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;
Date:   Fri, 27 May 2022 10:55:30 +0800
From:   Muchun Song <songmuchun@bytedance.com>
To:     Waiman Long <longman@redhat.com>
Cc:     Johannes Weiner <hannes@cmpxchg.org>, mhocko@kernel.org,
        roman.gushchin@linux.dev, shakeelb@google.com,
        cgroups@vger.kernel.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, duanxiongchun@bytedance.com
Subject: Re: [PATCH v4 03/11] mm: memcontrol: make lruvec lock safe when LRU
 pages are reparented
Message-ID: <YpA9omBZ8O4+szbY@FVFYT0MHHV2J.usts.net>
References: <20220524060551.80037-1-songmuchun@bytedance.com>
 <20220524060551.80037-4-songmuchun@bytedance.com>
 <Yo0xmKOkBkhRy+bq@cmpxchg.org>
 <Yo38mlkMBFz2h+yP@FVFYT0MHHV2J.googleapis.com>
 <Yo4hVw7B+bUlMzLX@cmpxchg.org>
 <Yo4pPw+IHPBZvZUv@FVFYT0MHHV2J.googleapis.com>
 <Yo5B1tLcYPUoaACS@cmpxchg.org>
 <Yo5NdncOsqL0xP8Q@FVFYT0MHHV2J.googleapis.com>
 <9fe57cf7-9d21-3f91-ef27-e046b426c219@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <9fe57cf7-9d21-3f91-ef27-e046b426c219@redhat.com>
Precedence: bulk

On Thu, May 26, 2022 at 04:17:27PM -0400, Waiman Long wrote:
> On 5/25/22 11:38, Muchun Song wrote:
> > On Wed, May 25, 2022 at 10:48:54AM -0400, Johannes Weiner wrote:
> > > On Wed, May 25, 2022 at 09:03:59PM +0800, Muchun Song wrote:
> > > > On Wed, May 25, 2022 at 08:30:15AM -0400, Johannes Weiner wrote:
> > > > > On Wed, May 25, 2022 at 05:53:30PM +0800, Muchun Song wrote:
> > > > > > On Tue, May 24, 2022 at 03:27:20PM -0400, Johannes Weiner wrote:
> > > > > > > On Tue, May 24, 2022 at 02:05:43PM +0800, Muchun Song wrote:
> > > > > > > > @@ -1230,10 +1213,23 @@ void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio)
> > > > > > > >    */
> > > > > > > >   struct lruvec *folio_lruvec_lock(struct folio *folio)
> > > > > > > >   {
> > > > > > > > -	struct lruvec *lruvec = folio_lruvec(folio);
> > > > > > > > +	struct lruvec *lruvec;
> > > > > > > > +	rcu_read_lock();
> > > > > > > > +retry:
> > > > > > > > +	lruvec = folio_lruvec(folio);
> > > > > > > >   	spin_lock(&lruvec->lru_lock);
> > > > > > > > -	lruvec_memcg_debug(lruvec, folio);
> > > > > > > > +
> > > > > > > > +	if (unlikely(lruvec_memcg(lruvec) != folio_memcg(folio))) {
> > > > > > > > +		spin_unlock(&lruvec->lru_lock);
> > > > > > > > +		goto retry;
> > > > > > > > +	}
> > > > > > > > +
> > > > > > > > +	/*
> > > > > > > > +	 * Preemption is disabled in the internal of spin_lock, which can serve
> > > > > > > > +	 * as RCU read-side critical sections.
> > > > > > > > +	 */
> > > > > > > > +	rcu_read_unlock();
> > > > > > > The code looks right to me, but I don't understand the comment: why do
> > > > > > > we care that the rcu read-side continues? With the lru_lock held,
> > > > > > > reparenting is on hold and the lruvec cannot be rcu-freed anyway, no?
> > > > > > > 
> > > > > > Right. We could hold rcu read lock until end of reparting.  So you mean
> > > > > > we do rcu_read_unlock in folio_lruvec_lock()?
> > > > > The comment seems to suggest that disabling preemption is what keeps
> > > > > the lruvec alive. But it's the lru_lock that keeps it alive. The
> > > > > cgroup destruction path tries to take the lru_lock long before it even
> > > > > gets to synchronize_rcu(). Once you hold the lru_lock, having an
> > > > > implied read-side critical section as well doesn't seem to matter.
> > > > > 
> > > > Well, I thought that spinlocks have implicit read-side critical sections
> > > > because it disables preemption (I learned from the comments above
> > > > synchronize_rcu() that says interrupts, preemption, or softirqs have been
> > > > disabled also serve as RCU read-side critical sections).  So I have a
> > > > question: is it still true in a PREEMPT_RT kernel (I am not familiar with
> > > > this)?
> > > Yes, but you're missing my point.
> > > 
> > > > > Should the comment be deleted?
> > > > I think we could remove the comments. If the above question is false, seems
> > > > like we should continue holding rcu read lock.
> > > It's true.
> > > 
> > Thanks for your answer.
> > 
> > > But assume it's false for a second. Why would you need to continue
> > > holding it? What would it protect? The lruvec would be pinned by the
> > > spinlock even if it DIDN'T imply an RCU lock, right?
> > > 
> > > So I don't understand the point of the comment. If the implied RCU
> > > lock is protecting something not covered by the bare spinlock itself,
> > > it should be added to the comment. Otherwise, the comment should go.
> > > 
> > Got it. Thanks for your nice explanation. I'll remove
> > the comment here.
> 
> Note that there is a similar comment in patch 6 which may have to be removed
> as well.
>

I have noticed that. Thank you for remindering me as well.