Re: [PATCH v2] Input: add 'safe' user switch codes

[Geert Uytterhoeven]

Hi Jeff,

On Sun, Nov 7, 2021 at 7:17 AM Jeff LaBundy <jeff@xxxxxxxxxxx> wrote:
> On Sat, Nov 06, 2021 at 10:13:15AM +0000, Kieran Bingham wrote:
> > Quoting Dmitry Torokhov (2021-11-05 23:04:23)
> > > On Fri, Nov 05, 2021 at 12:00:37PM -0500, Jeff LaBundy wrote:
> > > > On Fri, Nov 05, 2021 at 10:35:07AM +0000, Kieran Bingham wrote:
> > > > > All existing SW input codes define an action which can be interpreted by
> > > > > a user environment to adapt to the condition of the switch.
> > > > >
> > > > > For example, switches to define the audio mute, will prevent audio
> > > > > playback, and switches to indicate lid and covers being closed may
> > > > > disable displays.
> > > > >
> > > > > Many evaluation platforms provide switches which can be connected to the
> > > > > input system but associating these to an action incorrectly could
> > > > > provide inconsistent end user experiences due to unmarked switch
> > > > > positions.
> > > > >
> > > > > Define two custom user defined switches allowing hardware descriptions
> > > > > to be created whereby the position of the switch is not interpreted as
> > > > > any standard condition that will affect a user experience.
> > > > >
> > > > > This allows wiring up custom generic switches in a way that will allow
> > > > > them to be read and processed, without incurring undesired or otherwise
> > > > > undocumented (by the hardware) 'default' behaviours.
> > > > >
> > > > > Signed-off-by: Kieran Bingham <kieran.bingham+renesas@xxxxxxxxxxxxxxxx>
> > > > > ---
> > > > >
> > > > > Sigh, a compile test might have at least saved the buildbots the trouble
> > > > > of notifying me I also need to update the INPUT_DEVICE_ID_SW_MAX. But
> > > > > even so - I'm really looking for a discussion on the best ways to
> > > > > describe a non-defined switch in device tree.
> > > > >
> > > > > Here's a compiling v2 ;-) But the real questions are :
> > > > >
> > > > >  - Should an existing feature switch be used for generic switches?
> > > > >  - Should we even have a 'user' defined switch?
> > > > >  - If we add user switches, how many?
> > > > >
> > > >
> > > > This is merely my opinion, but if a hardware switch does not have a defined
> > > > purpose, it does not seem necessary to represent it with an input device.
> > >
> > > Yes, exactly. For input core we are trying to avoid generic events with
> > > no defined meaning.
> >
> > That's understandable, particularly as I could then ponder - how do we
> > even define generic switches, and how many ;-) I wanted to discuss it
> > because otherwise these switches will be defined in DT as buttons. And
> > they are not buttons...
> >
> > > What are these switches? GPIOs? Maybe it would be better to use GPIO
> > > layer to test the state for them?
> >
> > They are physical slide switches on the board. But they have no defined
> > purpose by the hardware designer. The purpose would be defined by the
> > end user, as otherwise they are generic test switches.
> >
> > These have been previously handled as gpio-key buttons, for instance
> > key-1 to key-4 at [0] are actually four slides switches.
> >
> > [0] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e3414b8c45afa5cdfb1ffd10f5334da3458c4aa5
> >
> > What I'm trying to determine/promote is that they are not push buttons,
> > and shouldn't be described as such. I have posted [1] to add support for
> > these switches, but I am limited to chosing 'functions' which will have
> > an impact on the system...
> >
> > [1] https://lore.kernel.org/all/20211025130457.935122-1-kieran.bingham+renesas@xxxxxxxxxxxxxxxx/
> >
> > Presently in [1] I have chosen SW_LID and SW_DOCK as very arbitrary
> > functions for the switches. But my concern is that in doing so, the
> > SW_LID position could for instance suggest to a window environment or
> > power management system that the lid is closed, and the system should
> > be suspended (of course depending upon configurations). That would mean
> > that the board would now be potentially always heading into a suspend
> > after power up which would not be at all clear from the switch.
> >
> > I believe a 'switch' is the correct way to define this hardware, so that
> > both positions can be determined, and read, and events generated on
> > state change - but that there shouldn't be any artificially imposed side
> > effects from the description.
> >
> > If the answer is "no we can't have generic switches" then so be it, but
> > it feels wrong to further propogate the definition of these test
> > switches as keys.
>
> I agree that a slide switch tied to a GPIO is indeed a switch in terms of
> input core. Note, however, that definitions from your first example (such
> as KEY_1) are not any less generic; those have specific meanings too.

But at least the KEY_* events are less likely to cause harmful side
effects than the SW_* events.  I have no idea which daemon in e.g. a
generic Ubuntu userspace would act on the SW_* events.

> If the concern is that toggling a switch effects undesired behavior, such
> as turning a display on or off, then the switch should not be represented
> with a gpio-keys node in the first place.
>
> Stated another way, the fact that the GPIO are connected to something does
> not necessarily mean they need to be supported. Only once they map to some
> function should they be defined, in my opinion.

Following the mantra "DT describes hardware, not software policy", I
think we should describe generic switches in DT, and perhaps have a way
to configure the actual event code from userspace (e.g. using sysfs?).
Without such configuration, perhaps they could emit a SW_* event with
the MSB set, so at least their state can be queried using e.g. evtest?

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds