Re: __update_max_tr: rcu_read_lock() used illegally while idle!

[Paul E. McKenney]

On Tue, Jul 31, 2012 at 10:51:51AM -0400, Steven Rostedt wrote:
> On Tue, 2012-07-31 at 07:44 -0700, Paul E. McKenney wrote:
> 
> > > Found it (and Cc'd David).
> > > 
> > > In __update_max_tr() we have:
> > > 
> > > 	max_data = task_uid(tsk);
> > > 
> > > where task_uid() is:
> > > 
> > > #define task_uid(task)		(task_cred_xxx((task), uid))
> > > 
> > > #define task_cred_xxx(task, xxx)			\
> > > ({							\
> > > 	__typeof__(((struct cred *)NULL)->xxx) ___val;	\
> > > 	rcu_read_lock();				\
> > > 	___val = __task_cred((task))->xxx;		\
> > > 	rcu_read_unlock();				\
> > > 	___val;						\
> > > })
> > > 
> > > The __update_max_tr() is called at every location interrupts are enabled
> > > (and a max time is discovered). But now this can include places that
> > > rcu_read_lock can not be called, I'm not sure how to handle this. Is
> > > there a non rcu way to get a tasks uid?
> > 
> > OK, I will bite.  How about using something like RCU_NONIDLE(), either
> > directly or open-coded, to make it a legal call site?
> 
> OK, then something like:
> 
> 	RCU_NONIDLE(max_data = task_uid(tsk));
> 
> would work when called normally or with idle?

As long as you don't nest too deeply, both.  There are seven bits for
the count, so you would have to nest quite deeply for there to be
a problem.  And I can safely add several more bits if needed.

							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/