From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH] ipv6: Enable RFS sk_rxhash tracking for ipv6 sockets Date: Wed, 06 Apr 2011 20:37:44 +0200 Message-ID: <1302115064.3209.132.camel@edumazet-laptop> References: <1302113253.3209.126.camel@edumazet-laptop> <20110406.111712.226782643.davem@davemloft.net> <1302114188.3209.128.camel@edumazet-laptop> <20110406.112712.71110955.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: nhorman@tuxdriver.com, netdev@vger.kernel.org, kuznet@ms2.inr.ac.ru, pekkas@netcore.fi, jmorris@namei.org, yoshfuji@linux-ipv6.org, kaber@trash.net, therbert@google.com To: David Miller Return-path: Received: from mail-ww0-f44.google.com ([74.125.82.44]:38045 "EHLO mail-ww0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756145Ab1DFShs (ORCPT ); Wed, 6 Apr 2011 14:37:48 -0400 Received: by wwa36 with SMTP id 36so2123248wwa.1 for ; Wed, 06 Apr 2011 11:37:47 -0700 (PDT) In-Reply-To: <20110406.112712.71110955.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-ID: Le mercredi 06 avril 2011 =C3=A0 11:27 -0700, David Miller a =C3=A9crit= : > From: Eric Dumazet > Date: Wed, 06 Apr 2011 20:23:08 +0200 >=20 > > Le mercredi 06 avril 2011 =C3=A0 11:17 -0700, David Miller a =C3=A9= crit : > >> From: Eric Dumazet > >> Date: Wed, 06 Apr 2011 20:07:33 +0200 > >>=20 > >> > This is why we used on ipv4 : > >> >=20 > >> > if (inet_sk(sk)->inet_daddr) > >> > sock_rps_save_rxhash(sk, skb->rxhash); > >> >=20 > >> >=20 > >> > Only arm RFS on UDP if socket is bound to a given remote peer. > >>=20 > >> Agreed, Neil please make this change to your ipv6 code. > >=20 > > BTW, do you guys know if NFS is using RFS right now (if TCP transpo= rt is > > used) ? >=20 > It ought to be. Are you specifically concerned that it might be > missing the rxhash setting calls because of the APIs it uses to > do socket I/O? >=20 > The SunRPC layer is a major user of tcp_read_sock(), for example. >=20 > But the packet processing will be done via the normal means, > therefore we'll hit tcp_v4_do_rcv() and therefore save the > rxhash. Hmm, are multiple tcp sessions used between NFS client and server (One per cpu on client ?)