Automating Linux Kernel Documentation for Safety-Critical Compliance through Large Language Models

Explore how Large Language Models can automatically generate safety-critical compliance documentation for the Linux kernel in this 31-minute conference talk from the Linux Plumbers Conference. Learn about the challenges of adopting Linux in safety-critical industries like aerospace, where standards such as DO-178C require standardized documentation that the kernel currently lacks. Discover the research methodology used to evaluate four different LLM families—Meta Llama, StarCoder2, Mistral Devstral, and Google Gemma—for generating compliant documentation, with a focus on the Linux kernel's ftrace utility as a case study. Examine the evaluation criteria including documentation validity, GPU utilization, and throughput performance across these models. Understand why Mistral Devstral emerged as the most effective solution for producing accurate and standards-aligned documentation. Consider the broader implications for bridging the gap between open-source software and regulated environments, and explore critical questions about integration with existing toolchains like Sphinx and SPDX, reproducibility requirements, traceability standards, and expert validation processes that certification authorities might accept for aerospace and other compliance-driven domains.