Return-path: Received: from gundega.hpl.hp.com ([192.6.19.190]:58117 "EHLO gundega.hpl.hp.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1030771AbXCHWL6 (ORCPT ); Thu, 8 Mar 2007 17:11:58 -0500 Date: Thu, 8 Mar 2007 14:11:28 -0800 To: Johannes Berg Cc: Jouni Malinen , Michael Buesch , linux-wireless@vger.kernel.org, netdev , Jeff Garzik , Dan Williams Subject: Re: wireless extensions vs. 64-bit architectures Message-ID: <20070308221128.GA24884@bougret.hpl.hp.com> Reply-To: jt@hpl.hp.com References: <1173144447.15891.93.camel@johannes.berg> <20070306171316.GA19669@bougret.hpl.hp.com> <200703061943.07350.mb@bu3sch.de> <20070307020310.GA20466@bougret.hpl.hp.com> <1173364747.14001.7.camel@johannes.berg> <20070308184954.GA24485@bougret.hpl.hp.com> <1173382042.3248.63.camel@johannes.berg> <20070308193412.GG23040@devicescape.com> <1173382802.3248.68.camel@johannes.berg> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <1173382802.3248.68.camel@johannes.berg> From: Jean Tourrilhes Sender: linux-wireless-owner@vger.kernel.org List-ID: On Thu, Mar 08, 2007 at 08:40:01PM +0100, Johannes Berg wrote: > On Thu, 2007-03-08 at 11:34 -0800, Jouni Malinen wrote: > > > Yes, workaround in just iwlib is not enough. If the only possible > > solution is user space workaround, it better be documented (and > > communicated to maintainers of user space apps) well so that > > all user space programs not using iwlib can be modified, too. > > The more I think about it the worse it gets. Think about wireless events > where both 32 and 64-bit userspace programs may be listening... That > means we can't even fix it in the kernel without breaking something. > > johannes This is exactly what I was pointing out earlier. Well, actually, there may be ways of fixing it in the kernel, but that would be real ugly, and I don't want to go there. I've just released wireless_tools.29.pre15.tar.gz. This is supposed to include a "band-aid" for that problem. To the best of my knowledge, it should catch the problem and not introduce false positive. I would be glad if you guys would have a quick look into it, because obviously I can't test it. Now, about the way forward... First possiblity, we could stick with this band-aid permanently. Second possiblity : we do the right thing and plan a API change to return struct always aligned on 32 bits. This way, when we get 128 bit processor, we don't have to add another band aid ;-) It would work like the ESSID changeover. We pick a WE version changeover. We introduce userspace that can deal with before and after. After 1 or 2 years, we flip the switch. After another 1 or 2 years, we get rid of backward compatibility. Third possibility : we declare 32 bit userspace on 64 bit kernel as not supported and advise users to get a 64 bit userspace. The number of bug report on that issue would suggest that very few users are in this case. I know the userspace guys will hate (1) and hate even more (2). Regards, Jean