Subject: Re: installation failure of 2.0F on netra t1-105
To: Geoff Adams <gadams@avernus.com>
From: Kurt J. Lidl <lidl@pix.net>
List: port-sparc64
Date: 06/25/2004 14:06:05
On Thu, Jun 24, 2004 at 06:46:20PM -0400, Geoff Adams wrote:
> On 22 Jun 2004, at 11:29 AM, Kurt J. Lidl wrote:
> > In particular, the command:
> >
> > # disklabel -w -r -f /tmp/disktab sd0 'dorkfs'
> > write: No such file or directory
> >
> > (where 'dorkfs' is what I told it to name the disk)
> >
> > If I try just:
> >
> > # disklabel -w -f /tmp/disktab sd0 'dorkfs'
> >
> > I don't get an error, but the label doesn't seem to get written
> > to the disk (nor, should it be, on my reading of the manual page).
> 
> Hi, Kurt.
> 
> I can't say I understand what's going wrong here. I disklabel my disks 
> all the time on my sparc64s, and I've never had problems like this. 
> However, I don't use sysinst (I build to the target machine's disk), so 
> I'll just offer some things to try.
> 
> First, I've never quite understood the -r option. I mean, I understand 
> what the docs *say* it's there for, but it's often had the opposite 
> apparent effect. In the end, I never use it. And it always works 
> without it. In particular, the following sequence has always gotten me 
> up and running in the face of otherwise impassable trouble:
> 
>      dd if=/dev/zero of=/dev/rsd0c count=1
>      disklabe -e sd0

Well, that doesn't work either:

# dd if=/dev/zero of=/dev/rsd0c count=1 bs=8k
dd: /dev/rsd0c: end of device
1+0 records in
0+0 records out
0 bytes transferred in 0.009 secs (0 bytes/sec)
# echo $?
1

> I always also use the -e option (to bring up $EDITOR), rather than use 
> a file with the label contents in it. Out of habit. That may or may not 
> be related to your difficulty.
> 
> Note the lack of '-r'. Also note that this works for me, despite that.

That didn't work either.

> One last thing you might try is writing the disklabel in Solaris. The 
> default disklabel format on Sparc hardware is the Sun PROM-compatible 
> disklabel format, so that you can boot from the disk. Since the label 
> format is identical, there's no real need to write it from NetBSD. Just 
> convenience.

Well, I tried this too.

It would appear that sysinst, even if you tell it to use the existing
partition tables, wants to go ahead and scribble a new disklabel on
disk anyways, which fails, and then I don't have the option of continuing.

> I understand that these suggestions do nothing to address the actual 
> problem. But they might get you up and running.

Thanks for the suggestions, but I'm still at a loss to move beyond this
fundemental roadblock.  I even went and upgraded my boot prom from
the older 3.10.25 to the current version:

ok .version
Release 3.10.27 ME created 2000/06/22 16:45
OBP 3.10.27 2000/06/22 16:45
POST 1.17.0 2000/03/06 12:18

And did a set-default to whack all the PROM values back to something
like a sane value.  The only thing I changed after that was the
boot-device, from "disk net" to just "disk".

If I put a copy of the label on sd0, from the one on sd1:
# disklabel sd1 > /tmp/sd1
# disklabel -W sd0
# disklabel -R sd0 /tmp/sd1

It appears to be labeled properly, but I can't even manually newfs
the disk:

# newfs -i 8192 /dev/rsd0a
/dev/rsd0a: 140.3MB (287280 sectors) block size 8192, fragment size 1024
        using 4 cylinder groups of 35.07MB, 4489 blks, 4416 inodes.
wtfs: write error for sector 287279: Undefined error: 0

Any other suggestions?

-Kurt