Re: [PATCH RT v3 5/5] rcutorture: Avoid problematic critical section nesting on RT

From: Scott Wood <swood@redhat.com>
To: Joel Fernandes <joel@joelfernandes.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
	linux-rt-users@vger.kernel.org, linux-kernel@vger.kernel.org,
	"Paul E . McKenney" <paulmck@linux.ibm.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Steven Rostedt <rostedt@goodmis.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Juri Lelli <juri.lelli@redhat.com>,
	Clark Williams <williams@redhat.com>
Subject: Re: [PATCH RT v3 5/5] rcutorture: Avoid problematic critical section nesting on RT
Date: Mon, 16 Sep 2019 11:55:57 -0500	[thread overview]
Message-ID: <500cabaa80f250b974409ee4a4fca59bf2e24564.camel@redhat.com> (raw)
In-Reply-To: <20190912221706.GC150506@google.com>

On Thu, 2019-09-12 at 18:17 -0400, Joel Fernandes wrote:
> On Wed, Sep 11, 2019 at 05:57:29PM +0100, Scott Wood wrote:
> > rcutorture was generating some nesting scenarios that are not
> > reasonable.  Constrain the state selection to avoid them.
> > 
> > Example #1:
> > 
> > 1. preempt_disable()
> > 2. local_bh_disable()
> > 3. preempt_enable()
> > 4. local_bh_enable()
> > 
> > On PREEMPT_RT, BH disabling takes a local lock only when called in
> > non-atomic context.  Thus, atomic context must be retained until after
> > BH
> > is re-enabled.  Likewise, if BH is initially disabled in non-atomic
> > context, it cannot be re-enabled in atomic context.
> > 
> > Example #2:
> > 
> > 1. rcu_read_lock()
> > 2. local_irq_disable()
> > 3. rcu_read_unlock()
> > 4. local_irq_enable()
> 
> If I understand correctly, these examples are not unrealistic in the real
> world unless RCU is used in the scheduler.

I hope you mean "not realistic", at least when it comes to explicit
preempt/irq disabling rather than spinlock variants that don't disable
preempt/irqs on PREEMPT_RT.

> > If the thread is preempted between steps 1 and 2,
> > rcu_read_unlock_special.b.blocked will be set, but it won't be
> > acted on in step 3 because IRQs are disabled.  Thus, reporting of the
> > quiescent state will be delayed beyond the local_irq_enable().
> 
> Yes, with consolidated RCU this can happen but AFAIK it has not seen to be
> a
> problem since deferred QS reporting will happen take care of it, which can
> also happen from subsequent rcu_read_unlock_special().

The defer_qs_iw_pending stuff isn't in 5.2-rt.  Still, given patch 4/5 (and
special.b.deferred_qs on mainline) this shouldn't present a deadlock concern
(letting the test run a bit now to double check) so this patch could
probably be limited to the "example #1" sequence.

> > For now, these scenarios will continue to be tested on non-PREEMPT_RT
> > kernels, until debug checks are added to ensure that they are not
> > happening elsewhere.
> 
> Are you seeing real issues that need this patch? It would be good to not
> complicate rcutorture if not needed.

rcutorture crashes on RT without this patch (in particular due to the
local_bh_disable misordering).

-Scott