Suggestion: Work around Xi8088 BIOS bug by setting DS = 40h for 13.41 call

[ecm-pushbx]

Circling back to the Xi8088 BIOS bug that affected at least some revisions of the Book8088:

The bug was described in http://www.bttr-software.de/forum/forum_entry.php?id=21275

This is where the kernel may call a ROM-BIOS that doesn't support LBA to detect that it doesn't support LBA:

kernel/kernel/initdisk.c

Lines 651 to 657 in 1b6de0f

	/* check for LBA support */
	regs.b.x = 0x55aa;
	regs.a.b.h = 0x41;
	regs.d.b.l = drive;
	regs.flags = FLG_CARRY; /* ensure carry is set to force error if unsupported */
	init_call_intr(0x13, &regs);

My idea is to force DS = 40h for this call, which will work around the bug of the BIOS writing to offset 41h without setting up DS.

I didn't have this idea yet because my use of 13.42 without the proper 13.41 check call makes this more difficult, however I just added a workaround to the Xi8088 bug to the callers that are not as size constrained as the lDOS boot sector loaders, eg lDOS iniload: https://hg.pushbx.org/ecm/ldosboot/rev/4c929733a229

(I will upload a blog post that also addresses this later.)

[boeckmann]

Sounds reasonable. That is directly related to my question on the developer mailinglist some time ago to what DS should be set in this case. There is a (closed) related FDISK issue FDOS/fdisk#80. I may implement setting DS as well for FDISK.

[ecm-pushbx]

Has this still not been added yet?

[ecm-pushbx]

Added a Pull Request with this change: #224

@boeckmann It appears you patched fdisk in FDOS/fdisk@8126a21 - right?

[boeckmann]

@boeckmann It appears you patched fdisk in FDOS/fdisk@8126a21 - right?

At least that was the intention. I think I tested it via 86box.

[ecm-pushbx]

The FAT12/FAT16 loaders using either CHS or LBA set ds to 1FE0h:

kernel/boot/boot.asm

Line 210 in e3bfd6a

mov ds, ax

So it corrupts the byte [1FE0h:41h] which is at linear 1FE41h, at 127 KiB. That's within the allocation for a large FAT or a large kernel file. If the FAT and file both are smaller, or the FAT entry that ends up at this address isn't used, then it will appear to work despite the memory corruption.

[ecm-pushbx]

SYS also uses 13.41:

kernel/sys/sys.c

Line 1265 in e3bfd6a

#ifdef __WATCOMC__

[ecm-pushbx]

SYS also uses 13.41:

kernel/sys/sys.c

Line 1265 in e3bfd6a
#ifdef WATCOMC

Addressed in today's #226