Re: upgrading from netbsd 4/xen 3.1 on i386 to HEAD on amd64

["Chris Brookes" <cbrookes%gmail.com@localhost>]

On 04/03/2008, Christoph Egger <Christoph_Egger%gmx.de@localhost> wrote:
> On Monday 03 March 2008 22:36:42 Chris Brookes wrote:
> > I'm currently running NetBSD 4/Xen 3.1 on a 1.0Ghz i386 box for DOM0.
> > I have a few NetBSD 4 DOMU's  and a single Linux (Ubuntu 7.10) DOMU.
> > I've got an Athlon X2 BE-2350 and a GIGABYTE GA-MA69VM-S2 mobo on the
> > way which I'm hoping to run the NetBSD HEAD/current amd64 port on for
> > Dom-0. Will I be able to boot and run my existing PV 32bit DOMU NetBSD
> > and Linux installations while running under the amd64 DOM0?  If so
> > what kernels do I use, my existing ones? This will greatly aide my
> > upgrade path to 64 bits for the DOMUs...
>
> Basically yes. Just exchange your 32bit kernels with 32bit PAE kernels and
> that's it. No changes in the userland are needed.


And it works, mostly. The ride hasn't been smooth though. Apart from
sysinst refusing to work with me on a partition greater than 1TB,
which I worked around, I still have a few lingering issues only a
couple of which are probably relevant to this list but I'll mention
the others in the hopes of some direction...

1) Using 
http://ftp.netbsd.org/pub/NetBSD-daily/HEAD/200803080000Z/i386/binary/kernel/netbsd-INSTALL_XEN3PAE_DOMU.gz
I got a core when I run nslookup or dig.

link# nslookup
[1]   Bad system call (core dumped) nslookup
link# Mar 11 09:59:57 link /netbsd: pid 1363 (nslookup), uid 0: exited on signal

An example is at http://txrx.org/nslookup.core

2) On the Dom-0 using a kernel compiled from source at
http://ftp.netbsd.org/pub/NetBSD-daily/HEAD/200803080000Z/source/sets/syssrc.tgz
I couldn't get the re0 nic to work properly with my vlan setup.

re0 at pci2 dev 15 function 0: RealTek 8169SC/8110SC Single-chip
Gigabit Ethernet (rev. 0x10)
re0: interrupting at ioapic0 pin 23, event channel 13
re0: Ethernet address 00:1a:4d:ff:e4:72
re0: using 256 tx descriptors
rgephy0 at re0 phy 7: RTL8169S/8110S/8211 1000BASE-T media interface, rev. 2

When the interface was brought up with a regular IP it worked without
issue. When I brought up re0 without an IP and instead configured a
vlan interface with the other end connected to a 801.1q trunk port on
my switch, things started to go awry. I could source traffic from the
dom-0 and get as far as installing packages over the network seemingly
without problems, but strangely inbound traffic wasn't working
properly. An inbound telnet on port 22 to the dom0 would produce the
SSH banner, but any subsequent traffic would vanish. Looking at a
trace in wireshark it showed a sucessful tcp connection setup
(obviously since I saw the banner) but after the banner there'd be a
couple of retransmissions and then nothing. I plugged in an old  3c905
and using the ex driver everything's fine, so I'm relatively confident
it's a re drive issue. Which would be the best list to report to with
this, or should I file a PR, or both?

3) I have a parition in the dom0 (/dev/ld1a) that is exported to a
domu with 'phy:/dev/ld1a,wd2d,w' and is then exported to another
(linux) domu with 'phy:/dev/ld1a,sda3,r'. Prior to the upgrade, I had
this working for a couple of months with (I think) xentools3.1.2. With
my new setup (xentols3.1.3 ) this simply doesn't work.  The first
dom-u mounts with no problem and then the second one prints time out
message from XENBUS. I haven't tried the xentools 3.1.2 to see if that
is the problem, I just mention the versions because (aside from the
dom-0 architecture change!) that's the only thing that's different...

Any thoughts appreciated.

Chris