Linux creator Linus Torvalds has published a kernel branch built with help from the AI coding assistant Claude Code. It aims to speed up small file reads from many threads.
A new branch called noref-filemap-read-with-llm has appeared in Torvalds' main torvalds/linux.git repository. The repository normally hosts little more than the master branch, and the "LLM" in the name points to the AI assistance behind the work.
The code adapts the kernel's memory management so that data can be read from the page cache without taking folio references. It targets a known problem: when many threads make small reads of the same file, they contend on folio reference counting. Avoiding those references is meant to remove that bottleneck. The source says the effect of the patch series is significant, although it does not give figures here.
Torvalds was open about his use of AI. In the main patch message he wrote that the code, the test programs, the review behind the list of changes and the first draft of the cover letter were done with an LLM coding assistant, Claude Code, under his direction. He added that the measurements were run on his own machine.
Torvalds has previously said he sees AI as a useful tool. He has reportedly used it for kernel debugging and has tried vibe coding outside the kernel tree with Google's Antigravity. The patches currently live only on the branch. It remains to be seen whether he will merge them during the Linux 7.4 merge window, which follows the Linux 7.3 release.
Source: Lobster
