From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932844Ab0DGP5g (ORCPT ); Wed, 7 Apr 2010 11:57:36 -0400 Received: from e9.ny.us.ibm.com ([32.97.182.139]:59415 "EHLO e9.ny.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932729Ab0DGP5c (ORCPT ); Wed, 7 Apr 2010 11:57:32 -0400 Date: Wed, 7 Apr 2010 08:57:29 -0700 From: "Paul E. McKenney" To: David Howells Cc: Eric Dumazet , Trond.Myklebust@netapp.com, linux-nfs@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH] NFS: Fix RCU warnings in nfs_inode_return_delegation_noreclaim() [ver #2] Message-ID: <20100407155729.GA2481@linux.vnet.ibm.com> Reply-To: paulmck@linux.vnet.ibm.com References: <2440.1269967151@redhat.com> <21972.1269993064@redhat.com> <10818.1270044273@redhat.com> <15371.1270057054@redhat.com> <19556.1270076008@redhat.com> <14003.1270122314@redhat.com> <4161.1270133211@redhat.com> <23331.1270570443@redhat.com> <26510.1270582446@redhat.com> <24225.1270646561@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <24225.1270646561@redhat.com> User-Agent: Mutt/1.5.20 (2009-06-14) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Apr 07, 2010 at 02:22:41PM +0100, David Howells wrote: > Paul E. McKenney wrote: > > > +#define rcu_access_pointer(p, c) \ > > Why is there a need for 'c'? An example use is where rcu_access_pointer() is legal because we are either initializing or cleaning up, so that no other CPU has access to the pointer. In these cases, you might do something like: q = rcu_access_pointer(p->a, p->refcnt == 0); > > +#define rcu_dereference_protect(p, c) \ > > I'd prefer rcu_dereference_protected(), I think. This macro doesn't protect > anything. Also, again, why the need for 'c'? Agreed on rcu_dereference_protected(). I succumbed to a fit of "make the identifier shorter", please accept my apologies. > For instance, in: > > static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid) > { > struct nfs_delegation *delegation = > rcu_dereference_protected(nfsi->delegation, ????); > > what would be the condition? That the spinlock is held? That's a condition > for calling the function. Yep, that the spinlock is held. I agree that it is a bit obvious in this case, but I have come across a number of RCU uses where the lock in question was acquired many function calls removed from the access, and where there other locks were held for other purposes. > And in: > > void nfs_inode_return_delegation_noreclaim(struct inode *inode) > { > struct nfs_client *clp = NFS_SERVER(inode)->nfs_client; > struct nfs_inode *nfsi = NFS_I(inode); > struct nfs_delegation *delegation; > > if (rcu_access_pointer(nfsi->delegation, ????) != NULL) { > > what would be the condition here? There's no lock to check - that's the whole > point of the macro. I also can't give it nfsi->delegation to check as the > value may change between the two accesses. I suggest something like the following: /* protected by double-check lock pattern. */ if (rcu_access_pointer(nfsi->delegation, 1) != NULL) { Thanx, Paul