Re: [PATCH tip/core/rcu 0/21] v6 add lockdep-based diagnostics to rcu_dereference()

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

On Tue, Feb 23, 2010 at 06:15:59PM +0100, Arnd Bergmann wrote:
> On Tuesday 23 February 2010, Paul E. McKenney wrote:
> > We would also need something for initialization of structure fields.
> > Does __force work in that case as well?
> 
> Yes, it will just need some syntactical sugar to avoid placing __force
> in device drivers.

Very good!

> > > If there are cases where it does not work, we need to come up with
> > > names for new primitives that just do the assignment or dereference
> > > with __force but no actual synchronization.
> > 
> > Some data structures are shared by RCU and non-RCU code, with struct
> > list_head being the most prominent example.  Making the "next" pointer
> > as __rcu might be OK, but there are a -lot- of non-RCU uses of struct
> > list_head.  Would we really want to introduce rcu_dereference() to all
> > non-RCU list-traversal primitives, or do we need to do something else?
> 
> I've just started an experimental implementation and got stuck at list rcu.
> The two to deal with it that I can see are
> - ignore list-rcu for now, and make all include/linux/rculist.h __force the
>   problem to be ignored.
> - introduce a new struct rcu_list_head that needs to be used for list rcu.
> 
> A nicer option might be if sparse would let you write
> 'struct list_head __rcu head' and interpret that as having the pointers
> inside it annotated as __rcu.

Only the "next" pointer, not the "prev" pointer, but yes.

Perhaps it would be best to see if the sparse guys are willing to take
this on?

							Thanx, Paul