search

mamedev / mame

MAME
https://www.mamedev.org/
Other
7.76k stars 1.95k forks source link

imagedev/floppy: fix a bug with reading the first 1-bit on a track #12439

Closed SpecLad closed 4 weeks ago

SpecLad commented 4 weeks ago

floppy_image_device::find_index uses binary search to find the index for which buf[spos] <= position < buf[spos + 1]. However, the algorithm behaves incorrectly when position < buf[0]. In this case, the algorithm returns 0, as if position was between buf[0] and buf[1].

The effect of this is that if get_next_transition is called with a timestamp that is between the start of the revolution and the first transition, then instead of returning the timestamp of that transition, it returns the timestamp of the second transition instead. Essentially, the first 1-bit on the track gets flipped to a 0.

I have encountered this in Apple II emulation, where this bug manifests as sporadic I/O errors.

Fix it by doing two things:

  1. Replace find_index with a call to upper_bound from the standard library, which behaves correctly in edge cases.

  2. If upper_bound signals that position < buf[0], then adjust base and index to point to the last transition of the previous revolution.

rb6502 commented 4 weeks ago

Tried this on a small assortment of systems including Apple II and Amiga 500 and it doesn't seem to have any ill effects. I'd still like @galibert to have a look as a sanity check though.

mgarlanger commented 4 weeks ago

Interesting, I wonder if this could be the cause of the glitch, when we tried to have hard-sectored tracks keep the index hole as the start of the track. This caused intermittent glitches on the last sector of the track which has the index hole in the middle of the sector.

Lord-Nightmare commented 4 weeks ago

I would not be at all surprised if this was the cause of that issue with hard sectored disks.

mgarlanger commented 4 weeks ago

Confirmed this fixes the hard-sectored glitches at the index hole. I added this change to my hard-sectored PR - https://github.com/mamedev/mame/pull/11952 and added back the half sector rotate, and things are working fine. The Vector 4 booted without problem, able to run programs off the disk, copied the program, erase the file, and format a new disk. All without any sign of disk errors.

galibert commented 4 weeks ago

Much thanks for finding that :-)