cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

power-management.rst (35394B)


      1.. _usb-power-management:
      2
      3Power Management for USB
      4~~~~~~~~~~~~~~~~~~~~~~~~
      5
      6:Author: Alan Stern <stern@rowland.harvard.edu>
      7:Date: Last-updated: February 2014
      8
      9..
     10	Contents:
     11	---------
     12	* What is Power Management?
     13	* What is Remote Wakeup?
     14	* When is a USB device idle?
     15	* Forms of dynamic PM
     16	* The user interface for dynamic PM
     17	* Changing the default idle-delay time
     18	* Warnings
     19	* The driver interface for Power Management
     20	* The driver interface for autosuspend and autoresume
     21	* Other parts of the driver interface
     22	* Mutual exclusion
     23	* Interaction between dynamic PM and system PM
     24	* xHCI hardware link PM
     25	* USB Port Power Control
     26	* User Interface for Port Power Control
     27	* Suggested Userspace Port Power Policy
     28
     29
     30What is Power Management?
     31-------------------------
     32
     33Power Management (PM) is the practice of saving energy by suspending
     34parts of a computer system when they aren't being used.  While a
     35component is ``suspended`` it is in a nonfunctional low-power state; it
     36might even be turned off completely.  A suspended component can be
     37``resumed`` (returned to a functional full-power state) when the kernel
     38needs to use it.  (There also are forms of PM in which components are
     39placed in a less functional but still usable state instead of being
     40suspended; an example would be reducing the CPU's clock rate.  This
     41document will not discuss those other forms.)
     42
     43When the parts being suspended include the CPU and most of the rest of
     44the system, we speak of it as a "system suspend".  When a particular
     45device is turned off while the system as a whole remains running, we
     46call it a "dynamic suspend" (also known as a "runtime suspend" or
     47"selective suspend").  This document concentrates mostly on how
     48dynamic PM is implemented in the USB subsystem, although system PM is
     49covered to some extent (see ``Documentation/power/*.rst`` for more
     50information about system PM).
     51
     52System PM support is present only if the kernel was built with
     53``CONFIG_SUSPEND`` or ``CONFIG_HIBERNATION`` enabled.  Dynamic PM support
     54
     55for USB is present whenever
     56the kernel was built with ``CONFIG_PM`` enabled.
     57
     58[Historically, dynamic PM support for USB was present only if the
     59kernel had been built with ``CONFIG_USB_SUSPEND`` enabled (which depended on
     60``CONFIG_PM_RUNTIME``).  Starting with the 3.10 kernel release, dynamic PM
     61support for USB was present whenever the kernel was built with
     62``CONFIG_PM_RUNTIME`` enabled.  The ``CONFIG_USB_SUSPEND`` option had been
     63eliminated.]
     64
     65
     66What is Remote Wakeup?
     67----------------------
     68
     69When a device has been suspended, it generally doesn't resume until
     70the computer tells it to.  Likewise, if the entire computer has been
     71suspended, it generally doesn't resume until the user tells it to, say
     72by pressing a power button or opening the cover.
     73
     74However some devices have the capability of resuming by themselves, or
     75asking the kernel to resume them, or even telling the entire computer
     76to resume.  This capability goes by several names such as "Wake On
     77LAN"; we will refer to it generically as "remote wakeup".  When a
     78device is enabled for remote wakeup and it is suspended, it may resume
     79itself (or send a request to be resumed) in response to some external
     80event.  Examples include a suspended keyboard resuming when a key is
     81pressed, or a suspended USB hub resuming when a device is plugged in.
     82
     83
     84When is a USB device idle?
     85--------------------------
     86
     87A device is idle whenever the kernel thinks it's not busy doing
     88anything important and thus is a candidate for being suspended.  The
     89exact definition depends on the device's driver; drivers are allowed
     90to declare that a device isn't idle even when there's no actual
     91communication taking place.  (For example, a hub isn't considered idle
     92unless all the devices plugged into that hub are already suspended.)
     93In addition, a device isn't considered idle so long as a program keeps
     94its usbfs file open, whether or not any I/O is going on.
     95
     96If a USB device has no driver, its usbfs file isn't open, and it isn't
     97being accessed through sysfs, then it definitely is idle.
     98
     99
    100Forms of dynamic PM
    101-------------------
    102
    103Dynamic suspends occur when the kernel decides to suspend an idle
    104device.  This is called ``autosuspend`` for short.  In general, a device
    105won't be autosuspended unless it has been idle for some minimum period
    106of time, the so-called idle-delay time.
    107
    108Of course, nothing the kernel does on its own initiative should
    109prevent the computer or its devices from working properly.  If a
    110device has been autosuspended and a program tries to use it, the
    111kernel will automatically resume the device (autoresume).  For the
    112same reason, an autosuspended device will usually have remote wakeup
    113enabled, if the device supports remote wakeup.
    114
    115It is worth mentioning that many USB drivers don't support
    116autosuspend.  In fact, at the time of this writing (Linux 2.6.23) the
    117only drivers which do support it are the hub driver, kaweth, asix,
    118usblp, usblcd, and usb-skeleton (which doesn't count).  If a
    119non-supporting driver is bound to a device, the device won't be
    120autosuspended.  In effect, the kernel pretends the device is never
    121idle.
    122
    123We can categorize power management events in two broad classes:
    124external and internal.  External events are those triggered by some
    125agent outside the USB stack: system suspend/resume (triggered by
    126userspace), manual dynamic resume (also triggered by userspace), and
    127remote wakeup (triggered by the device).  Internal events are those
    128triggered within the USB stack: autosuspend and autoresume.  Note that
    129all dynamic suspend events are internal; external agents are not
    130allowed to issue dynamic suspends.
    131
    132
    133The user interface for dynamic PM
    134---------------------------------
    135
    136The user interface for controlling dynamic PM is located in the ``power/``
    137subdirectory of each USB device's sysfs directory, that is, in
    138``/sys/bus/usb/devices/.../power/`` where "..." is the device's ID.  The
    139relevant attribute files are: wakeup, control, and
    140``autosuspend_delay_ms``.  (There may also be a file named ``level``; this
    141file was deprecated as of the 2.6.35 kernel and replaced by the
    142``control`` file.  In 2.6.38 the ``autosuspend`` file will be deprecated
    143and replaced by the ``autosuspend_delay_ms`` file.  The only difference
    144is that the newer file expresses the delay in milliseconds whereas the
    145older file uses seconds.  Confusingly, both files are present in 2.6.37
    146but only ``autosuspend`` works.)
    147
    148	``power/wakeup``
    149
    150		This file is empty if the device does not support
    151		remote wakeup.  Otherwise the file contains either the
    152		word ``enabled`` or the word ``disabled``, and you can
    153		write those words to the file.  The setting determines
    154		whether or not remote wakeup will be enabled when the
    155		device is next suspended.  (If the setting is changed
    156		while the device is suspended, the change won't take
    157		effect until the following suspend.)
    158
    159	``power/control``
    160
    161		This file contains one of two words: ``on`` or ``auto``.
    162		You can write those words to the file to change the
    163		device's setting.
    164
    165		- ``on`` means that the device should be resumed and
    166		  autosuspend is not allowed.  (Of course, system
    167		  suspends are still allowed.)
    168
    169		- ``auto`` is the normal state in which the kernel is
    170		  allowed to autosuspend and autoresume the device.
    171
    172		(In kernels up to 2.6.32, you could also specify
    173		``suspend``, meaning that the device should remain
    174		suspended and autoresume was not allowed.  This
    175		setting is no longer supported.)
    176
    177	``power/autosuspend_delay_ms``
    178
    179		This file contains an integer value, which is the
    180		number of milliseconds the device should remain idle
    181		before the kernel will autosuspend it (the idle-delay
    182		time).  The default is 2000.  0 means to autosuspend
    183		as soon as the device becomes idle, and negative
    184		values mean never to autosuspend.  You can write a
    185		number to the file to change the autosuspend
    186		idle-delay time.
    187
    188Writing ``-1`` to ``power/autosuspend_delay_ms`` and writing ``on`` to
    189``power/control`` do essentially the same thing -- they both prevent the
    190device from being autosuspended.  Yes, this is a redundancy in the
    191API.
    192
    193(In 2.6.21 writing ``0`` to ``power/autosuspend`` would prevent the device
    194from being autosuspended; the behavior was changed in 2.6.22.  The
    195``power/autosuspend`` attribute did not exist prior to 2.6.21, and the
    196``power/level`` attribute did not exist prior to 2.6.22.  ``power/control``
    197was added in 2.6.34, and ``power/autosuspend_delay_ms`` was added in
    1982.6.37 but did not become functional until 2.6.38.)
    199
    200
    201Changing the default idle-delay time
    202------------------------------------
    203
    204The default autosuspend idle-delay time (in seconds) is controlled by
    205a module parameter in usbcore.  You can specify the value when usbcore
    206is loaded.  For example, to set it to 5 seconds instead of 2 you would
    207do::
    208
    209	modprobe usbcore autosuspend=5
    210
    211Equivalently, you could add to a configuration file in /etc/modprobe.d
    212a line saying::
    213
    214	options usbcore autosuspend=5
    215
    216Some distributions load the usbcore module very early during the boot
    217process, by means of a program or script running from an initramfs
    218image.  To alter the parameter value you would have to rebuild that
    219image.
    220
    221If usbcore is compiled into the kernel rather than built as a loadable
    222module, you can add::
    223
    224	usbcore.autosuspend=5
    225
    226to the kernel's boot command line.
    227
    228Finally, the parameter value can be changed while the system is
    229running.  If you do::
    230
    231	echo 5 >/sys/module/usbcore/parameters/autosuspend
    232
    233then each new USB device will have its autosuspend idle-delay
    234initialized to 5.  (The idle-delay values for already existing devices
    235will not be affected.)
    236
    237Setting the initial default idle-delay to -1 will prevent any
    238autosuspend of any USB device.  This has the benefit of allowing you
    239then to enable autosuspend for selected devices.
    240
    241
    242Warnings
    243--------
    244
    245The USB specification states that all USB devices must support power
    246management.  Nevertheless, the sad fact is that many devices do not
    247support it very well.  You can suspend them all right, but when you
    248try to resume them they disconnect themselves from the USB bus or
    249they stop working entirely.  This seems to be especially prevalent
    250among printers and scanners, but plenty of other types of device have
    251the same deficiency.
    252
    253For this reason, by default the kernel disables autosuspend (the
    254``power/control`` attribute is initialized to ``on``) for all devices other
    255than hubs.  Hubs, at least, appear to be reasonably well-behaved in
    256this regard.
    257
    258(In 2.6.21 and 2.6.22 this wasn't the case.  Autosuspend was enabled
    259by default for almost all USB devices.  A number of people experienced
    260problems as a result.)
    261
    262This means that non-hub devices won't be autosuspended unless the user
    263or a program explicitly enables it.  As of this writing there aren't
    264any widespread programs which will do this; we hope that in the near
    265future device managers such as HAL will take on this added
    266responsibility.  In the meantime you can always carry out the
    267necessary operations by hand or add them to a udev script.  You can
    268also change the idle-delay time; 2 seconds is not the best choice for
    269every device.
    270
    271If a driver knows that its device has proper suspend/resume support,
    272it can enable autosuspend all by itself.  For example, the video
    273driver for a laptop's webcam might do this (in recent kernels they
    274do), since these devices are rarely used and so should normally be
    275autosuspended.
    276
    277Sometimes it turns out that even when a device does work okay with
    278autosuspend there are still problems.  For example, the usbhid driver,
    279which manages keyboards and mice, has autosuspend support.  Tests with
    280a number of keyboards show that typing on a suspended keyboard, while
    281causing the keyboard to do a remote wakeup all right, will nonetheless
    282frequently result in lost keystrokes.  Tests with mice show that some
    283of them will issue a remote-wakeup request in response to button
    284presses but not to motion, and some in response to neither.
    285
    286The kernel will not prevent you from enabling autosuspend on devices
    287that can't handle it.  It is even possible in theory to damage a
    288device by suspending it at the wrong time.  (Highly unlikely, but
    289possible.)  Take care.
    290
    291
    292The driver interface for Power Management
    293-----------------------------------------
    294
    295The requirements for a USB driver to support external power management
    296are pretty modest; the driver need only define::
    297
    298	.suspend
    299	.resume
    300	.reset_resume
    301
    302methods in its :c:type:`usb_driver` structure, and the ``reset_resume`` method
    303is optional.  The methods' jobs are quite simple:
    304
    305      - The ``suspend`` method is called to warn the driver that the
    306	device is going to be suspended.  If the driver returns a
    307	negative error code, the suspend will be aborted.  Normally
    308	the driver will return 0, in which case it must cancel all
    309	outstanding URBs (:c:func:`usb_kill_urb`) and not submit any more.
    310
    311      - The ``resume`` method is called to tell the driver that the
    312	device has been resumed and the driver can return to normal
    313	operation.  URBs may once more be submitted.
    314
    315      - The ``reset_resume`` method is called to tell the driver that
    316	the device has been resumed and it also has been reset.
    317	The driver should redo any necessary device initialization,
    318	since the device has probably lost most or all of its state
    319	(although the interfaces will be in the same altsettings as
    320	before the suspend).
    321
    322If the device is disconnected or powered down while it is suspended,
    323the ``disconnect`` method will be called instead of the ``resume`` or
    324``reset_resume`` method.  This is also quite likely to happen when
    325waking up from hibernation, as many systems do not maintain suspend
    326current to the USB host controllers during hibernation.  (It's
    327possible to work around the hibernation-forces-disconnect problem by
    328using the USB Persist facility.)
    329
    330The ``reset_resume`` method is used by the USB Persist facility (see
    331:ref:`usb-persist`) and it can also be used under certain
    332circumstances when ``CONFIG_USB_PERSIST`` is not enabled.  Currently, if a
    333device is reset during a resume and the driver does not have a
    334``reset_resume`` method, the driver won't receive any notification about
    335the resume.  Later kernels will call the driver's ``disconnect`` method;
    3362.6.23 doesn't do this.
    337
    338USB drivers are bound to interfaces, so their ``suspend`` and ``resume``
    339methods get called when the interfaces are suspended or resumed.  In
    340principle one might want to suspend some interfaces on a device (i.e.,
    341force the drivers for those interface to stop all activity) without
    342suspending the other interfaces.  The USB core doesn't allow this; all
    343interfaces are suspended when the device itself is suspended and all
    344interfaces are resumed when the device is resumed.  It isn't possible
    345to suspend or resume some but not all of a device's interfaces.  The
    346closest you can come is to unbind the interfaces' drivers.
    347
    348
    349The driver interface for autosuspend and autoresume
    350---------------------------------------------------
    351
    352To support autosuspend and autoresume, a driver should implement all
    353three of the methods listed above.  In addition, a driver indicates
    354that it supports autosuspend by setting the ``.supports_autosuspend`` flag
    355in its usb_driver structure.  It is then responsible for informing the
    356USB core whenever one of its interfaces becomes busy or idle.  The
    357driver does so by calling these six functions::
    358
    359	int  usb_autopm_get_interface(struct usb_interface *intf);
    360	void usb_autopm_put_interface(struct usb_interface *intf);
    361	int  usb_autopm_get_interface_async(struct usb_interface *intf);
    362	void usb_autopm_put_interface_async(struct usb_interface *intf);
    363	void usb_autopm_get_interface_no_resume(struct usb_interface *intf);
    364	void usb_autopm_put_interface_no_suspend(struct usb_interface *intf);
    365
    366The functions work by maintaining a usage counter in the
    367usb_interface's embedded device structure.  When the counter is > 0
    368then the interface is deemed to be busy, and the kernel will not
    369autosuspend the interface's device.  When the usage counter is = 0
    370then the interface is considered to be idle, and the kernel may
    371autosuspend the device.
    372
    373Drivers must be careful to balance their overall changes to the usage
    374counter.  Unbalanced "get"s will remain in effect when a driver is
    375unbound from its interface, preventing the device from going into
    376runtime suspend should the interface be bound to a driver again.  On
    377the other hand, drivers are allowed to achieve this balance by calling
    378the ``usb_autopm_*`` functions even after their ``disconnect`` routine
    379has returned -- say from within a work-queue routine -- provided they
    380retain an active reference to the interface (via ``usb_get_intf`` and
    381``usb_put_intf``).
    382
    383Drivers using the async routines are responsible for their own
    384synchronization and mutual exclusion.
    385
    386	:c:func:`usb_autopm_get_interface` increments the usage counter and
    387	does an autoresume if the device is suspended.  If the
    388	autoresume fails, the counter is decremented back.
    389
    390	:c:func:`usb_autopm_put_interface` decrements the usage counter and
    391	attempts an autosuspend if the new value is = 0.
    392
    393	:c:func:`usb_autopm_get_interface_async` and
    394	:c:func:`usb_autopm_put_interface_async` do almost the same things as
    395	their non-async counterparts.  The big difference is that they
    396	use a workqueue to do the resume or suspend part of their
    397	jobs.  As a result they can be called in an atomic context,
    398	such as an URB's completion handler, but when they return the
    399	device will generally not yet be in the desired state.
    400
    401	:c:func:`usb_autopm_get_interface_no_resume` and
    402	:c:func:`usb_autopm_put_interface_no_suspend` merely increment or
    403	decrement the usage counter; they do not attempt to carry out
    404	an autoresume or an autosuspend.  Hence they can be called in
    405	an atomic context.
    406
    407The simplest usage pattern is that a driver calls
    408:c:func:`usb_autopm_get_interface` in its open routine and
    409:c:func:`usb_autopm_put_interface` in its close or release routine.  But other
    410patterns are possible.
    411
    412The autosuspend attempts mentioned above will often fail for one
    413reason or another.  For example, the ``power/control`` attribute might be
    414set to ``on``, or another interface in the same device might not be
    415idle.  This is perfectly normal.  If the reason for failure was that
    416the device hasn't been idle for long enough, a timer is scheduled to
    417carry out the operation automatically when the autosuspend idle-delay
    418has expired.
    419
    420Autoresume attempts also can fail, although failure would mean that
    421the device is no longer present or operating properly.  Unlike
    422autosuspend, there's no idle-delay for an autoresume.
    423
    424
    425Other parts of the driver interface
    426-----------------------------------
    427
    428Drivers can enable autosuspend for their devices by calling::
    429
    430	usb_enable_autosuspend(struct usb_device *udev);
    431
    432in their :c:func:`probe` routine, if they know that the device is capable of
    433suspending and resuming correctly.  This is exactly equivalent to
    434writing ``auto`` to the device's ``power/control`` attribute.  Likewise,
    435drivers can disable autosuspend by calling::
    436
    437	usb_disable_autosuspend(struct usb_device *udev);
    438
    439This is exactly the same as writing ``on`` to the ``power/control`` attribute.
    440
    441Sometimes a driver needs to make sure that remote wakeup is enabled
    442during autosuspend.  For example, there's not much point
    443autosuspending a keyboard if the user can't cause the keyboard to do a
    444remote wakeup by typing on it.  If the driver sets
    445``intf->needs_remote_wakeup`` to 1, the kernel won't autosuspend the
    446device if remote wakeup isn't available.  (If the device is already
    447autosuspended, though, setting this flag won't cause the kernel to
    448autoresume it.  Normally a driver would set this flag in its ``probe``
    449method, at which time the device is guaranteed not to be
    450autosuspended.)
    451
    452If a driver does its I/O asynchronously in interrupt context, it
    453should call :c:func:`usb_autopm_get_interface_async` before starting output and
    454:c:func:`usb_autopm_put_interface_async` when the output queue drains.  When
    455it receives an input event, it should call::
    456
    457	usb_mark_last_busy(struct usb_device *udev);
    458
    459in the event handler.  This tells the PM core that the device was just
    460busy and therefore the next autosuspend idle-delay expiration should
    461be pushed back.  Many of the usb_autopm_* routines also make this call,
    462so drivers need to worry only when interrupt-driven input arrives.
    463
    464Asynchronous operation is always subject to races.  For example, a
    465driver may call the :c:func:`usb_autopm_get_interface_async` routine at a time
    466when the core has just finished deciding the device has been idle for
    467long enough but not yet gotten around to calling the driver's ``suspend``
    468method.  The ``suspend`` method must be responsible for synchronizing with
    469the I/O request routine and the URB completion handler; it should
    470cause autosuspends to fail with -EBUSY if the driver needs to use the
    471device.
    472
    473External suspend calls should never be allowed to fail in this way,
    474only autosuspend calls.  The driver can tell them apart by applying
    475the :c:func:`PMSG_IS_AUTO` macro to the message argument to the ``suspend``
    476method; it will return True for internal PM events (autosuspend) and
    477False for external PM events.
    478
    479
    480Mutual exclusion
    481----------------
    482
    483For external events -- but not necessarily for autosuspend or
    484autoresume -- the device semaphore (udev->dev.sem) will be held when a
    485``suspend`` or ``resume`` method is called.  This implies that external
    486suspend/resume events are mutually exclusive with calls to ``probe``,
    487``disconnect``, ``pre_reset``, and ``post_reset``; the USB core guarantees that
    488this is true of autosuspend/autoresume events as well.
    489
    490If a driver wants to block all suspend/resume calls during some
    491critical section, the best way is to lock the device and call
    492:c:func:`usb_autopm_get_interface` (and do the reverse at the end of the
    493critical section).  Holding the device semaphore will block all
    494external PM calls, and the :c:func:`usb_autopm_get_interface` will prevent any
    495internal PM calls, even if it fails.  (Exercise: Why?)
    496
    497
    498Interaction between dynamic PM and system PM
    499--------------------------------------------
    500
    501Dynamic power management and system power management can interact in
    502a couple of ways.
    503
    504Firstly, a device may already be autosuspended when a system suspend
    505occurs.  Since system suspends are supposed to be as transparent as
    506possible, the device should remain suspended following the system
    507resume.  But this theory may not work out well in practice; over time
    508the kernel's behavior in this regard has changed.  As of 2.6.37 the
    509policy is to resume all devices during a system resume and let them
    510handle their own runtime suspends afterward.
    511
    512Secondly, a dynamic power-management event may occur as a system
    513suspend is underway.  The window for this is short, since system
    514suspends don't take long (a few seconds usually), but it can happen.
    515For example, a suspended device may send a remote-wakeup signal while
    516the system is suspending.  The remote wakeup may succeed, which would
    517cause the system suspend to abort.  If the remote wakeup doesn't
    518succeed, it may still remain active and thus cause the system to
    519resume as soon as the system suspend is complete.  Or the remote
    520wakeup may fail and get lost.  Which outcome occurs depends on timing
    521and on the hardware and firmware design.
    522
    523
    524xHCI hardware link PM
    525---------------------
    526
    527xHCI host controller provides hardware link power management to usb2.0
    528(xHCI 1.0 feature) and usb3.0 devices which support link PM. By
    529enabling hardware LPM, the host can automatically put the device into
    530lower power state(L1 for usb2.0 devices, or U1/U2 for usb3.0 devices),
    531which state device can enter and resume very quickly.
    532
    533The user interface for controlling hardware LPM is located in the
    534``power/`` subdirectory of each USB device's sysfs directory, that is, in
    535``/sys/bus/usb/devices/.../power/`` where "..." is the device's ID. The
    536relevant attribute files are ``usb2_hardware_lpm`` and ``usb3_hardware_lpm``.
    537
    538	``power/usb2_hardware_lpm``
    539
    540		When a USB2 device which support LPM is plugged to a
    541		xHCI host root hub which support software LPM, the
    542		host will run a software LPM test for it; if the device
    543		enters L1 state and resume successfully and the host
    544		supports USB2 hardware LPM, this file will show up and
    545		driver will enable hardware LPM	for the device. You
    546		can write y/Y/1 or n/N/0 to the file to	enable/disable
    547		USB2 hardware LPM manually. This is for	test purpose mainly.
    548
    549	``power/usb3_hardware_lpm_u1``
    550	``power/usb3_hardware_lpm_u2``
    551
    552		When a USB 3.0 lpm-capable device is plugged in to a
    553		xHCI host which supports link PM, it will check if U1
    554		and U2 exit latencies have been set in the BOS
    555		descriptor; if the check is passed and the host
    556		supports USB3 hardware LPM, USB3 hardware LPM will be
    557		enabled for the device and these files will be created.
    558		The files hold a string value (enable or disable)
    559		indicating whether or not USB3 hardware LPM U1 or U2
    560		is enabled for the device.
    561
    562USB Port Power Control
    563----------------------
    564
    565In addition to suspending endpoint devices and enabling hardware
    566controlled link power management, the USB subsystem also has the
    567capability to disable power to ports under some conditions.  Power is
    568controlled through ``Set/ClearPortFeature(PORT_POWER)`` requests to a hub.
    569In the case of a root or platform-internal hub the host controller
    570driver translates ``PORT_POWER`` requests into platform firmware (ACPI)
    571method calls to set the port power state. For more background see the
    572Linux Plumbers Conference 2012 slides [#f1]_ and video [#f2]_:
    573
    574Upon receiving a ``ClearPortFeature(PORT_POWER)`` request a USB port is
    575logically off, and may trigger the actual loss of VBUS to the port [#f3]_.
    576VBUS may be maintained in the case where a hub gangs multiple ports into
    577a shared power well causing power to remain until all ports in the gang
    578are turned off.  VBUS may also be maintained by hub ports configured for
    579a charging application.  In any event a logically off port will lose
    580connection with its device, not respond to hotplug events, and not
    581respond to remote wakeup events.
    582
    583.. warning::
    584
    585   turning off a port may result in the inability to hot add a device.
    586   Please see "User Interface for Port Power Control" for details.
    587
    588As far as the effect on the device itself it is similar to what a device
    589goes through during system suspend, i.e. the power session is lost.  Any
    590USB device or driver that misbehaves with system suspend will be
    591similarly affected by a port power cycle event.  For this reason the
    592implementation shares the same device recovery path (and honors the same
    593quirks) as the system resume path for the hub.
    594
    595.. [#f1]
    596
    597  http://dl.dropbox.com/u/96820575/sarah-sharp-lpt-port-power-off2-mini.pdf
    598
    599.. [#f2]
    600
    601  http://linuxplumbers.ubicast.tv/videos/usb-port-power-off-kerneluserspace-api/
    602
    603.. [#f3]
    604
    605  USB 3.1 Section 10.12
    606
    607  wakeup note: if a device is configured to send wakeup events the port
    608  power control implementation will block poweroff attempts on that
    609  port.
    610
    611
    612User Interface for Port Power Control
    613-------------------------------------
    614
    615The port power control mechanism uses the PM runtime system.  Poweroff is
    616requested by clearing the ``power/pm_qos_no_power_off`` flag of the port device
    617(defaults to 1).  If the port is disconnected it will immediately receive a
    618``ClearPortFeature(PORT_POWER)`` request.  Otherwise, it will honor the pm
    619runtime rules and require the attached child device and all descendants to be
    620suspended. This mechanism is dependent on the hub advertising port power
    621switching in its hub descriptor (wHubCharacteristics logical power switching
    622mode field).
    623
    624Note, some interface devices/drivers do not support autosuspend.  Userspace may
    625need to unbind the interface drivers before the :c:type:`usb_device` will
    626suspend.  An unbound interface device is suspended by default.  When unbinding,
    627be careful to unbind interface drivers, not the driver of the parent usb
    628device.  Also, leave hub interface drivers bound.  If the driver for the usb
    629device (not interface) is unbound the kernel is no longer able to resume the
    630device.  If a hub interface driver is unbound, control of its child ports is
    631lost and all attached child-devices will disconnect.  A good rule of thumb is
    632that if the 'driver/module' link for a device points to
    633``/sys/module/usbcore`` then unbinding it will interfere with port power
    634control.
    635
    636Example of the relevant files for port power control.  Note, in this example
    637these files are relative to a usb hub device (prefix)::
    638
    639     prefix=/sys/devices/pci0000:00/0000:00:14.0/usb3/3-1
    640
    641                      attached child device +
    642                  hub port device +         |
    643     hub interface device +       |         |
    644                          v       v         v
    645                  $prefix/3-1:1.0/3-1-port1/device
    646
    647     $prefix/3-1:1.0/3-1-port1/power/pm_qos_no_power_off
    648     $prefix/3-1:1.0/3-1-port1/device/power/control
    649     $prefix/3-1:1.0/3-1-port1/device/3-1.1:<intf0>/driver/unbind
    650     $prefix/3-1:1.0/3-1-port1/device/3-1.1:<intf1>/driver/unbind
    651     ...
    652     $prefix/3-1:1.0/3-1-port1/device/3-1.1:<intfN>/driver/unbind
    653
    654In addition to these files some ports may have a 'peer' link to a port on
    655another hub.  The expectation is that all superspeed ports have a
    656hi-speed peer::
    657
    658  $prefix/3-1:1.0/3-1-port1/peer -> ../../../../usb2/2-1/2-1:1.0/2-1-port1
    659  ../../../../usb2/2-1/2-1:1.0/2-1-port1/peer -> ../../../../usb3/3-1/3-1:1.0/3-1-port1
    660
    661Distinct from 'companion ports', or 'ehci/xhci shared switchover ports'
    662peer ports are simply the hi-speed and superspeed interface pins that
    663are combined into a single usb3 connector.  Peer ports share the same
    664ancestor XHCI device.
    665
    666While a superspeed port is powered off a device may downgrade its
    667connection and attempt to connect to the hi-speed pins.  The
    668implementation takes steps to prevent this:
    669
    6701. Port suspend is sequenced to guarantee that hi-speed ports are powered-off
    671   before their superspeed peer is permitted to power-off.  The implication is
    672   that the setting ``pm_qos_no_power_off`` to zero on a superspeed port may
    673   not cause the port to power-off until its highspeed peer has gone to its
    674   runtime suspend state.  Userspace must take care to order the suspensions
    675   if it wants to guarantee that a superspeed port will power-off.
    676
    6772. Port resume is sequenced to force a superspeed port to power-on prior to its
    678   highspeed peer.
    679
    6803. Port resume always triggers an attached child device to resume.  After a
    681   power session is lost the device may have been removed, or need reset.
    682   Resuming the child device when the parent port regains power resolves those
    683   states and clamps the maximum port power cycle frequency at the rate the
    684   child device can suspend (autosuspend-delay) and resume (reset-resume
    685   latency).
    686
    687Sysfs files relevant for port power control:
    688
    689	``<hubdev-portX>/power/pm_qos_no_power_off``:
    690		This writable flag controls the state of an idle port.
    691		Once all children and descendants have suspended the
    692		port may suspend/poweroff provided that
    693		pm_qos_no_power_off is '0'.  If pm_qos_no_power_off is
    694		'1' the port will remain active/powered regardless of
    695		the stats of descendants.  Defaults to 1.
    696
    697	``<hubdev-portX>/power/runtime_status``:
    698		This file reflects whether the port is 'active' (power is on)
    699		or 'suspended' (logically off).  There is no indication to
    700		userspace whether VBUS is still supplied.
    701
    702	``<hubdev-portX>/connect_type``:
    703		An advisory read-only flag to userspace indicating the
    704		location and connection type of the port.  It returns
    705		one of four values 'hotplug', 'hardwired', 'not used',
    706		and 'unknown'.  All values, besides unknown, are set by
    707		platform firmware.
    708
    709		``hotplug`` indicates an externally connectable/visible
    710		port on the platform.  Typically userspace would choose
    711		to keep such a port powered to handle new device
    712		connection events.
    713
    714		``hardwired`` refers to a port that is not visible but
    715		connectable. Examples are internal ports for USB
    716		bluetooth that can be disconnected via an external
    717		switch or a port with a hardwired USB camera.  It is
    718		expected to be safe to allow these ports to suspend
    719		provided pm_qos_no_power_off is coordinated with any
    720		switch that gates connections.  Userspace must arrange
    721		for the device to be connected prior to the port
    722		powering off, or to activate the port prior to enabling
    723		connection via a switch.
    724
    725		``not used`` refers to an internal port that is expected
    726		to never have a device connected to it.  These may be
    727		empty internal ports, or ports that are not physically
    728		exposed on a platform.  Considered safe to be
    729		powered-off at all times.
    730
    731		``unknown`` means platform firmware does not provide
    732		information for this port.  Most commonly refers to
    733		external hub ports which should be considered 'hotplug'
    734		for policy decisions.
    735
    736		.. note::
    737
    738			- since we are relying on the BIOS to get this ACPI
    739			  information correct, the USB port descriptions may
    740			  be missing or wrong.
    741
    742			- Take care in clearing ``pm_qos_no_power_off``. Once
    743			  power is off this port will
    744			  not respond to new connect events.
    745
    746	Once a child device is attached additional constraints are
    747	applied before the port is allowed to poweroff.
    748
    749	``<child>/power/control``:
    750		Must be ``auto``, and the port will not
    751		power down until ``<child>/power/runtime_status``
    752		reflects the 'suspended' state.  Default
    753		value is controlled by child device driver.
    754
    755	``<child>/power/persist``:
    756		This defaults to ``1`` for most devices and indicates if
    757		kernel can persist the device's configuration across a
    758		power session loss (suspend / port-power event).  When
    759		this value is ``0`` (quirky devices), port poweroff is
    760		disabled.
    761
    762	``<child>/driver/unbind``:
    763		Wakeup capable devices will block port poweroff.  At
    764		this time the only mechanism to clear the usb-internal
    765		wakeup-capability for an interface device is to unbind
    766		its driver.
    767
    768Summary of poweroff pre-requisite settings relative to a port device::
    769
    770	echo 0 > power/pm_qos_no_power_off
    771	echo 0 > peer/power/pm_qos_no_power_off # if it exists
    772	echo auto > power/control # this is the default value
    773	echo auto > <child>/power/control
    774	echo 1 > <child>/power/persist # this is the default value
    775
    776Suggested Userspace Port Power Policy
    777-------------------------------------
    778
    779As noted above userspace needs to be careful and deliberate about what
    780ports are enabled for poweroff.
    781
    782The default configuration is that all ports start with
    783``power/pm_qos_no_power_off`` set to ``1`` causing ports to always remain
    784active.
    785
    786Given confidence in the platform firmware's description of the ports
    787(ACPI _PLD record for a port populates 'connect_type') userspace can
    788clear pm_qos_no_power_off for all 'not used' ports.  The same can be
    789done for 'hardwired' ports provided poweroff is coordinated with any
    790connection switch for the port.
    791
    792A more aggressive userspace policy is to enable USB port power off for
    793all ports (set ``<hubdev-portX>/power/pm_qos_no_power_off`` to ``0``) when
    794some external factor indicates the user has stopped interacting with the
    795system.  For example, a distro may want to enable power off all USB
    796ports when the screen blanks, and re-power them when the screen becomes
    797active.  Smart phones and tablets may want to power off USB ports when
    798the user pushes the power button.