I have a 250Gb hard disk partitioned into 2 equal parts and I have DMA enabled.

It works for me too.

I typically create my partitions with Linux fdisk. I use a Lubuntu 16.04 bootable CD for this purpose. Oldest that hardware I typically use LBA on is usually able to boot from CD, this might not be your case. I mention this because there can be caveats with doing the partitioning on a modern system, AFAICR, but those typically only come into play when the retro hardware has very old/buggy LBA support OR in the case of LBA48.

Are you creating the partitions with the same size in both your CHS and LBA tests or are you using a bigger size in one test ? Incidentally, what is/are those size(s) ?

Also, Fdisk has a now deprecated (or possibly even now removed) legacy option. Not sure what that changes off the top of my head, but if may be a worth looking into. I do not recall if I have needed to use it.

I also seem to recall an old bug that may or may not be XTIDE specific (can't recall). Try using the secondary IDE controller to see if it helps.

Some working scenarios I have experienced :
Adding XT-IDE option ROM to Asus P3B-F BIOS [Thanks to DenizOezmen, it actually works!!!]

Here's an idea: using high endurance (micro)SD cards meant for continuous video recording as storage for retro gear .

Hope this helps.

Thanks ElectroSoldier and darry for the answers.

@darry, I'm creating my partitions using a custom Puppy Linux CD with my backup scripts installed (boots and works well even with 256 MB RAM). I'm not using any disk geometry related options for sfdisk except for the size and start of the partition, something like type=b, start=63, size=2GB.

I think I've narrowed the cause of the problem to the fact that sfdisk write a somehow non-compatible MBR when invoked with "type=b" on my old BIOS Windows 98 machine. It works if I do the same thing on VirtualBox (with the same Puppy Linux CD and same HD size).

For now, I fixed the problem by replacing my sfdisk command by a raw MBR write:

1MBR_1Partition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
2
3# Main partition can't be created using standard sfdisk command because
4# of BIOS / Linux incompatibilites regarding CHS format...
5echo $MBR_1Partition_2GB | xxd -r -p | dd of=/dev/sda

It's a bit less versatile than sfdisk, but after this intialization, everything works well (including the file system creation with mkfs.vfat), and I have DMA enabled (as the partition is not LBA).

I've found no trace of deprecated option in my sfdisk man page. I'll try to do a binary diff of my MBR between the working version (MBR_1Partition_2GB in my script, created by MS-DOS FDISK), and the non-working one.

As you suggested, a BIOS modification should probably fix the issue (the VirtualBox BIOS works well), I'll see if I'm brave enough to try this 😀

Talking about BIOSes, I've seen some projects like CoreBoot or LibreBoot that should probably offer more compatibility with modern hardware and other OSes. Would they work on old harware too (had anybody tried them) ?

You can actually change the partition type from 0B to 0C and vice versa just by directly editing the type field directly in the MBR sector with DEBUG or whatever, nothing else is different

maxtherabbit wrote on 2023-12-30, 15:28:

You can actually change the partition type from 0B to 0C and vice versa just by directly editing the type field directly in the MBR sector with DEBUG or whatever, nothing else is different

Yes, I already tried something like that : create the partition in LBA mode with sfdisk, restore all Windows files on it, then remove the "LBA" flag (still using sfdisk, or maybe gparted), and it did not work. IIRC (I tried lots of different things), Windows was not booting.

I'm not 100% sure that "LBA" flag is just a byte in the MBR, or if it somehow influences the geometry at partition creation time or something else on Linux / Windows side ?

It's just a byte field in the MBR