Re: [PATCH] NETLABEL: Fix an RCU warning

[Eric Dumazet <eric.dumazet@gmail.com>]

Le lundi 29 mars 2010 à 16:05 -0400, Paul Moore a écrit :

> Okay, is there a recommended approach towards accessing RCU-protected pointers 
> both under a RCU read lock and under only a spinlock (or similar lock 
> construct)?  I know I could do something based on querying the state of the 
> RCU/etc. locks but that seems like a hack and could interfere with some of the 
> logic used to detect coding problems.
> 

Say you use a common function func1(), that use RCU protection, from a
pure reader, and from a pure writer.

pure_reader()
{
	rcu_read_lock();
	res = func1();
	rcu_read_unlock();
}

pure_writer()
{
	spin_lock(&some_lock);
	res = func1();
	spin_unlock(&some_lock);
}

then, func1 could use

func1()
{
 ..... ptr = rcu_dereference_check(xxx->ptr,
				   rcu_read_lock_held() || 
				   lockdep_is_held(&some_lock));
...
}


They are numerous examples in tree.
If some iterators use implicit rcu_dereference(), you'll probably need
to define new iterators, so that full condition can be tested.

rcu_dereference() is a shorthand for 
	rcu_dereference_check(p, rcu_read_lock_held())

This only deals for pure readers, not potential pure writer :)