It's fairly well known that Linus Torvalds, the creator of Linux, has made public his stance on using AI to help in the debugging of the Linux kernel…he's all for it. But until now, the use of AI was limited to that task.

Now, it seems, Torvalds has decided to use AI in the development of new kernel features. His tool of choice: Claude Code.

"The code, the test programs, the review behind the list above and the first draft of this text were done with an LLM coding assistant (Claude Code) under my direction." Said Torvalds, in this recent patch message. He adds, "The measurements were run on my machine."

The code to which Torvalds was referring was for noref-filemap-read-with-llm, which is an idea he had for "A folio removed from the page cache goes back to the page allocator only after an RCU grace period."

Michael Larabel, from Phoronix spotted a new kernel branch appear, to which he said, "I was surprised to find a new kernel branch appear. Typically Torvalds isn't doing much work on branches in his main repository and typically just hosting the master branch."

What Torvalds discovered, with the help of Claude Code, was that instead of tracking every reader with a reference count, better performance can be had by using Linux's Read-Copy-Update (RCU) synchronization mechanism. The previous approach suffered under heavy CPU locking, whereas the new model uses lockless RCU Read by way of deferred memory freeing and reference-free copy.

The patches can be viewed here, but there's no definitive declaration if they will be merged into the upcoming 7.4 kernel.