Re: [PATCH] input: uinput: Drop checks for abs_min > abs_max

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

Hi Paul,

On Wed, Dec 20, 2023 at 01:38:39AM +0100, Paul Cercueil wrote:
> Hi Peter,
> 
> Le mercredi 20 décembre 2023 à 09:51 +1000, Peter Hutterer a écrit :
> > On Mon, Dec 18, 2023 at 11:16:53AM -0600, Chris Morgan wrote:
> > > From: Chris Morgan <macromorgan@hotmail.com>
> > > 
> > > Stop checking if the minimum abs value is greater than the maximum
> > > abs
> > > value. When the axis is inverted this condition is allowed. Without
> > > relaxing this check, it is not possible to use uinput on devices in
> > > userspace with an inverted axis, such as the adc-joystick found on
> > > many handheld gaming devices.
> > 
> > As mentioned in the other thread [1] a fair bit of userspace relies
> > on
> > that general assumption so removing it will likely cause all sorts of
> > issues.
> 
> There is some userspace that works with it though, so why restrict it
> artificially?
> 
> The fact that some other userspace code wouldn't work with it sounds a
> bit irrelevant. They just never encountered that min>max usage before.
> 
> And removing this check won't cause all sort of issues, why would it?
> It's not like the current software actively probes min>max and crash
> badly if it doesn't return -EINVAL...

It will cause weird movements because calculations expect min be the
minimum, and max the maximum, and not encode left/right or up/down.
Putting this into adc joystick binding was a mistake.

This is the definition of absinfo:

/**
 * struct input_absinfo - used by EVIOCGABS/EVIOCSABS ioctls
 * @value: latest reported value for the axis.
 * @minimum: specifies minimum value for the axis.
 * @maximum: specifies maximum value for the axis.
 * @fuzz: specifies fuzz value that is used to filter noise from
 *	the event stream.
 * @flat: values that are within this value will be discarded by
 *	joydev interface and reported as 0 instead.
 * @resolution: specifies resolution for the values reported for
 *	the axis.
 *
 * Note that input core does not clamp reported values to the
 * [minimum, maximum] limits, such task is left to userspace.
...
 */

(We should change wording of the last sentence to "... does not always
clamp ..." since when we do inversion/swap we do clamp values.)

And note that the userspace that can handle swapped min and max will
work fine if the kernel provides normalized data, but other software
will/may not work.

Thanks.

-- 
Dmitry