A3000 and UVM

[David Hopper <dhop%globalesi.com@localhost>]

Hi all,

I've got a critical problem (or two unrelated problems?) right now with 1.4.1 on
my A3000 that I hope I can get some feedback on.  Problem One:  I've been seeing
UVM faults repeatedly upon execution of the startup scripts while using two
memory segments: 32 megs onboard a Warp Engine 4040, and 16 megs of static zips
on the motherboard.  NetBSD ran fine while using only the onboard 32 megs,
enough to configure the system, compile and install KDE, and run for three
days.  Once I enabled multiple segments via loadbsd -a -n2, I couldn't get to
the login prompt.  UVM faults occurred during execution of the first few
binaries.  The kernel on the root partition did match the loadbsd image; I even
used BFFS to loadbsd from the root partition directly.

Problem Two:  In trying to isolate the problem (and looking forward to
installing a new toy), I upgraded the whole kit to a Cyberstorm PPC 233/'060,
and added an additional 32 megs.  I had compiled and installed an '060 and
Cyberstorm SCSI kernel beforehand, but I saw uvm_faults even more frequently,
with getty, inetd, and others coring into the root.

I then started from scratch on the Cyberstorm with Ignatios' installation
kernel, and it fails at random points in the installation script with either a
segmentation fault and a lockup, or a uvm_fault dump.  This is all on one single
memory segment.  I'm using loadbsd -b netbsd with the installation kernel.  The
kernel boots fine (just like my earlier Warp Engine problem), it's only in
userland that the system blows up.

I have a couple of questions before I submit a really confused PR:  is anyone
out there running 1.4.1 using either multiple segments and the A3000, or a Phase
5 accelerator and the A3000?  Just wondering if I'm on an island, here.
Secondly, what should I be looking for?  What small memory tests I've run on the
Amiga side check out, although I won't rule out bad zips.  FWIW, the Amiga side
has been up for two days raytracing, but I'm sure the memory boundary hasn't
reached the lower priority zips.  This doesn't help the Phase 5 board UVM
problems, though, as that was only on the SIMM memory block.

Thanks much.  Whatever help can be offered will be graciously accepted:  if I
can't run NetBSD on my 68060, that means I have to run Linux APUS on the
PowerPC.  I don't want to do have to do that.  Really.  :^(

Dave Hopper
Global Event Services, Inc.
http://www.globalesi.com