At 2:17 PM -0400 4/11/05, Nathan J. Williams wrote:
>There are two things that would be useful:
>
>1. Make trap() dump register state if it's about to panic.
>
>2. Make the ddb backtrace dump trapframe register state when it
>   traverses a trap. This would be useful both in panic situations and
>   otherwise.

Motorola's Macsbug has a command, stdlog, that will open a file, write the
registers, two stack traces, some memory locations, and a couple other
things. While there are going to be problems getting to the file system
from ddb in a trap, a single command that relieves the user from worrying
about the right stuff to send back would help a lot.

>At a higher level, an issue here is that the code was not particularly
>designed with the goal of being accessable or debuggable by
>non-experts - usually, debug and panic messages are written by the
>author of the code to be useful to them, and perhaps secondairly to be
>useful for other people to report back.

Perhaps panic could be shimmed to prepend a standard NetBSD output
instruction manual and then the actual panic message displayed.

tim