Re: [Lse-tech] Subject: [RFC][PATCH] Fix for unsafe notifier chain mechanism

[Alan Stern <stern@rowland.harvard.edu>]

On Fri, 11 Nov 2005, Paul E. McKenney wrote:

> > > > +	down_write(&nh->rwsem);
> > > > +	nl = &nh->head;
> > > > +	while ((*nl) != NULL) {
> > > > +		if (n->priority > (*nl)->priority)
> > > > +			break;
> > > > +		nl = &((*nl)->next);
> > > > +	}
> > > > +	rcu_assign_pointer(n->next, *nl);
> > > 
> > > The above can simply be "n->next = *nl;".  The reason is that this change
> > > of state is not visible to RCU readers until after the following statement,
> > > and it therefore need not be an RCU-reader-safe assignment.  You only need
> > > to use rcu_assign_pointer() when the results of the assignment are
> > > immediately visible to RCU readers.
> > 
> > Correct, the rcu call isn't really needed.  It doesn't hurt perceptibly,
> > though, and part of the RCU documentation states:
> > 
> >  * ...  More importantly, this
> >  * call documents which pointers will be dereferenced by RCU read-side
> >  * code.
> > 
> > For that reason, I felt it was worth putting it in.
> 
> But the following statement does a much better job of documenting the
> pointer that is to be RCU-dereferenced.  Duplicate documentation can
> be just as confusing as no documentation.

It's not really duplicate documentation since _both_ pointers are to be 
RCU-dereferenced.  But maybe you mean that only the second pointer can be 
RCU-dereferenced at the time the write occurs?  I don't think that's what 
the documentation comment intended.

Alan Stern

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/