[PATCH 0/2] mm/huge_memory: optimize migration when huge PMD needs split

[Wei Yang <richard.weiyang@gmail.com>]

This is a follow up optimization for commit 939080834fef ("mm/huge_memory: 
fix early failure try_to_migrate() when split huge pmd for shared THP").

When split_huge_pmd_locked() successfully split the PMD entry with @freeze
= true, it means each PTE entry are properly set to migration entry.  And 
we can return from try_to_migrate_one() directly. 

Currently it is done in a sub-optimal way: it always restarts the walk and go
through each PTE entry by page_vma_mapped_walk() which then skip all of them
when the PMD is split to migration entry.  

Let split_huge_pmd_locked() indicate whether it split PMD to migration entry,
so that to optimize migration if huge PMD needs split.

Also add a selftest to check the bug fixed in commit 939080834fef
("mm/huge_memory: fix early failure try_to_migrate() when split huge pmd for
shared THP") will not be introduced.

Cc: Gavin Guo <gavinguo@igalia.com>
Cc: "David Hildenbrand (Red Hat)" <david@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Lance Yang <lance.yang@linux.dev>

Wei Yang (2):
  mm/huge_memory: return true if split_huge_pmd_locked() split PMD to
    migration entry
  mm/selftests: add split_shared_pmd()

 include/linux/huge_mm.h                       |  9 ++-
 mm/huge_memory.c                              | 21 ++++--
 mm/rmap.c                                     | 11 ++-
 .../selftests/mm/split_huge_page_test.c       | 73 ++++++++++++++++++-
 4 files changed, 99 insertions(+), 15 deletions(-)

-- 
2.34.1