Re: [RFC PATCH 0/6] mm/mglru: skip empty PUD subtrees during aging with PUD-level Bloom filter

On 08/06/26 at 06:29pm, Baoquan He wrote:
> Problem
> =======
> 
> MGLRU's aging walks every present PUD of every mm in the mm_list.  When
> an mm has no pages on a given NUMA node, all 512 PMDs in each PUD fail
> the existing PMD-level Bloom filter test, yet the walker still descends
> every PUD and iterates every PMD ??? pure waste.
> 
> These cross-node empty walks are structural: lru_gen_use_mm() sets
> mm->lru_gen.bitmap to -1 (all nodes) at each context switch, so every
> node's kswapd independently walks the same mm.  On a 2-node KVM guest,
> ~80% of all aging walks are empty; the worst-case leaf-traversal waste
> with a fully resident remote mm is ~73%.

A note on the design background and one limitation.

These patches replace my earlier per-mm empty_map approach, which skipped
an mm for K generations  (default 4) after an empty walk on a node. As
Kairui pointed out, that leaves a window of up to K generations during which
pages migrated into the node are not scanned by aging until the forced  rescan,
and it adds 16 bytes of per-mm state with awkward invalidation semantics.
Following his  suggestion, the series instead adds a PUD-level Bloom filter — a
coarser counterpart of the existing  PMD-level filter — so a 1GB subtree that
had no young entries for this lruvec last generation is skipped, with no per-mm
state and no blind window.

I will paste the empty_map patches to this thread for reviewers'
reference.