Review Request 112294: Implement multi-seat support in KDM

[Stefan Brüns]

> On March 28, 2014, 10:59 a.m., Oswald Buddenhagen wrote:
> > kdm/backend/server.c, line 79
> > <https://git.reviewboard.kde.org/r/112294/diff/2/?file=186613#file186613line79>
> >
> >     why the redundant supply of the seat as a layout?
> 
> Stefan Brüns wrote:
>     There are no config matches taking the "seat" into account, so the only possibility to apply a specific config is to use "layout".
> 
> Oswald Buddenhagen wrote:
>     something is wrong about this.
>     what does the -seat option do in the first place? if it is sufficient to actually launch the server correctly, then that's all that should be done - the user can manually specifiy ServerArgs if he needs to configure the layout for some reason. otherwise every user is forced to actually have a "seatN" layout in his config.
>     also, cannot the layout determine the seat? seems a bit stupid that these are entirely orthogonal.

The seat option is used for matching devices from udev, i.e. input devices, GPUs, ... It specifies *which* devices to use, but now *how to configure* these.

Unfortunately the config files lack a "MatchSeat" option, so the only possible options are either specifying a config file (or directory) or a layout. There is a patch from Oleg Samarin floating around which adds MatchSeat, but this is not upstream so currently no option.

If a layout is specified but not available, the default layout is chosen, so no problem here.

ServerArgs is no alternative as it is specific to a display, but the seats will use a more or less random display, dependent on when they become available (much like Reserve servers, which take the first display available).


> On March 28, 2014, 10:59 a.m., Oswald Buddenhagen wrote:
> > kdm/backend/dm.c, line 1397
> > <https://git.reviewboard.kde.org/r/112294/diff/2/?file=186612#file186612line1397>
> >
> >     that seems questionable to me. why are you re-defining the display to be permanent? when the seat goes away, kdm won't know what to do with it.
> 
> Stefan Brüns wrote:
>     When a seat goes away, rStopDisplay(d, DS_RESERVE) is called - maybe a "d->displayType = dLocal | dReserve" is missing for these cases.
>     But as long as the seat exists, you want it to behave like a static display (restart server after session exit), so dPermanent is IMHO correct.
> 
> Oswald Buddenhagen wrote:
>     i see what you mean ...
>     still, it seems wrong to change the display type just so. this is supposed to be a constant. this is of course a consequence of abusing reserve displays in the first place - technically, there should be a dDynamic display type. if having reserve displays on dynamic seats is technically possible at all, this would also need some thought.

So, what approach to take here:
1. Its somewhat ugly, but it works, so use it.
2. Introduce a dDynamic display type, and make it behave like dPermanent.
  a) add a DynamicServers option for this
  b) create list of displays on the fly.

Reserve displays on dynamic seats is not possible currently, you cannot change the VT. For this to work you need support for systemd-login, but this is more serious surgery.


- Stefan


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/112294/#review54418
-----------------------------------------------------------


On March 29, 2014, 5:38 p.m., Stefan Brüns wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/112294/
> -----------------------------------------------------------
> 
> (Updated March 29, 2014, 5:38 p.m.)
> 
> 
> Review request for kde-workspace and Oswald Buddenhagen.
> 
> 
> Repository: kde-workspace
> 
> 
> Description
> -------
> 
> This patch implements dynamic multiseat in KDM. It follows the description in:
> http://www.freedesktop.org/wiki/Software/systemd/writing-display-managers/
> 
> In case systemd is no found at compile time, nothing changes. If logind is not running, nothing changes. If no additional seats have been configured (some Plugable USB-GPUs are automatically added as additional seats), nothing changes.
> 
> In case there are additional seats beyond seat0, a reserved display is promoted to a local static one (and demoted if the seat is removed) and a new X-Server/greeter is spawned.
> 
> The code has been tested extensively, with a combination of [Radeon dedicated GPU|Intel iGPU], [Intel iGPU|Displaylink USB GPU] and others. For history of this patch, see https://bugzilla.redhat.com/show_bug.cgi?id=884271 and https://bugzilla.redhat.com/show_bug.cgi?id=975079
> 
> 
> Diffs
> -----
> 
>   CMakeLists.txt d5c76d8 
>   cmake/modules/CMakeLists.txt 117b3a5 
>   cmake/modules/FindSystemd.cmake PRE-CREATION 
>   kdm/backend/CMakeLists.txt 25f383f 
>   kdm/backend/client.c 26bb0b4 
>   kdm/backend/dm.h b2f8c61 
>   kdm/backend/dm.c 77a2ef7 
>   kdm/backend/server.c d8dd6f3 
>   kdm/backend/session.c 0e7901c 
> 
> Diff: https://git.reviewboard.kde.org/r/112294/diff/
> 
> 
> Testing
> -------
> 
> Single seat system, several multiseat systems
> 
> 
> Thanks,
> 
> Stefan Brüns
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-core-devel/attachments/20140330/a8bf64ad/attachment.htm>