From mboxrd@z Thu Jan 1 00:00:00 1970 From: Wang Chen Subject: Re: [PATCH net-next v2]netdevice: 82596: Convert netdev->priv to netdev_priv Date: Mon, 17 Nov 2008 22:42:16 +0800 Message-ID: <492182C8.6060704@cn.fujitsu.com> References: <4921095D.9010109@cn.fujitsu.com> <20081116.223637.108010315.davem@davemloft.net> <49213856.6040802@cn.fujitsu.com> <20081117.012939.72047312.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: jgarzik@pobox.com, netdev@vger.kernel.org To: David Miller Return-path: Received: from cn.fujitsu.com ([222.73.24.84]:58762 "EHLO song.cn.fujitsu.com" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1750783AbYKQOms (ORCPT ); Mon, 17 Nov 2008 09:42:48 -0500 In-Reply-To: <20081117.012939.72047312.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-ID: David Miller said the following on 2008-11-17 17:29: > From: Wang Chen > Date: Mon, 17 Nov 2008 17:24:38 +0800 > >> The memory, which pointed by netdev->priv, can be allocated by alloc_etherdev(). >> We don't need to use __get_free_pages() to get memory. >> >> Signed-off-by: Wang Chen > > I don't think this is right either. > > There might be device imposed alignment constraints, for example. > > And a whole page is also being allocated so that this cache mode > change doesn't have an effect on unrelated kernel memory. That could > kill performance if some critical other data structure sits in the > same page and now has it's cacheability disabled. > On this situation, I think that keeping mem_start and using ml_priv is the only way to remove directly reference of netdev->priv. I will send v3 tomorrow.