From mboxrd@z Thu Jan 1 00:00:00 1970 From: Len Brown Subject: Re: Main differences between drivers Date: Tue, 15 Aug 2006 10:56:45 -0400 Message-ID: <200608151056.45913.len.brown@intel.com> References: <59a2a3420608150722i15be71d4wfa014e18c38dbba5@mail.gmail.com> Reply-To: Len Brown Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <59a2a3420608150722i15be71d4wfa014e18c38dbba5@mail.gmail.com> Content-Disposition: inline List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: cpufreq-bounces@lists.linux.org.uk Errors-To: cpufreq-bounces+glkc-cpufreq=m.gmane.org+glkc-cpufreq=m.gmane.org@lists.linux.org.uk Content-Type: text/plain; charset="us-ascii" To: cpufreq@lists.linux.org.uk On Tuesday 15 August 2006 10:22, Guilherme Moro wrote: > Hello all, > > Can someone explain, or send a link or a document that explain the > real difference between the different drivers for Intel processors ? > Why the same processor can use speedstep-ich, speedstep-centrino and > acpi-cpufreq ? > the basic differences I can understand, like acpi-cpufreq use of acpi > tables, but what driver is the "better" or more "official" for Intel > platforms? good question -- the names are quite poor, and unfortunately it is a pain to re-name drivers. As of 2.6.17, both acpi-cpufreq and speedstep-centrino use ACPI tables, but acpi-cpufreq uses IO accesses and speedstep centrino uses MSR accesses. Since MSR accesses are faster, it is preferred to try speedstep-centrino first. speedstep centrino also has reverse engineered tables to fall back on if ACPI tables fail. I haven't seen this happen in practice in recent history. The hard-coded tables were there from before Intel published the PDC bits that allow ACPI tables to work with MSR accesses. Going forward, there has been a set of patches on the list to: 1. add MSR access to acpi-cpufreq 2. remove ACPI tables from speedstep-centrino. That would mean that acpi-cpufreq can handle any ACPI situation -- IO or MSR; and that speedstep-centrino would be only for the reverse-engineered case. Thus the preference would change to be acpi-cpufreq first, speedstep-centrino second -- though the reverse preference would probably be okay too. -Len ps. don't bother with speedstep-ich if the above work, and never use p4-clockmod.