On Tue, Apr 11, 2023 at 03:12:34PM +0200, Michal Hocko <mhocko@xxxxxxxx> wrote: > > Oh I see, you mean scanning all processes in all cpusets and scanning > > all processes globally are equivalent. > > Why cannot you simple select a process from the cpuset the allocating > process belongs to? I thought the whole idea was to handle well > partitioned workloads. Ah, I was confused by the top_cpuset implementation. The iteration should then start in nearest_hardwall_ancestor(task_cs(current)). (in the 1st approximation). The nodes_intersect/nodes_subset/nodes_equal/whatnot heuristics is secondary. HTH, Michal
Attachment:
signature.asc
Description: PGP signature
