Messages in this thread | | | Date | Sat, 5 Oct 2013 19:21:18 +0200 | From | Oleg Nesterov <> | Subject | Re: [PATCH 1/5] rcusync: introduce struct rcu_sync_ops |
| |
On 10/04, Peter Zijlstra wrote: > > On Fri, Oct 04, 2013 at 12:38:37PM -0700, Linus Torvalds wrote: > > On Fri, Oct 4, 2013 at 12:30 PM, Steven Rostedt <rostedt@goodmis.org> wrote: > > > > > > As all the rcu_synchronization() methods (on non UP) are quite > > > expensive, I doubt that this optimization is worth anything. > > > > Maybe. It just annoys me, because afaik, the function that gets called > > is always static per callsite. > > Yes, very much so indeed. Worst is that we have no users of the regular > RCU and RCU_BH variants and only included them for completeness since > the general operation is just as valid for those.
And personally I think we should keep type/ops for completeness anyway, even if we do not have RCU and RCU_BH users. But perhaps we can kill RCU_SYNC and RCU_BH_SYNC enums/entries until we have a user.
Oleg.
| |