Re: [hatari-devel] 68030 MMU emulation: TEMPLMON, ROMSPEED

[Thomas Huth <th.huth@xxxxxx>]

Am Fri, 19 Oct 2012 17:36:24 +0200
schrieb Andreas Grabher <andreas.grabher@xxxxxxxxxxxx>:

> That does not look very good ... seems there might be something wrong
> with the TRY/CATCH/THROW stuff.
> 
> I did not yet port this to Previous. You may try to undo it:
> 
> Go to cpummu030.c, line 1480:
> 
> In the function mmu030_put_long_atc there is a line:
> THROW(2); //M68000_BusError(addr, 0);
> 
> Comment or remove THROW(2) and re-activate M68000_BusError(addr, 0);
> 
> There are 5 more functions to do the same (put word, put byte, get
> long, get word, get byte).

Andreas,

the more I think about this, the more I think that the MMU code _must_
really throw the bus errors via THROW instead of M68000_BusError()
(i.e. not using SPCFLAG_BUSERROR). Imagine the following opcode:

   move.l   (a0,d0.w),d0

Let's assume that a0+d0 points to a memory location of a non-mapped
page, so the memory access should cause a bus error. When the MMU code
uses SPCFLAG_BUSERROR to emulate the bus error, d0 gets destroyed
before the bus error handler is called. Thus when the instruction is
resumed, a0+d0 points to the wrong memory location!

So it's crucial to use THROW instead of SPCFLAG_BUSERROR - and when
THROW is not working yet, we've got to find the right fix first.

 Thomas