Greetings,

Presently, UVM coredumps seem to work thus:

If I create a program that mmaps some IO space for example part of
/dev/mem, when the program core dumps then the mmap()ed region will get
core dumped too.

Now, on one architecture I have here, doing word reads from IO regions
will cause a TAE (a PowerPC machine check type). While I might access it
through my mmap()ed hole one byte at a time, the core dump code reads it
a word at a time and knocks out the computer. Clearly there's a
portability problem here.

Q: Is core dumping of any device really that useful anyway, most of the
time?

	Neil