From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH 1/3] net: introduce a list of device addresses dev_addr_list (v3) Date: Sat, 18 Apr 2009 10:06:42 +0200 Message-ID: <49E98A12.9020302@cosmosbay.com> References: <20090313183303.GF3436@psychotron.englab.brq.redhat.com> <20090415081720.GA21342@psychotron.englab.brq.redhat.com> <20090415081819.GB21342@psychotron.englab.brq.redhat.com> <20090415180215.GA22540@psychotron.englab.brq.redhat.com> <20090417115723.GE9556@psychotron.englab.brq.redhat.com> <20090417083315.2f089755@nehalam> <20090418070151.GA3370@psychotron.englab.brq.redhat.com> <49E982C4.8020407@cosmosbay.com> <20090418074427.GB3370@psychotron.englab.brq.redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Stephen Hemminger , linux-kernel@vger.kernel.org, netdev@vger.kernel.org, jgarzik@pobox.com, davem@davemloft.net, bridge@lists.linux-foundation.org, fubar@us.ibm.com, bonding-devel@lists.sourceforge.net, kaber@trash.net, mschmidt@redhat.com, ivecera@redhat.com To: Jiri Pirko Return-path: Received: from gw1.cosmosbay.com ([212.99.114.194]:38469 "EHLO gw1.cosmosbay.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751466AbZDRIIj convert rfc822-to-8bit (ORCPT ); Sat, 18 Apr 2009 04:08:39 -0400 In-Reply-To: <20090418074427.GB3370@psychotron.englab.brq.redhat.com> Sender: netdev-owner@vger.kernel.org List-ID: Jiri Pirko a =E9crit : > Sat, Apr 18, 2009 at 09:35:32AM CEST, dada1@cosmosbay.com wrote: >> Jiri Pirko a =E9crit : >>> Fri, Apr 17, 2009 at 05:33:15PM CEST, shemminger@vyatta.com wrote: >>> >>> >>> >>>>> +struct netdev_hw_addr { >>>>> + struct list_head list; >>>>> + unsigned char addr[MAX_ADDR_LEN]; >>>>> + int refcount; >>>>> + struct rcu_head rcu_head; >>>>> +}; >>>> Minor nit, the ordering of elements cause holes that might not be >>>> needed. >>> Agree that ordering might be done better. Will do. >>>> Space saving? is rcu_head needed or would using synchronize_net >>>> make code cleaner and save space.=20 >>>> >>> Well I originaly had this done by synchronize_rcu(). Eric argued th= at it might >>> cause especially __hw_addr_del_multiple_ii() to run long and sugges= ted to use >>> call_rcu() instead. I plan to switch this to kfree_rcu() (or whatev= er it's >>> called) once it hits the tree. >>> >> Yes, and dont forget we wont save space, as we allocate a full >> cache line to hold a 'struct netdev_hw_addr', since we dont want thi= s >> critical and read_mostly object polluted by a hot spot elsewhere in = kernel... >> >> Considering this, letting 'rcu_head' at the end of structure, even i= f we >> have an eventual hole on 64 bit arches is not really a problem, and = IMHO >> the best thing to do, as rcu_head is only used at dismantle time. >=20 > I will order the struct better, there are archs with small cache line= size where > it makes sense. How exactly ? If you consider a 32bit arch with 16 or 32 bytes cache line, sizeof(struct_list_dead) is 8 sizeof(addr) =3D 32 (but we really use 6 bytes for ethernet) struct netdev_hw_addr { unsigned char addr[MAX_ADDR_LEN]; struct list_head list; int refcount; struct rcu_head rcu_head; }; would cost more at lookup time, since we would use two cache lines struct netdev_hw_addr { struct list_head list; unsigned char addr[MAX_ADDR_LEN]; int refcount; struct rcu_head rcu_head; }; Is nicer, because at least 8 bytes of addr share the same cache line than list. So direct dev->dev_addr would be fast (for devices with one address), and is_etherdev_addr() would still use one cache line per item.