[$] Improving the per-CPU memory allocator
The 2026 Linux Storage, Filesystem, Memory Management, and BPF Summit addressed performance issues with the kernel's per-CPU memory allocator. Harry Yoo led discussions on improving allocation and initialization processes, highlighting scalability and initialization cost concerns. Proposed solutions included dual-mode per-CPU counters and integrating the allocator with the slab allocator.
- ▪Harry Yoo led a session on improving the per-CPU memory allocator at the 2026 Linux Summit.
- ▪The current allocator has scalability issues due to a global lock and expensive initialization costs.
- ▪Proposals for improvement include dual-mode counters and closer integration with the slab allocator.
Opening excerpt (first ~120 words) tap to expand
Ready to give LWN a try? With a subscription to LWN, you can stay current with what is happening in the Linux and free-software community and take advantage of subscriber-only site features. We are pleased to offer you a free trial subscription, no credit card required, so that you can see for yourself. Please, join us! By Jonathan CorbetMay 19, 2026 LSFMM+BPF There are many places in the kernel where performance can be improved by using per-CPU data. But, as it turns out, the kernel's allocator for per-CPU data has some performance problems of its own. Harry Yoo led a session in the memory-management track of the 2026 Linux Storage, Filesystem, Memory Management, and BPF Summit to explore ways to address those problems and accelerate the allocation and initialization of per-CPU data.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at LWN.net (Linux Weekly News).