[syslinux] MEMDISK issue with OptiPlex GX280,620

[Gene Cumm]

On Fri, Dec 10, 2010 at 20:28, Jason Vasquez <jvasquez at kennesaw.edu> wrote:
> Pardon my delay in responding. The past day had me reviewing the disk image being used - specifically, its geometry and its history.
>
> Ghost imaging has long been an integral part of the environment here. With that said, relying on Symantec's Ghost Solution Suite's Ghost Boot Wizard to create a Network Boot Package in the form of an ISO and CD had been the way for years. Enter SYSLINUX and an available, local PXE server, and CD booting was migrated to a less-archaic method of imaging. The ISO produced from the Boot Wizard was not to be dismissed though. The ISO file contains a 24 MB disk image; one that is larger than I prefer and loading slower than desired. That image is, in effect, the source of my hacked/modded 8MB disk image. I need something larger than 4 MB but less than 24 MB.
>
> Upon Shao's splendid advice of gzipping initrd and seeing the results, I have elected to start anew. I returned to the ISO (and its embedded 24 MB disk image) produced from the Symantec GSS Boot Wizard. After gzipping the disk image, I now have a ~6 MB file. Nicely, it loads rather quickly -  besting the 8 MB image.
>
> Additional to this clean approach, the disk image's geometry is explicit. Based on yours and Gene's comments about disk geometry, I feel much better now having a sound disk image. Progress, agree?
>
> Booting the new image though has changed little. Version 4.03 of MEMDISK (ISOLINUX and PXELINUX both invoked for testing), produced these results:
>
> MEMDISK 4.03:
>
> Ramdisk at 0x07f099000, length 0x005cc813
> moving compressed data 0x07f099000 to 0x7d939a00
> gzip image; decompressed addr 0x7df06400, len 0x01780800: ok
> command line: initrd=images/ghostclient/280_620/osbootc.img.gz BOOT_IMAGE=memdisk
> MEMDISK: Image seems to have fractional end cylinder
> Disk is hd0, 24066 K, C/H/S = 2/255/63 (MBR/MBR), EDD on, rw
> Using safe INT 15h access to high memory
> Code 1744, meminfo 264, cmdline 70, stack 512
> Total size needed = 2590 bytes, allocating 3K
> Old dos memory at 0x9fc00 (map says 0xa0000), loading at 0x9f000
> 1588: 0xffff  15E801: 0x3c00 0x07cf0
> INT 13 08: Success, count = 1, BPT = 0000:0000
> We lost the last drive in our class of drives.
> Drive probing gives drive shift limit: 0x01
> old: int13 = f0005c6f  int15 = f000f859  int1e = f000efc7
> new: int13 = 9f00000a  int15 = 9f0003ba  int1e = f000efc7
> Loading boot sector... booting...
> Starting PC DOS...
>
> Again, Ghost is unable to see the internal SATA drive from this image's booted environment.
>
> Reverting back to MEMDISK 3.83, the 'lost last drive' message does not occur and Ghost can see the SATA drive.
>
> To reiterate, this image does not work on OptiPlex GX280s and GX620s via MEMDISK versions newer than 3.83. The image will work on the OptiPlex 755,780, and 980 systems via all MEMDISK versions post 3.83. And as much as a superfloppy might assist here, its 3840 KB capacity is generally inadequate.

With regards to why those exact two models, I may have some insight
(it just dawned on me).  BCM5751 [14e4:1677] (rev 01).  Models GX260
and GX270 used Intel chips and the models 740 and 745 used BCM5754
chips.  Models 755, 760, 780 and 980 all use Intel chips.  A Dell
Latitude D610 also uses a BCM5751 and may experience the same
behavior.

Along the lines of superfloppies, I think the LS-120 drive _might_
have acted like a true superfloppy (big unpartitioned device).  You
can go bigger but you'll need to specify and track your geometry more
carefully.  The others are just standard sizes (even if rare like the
3840 kiB size).

> If it helps, I could provide a copy of my image for personal testing and examination. If you're interested, please let me know how it would be best to deliver the ~6 MB image.

I'd be interested in seeing it.  Private email or a web link are the
two quickest methods at least for me.  I might even be able to make
one like it.

> Thank you and Gene both for responding this week. Regards.

-- 
-Gene