Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:4 as permitted sender) client-ip=2620:137:e000::3:4;
MIME-Version: 1.0
References: <20231119194740.94101-1-ryncsn@gmail.com> <20231119194740.94101-5-ryncsn@gmail.com>
 <CAF8kJuNbcvwmojMB2Vp2LjTBxZtKFO_Zng4TODqADxmNPn2kCg@mail.gmail.com>
 <CAMgjq7D_OJiaNPWquFqxk5DKZXqoUMrD7Y_CRM7D0RnRnHmW7Q@mail.gmail.com>
 <CAF8kJuOzXAy8Er_knD0hi60Tb+XDMrijg-W83KD=Zdjqbu1Dmw@mail.gmail.com>
 <CAMgjq7DNVkyhwyXz9PRcp3XTBNvznZ97dOFFA_DGsJks12_pQw@mail.gmail.com> <CAF8kJuM05apUTd9Mpv6s4sZ6SfSAWD+hM_KLiDfatoXu5QwFbg@mail.gmail.com>
In-Reply-To: <CAF8kJuM05apUTd9Mpv6s4sZ6SfSAWD+hM_KLiDfatoXu5QwFbg@mail.gmail.com>
From:   Kairui Song <ryncsn@gmail.com>
Date:   Fri, 24 Nov 2023 16:14:41 +0800
Message-ID: <CAMgjq7Bax-EOiXXDRxpHFoyrQWRQdSWbZ0KN7-W8YsocV5C97w@mail.gmail.com>
Subject: Re: [PATCH 04/24] mm/swap: avoid setting page lock bit and doing
 extra unlock check
To:     Chris Li <chrisl@kernel.org>
Cc:     linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
        "Huang, Ying" <ying.huang@intel.com>,
        David Hildenbrand <david@redhat.com>,
        Hugh Dickins <hughd@google.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Matthew Wilcox <willy@infradead.org>,
        Michal Hocko <mhocko@suse.com>, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Chris Li <chrisl@kernel.org> =E4=BA=8E2023=E5=B9=B411=E6=9C=8823=E6=97=A5=
=E5=91=A8=E5=9B=9B 04:57=E5=86=99=E9=81=93=EF=BC=9A
>
> Hi Kairui,
>
> On Wed, Nov 22, 2023 at 9:33=E2=80=AFAM Kairui Song <ryncsn@gmail.com> wr=
ote:
>
> > > There are two different charges. Memcg charging and memcg swapin char=
ging.
> > > The folio_put will do the memcg discharge, the corresponding memcg
> > > charge is in follio allocation.
> >
> > Hi Chris,
> >
> > I didn't get your idea here... By "memcg swapin charge", do you mean
> > "memory.swap.*"? And "memcg charging" means "memory.*"?. There is no
>
> Sorry I should have used the function name then there is no ambiguity.
> "memcg swapin charge" I mean function mem_cgroup_swapin_charge_folio().
> This function will look up the swap entry and find the memcg by swap entr=
y then
> charge to that memcg.
>
> > memcg charge related code in folio allocation (alloc_pages_mpol),
> > actually the mem_cgroup_swapin_charge_folio here is doing memcg charge
> > not memcg swapin charge. Swapin path actually need to uncharge
> > "memory.swap" by mem_cgroup_swapin_uncharge_swap in later part of this
> > function.
>
> I still think you have a bug there.
>
> Take this make up example:
> Let say the for loop runs 3 times and the 3rd time breaks out the for loo=
p.
> The original code will call:
> filemap_get_folio() 3 times
> folio_put() 2 times
> mem_cgroup_swapin_charge_folio() 1 time.
>
> With your patch, it will call:
> filemap_get_folio() 3 times
> folio_put() 2 times
> mem_cgroup_swapin_charge_folio() 3 times.
>
> Do you see the behavior difference there?

Hi Chris.

folio_put will discharge a page if it's charged, in original code the
2 folio_put call simply free the page since it's not charged. But in
this patch, folio_put will cancel previous
mem_cgroup_swapin_charge_folio call, so actually the 3
mem_cgroup_swapin_charge_folio calls will only charge once. (2 calls
was cancelled by folio_put).

I think this is making it confusing indeed and causing more trouble in
error path (the uncharge could be more expensive than unlock check),
will rework this part.