Formalizing The LLVMLinux Project: Clang'ing Kernels

Interest in building the Linux kernel through the LLVM/Clang compiler rather than GCC continues to grow. The consolidated LLVMLinux project was announced last week.

Interest -- and some out-of-tree-success -- in building the Linux kernel with Clang is not new; it's been covered many times on Phoronix already. Since late 2010 (when it was first covered on Phoronix), it became possible to build the Linux kernel with LLVM/Clang, albeit at the time it relied upon non-mainline patches to LLVM/Clang and the Linux kernel. Upstream LLVM/Clang now works, but there's still non-mainline changes needed for supporting a Clang'ed Linux kernel on x86 and ARM.

Earlier this year within the ARM world, building the Linux kernel with Clang was extensively talked about at the Linux Foundation Collaboration Summit by Qualcomm's Mark Charlebois and David Kipping. Just a few weeks back, Clang'ing the Linux kernel was talked about at LinuxCon San Diego by Bryce Adelstein-Lelbach, one of the original developers interested in using this alternative compiler for producing Linux x86 kernel binaries.

So last week now what was announced on the llvmdev list to LLVM developers is the LLVMLinux project. "I'd like to announce the LLVMLinux project. This project aims to fully build the Linux kernel using Clang/LLVM on the various architectures supported by the Linux kernel. The project consolidates the work of the lll-project, the PAX team, and Mark Charlebois' work on the ARM kernel."

It's nothing really new to announce and the consolidated work continues to be available from llvm.linuxfoundation.org as it has been in recent months. "The goal is for this project to be a testing/proving ground for these updates and to get patches that make this possible upstream to their respective projects (Clang/LLVM and the Linux Kernel)."

Among the troubling spots still for the mainline Linux kernel in being built by non-GCC compilers namely down to the use of variable length arrays in structs, Kbuild support for Clang, use of explicit register variables, segment reference errors with __refdata annotations, and EXPORT_SYMBOL of inline functions. As it applies to LLVM developers, there's a few compiler flags that the Linux kernel prefers but are not handled by Clang: -fdelete-null-pointer-checks, --fno-inline-functions-called-once, --Wno-unused-but-set-variable, --mabi=aapcs-linux.

In the x86 world their work is currently based upon the Linux 3.3 kernel but is able to boot a console and desktop. With ARM, however, they're more closely following the latest upstream tip of Linus Torvalds' branch. Module unloading is also completely broken for Clang-built kernels. While an x86 Clang Linux kernel desktop may boot, there's also other open bugs preventing network / firewall / drivers from working properly.

While not the Linux kernel, FreeBSD 10 can be built using Clang and over there in the BSD world it's becoming their default compiler to deprecate GCC. There's also been experiments within the Debian camp.