From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754908AbaHAKrq (ORCPT ); Fri, 1 Aug 2014 06:47:46 -0400 Received: from stinky.trash.net ([213.144.137.162]:54238 "EHLO stinky.trash.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754528AbaHAKrn (ORCPT ); Fri, 1 Aug 2014 06:47:43 -0400 User-Agent: K-9 Mail for Android In-Reply-To: <20140801103901.GC7331@casper.infradead.org> References: <970429e5-2465-40f2-998b-b82dab3debe8@email.android.com> <20140801103901.GC7331@casper.infradead.org> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=UTF-8 Subject: Re: [PATCH net-next 3/3] nftables: Convert nft_hash to use generic rhashtable From: Patrick McHardy Date: Fri, 01 Aug 2014 11:47:28 +0100 To: Thomas Graf CC: davem@davemloft.net, netdev@vger.kernel.org, linux-kernel@vger.kernel.org, paulmck@linux.vnet.ibm.com, josh@joshtriplett.org, challa@noironetworks.com, walpole@cs.pdx.edu, dev@openvswitch.org, tklauser@distanz.ch, netfilter-devel@vger.kernel.org Message-ID: <173a5240-a4c6-4611-bc03-95d16080fdb5@email.android.com> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 1. August 2014 11:39:01 GMT+01:00, Thomas Graf wrote: >On 08/01/14 at 11:17am, Patrick McHardy wrote: >> On 1. August 2014 09:52:00 GMT+01:00, Thomas Graf >wrote: >> >If the set size is known in advance, the table is sized accordingly, >> >otherwise the table size will default to 64. This is a slight change >> >in behaviour as previously the default was 4 which eventually >required >> >many expansion iterations. >> >> That's assuming a sufficient number of entries will be added. I think >there >> will be many cases in nftables where the number will be lower. Since >> expansion is not very expansive this number was chosen very >deliberately >> and I'd prefer to keep it. >> >> Other than that: >> >> Acked-by: Patrick McHardy > >OK, I will change the nft_hash default size back to 4. Thanks. >I think the expansion is more expensive than it looks though as we >are potentially calling synchronize_rcu() multiple times while >holding a mutex to protect from concurrent mutations. True. On average it should only be a single grace period though IIRC.