Relies on POST=7 as a kernel initialization indicator, instead of
patching the BIOS.
Fixes EXE loading with OpenBIOS and PS2 BIOS, and fast boot getting
baked into save states.
See the comments in the diff. This **will** cause lag in Final Fantasy
VII during preload areas, but that has also been confirmed on console.
Duke Nukem - Total Meltdown does this silly Read -> Pause command chain,
except it sets its data/INT1 callback on the read, but never clears it
after the pause. Therefore, if it doesn't receive at least one sector,
the callback never gets cleared, and when the next read happens, it
stores the "current" callback in the "backup" variable, which should be
null, but now has the callback from the dud read.
The result is any INT1 during XA playback running the dud callback,
which says "hey, I'm not expecting any data, so pause, and stops the
background music playback. Making sure at least one sector from that
silly chain is delivered ensures the callback is cleared, and this does
not happen.
Since the pause first mentioned above will now error out until the first
sector is delievered, the game spams pause until it eventually does
succeed after the INT1. This behaviour has also been verified on
hardware, thanks to rama for the xStation logs.
Fixes Rock-Climbing - Mitouhou e no Chousen - Alps Hen (Japan) crashing
on boot. Silly game does a 512 word DMA from CDROM to 0x001FFF04, with the
mapped memory size set to 4MB, expecting the DMA to trigger a bus error,
because if it doesn't, then your kernel memory goes bye bye.
A bunch of other games do set the upper 4MB to be unmapped, but this
change appears to have no effect on them. As expected, they're probably
doing the correct thing.
Fixes pops in menu music in Simple 1500 Series Vol. 92 - The Tozan
RPG - Ginrei no Hasha.
I haven't seen anything that _properly_ uses 8-bit ADPCM yet. The above
game does, but only in the inaudible portion of the start of the track.
