1Device Power Management 2 3Copyright (c) 2010-2011 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc. 4Copyright (c) 2010 Alan Stern <stern@rowland.harvard.edu> 5Copyright (c) 2014 Intel Corp., Rafael J. Wysocki <rafael.j.wysocki@intel.com> 6 7 8Most of the code in Linux is device drivers, so most of the Linux power 9management (PM) code is also driver-specific. Most drivers will do very 10little; others, especially for platforms with small batteries (like cell 11phones), will do a lot. 12 13This writeup gives an overview of how drivers interact with system-wide 14power management goals, emphasizing the models and interfaces that are 15shared by everything that hooks up to the driver model core. Read it as 16background for the domain-specific work you'd do with any specific driver. 17 18 19Two Models for Device Power Management 20====================================== 21Drivers will use one or both of these models to put devices into low-power 22states: 23 24 System Sleep model: 25 Drivers can enter low-power states as part of entering system-wide 26 low-power states like "suspend" (also known as "suspend-to-RAM"), or 27 (mostly for systems with disks) "hibernation" (also known as 28 "suspend-to-disk"). 29 30 This is something that device, bus, and class drivers collaborate on 31 by implementing various role-specific suspend and resume methods to 32 cleanly power down hardware and software subsystems, then reactivate 33 them without loss of data. 34 35 Some drivers can manage hardware wakeup events, which make the system 36 leave the low-power state. This feature may be enabled or disabled 37 using the relevant /sys/devices/.../power/wakeup file (for Ethernet 38 drivers the ioctl interface used by ethtool may also be used for this 39 purpose); enabling it may cost some power usage, but let the whole 40 system enter low-power states more often. 41 42 Runtime Power Management model: 43 Devices may also be put into low-power states while the system is 44 running, independently of other power management activity in principle. 45 However, devices are not generally independent of each other (for 46 example, a parent device cannot be suspended unless all of its child 47 devices have been suspended). Moreover, depending on the bus type the 48 device is on, it may be necessary to carry out some bus-specific 49 operations on the device for this purpose. Devices put into low power 50 states at run time may require special handling during system-wide power 51 transitions (suspend or hibernation). 52 53 For these reasons not only the device driver itself, but also the 54 appropriate subsystem (bus type, device type or device class) driver and 55 the PM core are involved in runtime power management. As in the system 56 sleep power management case, they need to collaborate by implementing 57 various role-specific suspend and resume methods, so that the hardware 58 is cleanly powered down and reactivated without data or service loss. 59 60There's not a lot to be said about those low-power states except that they are 61very system-specific, and often device-specific. Also, that if enough devices 62have been put into low-power states (at runtime), the effect may be very similar 63to entering some system-wide low-power state (system sleep) ... and that 64synergies exist, so that several drivers using runtime PM might put the system 65into a state where even deeper power saving options are available. 66 67Most suspended devices will have quiesced all I/O: no more DMA or IRQs (except 68for wakeup events), no more data read or written, and requests from upstream 69drivers are no longer accepted. A given bus or platform may have different 70requirements though. 71 72Examples of hardware wakeup events include an alarm from a real time clock, 73network wake-on-LAN packets, keyboard or mouse activity, and media insertion 74or removal (for PCMCIA, MMC/SD, USB, and so on). 75 76 77Interfaces for Entering System Sleep States 78=========================================== 79There are programming interfaces provided for subsystems (bus type, device type, 80device class) and device drivers to allow them to participate in the power 81management of devices they are concerned with. These interfaces cover both 82system sleep and runtime power management. 83 84 85Device Power Management Operations 86---------------------------------- 87Device power management operations, at the subsystem level as well as at the 88device driver level, are implemented by defining and populating objects of type 89struct dev_pm_ops: 90 91struct dev_pm_ops { 92 int (*prepare)(struct device *dev); 93 void (*complete)(struct device *dev); 94 int (*suspend)(struct device *dev); 95 int (*resume)(struct device *dev); 96 int (*freeze)(struct device *dev); 97 int (*thaw)(struct device *dev); 98 int (*poweroff)(struct device *dev); 99 int (*restore)(struct device *dev); 100 int (*suspend_late)(struct device *dev); 101 int (*resume_early)(struct device *dev); 102 int (*freeze_late)(struct device *dev); 103 int (*thaw_early)(struct device *dev); 104 int (*poweroff_late)(struct device *dev); 105 int (*restore_early)(struct device *dev); 106 int (*suspend_noirq)(struct device *dev); 107 int (*resume_noirq)(struct device *dev); 108 int (*freeze_noirq)(struct device *dev); 109 int (*thaw_noirq)(struct device *dev); 110 int (*poweroff_noirq)(struct device *dev); 111 int (*restore_noirq)(struct device *dev); 112 int (*runtime_suspend)(struct device *dev); 113 int (*runtime_resume)(struct device *dev); 114 int (*runtime_idle)(struct device *dev); 115}; 116 117This structure is defined in include/linux/pm.h and the methods included in it 118are also described in that file. Their roles will be explained in what follows. 119For now, it should be sufficient to remember that the last three methods are 120specific to runtime power management while the remaining ones are used during 121system-wide power transitions. 122 123There also is a deprecated "old" or "legacy" interface for power management 124operations available at least for some subsystems. This approach does not use 125struct dev_pm_ops objects and it is suitable only for implementing system sleep 126power management methods. Therefore it is not described in this document, so 127please refer directly to the source code for more information about it. 128 129 130Subsystem-Level Methods 131----------------------- 132The core methods to suspend and resume devices reside in struct dev_pm_ops 133pointed to by the ops member of struct dev_pm_domain, or by the pm member of 134struct bus_type, struct device_type and struct class. They are mostly of 135interest to the people writing infrastructure for platforms and buses, like PCI 136or USB, or device type and device class drivers. They also are relevant to the 137writers of device drivers whose subsystems (PM domains, device types, device 138classes and bus types) don't provide all power management methods. 139 140Bus drivers implement these methods as appropriate for the hardware and the 141drivers using it; PCI works differently from USB, and so on. Not many people 142write subsystem-level drivers; most driver code is a "device driver" that builds 143on top of bus-specific framework code. 144 145For more information on these driver calls, see the description later; 146they are called in phases for every device, respecting the parent-child 147sequencing in the driver model tree. 148 149 150/sys/devices/.../power/wakeup files 151----------------------------------- 152All device objects in the driver model contain fields that control the handling 153of system wakeup events (hardware signals that can force the system out of a 154sleep state). These fields are initialized by bus or device driver code using 155device_set_wakeup_capable() and device_set_wakeup_enable(), defined in 156include/linux/pm_wakeup.h. 157 158The "power.can_wakeup" flag just records whether the device (and its driver) can 159physically support wakeup events. The device_set_wakeup_capable() routine 160affects this flag. The "power.wakeup" field is a pointer to an object of type 161struct wakeup_source used for controlling whether or not the device should use 162its system wakeup mechanism and for notifying the PM core of system wakeup 163events signaled by the device. This object is only present for wakeup-capable 164devices (i.e. devices whose "can_wakeup" flags are set) and is created (or 165removed) by device_set_wakeup_capable(). 166 167Whether or not a device is capable of issuing wakeup events is a hardware 168matter, and the kernel is responsible for keeping track of it. By contrast, 169whether or not a wakeup-capable device should issue wakeup events is a policy 170decision, and it is managed by user space through a sysfs attribute: the 171"power/wakeup" file. User space can write the strings "enabled" or "disabled" 172to it to indicate whether or not, respectively, the device is supposed to signal 173system wakeup. This file is only present if the "power.wakeup" object exists 174for the given device and is created (or removed) along with that object, by 175device_set_wakeup_capable(). Reads from the file will return the corresponding 176string. 177 178The "power/wakeup" file is supposed to contain the "disabled" string initially 179for the majority of devices; the major exceptions are power buttons, keyboards, 180and Ethernet adapters whose WoL (wake-on-LAN) feature has been set up with 181ethtool. It should also default to "enabled" for devices that don't generate 182wakeup requests on their own but merely forward wakeup requests from one bus to 183another (like PCI Express ports). 184 185The device_may_wakeup() routine returns true only if the "power.wakeup" object 186exists and the corresponding "power/wakeup" file contains the string "enabled". 187This information is used by subsystems, like the PCI bus type code, to see 188whether or not to enable the devices' wakeup mechanisms. If device wakeup 189mechanisms are enabled or disabled directly by drivers, they also should use 190device_may_wakeup() to decide what to do during a system sleep transition. 191Device drivers, however, are not supposed to call device_set_wakeup_enable() 192directly in any case. 193 194It ought to be noted that system wakeup is conceptually different from "remote 195wakeup" used by runtime power management, although it may be supported by the 196same physical mechanism. Remote wakeup is a feature allowing devices in 197low-power states to trigger specific interrupts to signal conditions in which 198they should be put into the full-power state. Those interrupts may or may not 199be used to signal system wakeup events, depending on the hardware design. On 200some systems it is impossible to trigger them from system sleep states. In any 201case, remote wakeup should always be enabled for runtime power management for 202all devices and drivers that support it. 203 204/sys/devices/.../power/control files 205------------------------------------ 206Each device in the driver model has a flag to control whether it is subject to 207runtime power management. This flag, called runtime_auto, is initialized by the 208bus type (or generally subsystem) code using pm_runtime_allow() or 209pm_runtime_forbid(); the default is to allow runtime power management. 210 211The setting can be adjusted by user space by writing either "on" or "auto" to 212the device's power/control sysfs file. Writing "auto" calls pm_runtime_allow(), 213setting the flag and allowing the device to be runtime power-managed by its 214driver. Writing "on" calls pm_runtime_forbid(), clearing the flag, returning 215the device to full power if it was in a low-power state, and preventing the 216device from being runtime power-managed. User space can check the current value 217of the runtime_auto flag by reading the file. 218 219The device's runtime_auto flag has no effect on the handling of system-wide 220power transitions. In particular, the device can (and in the majority of cases 221should and will) be put into a low-power state during a system-wide transition 222to a sleep state even though its runtime_auto flag is clear. 223 224For more information about the runtime power management framework, refer to 225Documentation/power/runtime_pm.txt. 226 227 228Calling Drivers to Enter and Leave System Sleep States 229====================================================== 230When the system goes into a sleep state, each device's driver is asked to 231suspend the device by putting it into a state compatible with the target 232system state. That's usually some version of "off", but the details are 233system-specific. Also, wakeup-enabled devices will usually stay partly 234functional in order to wake the system. 235 236When the system leaves that low-power state, the device's driver is asked to 237resume it by returning it to full power. The suspend and resume operations 238always go together, and both are multi-phase operations. 239 240For simple drivers, suspend might quiesce the device using class code 241and then turn its hardware as "off" as possible during suspend_noirq. The 242matching resume calls would then completely reinitialize the hardware 243before reactivating its class I/O queues. 244 245More power-aware drivers might prepare the devices for triggering system wakeup 246events. 247 248 249Call Sequence Guarantees 250------------------------ 251To ensure that bridges and similar links needing to talk to a device are 252available when the device is suspended or resumed, the device tree is 253walked in a bottom-up order to suspend devices. A top-down order is 254used to resume those devices. 255 256The ordering of the device tree is defined by the order in which devices 257get registered: a child can never be registered, probed or resumed before 258its parent; and can't be removed or suspended after that parent. 259 260The policy is that the device tree should match hardware bus topology. 261(Or at least the control bus, for devices which use multiple busses.) 262In particular, this means that a device registration may fail if the parent of 263the device is suspending (i.e. has been chosen by the PM core as the next 264device to suspend) or has already suspended, as well as after all of the other 265devices have been suspended. Device drivers must be prepared to cope with such 266situations. 267 268 269System Power Management Phases 270------------------------------ 271Suspending or resuming the system is done in several phases. Different phases 272are used for freeze, standby, and memory sleep states ("suspend-to-RAM") and the 273hibernation state ("suspend-to-disk"). Each phase involves executing callbacks 274for every device before the next phase begins. Not all busses or classes 275support all these callbacks and not all drivers use all the callbacks. The 276various phases always run after tasks have been frozen and before they are 277unfrozen. Furthermore, the *_noirq phases run at a time when IRQ handlers have 278been disabled (except for those marked with the IRQF_NO_SUSPEND flag). 279 280All phases use PM domain, bus, type, class or driver callbacks (that is, methods 281defined in dev->pm_domain->ops, dev->bus->pm, dev->type->pm, dev->class->pm or 282dev->driver->pm). These callbacks are regarded by the PM core as mutually 283exclusive. Moreover, PM domain callbacks always take precedence over all of the 284other callbacks and, for example, type callbacks take precedence over bus, class 285and driver callbacks. To be precise, the following rules are used to determine 286which callback to execute in the given phase: 287 288 1. If dev->pm_domain is present, the PM core will choose the callback 289 included in dev->pm_domain->ops for execution 290 291 2. Otherwise, if both dev->type and dev->type->pm are present, the callback 292 included in dev->type->pm will be chosen for execution. 293 294 3. Otherwise, if both dev->class and dev->class->pm are present, the 295 callback included in dev->class->pm will be chosen for execution. 296 297 4. Otherwise, if both dev->bus and dev->bus->pm are present, the callback 298 included in dev->bus->pm will be chosen for execution. 299 300This allows PM domains and device types to override callbacks provided by bus 301types or device classes if necessary. 302 303The PM domain, type, class and bus callbacks may in turn invoke device- or 304driver-specific methods stored in dev->driver->pm, but they don't have to do 305that. 306 307If the subsystem callback chosen for execution is not present, the PM core will 308execute the corresponding method from dev->driver->pm instead if there is one. 309 310 311Entering System Suspend 312----------------------- 313When the system goes into the freeze, standby or memory sleep state, 314the phases are: 315 316 prepare, suspend, suspend_late, suspend_noirq. 317 318 1. The prepare phase is meant to prevent races by preventing new devices 319 from being registered; the PM core would never know that all the 320 children of a device had been suspended if new children could be 321 registered at will. (By contrast, devices may be unregistered at any 322 time.) Unlike the other suspend-related phases, during the prepare 323 phase the device tree is traversed top-down. 324 325 After the prepare callback method returns, no new children may be 326 registered below the device. The method may also prepare the device or 327 driver in some way for the upcoming system power transition, but it 328 should not put the device into a low-power state. 329 330 For devices supporting runtime power management, the return value of the 331 prepare callback can be used to indicate to the PM core that it may 332 safely leave the device in runtime suspend (if runtime-suspended 333 already), provided that all of the device's descendants are also left in 334 runtime suspend. Namely, if the prepare callback returns a positive 335 number and that happens for all of the descendants of the device too, 336 and all of them (including the device itself) are runtime-suspended, the 337 PM core will skip the suspend, suspend_late and suspend_noirq suspend 338 phases as well as the resume_noirq, resume_early and resume phases of 339 the following system resume for all of these devices. In that case, 340 the complete callback will be called directly after the prepare callback 341 and is entirely responsible for bringing the device back to the 342 functional state as appropriate. 343 344 Note that this direct-complete procedure applies even if the device is 345 disabled for runtime PM; only the runtime-PM status matters. It follows 346 that if a device has system-sleep callbacks but does not support runtime 347 PM, then its prepare callback must never return a positive value. This 348 is because all devices are initially set to runtime-suspended with 349 runtime PM disabled. 350 351 2. The suspend methods should quiesce the device to stop it from performing 352 I/O. They also may save the device registers and put it into the 353 appropriate low-power state, depending on the bus type the device is on, 354 and they may enable wakeup events. 355 356 3 For a number of devices it is convenient to split suspend into the 357 "quiesce device" and "save device state" phases, in which cases 358 suspend_late is meant to do the latter. It is always executed after 359 runtime power management has been disabled for all devices. 360 361 4. The suspend_noirq phase occurs after IRQ handlers have been disabled, 362 which means that the driver's interrupt handler will not be called while 363 the callback method is running. The methods should save the values of 364 the device's registers that weren't saved previously and finally put the 365 device into the appropriate low-power state. 366 367 The majority of subsystems and device drivers need not implement this 368 callback. However, bus types allowing devices to share interrupt 369 vectors, like PCI, generally need it; otherwise a driver might encounter 370 an error during the suspend phase by fielding a shared interrupt 371 generated by some other device after its own device had been set to low 372 power. 373 374At the end of these phases, drivers should have stopped all I/O transactions 375(DMA, IRQs), saved enough state that they can re-initialize or restore previous 376state (as needed by the hardware), and placed the device into a low-power state. 377On many platforms they will gate off one or more clock sources; sometimes they 378will also switch off power supplies or reduce voltages. (Drivers supporting 379runtime PM may already have performed some or all of these steps.) 380 381If device_may_wakeup(dev) returns true, the device should be prepared for 382generating hardware wakeup signals to trigger a system wakeup event when the 383system is in the sleep state. For example, enable_irq_wake() might identify 384GPIO signals hooked up to a switch or other external hardware, and 385pci_enable_wake() does something similar for the PCI PME signal. 386 387If any of these callbacks returns an error, the system won't enter the desired 388low-power state. Instead the PM core will unwind its actions by resuming all 389the devices that were suspended. 390 391 392Leaving System Suspend 393---------------------- 394When resuming from freeze, standby or memory sleep, the phases are: 395 396 resume_noirq, resume_early, resume, complete. 397 398 1. The resume_noirq callback methods should perform any actions needed 399 before the driver's interrupt handlers are invoked. This generally 400 means undoing the actions of the suspend_noirq phase. If the bus type 401 permits devices to share interrupt vectors, like PCI, the method should 402 bring the device and its driver into a state in which the driver can 403 recognize if the device is the source of incoming interrupts, if any, 404 and handle them correctly. 405 406 For example, the PCI bus type's ->pm.resume_noirq() puts the device into 407 the full-power state (D0 in the PCI terminology) and restores the 408 standard configuration registers of the device. Then it calls the 409 device driver's ->pm.resume_noirq() method to perform device-specific 410 actions. 411 412 2. The resume_early methods should prepare devices for the execution of 413 the resume methods. This generally involves undoing the actions of the 414 preceding suspend_late phase. 415 416 3 The resume methods should bring the device back to its operating 417 state, so that it can perform normal I/O. This generally involves 418 undoing the actions of the suspend phase. 419 420 4. The complete phase should undo the actions of the prepare phase. Note, 421 however, that new children may be registered below the device as soon as 422 the resume callbacks occur; it's not necessary to wait until the 423 complete phase. 424 425 Moreover, if the preceding prepare callback returned a positive number, 426 the device may have been left in runtime suspend throughout the whole 427 system suspend and resume (the suspend, suspend_late, suspend_noirq 428 phases of system suspend and the resume_noirq, resume_early, resume 429 phases of system resume may have been skipped for it). In that case, 430 the complete callback is entirely responsible for bringing the device 431 back to the functional state after system suspend if necessary. [For 432 example, it may need to queue up a runtime resume request for the device 433 for this purpose.] To check if that is the case, the complete callback 434 can consult the device's power.direct_complete flag. Namely, if that 435 flag is set when the complete callback is being run, it has been called 436 directly after the preceding prepare and special action may be required 437 to make the device work correctly afterward. 438 439At the end of these phases, drivers should be as functional as they were before 440suspending: I/O can be performed using DMA and IRQs, and the relevant clocks are 441gated on. 442 443However, the details here may again be platform-specific. For example, 444some systems support multiple "run" states, and the mode in effect at 445the end of resume might not be the one which preceded suspension. 446That means availability of certain clocks or power supplies changed, 447which could easily affect how a driver works. 448 449Drivers need to be able to handle hardware which has been reset since the 450suspend methods were called, for example by complete reinitialization. 451This may be the hardest part, and the one most protected by NDA'd documents 452and chip errata. It's simplest if the hardware state hasn't changed since 453the suspend was carried out, but that can't be guaranteed (in fact, it usually 454is not the case). 455 456Drivers must also be prepared to notice that the device has been removed 457while the system was powered down, whenever that's physically possible. 458PCMCIA, MMC, USB, Firewire, SCSI, and even IDE are common examples of busses 459where common Linux platforms will see such removal. Details of how drivers 460will notice and handle such removals are currently bus-specific, and often 461involve a separate thread. 462 463These callbacks may return an error value, but the PM core will ignore such 464errors since there's nothing it can do about them other than printing them in 465the system log. 466 467 468Entering Hibernation 469-------------------- 470Hibernating the system is more complicated than putting it into the other 471sleep states, because it involves creating and saving a system image. 472Therefore there are more phases for hibernation, with a different set of 473callbacks. These phases always run after tasks have been frozen and memory has 474been freed. 475 476The general procedure for hibernation is to quiesce all devices (freeze), create 477an image of the system memory while everything is stable, reactivate all 478devices (thaw), write the image to permanent storage, and finally shut down the 479system (poweroff). The phases used to accomplish this are: 480 481 prepare, freeze, freeze_late, freeze_noirq, thaw_noirq, thaw_early, 482 thaw, complete, prepare, poweroff, poweroff_late, poweroff_noirq 483 484 1. The prepare phase is discussed in the "Entering System Suspend" section 485 above. 486 487 2. The freeze methods should quiesce the device so that it doesn't generate 488 IRQs or DMA, and they may need to save the values of device registers. 489 However the device does not have to be put in a low-power state, and to 490 save time it's best not to do so. Also, the device should not be 491 prepared to generate wakeup events. 492 493 3. The freeze_late phase is analogous to the suspend_late phase described 494 above, except that the device should not be put in a low-power state and 495 should not be allowed to generate wakeup events by it. 496 497 4. The freeze_noirq phase is analogous to the suspend_noirq phase discussed 498 above, except again that the device should not be put in a low-power 499 state and should not be allowed to generate wakeup events. 500 501At this point the system image is created. All devices should be inactive and 502the contents of memory should remain undisturbed while this happens, so that the 503image forms an atomic snapshot of the system state. 504 505 5. The thaw_noirq phase is analogous to the resume_noirq phase discussed 506 above. The main difference is that its methods can assume the device is 507 in the same state as at the end of the freeze_noirq phase. 508 509 6. The thaw_early phase is analogous to the resume_early phase described 510 above. Its methods should undo the actions of the preceding 511 freeze_late, if necessary. 512 513 7. The thaw phase is analogous to the resume phase discussed above. Its 514 methods should bring the device back to an operating state, so that it 515 can be used for saving the image if necessary. 516 517 8. The complete phase is discussed in the "Leaving System Suspend" section 518 above. 519 520At this point the system image is saved, and the devices then need to be 521prepared for the upcoming system shutdown. This is much like suspending them 522before putting the system into the freeze, standby or memory sleep state, 523and the phases are similar. 524 525 9. The prepare phase is discussed above. 526 527 10. The poweroff phase is analogous to the suspend phase. 528 529 11. The poweroff_late phase is analogous to the suspend_late phase. 530 531 12. The poweroff_noirq phase is analogous to the suspend_noirq phase. 532 533The poweroff, poweroff_late and poweroff_noirq callbacks should do essentially 534the same things as the suspend, suspend_late and suspend_noirq callbacks, 535respectively. The only notable difference is that they need not store the 536device register values, because the registers should already have been stored 537during the freeze, freeze_late or freeze_noirq phases. 538 539 540Leaving Hibernation 541------------------- 542Resuming from hibernation is, again, more complicated than resuming from a sleep 543state in which the contents of main memory are preserved, because it requires 544a system image to be loaded into memory and the pre-hibernation memory contents 545to be restored before control can be passed back to the image kernel. 546 547Although in principle, the image might be loaded into memory and the 548pre-hibernation memory contents restored by the boot loader, in practice this 549can't be done because boot loaders aren't smart enough and there is no 550established protocol for passing the necessary information. So instead, the 551boot loader loads a fresh instance of the kernel, called the boot kernel, into 552memory and passes control to it in the usual way. Then the boot kernel reads 553the system image, restores the pre-hibernation memory contents, and passes 554control to the image kernel. Thus two different kernels are involved in 555resuming from hibernation. In fact, the boot kernel may be completely different 556from the image kernel: a different configuration and even a different version. 557This has important consequences for device drivers and their subsystems. 558 559To be able to load the system image into memory, the boot kernel needs to 560include at least a subset of device drivers allowing it to access the storage 561medium containing the image, although it doesn't need to include all of the 562drivers present in the image kernel. After the image has been loaded, the 563devices managed by the boot kernel need to be prepared for passing control back 564to the image kernel. This is very similar to the initial steps involved in 565creating a system image, and it is accomplished in the same way, using prepare, 566freeze, and freeze_noirq phases. However the devices affected by these phases 567are only those having drivers in the boot kernel; other devices will still be in 568whatever state the boot loader left them. 569 570Should the restoration of the pre-hibernation memory contents fail, the boot 571kernel would go through the "thawing" procedure described above, using the 572thaw_noirq, thaw, and complete phases, and then continue running normally. This 573happens only rarely. Most often the pre-hibernation memory contents are 574restored successfully and control is passed to the image kernel, which then 575becomes responsible for bringing the system back to the working state. 576 577To achieve this, the image kernel must restore the devices' pre-hibernation 578functionality. The operation is much like waking up from the memory sleep 579state, although it involves different phases: 580 581 restore_noirq, restore_early, restore, complete 582 583 1. The restore_noirq phase is analogous to the resume_noirq phase. 584 585 2. The restore_early phase is analogous to the resume_early phase. 586 587 3. The restore phase is analogous to the resume phase. 588 589 4. The complete phase is discussed above. 590 591The main difference from resume[_early|_noirq] is that restore[_early|_noirq] 592must assume the device has been accessed and reconfigured by the boot loader or 593the boot kernel. Consequently the state of the device may be different from the 594state remembered from the freeze, freeze_late and freeze_noirq phases. The 595device may even need to be reset and completely re-initialized. In many cases 596this difference doesn't matter, so the resume[_early|_noirq] and 597restore[_early|_norq] method pointers can be set to the same routines. 598Nevertheless, different callback pointers are used in case there is a situation 599where it actually does matter. 600 601 602Device Power Management Domains 603------------------------------- 604Sometimes devices share reference clocks or other power resources. In those 605cases it generally is not possible to put devices into low-power states 606individually. Instead, a set of devices sharing a power resource can be put 607into a low-power state together at the same time by turning off the shared 608power resource. Of course, they also need to be put into the full-power state 609together, by turning the shared power resource on. A set of devices with this 610property is often referred to as a power domain. 611 612Support for power domains is provided through the pm_domain field of struct 613device. This field is a pointer to an object of type struct dev_pm_domain, 614defined in include/linux/pm.h, providing a set of power management callbacks 615analogous to the subsystem-level and device driver callbacks that are executed 616for the given device during all power transitions, instead of the respective 617subsystem-level callbacks. Specifically, if a device's pm_domain pointer is 618not NULL, the ->suspend() callback from the object pointed to by it will be 619executed instead of its subsystem's (e.g. bus type's) ->suspend() callback and 620analogously for all of the remaining callbacks. In other words, power 621management domain callbacks, if defined for the given device, always take 622precedence over the callbacks provided by the device's subsystem (e.g. bus 623type). 624 625The support for device power management domains is only relevant to platforms 626needing to use the same device driver power management callbacks in many 627different power domain configurations and wanting to avoid incorporating the 628support for power domains into subsystem-level callbacks, for example by 629modifying the platform bus type. Other platforms need not implement it or take 630it into account in any way. 631 632 633Device Low Power (suspend) States 634--------------------------------- 635Device low-power states aren't standard. One device might only handle 636"on" and "off", while another might support a dozen different versions of 637"on" (how many engines are active?), plus a state that gets back to "on" 638faster than from a full "off". 639 640Some busses define rules about what different suspend states mean. PCI 641gives one example: after the suspend sequence completes, a non-legacy 642PCI device may not perform DMA or issue IRQs, and any wakeup events it 643issues would be issued through the PME# bus signal. Plus, there are 644several PCI-standard device states, some of which are optional. 645 646In contrast, integrated system-on-chip processors often use IRQs as the 647wakeup event sources (so drivers would call enable_irq_wake) and might 648be able to treat DMA completion as a wakeup event (sometimes DMA can stay 649active too, it'd only be the CPU and some peripherals that sleep). 650 651Some details here may be platform-specific. Systems may have devices that 652can be fully active in certain sleep states, such as an LCD display that's 653refreshed using DMA while most of the system is sleeping lightly ... and 654its frame buffer might even be updated by a DSP or other non-Linux CPU while 655the Linux control processor stays idle. 656 657Moreover, the specific actions taken may depend on the target system state. 658One target system state might allow a given device to be very operational; 659another might require a hard shut down with re-initialization on resume. 660And two different target systems might use the same device in different 661ways; the aforementioned LCD might be active in one product's "standby", 662but a different product using the same SOC might work differently. 663 664 665Power Management Notifiers 666-------------------------- 667There are some operations that cannot be carried out by the power management 668callbacks discussed above, because the callbacks occur too late or too early. 669To handle these cases, subsystems and device drivers may register power 670management notifiers that are called before tasks are frozen and after they have 671been thawed. Generally speaking, the PM notifiers are suitable for performing 672actions that either require user space to be available, or at least won't 673interfere with user space. 674 675For details refer to Documentation/power/notifiers.txt. 676 677 678Runtime Power Management 679======================== 680Many devices are able to dynamically power down while the system is still 681running. This feature is useful for devices that are not being used, and 682can offer significant power savings on a running system. These devices 683often support a range of runtime power states, which might use names such 684as "off", "sleep", "idle", "active", and so on. Those states will in some 685cases (like PCI) be partially constrained by the bus the device uses, and will 686usually include hardware states that are also used in system sleep states. 687 688A system-wide power transition can be started while some devices are in low 689power states due to runtime power management. The system sleep PM callbacks 690should recognize such situations and react to them appropriately, but the 691necessary actions are subsystem-specific. 692 693In some cases the decision may be made at the subsystem level while in other 694cases the device driver may be left to decide. In some cases it may be 695desirable to leave a suspended device in that state during a system-wide power 696transition, but in other cases the device must be put back into the full-power 697state temporarily, for example so that its system wakeup capability can be 698disabled. This all depends on the hardware and the design of the subsystem and 699device driver in question. 700 701During system-wide resume from a sleep state it's easiest to put devices into 702the full-power state, as explained in Documentation/power/runtime_pm.txt. Refer 703to that document for more information regarding this particular issue as well as 704for information on the device runtime power management framework in general. 705