Re: [PATCH v3] mm/huge_memory: fix folio isn't locked in softleaf_to_folio()

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

On Fri, Mar 20, 2026 at 04:56:57PM +0800, Jinjiang Tu wrote:
>
> 在 2026/3/20 16:10, David Hildenbrand (Arm) 写道:
> > On 3/20/26 02:52, Jinjiang Tu wrote:
> > > 在 2026/3/20 6:51, Andrew Morton 写道:
> > > > On Thu, 19 Mar 2026 09:25:41 +0800 Jinjiang Tu <tujinjiang@huawei.com>
> > > > wrote:
> > > >
> > > > > On arm64 server, we found folio that get from migration entry isn't
> > > > > locked
> > > > > in softleaf_to_folio(). This issue triggers when mTHP splitting and
> > > > > zap_nonpresent_ptes() races, and the root cause is lack of memory
> > > > > barrier
> > > > > in softleaf_to_folio(). The race is as follows:
> > > > >
> > > > >      CPU0                                             CPU1
> > > > >
> > > > > deferred_split_scan()                              zap_nonpresent_ptes()
> > > > >     lock folio
> > > > >     split_folio()
> > > > >       unmap_folio()
> > > > >         change ptes to migration entries
> > > > >       __split_folio_to_order()
> > > > > softleaf_to_folio()
> > > > >         set flags(including PG_locked) for tail pages    folio =
> > > > > pfn_folio(softleaf_to_pfn(entry))
> > > > >         smp_wmb()
> > > > > VM_WARN_ON_ONCE(!folio_test_locked(folio))
> > > > >         prep_compound_page() for tail pages
> > > > >
> > > > > In __split_folio_to_order(), smp_wmb() guarantees page flags of tail
> > > > > pages
> > > > > are visible before the tail page becomes non-compound. smp_wmb() should
> > > > > be paired with smp_rmb() in softleaf_to_folio(), which is missed. As a
> > > > > result, if zap_nonpresent_ptes() accesses migration entry that stores
> > > > > tail pfn, softleaf_to_folio() may see the updated compound_head of tail
> > > > > page before page->flags.
> > > > Please describe the userspace-visible runtime effects of this bug.
> > > This issue will trigger VM_WARN_ON_ONCE() in pfn_swap_entry_folio().
> > But the impact is bigger, right, when callers rely on folio_test_anon() etc?
>
> Yes, the impact is unpredictable if CONFIG_DEBUG_VM isn't enabled.
>
> But for stable kernel before v6.19.9, this is a BUG_ON
>
>
>

I think important to note that the asserts are asserting consistency, and not
having consistency is a lot worse than a failed assert :)

I'd say more so that you're solving a race between folio split and
zap_nonpresent_ptes() leading to a folio incorrectly undergoing
modification without a folio lock being held.

Cheers, Lorenzo