Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230705171213.2843068-1-surenb@google.com> <20230705171213.2843068-2-surenb@google.com>
 <10c8fe17-fa9b-bf34-cb88-c758e07c9d72@redhat.com> <CAJuCfpFBh647trAjgPfr0Wcd=7V2gbHUnBe8mR4Pgdmrzh6Hxg@mail.gmail.com>
 <20230705230647.twq3n5nb2iabr7uk@revolver>
In-Reply-To: <20230705230647.twq3n5nb2iabr7uk@revolver>
From:   Suren Baghdasaryan <surenb@google.com>
Date:   Wed, 5 Jul 2023 17:20:06 -0700
Message-ID: <CAJuCfpG1GCwvvOWgtHkGCNLk-emOsb_sA2hki5dMriAQzRQQNg@mail.gmail.com>
Subject: Re: [PATCH v3 1/2] fork: lock VMAs of the parent process when forking
To:     "Liam R. Howlett" <Liam.Howlett@oracle.com>,
        Suren Baghdasaryan <surenb@google.com>,
        David Hildenbrand <david@redhat.com>,
        akpm@linux-foundation.org, jirislaby@kernel.org,
        jacobly.alt@gmail.com, holger@applied-asynchrony.com,
        hdegoede@redhat.com, michel@lespinasse.org, jglisse@google.com,
        mhocko@suse.com, vbabka@suse.cz, hannes@cmpxchg.org,
        mgorman@techsingularity.net, dave@stgolabs.net,
        willy@infradead.org, peterz@infradead.org, ldufour@linux.ibm.com,
        paulmck@kernel.org, mingo@redhat.com, will@kernel.org,
        luto@kernel.org, songliubraving@fb.com, peterx@redhat.com,
        dhowells@redhat.com, hughd@google.com, bigeasy@linutronix.de,
        kent.overstreet@linux.dev, punit.agrawal@bytedance.com,
        lstoakes@gmail.com, peterjung1337@gmail.com, rientjes@google.com,
        chriscli@google.com, axelrasmussen@google.com, joelaf@google.com,
        minchan@google.com, rppt@kernel.org, jannh@google.com,
        shakeelb@google.com, tatashin@google.com, edumazet@google.com,
        gthelen@google.com, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, stable@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, Jul 5, 2023 at 4:07=E2=80=AFPM Liam R. Howlett <Liam.Howlett@oracle=
.com> wrote:
>
> * Suren Baghdasaryan <surenb@google.com> [230705 13:24]:
> > On Wed, Jul 5, 2023 at 10:14=E2=80=AFAM David Hildenbrand <david@redhat=
.com> wrote:
> > >
> > > On 05.07.23 19:12, Suren Baghdasaryan wrote:
> > > > When forking a child process, parent write-protects an anonymous pa=
ge
> > > > and COW-shares it with the child being forked using copy_present_pt=
e().
> > > > Parent's TLB is flushed right before we drop the parent's mmap_lock=
 in
> > > > dup_mmap(). If we get a write-fault before that TLB flush in the pa=
rent,
> > > > and we end up replacing that anonymous page in the parent process i=
n
> > > > do_wp_page() (because, COW-shared with the child), this might lead =
to
> > > > some stale writable TLB entries targeting the wrong (old) page.
> > > > Similar issue happened in the past with userfaultfd (see flush_tlb_=
page()
> > > > call inside do_wp_page()).
> > > > Lock VMAs of the parent process when forking a child, which prevent=
s
> > > > concurrent page faults during fork operation and avoids this issue.
> > > > This fix can potentially regress some fork-heavy workloads. Kernel =
build
> > > > time did not show noticeable regression on a 56-core machine while =
a
> > > > stress test mapping 10000 VMAs and forking 5000 times in a tight lo=
op
> > > > shows ~5% regression. If such fork time regression is unacceptable,
> > > > disabling CONFIG_PER_VMA_LOCK should restore its performance. Furth=
er
> > > > optimizations are possible if this regression proves to be problema=
tic.
> > > >
> > > > Suggested-by: David Hildenbrand <david@redhat.com>
> > > > Reported-by: Jiri Slaby <jirislaby@kernel.org>
> > > > Closes: https://lore.kernel.org/all/dbdef34c-3a07-5951-e1ae-e9c6e3c=
df51b@kernel.org/
> > > > Reported-by: Holger Hoffst=C3=A4tte <holger@applied-asynchrony.com>
> > > > Closes: https://lore.kernel.org/all/b198d649-f4bf-b971-31d0-e8433ec=
2a34c@applied-asynchrony.com/
> > > > Reported-by: Jacob Young <jacobly.alt@gmail.com>
> > > > Closes: https://bugzilla.kernel.org/show_bug.cgi?id=3D217624
> > > > Fixes: 0bff0aaea03e ("x86/mm: try VMA lock-based page fault handlin=
g first")
> > > > Cc: stable@vger.kernel.org
> > > > Signed-off-by: Suren Baghdasaryan <surenb@google.com>
> > > > ---
> > > >   kernel/fork.c | 6 ++++++
> > > >   1 file changed, 6 insertions(+)
> > > >
> > > > diff --git a/kernel/fork.c b/kernel/fork.c
> > > > index b85814e614a5..403bc2b72301 100644
> > > > --- a/kernel/fork.c
> > > > +++ b/kernel/fork.c
> > > > @@ -658,6 +658,12 @@ static __latent_entropy int dup_mmap(struct mm=
_struct *mm,
> > > >               retval =3D -EINTR;
> > > >               goto fail_uprobe_end;
> > > >       }
> > > > +#ifdef CONFIG_PER_VMA_LOCK
> > > > +     /* Disallow any page faults before calling flush_cache_dup_mm=
 */
> > > > +     for_each_vma(old_vmi, mpnt)
> > > > +             vma_start_write(mpnt);
> > > > +     vma_iter_init(&old_vmi, oldmm, 0);
>
> vma_iter_set(&old_vmi, 0) is probably what you want here.

Ok, I send another version with that.

>
> > > > +#endif
> > > >       flush_cache_dup_mm(oldmm);
> > > >       uprobe_dup_mmap(oldmm, mm);
> > > >       /*
> > >
> > > The old version was most probably fine as well, but this certainly lo=
oks
> > > even safer.
> > >
> > > Acked-by: David Hildenbrand <david@redhat.com>
>
> I think this is overkill and believe setting the vma_start_write() will
> synchronize with any readers since it's using the per-vma rw semaphore
> in write mode. Anything faulting will need to finish before the fork
> continues and faults during the fork will fall back to a read lock of
> the mmap_lock.  Is there a possibility of populate happening outside the
> mmap_write lock/vma_lock?

Yes, I think we understand the loss of concurrency in the parent's
ability to fault pages while forking. Is that a real problem though?

>
> Was your benchmarking done with this loop at the start?

No, it was done with the initial version where the lock was inside the
existing loop. I just reran the benchmark and while kernel compilation
times did not change, the stress test shows ~7% regression now,
probably due to that additional tree walk. I'll update that number in
the new patch.
Thanks!

>
> Thanks,
> Liam