Linus Torvalds Posts First LLM-Assisted Kernel Patch Series

The Patch Series

Linus Torvalds has posted the first Linux kernel patch series he worked on with the help of a large language model. The fairly short series of ten patches addresses a longstanding problem in folio lookup.

The series enhances folio lookup by noticing when folios are hot and setting a delay that holds the folio until after an RCU period. Using that RCU delay, the code performs not just the folio lookup itself under RCU, which has been done for a long time, but also the actual folio data accesses. No refcount taking is needed for the fastpath, where the data can be read entirely within an RCU-protected region as long as the user copy does not cause page faults.

The end result is an order-of-magnitude speedup for a read()-heavy benchmark.

Revealed at the Maintainers Summit

Torvalds spoke about the work at the recently concluded Maintainers Summit, saying he had been using an LLM while developing the series. He has now posted the patches for review, as reported by LWN on October 14, 2026.

Review Finds a Race Condition

Review of the series has already identified a race condition in the code, so further work will be needed before the patches can move forward.

Community Debate Over AI Assistance

The article's comments section saw discussion about the example the move sets. One reader argued that Torvalds is in favor of AI for experienced, existing developers, but warned that the tool is being used to lower the bar for getting new patches accepted. That reader worried that beginners would see the kernel creator using AI and assume they could do the same without understanding the development process and its rules.

Another commenter saw no conflict between those positions, comparing the situation to an excavator that produces different results in the hands of a professional versus a layman.

Sources: lwn.net