[B2-devel] Basilisk II FPU problem with Speedometer 3.23

[Gwenole B. <gb....@fr...>]

Hi,

On fast machines, or with JIT enabled, Speedometer 3.23 can crash at 
the exit of FPU tests. With a Quadra 630 ROM, the relevant code snippet 
is:
pc 0200db76 | 0838 0000 1efc            btst    #$00,MMFlags
pc 0200db7c | 660e                      bne.s   $0200db8c
pc 0200db8c | 2008                      move.l  a0,d0
pc 0200db8e | 2240                      movea.l d0,a1
pc 0200db90 | 6702                      beq.s   $0200db94
pc 0200db92 | 2051                      movea.l (a1),a0
The crash occurs once you validate the "Tests are done" dialog.

A1 is somehow corrupted and always has 0xffff00ff when a crash occurs. 
I could reproduce the problem in 32-bit or 64-bit x86 mode. In any 
addressing mode. In 68030+FPU or 68040, which are the same. With any 
FPU emulation (UAE, IEEE).

Of course, it's easier (faster) to get a crash with JIT enabled. This 
yielded the question of possible problems with nested EMUL_OP 
processing or Execute68k()? I am not quite sure.

Workarounding the faulting 68k instruction will simply make B2 crash 
later. The OS is 8.1 with 32 MB.

Anobody got similar problems?