version=3.0.3
Sender: port-hp700-owner@NetBSD.org

Greetings all,

I finally got a chance to look at getting my 715/64 booting NetBSD,
and I figured I would share the results.  Since it was included in the
NetBSD 3.0 release, I first tried to use 3.0 following the release
notes, but I soon found that the instructions are perhaps a bit out of
date, and the release is lacking the SYSNBSD LIF file which is
mentioned for netbooting.  At that point, I figured I would just
include a hp700 build when I was building my 3.0 sandbox release from
scratch, and so I did not proceed much beyond this initial stage.

When building from source, my hp700 build proceeded nicely (just as I
expected), and I soon had a tree which I could use to produce my
diskless boot area.  Rather than unpacking a few set tarballs into the
root directory, I did the following:

0) I produced a SYSNBSD, which is not built by build.sh while building
   a release, using the steps listed in the hp700 section of the
   diskless booting how-to.  This was then copied to my tftpboot area,
   and my dhcpd.conf was changed to reflect this.

   NOTE: Neither this nor any other bootable media/file appears to be
   built via build.sh when building a release, contrary to what the
   news from 2005 May 18 seems to imply.

1) Copied the destdir from the build to my diskless filesystem area
   for a shared master area.  This is /u2/diskless/hp700.

2) Copied the generic kernel into /u2/diskless/hp700

3) Modified files in /u2/diskless/hp700/etc such as resolv.conf which
   would be identical across all clients.  At this point, I created
   mount points such as ./swap, setup accounts/groups, NIS, etc., and
   my fstab, which is:

	pell:/u2/diskless/mariner/swap  none	swap	sw,nfsmntpt=/swap
	pell:/u2/diskless/mariner/root  /	nfs	rw		0 0
	pell:/u2/diskless/hp700/usr	/usr	nfs	rw		0 0

4) Made a client-specific directory, /u2/diskless/mariner which
   contained a root subdirectory and a swap file, created as mentioned
   in both the release notes and the diskless how-to.

5) Unpacked base.tgz, but not etc.tgz into the client root tree.  

6) From the shared tree, I did:

       ls -1 | grep -v usr | pax -r -w -pe /u2/diskless/mariner/root

   The reason for doing 5 and then 6 is that it contains some stuff in
   /usr which appears to be needed before /usr was mounted (I did not
   note what, bad me!), but this combination gets all the other files
   which are not in base.tgz, including those from sets like xbase,
   xetc, etc.

7) Did a MAKEDEV in /dev

8) Made a few last client-specific changes to files under the
   directory /u2/diskless/mariner/root, such as setting the hostname,
   etc. in rc.conf.

9) Booted the my 715/64, which proceeded to generate its sshd host
   keys, etc.

All of this yielded a client root tree of around 94MB and shared tree
of about 430MB in size.  

The machine has been up for almost 4 days now, with me rebooting once
since the initial set of reboots to tweek some configuration
parameters.  The latest dmesg.boot is attached, and here are some
observations:

a) The system does not update the HW clock.  Probably a known TODO,
   but I get the following message when rebooting (which is no big deal):

    WARNING: clock lost 13016 days -- CHECK AND RESET THE DATE!

b) While talking about time, after awhile, ntpd starts reporting the
   following message:

	frequency error 512 PPM exceeds tolerance 500 PPM

   However, ntpd has still been staying pretty much synced with the 3
   servers which gateway NTP into my network, but the offset is just
   over 300 instead of well below 1.0 like on the dozen other machines
   I have.  It may just be this box... 

c) I have seen core dumps from newsyslog, sendmail, egrep and sshd.
   The first two may just be that I have not fully configured this
   system, though the same sendmail config files work on all my other
   machines to submit mail to my mail hub.  And perhaps newsyslog does
   not like the fact that there are no local log files. ;)  As for
   sshd and egrep, I had sshd crash on me after about a day and
   required me to restart, and egrep failed in the following command
   line, which I was using to watch my pkgsrc build for my 3.0/i386
   rdist repository.   The failing command was

	tail -f +1 mk_pkgs.log | egrep --line-buffered '^(\+\+\+)>' | grep SUCC

   and this was being done in an area mounted by amd from my build
   server. 

   I will probably be looking at these core files sometime in the next
   week if I get a chance.

d) I now find myself wishing that we had X11. ;)  But I realize there
   is still some other work to do, such as getting the HW clock
   updating. ;)

e) At some point, we probably should think about updating the FAQs,
   todos, release notes, etc.

Well, I need to get back to doing work.  But I would say we could
probably move the 715/64 from "make work" to "works".

- Doug

-- 
Douglas Wade Needham - KA8ZRT        UN*X Consultant & UW/BSD kernel programmer
Email:  cinnion @ ka8zrt . com       http://cinnion.ka8zrt.com
Disclaimer: My opinions are my own.  Since I don't want them, why
            should my employer, or anybody else for that matter!