[PATCH libinput 1/3] udev: Add name-based input device detection without dmi

[Peter Hutterer]

On Mon, Apr 24, 2017 at 03:31:39PM +0200, Paul Kocialkowski wrote:
> Le lundi 24 avril 2017 à 16:11 +1000, Peter Hutterer a écrit :
> > On Fri, Apr 21, 2017 at 01:31:56PM +0200, Paul Kocialkowski wrote:
> > > Some devices do not use dmi at all (this is the case on most non-x86
> > > platforms, such as arm and arm64) but should able to select specific
> > > quirks based on the input device name too.
> > > 
> > > This adds name-based input device detection without dmi to support
> > > these devices.
> > > 
> > > Signed-off-by: Paul Kocialkowski <contact at paulk.fr>
> > > ---
> > >  udev/90-libinput-model-quirks.rules.in | 4 ++++
> > >  1 file changed, 4 insertions(+)
> > > 
> > > diff --git a/udev/90-libinput-model-quirks.rules.in b/udev/90-libinput-
> > > model-quirks.rules.in
> > > index 8bff192..149ef0d 100644
> > > --- a/udev/90-libinput-model-quirks.rules.in
> > > +++ b/udev/90-libinput-model-quirks.rules.in
> > > @@ -37,4 +37,8 @@ ENV{ID_INPUT_MOUSE}=="1", \
> > >  KERNELS=="input*", \
> > >    IMPORT{builtin}="hwdb
> > > 'libinput:name:$attr{name}:$attr{[dmi/id]modalias}'"
> > >  
> > > +# libinput:name:<name>
> > > +KERNELS=="input*", \
> > > +  IMPORT{builtin}="hwdb 'libinput:name:$attr{name}'"
> > 
> > is there something else we can hook onto? the i2c bus maybe? I'm a bit
> > worried for these generic ones to start leaking into other devices.
> 
> I don't think the i2c bus would make any relevant difference here.
> 
> On the other hand, the device-tree model string (/proc/device-tree/model) should
> properly (and uniquely) identify end devices. However, I don't think there's a
> way to bind this to the attr udev event format, although it would be relevant to
> have rules depending on the device-tree format (not only for libinput).
> 
> Also, note that the name+dmi rules will have priority over the name-only one
> that I'm introducing. So the problem will likely only be between non-dmi devices
> that have the same driver name but need separate quirks. There are also already
> many rules where the dmi string is ignored (which is equivalent to what I'm
> introducing), only that this doesn't work on ARM because the dmi nodes are
> missing.
> 
> To solve this, I think the only relevant course of action would be to add
> device-tree model support in systemd/udev and then use that in libinput.
> 
> What do you think, should I prepare and submit such changes?

yeah, having this available would help a lot here, even if it ends up just
separating the device tree bits from the normal bits. I'm not overly happy
with the hwdb matching as it is, it behaves a bit unexpected at times and I
wouldn't be suprised if your current hwdb has some bugs where things don't
get applied correctly. One more reason to not over-expand the match rules.

Cheers,
   Peter