Re: [PATCH] mm/huge_memory: prevent losing PG_waiters in __split_folio_to_order()

["Lorenzo Stoakes (ARM)" <ljs@kernel.org>]

On Thu, Oct 29, 2026 at 10:08:44PM -0400, Zi Yan wrote:
> __split_folio_to_order() calls prep_compound_page() on each new head page
> to create a compound page after clear_compound_head(). prep_compound_page()
> uses non atomic __SetPageHead(). It is mostly OK, since the folio is locked
> and frozen and no one else is supposed to modify the folio's flags.
>
> However, code calling migration_entry_wait() can reach the folio by PFN,
> then access its flags. It is because softleaf_entry_wait_on_locked() used

Oh fun.

> by migration_entry_wait() sets PG_waiters concurrently without taking a
> reference and the frozen refcount during split cannot block it. Meanwhile,
> __SetPageHead() does a plain read and write, so PG_waiters atomically set
> by softleaf_entry_wait_on_locked() will be lost if the set happens after
> __SetPageHead()'s read and before its write.
>
> One example code interleaving:
>
> CPU0: __split_folio_to_order(new_order>0)  CPU1: do_swap_page()
>
> 1. clear_compound_head(new_head)
>    new_head->compound_info = 0
>    (page_folio(new_head) == new_head now)
>
>                                            2. migration_entry_wait()
>                                              softleaf_to_folio(entry)
>                                              page_folio() -> new_head
>                                              (not the old head any more)
>
> 3. prep_compound_page(new_head, new_order)
>      __SetPageHead(new_head)
>        tmp = new_head->flags.f  <-- load; PG_waiters == 0 in tmp
>
>                                            4. folio_set_waiters(new_head)
>                                              set_bit(PG_waiters, &flags.f)
>                                              enqueue on waitqueue, sleep
>
> 5.     new_head->flags.f = tmp | (1UL << PG_head)
>        <-- store; overwrites step 4, PG_waiters == 0 again
>
> Later, folio_unlock(new_folio) sees no PG_waiters, skips folio_wake_bit(),
> and CPU1 can remain blocked indefinitely.
>
> Fix it by adding the atomic version SetPageHead() and
> prep_compound_page_atomic() using it. __split_folio_to_order() uses
> prep_compound_page_atomic() to avoid the issue without adding an atomic ops
> to other prep_compound_page() users.
>
> Reported-by: David Hildenbrand <david@kernel.org>
> Closes: https://lore.kernel.org/all/DMHE8BEEER1H.2BIY52NXATPIA@nvidia.com/
> Fixes: c010d47f107f6 ("mm: thp: split huge page to any lower order pages")
> Signed-off-by: Zi Yan <ziy@nvidia.com>

Nasty one. Some nits below but change looks sensible to me so:

Acked-by: Lorenzo Stoakes (ARM) <ljs@kernel.org>

BTW AI flags an hw poison thing doing the same:

memory_failure() does TestSetPageHWPoison(p) on an arbitrary subpage before
taking a reference (mm/memory-failure.c:2458), which races the same way with the
plain flag clone and with folio_set_order()'s plain _flags_1 store.

> Cc: stable@vger.kernel.org
> ---
> The issue was raised by David when reviewing Ye Liu's "mm/huge_memory: use
> READ_ONCE() when cloning folio flags during split" [1].
>
> [1] https://lore.kernel.org/all/20261029-data_race_huge_mem-v1-1-112e331291c1@linux.dev/
> ---
>  include/linux/page-flags.h |  1 +
>  mm/huge_memory.c           | 10 +++++++++-
>  mm/internal.h              |  2 ++
>  mm/page_alloc.c            | 20 ++++++++++++++++++--
>  4 files changed, 30 insertions(+), 3 deletions(-)
>
> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> index b0ddc652e76cc..08492ab3da895 100644
> --- a/include/linux/page-flags.h
> +++ b/include/linux/page-flags.h
> @@ -854,6 +854,7 @@ static __always_inline int PageHead(const struct page *page)
>  	return test_bit(PG_head, &page->flags.f);
>  }
>
> +SETPAGEFLAG(Head, head, PF_ANY)
>  __SETPAGEFLAG(Head, head, PF_ANY)
>  __CLEARPAGEFLAG(Head, head, PF_ANY)
>  CLEARPAGEFLAG(Head, head, PF_ANY)
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index bd44d58e86845..319b038eba89d 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3779,7 +3779,15 @@ static void __split_folio_to_order(struct folio *folio, int old_order,
>  		 */
>  		clear_compound_head(new_head);
>  		if (new_order) {
> -			prep_compound_page(new_head, new_order);
> +			/*
> +			 * new_head is recognized as a folio after
> +			 * clear_compound_head().
> +			 * softleaf_entry_wait_on_locked() can set PG_waiters
> +			 * on it without taking a reference, so its frozen state
> +			 * cannot prevent it. Use the atomic version to avoid
> +			 * losing a concurrently set PG_waiters.
> +			 */
> +			prep_compound_page_atomic(new_head, new_order);
>  			folio_set_large_rmappable(new_folio);
>  		}
>
> diff --git a/mm/internal.h b/mm/internal.h
> index aeebb3e49ed11..9b8864a8540aa 100644
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -807,6 +807,8 @@ static inline void prep_compound_tail(struct page *tail,
>  	VM_WARN_ON_ONCE(tail->private);
>  }
>
> +void prep_compound_page_atomic(struct page *page, unsigned int order);
> +
>  #if defined CONFIG_COMPACTION || defined CONFIG_CMA
>
>  /*
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 143776e671abb..ace1359468046 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -682,18 +682,34 @@ static inline bool pcp_allowed_order(unsigned int order)
>   * This usage means that zero-order pages may not be compound.
>   */
>
> -void prep_compound_page(struct page *page, unsigned int order)
> +static void __prep_compound_head_and_tail(struct page *page, unsigned int order)

NIT: can this just be __prepare_compound_page()? I find the head + tail bit
quite confusing. The __ implies it's an internal function.

>  {
>  	int i;
>  	int nr_pages = 1 << order;
>
> -	__SetPageHead(page);
>  	for (i = 1; i < nr_pages; i++)
>  		prep_compound_tail(page + i, page, order);
>
>  	prep_compound_head(page, order);
>  }
>

NIT: Maybe add a comment describing this like prep_compound_page_atomic()?

> +void prep_compound_page(struct page *page, unsigned int order)
> +{
> +	__SetPageHead(page);
> +	__prep_compound_head_and_tail(page, order);
> +}
> +
> +/*
> + * PG_head is set atomically whereas the other metadata is set with plain stores.
> + * It should be used when @page's flags can be modified concurrently,
> + * e.g., in __split_folio_to_order().
> + */
> +void prep_compound_page_atomic(struct page *page, unsigned int order)
> +{
> +	SetPageHead(page);
> +	__prep_compound_head_and_tail(page, order);
> +}
> +
>  static inline void set_buddy_order(struct page *page, unsigned int order)
>  {
>  	set_page_private(page, order);
>
> ---
> base-commit: d865f2677c44cb2171b833d6775cf692756fa8be
> change-id: 20261029-fix-split_folio_to_order-race-087a2a9fbf7c
>
> Best regards,
> --
> Yan, Zi
>

--
Cheers, Lorenzo