Re: Weird rcu lockdep warning

[Paul E. McKenney]

On Wed, Apr 14, 2010 at 11:34:33AM +0800, Lai Jiangshan wrote:
> Paul E. McKenney wrote:
> > On Tue, Apr 13, 2010 at 05:13:06PM -0700, David Miller wrote:
> >> From: Frederic Weisbecker <fweisbec@xxxxxxxxx>
> >> Date: Wed, 14 Apr 2010 02:02:27 +0200
> >>
> >>> I just have a guess though....
> >>> This seems to always happen from NMI path, and lockdep is disabled on NMI.
> >>> I suspect the lock_acquire() performed by rcu_read_lock() is just ignored
> >>> and then the rcu_read_lock_held() check has the wrong result...
> >> Yeah, I bet that's it too.
> >>
> >> lock_is_held() can't return anything meaningful while lockdep is
> >> disabled, which it is during NMIs.
> > 
> > Ah!  So I just need to add a "current->lockdep_recursion"
> > check to debug_lockdep_rcu_enabled().  And move the function to
> > kernel/rcutree_plugin.h to avoid #include hell.
> > 
> > See below for (untested) patch.
> > 
> > 						Thanx, Paul
> > 
> > ------------------------------------------------------------------------
> > 
> >  include/linux/rcupdate.h |    5 +----
> >  kernel/rcutree_plugin.h  |   11 +++++++++++
> >  2 files changed, 12 insertions(+), 4 deletions(-)
> > 
> > commit 304d8da6cd791a81ce3164f867e1b3ef4f9af1d1
> > Author: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> > Date:   Tue Apr 13 18:45:51 2010 -0700
> > 
> >     rcu: Make RCU lockdep check the lockdep_recursion variable
> >     
> >     The lockdep facility temporarily disables lockdep checking by incrementing
> >     the current->lockdep_recursion variable.  Such disabling happens in NMIs
> >     and in other situations where lockdep might expect to recurse on itself.
> >     This patch therefore checks current->lockdep_recursion, disabling RCU
> >     lockdep splats when this variable is non-zero.
> >     
> >     Reported-by: Frederic Weisbecker <fweisbec@xxxxxxxxx>
> >     Reported-by: David Miller <davem@xxxxxxxxxxxxx>
> >     Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> > 
> > diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
> > index 9f1ddfe..07db2fe 100644
> > --- a/include/linux/rcupdate.h
> > +++ b/include/linux/rcupdate.h
> > @@ -101,10 +101,7 @@ extern struct lockdep_map rcu_sched_lock_map;
> >  # define rcu_read_release_sched() \
> >  		lock_release(&rcu_sched_lock_map, 1, _THIS_IP_)
> >  
> > -static inline int debug_lockdep_rcu_enabled(void)
> > -{
> > -	return likely(rcu_scheduler_active && debug_locks);
> > -}
> > +extern int debug_lockdep_rcu_enabled(void);
> >  
> >  /**
> >   * rcu_read_lock_held - might we be in RCU read-side critical section?
> > diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h
> > index 79b53bd..2169abe 100644
> > --- a/kernel/rcutree_plugin.h
> > +++ b/kernel/rcutree_plugin.h
> > @@ -1067,3 +1067,14 @@ static void rcu_needs_cpu_flush(void)
> >  }
> >  
> >  #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
> > +
> > +#ifdef CONFIG_DEBUG_LOCK_ALLOC
> > +
> > +int debug_lockdep_rcu_enabled(void)
> > +{
> > +	return likely(rcu_scheduler_active &&
> > +		      debug_locks &&
> > +		      current->lockdep_recursion == 0);
> > +}
> > +
> 
> Looks good to me too, but I think
> 'likely' is needless since the function is not inline.

Good point.  And to add injury to insult, I forgot EXPORT_SYMBOL_GPL().

Updated patch in the works.

							Thanx, Paul
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/