[CUWiN-Dev] upgrade/partition table problems

[Bill Comisky]

On Wed, 28 Feb 2007, Matthew Isaacs wrote:

> --[PinePGP]--------------------------------------------------[begin]--
> Bill Comisky wrote:
>> On 2/28/07, Bill Comisky <bcomisky at pobox.com> wrote:
>>> On 2/28/07, Matthew Isaacs <isaacsm at cuwireless.net> wrote:
>>   <snip>
>>>> Someone emailed the support list the other day with this problem.  I
>>>> thought it might be a bug with the memory disk image used to PXE boot
>>>> the person's node.  I'll have the run fdisk -vv to confirm that it is
>>>> the same problem.
>>>>
>>>> Regarding the corrupt partition table, I believe fixlabel was designed
>>>> to be used in the instances.  Alternatively, one could PXE boot the
>>> node
>>>> with and memory-disk image and run the upgrade from that.  The current
>>>> md image is at 0.7.0, so it *shouldn't* pose any problem.
>>>>
>>>> Hope that helps.
>>>
>>> I'll take a look at fixlabel when I get a chance.  As far as PXE
>>> booting, is there any way to initiate that without having serial
>>> console access in order to interrupt the boot process and enter the
>>> BIOS?
>>
>> I didn't read Dave's PXE boot HOWTO closely enough last time..
>>
>> "It is best if your CUWiN nodes always PXE boot, first.  To make
>> sure they do, you need to boot your node with a serial console
>> attached, break into the comBIOS monitor with Ctrl-P, and 'set
>> bootdrive=f0 80 81 ff'.  Not all comBIOS versions support 'set
>> bootdrive'.  You may need to update your comBIOS to 1.28 (or whatever
>> is the latest version)."
>>
>> Anyone know how long the 4526 will attempt to PXE boot this way before
>> failing and booting from flash?  That is, can I make this setting
>> permanent on the nodes without much (re)boot time penalty?
>>
>> Bill
>>
>
> It waits a few seconds.  All of the nodes in the Urbana testbed are
> setup in this manner, and I think it is enabled by default in the more
> recent bios releases from soekris.  If it is not currently setup boot
> from the network, that doesn't solve the problem of getting to the
> nodes.  Hopefully, they're already setup that way, and if not, then
> maybe fixlabel will do the trick.  Looking at the source, it basically
> destroys the MBR and partition table and rebuilds it for the 4526, so it
> should fix the problem, as long as the flash isn't corrupt/bad.

I'm pretty sure all the nodes are setup to boot from flash first, but 
it'll be a big win to setup the BIOS for default PXE booting from now on.
Unfortunately fixlabel didn't do the trick, I got a segfault and then lost 
my connection shortly afterward.

# fixlabel
old mbr and disklabel
Disk: /dev/rwd0d
NetBSD disklabel disk geometry:
cylinders: 977, heads: 4, sectors/track: 32 (128 sectors/cylinder)
total sectors: 125056

BIOS disk geometry:
cylinders: 977, heads: 4, sectors/track: 32 (128 sectors/cylinder)
total sectors: 727412841289204

Partition table:
0: <UNUSED>
1: <UNUSED>
2: <UNUSED>
3: <UNUSED>
# /dev/rwd0d:
type: ESDI
disk: net4526
label:
flags:
bytes/sector: 512
sectors/track: 32
tracks/cylinder: 4
sectors/cylinder: 128
cylinders: 977
total sectors: 125056
rpm: 3600
interleave: 1
trackskew: 0
cylinderskew: 0
headswitch: 0           # microseconds
track-to-track seek: 0  # microseconds
drivedata: 0

4 partitions:
#        size    offset     fstype [fsize bsize cpg/sgs]
  a:    125056         0     4.2BSD      0     0     0  # (Cyl.      0 - 
976)
  d:    125056         0     unused      0     0        # (Cyl.      0 - 
976)

1+0 records in
1+0 records out
65536 bytes transferred in 0.053 secs (1236528 bytes/sec)
[1]   Segmentation fault (core dumped) disklabel -rw wd...
# fdisk -vv
fdisk: primary partition table invalid, no magic in sector 0

After this last command I lost contact with the box, so it's likely a roof 
trip will be required after all.  Thanks for the tips,

Bill

--
Bill Comisky
bcomisky at pobox.com