Re: [PATCH 4/6] rcusync: Introduce struct rcu_sync_ops

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

On Fri, Oct 18, 2013 at 02:10:15PM +0200, Oleg Nesterov wrote:
> On 10/18, Lai Jiangshan wrote:
> >
> > On 10/17/2013 11:42 PM, Paul E. McKenney wrote:
> > > On Thu, Oct 17, 2013 at 10:07:15AM +0800, Lai Jiangshan wrote:
> > >> On 10/08/2013 06:25 PM, Peter Zijlstra wrote:
> > >>> From: Oleg Nesterov <oleg@redhat.com>
> > >>>
> > >>> Add the new struct rcu_sync_ops which holds sync/call methods, and
> > >>> turn the function pointers in rcu_sync_struct into an array of struct
> > >>> rcu_sync_ops.
> > >>
> > >> Hi, Paul
> > >>
> > >> I think this work should be done in rcupdate.[ch] side by introducing
> > >> struct rcu_flavor.
> > >
> > > I -do- have on my list to add an rcutorture test for rcu_sync, but
> > > what do you have in mind by adding struct rcu_flavor?  I am guessing
> > > that you do not mean to try to create an rcu_state and a set of
> >
> > No.
> > The thing what I need is just as same as Oleg Nesterov implemented.
> > It is just a structure with several function pointers for different RCU variants.
> > But it would be better if we implement in rcupdate.[ch],
> > and name it to struct rcu_flavor like the URCU.
> >
> > After we have struct rcu_flavor, we can replace the following code
> > to a pointer to struct rcu_flavor.
> >
> > struct rcu_state:
> > 	void (*call)(struct rcu_head *head,	/* call_rcu() flavor. */
> > 		     void (*func)(struct rcu_head *head));
> >
> > struct rcu_torture_ops {
> > 	int (*readlock)(void);
> > 	void (*readunlock)(int idx);
> > 	void (*sync)(void);
> > 	void (*exp_sync)(void);
> > 	void (*call)(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
> > 	void (*cb_barrier)(void);
> > };
> 
> Yes, probably. But it is not clear how/when this rcu_sync will be merged.
> (I hope it will be merged anyway, if nothing else I'll resend these patches
>  for percpu_rw_semaphore with other updates in percpu-rwsem.c).
> 
> Until then, perhaps you can add rcu_flavor/whatever in rcupdate.* ? Then
> rcu_sync can be triviallly updated to use the ops we have in rcupdate.
> And rcutorture.c of course.
> 
> IOW, I think that this should be a separate change, before or after
> rcu_sync.

Perhaps put it in the shiny new kernel/rcu directory in -tip?

							Thanx, Paul