Re: [hatari-devel] 68030 MMU debugging

[ Thread Index | Date Index | More Archives ]

Am 11.09.2012 um 19:06 schrieb Eero Tamminen:

> Hi,
> On tiistai 11 syyskuu 2012, Andreas Grabher wrote:
>> If one of you is testing the 68030 MMU with different programs and
>> experiences problems, that seem to be MMU related, feel free to write me
>> a message. I'll try to fix the problems if i can find it. But because
>> finding the problem is harder than fixing it, i need the maximum
>> information i can get about the failing code.
>> So please if you post MMU debugging code, check that all debugging
>> messages are enabled (in line 63 and 64 of cpummu030.c set both
>> MMU030_ATC_DBG_MSG and MMU030_REG_DBG_MSG to 1). If possible, post the
>> critical part of the debugging code, but please also include a complete
>> debugging output (if the size is not too big), for example in an
>> appended text file.
> Maybe you could consider adding a "mmu" trace option, if that debug
> output is one-liners and not in very performance critical place?
> For that you need to add few lines to src/debug/log.[ch], the actual
> trace output is done in either of these ways:
>                if (LOG_TRACE_LEVEL(TRACE_OS_AES))
>                {
>                        AES_OpcodeInfo(TraceFile, AESOpCode);
>                }
> ...
>                LOG_TRACE(TRACE_OS_VDI, "VDI call %3hd/%3hd (%s)\n",
>                          VDIOpCode, subcode,
>                          VDI_Opcode2Name(VDIOpCode, subcode));
> Alternatively/additionally you could also have "mmu" subcommand for
> the debugger "info" command which would output more details.  See 
> src/debug/debugInfo.c for examples.
> (One can use "info" option in breakpoints, so that you get the "info"
> command information printed every time a breakpoint is hit.)
> 	- Eero

Thanks for the suggestion. At the moment some of the debugging messages are combined from two lines, depending on the MMU translation process. Also especially the ATC part is at performance critical points (called with every memory access). I can disable them via preprocessor, to get full speed for the ATC access.
At the moment i prefer to keep it the way it is. That is true for Previous, but feel free to change things in Hatari. But for now it would be good if you could keep the messages text the same, so i can easily understand and analyze the debugging code and compare to Previous' output.

I'll further think about the debugging messages once the MMU comes closer to its final state.

Mail converted by MHonArc 2.6.19+