This is motivated by an excessive reclaim issue we saw in a customer
workload and was associated with THP page faults and an unexpected
zone_reclaim_mode-like behavior that can happen for those.

The first patch is intended as a minimal targeted backportable fix for
the observed issue. I hope I'll be able to confirm it has fixed the
problem, but it might take some time.

The second patch is a more "let's take a step back and rethink what
we're doing here" so it's separate, as it changes a bit the semantics
of __GFP_NORETRY.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
---
Vlastimil Babka (2):
      mm, page_alloc, thp: prevent reclaim for __GFP_THISNODE THP allocations
      mm, page_alloc: fail costly __GFP_NORETRY allocations faster

 include/linux/gfp_types.h |  2 ++
 mm/page_alloc.c           | 33 +++------------------------------
 2 files changed, 5 insertions(+), 30 deletions(-)
---
base-commit: 8f0b4cce4481fb22653697cced8d0d04027cb1e8
change-id: 20251216-thp-thisnode-tweak-c9c2acb3a627

Best regards,
-- 
Vlastimil Babka <vbabka@suse.cz>