From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754813AbbE1VRA (ORCPT ); Thu, 28 May 2015 17:17:00 -0400 Received: from e37.co.us.ibm.com ([32.97.110.158]:35501 "EHLO e37.co.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754743AbbE1VQw (ORCPT ); Thu, 28 May 2015 17:16:52 -0400 Date: Thu, 28 May 2015 14:16:47 -0700 From: "Paul E. McKenney" To: Dan Streetman Cc: Steven Rostedt , Andrew Morton , Josh Triplett , Mathieu Desnoyers , Lai Jiangshan , linux-kernel Subject: Re: [PATCH 1/2] rcu: introduce list_last_or_null_rcu Message-ID: <20150528211647.GO5989@linux.vnet.ibm.com> Reply-To: paulmck@linux.vnet.ibm.com References: <1432845328-27932-1-git-send-email-ddstreet@ieee.org> <20150528170542.3b3a7b48@gandalf.local.home> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) X-TM-AS-MML: disable X-Content-Scanned: Fidelis XPS MAILER x-cbid: 15052821-0025-0000-0000-00000B0200DF Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, May 28, 2015 at 05:12:00PM -0400, Dan Streetman wrote: > On Thu, May 28, 2015 at 5:05 PM, Steven Rostedt wrote: > > On Thu, 28 May 2015 16:35:27 -0400 > > Dan Streetman wrote: > > > >> Add list_last_or_null_rcu(), to simplify getting the last entry from a > >> rcu-protected list. The standard list_last_entry() can't be used as it > >> is not rcu-protected; the list may be modified concurrently. And the > >> ->prev pointer can't be used, as only the ->next pointers are protected > >> by rcu. > >> > >> This simply iterates forward through the entire list, to get to the last > >> entry. If the list is empty, it returns NULL. > > > > May I asked what this would be used for? It seems awfully inefficient > > in its implementation. What use cases would this be for? I hate to add > > something like this as a generic function which would encourage people > > to use it. Iterating over an entire list to find the last element is > > just nasty. > > i have a patch series that will update zswap to be able to change its > parameters at runtime, instead of only at boot time. To do that, it > creates new "pools" dynamically, and keeps them all in a list, with > only the 1st pool being actively used; any following pools still have > everything that was stored in them, but they aren't added to. When > zswap has to "shrink" - by telling one of the pools to get rid of 1 or > more items - it picks the last on the list. Once a pool is empty, > it's removed/freed. > > So zswap *could* just manually iterate the list to the last element, > instead of using this new function. But if rcu lists are ever > improved later on, e.g. if ->prev is somehow rcu-protected as well as > ->next, then this function should be faster than manually iterating. > > if there's a better rcu-way to get to the last list entry, then we > should definitely use it, although based on my understanding of the > rcu list implementation, you can only iterate forwards, safely > (without locking). The usual approach would be to maintain a tail pointer. How big are these lists likely to get? Thanx, Paul