Passing multi-screen layout to KMS driver

[Michael Thayer]

Hello Gerd,

14.12.2016 11:11, Gerd Hoffmann wrote:
>> So I would be interested to know whether anyone else has thought about
>> this problem, and possibly even about an interface to let the compositor
>> pass the information.  If not, would people be open to the idea?  I
>> would much rather have something generally agreed on than hack something up.
>
> I think the best way to tackle this is to have multiple tablets, one per
> display device (touchscreen-style setup).
>
> qemu can do that, with input routing (must configure on the host which
> display belongs to which tablet).  Here is some info on that:
>
> http://git.qemu-project.org/?p=qemu.git;a=blob;f=docs/multiseat.txt;hb=HEAD
>
> The setup on the guest side is completely manual.  You have to use
> "xinput --map-to-output" to tell Xorg which tablet belongs to which
> display.  Maybe it is also possible to stick that into xorg.conf.
> Should be improved, and it surely makes sense that qemu and virtualbox
> use the same approach here.
>
> Not sure if and how this works automatically with physical touchscreens.
> Any clues are welcome.

Thanks for the answer.  That was the direction I was initially expecting 
to go too.  In theory libinput lets you map input devices to heads[1] 
using a udev property[2], though I have yet to test whether anyone 
supports that yet (couldn't find it in the Mutter/GNOME Shell source, 
but I'm not familiar with it).  As the API description says, the default 
without that property is still to map the input device to all screens 
like X.Org on Linux does.

[1] 
https://cgit.freedesktop.org/wayland/libinput/tree/src/udev-seat.c?id=1.5.3#n97
[2] 
https://wayland.freedesktop.org/libinput/doc/0.99.1/group__device.html#gaf48626f6190e9c9bc14abb704e66cc22


> Another option is to use a guest agent, spice does that since years to
> handle multihead.  The guest agent queries the display layout using
> xrandr, gets x + y + displayid from the spice client and generates
> pointer events from that.  But I expect that scheme breaks with wayland
> because wayland is by design alot more restrictive, so spice-agent
> probably isn't allowed to send pointer events.  So not really an option
> these days ...

We actually do something similar in Windows guests: older versions 
provided the layout information to the driver directly, but at least 
Windows 10 does not, so we query it with a user-space agent which passes 
it to the driver.  (We send our pointer events from a driver, in all 
supported guest types, which works fine with Wayland too.)  It would be 
nice though as I said if the compositor (or whoever is controlling the 
display) could just provide the layout information to the driver itself. 
  We already have "suggested X" and "suggested Y" for the other 
direction, and for now I have solved it by always providing "suggested 
X" and "suggested Y" hints in the driver.  That which works well enough 
in a first approximation - if the user changes the layout inside the 
virtual machine the mapping breaks, and as soon as they change it 
outside it mends again.  So my idea was to try to have people agree on 
on interface for that.

Regards
Michael

> cheers,
>   Gerd
-- 
Michael Thayer | VirtualBox engineer
ORACLE Deutschland B.V. & Co. KG | Werkstr. 24 | D-71384 Weinstadt

ORACLE Deutschland B.V. & Co. KG
Hauptverwaltung: Riesstraße 25, D-80992 München
Registergericht: Amtsgericht München, HRA 95603

Komplementärin: ORACLE Deutschland Verwaltung B.V.
Hertogswetering 163/167, 3543 AS Utrecht, Niederlande Handelsregister 
der Handelskammer Midden-Nederland, Nr. 30143697
Geschäftsführer: Alexander van der Ven, Jan Schultheiss, Val Maher