Hi Daniel,
On Tuesday 06 March 2007, Daniel Vandersee wrote:
I noticed that the Fiasco kernel hangs under some circumstances so the whole system stops working. Unfortunately I wasn't even able to enter the kernel debugger via a serial cable in this case. The error is reproducible on different plattforms (e.g. Thinkpad T43, 1GB RAM, Pentium M 1.8 GHz or Thinkpad T41) even though it only appears from time to time (probability ~50% or even better). Within VMWare or Qemu this deadlock does *not* appear. Furthermore this error does not occur if the kernel was started using the "-esc" option. I wasn't able to reproduce this bug on a HP nx7400 (Dual Core T2300, 1GB RAM). We use an own dynamic loader to load some L4 tasks whereas other L4 tasks, as the mGUI, are loaded by GRUB.
The bug can be reproduced on real hardware using the iso-image which I will upload to the SLOX server (deadlock.iso). You have to choose GRUB entry two or three. If the error occurs the system hangs in text mode otherwise a graphical mode is entered. Sometimes you have to reboot up to 20 times but it usually occurs immediately.
To debug such issues the builtin watchdog is very helpful. Make sure that CONFIG_WATCHDOG is enabled in the kernel configuration and pass the -watchdog parameter at the kernel command line. A builtin Local APIC is required. If the kernel deadlocks for some reason and the interrupts are disabled, the watchdog will force the CPU to enter the Fiasco kernel debugger.
Kind regards,
Frank