Re: [PATCH] rcu: call __rcu_read_unlock() in exit_rcu()

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Tue, Jan 04, 2011 at 04:32:25PM +0800, Lai Jiangshan wrote:
> 
> rcu_read_unlock() is a wrapper of many things:
> 	the core __rcu_read_unlock(), debugs...etc.
> 
> It is better that we only calls the core __rcu_read_unlock() here.

Hello, Lai!

Last time I checked, a task exiting in an RCU read-side critical section
got an error before it got to this point.  That said, the "__" does a
good job of indicating that this is not a normal end of an RCU read-side
critical section.

I have queued this at least for the moment, and also created a TINY_RCU
counterpart.

							Thanx, Paul

> signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> ---
> diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h
> index a363871..94035a0 100644
> --- a/kernel/rcutree_plugin.h
> +++ b/kernel/rcutree_plugin.h
> @@ -802,7 +802,7 @@ void exit_rcu(void)
>  	if (t->rcu_read_lock_nesting == 0)
>  		return;
>  	t->rcu_read_lock_nesting = 1;
> -	rcu_read_unlock();
> +	__rcu_read_unlock();
>  }
> 
>  #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */