AI Code Review Prompts Initiative Making Progress For The Linux Kernel

Longtime Linux kernel developer Chris Mason, known for creating Btrfs, is making progress on an initiative using AI review prompts for LLM-assisted code review of kernel patches. He recently posted updates to the Linux kernel mailing list, introducing changes that break the review process into individual tasks. This task-based approach aims to review smaller diff chunks separately, which often reduces token costs by avoiding sending context for the entire diff repeatedly, and is reported to catch more bugs. Mason uses a Python script to process changes in bulk, extracting modified functions and call graphs to further minimize token usage. The new setup includes specific tasks for code chunk review, checking past lore threads, verifying Fixes: tags, deep-diving into syzkaller fixes, and generating a final report. Mason, who is a Meta engineer, is seeking feedback on these changes, particularly regarding context window management across tasks, as this effort aims to make AI/LLM-assisted review accurate and helpful for upstream Linux kernel development.

(Source：Phoronix)