I was under the impression that the the crash dump was generated some number
   of bytes past the beginning of the swap area; that may actually be true, but
   I'm not sure where "dumplo" is defined.  Do you think any paging is done before
   savecore gets called, though?

from what i remember of last night, dumplo gets set to 8 (i think
it arch/sparc/machdep.c, dumpconf(), but i disremember and aren't
near my source right now).  `pstat -s' run when the system comes
up in single user says that sd0b is using 8 blocks.  looks fairly
sensible to me.
   
   I see that you're working on a sparc; I'm interested in getting this working on
   the i386.  From looking at i386/isa/wd.c, I get the impression that the
   pmap_enter() routine to translate the kernel address in sddump() won't work on
   the i386; is that true?  (I'll admit to being pretty ignorant of the VM system
   in general).  Maybe the reason your core was bad was because the data you wrote
   was bogus?

i wouldn't have a clue about the i386 -- i avoid it like the
plague ;-)

i think the problem is what deraadt/pk/cgd said -- the dump format
isn't understood properly or something.  maybe tonight i'll look at
it again.  maybe not.  (it's seinfeld and northern exposure night,
my 1.5 hours/week of tv ;-)