preempt cause series of oopses in 2.5.22 and 2.5.22-dj1

preempt cause series of oopses in 2.5.22 and 2.5.22-dj1

[Helge Hafting]

Booting a UP kernel compiled with preempt
(2.5.22, 2.5.22-dj1 and several earlier ones)
results in a series of oopses when running boot scripts, typically
when xdm is about to start.  I can start up single-
user and run X though, I haven't figured out exactly where
it goes wrong.

The oopses scrolling by all starts with
init[454] exited with preempt-count xx
where xx increase by 3 for each oops.  
The call trace is longer each time.
The scrolling stops when this counter reach 63 or 66, 
probably running out of some resource such as kernel stack.

I haven't written down the last oops assuming it is the first
one that is interesting, but I can reproduce and write
down details if someone thinks it makes a difference.

The kernels are compiled with preempt, vesafb, devfs.
Recompiling without preempt gives me a kernel
that works fine.

This preempt problem seems to have existed at least
since 2.5.12-dj1, it took some time before I figured
out it was a preempt problem.

Helge Hafting

Re: preempt cause series of oopses in 2.5.22 and 2.5.22-dj1

[Dave Jones]

On Wed, Jun 19, 2002 at 12:47:24PM +0200, Helge Hafting wrote:
 > The oopses scrolling by all starts with
 > init[454] exited with preempt-count xx
 > where xx increase by 3 for each oops.  
 > The call trace is longer each time.
 > The scrolling stops when this counter reach 63 or 66, 
 > probably running out of some resource such as kernel stack.

As I wrote to you offlist a week or two ago, I've tried tracking
this down, but found no real fix. I've a box here that doesn't
like preempt at all, but it's only my tree that the oops seems
to show up in. 

Robert Love and myself stared at the (minimal) backtrace that
it caused a week or two ago and got nowhere. The only solution
I found so far was to disable preempt on that box.

If you can rig this box up to a serial terminal and capture
the first oops and decode it, it'll be interesting to see if
we're hitting the same bug (Which I suspect we are, the symptoms
sound identical).

I'll investigate some more when I get time.

        Dave

-- 
| Dave Jones.        http://www.codemonkey.org.uk
| SuSE Labs