Bug#339790: Automatic VT allocation is unreliable

[Linas Žvirblis]

Package: gdm
Version: 2.8.0.6-1

The default automatic VT allocation (VTAllocation option in 
/etc/gdm/gdm.conf) is unreliable. It assigns all X servers VT8. This 
leads to a hang when a second X server is started via gdmflexiserver.

To reproduce:
1. Start GDM via "/etc/init.d/gdm".
2. Log in as normal user.
3. Start gdmflexiserver.

The newly started X server is also assigned VT8. It becomes highly 
unresponsive and usually hangs after a while. It is usually possible to 
get back to a usable system by killing all instances of X server and GDM 
via root console, but sometimes screen just turns black and stays that 
way until a reboot. No, the system does not crash. I can reboot via 
CTRL+ALT+DEL and I can hear MOC playing music.

The fix is to simply set VTAllocation=false by default in 
/etc/gdm/gdm.conf. I did not dig deep into this but a comment in 
gdm.conf saying...

Automatic VT allocation.  Right now only works on Linux.  This way
we force X to use specific vts.  turn VTAllocation to false if this
is causing problems.

...does suggest that this feature might not be reliable.

My system:
Up to date Debian Sid (except binutils held at 2.16.1-2);
Vanilla kernel 2.6.14.2 I compiled myself;
Binary NVIDIA drivers 1.0-7676 from nvidia.com;