Re: [PATCH v2 5.15 1/2] mm: replace free hugepage folios after migration

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [High] mm/hugetlb: isolated free pages in free_huge_pages break
  reservations and offline
- [Low] mm: missing Fixes trailer on commit fixing hugetlb migration in
  alloc_contig_range()
--

commit d4646288ad2f50427e782b8be06fdbc2d46b6d5b
Author: yangge <yangge1116@126.com>

mm: replace free hugepage folios after migration

This commit adds replace_free_hugepage_folios() after migrating in-use
hugetlb pages in alloc_contig_range() and skips MIGRATE_ISOLATE pages in
dequeue_huge_page_node_exact() so freed hugepages in the isolated range are
returned to the buddy allocator instead of being reused from the hugetlb
pool.

Link: https://lkml.kernel.org/r/1734503588-16254-1-git-send-email-yangge1116@126.com
Link: https://lkml.kernel.org/r/1736582300-11364-1-git-send-email-yangge1116@126.com

[Severity: Low]
In the commit message, this is a bug fix for migrating in-use hugetlb pages
in alloc_contig_range(). Should it include:

  Fixes: ae37c7ff79f1 ("mm: make alloc_contig_range handle in-use hugetlb pages")

[Severity: High]
In mm/hugetlb.c at line 1099, does skipping MIGRATE_ISOLATE pages in
dequeue_huge_page_node_exact() while still counting them in
h->free_huge_pages allow migration to consume non-isolated free pages
backing h->resv_huge_pages?

When an in-use hugepage inside an isolated range (from offline_pages() or
alloc_contig_range()) is migrated and freed back to h->hugepage_freelists,
it remains counted in h->free_huge_pages. On the next hugepage migration,
alloc_huge_page_nodemask() (mm/hugetlb.c, line 2184) sees
h->free_huge_pages - h->resv_huge_pages > 0, and
dequeue_huge_page_node_exact() skips the isolated free page and dequeues a
non-isolated free page that was needed to back h->resv_huge_pages.

Once the free pages backing h->resv_huge_pages have been swapped into the
MIGRATE_ISOLATE range (or accounted as available by gather_surplus_pages()
at mm/hugetlb.c, line 2231), dissolve_free_huge_page() (mm/hugetlb.c, line
1971) hits h->free_huge_pages - h->resv_huge_pages == 0 and fails
offline_pages() with -EBUSY after already shrinking h->max_huge_pages. In
addition, page faults on reserved hugetlb mappings fail in
alloc_huge_page() (mm/hugetlb.c, line 2836) with -ENOSPC (delivering SIGBUS)
when dequeue_huge_page_vma() skips the isolated pages and surplus allocation
is disallowed (such as when nr_overcommit_huge_pages is 0).

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106220445.2859944-1-mkbund@amazon.com?part=1