From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S266297AbUANFJk (ORCPT ); Wed, 14 Jan 2004 00:09:40 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S266301AbUANFJk (ORCPT ); Wed, 14 Jan 2004 00:09:40 -0500 Received: from delerium.codemonkey.org.uk ([81.187.208.145]:12247 "EHLO delerium.codemonkey.org.uk") by vger.kernel.org with ESMTP id S266297AbUANFJj (ORCPT ); Wed, 14 Jan 2004 00:09:39 -0500 Date: Wed, 14 Jan 2004 05:08:18 +0000 From: Dave Jones To: Linux Kernel Mailing List Subject: Re: BIOS Flash changes PowerNOW frequencies? Message-ID: <20040114050818.GC23845@redhat.com> Mail-Followup-To: Dave Jones , Linux Kernel Mailing List References: <20040111175610.GA26855@dotnetslash.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20040111175610.GA26855@dotnetslash.net> User-Agent: Mutt/1.4.1i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Sun, Jan 11, 2004 at 12:56:10PM -0500, Mark W. Alexander wrote: > I'm not currently subscribed. Please cc: me on responses. > > I'm running 2.6.0 on an HP Pavilion ze4420 Athlon version (lspci -v below). I > recently flashed the BIOS (hoping against all odds for suspend to ram > capability) and the CPU frequencies discovered by PowerNOW (K7) has changed. > This is obviously caused by the BIOS update, but the stupid question of the day > is "Why?". If the CPU and chipset support both sets of frequencies with > different BIOS, wouldn't the _real_ set of supported frequencies be the union > of the 2? In reality, yes. However BIOS programmers have a different perception of reality to the rest of us. The spec for PST tables allows for up to 256 FID/VID pairs, yet everyone just seems to offer 5-6 as maximum. I guess they figured no-one needed the granularity of the full range. > As startling as it was to come up at 532Mhz the first boot, I can see where > this could provide some dramatic power savings (say, while using vi), > but the now missing 1064, 1463 and 1596 frequencies were more practical > for actually doing something worthwhile (say, using vi while watching a > DVD ;). > Is there anything I can do to persuade PowerNOW/frequency scaling to see the > full range of frequencies that I've seen this box can do? Something that has been planned for quite a while has been a means of overriding the tables using sysfs. I haven't had time to implement this, and no-one else has found the time/motivation to do so either it seems. Something I was tempted to do at one point (due to the number of broken PST's out there) was to offer a 'ignore_pst' module parameter, which exposed the full table to sysfs. The only problem being some VRMs can't handle certain frequencies at certain voltages whilst some can, making it hard to find a set of 'safe' values for each frequency. How to find out which one VRM can handle frequency X at voltage Y ? Through the PST tables. *sigh*, back to the drawing board. Dave