Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Tue, 26 Jun 2018 11:03:03 -0700
From:   "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     linux-kernel@vger.kernel.org, mingo@kernel.org,
        jiangshanlai@gmail.com, dipankar@in.ibm.com,
        akpm@linux-foundation.org, mathieu.desnoyers@efficios.com,
        josh@joshtriplett.org, tglx@linutronix.de, rostedt@goodmis.org,
        dhowells@redhat.com, edumazet@google.com, fweisbec@gmail.com,
        oleg@redhat.com, joel@joelfernandes.org
Subject: Re: [PATCH tip/core/rcu 06/27] rcu: Mark task as .need_qs less
 aggressively
Reply-To: paulmck@linux.vnet.ibm.com
References: <20180626003448.GA26209@linux.vnet.ibm.com>
 <20180626003513.27812-6-paulmck@linux.vnet.ibm.com>
 <20180626170812.GH2494@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180626170812.GH2494@hirez.programming.kicks-ass.net>
User-Agent: Mutt/1.5.21 (2010-09-15)
Message-Id: <20180626180303.GD3593@linux.vnet.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Jun 26, 2018 at 07:08:12PM +0200, Peter Zijlstra wrote:
> On Mon, Jun 25, 2018 at 05:34:52PM -0700, Paul E. McKenney wrote:
> > If any scheduling-clock interrupt interrupts an RCU-preempt read-side
> > critical section, the interrupted task's ->rcu_read_unlock_special.b.need_qs
> > field is set.  This causes the outermost rcu_read_unlock() to incur the
> > extra overhead of calling into rcu_read_unlock_special().  This commit
> > reduces that overhead by setting ->rcu_read_unlock_special.b.need_qs only
> > if the grace period has been in effect for more than one second.
> 
> Even less agressive is never setting it at all.

True, but if the CPU has been in an RCU read-side critical section for
a full second (which is the case with high probability when .b.need_qs
is set after this change), we might want to respond to the end of that
critical section sooner rather than later.

> Changelog fails to explain why not setting it every tick is correct, nor
> why 1s is a 'safe' value to use.

The RCU CPU stall warning cannot be set to less than 3s, so 1s is
reasonable.  It is a tradeoff -- setting it lower causes a greater
fraction of RCU read-side critical sections to incur extra overhead at
rcu_read_unlock() time, while setting it higher keeps a lazy approach
to reporting the quiescent state to core RCU for longer critical sections.

The upcoming RCU-bh/RCU-preempt/RCU-sched consolidation will raise
contention and overhead, so this is one of several things done to
lower overhead and contention to compensate for that.

							Thanx, Paul