Re: [RFC] Proposal: common kernel-wide GPIO interface

From: Jim Cromie
Date: Tue Aug 01 2006 - 11:51:32 EST

Ben Dooks wrote:
On Sun, Jul 30, 2006 at 03:08:11PM +0200, Robert Schwebel wrote:

On Fri, Jul 28, 2006 at 09:44:40PM +0100, Chris Boot wrote:
I propose to develop a common way of registering and accessing GPIO pins on various devices.
I've attached the gpio framework we have developed a while ago; it is
not ready for upstream, only tested on pxa and has probably several
other drawbacks, but may be a start for your activities. One of the
problems we've recently seen is that for example on PowerPCs you don't
have such a clear "this is gpio pin x" nomenclature, so the question
would be how to do the mapping here.

Right, my $0.02 worth:

$2.00 at least.
I have a patch which adds a sysfs interface much as youve described below.
an old version of patch is here:
Its far from complete, but I think it belongs in this discussion !

1) The system does not currently allow for other GPIO sources
than the CPU. There are a variety of GPIOs, that could come
from expansion chips, on board CPLDs, etc.

Im not sure what you mean here -
the above patch manages to add a sysfs-gpio interface to 2 drivers: scx200_gpio, and pc8736x_gpio.

ISTM that you've described a limitation of Robert Schwebel's patch,
since his examples use a single path in sysfs.

+ Or to stop the motor again:
+ $ echo 1 > /sys/class/gpio/gpio63/level

Heres my sysfs-gpio interface, which obviously covers both drivers:

soekris:/sys/devices/platform# ls *.0/bit_0.0_*
pc8736x_gpio.0/bit_0.0_current_output scx200_gpio.0/bit_0.0_current_output
pc8736x_gpio.0/bit_0.0_debounced scx200_gpio.0/bit_0.0_debounced
pc8736x_gpio.0/bit_0.0_locked scx200_gpio.0/bit_0.0_locked
pc8736x_gpio.0/bit_0.0_output_enabled scx200_gpio.0/bit_0.0_output_enabled
pc8736x_gpio.0/bit_0.0_pullup_enabled scx200_gpio.0/bit_0.0_pullup_enabled
pc8736x_gpio.0/bit_0.0_status scx200_gpio.0/bit_0.0_status
pc8736x_gpio.0/bit_0.0_totem scx200_gpio.0/bit_0.0_totem
pc8736x_gpio.0/bit_0.0_value scx200_gpio.0/bit_0.0_value

Robert, sysfs seems to populate lots of symlinks underneath /sys,
do any of them give a device-centric organization that lets you address separate devices ?

2) The GPIO configuration from my last thought experiment have the
following properties for each pin:

- input
- inverted input

- normal output
- inverted output
- tristatable output
- open collector (can only pull to zero)
- open emmitor (can only pull to high)

The allowance of inverted outputs, is very useful to allow
drivers to assume either '0' or '1' is an active signal, allowing
per-board fixups when the designer suddely decides the best way
of connecting device A to B is via a spare inverter...

The other way would be to allow the mapping of '0' and '1' states
to either of the states:

- output 1
- output 0
- tri-state

The classing of tri-state as a seperate from input, is in case the
hardware does not see input as a valid state, or that input and
output are somehow different. pull resistor:
- tristate (no resistor)
- pull low
- pull high

The input and output are seperate, assuming that there is the
possiblity the system can read back the line even if the GPIO
is set as an output.

heres how Im doing it..

struct gpio_attributes {
struct sensor_device_attribute_2 value;
struct sensor_device_attribute_2 curr;
struct sensor_device_attribute_2 output_enabled;
struct sensor_device_attribute_2 totem_pole;
struct sensor_device_attribute_2 pullup_enabled;
struct sensor_device_attribute_2 debounced;
struct sensor_device_attribute_2 locked;
struct sensor_device_attribute_2 status;

3) The sysfs interface should be configurable, as systems
with lots of GPIO would end up with large numbers of
files and directories in sysfs.

static int nobits = 0;
module_param(nobits, int, 0);
MODULE_PARM_DESC(nobits, "nobits=1 to suppress sysfs bits interface");

static int noports = 0;
module_param(noports, int, 0);
MODULE_PARM_DESC(noports, "noports=1 to supress sysfs ports interface");

4) you probably want to ensure pull-up resistors are off if the
output is being driven.

Sounds smart. Is there any reason to make it overrideable ?

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at