In message <20041127091152.GA1619@snowdrop.l8s.co.uk>, David Laight writes:
>On Fri, Nov 26, 2004 at 09:26:08PM -0500, Steve Bellovin wrote:
>
>Some responses to earlier in the thread...
>
>The "no operating system" error (reported on the original mail) was
>probably the "No operating system" error message from the NetBSD 'normal'
>mbr code.  The 'bootselect' versions report 'Error 3' instead.
>
>So the 'use entire disk' made to difference.
>
>That error is reported if the first sector of the 'root' partition
>definitely doesn't contain the boot code.
>This is usually because 'installboot' failed to run properly.
>(note that disklabel -B isn't the way to set up that area - except on hp300)
>
>In this case the mbr code will have read sector 63 - and not found the
>0xaa55 at the end.  This read should be largely independant of the disk
>geometry in use - the sectors/track is almost always 63.
>
>> I changed the BIOS setting for the disk to LBA, from "Auto" (the other 
>> choices were "Large" and "C/H/S"); after that, it worked just fine.
>
>That makes this all look like a BIOS bug!
>Probably something to do with the geometry translations!
>I don't actually know what the BIOS means by 'LBA' and 'Auto', but you
>might see the reported geometry change when you select different options.
>
>> Looking back at my postings I see this:
>> 
>> # fdisk wd1
>> Disk: /dev/rwd1d
>> NetBSD disklabel disk geometry:
>> cylinders: 310101, heads: 16, sectors/track: 63 (1008 sectors/cylinder)
>> total sectors: 312581808
>> 
>> BIOS disk geometry:
>> cylinders: 1024, heads: 240, sectors/track: 63 (15120 sectors/cylinder)
>> total sectors: 268435455
>> 
>> Why do the BIOS and NetBSD number of sectors disagree?  The same is 
>> true now, with the system running.  Yes, it's a 160G drive, but I have 
>> a 160G drive on another machine, and the two geometries give the same 
>> total on it.  Is this some BIOS limitation in dealing with a large 
>> drive?  And might that have been at the root of my problem?
>
>Well, 268435455 is 0xfffff - which probably means that the BIOS doesn't
>support LBA48 reads for disks > 137GB.  You aren't trying to read beyond
>that point - at least the failure is being reported before any attempt to
>read such sectors (/boot and /netbsd probably have to be below the 137GB
>boundary).

Right -- and now that I have it working, I'm quite certain that the 
problem is the 137G boundary.  (The disk has a jumper to limit it to 
32G -- which isn't what I bought a 160G drive for...)  

I have two questions now.  The first is whether or not some part of 
NetBSD -- sysinst, fdisk, installboot -- should detect the problem and 
issue a warning mesage.  To me, the clue is the differing numbers of 
sectors reported by NetBSD and the BIOS.  The second is what I should 
do now: should I try a BIOS update, or should I just repartition the 
disk so that wd0a is within the 137G limit?  (Actually, the INSTALL 
documentation should warn of this, too, and suggest that the default 
partitioning -- everything in 'a' -- is a bad idea if you see the 
warning about the BIOS versus your disks....)

		--Steve Bellovin, http://www.research.att.com/~smb