Alan Stern wrote:
[snip]
> My impression is that this problem arises somewhere within or
> below the free_irq routine.  I don't have the -mm2 sources, so I
> can't be any more precise than that.

Here is an updated dmesg for kernel 2.6.9-rc4-mm1. But I'm afraid it 
won't give more information, as the call stack is identical to the 
2.6.9-rc3-mm2 one.

I will try a vanilla kernel if it's needed.

-- 
laurent