1GPIO Interfaces
   3This provides an overview of GPIO access conventions on Linux.
   5These calls use the gpio_* naming prefix.  No other calls should use that
   6prefix, or the related __gpio_* prefix.
   9What is a GPIO?
  11A "General Purpose Input/Output" (GPIO) is a flexible software-controlled
  12digital signal.  They are provided from many kinds of chip, and are familiar
  13to Linux developers working with embedded and custom hardware.  Each GPIO
  14represents a bit connected to a particular pin, or "ball" on Ball Grid Array
  15(BGA) packages.  Board schematics show which external hardware connects to
  16which GPIOs.  Drivers can be written generically, so that board setup code
  17passes such pin configuration data to drivers.
  19System-on-Chip (SOC) processors heavily rely on GPIOs.  In some cases, every
  20non-dedicated pin can be configured as a GPIO; and most chips have at least
  21several dozen of them.  Programmable logic devices (like FPGAs) can easily
  22provide GPIOs; multifunction chips like power managers, and audio codecs
  23often have a few such pins to help with pin scarcity on SOCs; and there are
  24also "GPIO Expander" chips that connect using the I2C or SPI serial busses.
  25Most PC southbridges have a few dozen GPIO-capable pins (with only the BIOS
  26firmware knowing how they're used).
  28The exact capabilities of GPIOs vary between systems.  Common options:
  30  - Output values are writable (high=1, low=0).  Some chips also have
  31    options about how that value is driven, so that for example only one
  32    value might be driven ... supporting "wire-OR" and similar schemes
  33    for the other value (notably, "open drain" signaling).
  35  - Input values are likewise readable (1, 0).  Some chips support readback
  36    of pins configured as "output", which is very useful in such "wire-OR"
  37    cases (to support bidirectional signaling).  GPIO controllers may have
  38    input de-glitch/debounce logic, sometimes with software controls.
  40  - Inputs can often be used as IRQ signals, often edge triggered but
  41    sometimes level triggered.  Such IRQs may be configurable as system
  42    wakeup events, to wake the system from a low power state.
  44  - Usually a GPIO will be configurable as either input or output, as needed
  45    by different product boards; single direction ones exist too.
  47  - Most GPIOs can be accessed while holding spinlocks, but those accessed
  48    through a serial bus normally can't.  Some systems support both types.
  50On a given board each GPIO is used for one specific purpose like monitoring
  51MMC/SD card insertion/removal, detecting card writeprotect status, driving
  52a LED, configuring a transceiver, bitbanging a serial bus, poking a hardware
  53watchdog, sensing a switch, and so on.
  56GPIO conventions
  58Note that this is called a "convention" because you don't need to do it this
  59way, and it's no crime if you don't.  There **are** cases where portability
  60is not the main issue; GPIOs are often used for the kind of board-specific
  61glue logic that may even change between board revisions, and can't ever be
  62used on a board that's wired differently.  Only least-common-denominator
  63functionality can be very portable.  Other features are platform-specific,
  64and that can be critical for glue logic.
  66Plus, this doesn't require any implementation framework, just an interface.
  67One platform might implement it as simple inline functions accessing chip
  68registers; another might implement it by delegating through abstractions
  69used for several very different kinds of GPIO controller.  (There is some
  70optional code supporting such an implementation strategy, described later
  71in this document, but drivers acting as clients to the GPIO interface must
  72not care how it's implemented.)
  74That said, if the convention is supported on their platform, drivers should
  75use it when possible.  Platforms must declare GENERIC_GPIO support in their
  76Kconfig (boolean true), and provide an <asm/gpio.h> file.  Drivers that can't
  77work without standard GPIO calls should have Kconfig entries which depend
  78on GENERIC_GPIO.  The GPIO calls are available, either as "real code" or as
  79optimized-away stubs, when drivers use the include file:
  81        #include <linux/gpio.h>
  83If you stick to this convention then it'll be easier for other developers to
  84see what your code is doing, and help maintain it.
  86Note that these operations include I/O barriers on platforms which need to
  87use them; drivers don't need to add them explicitly.
  90Identifying GPIOs
  92GPIOs are identified by unsigned integers in the range 0..MAX_INT.  That
  93reserves "negative" numbers for other purposes like marking signals as
  94"not available on this board", or indicating faults.  Code that doesn't
  95touch the underlying hardware treats these integers as opaque cookies.
  97Platforms define how they use those integers, and usually #define symbols
  98for the GPIO lines so that board-specific setup code directly corresponds
  99to the relevant schematics.  In contrast, drivers should only use GPIO
 100numbers passed to them from that setup code, using platform_data to hold
 101board-specific pin configuration data (along with other board specific
 102data they need).  That avoids portability problems.
 104So for example one platform uses numbers 32-159 for GPIOs; while another
 105uses numbers 0..63 with one set of GPIO controllers, 64-79 with another
 106type of GPIO controller, and on one particular board 80-95 with an FPGA.
 107The numbers need not be contiguous; either of those platforms could also
 108use numbers 2000-2063 to identify GPIOs in a bank of I2C GPIO expanders.
 110If you want to initialize a structure with an invalid GPIO number, use
 111some negative number (perhaps "-EINVAL"); that will never be valid.  To
 112test if a number could reference a GPIO, you may use this predicate:
 114        int gpio_is_valid(int number);
 116A number that's not valid will be rejected by calls which may request
 117or free GPIOs (see below).  Other numbers may also be rejected; for
 118example, a number might be valid but unused on a given board.
 120Whether a platform supports multiple GPIO controllers is currently a
 121platform-specific implementation issue.
 124Using GPIOs
 126The first thing a system should do with a GPIO is allocate it, using
 127the gpio_request() call; see later.
 129One of the next things to do with a GPIO, often in board setup code when
 130setting up a platform_device using the GPIO, is mark its direction:
 132        /* set as input or output, returning 0 or negative errno */
 133        int gpio_direction_input(unsigned gpio);
 134        int gpio_direction_output(unsigned gpio, int value);
 136The return value is zero for success, else a negative errno.  It should
 137be checked, since the get/set calls don't have error returns and since
 138misconfiguration is possible.  You should normally issue these calls from
 139a task context.  However, for spinlock-safe GPIOs it's OK to use them
 140before tasking is enabled, as part of early board setup.
 142For output GPIOs, the value provided becomes the initial output value.
 143This helps avoid signal glitching during system startup.
 145For compatibility with legacy interfaces to GPIOs, setting the direction
 146of a GPIO implicitly requests that GPIO (see below) if it has not been
 147requested already.  That compatibility is being removed from the optional
 148gpiolib framework.
 150Setting the direction can fail if the GPIO number is invalid, or when
 151that particular GPIO can't be used in that mode.  It's generally a bad
 152idea to rely on boot firmware to have set the direction correctly, since
 153it probably wasn't validated to do more than boot Linux.  (Similarly,
 154that board setup code probably needs to multiplex that pin as a GPIO,
 155and configure pullups/pulldowns appropriately.)
 158Spinlock-Safe GPIO access
 160Most GPIO controllers can be accessed with memory read/write instructions.
 161That doesn't need to sleep, and can safely be done from inside IRQ handlers.
 162(That includes hardirq contexts on RT kernels.)
 164Use these calls to access such GPIOs:
 166        /* GPIO INPUT:  return zero or nonzero */
 167        int gpio_get_value(unsigned gpio);
 169        /* GPIO OUTPUT */
 170        void gpio_set_value(unsigned gpio, int value);
 172The values are boolean, zero for low, nonzero for high.  When reading the
 173value of an output pin, the value returned should be what's seen on the
 174pin ... that won't always match the specified output value, because of
 175issues including open-drain signaling and output latencies.
 177The get/set calls have no error returns because "invalid GPIO" should have
 178been reported earlier from gpio_direction_*().  However, note that not all
 179platforms can read the value of output pins; those that can't should always
 180return zero.  Also, using these calls for GPIOs that can't safely be accessed
 181without sleeping (see below) is an error.
 183Platform-specific implementations are encouraged to optimize the two
 184calls to access the GPIO value in cases where the GPIO number (and for
 185output, value) are constant.  It's normal for them to need only a couple
 186of instructions in such cases (reading or writing a hardware register),
 187and not to need spinlocks.  Such optimized calls can make bitbanging
 188applications a lot more efficient (in both space and time) than spending
 189dozens of instructions on subroutine calls.
 192GPIO access that may sleep
 194Some GPIO controllers must be accessed using message based busses like I2C
 195or SPI.  Commands to read or write those GPIO values require waiting to
 196get to the head of a queue to transmit a command and get its response.
 197This requires sleeping, which can't be done from inside IRQ handlers.
 199Platforms that support this type of GPIO distinguish them from other GPIOs
 200by returning nonzero from this call (which requires a valid GPIO number,
 201which should have been previously allocated with gpio_request):
 203        int gpio_cansleep(unsigned gpio);
 205To access such GPIOs, a different set of accessors is defined:
 207        /* GPIO INPUT:  return zero or nonzero, might sleep */
 208        int gpio_get_value_cansleep(unsigned gpio);
 210        /* GPIO OUTPUT, might sleep */
 211        void gpio_set_value_cansleep(unsigned gpio, int value);
 213Other than the fact that these calls might sleep, and will not be ignored
 214for GPIOs that can't be accessed from IRQ handlers, these calls act the
 215same as the spinlock-safe calls.
 218Claiming and Releasing GPIOs
 220To help catch system configuration errors, two calls are defined.
 222        /* request GPIO, returning 0 or negative errno.
 223         * non-null labels may be useful for diagnostics.
 224         */
 225        int gpio_request(unsigned gpio, const char *label);
 227        /* release previously-claimed GPIO */
 228        void gpio_free(unsigned gpio);
 230Passing invalid GPIO numbers to gpio_request() will fail, as will requesting
 231GPIOs that have already been claimed with that call.  The return value of
 232gpio_request() must be checked.  You should normally issue these calls from
 233a task context.  However, for spinlock-safe GPIOs it's OK to request GPIOs
 234before tasking is enabled, as part of early board setup.
 236These calls serve two basic purposes.  One is marking the signals which
 237are actually in use as GPIOs, for better diagnostics; systems may have
 238several hundred potential GPIOs, but often only a dozen are used on any
 239given board.  Another is to catch conflicts, identifying errors when
 240(a) two or more drivers wrongly think they have exclusive use of that
 241signal, or (b) something wrongly believes it's safe to remove drivers
 242needed to manage a signal that's in active use.  That is, requesting a
 243GPIO can serve as a kind of lock.
 245Some platforms may also use knowledge about what GPIOs are active for
 246power management, such as by powering down unused chip sectors and, more
 247easily, gating off unused clocks.
 249Note that requesting a GPIO does NOT cause it to be configured in any
 250way; it just marks that GPIO as in use.  Separate code must handle any
 251pin setup (e.g. controlling which pin the GPIO uses, pullup/pulldown).
 253Also note that it's your responsibility to have stopped using a GPIO
 254before you free it.
 257GPIOs mapped to IRQs
 259GPIO numbers are unsigned integers; so are IRQ numbers.  These make up
 260two logically distinct namespaces (GPIO 0 need not use IRQ 0).  You can
 261map between them using calls like:
 263        /* map GPIO numbers to IRQ numbers */
 264        int gpio_to_irq(unsigned gpio);
 266        /* map IRQ numbers to GPIO numbers (avoid using this) */
 267        int irq_to_gpio(unsigned irq);
 269Those return either the corresponding number in the other namespace, or
 270else a negative errno code if the mapping can't be done.  (For example,
 271some GPIOs can't be used as IRQs.)  It is an unchecked error to use a GPIO
 272number that wasn't set up as an input using gpio_direction_input(), or
 273to use an IRQ number that didn't originally come from gpio_to_irq().
 275These two mapping calls are expected to cost on the order of a single
 276addition or subtraction.  They're not allowed to sleep.
 278Non-error values returned from gpio_to_irq() can be passed to request_irq()
 279or free_irq().  They will often be stored into IRQ resources for platform
 280devices, by the board-specific initialization code.  Note that IRQ trigger
 281options are part of the IRQ interface, e.g. IRQF_TRIGGER_FALLING, as are
 282system wakeup capabilities.
 284Non-error values returned from irq_to_gpio() would most commonly be used
 285with gpio_get_value(), for example to initialize or update driver state
 286when the IRQ is edge-triggered.  Note that some platforms don't support
 287this reverse mapping, so you should avoid using it.
 290Emulating Open Drain Signals
 292Sometimes shared signals need to use "open drain" signaling, where only the
 293low signal level is actually driven.  (That term applies to CMOS transistors;
 294"open collector" is used for TTL.)  A pullup resistor causes the high signal
 295level.  This is sometimes called a "wire-AND"; or more practically, from the
 296negative logic (low=true) perspective this is a "wire-OR".
 298One common example of an open drain signal is a shared active-low IRQ line.
 299Also, bidirectional data bus signals sometimes use open drain signals.
 301Some GPIO controllers directly support open drain outputs; many don't.  When
 302you need open drain signaling but your hardware doesn't directly support it,
 303there's a common idiom you can use to emulate it with any GPIO pin that can
 304be used as either an input or an output:
 306 LOW:   gpio_direction_output(gpio, 0) ... this drives the signal
 307        and overrides the pullup.
 309 HIGH:  gpio_direction_input(gpio) ... this turns off the output,
 310        so the pullup (or some other device) controls the signal.
 312If you are "driving" the signal high but gpio_get_value(gpio) reports a low
 313value (after the appropriate rise time passes), you know some other component
 314is driving the shared signal low.  That's not necessarily an error.  As one
 315common example, that's how I2C clocks are stretched:  a slave that needs a
 316slower clock delays the rising edge of SCK, and the I2C master adjusts its
 317signaling rate accordingly.
 320What do these conventions omit?
 322One of the biggest things these conventions omit is pin multiplexing, since
 323this is highly chip-specific and nonportable.  One platform might not need
 324explicit multiplexing; another might have just two options for use of any
 325given pin; another might have eight options per pin; another might be able
 326to route a given GPIO to any one of several pins.  (Yes, those examples all
 327come from systems that run Linux today.)
 329Related to multiplexing is configuration and enabling of the pullups or
 330pulldowns integrated on some platforms.  Not all platforms support them,
 331or support them in the same way; and any given board might use external
 332pullups (or pulldowns) so that the on-chip ones should not be used.
 333(When a circuit needs 5 kOhm, on-chip 100 kOhm resistors won't do.)
 334Likewise drive strength (2 mA vs 20 mA) and voltage (1.8V vs 3.3V) is a
 335platform-specific issue, as are models like (not) having a one-to-one
 336correspondence between configurable pins and GPIOs.
 338There are other system-specific mechanisms that are not specified here,
 339like the aforementioned options for input de-glitching and wire-OR output.
 340Hardware may support reading or writing GPIOs in gangs, but that's usually
 341configuration dependent:  for GPIOs sharing the same bank.  (GPIOs are
 342commonly grouped in banks of 16 or 32, with a given SOC having several such
 343banks.)  Some systems can trigger IRQs from output GPIOs, or read values
 344from pins not managed as GPIOs.  Code relying on such mechanisms will
 345necessarily be nonportable.
 347Dynamic definition of GPIOs is not currently standard; for example, as
 348a side effect of configuring an add-on board with some GPIO expanders.
 351GPIO implementor's framework (OPTIONAL)
 353As noted earlier, there is an optional implementation framework making it
 354easier for platforms to support different kinds of GPIO controller using
 355the same programming interface.  This framework is called "gpiolib".
 357As a debugging aid, if debugfs is available a /sys/kernel/debug/gpio file
 358will be found there.  That will list all the controllers registered through
 359this framework, and the state of the GPIOs currently in use.
 362Controller Drivers: gpio_chip
 364In this framework each GPIO controller is packaged as a "struct gpio_chip"
 365with information common to each controller of that type:
 367 - methods to establish GPIO direction
 368 - methods used to access GPIO values
 369 - flag saying whether calls to its methods may sleep
 370 - optional debugfs dump method (showing extra state like pullup config)
 371 - label for diagnostics
 373There is also per-instance data, which may come from device.platform_data:
 374the number of its first GPIO, and how many GPIOs it exposes.
 376The code implementing a gpio_chip should support multiple instances of the
 377controller, possibly using the driver model.  That code will configure each
 378gpio_chip and issue gpiochip_add().  Removing a GPIO controller should be
 379rare; use gpiochip_remove() when it is unavoidable.
 381Most often a gpio_chip is part of an instance-specific structure with state
 382not exposed by the GPIO interfaces, such as addressing, power management,
 383and more.  Chips such as codecs will have complex non-GPIO state,
 385Any debugfs dump method should normally ignore signals which haven't been
 386requested as GPIOs.  They can use gpiochip_is_requested(), which returns
 387either NULL or the label associated with that GPIO when it was requested.
 390Platform Support
 392To support this framework, a platform's Kconfig will "select" either
 394and arrange that its <asm/gpio.h> includes <asm-generic/gpio.h> and defines
 395three functions: gpio_get_value(), gpio_set_value(), and gpio_cansleep().
 396They may also want to provide a custom value for ARCH_NR_GPIOS.
 398ARCH_REQUIRE_GPIOLIB means that the gpio-lib code will always get compiled
 399into the kernel on that architecture.
 401ARCH_WANT_OPTIONAL_GPIOLIB means the gpio-lib code defaults to off and the user
 402can enable it and build it into the kernel optionally.
 404If neither of these options are selected, the platform does not support
 405GPIOs through GPIO-lib and the code cannot be enabled by the user.
 407Trivial implementations of those functions can directly use framework
 408code, which always dispatches through the gpio_chip:
 410  #define gpio_get_value        __gpio_get_value
 411  #define gpio_set_value        __gpio_set_value
 412  #define gpio_cansleep         __gpio_cansleep
 414Fancier implementations could instead define those as inline functions with
 415logic optimizing access to specific SOC-based GPIOs.  For example, if the
 416referenced GPIO is the constant "12", getting or setting its value could
 417cost as little as two or three instructions, never sleeping.  When such an
 418optimization is not possible those calls must delegate to the framework
 419code, costing at least a few dozen instructions.  For bitbanged I/O, such
 420instruction savings can be significant.
 422For SOCs, platform-specific code defines and registers gpio_chip instances
 423for each bank of on-chip GPIOs.  Those GPIOs should be numbered/labeled to
 424match chip vendor documentation, and directly match board schematics.  They
 425may well start at zero and go up to a platform-specific limit.  Such GPIOs
 426are normally integrated into platform initialization to make them always be
 427available, from arch_initcall() or earlier; they can often serve as IRQs.
 430Board Support
 432For external GPIO controllers -- such as I2C or SPI expanders, ASICs, multi
 433function devices, FPGAs or CPLDs -- most often board-specific code handles
 434registering controller devices and ensures that their drivers know what GPIO
 435numbers to use with gpiochip_add().  Their numbers often start right after
 436platform-specific GPIOs.
 438For example, board setup code could create structures identifying the range
 439of GPIOs that chip will expose, and passes them to each GPIO expander chip
 440using platform_data.  Then the chip driver's probe() routine could pass that
 441data to gpiochip_add().
 443Initialization order can be important.  For example, when a device relies on
 444an I2C-based GPIO, its probe() routine should only be called after that GPIO
 445becomes available.  That may mean the device should not be registered until
 446calls for that GPIO can work.  One way to address such dependencies is for
 447such gpio_chip controllers to provide setup() and teardown() callbacks to
 448board specific code; those board specific callbacks would register devices
 449once all the necessary resources are available, and remove them later when
 450the GPIO controller device becomes unavailable.
 453Sysfs Interface for Userspace (OPTIONAL)
 455Platforms which use the "gpiolib" implementors framework may choose to
 456configure a sysfs user interface to GPIOs.  This is different from the
 457debugfs interface, since it provides control over GPIO direction and
 458value instead of just showing a gpio state summary.  Plus, it could be
 459present on production systems without debugging support.
 461Given approprate hardware documentation for the system, userspace could
 462know for example that GPIO #23 controls the write protect line used to
 463protect boot loader segments in flash memory.  System upgrade procedures
 464may need to temporarily remove that protection, first importing a GPIO,
 465then changing its output state, then updating the code before re-enabling
 466the write protection.  In normal use, GPIO #23 would never be touched,
 467and the kernel would have no need to know about it.
 469Again depending on appropriate hardware documentation, on some systems
 470userspace GPIO can be used to determine system configuration data that
 471standard kernels won't know about.  And for some tasks, simple userspace
 472GPIO drivers could be all that the system really needs.
 474Note that standard kernel drivers exist for common "LEDs and Buttons"
 475GPIO tasks:  "leds-gpio" and "gpio_keys", respectively.  Use those
 476instead of talking directly to the GPIOs; they integrate with kernel
 477frameworks better than your userspace code could.
 480Paths in Sysfs
 482There are three kinds of entry in /sys/class/gpio:
 484   -    Control interfaces used to get userspace control over GPIOs;
 486   -    GPIOs themselves; and
 488   -    GPIO controllers ("gpio_chip" instances).
 490That's in addition to standard files including the "device" symlink.
 492The control interfaces are write-only:
 494    /sys/class/gpio/
 496        "export" ... Userspace may ask the kernel to export control of
 497                a GPIO to userspace by writing its number to this file.
 499                Example:  "echo 19 > export" will create a "gpio19" node
 500                for GPIO #19, if that's not requested by kernel code.
 502        "unexport" ... Reverses the effect of exporting to userspace.
 504                Example:  "echo 19 > unexport" will remove a "gpio19"
 505                node exported using the "export" file.
 507GPIO signals have paths like /sys/class/gpio/gpio42/ (for GPIO #42)
 508and have the following read/write attributes:
 510    /sys/class/gpio/gpioN/
 512        "direction" ... reads as either "in" or "out".  This value may
 513                normally be written.  Writing as "out" defaults to
 514                initializing the value as low.  To ensure glitch free
 515                operation, values "low" and "high" may be written to
 516                configure the GPIO as an output with that initial value.
 518                Note that this attribute *will not exist* if the kernel
 519                doesn't support changing the direction of a GPIO, or
 520                it was exported by kernel code that didn't explicitly
 521                allow userspace to reconfigure this GPIO's direction.
 523        "value" ... reads as either 0 (low) or 1 (high).  If the GPIO
 524                is configured as an output, this value may be written;
 525                any nonzero value is treated as high.
 527GPIO controllers have paths like /sys/class/gpio/chipchip42/ (for the
 528controller implementing GPIOs starting at #42) and have the following
 529read-only attributes:
 531    /sys/class/gpio/gpiochipN/
 533        "base" ... same as N, the first GPIO managed by this chip
 535        "label" ... provided for diagnostics (not always unique)
 537        "ngpio" ... how many GPIOs this manges (N to N + ngpio - 1)
 539Board documentation should in most cases cover what GPIOs are used for
 540what purposes.  However, those numbers are not always stable; GPIOs on
 541a daughtercard might be different depending on the base board being used,
 542or other cards in the stack.  In such cases, you may need to use the
 543gpiochip nodes (possibly in conjunction with schematics) to determine
 544the correct GPIO number to use for a given signal.
 547Exporting from Kernel code
 549Kernel code can explicitly manage exports of GPIOs which have already been
 550requested using gpio_request():
 552        /* export the GPIO to userspace */
 553        int gpio_export(unsigned gpio, bool direction_may_change);
 555        /* reverse gpio_export() */
 556        void gpio_unexport();
 558After a kernel driver requests a GPIO, it may only be made available in
 559the sysfs interface by gpio_export().  The driver can control whether the
 560signal direction may change.  This helps drivers prevent userspace code
 561from accidentally clobbering important system state.
 563This explicit exporting can help with debugging (by making some kinds
 564of experiments easier), or can provide an always-there interface that's
 565suitable for documenting as part of a board support package.