>>   How about even only reserving space in the trapframe for those registers
>> when DDB is defined as well?  Then you don't have the stack space taken up
>> as well when DDB isn't in the kernel.
>
> I guess... tho debugging crash dumps might be a bit more difficult that
> way (e.g. backtrace code would have to know different trapframe sizes).

Having kdbaux variable is shame, although it was me who introduced it
(I changed twice trapframe design on kernel stack).  When I started to
change trapframe design on kernel stack, backtrace code was severely
handicappped.  I reordered code sequence and made it more useful
rather than showing improper values.  I believe it's time to redo all
of DDB stuff because it could gather attentention again. 

Tohru Nishimura