[RFC,0/3] Fix and refactor do_{huge_pmd_}numa_page()

Message

Zi Yan July 12, 2024, 2:44 a.m. UTC

From: Zi Yan <[e-mail]>

First patch adds the missing sysctl_numa_balancing_mode check. Second patch
introduces folio_has_cpupid() to replace open coded folio last_cpupid check.
Third patch consolidates code in do_numa_page() and do_huge_pmd_numa_page()
by moving more common code into numa_migrate_prep() (renamed to
numa_migrate_check() in the patch).

The RFC is mainly for third patch. It changes the original code behavior:
1. do_huge_pmd_numa_page() did not check shared folios to set TNF_SHARED.
2. do_huge_pmd_numa_page() did not check and skip zone device folios.



Zi Yan (3):
  memory tiering: read last_cpupid correctly in do_huge_pmd_numa_page()
  memory tiering: introduce folio_has_cpupid() check
  mm/migrate: move common code to numa_migrate_check (was
    numa_migrate_prep)

 include/linux/memory-tiers.h |  8 ++++
 kernel/sched/fair.c          |  3 +-
 mm/huge_memory.c             | 31 +++++---------
 mm/internal.h                |  5 ++-
 mm/memory-tiers.c            | 17 ++++++++
 mm/memory.c                  | 82 +++++++++++++++++-------------------
 mm/mprotect.c                |  3 +-
 7 files changed, 80 insertions(+), 69 deletions(-)