Re: [RFC][PATCH] POSKeyboard driver for exclusive keyboard access

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

Hi Niels,

On Fri, Aug 22, 2008 at 06:17:28PM +0200, Niels de Vos wrote:
> This new driver makes it possible for middleware like JavaPOS to use
> a POSKeyboard (connected to PS/2) with exclusive access. This is
> required by the UnifiedPOS Specification which is available from
> http://www.nrf-arts.org/UnifiedPOS. Any middleware using this driver
> should implement the full exception-handling in user-space. Therefor
> it is possible to use specific POS-extensions of POSKeyboards, without
> abusing other keyboard-drivers.
> 
> Opening /dev/poskeyboard will route all scancodes to this device. The
> scancodes will not be processes by the input-subsystem anymore. Reading
> /dev/poskeyboard results in receiving the scancodes as raw data for
> further processing by the reader. Sending commands to the hardware can
> be done by writing to /dev/poskeyboard.
> 
> If the driver is loaded and /dev/poskeyboard is not opened, all
> scancodes are given to the input-subsystem. This allows 'normal' use of
> the keyboard.
> 
> Making the driver active involves some commands like the following:
>    echo -n serio1 > /sys/bus/serio/drivers/atkbd/unbind
>    echo -n serio1 > /sys/bus/serio/drivers/poskbd/bind
> 

It seems you have just reimplemented serio_raw driver, therefore NAK.
Right now serio_raw is limited to untranslated ports but adding
SERIO_I8042XL signature should be easy.

> Open questions:
> - How to achieve the same with USB-HID complaint keyboards?

Thsi is probably is the candidate for legitimate use of EVIOCGRAB.
Just grab device for exclusive access and do your processing, probably
looking at MSC_SCAN (or KEY_*, depends on how hardware-sepcific this
thing is). This approach shoudl work equally well for HID, PS/2,
serial or any other kind of keyboard/input device actually. You can
also re-inject the keystrokes you are not interested in back into
input core via uinput.

> - Does serio_unregister_port() a kfree() on the port?

Eventually, after the last user drops off.

-- 
Dmitry