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:   Wed, 23 May 2018 10:03:03 -0700
From:   "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
To:     Steven Rostedt <rostedt@goodmis.org>
Cc:     Joel Fernandes <joelaf@google.com>, linux-kernel@vger.kernel.org,
        "Joel Fernandes (Google)" <joel@joelfernandes.org>,
        Peter Zilstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Boqun Feng <boqun.feng@gmail.com>, byungchul.park@lge.com,
        kernel-team@android.com, Josh Triplett <josh@joshtriplett.org>,
        Lai Jiangshan <jiangshanlai@gmail.com>,
        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Subject: Re: [PATCH 1/4] rcu: Speed up calling of RCU tasks callbacks
Reply-To: paulmck@linux.vnet.ibm.com
References: <20180523063815.198302-1-joel@joelfernandes.org>
 <20180523063815.198302-2-joel@joelfernandes.org>
 <20180523155734.GK3803@linux.vnet.ibm.com>
 <20180523124531.7b0e972a@gandalf.local.home>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180523124531.7b0e972a@gandalf.local.home>
User-Agent: Mutt/1.5.21 (2010-09-15)
Message-Id: <20180523170303.GR3803@linux.vnet.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, May 23, 2018 at 12:45:31PM -0400, Steven Rostedt wrote:
> On Wed, 23 May 2018 08:57:34 -0700
> "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> wrote:
> 
> > On Tue, May 22, 2018 at 11:38:12PM -0700, Joel Fernandes wrote:
> > > From: "Joel Fernandes (Google)" <joel@joelfernandes.org>
> > > 
> > > RCU tasks callbacks can take at least 1 second before the callbacks are
> > > executed. This happens even if the hold-out tasks enter their quiescent states
> > > quickly. I noticed this when I was testing trampoline callback execution.
> > > 
> > > To test the trampoline freeing, I wrote a simple script:
> > > cd /sys/kernel/debug/tracing/
> > > echo '__schedule_bug:traceon' > set_ftrace_filter;
> > > echo '!__schedule_bug:traceon' > set_ftrace_filter;
> > > 
> > > In the background I had simple bash while loop:
> > > while [ 1 ]; do x=1; done &
> > > 
> > > Total time of completion of above commands in seconds:
> > > 
> > > With this patch:
> > > real    0m0.179s
> > > user    0m0.000s
> > > sys     0m0.054s
> > > 
> > > Without this patch:
> > > real    0m1.098s
> > > user    0m0.000s
> > > sys     0m0.053s
> > > 
> > > That's a greater than 6X speed up in performance. In order to accomplish
> > > this, I am waiting for HZ/10 time before entering the hold-out checking
> > > loop. The loop still preserves its checking of held tasks every 1 second
> > > as before, in case this first test doesn't succeed.
> > > 
> > > Cc: Steven Rostedt <rostedt@goodmis.org>  
> > 
> > Given an ack from Steven, I would be happy to take this, give or take
> > some nits below.
> 
> I'm currently testing it, and trying to understand it better.

Very good, thank you!

> > > Cc: Peter Zilstra <peterz@infradead.org>
> > > Cc: Ingo Molnar <mingo@redhat.com>
> > > Cc: Boqun Feng <boqun.feng@gmail.com>
> > > Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
> > > Cc: byungchul.park@lge.com
> > > Cc: kernel-team@android.com
> > > Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
> > > ---
> > >  kernel/rcu/update.c | 12 +++++++++++-
> > >  1 file changed, 11 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/kernel/rcu/update.c b/kernel/rcu/update.c
> > > index 5783bdf86e5a..a28698e44b08 100644
> > > --- a/kernel/rcu/update.c
> > > +++ b/kernel/rcu/update.c
> > > @@ -743,6 +743,12 @@ static int __noreturn rcu_tasks_kthread(void *arg)
> > >  		 */
> > >  		synchronize_srcu(&tasks_rcu_exit_srcu);
> > > 
> > > +		/*
> > > +		 * Wait a little bit incase held tasks are released  
> > 
> > 				in case
> > 
> > > +		 * during their next timer ticks.
> > > +		 */
> > > +		schedule_timeout_interruptible(HZ/10);
> > > +
> > >  		/*
> > >  		 * Each pass through the following loop scans the list
> > >  		 * of holdout tasks, removing any that are no longer
> > > @@ -755,7 +761,6 @@ static int __noreturn rcu_tasks_kthread(void *arg)
> > >  			int rtst;
> > >  			struct task_struct *t1;
> > > 
> > > -			schedule_timeout_interruptible(HZ);
> > >  			rtst = READ_ONCE(rcu_task_stall_timeout);
> > >  			needreport = rtst > 0 &&
> > >  				     time_after(jiffies, lastreport + rtst);
> > > @@ -768,6 +773,11 @@ static int __noreturn rcu_tasks_kthread(void *arg)
> > >  				check_holdout_task(t, needreport, &firstreport);
> > >  				cond_resched();
> > >  			}
> > > +
> > > +			if (list_empty(&rcu_tasks_holdouts))
> > > +				break;
> > > +
> > > +			schedule_timeout_interruptible(HZ);  
> 
> Why is this a full second wait and not the HZ/10 like the others?

The idea is to respond quickly on small idle systems and to reduce the
number of possibly quite lengthy traversals of the task list otherwise.
I actually considered exponential backoff, but decided to keep it simple,
at least to start with.

							Thanx, Paul

> -- Steve
> 
> > 
> > Is there a better way to do this?  Can this be converted into a for-loop?
> > Alternatively, would it make sense to have a firsttime local variable
> > initialized to true, to keep the schedule_timeout_interruptible() at
> > the beginning of the loop, but skip it on the first pass through the loop?
> > 
> > Don't get me wrong, what you have looks functionally correct, but
> > duplicating the condition might cause problems later on, for example,
> > should a bug fix be needed in the condition.
> > 
> > >  		}
> > > 
> > >  		/*
> > > -- 
> > > 2.17.0.441.gb46fe60e1d-goog
> > >   
>