[syslinux] Looking for debugging ideas and help with testing

[Gene Cumm]

On Mon, Dec 13, 2010 at 20:09, Ferenc Wagner <wferi at niif.hu> wrote:
> "H. Peter Anvin" <hpa at zytor.com> writes:
>
>> On 12/13/2010 02:34 PM, Ferenc Wagner wrote:
>>>
>>> The theory is that the boot.img.gz-s fail to boot correctly if written
>>> to a full disk device (/dev/sda), but work fine if written to a
>>> partition (/dev/sda1) and invoked by some MBR (see #604245).  No theory
>>> about the ISO, but the reported symptoms are rather similar.  Anybody
>>> inclined to help, please test whether the above images (or other similar
>>> ones) fail to boot the Debian Installer on their systems.  Thanks.
>>
>> Most likely the problem is that the newer Syslinux misidentifies
>> something in memory as a partition handover structure.
>>
>> Someone -- and I can't remember who right now -- built a debugging MBR
>> for testing this case.
>
> Do you mean Gene Cumm's mbrho.S?  I built that right now, under Qemu it
> prints
>
> DL: 80  DS: 0000  SI: 0000
>  53 FF 00 F0 53 FF 00 F0 53 FF 00 F0 53 FF 00 F0
> ES: 0000  DI: FFAC
>  00 00 D4 CC
>
> I can't make much sense of this, but anyway...  Should I get the
> submitters run this?  Could this affect Isolinux as well?

All the standard BIOS handoff.  DL= BIOS drive number (0x80 is the
first HDD).  [DS:DI] is supposed to be the partition table entry
(hence 16 bytes).  ES:DI may be a pointer to "$PnP" (not in this
case).  It also performs a standard scan for "$PnP".  I know it'll
show properly on real machines and bochs and/or qemu fake it and it's
not really there.

Unfortunately, I just found there's either a bug in the version at the
top of that branch or my compiler went wonky.

-- 
-Gene