Sigh.

I had a severe version of this same problem when you first reported it, a
bit over a month ago, and "solved" it (so I thought) by recompiling with
even more -current sources than what I was running.

Alas, the hangs are still happening to me on my SPARCclassic (48MB),
running a stripped down (sun4m only, DIAGNOSTIC, but no DDB or DEBUG)
kernel with the standard MACH VM. My beast is headless, with a serial
console, and I very rarely use that console. I telnet in from the same
Ethernet. My home directory is served up via NFS from a SPARC LX running
NetBSD 1.2G (DIGITAL) #7: Wed Jun 18 13:56:59 PDT 1997 (old, I know, but
it's completely stable) over a private Ethernet on Sun X1053A 501-2015
FSBE/S cards (one in the Classic, one in the LX).

The hang is quite complete - no response to BREAK on the console (or any
character echo at all) - gotta power cycle the beastie to get it back.

At the time that I had these problems most severely, it seemed to be an
interaction between disk activity and network activity - do them both
heavily at the same time, and you'd get death, quick. I based this in part
on the observation that if I FTP'd kernel sources to local disk, I got
death, but if I FTP'd into suitably enlarged /tmp MFS, I won. I would also
get hangs during "large" sup downloads (getting the new gcc stuff down was
a pain). This could be a coincidence.

I still get the hangs. They are not as frequent as before, but they sure
are annoying. Since I've been using the thing more heavily lately, it's
been about once a day. Oddly, full world builds on local disk have no
problems.

Anyone else running current on a classic or other sun4m having this problem?

	Erik <fair@clock.org>