X Window System (X11)

The participants in the GNOME documentation hackfest led a great lunchtime talk on Friday, introducing GNOME 3 to about two dozen Senecans.

GNOME 3 embodies a complete re-design of the desktop. Clutter has been replaced with discoverable behaviours, visual cues, and generally streamlined operation. It's a bold experiment that has already attracted some detractors, but it was fascinating and enlightening to hear the environment explained by members of the community that created it. I'm looking forward to using GNOME 3 in the upcoming release of Fedora 15.

There were many who expressed an interest in attending but were unable to do so. Here are a couple of links:

• GNOME 3 site, with features, screenshots, and FAQ: http://www.gnome3.org/
• Main GNOME site: http://www.gnome.org/

As far as I know, there is no way to reliably get a process ID from an X window ID for local clients (to implement Richard's View Source idea). I would love to be wrong!

Questions:

(1) Did I miss something? Can this be done now?

(2) If this can't be done now, what would it take? Could we create an X extension so that the server can supply connection info for a window, and then trace that connection info back to a specific process?

I haven't written about multiseat for a while.

Bad news first: you can't do (multi-card) multiseat on F10 using open-source drivers, and you won't be able to do it in F11 either. David Airlie's work on the Radeon driver is getting close, but there's still a crash-on-exit last time I checked.

Slightly good news: you can sometimes do multi-card multiseat on Nvidia hardware using the proprietary driver on F10/F11, if the phase of the moon is right. I have it working on a dual-PCIe system, but on another dual-PCIe+dual-PCI system (4 seats), I can only get three of the seats active at once.

Better news: I'm hopeful for the F12 timeframe.

More to follow...

I haven't had a clear picture of where the current X drivers stand in terms of multiseat and multicard support. So, after a week of fighting with drivers, video cards, feisty GPU power cables and hard disk drives mounted in such a way that they block PCIe slots, I've compiled this table of the state of current ATI and NVIDIA drivers (excuse the width of the table):

Testing was performed using Fedora 10 and Xorg 1.5.3 on an Asus P5E3 Premium motherboard with an Intel Q6600 processor, using BusID sections in X server configuration files to start a separate X server for each seat, turning off AutoAddDevices, using the evdev driver with device lines for each manually-configured keyboard/mouse input device (/dev/input/eventX), and starting the X servers with --sharevts. Initial tests did not use a display manager; later tests on drivers that passed the first tests used xdm (with mixed success).

At this time, the radeon driver appears to be the Open Source driver closest to being able to run a multiseat, multi-card system. I tried to find out more about the cause of the radeon lockup, but single-stepping with gdb yielded the only successful exit from the X server (!!!), and in every other case the system lockup has prevented me from getting useful information.

Some notes:

• xdm causes SELinux AVC denials in this configuration as packaged in F10 (writing a temp file), and displays a message on the login dialog about the displays being insecure (?).
• Current console policy assumes a single seat. This has an impact on sound use; it probably also affects removable storage, but I did not test this.
• The radeon driver runs glxgears at about 100-200 fps on each of two seats. The nvidia driver runs glxgears at about 600 fps on each of the seats. Oddly, glxgear startup under the nvidia driver complains about /dev/nvidiactl being unwritable, but it runs; however, making /dev/nvidiactl writable prevents glxgears from starting successfully.

Next round: layered (Xephyr-on-X) servers for multiseat on a single card.

I finally got a 2-seat system working reliably on dual ATI R710's (Asus EAH 4550s) this afternoon, using F10. The stability problems I mentioned earlier cleared up when I reduced the set of services running -- the important stuff (dbus, NM, hal) are all still enabled so I need to do some more checking to find the conflict (isdn? bluetooth?).

More tests to follow, but it's encouraging to know that there is at least one Open Source driver that can be used to build a multiseat system with the current X server. I'll blog details in the next day or two.

Two users, two monitors, two mice, two keyboards, two ATI video cards, one computer. Now if it would just stay stable while the mice and keyboards are being used, like in the good old (F8) days...