From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S263242AbUB1Bqf (ORCPT ); Fri, 27 Feb 2004 20:46:35 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262951AbUB1Bqf (ORCPT ); Fri, 27 Feb 2004 20:46:35 -0500 Received: from gate.crashing.org ([63.228.1.57]:14268 "EHLO gate.crashing.org") by vger.kernel.org with ESMTP id S263244AbUB1Bqe (ORCPT ); Fri, 27 Feb 2004 20:46:34 -0500 Subject: Re: Radeon Framebuffer Driver in 2.6.3? From: Benjamin Herrenschmidt To: James Simmons Cc: arief# , Linux Kernel list In-Reply-To: References: Content-Type: text/plain Message-Id: <1077932239.23405.71.camel@gaston> Mime-Version: 1.0 X-Mailer: Ximian Evolution 1.4.5 Date: Sat, 28 Feb 2004 12:37:21 +1100 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org > Rememeber we have to modify every driver then to support FB_ACTIVATE_FORCE. > You have to ask yourself what do you want to do exactly? No we don't. Only fbmem, and that's part of the patch. What we could do is strip the FB_ACTIVATE_FORCE (and actually clear out the activate field completely) when copying to the driver's var structure in fb_set_var(). In fact, we should certainly fix fb_set_var to _ignore_ the activate field when comparing the var structures... this is a bug in the current version imho. It's a bit difficult to fix it while keeping memcmp, except if we do a local copy of the var structure, which would eat stack space... One problem with unblank is that it can be called at interrupt time, but I don't think that ever happens with mode_switch set to true. So we may just kmalloc a local copy of the var... Ben.