The first step is setting up the build system. Let’s assume you’re about
to port Frida to run on Linux/MIPS. As Frida already supports Linux, all
we need to do is add the architecture-specific bits.
Porting the build system
Depending on the architecture, you may need to tweak releng/machine_spec.py.
Review the generated machine file in build/, e.g. build/frida-linux-mips.txt,
to make sure the toolchain is configured correctly.
Building frida-gum
This is the lowest level component and where most of the porting effort is
typically needed. To build it, run:
$ git clone https://github.com/frida/frida-gum.git
$ cd frida-gum
$ make
This will probably not succeed, but should at least get the environment set up.
Once you get the code to compile, it’s time to focus on getting tests passing:
$ make test
You can also run a single test, e.g.
$ FRIDA_TEST_OPTIONS="--test-args='-p /Core/Process/process_modules' -v" make test
Porting frida-gum
Add the directory gum/backend-mips by duplicating for example
gum/backend-arm64,
and then search-replace everything. The important part to port here is
guminterceptor-mips.c and gumspinlock-mips.c. You should leave
gumstalker-mips.c as a stub, as it’s an advanced feature that takes a lot
of effort to port.
Building frida-core
Now that frida-gum works, it’s time to repeat the same process for frida-core.
Porting frida-core
This should only be a matter of porting the injector. The implementation is here
and the recommended approach is to follow the #if X86 breadcrumbs to port
the architecture-specific bits. For a walkthrough of the Linux injector, check
out our presentation here.
