IT之家 AIUpdated

Linus Torvalds, father of Linux, releases the first AI-assisted developed patch: concurrent file read throughput improved by about 3x

ITHome, October 15 — Tech media outlet lwn published a blog post yesterday (October 14), reporting that Linux creator Linus Torvalds pushed a new development branch named noref-filemap-read-with-llm in his personal Git…

Image source · IT之家 AI

IT之家 News from October 15: the tech media site lwn published a blog post yesterday (October 14), reporting that the father of Linux, Linus Torvalds, pushed a new development branch named noref-filemap-read-with-llm in his personal Git repository.Formally submitted his first kernel patch series developed with the assistance of a large language model (LLM).

The outlet noted that under normal circumstances, in the torvalds / linux.git repository, Torvalds only maintains the master branch and rarely develops on branches; however, a new noref-filemap-read-with-llm branch has recently appeared, and its suffix contains "llm" — these are some kernel improvements he made with the help of AI / LLM (specifically Claude Code).

Torvalds committed a new patch to this branch on October 13, titled "improving page reference count scalability," mainly addressing a long-standing bottleneck in Linux memory management. When multiple threads simultaneously perform high-frequency, small reads of the same file, the traditional mechanism repeatedly triggers reference count updates on the page cache, which easily causes severe race contention among multi-core CPUs.

Torvalds candidly stated that the core code of this patch set, the test programs, the vulnerability review, and the initial draft of the announcement documentation were all completed in collaboration with Claude Code, the AI programming tool launched by Anthropic, under his personal guidance.

After testing with some of his computer's security protections disabled, under a specific test workload of 8 threads simultaneously reading the same file, read throughput jumped directly from the original 9.9 million to 10.4 million operations per second to 30 million to 33 million operations per second (equivalent to soaring from about 10M to 30M+), a performance improvement of roughly 3 times; meanwhile, the kernel clock cycles consumed per read also dropped sharply from about 2950 to 725.

IT之家 note: this patch is currently still on an experimental branch and has not yet been merged into the official mainline of Linux 7.4.

Moreover, in the recent Maintainers Summit and in another cleanup of the assoc_array codebase, Torvalds likewise used an LLM to refactor obscure legacy code, even uncovering in one go 3 hidden bugs that had been lurking in the Linux system for as long as 13 years (traceable to the first commit in 2013).

Original source

IT之家 AI

Content notes

Original publication and rights belong to the source.

Machine translation · Refer to the original