From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S263654AbTEMWzd (ORCPT ); Tue, 13 May 2003 18:55:33 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S263657AbTEMWzd (ORCPT ); Tue, 13 May 2003 18:55:33 -0400 Received: from deviant.impure.org.uk ([195.82.120.238]:41890 "EHLO deviant.impure.org.uk") by vger.kernel.org with ESMTP id S263654AbTEMWzc (ORCPT ); Tue, 13 May 2003 18:55:32 -0400 Date: Wed, 14 May 2003 00:08:36 +0100 From: Dave Jones To: James Simmons Cc: Linux Kernel Subject: framebuffer initialisation. Message-ID: <20030513230836.GA15158@suse.de> Mail-Followup-To: Dave Jones , James Simmons , Linux Kernel Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.4i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org James, The 2.5 agpgart code broke the i810fb a while back. The reason behind this, is that previously (i.e. 2.4) agp_init() would call the chipset specific backend init routines. In 2.5, that stuff got pushed out into seperate modules, so now i810fb does a (superflous) agp_init() which just printk's the agpgart banner, and then calls the intel_agp_init() function directly. This *should* work, but doesn't. The agpgart code *absolutely must* be set up before the framebuffer code. Previously this was going to be solved with link order shuffling, however that won't work, as the drivers/video stuff is initialised from drivers/char/mem.c (Really non-intuitive place to hide fb init IMO) mem.c gets linked before agp/ so we lose there. Hence the explicit init in i810fb. I've tried to untangle this and figure out why that isn't working, at a complete loss. (The lack of hardware to test this on also doesn't help). Any ideas ? Dave