Re: [1/4] rcu: Detect uses of rcu read side in extended quiescent states

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Jun 06, 2011 at 05:42:50PM -0700, Paul E. McKenney wrote:
> On Tue, Jun 07, 2011 at 02:19:07AM +0200, Frederic Weisbecker wrote:
> > On Mon, Jun 06, 2011 at 11:10:21AM -0700, Paul E. McKenney wrote:
> > > commit c15d76f26712bd5228aa0c6af7a7e7c492a812c9
> > > Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > > Date:   Tue May 24 08:31:09 2011 -0700
> > > 
> > >     rcu: Restore checks for blocking in RCU read-side critical sections
> > >     
> > >     Long ago, using TREE_RCU with PREEMPT would result in "scheduling
> > >     while atomic" diagnostics if you blocked in an RCU read-side critical
> > >     section.  However, PREEMPT now implies TREE_PREEMPT_RCU, which defeats
> > >     this diagnostic.  This commit therefore adds a replacement diagnostic
> > >     based on PROVE_RCU.
> > >     
> > >     Because rcu_lockdep_assert() and lockdep_rcu_dereference() are now being
> > >     used for things that have nothing to do with rcu_dereference(), rename
> > >     lockdep_rcu_dereference() to lockdep_rcu_suspicious() and add a third
> > >     argument that is a string indicating what is suspicious.  This third
> > >     argument is passed in from a new third argument to rcu_lockdep_assert().
> > >     Update all calls to rcu_lockdep_assert() to add an informative third
> > >     argument.
> > >     
> > >     Finally, add a pair of rcu_lockdep_assert() calls from within
> > >     rcu_note_context_switch(), one complaining if a context switch occurs
> > >     in an RCU-bh read-side critical section and another complaining if a
> > >     context switch occurs in an RCU-sched read-side critical section.
> > >     These are present only if the PROVE_RCU kernel parameter is enabled.
> > >     
> > >     Again, you must enable PROVE_RCU to see these new diagnostics.  But you
> > >     are enabling PROVE_RCU to check out new RCU uses in any case, aren't you?
> > >     
> > >     Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > 
> > A little comment about this patch:
> > 
> > <snip>
> > > diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> > > index 88547c8..8b4b3da 100644
> > > --- a/kernel/rcutree.c
> > > +++ b/kernel/rcutree.c
> > > @@ -153,6 +153,12 @@ void rcu_bh_qs(int cpu)
> > >   */
> > >  void rcu_note_context_switch(int cpu)
> > >  {
> > > +	rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map),
> > > +			   "Illegal context switch in RCU-bh"
> > > +			   " read-side critical section");
> > > +	rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map),
> > > +			   "Illegal context switch in RCU-sched"
> > > +			   " read-side critical section");
> > 
> > This looks like more a check to make inside might_sleep().
> > It's better because might_sleep() triggers the check even if
> > we don't actually go to sleep.
> 
> This does make quite a bit of sense.
> 
> > In fact I believe might_sleep() already does the job fine:
> > 
> > If !PREEMPT, might_sleep() detects that preemption is disabled
> > by rcu_read_lock().
> 
> If !PREEMPT, isn't the preempt_disable() called by rcu_read_lock()
> implemented as follows?
> 
> #define preempt_disable()               do { } while (0)
> 
> Unless I am missing something, __might_sleep() won't detect that.

Ah, right.
 
> > If PREEMPT, might_sleep() checks rcu_preempt_depth().
> 
> Agreed, for CONFIG_TREE_PREEMPT_RCU and CONFIG_TINY_PREEMPT_RCU,
> the existing might_sleep() checks do cover it.
> 
> So I could export an rcu_might_sleep() or some such that contained
> the above two rcu_lockdep_assert()s, but only if !PREEMPT_RCU.
> If PREEMPT_RCU, rcu_might_sleep() would be a no-op.
> 
> Seem reasonable, or am I missing something?

Ok but that only improves the rcu debugging. What about instead improving
might_sleep() to also work in !PREEMPT, so that it profits to any detection
of forbidden sleeping (sleep inside spinlock, preempt_disable, might_fault, etc...)

We could define a new config:

config PREEMPT_COUNT
       default PREEMPT || DEBUG_SPINLOCK_SLEEP

and build preempt_disable()/preempt_enable() on top of that instead
of using CONFIG_PREEMPT directly.

Does that look sane?