Messages in this thread | | | Date | Fri, 4 Sep 2009 13:43:35 -0700 | From | "Paul E. McKenney" <> | Subject | Re: [PATCH] slub: fix slab_pad_check() |
| |
On Fri, Sep 04, 2009 at 11:42:17AM -0400, Christoph Lameter wrote: > On Thu, 3 Sep 2009, Paul E. McKenney wrote: > > > If it were the user of the slab who was invoking some variant of > > call_rcu(), then I would agree with you. > > The user already has to deal with it as explained by Eric.
I didn't read his email that way. Perhaps I misinterpreted it.
> > However, call_rcu() is instead being invoked by the slab itself in the > > case of SLAB_DESTROY_BY_RCU, so that there is no variation in usage. > > Requiring that the user call rcu_barrier() is asking for subtle bugs. > > Therefore, the best approach is to have kmem_cache_destroy() handle > > the RCU cleanup, given that this cleanup is for actions taken by > > kmem_cache_free(), not by the user. > > The user already has to properly handle barriers and rcu logic in order to > use objects handled with RCU properly. Moreover the user even has to check > that the object is not suddenly checked under it. Its already complex.
mm/slab.c has had RCU calls since 2.6.9, so this is not new.
> Guess we are doing this ... Sigh. Are you going to add support other rcu > versions to slab as well as time permits and as the need arises? Should > we add you as a maintainer? ;-)
I don't see any point in adding anything resembling SLAB_DESTROY_BY_RCU_BH, SLAB_DESTROY_BY_RCU_SCHED, or SLAB_DESTROY_BY_SRCU unless and until someone needs it. And I most especially don't see the point of adding (say) SLAB_DESTROY_BY_RCU_BH_SCHED until someone convinces me of the need for it. I would prefer to put the energy into further streamlining the underlying RCU implementation, maybe someday collapsing RCU-BH back into RCU. ;-)
We have gotten along fine with only SLAB_DESTROY_BY_RCU for almost five years, so I think we are plenty fine with what we have. So, as you say, "as the need arises".
I don't see any more need to add me as maintainer of slab and friends than of btrfs, netfilter, selinux, decnet, afs, wireless, or any of a number of other subsystems that use RCU.
Thanx, Paul
| |