[ofa-general] Re: [PATCH v2 2/4] rdma/cma: simplify locking needed for serialization of callbacks

[Roland Dreier]

 > @@ -2570,9 +2557,12 @@ static int cma_ib_mc_handler(int status,
 >  	int ret;
 > 
 >  	id_priv = mc->id_priv;
 > -	if (cma_disable_remove(id_priv, CMA_ADDR_BOUND) &&
 > -	    cma_disable_remove(id_priv, CMA_ADDR_RESOLVED))
 > +	mutex_lock(&id_priv->handler_mutex);
 > +	if ((id_priv->state != CMA_ADDR_BOUND) &&
 > + 	    (id_priv->state != CMA_ADDR_RESOLVED)) {
 > +		mutex_unlock(&id_priv->handler_mutex);
 >  		return 0;
 > +	}

why don't you just do

	if (cma_disable_callback(id_priv, CMA_ADDR_BOUND) &&
	    cma_disable_callback(id_priv, CMA_ADDR_RESOLVED))

instead of open-coding here?  (&& is specified to have short-circuit
semantics, so if the first call takes the mutex, the second call won't
be made)

It seems your change is a subtle change in semantics and breaks the
encapsulation of this callback disabling here.

 - R.