Subject: hp300 port install problems / fixes / workarounds
To: None <Herb.Peyerl@sidney.novatel.ca>
From: Havard Eidnes <Havard.Eidnes@runit.sintef.no>
Date: 09/25/1994 17:34:10
I recently installed the HP-300 port on a HP-9000/425 we have. Luckily
before doing this I read the port-hp300 archive, and saw that on 10 Aug
Scott Reynolds wrote a message asking whether the root (a) and swap (b)
partitions in the label on the distributed rootimage.gz overlaps (they do).
This probably saved me a lot of frustration when doing this installation.
To overcome this problem the installation routine I used was a little
different from that described in the INSTALL file, here is the procedure I
When booting the newly installed root image, I replied with "/sdnetbsd -s"
to the bootsrap loader. Apparently, this causes netbsd not to open the
swap partition, so that it's offset and size can be adjusted (I tried
without doing that at first, but it failed with a sensible error message).
Then, when netbsd has booted up, you immediately need to re-mount the root
file system, because without a writable file system you won't be able to
"disklabel -e" (ed won't be able to create /tmp/xxx):
# mount -u /dev/sd0a /
Then enable writing the disklabel:
# disklabel -W sd0
Then set the editor to "ed" since the default editor (vi) isn't installed
at this point:
# EDITOR=/bin/ed; export EDITOR
Now edit the disk label:
# disklabel -e sd0
Luckily my disk was the same type as what was distributed in the root image
(a Rodime-3000T (?) 200MB SCSI disk), so I only had to modify the
size/offsets of the partitions b (offset), e (offset) and g (offset +
size). I say "luckily" because either the SCSI code in hp-NetBSD is doing
something wrong when calculating the total number of blocks on the disk to
report during autoconfig or the disk drive itself is returning garbage:
sd0: HP C2235-300 rev 0B16 (SCSI-2), 825012 512 byte blocks
The other small problem with the distribution (pointed out before on this
list, I think) is that the usr.share archive in the distribution is
corrupted, and I had to bring over the /usr/share tree from my i486 box
When upgrading this box to NetBSD 1.0_Beta from the July snapshot I noticed
that I first had to install the new assembler, because some of the
assembler soruce files in the fpsp library now contain some "long" integer
constants. To complete the installation I also had to install a new ld.so;
without it xargs gave me problems (like "xargs: chmod: Unknown error:
67453686." -- execvp was failing). I'm not sure at what stage in the
upgrade proces it is safe to install ld.so, though, I did it first when
xargs gave me problems during the "make install" from /usr/src.
The most serious problem I have is that whenever i do "reboot", the system
will not come up properly -- after printing
real mem = 33546240
during the autoconfig phase of the rebooted kernel, it panics with
"pmap_enter_ptpage: PT page not entered" and it drops into the debugger
with a frozen keyboard. (This particular bug has been reported with
send-pr, BTW.) Thus, I have to be physically present at the machine to
reboot it (pressing the "reset" button is necessary).
Other than that I must say that I am quite impressed by how easy this
really went -- I am now up and running a complete NetBSD 1.0_Beta system
with X11R5 installed.