As of a day or so ago, my Sun-3/150 finished a full build of the world.
That means "make build" in /usr/src, build a kernel, build the
bootblocks, install everything.

In particular, the new SCSI driver ran flawlessly for two or three days
solid.

Erik Fair remarked
> One other item: the make died twice trying to rebuild libc, just as
> before, and I had sup'd and reinstalled "make" before starting this
> whole process, so the bug is still there.

This is almost certainly two things collaborating: (a) a memory leak
somewhere in make and (b) a default 8-megabyte datasize limit.  My
"make build" was done after "unlimit datasize", which boosted the limit
to (as I recall) 32 megs.  And according to ps awwlx, when make was
chewing through libc its VSIZE was well above 9 megs and coming up on
10 megs.

> I also have some kind of libkvm mismatch because vmstat will not run,
> and ps reports funny results in places.

In my experience:

- ps invariably reports only the command name, and that in parentheses;
   "ps ax" output might look like

	  PID TT  STAT      TIME COMMAND
	    0 ??  DLs    0:01.31 (swapper)
	    1 ??  Is     0:00.49 (init)
	    2 ??  DL     0:12.92 (pagedaemon)
	   60 co  Ss     0:05.08 (sh)
	   73 co  R+     0:00.09 (ps)

- dmesg works immediately after boot, but at other times it says
	dmesg: kvm_read: kvm_read: Bad address
   I don't know when the change occurs.

This is despite doing at least two "make build"s; ps, dmesg, libkvm,
kernel, etc, are all up-to-date with respect to one another....

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu