Re: Using EV_MSC or extending KEY_*

["Rick L. Vinyard, Jr." <rvinyard@cs.nmsu.edu>]

Henrique de Moraes Holschuh wrote:
> On Thu, 17 Sep 2009, Dmitry Torokhov wrote:
>> On Thu, Sep 17, 2009 at 02:57:07PM -0300, Henrique de Moraes Holschuh
>> wrote:
>> > On Wed, 16 Sep 2009, Rick L. Vinyard, Jr. wrote:
>> > > The M* keys are intended to provide a quick way to switch between
>> key
>> > > mappings, with each mode having their own user-defined mappings.
>> >
>> > What I'd do in this case would be this:
>> >
>> > 1. Initially have the M* level-shift keys assigned KEY_RESERVED
>> >
>> > 2. Have a big enough keymap to map all keys in all M*-level shift
>> states
>> > possible.
>> >
>> > Eg:
>> >    START OF KEYMAP
>> >    M* keys
>> >    1st set of G* keys
>> >    2nd set of G* keys
>> >    3rd set of G* keys...
>> >    ...
>> >    last set of G* keys
>> >    END OF KEYMAP
>> >
>> > 3. Have the driver special-process M* level-shift keys *as long as
>> they are
>> > still set to KEY_RESERVED* to select which part of the keymap is used
>> to
>> > translate the other keys.  Note that this likely means pressing a M*
>> key
>> > would be transparent to userspace in this case, i.e. no events would
>> be
>> > issued when a M* key is doing a level shift.
>> >
>> > So, you'd be able to set all mappings you want in the driver, and the
>> M*
>> > keys would do what they're expected to do without any userland help at
>> all,
>> > but you'd still be able to program the M* keys to be normal keys if
>> you
>> > want.
>> >
>> > Of course, this assumes you don't do chording on multiple M* keys to
>> end up
>> > with a huge number of keymaps :p
>>
>> Actually I think that the device should just emit KEY_PROG1..KEY_PROG4
>> for the M keys and have userspace daemon load alternate keymaps on the
>> fly in resaponse to KEY_PROGx. The device is just a set of completely
>> generic buttons... User will have to tell the kernel what to map them
>> to.
>
> It would work, but it is a big trip through userspace.  If quickly
> pressing
> M#+G# is a common use pattern (and it will be, for gaming), i.e. you often
> want to access quickly a function on one level then another on a different
> level, asking userspace to upload a new keymap to switch levels at every
> M#
> press is going to be way too racy.
>
> If it is to be used like that, I'd advocate either doing the entire
> map-switching thing in kernel space, or doing the entire mapping in
> userspace.  In the later case, you don't issue KEY_* in the kernel driver,
> you just issue MSC_SCAN events, and the userspace driver should open the
> input device in exclusive mode, do its magic, and use uinput to generate
> translated events (KEY_*, and even BTN_*, etc).
>
> I understand the current version allows for an all-userspace enhanced
> driver
> if one sets the entire keymap to KEY_RESERVED, since it will issue
> MSC_SCAN
> events for all keys (if it doesn't, I suggest doing so).

It does.

> That might
> indeed
> be the best option if one doesn't want a more complex kernel driver.  And
> one could still use the device in degraded mode by not loading the
> userspace
> driver, and uploading a regular keymap to it.
>

A keymap for three modes wouldn't be too bad space wise. I'd prefer to
keep the input events in kernel space.

I was trying to minimize the code in the driver, but I don't think it will
increase it by that much.