Re: [PATCH] jump label: close race in jump_label_inc() vs. jump_label_dec()

[Gleb Natapov <gleb@redhat.com>]

On Wed, Jan 04, 2012 at 10:32:37AM -0500, Jason Baron wrote:
> The previous fix to ensure that jump_label_inc() does not return until the jump
> is completely patched, opened up a race in the inc/dec path. The scenario is:
> 
> key->enabled = 0;
> 
> 	CPU 0					CPU 1
>         ----- 					-----
> 
> jump_label_inc():			jump_label_dec():
> 
> 1) if (atomic_read(&key->enabled) == 0)
> 	jump_label_update(key, JUMP_LABEL_ENABLE);
> 
> 					 2) if (!atomic_dec_and_mutex_lock(&key->enabled, &jump_label_mutex))
> 					 	return;
> 
> 3) atomic_inc(&key->enabled);
> 
> So now, key->enabled = 0, but the jump has been enabled, which is an invalid
> state.
> 
Isn't it an indication of a higher level bug if jump_label_dec() is
called on a disabled jump label? In other words isn't key->enabled == -1
invalid sate by itself? I do not see how the call sequence above can
happen with perf events for instance. jump_label_dec() is called on
event destruction and if key->enabled = 0 then there is no events to
destroy.

--
			Gleb.