Bill,

>>When that happens, the program in question will either produce incorrect
>>output or crash.  But the system will keep up and running.
> 
> As noted, /sbin/init can cause issues. Another problem may be if libc is 
> updated to use a new call that isn't there; ALL programs can crash. I 
> remember when we went to having getcwd() based on a system call. If you 
> used a new userland (that expected the system call) with an old kernel 
> (that didn't have it), most every program crashed. I don't know if init 
> crashed then, but I think every shell did. :-)

I was specifically thinking about that.
Anyway, related question : how to debug the kernel when the failure 
consists in unescapable reboots after the "acpibat" message ?

If anybody has a clue, I'll be glad to catch it ;)

Vincent