From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: Patch: [NET]: Remove CONFIG_PROC_FS depency for pcounter inuse Date: Sat, 01 Mar 2008 14:45:54 +0100 Message-ID: <47C95E12.609@cosmosbay.com> References: <47C8E777.5050804@hartkopp.net> <47C91BC6.6050603@cosmosbay.com> <47C93C76.5090905@hartkopp.net> <47C945D7.7050601@cosmosbay.com> <47C950B4.5040703@hartkopp.net> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Arnaldo Carvalho de Melo , David Miller , netdev@vger.kernel.org To: Oliver Hartkopp Return-path: Received: from neuf-infra-smtp-out-sp604007av.neufgp.fr ([84.96.92.120]:40944 "EHLO neuf-infra-smtp-out-sp604007av.neufgp.fr" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755382AbYCANqQ (ORCPT ); Sat, 1 Mar 2008 08:46:16 -0500 In-Reply-To: <47C950B4.5040703@hartkopp.net> Sender: netdev-owner@vger.kernel.org List-ID: Oliver Hartkopp a =E9crit : > Eric Dumazet wrote: >> Oliver Hartkopp a =E9crit : >>> It's not about the counter implementation but the integration/usage= =20 >>> in the networking subsystem. >>> >>> Or does your mentioned patch mean, that the added functions in=20 >>> proto_[un|]register() will also be reverted? >>> >> >> A patch will make inet use percpu_counter instead of pcounter. >> >> Then a zap patch will delete lib/pcounter.c & include/linux/pcounter= =2Eh >> >> I dont understand why you say CONFIG_PROC_FS is *forced*. >> I can build a kernel with CONFIG_PROC_FS=3Dn, with working INET. > Right. With enabled CONFIG_EMBEDDED you might have CONFIG_INET with=20 > CONFIG_PROC_FS=3Dn. >=20 > But this is not the thing, i wanted to point out. >=20 > My major concern was, that "whatever-per-cpu-counters" are=20 > allocated/initialized in "proto_register()" for *every* network proto= col=20 > but *only* IPv[4|6] is using these counters (when CONFIG_PROC_FS is s= et). >=20 > I just wanted to point out the situation for network protocols that d= o=20 > not need any inuse counters. In the current implementation the pcount= ers=20 > are allocated for every networking protocol in proto_register() which= =20 > does not look optimized to me. >=20 > Will this change with your patch that uses percpu_counter instead of=20 > pcounter?? >=20 Yes, everything will be cleaned by me. We dont need to *optimize* something that will die very soon. Thank you