Messages in this thread | | | Date | Tue, 6 Apr 2010 10:29:44 -0700 | From | "Paul E. McKenney" <> | Subject | Re: [PATCH] NFS: Fix RCU warnings in nfs_inode_return_delegation_noreclaim() [ver #2] |
| |
On Tue, Apr 06, 2010 at 05:14:03PM +0100, David Howells wrote: > Paul E. McKenney <paulmck@linux.vnet.ibm.com> wrote: > > > > > So you have objected to needless memory barriers. How do you feel > > > > about possibly needless ACCESS_ONCE() calls? > > > > > > That would work here since it shouldn't emit any excess instructions. > > > > And here is the corresponding patch. Seem reasonable? > > Actually, now I've thought about it some more. No, it's not reasonable. > You've written: > > This patch adds a variant of rcu_dereference() that handles situations > where the RCU-protected data structure cannot change, perhaps due to > our holding the update-side lock, or where the RCU-protected pointer is > only to be tested, not dereferenced. > > But if we hold the update-side lock, then why should we be forced to use > ACCESS_ONCE()? > > In fact, if we don't hold the lock, but we want to test the pointer twice in > succession, why should we be required to use ACCESS_LOCK()?
OK, just to make sure I understand you... You are asking for two additional RCU API members:
1. rcu_access_pointer() or some such that includes ACCESS_ONCE(), but not smp_read_barrier_depends(), which may be used when we are simply examining the value of the RCU-protected pointer (as in the NFS case). It could also be used when the appropriate update-side lock is held, but for that we have:
2. rcu_dereference_protected() or some such that includes neither ACCESS_ONCE() nor smp_read_barrier_depends(), and that may only be used if updates are prevented, for example, by holding the appropriate update-side lock.
Does this fit?
Thanx, Paul
| |