Barry McGuigan World Championship Boxing (1985)(Activision).crt freezes

[dirkwhoffmann]

Reported by @chris70c.

At some point, the CPU halts:

Interestingly, the cartridge has type "Magic Desk":

Also, the CPU instruction view is broken in latest beta:

[dirkwhoffmann]

Update: This CRT is part of OneLoad64 Games Collection v2 Now Released - 1001 unique games to play via Magic Desk CRT format which contains lots of other CRTs in Magic Desk format. All other titles I've tried seem to work. Hence, I expect the issue is not related to the cartridge format. However, the D64 version of Barry McGuigan works just fine.

[chris70c]

Yes is part of OneLoad64 v4, I've tried 100s of other CRT from the same collection and they seem to work fine, maybe the D64 works fine because of a different crack or the way the memory dump has been made for the CRT... Sent you a d64 of Doughboy who also crash after the trainer (works in VICE)...

[dirkwhoffmann]

I cannot reproduce the Doughboy bug here (although I've spotted some other bugs in the latest beta (v4.6b1 is essentially broken):

[dirkwhoffmann]

More findings:

VirtualC64 eventually takes the jump at address $5127 which seems to be a ticket to death (it eventually executes an illegal instruction). VICE doesn't seem to take this jump. The culprit might be the LDA instruction at $5116 which reads from IO2 space. The current implementation of VirtualC64 always returns 0 here (only IO1 is used by this cartridge type).

However, VICE seems to return different values (values behind A:) :

Now, the question is where those values come from.

[chris70c]

Maybe some changes in 4.6 fixed it, I get a corrupted screen and the emulator stop working completely after the trainer in Doughboy...

[dirkwhoffmann]

Preliminary findings:

Now, the question is where those values come from.

It should behave the same way as if no cartridge is attached (open IO returns the bus value during phi1).

I have no idea what the code is doing. Maybe it's a cartridge check and somehow related to VICE bug 1011

I've tried the following two code changes, but both didn't work

u8
MagicDesk::peekIO2(u16 addr) const
{
    return vic.getDataBusPhi1();
}

u8
MagicDesk::peekIO2(u16 addr) const
{
    return rand() & 0xFF;
}

I'll stop working on this issue for now as I don't have enough time to dig deeper.

[dirkwhoffmann]

Nailed it. Returning phi1 is the right thing to do, but I only did this for the IO 2 space. It also has to be done for IO 1:

u8
MagicDesk::peekIO1(u16 addr)
{
    return spypeekIO1(addr);
}

u8
MagicDesk::spypeekIO1(u16 addr) const
{
    return disabled() ? vic.getDataBusPhi1() : control;
}

I was right with assuming the game implements a cartridge check. But with my first code change, it passed the IO 2 check, only. With the latest changes, it passes the IO 1 check, too.

[dirkwhoffmann]

Fixed in v4.6.