mycroft> Yes.  Currently that works, but at the cost of DDB being *SEVERELY*
    mycroft> broken, such that a user program can caused a trap into DDB.

Is that what's happening to me?  I am trying to debug fsck with gdb,
because I have a file system that causes fsck to dump core, but instead
of trapping into gdb on either a breakpoint or a segv, I trap into
ddb (for the trace trap) or end up dereferencing zero when I try a
backtrace.

I'm building a kernel with no ddb to see if this will work better.