grok the linux kernel via a cmake shim
I'm sure there are great ways to spelunk the kernel with an ide... This is my attempt to rig something together using a simple ruby script and a roll of duct tape.
- create
compile_commands.json
for kernel using an intercepted build - use ruby script to convert
compile_commands.json
into an IDE friendlyCMakeLists.txt
(Note: not fully tested. Requires some prior knowledge to work around typos etc)
- Install
scan-build
(a build interceptor)
## requires python
pip install scan-build --user
## installs intercept-build tool in ~/.local/bin
OR
- Install Bear
(Another build interceptor; wont work on recent MacOs, likely thwarted by
selinux
)
git clone https://github.com/rizsotto/Bear && cd Bear
mkdir build && cd build
cmake .. && make
sudo make install
cd ../..
- Grab the kernel and build
git clone git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git
cd linux-stable
git reset --hard v4.14-rc2 ## random known good version
make defconfig
## Note that we intercept the build
## _Either_
bear make -j 12 ## create compile_commands.json
## _Or_
~/.local/bin/intercept-build make -j 12 ## create compile_commands.json
cd ..
- Use our ruby script
git clone https://github.com/habemus-papadum/kernel-grok.git
cd linux-stable
../kernel-grok/generate_cmake ## creates CMakeLists.txt
## test that generated cmake is valid
mkdir build
cd build
cmake .. && make -j12
- Now open linux-stable/CMakeLists.txt in an ide (tried testing with clion; though takes about 10 minutes to index the project initially; at first it seemed like an epic fail, but seems quite usable after initial indexing and then restarting)
- note
generate-cmake -d
option if not using clion or if cmake build directory is not one level under kernel source root
- using compilation database generated by
bear
/intercept-build
, a cmake shim is created that creates a single cmakeOBJECT
library (this is a somewhat obscure library type that as opposed toSHARED
performs no linking). - The shim should properly properly compile -- though it just creates object files that are then abandoned (not linked into a final kernel). This means your ide can be used not only for semantic navigation e.g. it knows where to find headers and knows about
-Ddefines
and so can jump to defintions and find usages etc, but also allows for testing whether edits to existing files compile -- perhaps 90% of the typical dev workflow. - Analogy might be Xcode projects generated by CMake. Can be used to make some edits, but larger things like adding files, etc cannot be done in Xcode. So after big changes e.g. new file, new kernel config, you must repeat intercepted KBuild, regenerate shim, and then reload project in ide
- Use ide compilation to test if your edits compile (only worth it because it is easy to navigate to errors when you ide does the build). Use KBuild to create a proper image to run in qemu etc.
- NOTE: Everything depends on the kernel actually being built at least once with normal Kbuild. There are generated files that are needed.
- NOTE: Although a CMakeLists.txt is included in the repo for providing a quick look at what is being done, it should not be used -- GENERATE YOUR OWN
- NOTE: linux assumes builds are done from the root src directory e.g. "in-tree". Some hacking is done to allow the cmake shim to work out of tree (again, cf.
-d
option as well as the actual ruby code (which is pretty simple)) - NOTE: An alternative would be to use {spac}emacs + rtags -- TODO: compare and constrast vs clion
- A similar approach can be taken to spelunk through other non-cmake projects , though each project will require some tinkering