From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from natsmtp00.rzone.de ([81.169.145.165]:46766 "EHLO natsmtp00.rzone.de") by vger.kernel.org with ESMTP id S264287AbUEDI6T convert rfc822-to-8bit (ORCPT ); Tue, 4 May 2004 04:58:19 -0400 From: Arnd Bergmann Subject: Re: static DEFINE_PER_CPU vs. modules Date: Tue, 4 May 2004 10:56:39 +0200 References: <200405031741.52504.arnd@arndb.de> <200405040024.03906.arnd@arndb.de> <16534.53723.524354.936428@napali.hpl.hp.com> In-Reply-To: <16534.53723.524354.936428@napali.hpl.hp.com> MIME-Version: 1.0 Content-Disposition: inline Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 8BIT Message-Id: <200405041056.43748.arnd@arndb.de> To: davidm@hpl.hp.com Cc: Richard Henderson , Rusty Russell , linux-arch@vger.kernel.org, epasch@de.ibm.com, hare@suse.de List-ID: On Tuesday 04 May 2004 01:12, David Mosberger wrote: > Why not use a GCC attribute?  Perhaps it's a bit of a stretch, but if > you interpret "model(small)" as indicating "small positive/negative > _absolute_ address", you could even use the same attribute name. Yes, that might solve it for gcc-3.5, but not for any older compiler that does not support this, so it's not an option. Besides, it would probably mean less efficient code (three loads to get to a per-cpu variable instead of the current two). Arnd <><