Disassemble some M68000 machine code into what an assembler can assemble again
This project aims to be a disassembler that is capable to produce assembly code that an assembler will translate into the same original machine code. It could be called a matching disassembling. GNU AS and Sierra ASM68 assemblers are currently supported.
It's only use case for now is Sega Mega Drive / Genesis ROM hacking. I failed to find any way to disassemble SMD ROMs in such a way that it would be possible to assemble it back into binary. All disassemblers I tried produce either syntactically incompatible assembly listing, or it is not the same as original binary after translation into machine code. So I decided to build my own disassembler, that will do exactly what I need with full control over the process and maybe more.
To build this project, you will need CMake and some modern C++ compiler like GCC or Clang. Here is how to build it using CMake:
cmake -B cmake-build -S .
cmake --build cmake-build
It will produce a binary named m68k-disasm
inside the cmake-build
directory.
You can copy it somewhere to ~/.local/bin/
, /usr/local/bin/
or other
directory that is added to your shell's PATH
environment variable.
It works for me on Linux, It may work the same way on OSX either and it may be not so easy on Windows. I can't see why it could impossible on OSX and Windows to build this project, but it is not tested.
You may want to run this on a random binary file just to see how it works. You can do it like this:
./cmake-build/m68k-disasm -o disasm.S /path/to/file.bin
This command will produce disasm.S
file, that contains assembly listing.
You may assemble it back with m68k-none-elf-as
to see if it is valid asm code
with the following command sequence. Note that you need to obtain or build by
yourself m68k-none-elf-gcc
toolchain to run the following command sequence,
test.ld
is already provided in this repo.
m68k-none-elf-as disasm.S -o a.o
m68k-none-elf-ld -T test.ld -o a.elf a.o
m68k-none-elf-objcopy -O binary a.elf a.bin
cmp /path/to/file.bin a.bin
This command sequence will produce a.o
, a.elf
intermediate files and a.bin
being the same binary as the /path/to/file.bin
file, which is tested by cmp
command.
Speaking of the real use case: you can disassemble Sega Mega Drive (Genesis) ROM with PC trace table to start hacking it. PC trace table is a text file containing one decimal number per line, representing a program counter value that it had at least once during the ROM execution. Every number must be unique to the file. It may look like this:
512
518
520
526
528
532
536
540
544
548
...
It may contain thousands of lines (real example).
I personally got one by playing a game on a specifically modified version of
picodrive
for this purpose. I added 4MiB table and made emulator write all
program counter values in it and then dumped the table into a file using
printf
function in picodrive
C source code. You can do this with you
favorite open source emulator too.
When PC trace table file is obtained, pass it with option -t pc-trace.txt
alongside with the ROM you were playing off of (rom.bin
) while gathering the
trace table.
./cmake-build/m68k-disasm -t pc-trace.txt -o disasm.S rom.bin
Or better with labeled locations analysis and some fancy raw comments:
./cmake-build/m68k-disasm -frdc -fxrefs-to -fxrefs-from -flabels -fabs-labels -frel-labels -fexport-labels -fexport-functions -fimm-hex -follow-jumps -fwalk -t pc-trace.txt -o disasm.S rom.bin
It will produce disasm.S
which you can modify and assemble as shown in
previous examples.
To get detailed help you can run:
./cmake-build/m68k-disasm -h
Goals of this Motorola 68000 disassembler project in this particular repo:
- Support all Motorola 68000 ISA instructions.
- Flawless compatibility with GNU AS and Sierra ASM68 syntax. It should always emit the code on which GNU AS produces absolutely identical binary (with or without linkage) without errors or warnings, unless some peculiar flags has been specified.
- Support PC trace tables. With trace tables it will disassemble traced PC locations only, without attempt to disassemble everything, because not everything is instruction, some code is just data.
- Label jump locations and bind jumps and calls to them. Label obvious ROM read accessed locations and bind the instructions to the labeled locations. To make it possible to split and reorganize the binary.
What could become a goal (possible features):
- Other Motorola 680x0 instruction sets support, including ColdFire.
- Functions and function boundaries detection.
- Static analysis of call graph of branches and subroutine calls.
- PC trace aided static analysis of dynamic branches and subroutine calls.
- Base address other than
0x00000000
. It is only zero for now because it is sufficient for Sega Mega Drive / Genesis ROM images. - Support for more than 4MiB of code size.
- Sparse address space support (instead of single continuous 4MiB block that
starts at
0x00000000
offset). - Other assembler syntaxes (e.g. ASM68K.EXE).
- Expose a library API.
What is not the goal (at least not in this repo):
- Decompilation into some high level language like C or C++.
- Other instruction set architectures support like MIPS, x86, amd64, ARM, PowerPC and so on.
- All M68000 instructions supported
- It generates GNU AS compatible listing, that may be translated back to machine
code using
m68k-none-elf-as
in the way that it matches original binary file, no matter what. - It generates Sierra ASM68 compatible listing, that may be translated back to
machine code using
ASM68
in DOS environment in the way that it matches original binary file, no matter what. - It generates labels for all jump instructions (JSR, JMP, BRA, Bcc and DBcc) if
jump location is inside the code being disassembled. This feature can be
enabled with
-flabels
,-frel-labels
and-fabs-labels
options, all at once. It also generates labels for some data accessing instructions (namely: NBCD, PEA, LEA, CMP, SUB, ADD, MOVEM and MOVE) and this behavior enabled with the same options as per jump instructions. It is possible to implement this for all of the rest instructions, but it just has to be done if someone needs it. - Traced disassembling - you can provide a PC trace table file with option
--pc-trace=file
to disassemble only what is supposed to be instructions and leave all the rest as raw data. Otherwise it will try to disassemble everything. - Automatic and non-committing code discovery with jump follow with
-fwalk
and-ffollow-jumps
. I.e. when PC trace table provided, the disassembler will try it's best to disassemble non-traced continuous code spans, until unknown instruction or unconditional branch encountered.
- Motorola 68000 only supported. That means no 68010 support or any 680x0, except 68000. No ColdFire support.
- Code must be aligned to 2 bytes boundary. All PC trace values have to be
dividable by 2 without remaining. That limitation has been put into design
from the beginning, because 68000 cannot perform unaligned
word
andlong
memory access, including instruction fetch. This is not true for some of 680x0 variations. - Base address is always assumed to be
0x00000000
. - Maximum binary size is 4MiB.
- Labels for locations outside of the code being disassembled are not generated, they remain as raw address arguments and/or PC-relative offset arguments.
Authors:
- Vladimir Novikov – [email protected]
This is free and unencumbered software released into the public domain. See
UNLICENSE
for more information.
This repository includes source code of other projects:
- Optparse (gh:skeeto/optparse) - Unlicense
Resources used to implement the disassembler (this set is more than sufficient to support all of M68000 instructions):
- The 68000's Instruction Set - Appendix of an unrecognized book. Basically the same information also could be found in gh:prb28/m68k-instructions-documentation.
- Motorola 68000 CPU Opcodes.
- GNU assembler (GNU Binutils) 2.40 (
m68k-none-elf-as
). - GNU objdump (GNU Binutils) 2.40 (
m68k-none-elf-objdump
).
I will eventually put this repo on GitHub, I guess. You can create issues and pull requests there. You can email me directly to ask a question, send a patch or discuss problems if you prefer this over GitHub.
I decided to go without code formatting standard for now. Just be nice to not mix up spaces with tabs (use spaces everywhere) if you are sending patch or pull request and that's it.
C++ STL is not welcomed here. Almost every STL header (besides C standard
library wrappers like cstring
or cstdio
) increases compilation times
significantly. This disassembler is used to be developed on Celeron N4000
machine with eMMC memory instead of SSD and it is very sensitive to STL bullshit
increasing compile times. Please, make sure you don't use any compile time heavy
headers. If you really need something like hashmap or RB-tree, then bring some
tiny MIT/BSD/Unlicense library from somewhere or write it yourself.
Run tests when the work is done to make sure you didn't break anything.