From: "Paul E. McKenney" <paulmck@kernel.org>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>, Qian Cai <cai@redhat.com>,
Steven Rostedt <rostedt@goodmis.org>,
Ingo Molnar <mingo@kernel.org>, x86 <x86@kernel.org>,
linux-kernel@vger.kernel.org, linux-tip-commits@vger.kernel.org,
Linux Next Mailing List <linux-next@vger.kernel.org>,
Stephen Rothwell <sfr@canb.auug.org.au>
Subject: Re: [tip: locking/core] lockdep: Fix lockdep recursion
Date: Tue, 13 Oct 2020 09:26:50 -0700 [thread overview]
Message-ID: <20201013162650.GN3249@paulmck-ThinkPad-P72> (raw)
In-Reply-To: <20201013112544.GZ2628@hirez.programming.kicks-ass.net>
On Tue, Oct 13, 2020 at 01:25:44PM +0200, Peter Zijlstra wrote:
> On Tue, Oct 13, 2020 at 12:44:50PM +0200, Peter Zijlstra wrote:
> > On Tue, Oct 13, 2020 at 12:34:06PM +0200, Peter Zijlstra wrote:
> > > On Mon, Oct 12, 2020 at 02:28:12PM -0700, Paul E. McKenney wrote:
> > > > It is certainly an accident waiting to happen. Would something like
> > > > the following make sense?
> > >
> > > Sadly no.
> > >
> > > > ------------------------------------------------------------------------
> > > >
> > > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > > > index bfd38f2..52a63bc 100644
> > > > --- a/kernel/rcu/tree.c
> > > > +++ b/kernel/rcu/tree.c
> > > > @@ -4067,6 +4067,7 @@ void rcu_cpu_starting(unsigned int cpu)
> > > >
> > > > rnp = rdp->mynode;
> > > > mask = rdp->grpmask;
> > > > + lockdep_off();
> > > > raw_spin_lock_irqsave_rcu_node(rnp, flags);
> > > > WRITE_ONCE(rnp->qsmaskinitnext, rnp->qsmaskinitnext | mask);
> > > > newcpu = !(rnp->expmaskinitnext & mask);
> > > > @@ -4086,6 +4087,7 @@ void rcu_cpu_starting(unsigned int cpu)
> > > > } else {
> > > > raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
> > > > }
> > > > + lockdep_on();
> > > > smp_mb(); /* Ensure RCU read-side usage follows above initialization. */
> > > > }
> > >
> > > This will just shut it up, but will not fix the actual problem of that
> > > spin-lock ending up in trace_lock_acquire() which relies on RCU which
> > > isn't looking.
> > >
> > > What we need here is to supress tracing not lockdep. Let me consider.
> >
> > We appear to have a similar problem with rcu_report_dead(), it's
> > raw_spin_unlock()s can end up in trace_lock_release() while we just
> > killed RCU.
>
> So we can deal with the explicit trace_*() calls like the below, but I
> really don't like it much. It also doesn't help with function tracing.
> This is really early/late in the hotplug cycle and should be considered
> entry, we shouldn't be tracing anything here.
>
> Paul, would it be possible to use a scheme similar to IRQ/NMI for
> hotplug? That seems to mostly rely on atomic ops, not locks.
The rest of the rcu_node tree and the various grace-period/hotplug races
makes that question non-trivial. I will look into it, but I have no
reason for optimism.
But there is only one way to find out... ;-)
Thanx, Paul
> ---
> diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h
> index d05db575f60f..22e3a3523ad3 100644
> --- a/include/linux/lockdep.h
> +++ b/include/linux/lockdep.h
> @@ -159,7 +159,7 @@ extern void lockdep_init_task(struct task_struct *task);
> */
> #define LOCKDEP_RECURSION_BITS 16
> #define LOCKDEP_OFF (1U << LOCKDEP_RECURSION_BITS)
> -#define LOCKDEP_RECURSION_MASK (LOCKDEP_OFF - 1)
> +#define LOCKDEP_TRACE_MASK (LOCKDEP_OFF - 1)
>
> /*
> * lockdep_{off,on}() are macros to avoid tracing and kprobes; not inlines due
> @@ -176,6 +176,16 @@ do { \
> current->lockdep_recursion -= LOCKDEP_OFF; \
> } while (0)
>
> +#define lockdep_trace_off() \
> +do { \
> + current->lockdep_recursion++; \
> +} while (0)
> +
> +#define lockdep_trace_on() \
> +do { \
> + current->lockdep_recursion-- \
> +} while (0)
> +
> extern void lockdep_register_key(struct lock_class_key *key);
> extern void lockdep_unregister_key(struct lock_class_key *key);
>
> diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
> index 3e99dfef8408..2df98abee82e 100644
> --- a/kernel/locking/lockdep.c
> +++ b/kernel/locking/lockdep.c
> @@ -87,7 +87,7 @@ static inline bool lockdep_enabled(void)
> if (raw_cpu_read(lockdep_recursion))
> return false;
>
> - if (current->lockdep_recursion)
> + if (current->lockdep_recursion >> LOCKDEP_RECURSION_BITS)
> return false;
>
> return true;
> @@ -5410,7 +5410,8 @@ void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
> {
> unsigned long flags;
>
> - trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
> + if (!(current->lockdep_recursion & LOCKDEP_TRACE_MASK))
> + trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
>
> if (!debug_locks)
> return;
> @@ -5450,7 +5451,8 @@ void lock_release(struct lockdep_map *lock, unsigned long ip)
> {
> unsigned long flags;
>
> - trace_lock_release(lock, ip);
> + if (!(current->lockdep_recursion & LOCKDEP_TRACE_MASK))
> + trace_lock_release(lock, ip);
>
> if (unlikely(!lockdep_enabled()))
> return;
> @@ -5662,7 +5664,8 @@ void lock_contended(struct lockdep_map *lock, unsigned long ip)
> {
> unsigned long flags;
>
> - trace_lock_acquired(lock, ip);
> + if (!(current->lockdep_recursion & LOCKDEP_TRACE_MASK))
> + trace_lock_acquired(lock, ip);
>
> if (unlikely(!lock_stat || !lockdep_enabled()))
> return;
> @@ -5680,7 +5683,8 @@ void lock_acquired(struct lockdep_map *lock, unsigned long ip)
> {
> unsigned long flags;
>
> - trace_lock_contended(lock, ip);
> + if (!(current->lockdep_recursion & LOCKDEP_TRACE_MASK))
> + trace_lock_contended(lock, ip);
>
> if (unlikely(!lock_stat || !lockdep_enabled()))
> return;
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index edeabc232c21..dbd56603fc0a 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -4047,6 +4047,11 @@ void rcu_cpu_starting(unsigned int cpu)
>
> rnp = rdp->mynode;
> mask = rdp->grpmask;
> +
> + /*
> + * Lockdep will call tracing, which requires RCU, but RCU isn't on yet.
> + */
> + lockdep_trace_off();
> raw_spin_lock_irqsave_rcu_node(rnp, flags);
> WRITE_ONCE(rnp->qsmaskinitnext, rnp->qsmaskinitnext | mask);
> newcpu = !(rnp->expmaskinitnext & mask);
> @@ -4064,6 +4069,7 @@ void rcu_cpu_starting(unsigned int cpu)
> } else {
> raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
> }
> + lockdep_trace_on();
> smp_mb(); /* Ensure RCU read-side usage follows above initialization. */
> }
>
> @@ -4091,6 +4097,11 @@ void rcu_report_dead(unsigned int cpu)
>
> /* Remove outgoing CPU from mask in the leaf rcu_node structure. */
> mask = rdp->grpmask;
> +
> + /*
> + * Lockdep will call tracing, which requires RCU, but we're switching RCU off.
> + */
> + lockdep_trace_off();
> raw_spin_lock(&rcu_state.ofl_lock);
> raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */
> rdp->rcu_ofl_gp_seq = READ_ONCE(rcu_state.gp_seq);
> @@ -4101,8 +4112,10 @@ void rcu_report_dead(unsigned int cpu)
> raw_spin_lock_irqsave_rcu_node(rnp, flags);
> }
> WRITE_ONCE(rnp->qsmaskinitnext, rnp->qsmaskinitnext & ~mask);
> + /* RCU is off, locks must not call into tracing */
> raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
> raw_spin_unlock(&rcu_state.ofl_lock);
> + lockdep_trace_on();
>
> rdp->cpu_started = false;
> }
> diff --git a/kernel/rcu/update.c b/kernel/rcu/update.c
> index 39334d2d2b37..403b138f7cd4 100644
> --- a/kernel/rcu/update.c
> +++ b/kernel/rcu/update.c
> @@ -275,8 +275,8 @@ EXPORT_SYMBOL_GPL(rcu_callback_map);
>
> noinstr int notrace debug_lockdep_rcu_enabled(void)
> {
> - return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && debug_locks &&
> - current->lockdep_recursion == 0;
> + return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && __lockdep_enabled;
> +
> }
> EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
>
next prev parent reply other threads:[~2020-10-13 16:26 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <160223032121.7002.1269740091547117869.tip-bot2@tip-bot2>
2020-10-09 13:41 ` [tip: locking/core] lockdep: Fix lockdep recursion Qian Cai
2020-10-09 13:58 ` Paul E. McKenney
2020-10-09 15:30 ` Qian Cai
2020-10-09 16:11 ` Paul E. McKenney
2020-10-09 16:23 ` Peter Zijlstra
2020-10-09 16:37 ` Paul E. McKenney
2020-10-09 17:36 ` Qian Cai
2020-10-09 17:50 ` Paul E. McKenney
2020-10-09 17:54 ` Qian Cai
2020-10-09 18:21 ` Paul E. McKenney
2020-10-12 3:11 ` Boqun Feng
2020-10-12 14:14 ` Qian Cai
2020-10-12 21:28 ` Paul E. McKenney
2020-10-13 10:34 ` Peter Zijlstra
2020-10-13 10:44 ` Peter Zijlstra
2020-10-13 11:25 ` Peter Zijlstra
2020-10-13 16:26 ` Paul E. McKenney [this message]
2020-10-13 19:30 ` Paul E. McKenney
2020-10-14 18:34 ` Paul E. McKenney
2020-10-14 21:53 ` Peter Zijlstra
2020-10-14 22:11 ` Paul E. McKenney
2020-10-14 22:39 ` Peter Zijlstra
2020-10-14 23:55 ` Paul E. McKenney
2020-10-15 3:41 ` Paul E. McKenney
2020-10-15 9:49 ` Peter Zijlstra
2020-10-15 9:50 ` Peter Zijlstra
2020-10-15 16:15 ` Paul E. McKenney
2020-10-15 9:52 ` Peter Zijlstra
2020-10-15 16:20 ` Paul E. McKenney
2020-10-15 16:15 ` Paul E. McKenney
2020-10-15 17:23 ` Paul E. McKenney
2020-10-13 16:15 ` Paul E. McKenney
2020-10-13 10:27 ` Peter Zijlstra
2020-10-13 16:24 ` Boqun Feng
2020-10-27 19:31 ` Qian Cai
2020-10-28 3:01 ` Paul E. McKenney
2020-10-28 14:39 ` Qian Cai
2020-10-28 15:53 ` Paul E. McKenney
2020-10-28 20:08 ` Qian Cai
2020-10-28 21:02 ` Paul E. McKenney
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20201013162650.GN3249@paulmck-ThinkPad-P72 \
--to=paulmck@kernel.org \
--cc=boqun.feng@gmail.com \
--cc=cai@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-next@vger.kernel.org \
--cc=linux-tip-commits@vger.kernel.org \
--cc=mingo@kernel.org \
--cc=peterz@infradead.org \
--cc=rostedt@goodmis.org \
--cc=sfr@canb.auug.org.au \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox