Received-SPF: pass (google.com: domain of linux-kernel+bounces-160979-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240426190253.541419-1-zi.yan@sent.com> <e274ad3b-bbeb-49b0-81bf-24fd33106f05@redhat.com>
 <CB058583-C99A-4897-B3AE-11C56D6E3A59@nvidia.com>
In-Reply-To: <CB058583-C99A-4897-B3AE-11C56D6E3A59@nvidia.com>
From: Barry Song <21cnbao@gmail.com>
Date: Sat, 27 Apr 2024 17:32:04 +0800
Message-ID: <CAGsJ_4wEGhL7rRPP_soiEt5bjnpa3zhPnU+8JrNLZvWdcRNuqg@mail.gmail.com>
Subject: Re: [PATCH v5] mm/rmap: do not add fully unmapped large folio to
 deferred split list
To: Zi Yan <ziy@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>, Andrew Morton <akpm@linux-foundation.org>, linux-mm@kvack.org, 
	"Matthew Wilcox (Oracle)" <willy@infradead.org>, Yang Shi <shy828301@gmail.com>, 
	Ryan Roberts <ryan.roberts@arm.com>, Lance Yang <ioworker0@gmail.com>, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Sat, Apr 27, 2024 at 3:20=E2=80=AFAM Zi Yan <ziy@nvidia.com> wrote:
>
> On 26 Apr 2024, at 15:08, David Hildenbrand wrote:
>
> > On 26.04.24 21:02, Zi Yan wrote:
> >> From: Zi Yan <ziy@nvidia.com>
> >>
> >> In __folio_remove_rmap(), a large folio is added to deferred split lis=
t
> >> if any page in a folio loses its final mapping. But it is possible tha=
t
> >> the folio is fully unmapped and adding it to deferred split list is
> >> unnecessary.
> >>
> >> For PMD-mapped THPs, that was not really an issue, because removing th=
e
> >> last PMD mapping in the absence of PTE mappings would not have added t=
he
> >> folio to the deferred split queue.
> >>
> >> However, for PTE-mapped THPs, which are now more prominent due to mTHP=
,
> >> they are always added to the deferred split queue. One side effect
> >> is that the THP_DEFERRED_SPLIT_PAGE stat for a PTE-mapped folio can be
> >> unintentionally increased, making it look like there are many partiall=
y
> >> mapped folios -- although the whole folio is fully unmapped stepwise.
> >>
> >> Core-mm now tries batch-unmapping consecutive PTEs of PTE-mapped THPs
> >> where possible starting from commit b06dc281aa99 ("mm/rmap: introduce
> >> folio_remove_rmap_[pte|ptes|pmd]()"). When it happens, a whole PTE-map=
ped
> >> folio is unmapped in one go and can avoid being added to deferred spli=
t
> >> list, reducing the THP_DEFERRED_SPLIT_PAGE noise. But there will still=
 be
> >> noise when we cannot batch-unmap a complete PTE-mapped folio in one go
> >> -- or where this type of batching is not implemented yet, e.g., migrat=
ion.
> >>
> >> To avoid the unnecessary addition, folio->_nr_pages_mapped is checked
> >> to tell if the whole folio is unmapped. If the folio is already on
> >> deferred split list, it will be skipped, too.
> >>
> >> Note: commit 98046944a159 ("mm: huge_memory: add the missing
> >> folio_test_pmd_mappable() for THP split statistics") tried to exclude
> >> mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does no=
t
> >> fix the above issue. A fully unmapped PTE-mapped order-9 THP was still
> >> added to deferred split list and counted as THP_DEFERRED_SPLIT_PAGE,
> >> since nr is 512 (non zero), level is RMAP_LEVEL_PTE, and inside
> >> deferred_split_folio() the order-9 folio is folio_test_pmd_mappable().
> >>
> >> Suggested-by: David Hildenbrand <david@redhat.com>
> >> Signed-off-by: Zi Yan <ziy@nvidia.com>
> >> ---
> >>   mm/rmap.c | 12 +++++++++---
> >>   1 file changed, 9 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/mm/rmap.c b/mm/rmap.c
> >> index 2608c40dffad..a9bd64ebdd9a 100644
> >> --- a/mm/rmap.c
> >> +++ b/mm/rmap.c
> >> @@ -1495,6 +1495,7 @@ static __always_inline void __folio_remove_rmap(=
struct folio *folio,
> >>   {
> >>      atomic_t *mapped =3D &folio->_nr_pages_mapped;
> >>      int last, nr =3D 0, nr_pmdmapped =3D 0;
> >> +    bool partially_mapped =3D false;
> >>      enum node_stat_item idx;
> >>      __folio_rmap_sanity_checks(folio, page, nr_pages, level);
> >> @@ -1515,6 +1516,8 @@ static __always_inline void __folio_remove_rmap(=
struct folio *folio,
> >>                                      nr++;
> >>                      }
> >>              } while (page++, --nr_pages > 0);
> >> +
> >> +            partially_mapped =3D !!nr && !!atomic_read(mapped);
> >
> > Nit: The && should remove the need for both !!.
>
> My impression was that !! is needed to convert from int to bool and I do
> find "!!int && !!int" use in the kernel. If this is unnecessary, Andrew
> can apply the fixup below. I can send a new version if it is really neede=
d.
>
> diff --git a/mm/rmap.c b/mm/rmap.c
> index a9bd64ebdd9a..c1fd5828409b 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1517,7 +1517,7 @@ static __always_inline void __folio_remove_rmap(str=
uct folio *folio,
>                         }
>                 } while (page++, --nr_pages > 0);
>
> -               partially_mapped =3D !!nr && !!atomic_read(mapped);
> +               partially_mapped =3D nr && atomic_read(mapped

Reviewed-by=EF=BC=9ABarry Song <baohua@kernel.org>

>                 break;
>         case RMAP_LEVEL_PMD:
>                 atomic_dec(&folio->_large_mapcount);
>
> >
> > Reviewed-by: David Hildenbrand <david@redhat.com>
>
> Thanks.
>
> --
> Best Regards,
> Yan, Zi