Re: [patch 01/15] Kernel Tracepoints

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

On Sat, Aug 02, 2008 at 02:03:57AM +0200, Peter Zijlstra wrote:
> On Fri, 2008-08-01 at 14:10 -0700, Paul E. McKenney wrote:
> 
> > Yeah, I was thinking in terms of rcu_dereference() working with both
> > rcu_assign_pointer() and an as-yet-mythical rcu_assign_index().  Perhaps
> > this would be a good time to get better names:
> > 
> > Current:	rcu_assign_pointer()	rcu_dereference()
> > New Pointers:	rcu_publish_pointer()	rcu_subscribe_pointer()
> > New Indexes:	rcu_publish_index()	rcu_subscribe_index()
> 
> Is it really worth the effort, splitting it out into these two cases?

Either we should split out into the pointer/index cases, or the
definition of rcu_assign_pointer() should probably lose its current
check for NULL...

> > And, while I am at it, work in a way of checking for either being in
> > the appropriate RCU read-side critical section and/or having the
> > needed lock/mutex/whatever held -- something I believe PeterZ was
> > prototyping some months back.
> 
> Yeah - I have (bitrotted a bit, but should be salvageable) lockdep
> annotations for rcu_dereference().
> 
> The problem with them is the huge amount of false positives.. Take for
> example the Radix tree code, its perfectly fine to use the radix tree
> code without RCU - say you do the old rwlock style, still it uses
> rcu_dereference().
> 
> I never figured out a suitable way to annotate that.

My thought was to add a second argument that contained a boolean.  If
the rcu_dereference() was either within an RCU read-side critical
section on the one hand or if the boolean evaluated to "true" on the
other, then no assertion.  This would require SPIN_LOCK_HELD() or
similar primitives.  (And one of the reasons for the renaming

Of course, in the case of radix tree, it would be necessary to pass the
boolean in through the radix-tree read-side APIs, which would perhaps be
a bit annoying.

							Thanx, Paul