mm/memory: fix mapcount / refcount sanity check for mTHP reuse

The following WARNING was triggered during swap stress test with mTHP
enabled:

[ 6609.335758] ------------[ cut here ]------------
[ 6609.337758] WARNING: CPU: 82 PID: 755116 at mm/memory.c:3794 do_wp_page+0x1084/0x10e0
[ 6609.340922] Modules linked in: zram virtiofs
[ 6609.342699] CPU: 82 UID: 0 PID: 755116 Comm: sh Kdump: loaded Not tainted 6.15.0-rc1+ #1429 PREEMPT(voluntary)
[ 6609.347620] Hardware name: Red Hat KVM/RHEL-AV, BIOS 0.0.0 02/06/2015
[ 6609.349909] RIP: 0010:do_wp_page+0x1084/0x10e0
[ 6609.351532] Code: ff ff 48 c7 c6 80 ba 49 82 4c 89 ef e8 95 fd fe ff 0f 0b bd f5 ff ff ff e9 43 fb ff ff 41 83 a9 bc 12 00 00 01 e9 5c fb ff ff <0f> 0b e9 a6 fc ff ff 65 ff 00 f0 48 0f b
a 6d 00 1f 0f 83 82 fc ff
[ 6609.357959] RSP: 0000:ffffc90002273d40 EFLAGS: 00010287
[ 6609.359915] RAX: 000000000000000f RBX: 0000000000000000 RCX: 000fffffffe00000
[ 6609.362606] RDX: 0000000000000010 RSI: 000055a119ac1000 RDI: ffffea000ae6ec00
[ 6609.365143] RBP: ffffea000ae6ec68 R08: 84000002b9bb1025 R09: 000055a119ab6000
[ 6609.367569] R10: ffff8881caa2ad80 R11: 0000000000000000 R12: ffff8881caa2ad80
[ 6609.370255] R13: ffffea000ae6ec00 R14: 000055a119ac1c9c R15: ffffc90002273dd8
[ 6609.373007] FS:  00007f08e467f740(0000) GS:ffff88a07c214000(0000) knlGS:0000000000000000
[ 6609.375999] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 6609.377946] CR2: 000055a119ac1c9c CR3: 00000001adfd6005 CR4: 0000000000770eb0
[ 6609.380376] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 6609.382853] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 6609.385216] PKRU: 55555554
[ 6609.386141] Call Trace:
[ 6609.387017]  <TASK>
[ 6609.387718]  ? ___pte_offset_map+0x1b/0x110
[ 6609.389056]  __handle_mm_fault+0xa51/0xf00
[ 6609.390363]  ? exc_page_fault+0x6a/0x140
[ 6609.391629]  handle_mm_fault+0x13d/0x360
[ 6609.392856]  do_user_addr_fault+0x2f2/0x7f0
[ 6609.394160]  ? sigprocmask+0x77/0xa0
[ 6609.395375]  exc_page_fault+0x6a/0x140
[ 6609.396735]  asm_exc_page_fault+0x26/0x30
[ 6609.398224] RIP: 0033:0x55a1050bc18b
[ 6609.399567] Code: 8b 3f 4d 85 ff 74 40 41 39 5f 18 75 f2 49 8b 7f 08 44 38 27 75 e9 4c 89 c6 4c 89 45 c8 e8 bd 83 fa ff 4c 8b 45 c8 85 c0 75 d5 <41> 83 47 1c 01 48 83 c4 28 4c 89 f8 5b 4
1 5c 41 5d 41 5e 41 5f 5d
[ 6609.405971] RSP: 002b:00007ffcf5f37d90 EFLAGS: 00010246
[ 6609.407737] RAX: 0000000000000000 RBX: 00000000182768fa RCX: 0000000000000000
[ 6609.410151] RDX: 00000000000000fa RSI: 000055a105175c7b RDI: 000055a119ac1c60
[ 6609.412606] RBP: 00007ffcf5f37de0 R08: 000055a105175c7b R09: 0000000000000000
[ 6609.414998] R10: 000000004d2dfb5a R11: 0000000000000246 R12: 0000000000000050
[ 6609.417193] R13: 00000000000000fa R14: 000055a119abaf60 R15: 000055a119ac1c80
[ 6609.419268]  </TASK>
[ 6609.419928] ---[ end trace 0000000000000000 ]---

The WARN_ON here is simply incorrect.  The refcount here must be at least
the mapcount, not the opposite.  Each mapcount must have a corresponding
refcount, but the refcount may increase if other components grab the
folio, which is acceptable.  Meanwhile, having a mapcount larger than
refcount is a real problem.

So fix the WARN_ON condition.

Link: https://lkml.kernel.org/r/20250425074325.61833-1-ryncsn@gmail.com
Fixes: 1da190f4d0a6 ("mm: Copy-on-Write (COW) reuse support for PTE-mapped THP")
Signed-off-by: Kairui Song <kasong@tencent.com>
Reported-by: Kairui Song <kasong@tencent.com>
Closes: https://lore.kernel.org/all/CAMgjq7D+ea3eg9gRCVvRnto3Sv3_H3WVhupX4e=k8T5QAfBHbw@mail.gmail.com/
Suggested-by: David Hildenbrand <david@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/mm/memory.c b/mm/memory.c
index ba3ea0a82f7f770b6cacba4526ba9c8c4929ddad..49199410805cd043b660be09cfddb3184852bbe4 100644 (file)
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3751,7 +3751,7 @@ static bool __wp_can_reuse_large_anon_folio(struct folio *folio,
 
        /* Stabilize the mapcount vs. refcount and recheck. */
        folio_lock_large_mapcount(folio);
-       VM_WARN_ON_ONCE(folio_large_mapcount(folio) < folio_ref_count(folio));
+       VM_WARN_ON_ONCE_FOLIO(folio_large_mapcount(folio) > folio_ref_count(folio), folio);
 
        if (folio_test_large_maybe_mapped_shared(folio))
                goto unlock;