On 10 July 2013 16:20, Dmitry Torokhov <dmitry.torokhov-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:

> On Wednesday, July 10, 2013 10:50:26 PM Grant Likely wrote:
> > On Wed, Jul 10, 2013 at 5:52 PM, Dmitry Torokhov
> >
> > <dmitry.torokhov-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
> > > On Wed, Jul 10, 2013 at 04:14:57PM +0100, Grant Likely wrote:
> > >> On Fri, 28 Jun 2013 07:19:06 -0600, Mathieu Poirier
> <mathieu.poirier-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
> > >> > On 13-06-28 12:09 AM, Dmitry Torokhov wrote:
> > >> > >>>> I do not agree.  We want the binding to be generic and not tied
> > >> > >>>> specifically to the keyreset functionality.  As such
> > >> > >>>> 'input-keyset' or
> > >> > >>>> 'input-keychord' are more appropriate.
> > >> > >>>
> > >> > >>> The binding is defined specifically for sysrq and specifically
> to
> > >> > >>> perform reset action.
> > >> > >>
> > >> > >> Yes for now but as the examples in the binding show, it is easy
> to
> > >> > >> envision how other drivers could use it.
> > >> > >
> > >> > > I think you over-complicate things here. Unlike matrix-keypad
> > >> > > binding,
> > >> > > where you have a common parsing code, here we have an individual
> > >> > > driver.
> > >> > > I really do not see anyone else using such sequences or chords as
> > >> > > such
> > >> > > processing should be done in userspace. Sysrq is quite an
> exception.
> > >> >
> > >> > To be honest I don't have a very strong opinion on the binding.  I
> made
> > >> > it as generic as possible on the guidance of the DT people.  Let's
> see
> > >> > what they think of it.
> > >>
> > >> Hi Mathieu,
> > >>
> > >> As per our conversation just now at Connect, the binding should
> probably
> > >> look like this:
> > >>
> > >> Sysrq keyset binding:
> > >>
> > >> The /chosen node can contain a linux,input-keyset-sysrq child node to
> > >> define a set of keys that will generate a sysrq when pressed together.
> > >
> > > Hmm, we would have only one such node, /sysrq, or /linux,sysrq,
> > > whatever. The sysrq setting is system-wide and applicable to all
> > > devices. Given that it is used only on mobile, where there not that
> > > many input devices (a few keys and touchscreen) I do not believe we
> > > should consider adding per-device settings.
> >
> > It's in /chosen, that isn't per-device.
> >
> > >> Required properties:
> > >> keyset: array of keycodes
> > >
> > > Please, let's call it 'key-reset-seq', because it is exactly the reset
> > > sequence. There won't be any additional sequences or chords as those
> > > should be handled in userspace, sysrq is a special case here.
> >
> > This is absolutely a linux-specific binding. It encodes the Linux
> > keycodes, and generates a linux meaning. I'm usually all about
> > carrying the OS-independent banner when defining DT bindings, but in
> > this case the linux prefix and sysrq reference is completely
> > appropriate.
>
> OK, I have no idea what "/chosen" actually means. What I am trying to say
> that there should be either "sysrq" or "linux,sysrq" node and that is what
> sysrq driver will be looking for.
>


Chosen pertains to system wide parameters that aren't related to hw
specific devices.  Correct, the driver will look exactly for
"linux,sysrq-reset-seq" in the "chosen" node and nowhere else.


>
> The entire node is Linux-specific and therefore there is no point in
> marking only one of the properties (the key sequence) Linux-specific while
> leaving other ones generic.
>
> Thanks.
>
> --
> Dmitry
>