Linus Torvalds publishes Linux kernel page-cache patches largely written with Anthropic's Claude Code
Linus Torvalds submitted a set of Linux kernel patches that were largely written with Anthropic's Claude Code, according to a ZDNET report. The series targets cache-line contention in the page cache, a known issue since 2024 that slows multi-core systems. Torvalds disclosed in his Git submission that the code, test programs, review and first draft of the text were produced by the LLM coding assistant under his direction. This is a development and tooling story, not a security incident, and no vulnerabilities, exploitation or threat actors are reported. As of Oct. 13, the branch had not been merged into mainline Linux 7.3-rc6, and its merge timetable is unknown.
Technical Details
- When several threads repeatedly read small amounts of data from the same cached file, they take and release references to the backing folio. The resulting atomic reference-count updates force cores to compete for a shared cache line.
- Torvalds measured that with eight threads issuing 16-byte
pread()calls against the same page on a four-core, eight-thread Intel Core i7-8565U laptop, the shared cache line accounted for 99.92% of measured cross-core load (HITM). - The patches defer reuse of memory removed from the page cache until a read-copy-update (RCU) grace period has elapsed. This lets a reader in an RCU read-side section copy data directly into a user buffer without taking a folio reference.
- Reported results: eight-thread throughput rose from about 9.9-10.4 million to 30-33 million reads per second, and kernel cycles per read fell from about 2,950 to 725. Single-thread reads improved from about 452-454 to 408-409 cycles. The gains come from a narrow small-read test on a laptop, not general Linux performance.
- Known trade-offs: delayed memory reuse leaves pages unavailable during the grace period, which Torvalds called "the part of this series that is least solved." There is also a regression when a read targets a user page that is not yet mapped, which incurs an extra failed exception before falling back to the existing path.
The article also traces Torvalds' shift from AI skepticism to hands-on use. He cautioned that AI must be used very carefully on real and important code.
Linha do tempo
-
Linus Torvalds publishes Linux kernel page-cache patches largely written with Anthropic's Claude Code
ler a análise desta fonte ocultar
Linus Torvalds submitted a set of Linux kernel patches that were largely written with Anthropic's Claude Code, according to a ZDNET report. The series targets cache-line contention in the page cache, a known issue since 2024 that slows multi-core systems. Torvalds disclosed in his Git submission that the code, test programs, review and first draft of the text were produced by the LLM coding assistant under his direction. This is a development and tooling story, not a security incident, and no vulnerabilities, exploitation or threat actors are reported. As of Oct. 13, the branch had not been merged into mainline Linux 7.3-rc6, and its merge timetable is unknown.
Technical Details
- When several threads repeatedly read small amounts of data from the same cached file, they take and release references to the backing folio. The resulting atomic reference-count updates force cores to compete for a shared cache line.
- Torvalds measured that with eight threads issuing 16-byte
pread()calls against the same page on a four-core, eight-thread Intel Core i7-8565U laptop, the shared cache line accounted for 99.92% of measured cross-core load (HITM). - The patches defer reuse of memory removed from the page cache until a read-copy-update (RCU) grace period has elapsed. This lets a reader in an RCU read-side section copy data directly into a user buffer without taking a folio reference.
- Reported results: eight-thread throughput rose from about 9.9-10.4 million to 30-33 million reads per second, and kernel cycles per read fell from about 2,950 to 725. Single-thread reads improved from about 452-454 to 408-409 cycles. The gains come from a narrow small-read test on a laptop, not general Linux performance.
- Known trade-offs: delayed memory reuse leaves pages unavailable during the grace period, which Torvalds called "the part of this series that is least solved." There is also a regression when a read targets a user page that is not yet mapped, which incurs an extra failed exception before falling back to the existing path.
The article also traces Torvalds' shift from AI skepticism to hands-on use. He cautioned that AI must be used very carefully on real and important code.