Linus Torvalds Uses Claude Code on Linux Kernel Work, With 8-Thread Read Test Showing Roughly 3x Throughput
A new branch in Linus Torvalds’ official Linux kernel repository, named noref-filemap-read-with-llm, has drawn attention because the suffix explicitly signals large language model involvement. In the patch notes, Torvalds said the core code, test cases, review work and draft documentation were produced with Claude Code under his direction. The branch targets a memory-management bottleneck in Linux’s page cache read path, where multiple threads repeatedly reading the same file contend on atomic reference counting for a folio. On Torvalds’ test laptop, an Intel Core i7-8565U system with 4 cores and 8 threads running ext4, the patch reportedly lifted throughput to about three times the previous level in an 8-thread test, while single-thread performance improved by about 10%. The repository also showed another branch, assoc-array-fixes, shortly afterward. According to the report, Torvalds used a large model to help untangle hard-to-read assoc-array code after developer Nils Ollrogge submitted a bug report and a fix. The code was then exercised in a user-space test framework. That process confirmed the reported issue and uncovered three more bugs, all traced back to the data structure’s original 2013 merge into the kernel.
