[RFC PATCH 4/4] mm/vmscan: do not skip node reclaim when only anon is reclaimable

[Ridong Chen <ridong.chen@linux.dev>]

From: Ridong Chen <chenridong@xiaomi.com>

node_reclaim() bailed out with "return 0" whenever both unmapped page
cache and reclaimable slab were at or below their limits:

	if (node_pagecache_reclaimable(pgdat) <= pgdat->min_unmapped_pages &&
	    node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) <=
	    pgdat->min_slab_pages)
		return 0;

But min_unmapped_pages and min_slab_pages only describe file and slab;
they say nothing about anonymous memory, which node reclaim is otherwise
happy to reclaim (may_swap is set, swappiness applies).  So a node with
little page cache and slab but plenty of reclaimable anon was skipped
entirely, even though there was anon to free.

Require in addition that anon cannot be reclaimed before bailing out, so
the early return only triggers when none of the three types has anything
to reclaim.  When anon is reclaimable the per-type flags leave file and
slab alone, so shrink_node reclaims anon only; checking
can_reclaim_anon_pages() here also avoids entering reclaim just to scan
without freeing anything when swap and demotion are both unavailable.

Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Ridong Chen <chenridong@xiaomi.com>
---
 mm/vmscan.c | 14 +++++---------
 1 file changed, 5 insertions(+), 9 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 5a3f67b3ba32..8c706807da44 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -7945,18 +7945,14 @@ unsigned long node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned i
 	};
 
 	/*
-	 * Node reclaim reclaims unmapped file backed pages and
-	 * slab pages if we are over the defined limits.
-	 *
-	 * A small portion of unmapped file backed pages is needed for
-	 * file I/O otherwise pages read by file I/O will be immediately
-	 * thrown out if the node is overallocated. So we do not reclaim
-	 * if less than a specified percentage of the node is used by
-	 * unmapped file backed pages.
+	 * min_unmapped_pages and min_slab_pages only gate file and slab.
+	 * Bail out only when both are under their limits and anon cannot
+	 * be reclaimed either, so we do not skip reclaimable anon.
 	 */
 	if (node_pagecache_reclaimable(pgdat) <= pgdat->min_unmapped_pages &&
 	    node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B) <=
-	    pgdat->min_slab_pages)
+	    pgdat->min_slab_pages &&
+	    !can_reclaim_anon_pages(NULL, pgdat->node_id, &sc))
 		return 0;
 
 	/*
-- 
2.34.1