I have two amd64 computers on OpenBSD 4.3. Both had uvm fault problems
today. There were panics with a message:
kernel diagnostic assertion "anon->an_page == NULL" failed: file
"/usr/src/sys/uvm/uvm_anon.c", line 169
I did trace and ps in ddb, but another crash before savecore could
capture the result of "boot dump" lost the crash dump, and the
results of those commands. I changed the sysctl for ddb:panic, but
the expected automatic unattended reboot with dump and savecore did
not happen, hands-on attention was required at every crash. I had
to do manual fsck of /usr each time. I would think /usr is usually
read-only, so there should be filesystem inconsistencies requiring
a manual fsck. Some file in ./src/usr.sbin had confusion, one
message said. Not anywhere there s.hould have been any activity.
So I searched google and openbsd mailing list archives. It seems
netbsd and openbsd have had nebulous uvm fault problems for years.
A thread from June 2008 inferred that a series of almost reproducible
crashes seems to have been solved by some patch, on netbsd.
These computers have the 3-core AMD CPUs. I had been running on the
uni-processor /bsd. I switched everything but wd0 today after several
crashes on one (case, power supply, cables, NICs, display adapter, yes,
everything but wd0). That alternate hardware crashed on every attempt
to boot until I booted single-user and did
mv /bsd /bsd.up
mv /bsd.mp /bsd
and rebooted. And it has been running since, but that's only about
3 hours so far. These are the amd64 4.3 GENERIC kernels, vintage
March 2008.
bsd.up:
OpenBSD 4.3 (GENERIC) #1368: Wed Mar 12 11:05:31 MDT 2008
bsd.mp:
OpenBSD 4.3 (GENERIC.MP) #1582: Wed Mar 12 11:16:45 MDT 2008
Does anybody have a handle on the uvm fault crashes? The bug reports
indicate it is very difficult to get a dump through savecore on these,
and there don't seem to be any closing of the bug reports, though many
independent ones seem to have been filed. Maybe it's several independent
problems and it's foolish of me to think they're related.
