A proposed change to Linux’s slab allocator was reportedly queued in slab/for-next during the week of September 26, 2026, ahead of the Linux 7.4 merge window expected in late October. The reported results include a 24% improvement in the will-it-scale mmap1 test.
How the proposed sheaf refill works
Sheaves are per-CPU, array-based caches used by the kernel memory allocator. The proposed change affects their prefill path: instead of drawing objects from partial slabs, it can take objects from full sheaves in the allocator’s “barn.” Any leftovers stay there in a partial sheaf.
The sheaf being refilled stays in place. The design is intended to free space on the barn’s full list for future RCU sheaves and to avoid a more costly refill from partial slabs under list_lock.
What the benchmark figures measure
The reported 24% improvement applies to will-it-scale mmap1. Separate results put barn_get at +22,904,539.81% and barn_put at +21,422,544.07%. These are figures for distinct, specifically named benchmarks; they do not establish gains of the same size in everyday applications.
