From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: dst->obsolete has become pointless Date: Thu, 10 Nov 2011 01:33:43 +0100 Message-ID: <1320885223.5825.37.camel@edumazet-laptop> References: <1320842948.6923.3.camel@Joe-Laptop> <20111109.142044.1307358041881082836.davem@davemloft.net> <1320882969.6923.29.camel@Joe-Laptop> <20111109.192427.661860445333469011.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: joe@perches.com, steffen.klassert@secunet.com, netdev@vger.kernel.org, timo.teras@iki.fi To: David Miller Return-path: Received: from mail-ww0-f44.google.com ([74.125.82.44]:62636 "EHLO mail-ww0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932214Ab1KJAds (ORCPT ); Wed, 9 Nov 2011 19:33:48 -0500 Received: by wwh12 with SMTP id 12so1515016wwh.1 for ; Wed, 09 Nov 2011 16:33:47 -0800 (PST) In-Reply-To: <20111109.192427.661860445333469011.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-ID: Le mercredi 09 novembre 2011 =C3=A0 19:24 -0500, David Miller a =C3=A9c= rit : > We can only use it in this scenerio if it universally > uses the same type. This is due to the by-hand layout > of the rest of the structure such that __refcnt ends > up on a different cache line. > -- tbench (or any tcp on loopback workload) is currently hitting a single dst refcnt because a single dst is shared by all sockets... Now some information was migrated to inetpeer, I wonder if we could not allocate a private dst for a socket ? Just an idea before the night...