Received-SPF: pass (google.com: domain of linux-kernel+bounces-81540-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <20240225123215.86503-1-ioworker0@gmail.com> <c9d4fc77-0121-4415-a3f8-9fd7159992e2@arm.com>
In-Reply-To: <c9d4fc77-0121-4415-a3f8-9fd7159992e2@arm.com>
From: Lance Yang <ioworker0@gmail.com>
Date: Mon, 26 Feb 2024 21:54:07 +0800
Message-ID: <CAK1f24kwiH_tw-y+gL=Sgjkms3AxriAhoWvA0tTJh-Mc2Ek4zQ@mail.gmail.com>
Subject: Re: [PATCH 1/1] mm/madvise: enhance lazyfreeing with mTHP in madvise_free
To: Ryan Roberts <ryan.roberts@arm.com>
Cc: akpm@linux-foundation.org, zokeefe@google.com, shy828301@gmail.com, 
	david@redhat.com, mhocko@suse.com, wangkefeng.wang@huawei.com, 
	songmuchun@bytedance.com, peterx@redhat.com, minchan@kernel.org, 
	linux-mm@kvack.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hey Ryan,

Thanks for taking time to review!

On Mon, Feb 26, 2024 at 9:00=E2=80=AFPM Ryan Roberts <ryan.roberts@arm.com>=
 wrote:
>
> Hi Lance,
>
> Thanks for working on this!
>
>
> On 25/02/2024 12:32, Lance Yang wrote:
> > This patch improves madvise_free_pte_range() to correctly
> > handle large folio that is smaller than PMD-size
>
> When you say "correctly handle" are you implying there is a bug with the =
current
> implementation or are you just saying you can optimize this to improve
> performance? I'm not convinced there is a bug, but I agree there is certa=
inly
> room for performance improvement.

I agree with your point, and will update the changelog in v2.

Thanks again for your time!

Best,
Lance

>
> Thanks,
> Ryan
>
> > (for example, 16KiB to 1024KiB[1]). It=E2=80=99s probably part of
> > the preparation to support anonymous multi-size THP.
> >
> > Additionally, when the consecutive PTEs are mapped to
> > consecutive pages of the same large folio (mTHP), if the
> > folio is locked before madvise(MADV_FREE) or cannot be
> > split, then all subsequent PTEs within the same PMD will
> > be skipped. However, they should have been MADV_FREEed.
> >
> > Moreover, this patch also optimizes lazyfreeing with
> > PTE-mapped mTHP (Inspired by David Hildenbrand[2]). We
> > aim to avoid unnecessary folio splitting if the large
> > folio is entirely within the given range.
> >
> > On an Intel I5 CPU, lazyfreeing a 1GiB VMA backed by
> > PTE-mapped folios of the same size results in the following
> > runtimes for madvise(MADV_FREE) in seconds (shorter is better):
> >
> > Folio Size  |    Old     |    New     |  Change
> > ----------------------------------------------
> >       4KiB  |  0.590251  |  0.590264  |     0%
> >      16KiB  |  2.990447  |  0.182167  |   -94%
> >      32KiB  |  2.547831  |  0.101622  |   -96%
> >      64KiB  |  2.457796  |  0.049726  |   -98%
> >     128KiB  |  2.281034  |  0.030109  |   -99%
> >     256KiB  |  2.230387  |  0.015838  |   -99%
> >     512KiB  |  2.189106  |  0.009149  |   -99%
> >    1024KiB  |  2.183949  |  0.006620  |   -99%
> >    2048KiB  |  0.002799  |  0.002795  |     0%
> >
> > [1] https://lkml.kernel.org/r/20231207161211.2374093-5-ryan.roberts@arm=
com
> > [2] https://lore.kernel.org/linux-mm/20240214204435.167852-1-david@redh=
at.com/
> >
> > Signed-off-by: Lance Yang <ioworker0@gmail.com>
> > ---
> >  mm/madvise.c | 69 +++++++++++++++++++++++++++++++++++++++++++---------
> >  1 file changed, 58 insertions(+), 11 deletions(-)
> >
> > diff --git a/mm/madvise.c b/mm/madvise.c
> > index cfa5e7288261..bcbf56595a2e 100644
> > --- a/mm/madvise.c
> > +++ b/mm/madvise.c
> > @@ -676,11 +676,43 @@ static int madvise_free_pte_range(pmd_t *pmd, uns=
igned long addr,
> >                */
> >               if (folio_test_large(folio)) {
> >                       int err;
> > +                     unsigned long next_addr, align;
> >
> > -                     if (folio_estimated_sharers(folio) !=3D 1)
> > -                             break;
> > -                     if (!folio_trylock(folio))
> > -                             break;
> > +                     if (folio_estimated_sharers(folio) !=3D 1 ||
> > +                         !folio_trylock(folio))
> > +                             goto skip_large_folio;
> > +
> > +                     align =3D folio_nr_pages(folio) * PAGE_SIZE;
> > +                     next_addr =3D ALIGN_DOWN(addr + align, align);
> > +
> > +                     /*
> > +                      * If we mark only the subpages as lazyfree,
> > +                      * split the large folio.
> > +                      */
> > +                     if (next_addr > end || next_addr - addr !=3D alig=
n)
> > +                             goto split_large_folio;
> > +
> > +                     /*
> > +                      * Avoid unnecessary folio splitting if the large
> > +                      * folio is entirely within the given range.
> > +                      */
> > +                     folio_test_clear_dirty(folio);
> > +                     folio_unlock(folio);
> > +                     for (; addr !=3D next_addr; pte++, addr +=3D PAGE=
_SIZE) {
> > +                             ptent =3D ptep_get(pte);
> > +                             if (pte_young(ptent) || pte_dirty(ptent))=
 {
> > +                                     ptent =3D ptep_get_and_clear_full=
(
> > +                                             mm, addr, pte, tlb->fullm=
m);
> > +                                     ptent =3D pte_mkold(ptent);
> > +                                     ptent =3D pte_mkclean(ptent);
> > +                                     set_pte_at(mm, addr, pte, ptent);
> > +                                     tlb_remove_tlb_entry(tlb, pte, ad=
dr);
> > +                             }
> > +                     }
> > +                     folio_mark_lazyfree(folio);
> > +                     goto next_folio;
> > +
> > +split_large_folio:
> >                       folio_get(folio);
> >                       arch_leave_lazy_mmu_mode();
> >                       pte_unmap_unlock(start_pte, ptl);
> > @@ -688,13 +720,28 @@ static int madvise_free_pte_range(pmd_t *pmd, uns=
igned long addr,
> >                       err =3D split_folio(folio);
> >                       folio_unlock(folio);
> >                       folio_put(folio);
> > -                     if (err)
> > -                             break;
> > -                     start_pte =3D pte =3D
> > -                             pte_offset_map_lock(mm, pmd, addr, &ptl);
> > -                     if (!start_pte)
> > -                             break;
> > -                     arch_enter_lazy_mmu_mode();
> > +
> > +                     /*
> > +                      * If the large folio is locked before madvise(MA=
DV_FREE)
> > +                      * or cannot be split, we just skip it.
> > +                      */
> > +                     if (err) {
> > +skip_large_folio:
> > +                             if (next_addr >=3D end)
> > +                                     break;
> > +                             pte +=3D (next_addr - addr) / PAGE_SIZE;
> > +                             addr =3D next_addr;
> > +                     }
> > +
> > +                     if (!start_pte) {
> > +                             start_pte =3D pte =3D pte_offset_map_lock=
(
> > +                                     mm, pmd, addr, &ptl);
> > +                             if (!start_pte)
> > +                                     break;
> > +                             arch_enter_lazy_mmu_mode();
> > +                     }
> > +
> > +next_folio:
> >                       pte--;
> >                       addr -=3D PAGE_SIZE;
> >                       continue;
>