Received-SPF: pass (google.com: domain of linux-kernel+bounces-105329-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Date: Sat, 16 Mar 2024 21:55:16 -0700
From: "Paul E. McKenney" <paulmck@kernel.org>
To: Zqiang <qiang.zhang1211@gmail.com>
Cc: frederic@kernel.org, joel@joelfernandes.org, neeraj.upadhyay@kernel.org,
	rcu@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] rcu-tasks: Avoid rtp_irq_work triggering when the
 rcu-tasks GP is ongoing
Message-ID: <49c73303-fdda-4912-b26a-75f2235b3a62@paulmck-laptop>
Reply-To: paulmck@kernel.org
References: <20240311035502.7112-1-qiang.zhang1211@gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240311035502.7112-1-qiang.zhang1211@gmail.com>

On Mon, Mar 11, 2024 at 11:55:02AM +0800, Zqiang wrote:
> This commit generate rcu_task_gp_in_progress() to check whether
> the rcu-tasks GP is ongoing, if is ongoing, avoid trigger
> rtp_irq_work to wakeup rcu tasks kthreads in call_rcu_tasks_generic().
> 
> The test results are as follows:
> 
> echo call_rcu_tasks_iw_wakeup > /sys/kernel/debug/tracing/set_ftrace_filter
> echo 1 > /sys/kernel/debug/tracing/function_profile_enabled
> insmod rcutorture.ko torture_type=tasks-tracing fwd_progress=4
> sleep 600
> rmmod rcutorture.ko
> echo 0 > /sys/kernel/debug/tracing/function_profile_enabled
> echo > /sys/kernel/debug/tracing/set_ftrace_filter
> 
> head /sys/kernel/debug/tracing/trace_stat/function*
> 
> original: 56376  apply patch: 33521
> 
> Signed-off-by: Zqiang <qiang.zhang1211@gmail.com>

Note that rcu_seq_current() does not provide ordering.  So are you
sure that this change is safe on weakly ordered systems such as ARM?

For example, consider the following sequence of events:

o	The call_rcu_tasks_generic() function picks up the grace-period
	sequence number, which shows that there is a grace period in
	progress.

o	The grace period ends, and sees no reason to start a new grace
	period.

o	The call_rcu_tasks_generic() function sees no reason to wake
	up the grace-period kthread.  There are no more calls to
	call_rcu_tasks*(), so the callback is never invoked.

Or is there something that prevents this sequence of events from
ever happening on weakly ordered systems?

							Thanx, Paul

> ---
>  
> original:
> ==> /sys/kernel/debug/tracing/trace_stat/function0 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup             13217    19292.52 us     1.459 us        8.834 us    
> 
> ==> /sys/kernel/debug/tracing/trace_stat/function1 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup             15146    22377.01 us     1.477 us        22.873 us   
> 
> ==> /sys/kernel/debug/tracing/trace_stat/function2 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup             12561    18125.76 us     1.443 us        6.372 us    
> 
> ==> /sys/kernel/debug/tracing/trace_stat/function3 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup             15452    21770.57 us     1.408 us        6.710 us
> 
> apply patch:
> ==> /sys/kernel/debug/tracing/trace_stat/function0 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup              8334    15121.13 us     1.814 us        4.457 us    
> 
> ==> /sys/kernel/debug/tracing/trace_stat/function1 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup              8355    15760.51 us     1.886 us        14.775 us   
> 
> ==> /sys/kernel/debug/tracing/trace_stat/function2 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup              7219    14194.27 us     1.966 us        42.440 us   
> 
> ==> /sys/kernel/debug/tracing/trace_stat/function3 <==
>   Function                               Hit    Time            Avg             s^2
>   --------                               ---    ----            ---             ---
>   call_rcu_tasks_iw_wakeup              9613    19850.04 us     2.064 us        91.023 us  
> 
>  kernel/rcu/tasks.h | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/rcu/tasks.h b/kernel/rcu/tasks.h
> index 147b5945d67a..36c7e1d441d0 100644
> --- a/kernel/rcu/tasks.h
> +++ b/kernel/rcu/tasks.h
> @@ -317,6 +317,11 @@ static void call_rcu_tasks_iw_wakeup(struct irq_work *iwp)
>  	rcuwait_wake_up(&rtp->cbs_wait);
>  }
>  
> +static int rcu_task_gp_in_progress(struct rcu_tasks *rtp)
> +{
> +	return rcu_seq_state(rcu_seq_current(&rtp->tasks_gp_seq));
> +}
> +
>  // Enqueue a callback for the specified flavor of Tasks RCU.
>  static void call_rcu_tasks_generic(struct rcu_head *rhp, rcu_callback_t func,
>  				   struct rcu_tasks *rtp)
> @@ -375,7 +380,8 @@ static void call_rcu_tasks_generic(struct rcu_head *rhp, rcu_callback_t func,
>  	}
>  	rcu_read_unlock();
>  	/* We can't create the thread unless interrupts are enabled. */
> -	if (needwake && READ_ONCE(rtp->kthread_ptr))
> +	if (needwake && READ_ONCE(rtp->kthread_ptr) &&
> +			!rcu_task_gp_in_progress(rtp))
>  		irq_work_queue(&rtpcp->rtp_irq_work);
>  }
>  
> -- 
> 2.17.1
>