[ros-kernel] Kernel crash: Timing dependent error on startup

Mike Nordell tamlin at algonet.se
Fri Apr 30 22:25:48 CEST 2004


If you've got a "too fast" machine, the kernel will crash on startup. This
has been observed by at least one user on IRC testing ROS on a 2GHz machine.

The fix? line 781/783 of ntoskrnl\ke\main.c.

There's an assumption that the "initial process" will always take at least
five seconds. Clearly that's wrong, especially if we look at
LdrLoadInitialProcess in ntoskrnl\ldr\init.c.

Since we now require at least Pentium, I'd say either scrap that timeout and
look at the TSC from the CPU, or at least lower the timeout to something
reasonable. 50ms (even that it too seems a bit large)?

To be precise: During installation from a freshly burnt CD the user got:
Bug Detected (code 71 param 1 0 0 0)
SESSION5_INITIALIZATION_FAILED

Lowering the timeout to 2 seconds worked for this user, but considering 5
seconds is too much for a 2GHz machine, 2 seconds would be too much for the
top-of-the-line machine in just a couple of years.

As I'd first like some discussion on this - personally I think it's insanity
to expect something CPU bound to take at least a specific amount of time
when we know hardware is only getting faster and faster - why I didn't
commit anything.

/Mike



More information about the Ros-kernel mailing list