Linux kernel -stable discussions
 help / color / mirror / Atom feed
From: Usama Arif <usama.arif@linux.dev>
To: "David Hildenbrand (Arm)" <david@kernel.org>,
	Zi Yan <ziy@nvidia.com>,
	Baolin Wang <baolin.wang@linux.alibaba.com>,
	"Liam R. Howlett" <liam@infradead.org>,
	Nico Pache <nico.pache@linux.dev>,
	Ryan Roberts <ryan.roberts@arm.com>, Dev Jain <dev.jain@arm.com>,
	Barry Song <baohua@kernel.org>, Lance Yang <lance.yang@linux.dev>,
	Kiryl Shutsemau <kas@kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Lorenzo Stoakes <ljs@kernel.org>,
	Vlastimil Babka <vbabka@kernel.org>,
	Mike Rapoport <rppt@kernel.org>,
	Suren Baghdasaryan <surenb@google.com>,
	Michal Hocko <mhocko@suse.com>, Jann Horn <jannh@google.com>,
	Ying Huang <huang.ying.caritas@gmail.com>
Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	stable@vger.kernel.org, Chris Mason <mason@kernel.org>,
	Sashiko <sashiko-bot@kernel.org>
Subject: Re: [PATCH] mm/madvise: fix MADV_FREE handling for PMD-mapped THPs
Date: Thu, 19 Nov 2026 09:21:04 -0800	[thread overview]
Message-ID: <299b3f93-f7dd-490a-8a6b-68afe4b0e473@linux.dev> (raw)
In-Reply-To: <20261119-madv_free_thp-v1-1-a312919e1ee5@kernel.org>



On 19/11/2026 08:04, David Hildenbrand (Arm) wrote:
> After MADV_FREE cleared the dirty flag of an anonymous folio, reclaim will
> later free the folio without writing it back if no mappings are dirty.
> 
> If our folio is still in the swap cache, the swap entries then point at
> swap slots that were never written, exposing stale content of the swap
> device. Bad.
> 
> madvise_free_pte_range() therefore frees the swap slots first, and skips
> the folio if that is impossible. madvise_free_huge_pmd() does not, and a
> PMD-mapped THP can be in the swap cache if try_to_unmap() failed during
> reclaim, for example due to a racing mlock().
> 
> Further, madvise_free_huge_pmd() checks whether the THP is mapped by
> other processes before taking the folio lock. Another process that still
> has swap entries for the THP can concurrently map it from the swap cache,
> and we might end up clearing the dirty flag of a shared folio.
> 
> Fix it by factoring out the folio handling of madvise_free_pte_range()
> into madvise_free_try_prepare_folio(), and using it for PMD-mapped THPs
> as well. Hopefully this prevents further bugs.
> 
> For PMD-mapped THPs, this change implies:
> 
> 1. We free the swap slots of a THP in the swap cache, and skip the THP if
>    that is impossible.
> 2. We check the mapcount under the folio lock.
> 3. We no longer lock and check THPs that are neither dirty nor in the
>    swap cache, just like for PTE-mapped folios.
> 
> For PTE-mapped folios, this means:
> 
> 4. We check the mapcount once before taking the folio lock, to not lock
>    folios that are mapped by other processes.
> 
> Only compile madvise_free_huge_pmd() with CONFIG_ADVISE_SYSCALLS to keep
> !CONFIG_ADVISE_SYSCALLS builds working.
> 
> Chris and Sashiko reported the missing swap cache handling for PMD-mapped
> THPs, the other issue was found by code inspection.
> 
> This has medium urgency.
> 
> Fixes: bd4c82c22c36 ("mm, THP, swap: delay splitting THP after swapped out")
> Cc: stable@vger.kernel.org
> Reported-by: Chris Mason <mason@kernel.org>
> Reported-by: Sashiko <sashiko-bot@kernel.org>
> Closes: https://lore.kernel.org/sashiko-outbox-202086@kernel.org/
> Signed-off-by: David Hildenbrand (Arm) <david@kernel.org>
> ---
> I would queue this fairly soon on mm-core/for-next-fixes
> ---
>  mm/huge_memory.c | 19 ++++++-----------
>  mm/internal.h    |  1 +
>  mm/madvise.c     | 65 +++++++++++++++++++++++++++++---------------------------
>  3 files changed, 41 insertions(+), 44 deletions(-)
> 

Yeah looking at pmd swap, I imagine there might be more cases
which are correctly handled by pte path but not pmd.

Acked-by: Usama Arif <usama.arif@linux.dev>

> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 79d400ea9553d..e22daf846dbe3 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -2421,6 +2421,7 @@ vm_fault_t do_huge_pmd_numa_page(struct vm_fault *vmf)
>  	return 0;
>  }
>  
> +#ifdef CONFIG_ADVISE_SYSCALLS
>  /*
>   * Return true if we do MADV_FREE successfully on entire pmd page.
>   * Otherwise, return false.
> @@ -2455,21 +2456,13 @@ bool madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
>  	if (folio_is_zone_device(folio))
>  		goto out;
>  
> -	/*
> -	 * If other processes are mapping this folio, we couldn't discard
> -	 * the folio unless they all do MADV_FREE so let's skip the folio.
> -	 */
> -	if (folio_maybe_mapped_shared(folio))
> -		goto out;
> -
> -	if (!folio_trylock(folio))
> -		goto out;
> -
>  	/*
>  	 * If user want to discard part-pages of THP, split it so MADV_FREE
>  	 * will deactivate only them.
>  	 */
>  	if (next - addr != HPAGE_PMD_SIZE) {
> +		if (folio_maybe_mapped_shared(folio) || !folio_trylock(folio))
> +			goto out;
>  		folio_get(folio);
>  		spin_unlock(ptl);
>  		split_folio(folio);
> @@ -2478,9 +2471,8 @@ bool madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
>  		goto out_unlocked;
>  	}
>  
> -	if (folio_test_dirty(folio))
> -		folio_clear_dirty(folio);
> -	folio_unlock(folio);
> +	if (!madvise_free_try_prepare_folio(folio, 1))
> +		goto out;
>  
>  	if (pmd_young(orig_pmd) || pmd_dirty(orig_pmd)) {
>  		pmdp_invalidate(vma, addr, pmd);
> @@ -2498,6 +2490,7 @@ bool madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
>  out_unlocked:
>  	return ret;
>  }
> +#endif /* CONFIG_ADVISE_SYSCALLS */
>  
>  static inline void zap_deposited_table(struct mm_struct *mm, pmd_t *pmd)
>  {
> diff --git a/mm/internal.h b/mm/internal.h
> index 85cf7becf8588..96cb55d7816b1 100644
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -61,6 +61,7 @@ void lru_add_drain_cpu(int cpu);
>  void lru_add_drain_cpu_zone(struct zone *zone);
>  void folio_deactivate(struct folio *folio);
>  void folio_mark_lazyfree(struct folio *folio);
> +bool madvise_free_try_prepare_folio(struct folio *folio, int nr_mappings);
>  
>  /* mm/vmscan.c */
>  unsigned long zone_reclaimable_pages(struct zone *zone);
> diff --git a/mm/madvise.c b/mm/madvise.c
> index 0214efd2a014f..ebb4ce7daf6cb 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -661,6 +661,38 @@ static long madvise_pageout(struct madvise_behavior *madv_behavior)
>  	return 0;
>  }
>  
> +bool madvise_free_try_prepare_folio(struct folio *folio, int nr_mappings)
> +{
> +	if (!folio_test_swapcache(folio) && !folio_test_dirty(folio))
> +		return true;
> +
> +	if (folio_mapcount(folio) != nr_mappings || !folio_trylock(folio))
> +		return false;
> +
> +	/*
> +	 * KSM folios can gain new mappings at any time, so we must
> +	 * not clear their dirty flag.
> +	 */
> +	if (folio_test_ksm(folio))
> +		goto unlock;
> +
> +	/*
> +	 * Check under the folio lock that the folio is exclusive: we own all
> +	 * mappings, and there are no references from swap entries.
> +	 */
> +	if (folio_mapcount(folio) != nr_mappings)
> +		goto unlock;
> +	if (folio_test_swapcache(folio) && !folio_free_swap(folio))
> +		goto unlock;
> +
> +	folio_clear_dirty(folio);
> +	folio_unlock(folio);
> +	return true;
> +unlock:
> +	folio_unlock(folio);
> +	return false;
> +}
> +
>  static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
>  				unsigned long end, struct mm_walk *walk)
>  
> @@ -753,37 +785,8 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
>  			}
>  		}
>  
> -		if (folio_test_swapcache(folio) || folio_test_dirty(folio)) {
> -			if (!folio_trylock(folio))
> -				continue;
> -			/*
> -			 * KSM folios can gain new mappings at any time, so we must
> -			 * not clear their dirty flag.
> -			 */
> -			if (folio_test_ksm(folio)) {
> -				folio_unlock(folio);
> -				continue;
> -			}
> -
> -			/*
> -			 * If we have a large folio at this point, we know it is
> -			 * fully mapped so if its mapcount is the same as its
> -			 * number of pages, it must be exclusive.
> -			 */
> -			if (folio_mapcount(folio) != folio_nr_pages(folio)) {
> -				folio_unlock(folio);
> -				continue;
> -			}
> -
> -			if (folio_test_swapcache(folio) &&
> -			    !folio_free_swap(folio)) {
> -				folio_unlock(folio);
> -				continue;
> -			}
> -
> -			folio_clear_dirty(folio);
> -			folio_unlock(folio);
> -		}
> +		if (!madvise_free_try_prepare_folio(folio, folio_nr_pages(folio)))
> +			continue;
>  
>  		if (pte_young(ptent) || pte_dirty(ptent)) {
>  			clear_young_dirty_ptes(vma, addr, pte, nr, cydp_flags);
> 
> ---
> base-commit: 5e600d79135124a4eb8e9db30ad524224c267c07
> change-id: 20261119-madv_free_thp-3c2a9887a722
> 
> Best regards,
> --  
> David Hildenbrand (Arm) <david@kernel.org>


      parent reply	other threads:[~2026-11-19 17:21 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-19 16:04 [PATCH] mm/madvise: fix MADV_FREE handling for PMD-mapped THPs David Hildenbrand (Arm)
2026-11-19 16:44 ` Michal Hocko
2026-11-19 17:21 ` Usama Arif [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=299b3f93-f7dd-490a-8a6b-68afe4b0e473@linux.dev \
    --to=usama.arif@linux.dev \
    --cc=akpm@linux-foundation.org \
    --cc=baohua@kernel.org \
    --cc=baolin.wang@linux.alibaba.com \
    --cc=david@kernel.org \
    --cc=dev.jain@arm.com \
    --cc=huang.ying.caritas@gmail.com \
    --cc=jannh@google.com \
    --cc=kas@kernel.org \
    --cc=lance.yang@linux.dev \
    --cc=liam@infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=ljs@kernel.org \
    --cc=mason@kernel.org \
    --cc=mhocko@suse.com \
    --cc=nico.pache@linux.dev \
    --cc=rppt@kernel.org \
    --cc=ryan.roberts@arm.com \
    --cc=sashiko-bot@kernel.org \
    --cc=stable@vger.kernel.org \
    --cc=surenb@google.com \
    --cc=vbabka@kernel.org \
    --cc=ziy@nvidia.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox