Linus Torvalds used Anthropic’s Claude Code to attack a page-cache performance problem that kernel developers had struggled with for more than two years, submitting a 10-patch series that delivered a 23-fold throughput gain in a synthetic benchmark on his 64-thread Ryzen Threadripper. Folio reference-count contention had been flagged in 2024 by Oracle engineer Matthew Wilcox after surfacing in a customer workload, and on Torvalds’ laptop it accounted for 99.92 percent of cross-core load HITMs in eight-thread tests. Torvalds said Claude Code wrote the implementation, tests and initial documentation under his direction; he admitted he had spent 18 months running a broken version of his own fix. The original eight-patch branch roughly tripled eight-thread laptop throughput (10 million to 33 million reads/second, cycles per read about 2,950 to 725), while the revised series moved Threadripper throughput from 13.3 million to 304 million reads/second. Reviewers identified a race in filemap_read_rcu_batch() and questioned the new PG_rcu_free flag and memory reclamation; Torvalds rejected alternatives from Wilcox and passed further development to Kiryl Shutsemau, with Arm maintainer David Hildenbrand backing a 64-bit-only flag restriction.
Linus Torvalds let Claude Code fix the Linux page cache. Then reviewers found a race.
Linus Torvalds used Anthropic's Claude Code to attack a page-cache performance problem that kernel developers had struggled with for more than two years, submitting a 10-patch series that delivered a 23-fold throughput gain in a synthetic benchmark on his 64-thread Ryzen Threadripper. Folio reference-count contention had been flagged in 2024 by Oracle engineer Matthew Wilcox after surfacing in a customer workload, and on Torvalds' laptop it accounted for 99.92 percent of cross-core load HITMs in eight-thread tests. Torvalds said Claude Code wrote the implementation, tests and initial documentation under his direction; he admitted he had spent 18 months running a broken version of his own fix. The original eight-patch branch roughly tripled eight-thread laptop throughput (10 million to 33 million reads/second, cycles per read about 2,950 to 725), while the revised series moved Threadripper throughput from 13.3 million to 304 million reads/second. Reviewers identified a race in filemap_read_rcu_batch() and questioned the new PG_rcu_free flag and memory reclamation; Torvalds rejected alternatives from Wilcox and passed further development to Kiryl Shutsemau, with Arm maintainer David Hildenbrand backing a 64-bit-only flag restriction.
Source: The New Stack