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:   Wed, 12 Jul 2023 21:27:25 -0700
From:   "Paul E. McKenney" <paulmck@kernel.org>
To:     Gao Xiang <hsiangkao@linux.alibaba.com>
Cc:     Joel Fernandes <joel@joelfernandes.org>,
        Sandeep Dhavale <dhavale@google.com>,
        Frederic Weisbecker <frederic@kernel.org>,
        Neeraj Upadhyay <quic_neeraju@quicinc.com>,
        Josh Triplett <josh@joshtriplett.org>,
        Boqun Feng <boqun.feng@gmail.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
        Lai Jiangshan <jiangshanlai@gmail.com>,
        Zqiang <qiang.zhang1211@gmail.com>,
        Matthias Brugger <matthias.bgg@gmail.com>,
        AngeloGioacchino Del Regno 
        <angelogioacchino.delregno@collabora.com>,
        linux-erofs@lists.ozlabs.org, xiang@kernel.org,
        Will Shiu <Will.Shiu@mediatek.com>, kernel-team@android.com,
        rcu@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-arm-kernel@lists.infradead.org,
        linux-mediatek@lists.infradead.org
Subject: Re: [PATCH v1] rcu: Fix and improve RCU read lock checks when
 !CONFIG_DEBUG_LOCK_ALLOC
Message-ID: <0d9e7b4d-6477-47a6-b3d2-2c9d9b64903d@paulmck-laptop>
Reply-To: paulmck@kernel.org
References: <20230711233816.2187577-1-dhavale@google.com>
 <CAEXW_YQvpiFEaaNoS=Msgi17mU3kZD+q8bNBaHYasMArG9aPig@mail.gmail.com>
 <CAB=BE-Rm0ycTZXj=wHW_FBCCKbswG+dh3L+o1+CUW=Pg_oWnyw@mail.gmail.com>
 <20230713003201.GA469376@google.com>
 <161f1615-3d85-cf47-d2d5-695adf1ca7d4@linux.alibaba.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <161f1615-3d85-cf47-d2d5-695adf1ca7d4@linux.alibaba.com>
Precedence: bulk

On Thu, Jul 13, 2023 at 10:02:17AM +0800, Gao Xiang wrote:
> 
> 
> On 2023/7/13 08:32, Joel Fernandes wrote:
> > On Wed, Jul 12, 2023 at 02:20:56PM -0700, Sandeep Dhavale wrote:
> > [..]
> > > > As such this patch looks correct to me, one thing I noticed is that
> > > > you can check rcu_is_watching() like the lockdep-enabled code does.
> > > > That will tell you also if a reader-section is possible because in
> > > > extended-quiescent-states, RCU readers should be non-existent or
> > > > that's a bug.
> > > > 
> > > Please correct me if I am wrong, reading from the comment in
> > > kernel/rcu/update.c rcu_read_lock_held_common()
> > > ..
> > >    * The reason for this is that RCU ignores CPUs that are
> > >   * in such a section, considering these as in extended quiescent state,
> > >   * so such a CPU is effectively never in an RCU read-side critical section
> > >   * regardless of what RCU primitives it invokes.
> > > 
> > > It seems rcu will treat this as lock not held rather than a fact that
> > > lock is not held. Is my understanding correct?
> > 
> > If RCU treats it as a lock not held, that is a fact for RCU ;-). Maybe you
> > mean it is not a fact for erofs?
> 
> I'm not sure if I get what you mean, EROFS doesn't take any RCU read lock
> here:

The key point is that we need lockdep to report errors when
rcu_read_lock(), rcu_dereference(), and friends are used when RCU is
not watching.  We also need lockdep to report an error when someone
uses rcu_dereference() when RCU is not watching, but also forgets the
rcu_read_lock().

And this is the job of rcu_read_lock_held(), which is one reason why
that rcu_is_watching() is needed.

> z_erofs_decompressqueue_endio() is actually a "bio->bi_end_io", previously
> which can be called under two scenarios:
> 
>  1) under softirq context, which is actually part of device I/O compleltion;
> 
>  2) under threaded context, like what dm-verity or likewise calls.
> 
> But EROFS needs to decompress in a threaded context anyway, so we trigger
> a workqueue to resolve the case 1).
> 
> Recently, someone reported there could be some case 3) [I think it was
> introduced recently but I have no time to dig into it]:
> 
>  case 3: under RCU read lock context, which is shown by this:
> https://lore.kernel.org/r/4a8254eb-ac39-1e19-3d82-417d3a7b9f94@linux.alibaba.com/T/#u
> 
>  and such RCU read lock is taken in __blk_mq_run_dispatch_ops().
> 
> But as the commit shown, we only need to trigger a workqueue for case 1)
> and 3) due to performance reasons.

Just out of curiosity, exactly how much is it costing to trigger the
workqueue?

> Hopefully I show it more clear.

One additional question...  What is your plan for kernels built with
CONFIG_PREEMPT_COUNT=n?  After all, in such kernels, there is no way
that I know of for code to determine whether it is in an RCU read-side
critical section, holding a spinlock, or running with preemption disabled.

						Thanx, Paul