The commits
98b211d6415f ("madvise: convert madvise_free_pte_range() to use
a folio")
fc986a38b670 ("mm: huge_memory: convert madvise_free_huge_pmd to
use a folio")

replaced the page_mapcount() with folio_mapcount() to check whether
the folio is shared by other mapping.

But it's not correct for large folio. folio_mapcount() returns the
total mapcount of large folio which is not suitable to detect whether
the folio is shared.

Use folio_estimated_sharers() which returns a estimated number of
shares. That means it's not 100% correct. But it should be OK for
madvise case here.

Fixes: 98b211d6415f ("madvise: convert madvise_free_pte_range() to use a folio")
Fixes: fc986a38b670 ("mm: huge_memory: convert madvise_free_huge_pmd to use a folio")
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Reviewed-by: Yu Zhao <yuzhao@google.com>
---
 mm/huge_memory.c | 2 +-
 mm/madvise.c     | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index eb3678360b97..68c890875257 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1613,7 +1613,7 @@ bool madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 	 * 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_mapcount(folio) != 1)
+	if (folio_estimated_sharers(folio) != 1)
 		goto out;
 
 	if (!folio_trylock(folio))
diff --git a/mm/madvise.c b/mm/madvise.c
index 148b46beb039..55bdf641abfa 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -678,7 +678,7 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
 		if (folio_test_large(folio)) {
 			int err;
 
-			if (folio_mapcount(folio) != 1)
+			if (folio_estimated_sharers(folio) != 1)
 				break;
 			if (!folio_trylock(folio))
 				break;
-- 
2.39.2

On Sat, 29 Jul 2023 00:13:56 +0800 Yin Fengwei <fengwei.yin@intel.com> wrote:

> Fixes: 98b211d6415f ("madvise: convert madvise_free_pte_range() to use a folio")
> Fixes: fc986a38b670 ("mm: huge_memory: convert madvise_free_huge_pmd to use a folio")

Having two Fixes: for one patch presumably makes backporting more
complicated and adds risk of making mistakes.

So I have split this into a three-patch series and I've fixed up the patch naming:

Subject: madvise:madvise_cold_or_pageout_pte_range(): don't use mapcount() against large folio for sharing check
Subject: madvise:madvise_free_huge_pmd(): don't use mapcount() against large folio for sharing check
Subject: madvise:madvise_free_pte_range(): don't use mapcount() against large folio for sharing check

I haven't added cc:stable at this time - that awaits the description of
user-visible effects.

Hi Andrew,

On 7/29/2023 1:41 AM, Andrew Morton wrote:
> On Sat, 29 Jul 2023 00:13:56 +0800 Yin Fengwei <fengwei.yin@intel.com> wrote:
> 
>> Fixes: 98b211d6415f ("madvise: convert madvise_free_pte_range() to use a folio")
>> Fixes: fc986a38b670 ("mm: huge_memory: convert madvise_free_huge_pmd to use a folio")
> 
> Having two Fixes: for one patch presumably makes backporting more
> complicated and adds risk of making mistakes.
> 
> So I have split this into a three-patch series and I've fixed up the patch naming:
> 
> Subject: madvise:madvise_cold_or_pageout_pte_range(): don't use mapcount() against large folio for sharing check
> Subject: madvise:madvise_free_huge_pmd(): don't use mapcount() against large folio for sharing check
> Subject: madvise:madvise_free_pte_range(): don't use mapcount() against large folio for sharing check
Thanks a lot for your kind help. Will be careful for the future patches.

> 
> I haven't added cc:stable at this time - that awaits the description of
> user-visible effects.
The impact of the patch:
  Without the patch, when user calls madvise() with MADV_COLD, MADV_PAGEOUT
  and MADV_FREE, it's likely THP pages will be skipped. With the patch,
  It's likely the THP pages will be split to pages which will be made code,
  reclaimed and freed.


Regards
Yin, Fengwei