Alex, my experiments show that the problem is caused by wifi and bluetooth initialization in rfkill. I've noticed that the problem can be reproduced with these turned on (uncommented). I tried enabling/disabling wifi and bluetooth NOT simultaneously, one is on and other is off. After reboot I could get kernel crash almost 100%. However, when they're turned on or off at the same time, booting is stable in most cases. As for me, it seems like some kind of race condition. Do rfkills have something in common with interrupt handling? 2012/9/18 Carsten Otto