Linux 6.10 To Account For NUMA Node When Allocating Per-CPU Cpumasks

In addition to a SLUB optimization for extreme scenarios, faster AES-XTS disk/file encryption for modern Intel/AMD CPUs, and other performance optimizations on the way for Linux 6.10, another minor one was queued up this week.

Queued into TIP.git's x86/cpu branch this week is a change to take the NUMA node into account when allocating per-CPU cpumasks.

With most per-CPU cpumasks being accessed from their own local processor cores, it makes sense for them to be allocated local to the given NUMA node.

The patch does acknowledge the intent to improve performance for the per-CPU cpumasks but without any indication of any observed performance benefits. The upstream mailing list thread isn't resolving for any additional insight, but in any event this change just touches a dozen lines of code and is now queued via x86/cpu for introduction in the Linux 6.10 merge window that will kick off in May.