ARM Linux engineer Lorenzo Stoakes leveraged artificial intelligence to identify and address bottlenecks in the Linux kernel build process, resulting in significant speed improvements. While the AI generated substantial amounts of flawed code, it proved effective in pinpointing areas for optimization.
Stoakes posted a set of 23 patches designed to “significantly speed up kernel builds” by tackling single-threaded bottlenecks. Testing revealed a 36% faster build time for kernels with all modules enabled, up to 70% faster incremental builds, and approximately 90% faster “noop” builds.
The process involved using an unnamed large language model (LLM) to both identify bottlenecks and propose solutions. According to Stoakes, the LLM “generated a lot of code, much of it hideous,” requiring extensive auditing, rewriting, and editing of commit messages and comments. He also noted the LLM assisted with build runs, testing, debugging, and analysis, with Stoakes manually verifying the correctness of both the build and running kernels.
Specifically, the improvements affect areas such as Kbuild, kallsyms, modpost, objtool, mksysmap, and the Rust build system. The patches aim to parallelize single-threaded tasks and enhance the overall efficiency of the build process. Stoakes hopes the work will be incorporated into the mainline Linux kernel, leading to faster build times across various hardware configurations and kernel configurations. Detailed benchmark results are available with the patch series.
Read the original coverage
💬 Comments
📜 Comment Policy