Re: [PATCH 0/4] mm: fix reclaim storms in defrag_mode

[Johannes Weiner <hannes@cmpxchg.org>]

On Wed, Jul 01, 2026 at 05:31:11PM -0700, Shakeel Butt wrote:
> On Fri, Jun 26, 2026 at 02:21:16PM -0400, Johannes Weiner wrote:
> > As we deployed vm.defrag_mode=1 into Meta production, some workloads
> > regressed with recurring pressure spikes and swap storms (which in turn
> > triggered userspace OOM rules on pressure and swap utilization levels).
> > 
> > Tracing pinned this to non-movable 
> 
> allocation?
> 
> > requests spinning and reclaiming

    ^^^

non-movable requests :)

> > unproductively when kswapd/kcompactd are overwhelmed. Direct reclaim
> > predominantly frees up pages in movable blocks, but those requests
> > cannot use that space under defrag_mode rules; 
> 
> Do we have these rules documented somewhere?

Kind of. The crux is that defrag_mode enforces ALLOC_NOFRAGMENT in the
slowpath as well (whereas vanilla just does it in the fastpath). So
it's documented to the extent that ALLOC_NOFRAGMENT is.

> > and it is unlikely to
> > free up whole blocks incidentally for __rmqueue_claim() to work.
> > 
> > This series fixes it by making non-movable requests participate in
> > pageblock production in the allocator slowpath.
> 
> Sorry after reading above sentence I didn't get what those allocators will do
> things differently after the series (I still have to go through the series).

Let me add:

[...] - meaning, they will invoke direct reclaim and direct compaction
with pageblock_order.

Does that clarify it a bit?