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;
Sender: Ingo Molnar <mingo.kernel.org@gmail.com>
Date:   Wed, 3 Aug 2022 19:13:01 +0200
From:   Ingo Molnar <mingo@kernel.org>
To:     Steven Rostedt <rostedt@goodmis.org>
Cc:     Liu Song <liusong@linux.alibaba.com>, mingo@redhat.com,
        peterz@infradead.org, juri.lelli@redhat.com,
        vincent.guittot@linaro.org, dietmar.eggemann@arm.com,
        bsegall@google.com, mgorman@suse.de, bristot@redhat.com,
        vschneid@redhat.com, linux-kernel@vger.kernel.org,
        "Paul E. McKenney" <paulmck@kernel.org>
Subject: Re: [PATCH] sched/debug: avoid executing show_state and causing rcu
 stall warning
Message-ID: <YuqsnX6LYxe34YN3@gmail.com>
References: <1659489525-82994-1-git-send-email-liusong@linux.alibaba.com>
 <20220803084235.5d56d1e4@gandalf.local.home>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220803084235.5d56d1e4@gandalf.local.home>
Precedence: bulk


* Steven Rostedt <rostedt@goodmis.org> wrote:

> 
> [ Adding Paul ]
> 
> On Wed,  3 Aug 2022 09:18:45 +0800
> Liu Song <liusong@linux.alibaba.com> wrote:
> 
> > From: Liu Song <liusong@linux.alibaba.com>
> > 
> > If the number of CPUs is large, "sysrq_sched_debug_show" will execute for
> > a long time. Every time I execute "echo t > /proc/sysrq-trigger" on my
> > 128-core machine, the rcu stall warning will be triggered. Moreover,
> > sysrq_sched_debug_show does not need to be protected by rcu_read_lock,
> > and no rcu stall warning will appear after adjustment.
> > 
> > Signed-off-by: Liu Song <liusong@linux.alibaba.com>
> > ---
> >  kernel/sched/core.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 5555e49..82c117e 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -8879,11 +8879,11 @@ void show_state_filter(unsigned int state_filter)
> >  			sched_show_task(p);
> >  	}
> >  
> > +	rcu_read_unlock();
> >  #ifdef CONFIG_SCHED_DEBUG
> >  	if (!state_filter)
> >  		sysrq_sched_debug_show();
> 
> If this is just because sysrq_sched_debug_show() is very slow, does RCU
> have a way to "touch" it? Like the watchdogs have? That is, to tell RCU
> "Yes I know I'm taking a long time, but I'm still making forward progress,
> don't complain about me". Then the sysrq_sched_debug_show() could have:
> 
> 	for_each_online_cpu(cpu) {
> 		/*
> 		 * Need to reset softlockup watchdogs on all CPUs, because
> 		 * another CPU might be blocked waiting for us to process
> 		 * an IPI or stop_machine.
> 		 */
> 		touch_nmi_watchdog();
> 		touch_all_softlockup_watchdogs();
> +		touch_rcu();
> 		print_cpu(NULL, cpu);
> 	}

I'd much rather we use the specific exclusion primitive suitable for that 
sequence - in that case it should be cpus_read_lock()/unlock() I suspect.

But the entire code sequence should be reviewed - do we anywhere walk task 
lists that need RCU protection?

My main complaint was that we cannot just randomly drop the RCU lock with 
no inspection of the underlying code.

	Ingo