Re: [patch] lockdep: fix possible races while disabling lock-debugging

[Jarek Poplawski <jarkao2@o2.pl>]

On Thu, Dec 07, 2006 at 02:29:03PM +0100, Ingo Molnar wrote:
> Subject: [patch] lockdep: fix possible races while disabling lock-debugging
> From: Ingo Molnar <mingo@elte.hu>
...
> (also note that as we all know the Linux kernel is, by definition, 
> bug-free and perfect, so this code never triggers, so these fixes are 
> highly theoretical. I wrote this patch for aesthetic reasons alone.)

Now it's too sexy! I can't work.

PS: I had some problems with patching - probably
because of something from -mm, but:

> Signed-off-by: Ingo Molnar <mingo@elte.hu>
...
> @@ -567,12 +601,10 @@ static noinline int print_circular_bug_t
>  	if (debug_locks_silent)
>  		return 0;
>  
> -	/* hash_lock unlocked by the header */
> -	__raw_spin_lock(&hash_lock);
>  	this.class = check_source->class;
>  	if (!save_trace(&this.trace))
>  		return 0;
> -	__raw_spin_unlock(&hash_lock);
> +

IMHO lock is needed here for save_trace.

...
> @@ -1212,7 +1244,8 @@ register_lock_class(struct lockdep_map *
>  	hash_head = classhashentry(key);
>  
>  	raw_local_irq_save(flags);
> -	__raw_spin_lock(&hash_lock);
> +	if (!graph_lock())

	! raw_local_irq_restore(flags);

> +		return NULL;
>  	/*
>  	 * We have to do the hash-walk again, to avoid races
>  	 * with another CPU:

Jarek P.