Based on kernel version 2.6.26. Page generated on 2008-07-16 21:13 EST.
1 Most of the code in Linux is device drivers, so most of the Linux power 2 management code is also driver-specific. Most drivers will do very little; 3 others, especially for platforms with small batteries (like cell phones), 4 will do a lot. 5 6 This writeup gives an overview of how drivers interact with system-wide 7 power management goals, emphasizing the models and interfaces that are 8 shared by everything that hooks up to the driver model core. Read it as 9 background for the domain-specific work you'd do with any specific driver. 10 11 12 Two Models for Device Power Management 13 ====================================== 14 Drivers will use one or both of these models to put devices into low-power 15 states: 16 17 System Sleep model: 18 Drivers can enter low power states as part of entering system-wide 19 low-power states like "suspend-to-ram", or (mostly for systems with 20 disks) "hibernate" (suspend-to-disk). 21 22 This is something that device, bus, and class drivers collaborate on 23 by implementing various role-specific suspend and resume methods to 24 cleanly power down hardware and software subsystems, then reactivate 25 them without loss of data. 26 27 Some drivers can manage hardware wakeup events, which make the system 28 leave that low-power state. This feature may be disabled using the 29 relevant /sys/devices/.../power/wakeup file; enabling it may cost some 30 power usage, but let the whole system enter low power states more often. 31 32 Runtime Power Management model: 33 Drivers may also enter low power states while the system is running, 34 independently of other power management activity. Upstream drivers 35 will normally not know (or care) if the device is in some low power 36 state when issuing requests; the driver will auto-resume anything 37 that's needed when it gets a request. 38 39 This doesn't have, or need much infrastructure; it's just something you 40 should do when writing your drivers. For example, clk_disable() unused 41 clocks as part of minimizing power drain for currently-unused hardware. 42 Of course, sometimes clusters of drivers will collaborate with each 43 other, which could involve task-specific power management. 44 45 There's not a lot to be said about those low power states except that they 46 are very system-specific, and often device-specific. Also, that if enough 47 drivers put themselves into low power states (at "runtime"), the effect may be 48 the same as entering some system-wide low-power state (system sleep) ... and 49 that synergies exist, so that several drivers using runtime pm might put the 50 system into a state where even deeper power saving options are available. 51 52 Most suspended devices will have quiesced all I/O: no more DMA or irqs, no 53 more data read or written, and requests from upstream drivers are no longer 54 accepted. A given bus or platform may have different requirements though. 55 56 Examples of hardware wakeup events include an alarm from a real time clock, 57 network wake-on-LAN packets, keyboard or mouse activity, and media insertion 58 or removal (for PCMCIA, MMC/SD, USB, and so on). 59 60 61 Interfaces for Entering System Sleep States 62 =========================================== 63 Most of the programming interfaces a device driver needs to know about 64 relate to that first model: entering a system-wide low power state, 65 rather than just minimizing power consumption by one device. 66 67 68 Bus Driver Methods 69 ------------------ 70 The core methods to suspend and resume devices reside in struct bus_type. 71 These are mostly of interest to people writing infrastructure for busses 72 like PCI or USB, or because they define the primitives that device drivers 73 may need to apply in domain-specific ways to their devices: 74 75 struct bus_type { 76 ... 77 int (*suspend)(struct device *dev, pm_message_t state); 78 int (*suspend_late)(struct device *dev, pm_message_t state); 79 80 int (*resume_early)(struct device *dev); 81 int (*resume)(struct device *dev); 82 }; 83 84 Bus drivers implement those methods as appropriate for the hardware and 85 the drivers using it; PCI works differently from USB, and so on. Not many 86 people write bus drivers; most driver code is a "device driver" that 87 builds on top of bus-specific framework code. 88 89 For more information on these driver calls, see the description later; 90 they are called in phases for every device, respecting the parent-child 91 sequencing in the driver model tree. Note that as this is being written, 92 only the suspend() and resume() are widely available; not many bus drivers 93 leverage all of those phases, or pass them down to lower driver levels. 94 95 96 /sys/devices/.../power/wakeup files 97 ----------------------------------- 98 All devices in the driver model have two flags to control handling of 99 wakeup events, which are hardware signals that can force the device and/or 100 system out of a low power state. These are initialized by bus or device 101 driver code using device_init_wakeup(dev,can_wakeup). 102 103 The "can_wakeup" flag just records whether the device (and its driver) can 104 physically support wakeup events. When that flag is clear, the sysfs 105 "wakeup" file is empty, and device_may_wakeup() returns false. 106 107 For devices that can issue wakeup events, a separate flag controls whether 108 that device should try to use its wakeup mechanism. The initial value of 109 device_may_wakeup() will be true, so that the device's "wakeup" file holds 110 the value "enabled". Userspace can change that to "disabled" so that 111 device_may_wakeup() returns false; or change it back to "enabled" (so that 112 it returns true again). 113 114 115 EXAMPLE: PCI Device Driver Methods 116 ----------------------------------- 117 PCI framework software calls these methods when the PCI device driver bound 118 to a device device has provided them: 119 120 struct pci_driver { 121 ... 122 int (*suspend)(struct pci_device *pdev, pm_message_t state); 123 int (*suspend_late)(struct pci_device *pdev, pm_message_t state); 124 125 int (*resume_early)(struct pci_device *pdev); 126 int (*resume)(struct pci_device *pdev); 127 }; 128 129 Drivers will implement those methods, and call PCI-specific procedures 130 like pci_set_power_state(), pci_enable_wake(), pci_save_state(), and 131 pci_restore_state() to manage PCI-specific mechanisms. (PCI config space 132 could be saved during driver probe, if it weren't for the fact that some 133 systems rely on userspace tweaking using setpci.) Devices are suspended 134 before their bridges enter low power states, and likewise bridges resume 135 before their devices. 136 137 138 Upper Layers of Driver Stacks 139 ----------------------------- 140 Device drivers generally have at least two interfaces, and the methods 141 sketched above are the ones which apply to the lower level (nearer PCI, USB, 142 or other bus hardware). The network and block layers are examples of upper 143 level interfaces, as is a character device talking to userspace. 144 145 Power management requests normally need to flow through those upper levels, 146 which often use domain-oriented requests like "blank that screen". In 147 some cases those upper levels will have power management intelligence that 148 relates to end-user activity, or other devices that work in cooperation. 149 150 When those interfaces are structured using class interfaces, there is a 151 standard way to have the upper layer stop issuing requests to a given 152 class device (and restart later): 153 154 struct class { 155 ... 156 int (*suspend)(struct device *dev, pm_message_t state); 157 int (*resume)(struct device *dev); 158 }; 159 160 Those calls are issued in specific phases of the process by which the 161 system enters a low power "suspend" state, or resumes from it. 162 163 164 Calling Drivers to Enter System Sleep States 165 ============================================ 166 When the system enters a low power state, each device's driver is asked 167 to suspend the device by putting it into state compatible with the target 168 system state. That's usually some version of "off", but the details are 169 system-specific. Also, wakeup-enabled devices will usually stay partly 170 functional in order to wake the system. 171 172 When the system leaves that low power state, the device's driver is asked 173 to resume it. The suspend and resume operations always go together, and 174 both are multi-phase operations. 175 176 For simple drivers, suspend might quiesce the device using the class code 177 and then turn its hardware as "off" as possible with late_suspend. The 178 matching resume calls would then completely reinitialize the hardware 179 before reactivating its class I/O queues. 180 181 More power-aware drivers drivers will use more than one device low power 182 state, either at runtime or during system sleep states, and might trigger 183 system wakeup events. 184 185 186 Call Sequence Guarantees 187 ------------------------ 188 To ensure that bridges and similar links needed to talk to a device are 189 available when the device is suspended or resumed, the device tree is 190 walked in a bottom-up order to suspend devices. A top-down order is 191 used to resume those devices. 192 193 The ordering of the device tree is defined by the order in which devices 194 get registered: a child can never be registered, probed or resumed before 195 its parent; and can't be removed or suspended after that parent. 196 197 The policy is that the device tree should match hardware bus topology. 198 (Or at least the control bus, for devices which use multiple busses.) 199 200 201 Suspending Devices 202 ------------------ 203 Suspending a given device is done in several phases. Suspending the 204 system always includes every phase, executing calls for every device 205 before the next phase begins. Not all busses or classes support all 206 these callbacks; and not all drivers use all the callbacks. 207 208 The phases are seen by driver notifications issued in this order: 209 210 1 class.suspend(dev, message) is called after tasks are frozen, for 211 devices associated with a class that has such a method. This 212 method may sleep. 213 214 Since I/O activity usually comes from such higher layers, this is 215 a good place to quiesce all drivers of a given type (and keep such 216 code out of those drivers). 217 218 2 bus.suspend(dev, message) is called next. This method may sleep, 219 and is often morphed into a device driver call with bus-specific 220 parameters and/or rules. 221 222 This call should handle parts of device suspend logic that require 223 sleeping. It probably does work to quiesce the device which hasn't 224 been abstracted into class.suspend() or bus.suspend_late(). 225 226 3 bus.suspend_late(dev, message) is called with IRQs disabled, and 227 with only one CPU active. Until the bus.resume_early() phase 228 completes (see later), IRQs are not enabled again. This method 229 won't be exposed by all busses; for message based busses like USB, 230 I2C, or SPI, device interactions normally require IRQs. This bus 231 call may be morphed into a driver call with bus-specific parameters. 232 233 This call might save low level hardware state that might otherwise 234 be lost in the upcoming low power state, and actually put the 235 device into a low power state ... so that in some cases the device 236 may stay partly usable until this late. This "late" call may also 237 help when coping with hardware that behaves badly. 238 239 The pm_message_t parameter is currently used to refine those semantics 240 (described later). 241 242 At the end of those phases, drivers should normally have stopped all I/O 243 transactions (DMA, IRQs), saved enough state that they can re-initialize 244 or restore previous state (as needed by the hardware), and placed the 245 device into a low-power state. On many platforms they will also use 246 clk_disable() to gate off one or more clock sources; sometimes they will 247 also switch off power supplies, or reduce voltages. Drivers which have 248 runtime PM support may already have performed some or all of the steps 249 needed to prepare for the upcoming system sleep state. 250 251 When any driver sees that its device_can_wakeup(dev), it should make sure 252 to use the relevant hardware signals to trigger a system wakeup event. 253 For example, enable_irq_wake() might identify GPIO signals hooked up to 254 a switch or other external hardware, and pci_enable_wake() does something 255 similar for PCI's PME# signal. 256 257 If a driver (or bus, or class) fails it suspend method, the system won't 258 enter the desired low power state; it will resume all the devices it's 259 suspended so far. 260 261 Note that drivers may need to perform different actions based on the target 262 system lowpower/sleep state. At this writing, there are only platform 263 specific APIs through which drivers could determine those target states. 264 265 266 Device Low Power (suspend) States 267 --------------------------------- 268 Device low-power states aren't very standard. One device might only handle 269 "on" and "off, while another might support a dozen different versions of 270 "on" (how many engines are active?), plus a state that gets back to "on" 271 faster than from a full "off". 272 273 Some busses define rules about what different suspend states mean. PCI 274 gives one example: after the suspend sequence completes, a non-legacy 275 PCI device may not perform DMA or issue IRQs, and any wakeup events it 276 issues would be issued through the PME# bus signal. Plus, there are 277 several PCI-standard device states, some of which are optional. 278 279 In contrast, integrated system-on-chip processors often use irqs as the 280 wakeup event sources (so drivers would call enable_irq_wake) and might 281 be able to treat DMA completion as a wakeup event (sometimes DMA can stay 282 active too, it'd only be the CPU and some peripherals that sleep). 283 284 Some details here may be platform-specific. Systems may have devices that 285 can be fully active in certain sleep states, such as an LCD display that's 286 refreshed using DMA while most of the system is sleeping lightly ... and 287 its frame buffer might even be updated by a DSP or other non-Linux CPU while 288 the Linux control processor stays idle. 289 290 Moreover, the specific actions taken may depend on the target system state. 291 One target system state might allow a given device to be very operational; 292 another might require a hard shut down with re-initialization on resume. 293 And two different target systems might use the same device in different 294 ways; the aforementioned LCD might be active in one product's "standby", 295 but a different product using the same SOC might work differently. 296 297 298 Meaning of pm_message_t.event 299 ----------------------------- 300 Parameters to suspend calls include the device affected and a message of 301 type pm_message_t, which has one field: the event. If driver does not 302 recognize the event code, suspend calls may abort the request and return 303 a negative errno. However, most drivers will be fine if they implement 304 PM_EVENT_SUSPEND semantics for all messages. 305 306 The event codes are used to refine the goal of suspending the device, and 307 mostly matter when creating or resuming system memory image snapshots, as 308 used with suspend-to-disk: 309 310 PM_EVENT_SUSPEND -- quiesce the driver and put hardware into a low-power 311 state. When used with system sleep states like "suspend-to-RAM" or 312 "standby", the upcoming resume() call will often be able to rely on 313 state kept in hardware, or issue system wakeup events. 314 315 PM_EVENT_HIBERNATE -- Put hardware into a low-power state and enable wakeup 316 events as appropriate. It is only used with hibernation 317 (suspend-to-disk) and few devices are able to wake up the system from 318 this state; most are completely powered off. 319 320 PM_EVENT_FREEZE -- quiesce the driver, but don't necessarily change into 321 any low power mode. A system snapshot is about to be taken, often 322 followed by a call to the driver's resume() method. Neither wakeup 323 events nor DMA are allowed. 324 325 PM_EVENT_PRETHAW -- quiesce the driver, knowing that the upcoming resume() 326 will restore a suspend-to-disk snapshot from a different kernel image. 327 Drivers that are smart enough to look at their hardware state during 328 resume() processing need that state to be correct ... a PRETHAW could 329 be used to invalidate that state (by resetting the device), like a 330 shutdown() invocation would before a kexec() or system halt. Other 331 drivers might handle this the same way as PM_EVENT_FREEZE. Neither 332 wakeup events nor DMA are allowed. 333 334 To enter "standby" (ACPI S1) or "Suspend to RAM" (STR, ACPI S3) states, or 335 the similarly named APM states, only PM_EVENT_SUSPEND is used; the other event 336 codes are used for hibernation ("Suspend to Disk", STD, ACPI S4). 337 338 There's also PM_EVENT_ON, a value which never appears as a suspend event 339 but is sometimes used to record the "not suspended" device state. 340 341 342 Resuming Devices 343 ---------------- 344 Resuming is done in multiple phases, much like suspending, with all 345 devices processing each phase's calls before the next phase begins. 346 347 The phases are seen by driver notifications issued in this order: 348 349 1 bus.resume_early(dev) is called with IRQs disabled, and with 350 only one CPU active. As with bus.suspend_late(), this method 351 won't be supported on busses that require IRQs in order to 352 interact with devices. 353 354 This reverses the effects of bus.suspend_late(). 355 356 2 bus.resume(dev) is called next. This may be morphed into a device 357 driver call with bus-specific parameters; implementations may sleep. 358 359 This reverses the effects of bus.suspend(). 360 361 3 class.resume(dev) is called for devices associated with a class 362 that has such a method. Implementations may sleep. 363 364 This reverses the effects of class.suspend(), and would usually 365 reactivate the device's I/O queue. 366 367 At the end of those phases, drivers should normally be as functional as 368 they were before suspending: I/O can be performed using DMA and IRQs, and 369 the relevant clocks are gated on. The device need not be "fully on"; it 370 might be in a runtime lowpower/suspend state that acts as if it were. 371 372 However, the details here may again be platform-specific. For example, 373 some systems support multiple "run" states, and the mode in effect at 374 the end of resume() might not be the one which preceded suspension. 375 That means availability of certain clocks or power supplies changed, 376 which could easily affect how a driver works. 377 378 379 Drivers need to be able to handle hardware which has been reset since the 380 suspend methods were called, for example by complete reinitialization. 381 This may be the hardest part, and the one most protected by NDA'd documents 382 and chip errata. It's simplest if the hardware state hasn't changed since 383 the suspend() was called, but that can't always be guaranteed. 384 385 Drivers must also be prepared to notice that the device has been removed 386 while the system was powered off, whenever that's physically possible. 387 PCMCIA, MMC, USB, Firewire, SCSI, and even IDE are common examples of busses 388 where common Linux platforms will see such removal. Details of how drivers 389 will notice and handle such removals are currently bus-specific, and often 390 involve a separate thread. 391 392 393 Note that the bus-specific runtime PM wakeup mechanism can exist, and might 394 be defined to share some of the same driver code as for system wakeup. For 395 example, a bus-specific device driver's resume() method might be used there, 396 so it wouldn't only be called from bus.resume() during system-wide wakeup. 397 See bus-specific information about how runtime wakeup events are handled. 398 399 400 System Devices 401 -------------- 402 System devices follow a slightly different API, which can be found in 403 404 include/linux/sysdev.h 405 drivers/base/sys.c 406 407 System devices will only be suspended with interrupts disabled, and after 408 all other devices have been suspended. On resume, they will be resumed 409 before any other devices, and also with interrupts disabled. 410 411 That is, IRQs are disabled, the suspend_late() phase begins, then the 412 sysdev_driver.suspend() phase, and the system enters a sleep state. Then 413 the sysdev_driver.resume() phase begins, followed by the resume_early() 414 phase, after which IRQs are enabled. 415 416 Code to actually enter and exit the system-wide low power state sometimes 417 involves hardware details that are only known to the boot firmware, and 418 may leave a CPU running software (from SRAM or flash memory) that monitors 419 the system and manages its wakeup sequence. 420 421 422 Runtime Power Management 423 ======================== 424 Many devices are able to dynamically power down while the system is still 425 running. This feature is useful for devices that are not being used, and 426 can offer significant power savings on a running system. These devices 427 often support a range of runtime power states, which might use names such 428 as "off", "sleep", "idle", "active", and so on. Those states will in some 429 cases (like PCI) be partially constrained by a bus the device uses, and will 430 usually include hardware states that are also used in system sleep states. 431 432 However, note that if a driver puts a device into a runtime low power state 433 and the system then goes into a system-wide sleep state, it normally ought 434 to resume into that runtime low power state rather than "full on". Such 435 distinctions would be part of the driver-internal state machine for that 436 hardware; the whole point of runtime power management is to be sure that 437 drivers are decoupled in that way from the state machine governing phases 438 of the system-wide power/sleep state transitions. 439 440 441 Power Saving Techniques 442 ----------------------- 443 Normally runtime power management is handled by the drivers without specific 444 userspace or kernel intervention, by device-aware use of techniques like: 445 446 Using information provided by other system layers 447 - stay deeply "off" except between open() and close() 448 - if transceiver/PHY indicates "nobody connected", stay "off" 449 - application protocols may include power commands or hints 450 451 Using fewer CPU cycles 452 - using DMA instead of PIO 453 - removing timers, or making them lower frequency 454 - shortening "hot" code paths 455 - eliminating cache misses 456 - (sometimes) offloading work to device firmware 457 458 Reducing other resource costs 459 - gating off unused clocks in software (or hardware) 460 - switching off unused power supplies 461 - eliminating (or delaying/merging) IRQs 462 - tuning DMA to use word and/or burst modes 463 464 Using device-specific low power states 465 - using lower voltages 466 - avoiding needless DMA transfers 467 468 Read your hardware documentation carefully to see the opportunities that 469 may be available. If you can, measure the actual power usage and check 470 it against the budget established for your project. 471 472 473 Examples: USB hosts, system timer, system CPU 474 ---------------------------------------------- 475 USB host controllers make interesting, if complex, examples. In many cases 476 these have no work to do: no USB devices are connected, or all of them are 477 in the USB "suspend" state. Linux host controller drivers can then disable 478 periodic DMA transfers that would otherwise be a constant power drain on the 479 memory subsystem, and enter a suspend state. In power-aware controllers, 480 entering that suspend state may disable the clock used with USB signaling, 481 saving a certain amount of power. 482 483 The controller will be woken from that state (with an IRQ) by changes to the 484 signal state on the data lines of a given port, for example by an existing 485 peripheral requesting "remote wakeup" or by plugging a new peripheral. The 486 same wakeup mechanism usually works from "standby" sleep states, and on some 487 systems also from "suspend to RAM" (or even "suspend to disk") states. 488 (Except that ACPI may be involved instead of normal IRQs, on some hardware.) 489 490 System devices like timers and CPUs may have special roles in the platform 491 power management scheme. For example, system timers using a "dynamic tick" 492 approach don't just save CPU cycles (by eliminating needless timer IRQs), 493 but they may also open the door to using lower power CPU "idle" states that 494 cost more than a jiffie to enter and exit. On x86 systems these are states 495 like "C3"; note that periodic DMA transfers from a USB host controller will 496 also prevent entry to a C3 state, much like a periodic timer IRQ. 497 498 That kind of runtime mechanism interaction is common. "System On Chip" (SOC) 499 processors often have low power idle modes that can't be entered unless 500 certain medium-speed clocks (often 12 or 48 MHz) are gated off. When the 501 drivers gate those clocks effectively, then the system idle task may be able 502 to use the lower power idle modes and thereby increase battery life. 503 504 If the CPU can have a "cpufreq" driver, there also may be opportunities 505 to shift to lower voltage settings and reduce the power cost of executing 506 a given number of instructions. (Without voltage adjustment, it's rare 507 for cpufreq to save much power; the cost-per-instruction must go down.)