze(), ->thaw(), ->poweroff(), 516or ->restore() callback pointers in the subsystem-level dev_pm_ops structures. 517 518If a subsystem wishes to use all of them at the same time, it can simply assign 519the GENERIC_SUBSYS_PM_OPS macro, defined in include/linux/pm.h, to its 520dev_pm_ops structure pointer. 521 522Device drivers that wish to use the same function as a system suspend, freeze, 523poweroff and run-time suspend callback, and similarly for system resume, thaw, 524restore, and run-time resume, can achieve this with the help of the 525UNIVERSAL_DEV_PM_OPS macro defined in include/linux/pm.h (possibly setting its 526last argument to NULL). 527