m2c Decompilerm2c ("Machine code to C") is a decompiler for MIPS and PowerPC assembly that produces C code, with partial support for C++.
This project, initially named mips_to_c, has the goal to support decompilation projects, which aim to write source code that yields byte-identical output when compiled with a particular build system.
It originally targeted popular compilers of the late 1990's, but it also works well with newer compilers or hand-written assembly.
m2c is often used in decompilation workflows with splat, asm-differ, and decomp-permuter.
Its focus on finding "matching" C source differentiates it from other decompilation suites, such as IDA or Ghidra.
Right now the decompiler is fairly functional, though it sometimes generates suboptimal code (especially for loops).
The input is expected to match the GNU as assembly format, produced by tools like spimdisasm.
See the tests/ directory for some example input and output.
An online version is also available.
This project requires Python 3.7 or later. To install the Python dependencies:
python3 -m pip install --upgrade pycparserYou might need to install pip first; on Ubuntu this can be done with:
sudo apt update
sudo apt install python3-pippython3 m2c.py [options] [-t <target>] [--context <context file>] [-f <function name>] <asmfile>...Run with --help to see which options are available.
Context files provided with --context are parsed and cached, so subsequent runs with the same file are faster. The cache for foo/bar.c is stored in foo/bar.m2c. These files can be ignored (added to .gitignore), and are automatically regenerated if context files change. Caching can be disabled with the --no-cache argument.
m2c has support for both MIPS and PowerPC assembly.
It also has some compiler-specific heuristics and language-specific behavior.
For example, it can demangle C++ symbol names as used by CodeWarrior.
Collectively, the output's architecture, compiler, and source language are referred to as a target. The following target triples are supported:
--target mips-ido-c: MIPS (with O32 ABI), IDO toolchain, C language--target mips-gcc-c: MIPS (with O32 ABI), GCC toolchain, C language--target ppc-mwcc-c: PowerPC, MetroWerks CodeWarrior toolchain (mwccecpp.exe), C language--target ppc-mwcc-c++: PowerPC, MetroWerks CodeWarrior toolchain (mwccecpp.exe), C++ languageBy default, m2c decompiles all functions in the text sections from the input assembly files.
m2c is able to perform a small amount of cross-function type inference, if the functions call each other.
You can limit the function(s) that decompiled by providing the -f <function name> flags (or the "Function" dropdown on the website).
When provided input files with data, rodata, and/or bss sections, m2c can generate the initializers for variables it knows the types of.
Qualifier hints such as const, static, and extern are based on which sections the symbols appear in, or if they aren't provided at all.
The output also includes prototypes for functions not declared in the context.
m2c cannot generate initializers for structs with bitfields (e.g. unsigned foo: 3;) or for symbols that it cannot infer the type of.
For the latter, you can provide a type for the symbol the context.
This feature is controlled with the --globals option (or "Global declarations" on the website):
--globals=used is the default behavior, global declarations are emitted for referenced symbols. Initializers are generated when the data/rodata sections are provided.--globals=none disables globals entirely; only function definitions are emitted.--globals=all includes all of the output in used, but also includes initializers for unreferenced symbols. This can be used to convert data/rodata files without decompiling any functions.By default, m2c can use type information from decompiled functions to help fill in unknown struct fields.
This behavior can be disabled with --no-unk-inference ("Disable unknown struct/type inference" on the website).
For structs in the context, the following fields treated as "unknown" space that can be inferred:
char arrays with a name starting with unk_, e.g. char unk_10[4];UNK_ or MIPS2C_UNK, e.g. UNK_TYPE4 foo;Currently, struct field inference only works on structs without bitfields or unnamed union fields.
The output will include declarations for any struct with at least one inferred field.
By default, m2c infers the types of stack (local) variables, and names them with the sp prefix based on their offset.
Internally, the stack is represented as a struct, so it is possible to manually specify the names & types of stack variables by providing a struct declaration in the context. m2c looks in the context for a struct with the tag name _m2c_stack_<function name> (e.g. struct _m2c_stack_test for a function test()).
The size of the stack must exactly match the detected frame size, or m2c will return an error.
If you run m2c with the --stack-structs option ("Stack struct templates" on the website), the output will include the inferred stack declaration, which can then be edited and provided as context by re-running m2c.
Here is an example for specifying the stack for the custom_stack end-to-end test.
First, run m2c with the --stack-structs option to get the inferred struct for the test() function:
python3 m2c.py tests/end_to_end/custom_stack/irix-o2.s -f test --stack-structsNow, based on the body of the test() function, we can make some guesses about the types of these variables, and give them more descriptive names:
// Save this file as `test_context.c`
struct Vec {
s32 x, y, z;
};
struct _m2c_stack_test {
char pad0[0x20];
struct Vec vec;
struct Vec *vec_ptr;
s32 scale_z;
s16 scale_y;
char pad36[1];
s8 scale_x;
}; /* size 0x38 */
int test(struct Vec *vec_arg);Finally, re-run m2c with our custom stack as part of the --context. The --context option can be specified multiple times to combine files.
python3 m2c.py tests/end_to_end/custom_stack/irix-o2.s -f test --context test_context.cThe following options control the formatting details of the output, such as braces style or numeric format. See ./m2c.py --help for more details.
(The option name on the website, if available, is in parentheses.)
--valid-syntax--allman ("Allman braces")--knr ("K&R braces")--pointer-style ("* to the left")--unk-underscore--hex-case--comment-style {multiline,oneline,none} ("Comment style")--comment-column N ("Comment style")--no-casts--zfill-constants ("0-fill constants")--deterministic-varsNote: --valid-syntax is used to produce output that is less human-readable, but is likely to directly compile without edits. This can be used to go directly from assembly to the permuter without human intervention.
There are several options to m2c which can be used to troubleshoot poor results. Many of these options produce more "primitive" output or debugging information.
--no-andor ("Disable &&/||"): Disable complex conditional detection, such as if (a && b). Instead, emit each part of the conditional as a separate if statement. Ands, ors, nots, etc. are usually represented with gotos.--no-switches ("Disable irregular switch detection"): Disable "irregular" switch statements, where the compiler emits a single switch as a series of branches and/or jump tables. By default, these are coalesced into a single switch and marked with an /* irregular */ comment.--no-unk-inference ("Disable unknown struct/type inference"): Disable attempting to infer struct fields/types in unknown struct sections and global symbols. See the Struct Field Inference section above.--gotos-only ("Use gotos for everything"): Do not detect loops or complex conditionals. This format is close to a 1-1 translation of the assembly.
# GOTO to the assembly input.--debug ("Debug info"): include debug information inline with the code, such as basic block boundaries & labels.--void ("Force void return type"): assume that the decompiled function has return type void. Alternatively: provide the function prototype in the context.m2c can generate an SVG representation of the control flow of a function, which can sometimes be helpful to untangle complex loops or early returns.
Pass --visualize on the command line, or use the "Visualize" button on the website. The output will be an SVG file.
Example to produce C & assembly visualizations of my_fn():
python3 ./m2c.py --visualize=c --context ctx.c -f my_fn my_asm.s > my_fn_c.svg
python3 ./m2c.py --visualize=asm --context ctx.c -f my_fn my_asm.s > my_fn_asm.svgmips_to_c.pyThis tool was originally known as mips_to_c. As part of the rename, deprecated command line arguments were removed.
When migrating to m2c, note the following changes to the CLI:
./mips_to_c.py becomes ./m2c.py-f FN, --function FN--rodata is no longer needed: my_text.s --rodata my_data.s becomes my_text.s my_data.s--compiler has been replaced by --target: --compiler gcc becomes --target mips-gcc-c--structs is now the default behavior: remove --structs from the argumentsThere is much low-hanging fruit still. Take a look at the issues if you want to help out.
We use black to auto-format our code, mypy for type checking and coverage for unit tests. We recommend using pre-commit to ensure only auto-formatted code is committed. To set these up, run:
pip install pre-commit black mypy coverage
pre-commit installYour commits will then be automatically formatted per commit. You can also manually run black on the command-line.
Type annotations are used for all Python code. mypy should pass without any errors.
To get pretty graph visualizations, install graphviz using pip and globally on your system (e.g. sudo apt install graphviz), and pass the --visualize flag.
There is a small test suite, which works as follows:
./run_tests.py to see if any tests have changed output.
These tests run the decompiler on a small corpus of assembly../run_tests.py --overwrite to write changed tests to disk, and commit resultant changes.It's possible to use the entire corpus of assembly files from decompilation projects as regression tests.
For now, the output of these tests are not tracked in version control.
You need to run ./run_tests.py --overwrite ... before making any code changes to create the baseline output.
As an example, if you have the oot project cloned locally in the parent directory containing m2c, the following will decompile all of its assembly files.
./run_tests.py --project ../oot --project-with-context ../ootThis has been tested with:
ver/us or ver/jp subfolder, e.g. --project ../papermario/ver/usThe following bash can be used in each decompilation project to create a "universal" context file that can be used for decompiling any assembly file in the project.
This creates ctx.c in the project directory.
cd mm # Or oot, papermario, etc.
find include/ src/ -type f -name "*.h" | sed -e 's/.*/#include "\0"/' > ctx_includes.c
tools/m2ctx.py ctx_includes.cThe build system in the MM decompilation project is currently being re-written.
It uses "transient" assembly that is not checked in, and in the normal build process it re-groups .rodata sections by function.
To use the MM project, run the following to just build the transient assembly files (and avoid running split_asm.py).
cd mm
make distclean
make setup
make asm/disasm.depThe repository should be setup correctly if there are asm/code, asm/boot, and asm/overlays folders with .asm files, but there should not be an asm/non_matchings folder.
Code branch coverage can be computed by running ./run_tests.py --coverage.
By default, this will generate an HTML coverage report ./htmlcov/index.html.
You are encouraged to add new end-to-end tests using the ./tests/add_test.py script.
For MIPS tests, you'll need the IDO cc compiler and the spimdisasm pip package.
A good reference test to base your new test on is array-access.
Create a new directory in tests/end_to_end, and write the orig.c test case.
If you want the test to pass in C context, also add irix-o2-flags.txt & irix-g-flags.txt files.
After writing these files, run add_test.py with the path to the new orig.c file, as shown below.
This example assumes that the IDO compiler is available from the OOT decompilation project.
You should change this exported path to match your system.
export IDO_CC=$HOME/oot/tools/ido_recomp/linux/7.1/cc
./tests/add_test.py $PWD/tests/end_to_end/my-new-test/orig.cThis should create irix-o2.s and irix-g.s files in your test directory.
Now, run ./run_tests.py --overwrite to invoke the decompiler and write the output to irix-o2-out.c and irix-g-out.c.
Finally, git add your test to track it.
./run_tests.py --overwrite
git add tests/end_to_end/my-new-testFor PowerPC, the MWCC_CC environment variable should be set to point to a PPC cc binary (mwcceppc.exe),
and on non-Windows, WINE set to point to wine or equivalent (wibo also works).
You can include m2c as a dependency in your project with Poetry
by adding the following to your pyproject.toml:
[tool.poetry.dependencies]
m2c = {git="https://github.com/matt-kempster/m2c.git"}If your project does not use pyproject.toml for dependencies, you can add the following
to your requirements.txt file instead:
m2c @ git+https://github.com/matt-kempster/m2c.git
# To specify a specific Git ref, such as a commit, tag, or branch:
m2c @ git+https://github.com/matt-kempster/m2c.git@[YOUR REF HERE]When installed as a Python package, a standalone command entrypoint is provided which can run the CLI.
m2c [options] [-t <target>] [--context <context file>] [-f <function name>] <asmfile>...