> The versions 0.9.4 - 0.9.8 of 8088_bios [1] seem to be affected by a > bug I described in an GitHub issue [2]. It causes DS:0041 to be > overwritten by the BIOS. Problem is, that DS points to whatever it > is set when entering the BIOS and not to the BIOS data area. The bug > is that the BIOS fails to set DS if it encounteres an unsupported > function request. It sets DS correctly for supported functions.
This is a pretty basic thing that a BIOS should never do -- "assume" something about what a segment register is. It makes me wonder how many similar issues might still be in that BIOS? > In the case of Free FDISK, this behaviour trashes INT 10 vector. > This is because FDISK uses the intr() function of the Watcom libc, > and setting the segment values (DS, ES) of the REGPACK structure to > zero before invoking intr(). This is recommended by the Watcom > documentation to prevent loading of garbage segment selectors if > running in protected mode [4]. That doesn't quite make sense, since 0 is just as much of a "garbage" segment as any other random selector, and a value of 0 is almost guaranteed to affect data in the IVT and/or BIOS and/or DOS kernel. I think the recommendation is there more to cause the user to stop and think about what DS & ES need to be when calling RM code (and not necessarily just an ISR). In this particular case, no matter what value you put in DS it would have been a problem, though the fact that it was 0 probably made it easier to find (more reproducible) since it was almost guaranteed to be catastrophic. _______________________________________________ Freedos-devel mailing list Freedos-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/freedos-devel
