Linus Torvalds, creator of the Linux kernel, is developing a patch to improve file reads using Anthropic’s AI coding assistant Claude Code, Phoronix reported on October 11, 2026, U.S. time. Torvalds’ development notes say he used Claude Code for code, test programs and initial drafts of explanatory text.

Phoronix reviewed the “noref-filemap-read-with-llm” development branch published in Torvalds’ repository. The changes aim to reduce contention when multiple threads read small amounts of data from the same file. In a commit dated October 10, Torvalds said that, under his direction, Claude Code helped write code and test programs, review known issues and draft explanatory text. He said he ran the performance measurements on his own machine.

The change concerns reads from the page cache, which temporarily keeps file data in memory. Previously, when multiple threads read the same data, competing updates to memory-management reference counts became a bottleneck. The new path delays reuse of memory removed from the cache, skips reference-count updates during reads and copies the data directly.

According to the patch description, Torvalds tested reading 16 bytes at a time from the start of the same file on a laptop with a four-core, eight-thread Intel Core i7-8565U. With eight threads, reads increased from about 9.9–10.4 million per second before the change to about 31–33 million afterward. These results apply to a specific operation and test setup; they do not show that overall Linux performance will be about three times faster.

The description also says some security mitigations and CONFIG_HARDENED_USERCOPY, which checks the safety of memory copies, were disabled during testing. Costs rise under conditions that trigger a page fault, the process of loading needed memory, such as when physical memory must be allocated for the destination region. The change is available on a development branch; Phoronix said whether it will be accepted during the Linux 7.4 merge window remains to be seen.

Linus Torvalds, creator of the Linux kernel, is developing a patch to improve file reads using Anthropic’s AI coding assistant Claude Code, Phoronix reported on October 11, 2026, U.S. time. Torvalds’ development notes say he used Claude Code for code, test programs and initial drafts of explanatory text.

Phoronix reviewed the “noref-filemap-read-with-llm” development branch published in Torvalds’ repository. The changes aim to reduce contention when multiple threads read small amounts of data from the same file. In a commit dated October 10, Torvalds said that, under his direction, Claude Code helped write code and test programs, review known issues and draft explanatory text. He said he ran the performance measurements on his own machine.

The change concerns reads from the page cache, which temporarily keeps file data in memory. Previously, when multiple threads read the same data, competing updates to memory-management reference counts became a bottleneck. The new path delays reuse of memory removed from the cache, skips reference-count updates during reads and copies the data directly.

According to the patch description, Torvalds tested reading 16 bytes at a time from the start of the same file on a laptop with a four-core, eight-thread Intel Core i7-8565U. With eight threads, reads increased from about 9.9–10.4 million per second before the change to about 31–33 million afterward. These results apply to a specific operation and test setup; they do not show that overall Linux performance will be about three times faster.

The description also says some security mitigations and CONFIG_HARDENED_USERCOPY, which checks the safety of memory copies, were disabled during testing. Costs rise under conditions that trigger a page fault, the process of loading needed memory, such as when physical memory must be allocated for the destination region. The change is available on a development branch; Phoronix said whether it will be accepted during the Linux 7.4 merge window remains to be seen.