[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <26dbecfee2c02456ddfda3647df1bcd56d9cc520.camel@redhat.com>
Date: Tue, 17 Sep 2019 09:36:22 -0500
From: Scott Wood <swood@...hat.com>
To: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
Cc: Joel Fernandes <joel@...lfernandes.org>,
linux-rt-users@...r.kernel.org, linux-kernel@...r.kernel.org,
"Paul E . McKenney" <paulmck@...ux.ibm.com>,
Thomas Gleixner <tglx@...utronix.de>,
Steven Rostedt <rostedt@...dmis.org>,
Peter Zijlstra <peterz@...radead.org>,
Juri Lelli <juri.lelli@...hat.com>,
Clark Williams <williams@...hat.com>
Subject: Re: [PATCH RT v3 5/5] rcutorture: Avoid problematic critical
section nesting on RT
On Tue, 2019-09-17 at 12:07 +0200, Sebastian Andrzej Siewior wrote:
> On 2019-09-16 11:55:57 [-0500], Scott Wood wrote:
> > 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.
>
> We have:
> - local_irq_disable() (+save)
> - spin_lock()
> - local_bh_disable()
> - preempt_disable()
>
> On non-RT you can (but should not) use the counter part of the function
> in random order like:
> local_bh_disable();
> local_irq_disable();
> local_bh_enable();
> local_irq_enable();
Actually even non-RT will assert if you do local_bh_enable() with IRQs
disabled -- but the other combinations do work, and are used some places via
spinlocks. If they are used via direct calls to preempt_disable() or
local_irq_disable() (or via raw spinlocks), then that will not go away on RT
and we'll have a problem.
> The non-RT will survive this. On RT the counterpart functions have to be
> used in reverse order:
> local_bh_disable();
> local_irq_disable();
> local_irq_enable();
> local_bh_enable();
>
> or the kernel will fall apart.
>
> Since you _can_ use it in random order Paul wants to test that the
> random use of those function does not break RCU in any way. Since they
> can not be used on RT in random order it has been agreed that we keep
> the test for !RT but disable it on RT.
For now, yes. Long term it would be good to keep track of when
preemption/irqs would be disabled on RT, even when running a non-RT debug
kernel, and assert when bad things are done with it (assuming an RT-capable
arch). Besides detecting these fairly unusual patterns, it could also
detect earlier the much more common problem of nesting a non-raw spinlock
inside a raw spinlock or other RT-atomic context.
-Scott
Powered by blists - more mailing lists