Re: [PATCH 0/7] Add rcu_sync infrastructure to avoid _expedited() in percpu-rwsem

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

On Sat, Jul 18, 2015 at 01:29:30AM +0200, Oleg Nesterov wrote:
> On 07/15, Paul E. McKenney wrote:
> >
> > On Wed, Jul 15, 2015 at 09:36:01PM +0200, Oleg Nesterov wrote:
> > >
> > > Do you mean you need another user except percpu_rw_semaphore? I do
> > > not see any right now...
> >
> > Not asking for more than one use, but it does need a use.  I believe
> > that percpu_rw_semaphore suffices.
> >
> > > Let me remind about sb_writers again. It actually has 3 rw_sem's
> > > and I am trying to turn then into percpu_rw_semaphore's.
> > >
> > > In this case freeze_super() will need 6 synchronize_sched_expedited().
> > > This just looks ugly. But if we have rcu_sync primitives, all 3 sem's
> > > in struct super_block can share the same "struct rcu_sync", and
> > > freeze_super() will need only once synchronize_sched().
> >
> > Makes sense.
> 
> Great, thanks. And iiuc Linus doesn't object to this particular change.
> Plus I see the "Make checkpatch.pl warn on expedited RCU grace periods"
> patch ;)

Note that it is a warning rather than an error.  ;-)

> So can I assume you will take these changes?
> 
> I do not need them right now, just I need to know what should I do in
> destroy_super() and (much more importantly) what should I say in the
> changelogs if I try to convert sb_writers to use percpu_rw_semaphore.

Yes, given a real use case, which you do appear to have, I will take
these changes.

								Thanx, Paul