[PATCH 3/6 v14] gpio: Add userland device interface to block GPIO

[alexander.stein@systec-electronic.com (Alexander Stein)]

On Tuesday 22 January 2013 13:06:41, Roland Stigge wrote:
> This patch adds a character device interface to the block GPIO system.
> 
> Signed-off-by: Roland Stigge <stigge@antcom.de>
> ---
>  Documentation/ABI/testing/dev-gpioblock |   34 ++++
>  drivers/gpio/gpiolib.c                  |  225 +++++++++++++++++++++++++++++++-
>  include/linux/gpio.h                    |   13 +
>  3 files changed, 271 insertions(+), 1 deletion(-)
> 
> --- /dev/null
> +++ linux-2.6/Documentation/ABI/testing/dev-gpioblock
> @@ -0,0 +1,34 @@
> +What:		/dev/<gpioblock>
> +Date:		Nov 2012
> +KernelVersion:	3.7
> +Contact:	Roland Stigge <stigge@antcom.de>
> +Description:	The /dev/<gpioblock> character device node provides userspace
> +		access to GPIO blocks, named exactly as the block, e.g.
> +		/dev/block0.
> +
> +		Reading:
> +		When reading sizeof(unsigned long) bytes from the device, the
> +		current state of the block, masked by the current mask (see
> +		below) can be obtained as a word. When the device is opened
> +		with O_NONBLOCK, read() always returns with data immediately,
> +		otherwise it blocks until data is available, see IRQ handling
> +		below.
> +
> +		Writing:
> +		By writing sizeof(unsigned long) bytes to the device, the
> +		current state of the block can be set. This operation is
> +		masked by the current mask (see below).
> +
> +		IRQ handling:
> +		When one or more IRQs in the block are IRQ capable, you can
                          ^^^^
I think this should be GPIOs

> +static long gpio_block_fop_ioctl(struct file *f, unsigned int cmd,
> +				 unsigned long arg)
> +{
> +	struct gpio_block *block = (struct gpio_block *)f->private_data;
> +	unsigned long __user *x = (unsigned long __user *)arg;
> +
> +	if (cmd == 0)
> +		return get_user(block->cur_mask, x);
> +
> +	return -EINVAL;
> +}

So there is no way from userspace to create/remove GPIO blocks? I know support in sysfs is problematic due to formatting, but an IOCTL for that would be nice.

Best regards,
Alexander