[RFC PATCH v2 5/5] mm/vmscan: scale hopeless threshold for kswapd workers

[Ehab Ababneh]

The per-node failure counter advances once per worker balance pass
that makes no reclaim progress. With multiple kswapd workers, it can
reach MAX_RECLAIM_RETRIES while others are still reclaiming, marking
the node hopeless prematurely.

Scale the threshold by the active kswapd worker count so each worker
keeps the existing retry budget.

Signed-off-by: Ehab Ababneh <ehab.ababneh@xxxxxxxxx>
---
 mm/vmscan.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 9e051aac2203..61d46a376eb9 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -7940,7 +7940,8 @@ void kswapd_try_clear_hopeless(struct pglist_data *pgdat,
 
 bool kswapd_test_hopeless(pg_data_t *pgdat)
 {
-	return atomic_read(&pgdat->kswapd_failures) >= MAX_RECLAIM_RETRIES;
+	return atomic_read(&pgdat->kswapd_failures) >=
+		MAX_RECLAIM_RETRIES * READ_ONCE(current_kswapds_per_node);
 }
 
 #ifdef CONFIG_HIBERNATION
-- 
2.43.0