1 Power Management for USB 2 3 Alan Stern <stern@rowland.harvard.edu> 4 5 October 5, 2007 6 7 8 9 What is Power Management? 10 ------------------------- 11 12Power Management (PM) is the practice of saving energy by suspending 13parts of a computer system when they aren't being used. While a 14component is "suspended" it is in a nonfunctional low-power state; it 15might even be turned off completely. A suspended component can be 16"resumed" (returned to a functional full-power state) when the kernel 17needs to use it. (There also are forms of PM in which components are 18placed in a less functional but still usable state instead of being 19suspended; an example would be reducing the CPU's clock rate. This 20document will not discuss those other forms.) 21 22When the parts being suspended include the CPU and most of the rest of 23the system, we speak of it as a "system suspend". When a particular 24device is turned off while the system as a whole remains running, we 25call it a "dynamic suspend" (also known as a "runtime suspend" or 26"selective suspend"). This document concentrates mostly on how 27dynamic PM is implemented in the USB subsystem, although system PM is 28covered to some extent (see Documentation/power/*.txt for more 29information about system PM). 30 31Note: Dynamic PM support for USB is present only if the kernel was 32built with CONFIG_USB_SUSPEND enabled. System PM support is present 33only if the kernel was built with CONFIG_SUSPEND or CONFIG_HIBERNATION 34enabled. 35 36 37 What is Remote Wakeup? 38 ---------------------- 39 40When a device has been suspended, it generally doesn't resume until 41the computer tells it to. Likewise, if the entire computer has been 42suspended, it generally doesn't resume until the user tells it to, say 43by pressing a power button or opening the cover. 44 45However some devices have the capability of resuming by themselves, or 46asking the kernel to resume them, or even telling the entire computer 47to resume. This capability goes by several names such as "Wake On 48LAN"; we will refer to it generically as "remote wakeup". When a 49device is enabled for remote wakeup and it is suspended, it may resume 50itself (or send a request to be resumed) in response to some external 51event. Examples include a suspended keyboard resuming when a key is 52pressed, or a suspended USB hub resuming when a device is plugged in. 53 54 55 When is a USB device idle? 56 -------------------------- 57 58A device is idle whenever the kernel thinks it's not busy doing 59anything important and thus is a candidate for being suspended. The 60exact definition depends on the device's driver; drivers are allowed 61to declare that a device isn't idle even when there's no actual 62communication taking place. (For example, a hub isn't considered idle 63unless all the devices plugged into that hub are already suspended.) 64In addition, a device isn't considered idle so long as a program keeps 65its usbfs file open, whether or not any I/O is going on. 66 67If a USB device has no driver, its usbfs file isn't open, and it isn't 68being accessed through sysfs, then it definitely is idle. 69 70 71 Forms of dynamic PM 72 ------------------- 73 74Dynamic suspends can occur in two ways: manual and automatic. 75"Manual" means that the user has told the kernel to suspend a device, 76whereas "automatic" means that the kernel has decided all by itself to 77suspend a device. Automatic suspend is called "autosuspend" for 78short. In general, a device won't be autosuspended unless it has been 79idle for some minimum period of time, the so-called idle-delay time. 80 81Of course, nothing the kernel does on its own initiative should 82prevent the computer or its devices from working properly. If a 83device has been autosuspended and a program tries to use it, the 84kernel will automatically resume the device (autoresume). For the 85same reason, an autosuspended device will usually have remote wakeup 86enabled, if the device supports remote wakeup. 87 88It is worth mentioning that many USB drivers don't support 89autosuspend. In fact, at the time of this writing (Linux 2.6.23) the 90only drivers which do support it are the hub driver, kaweth, asix, 91usblp, usblcd, and usb-skeleton (which doesn't count). If a 92non-supporting driver is bound to a device, the device won't be 93autosuspended. In effect, the kernel pretends the device is never 94idle. 95 96We can categorize power management events in two broad classes: 97external and internal. External events are those triggered by some 98agent outside the USB stack: system suspend/resume (triggered by 99userspace), manual dynamic suspend/resume (also triggered by 100userspace), and remote wakeup (triggered by the device). Internal 101events are those triggered within the USB stack: autosuspend and 102autoresume. 103 104 105 The user interface for dynamic PM 106 --------------------------------- 107 108The user interface for controlling dynamic PM is located in the power/ 109subdirectory of each USB device's sysfs directory, that is, in 110/sys/bus/usb/devices/.../power/ where "..." is the device's ID. The 111relevant attribute files are: wakeup, level, and autosuspend. 112 113 power/wakeup 114 115 This file is empty if the device does not support 116 remote wakeup. Otherwise the file contains either the 117 word "enabled" or the word "disabled", and you can 118 write those words to the file. The setting determines 119 whether or not remote wakeup will be enabled when the 120 device is next suspended. (If the setting is changed 121 while the device is suspended, the change won't take 122 effect until the following suspend.) 123 124 power/level 125 126 This file contains one of three words: "on", "auto", 127 or "suspend". You can write those words to the file 128 to change the device's setting. 129 130 "on" means that the device should be resumed and 131 autosuspend is not allowed. (Of course, system 132 suspends are still allowed.) 133 134 "auto" is the normal state in which the kernel is 135 allowed to autosuspend and autoresume the device. 136 137 "suspend" means that the device should remain 138 suspended, and autoresume is not allowed. (But remote 139 wakeup may still be allowed, since it is controlled 140 separately by the power/wakeup attribute.) 141 142 power/autosuspend 143 144 This file contains an integer value, which is the 145 number of seconds the device should remain idle before 146 the kernel will autosuspend it (the idle-delay time). 147 The default is 2. 0 means to autosuspend as soon as 148 the device becomes idle, and -1 means never to 149 autosuspend. You can write a number to the file to 150 change the autosuspend idle-delay time. 151 152Writing "-1" to power/autosuspend and writing "on" to power/level do 153essentially the same thing -- they both prevent the device from being 154autosuspended. Yes, this is a redundancy in the API. 155 156(In 2.6.21 writing "0" to power/autosuspend would prevent the device 157from being autosuspended; the behavior was changed in 2.6.22. The 158power/autosuspend attribute did not exist prior to 2.6.21, and the 159power/level attribute did not exist prior to 2.6.22.) 160 161 162 Changing the default idle-delay time 163 ------------------------------------ 164 165The default autosuspend idle-delay time is controlled by a module 166parameter in usbcore. You can specify the value when usbcore is 167loaded. For example, to set it to 5 seconds instead of 2 you would 168do: 169 170 modprobe usbcore autosuspend=5 171 172Equivalently, you could add to /etc/modprobe.conf a line saying: 173 174 options usbcore autosuspend=5 175 176Some distributions load the usbcore module very early during the boot 177process, by means of a program or script running from an initramfs 178image. To alter the parameter value you would have to rebuild that 179image. 180 181If usbcore is compiled into the kernel rather than built as a loadable 182module, you can add 183 184 usbcore.autosuspend=5 185 186to the kernel's boot command line. 187 188Finally, the parameter value can be changed while the system is 189running. If you do: 190 191 echo 5 >/sys/module/usbcore/parameters/autosuspend 192 193then each new USB device will have its autosuspend idle-delay 194initialized to 5. (The idle-delay values for already existing devices 195will not be affected.) 196 197Setting the initial default idle-delay to -1 will prevent any 198autosuspend of any USB device. This is a simple alternative to 199disabling CONFIG_USB_SUSPEND and rebuilding the kernel, and it has the 200added benefit of allowing you to enable autosuspend for selected 201devices. 202 203 204 Warnings 205 -------- 206 207The USB specification states that all USB devices must support power 208management. Nevertheless, the sad fact is that many devices do not 209support it very well. You can suspend them all right, but when you 210try to resume them they disconnect themselves from the USB bus or 211they stop working entirely. This seems to be especially prevalent 212among printers and scanners, but plenty of other types of device have 213the same deficiency. 214 215For this reason, by default the kernel disables autosuspend (the 216power/level attribute is initialized to "on") for all devices other 217than hubs. Hubs, at least, appear to be reasonably well-behaved in 218this regard. 219 220(In 2.6.21 and 2.6.22 this wasn't the case. Autosuspend was enabled 221by default for almost all USB devices. A number of people experienced 222problems as a result.) 223 224This means that non-hub devices won't be autosuspended unless the user 225or a program explicitly enables it. As of this writing there aren't 226any widespread programs which will do this; we hope that in the near 227future device managers such as HAL will take on this added 228responsibility. In the meantime you can always carry out the 229necessary operations by hand or add them to a udev script. You can 230also change the idle-delay time; 2 seconds is not the best choice for 231every device. 232 233Sometimes it turns out that even when a device does work okay with 234autosuspend there are still problems. For example, there are 235experimental patches adding autosuspend support to the usbhid driver, 236which manages keyboards and mice, among other things. Tests with a 237number of keyboards showed that typing on a suspended keyboard, while 238causing the keyboard to do a remote wakeup all right, would 239nonetheless frequently result in lost keystrokes. Tests with mice 240showed that some of them would issue a remote-wakeup request in 241response to button presses but not to motion, and some in response to 242neither. 243 244The kernel will not prevent you from enabling autosuspend on devices 245that can't handle it. It is even possible in theory to damage a 246device by suspending it at the wrong time -- for example, suspending a 247USB hard disk might cause it to spin down without parking the heads. 248(Highly unlikely, but possible.) Take care. 249 250 251 The driver interface for Power Management 252 ----------------------------------------- 253 254The requirements for a USB driver to support external power management 255are pretty modest; the driver need only define 256 257 .suspend 258 .resume 259 .reset_resume 260 261methods in its usb_driver structure, and the reset_resume method is 262optional. The methods' jobs are quite simple: 263 264 The suspend method is called to warn the driver that the 265 device is going to be suspended. If the driver returns a 266 negative error code, the suspend will be aborted. Normally 267 the driver will return 0, in which case it must cancel all 268 outstanding URBs (usb_kill_urb()) and not submit any more. 269 270 The resume method is called to tell the driver that the 271 device has been resumed and the driver can return to normal 272 operation. URBs may once more be submitted. 273 274 The reset_resume method is called to tell the driver that 275 the device has been resumed and it also has been reset. 276 The driver should redo any necessary device initialization, 277 since the device has probably lost most or all of its state 278 (although the interfaces will be in the same altsettings as 279 before the suspend). 280 281If the device is disconnected or powered down while it is suspended, 282the disconnect method will be called instead of the resume or 283reset_resume method. This is also quite likely to happen when 284waking up from hibernation, as many systems do not maintain suspend 285current to the USB host controllers during hibernation. (It's 286possible to work around the hibernation-forces-disconnect problem by 287using the USB Persist facility.) 288 289The reset_resume method is used by the USB Persist facility (see 290Documentation/usb/persist.txt) and it can also be used under certain 291circumstances when CONFIG_USB_PERSIST is not enabled. Currently, if a 292device is reset during a resume and the driver does not have a 293reset_resume method, the driver won't receive any notification about 294the resume. Later kernels will call the driver's disconnect method; 2952.6.23 doesn't do this. 296 297USB drivers are bound to interfaces, so their suspend and resume 298methods get called when the interfaces are suspended or resumed. In 299principle one might want to suspend some interfaces on a device (i.e., 300force the drivers for those interface to stop all activity) without 301suspending the other interfaces. The USB core doesn't allow this; all 302interfaces are suspended when the device itself is suspended and all 303interfaces are resumed when the device is resumed. It isn't possible 304to suspend or resume some but not all of a device's interfaces. The 305closest you can come is to unbind the interfaces' drivers. 306 307 308 The driver interface for autosuspend and autoresume 309 --------------------------------------------------- 310 311To support autosuspend and autoresume, a driver should implement all 312three of the methods listed above. In addition, a driver indicates 313that it supports autosuspend by setting the .supports_autosuspend flag 314in its usb_driver structure. It is then responsible for informing the 315USB core whenever one of its interfaces becomes busy or idle. The 316driver does so by calling these five functions: 317 318 int usb_autopm_get_interface(struct usb_interface *intf); 319 void usb_autopm_put_interface(struct usb_interface *intf); 320 int usb_autopm_set_interface(struct usb_interface *intf); 321 int usb_autopm_get_interface_async(struct usb_interface *intf); 322 void usb_autopm_put_interface_async(struct usb_interface *intf); 323 324The functions work by maintaining a counter in the usb_interface 325structure. When intf->pm_usage_count is > 0 then the interface is 326deemed to be busy, and the kernel will not autosuspend the interface's 327device. When intf->pm_usage_count is <= 0 then the interface is 328considered to be idle, and the kernel may autosuspend the device. 329 330(There is a similar pm_usage_count field in struct usb_device, 331associated with the device itself rather than any of its interfaces. 332This field is used only by the USB core.) 333 334The driver owns intf->pm_usage_count; it can modify the value however 335and whenever it likes. A nice aspect of the non-async usb_autopm_* 336routines is that the changes they make are protected by the usb_device 337structure's PM mutex (udev->pm_mutex); however drivers may change 338pm_usage_count without holding the mutex. Drivers using the async 339routines are responsible for their own synchronization and mutual 340exclusion. 341 342 usb_autopm_get_interface() increments pm_usage_count and 343 attempts an autoresume if the new value is > 0 and the 344 device is suspended. 345 346 usb_autopm_put_interface() decrements pm_usage_count and 347 attempts an autosuspend if the new value is <= 0 and the 348 device isn't suspended. 349 350 usb_autopm_set_interface() leaves pm_usage_count alone. 351 It attempts an autoresume if the value is > 0 and the device 352 is suspended, and it attempts an autosuspend if the value is 353 <= 0 and the device isn't suspended. 354 355 usb_autopm_get_interface_async() and 356 usb_autopm_put_interface_async() do almost the same things as 357 their non-async counterparts. The differences are: they do 358 not acquire the PM mutex, and they use a workqueue to do their 359 jobs. As a result they can be called in an atomic context, 360 such as an URB's completion handler, but when they return the 361 device will not generally not yet be in the desired state. 362 363There also are a couple of utility routines drivers can use: 364 365 usb_autopm_enable() sets pm_usage_cnt to 0 and then calls 366 usb_autopm_set_interface(), which will attempt an autosuspend. 367 368 usb_autopm_disable() sets pm_usage_cnt to 1 and then calls 369 usb_autopm_set_interface(), which will attempt an autoresume. 370 371The conventional usage pattern is that a driver calls 372usb_autopm_get_interface() in its open routine and 373usb_autopm_put_interface() in its close or release routine. But 374other patterns are possible. 375 376The autosuspend attempts mentioned above will often fail for one 377reason or another. For example, the power/level attribute might be 378set to "on", or another interface in the same device might not be 379idle. This is perfectly normal. If the reason for failure was that 380the device hasn't been idle for long enough, a delayed workqueue 381routine is automatically set up to carry out the operation when the 382autosuspend idle-delay has expired. 383 384Autoresume attempts also can fail. This will happen if power/level is 385set to "suspend" or if the device doesn't manage to resume properly. 386Unlike autosuspend, there's no delay for an autoresume. 387 388 389 Other parts of the driver interface 390 ----------------------------------- 391 392Sometimes a driver needs to make sure that remote wakeup is enabled 393during autosuspend. For example, there's not much point 394autosuspending a keyboard if the user can't cause the keyboard to do a 395remote wakeup by typing on it. If the driver sets 396intf->needs_remote_wakeup to 1, the kernel won't autosuspend the 397device if remote wakeup isn't available or has been disabled through 398the power/wakeup attribute. (If the device is already autosuspended, 399though, setting this flag won't cause the kernel to autoresume it. 400Normally a driver would set this flag in its probe method, at which 401time the device is guaranteed not to be autosuspended.) 402 403The usb_autopm_* routines have to run in a sleepable process context; 404they must not be called from an interrupt handler or while holding a 405spinlock. In fact, the entire autosuspend mechanism is not well geared 406toward interrupt-driven operation. However there is one thing a 407driver can do in an interrupt handler: 408 409 usb_mark_last_busy(struct usb_device *udev); 410 411This sets udev->last_busy to the current time. udev->last_busy is the 412field used for idle-delay calculations; updating it will cause any 413pending autosuspend to be moved back. The usb_autopm_* routines will 414also set the last_busy field to the current time. 415 416Calling urb_mark_last_busy() from within an URB completion handler is 417subject to races: The kernel may have just finished deciding the 418device has been idle for long enough but not yet gotten around to 419calling the driver's suspend method. The driver would have to be 420responsible for synchronizing its suspend method with its URB 421completion handler and causing the autosuspend to fail with -EBUSY if 422an URB had completed too recently. 423 424External suspend calls should never be allowed to fail in this way, 425only autosuspend calls. The driver can tell them apart by checking 426udev->auto_pm; this flag will be set to 1 for internal PM events 427(autosuspend or autoresume) and 0 for external PM events. 428 429Many of the ingredients in the autosuspend framework are oriented 430towards interfaces: The usb_interface structure contains the 431pm_usage_cnt field, and the usb_autopm_* routines take an interface 432pointer as their argument. But somewhat confusingly, a few of the 433pieces (usb_mark_last_busy() and udev->auto_pm) use the usb_device 434structure instead. Drivers need to keep this straight; they can call 435interface_to_usbdev() to find the device structure for a given 436interface. 437 438 439 Locking requirements 440 -------------------- 441 442All three suspend/resume methods are always called while holding the 443usb_device's PM mutex. For external events -- but not necessarily for 444autosuspend or autoresume -- the device semaphore (udev->dev.sem) will 445also be held. This implies that external suspend/resume events are 446mutually exclusive with calls to probe, disconnect, pre_reset, and 447post_reset; the USB core guarantees that this is true of internal 448suspend/resume events as well. 449 450If a driver wants to block all suspend/resume calls during some 451critical section, it can simply acquire udev->pm_mutex. Note that 452calls to resume may be triggered indirectly. Block IO due to memory 453allocations can make the vm subsystem resume a device. Thus while 454holding this lock you must not allocate memory with GFP_KERNEL or 455GFP_NOFS. 456 457Alternatively, if the critical section might call some of the 458usb_autopm_* routines, the driver can avoid deadlock by doing: 459 460 down(&udev->dev.sem); 461 rc = usb_autopm_get_interface(intf); 462 463and at the end of the critical section: 464 465 if (!rc) 466 usb_autopm_put_interface(intf); 467 up(&udev->dev.sem); 468 469Holding the device semaphore will block all external PM calls, and the 470usb_autopm_get_interface() will prevent any internal PM calls, even if 471it fails. (Exercise: Why?) 472 473The rules for locking order are: 474 475 Never acquire any device semaphore while holding any PM mutex. 476 477 Never acquire udev->pm_mutex while holding the PM mutex for 478 a device that isn't a descendant of udev. 479 480In other words, PM mutexes should only be acquired going up the device 481tree, and they should be acquired only after locking all the device 482semaphores you need to hold. These rules don't matter to drivers very 483much; they usually affect just the USB core. 484 485Still, drivers do need to be careful. For example, many drivers use a 486private mutex to synchronize their normal I/O activities with their 487disconnect method. Now if the driver supports autosuspend then it 488must call usb_autopm_put_interface() from somewhere -- maybe from its 489close method. It should make the call while holding the private mutex, 490since a driver shouldn't call any of the usb_autopm_* functions for an 491interface from which it has been unbound. 492 493But the usb_autpm_* routines always acquire the device's PM mutex, and 494consequently the locking order has to be: private mutex first, PM 495mutex second. Since the suspend method is always called with the PM 496mutex held, it mustn't try to acquire the private mutex. It has to 497synchronize with the driver's I/O activities in some other way. 498 499 500 Interaction between dynamic PM and system PM 501 -------------------------------------------- 502 503Dynamic power management and system power management can interact in 504a couple of ways. 505 506Firstly, a device may already be manually suspended or autosuspended 507when a system suspend occurs. Since system suspends are supposed to 508be as transparent as possible, the device should remain suspended 509following the system resume. The 2.6.23 kernel obeys this principle 510for manually suspended devices but not for autosuspended devices; they 511do get resumed when the system wakes up. (Presumably they will be 512autosuspended again after their idle-delay time expires.) In later 513kernels this behavior will be fixed. 514 515(There is an exception. If a device would undergo a reset-resume 516instead of a normal resume, and the device is enabled for remote 517wakeup, then the reset-resume takes place even if the device was 518already suspended when the system suspend began. The justification is 519that a reset-resume is a kind of remote-wakeup event. Or to put it 520another way, a device which needs a reset won't be able to generate 521normal remote-wakeup signals, so it ought to be resumed immediately.) 522 523Secondly, a dynamic power-management event may occur as a system 524suspend is underway. The window for this is short, since system 525suspends don't take long (a few seconds usually), but it can happen. 526For example, a suspended device may send a remote-wakeup signal while 527the system is suspending. The remote wakeup may succeed, which would 528cause the system suspend to abort. If the remote wakeup doesn't 529succeed, it may still remain active and thus cause the system to 530resume as soon as the system suspend is complete. Or the remote 531wakeup may fail and get lost. Which outcome occurs depends on timing 532and on the hardware and firmware design. 533 534More interestingly, a device might undergo a manual resume or 535autoresume during system suspend. With current kernels this shouldn't 536happen, because manual resumes must be initiated by userspace and 537autoresumes happen in response to I/O requests, but all user processes 538and I/O should be quiescent during a system suspend -- thanks to the 539freezer. However there are plans to do away with the freezer, which 540would mean these things would become possible. If and when this comes 541about, the USB core will carefully arrange matters so that either type 542of resume will block until the entire system has resumed. 543