From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Michael Frank" Subject: Re: problem with cpu eating too much power, prize given if solved Date: Thu, 01 Jul 2004 08:03:34 +0800 Sender: acpi-devel-admin-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org Message-ID: References: <20040624153513.GB19545@rap.rap.dk> <20040628204008.GK698@openzaurus.ucw.cz> <20040629221111.GB25464@elf.ucw.cz> <20040630173754.GA29587@rap.rap.dk> Mime-Version: 1.0 Content-Type: text/plain; format=flowed; delsp=yes; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: <20040630173754.GA29587-EittdKFJ/bZ/SzgSGea1oA@public.gmane.org> Errors-To: acpi-devel-admin-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , List-Archive: To: =?iso-8859-1?Q?Keld_J=F8rn_Simonsen?= Cc: Pavel Machek , acpi-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org List-Id: linux-acpi@vger.kernel.org On Wed, 30 Jun 2004 19:37:54 +0200, Keld J=F8rn Simonsen = wrote: > On Wed, Jun 30, 2004 at 10:16:31PM +0800, Michael Frank wrote: >> On Wed, 30 Jun 2004 00:11:11 +0200, Pavel Machek wrote: >> >> >Hi! >> > >> >>>>Many times I just want to read the screen, eg emails or documents, >> >>>>and then still the cpu will be in idle mode, where >> >>>>it uses 7.5 W out of the about 15 W that the whole machine uses. >> >>>>I have found out from Intel data sheets that my >> >>>>Mobile Intel Celeron 2.0 GHz CPU consumes 7.5 W in idle loop/ >> >>>>stop grant/sleep states, (and 5.0 W i deep sleep state). >> >>>> >> >>>>So I would like to have the CPU stopped completely (consuming 0 W) >> >>>>while I still could read the screen. In that way I should >> >>>>be able to have about double the lifetime on the battery, >> >>>>going to about 7 hours from the 3.5 hours I have now. >> >>>> >> >>>>I would like the machine to wake up again by any keystroke >> >>>>on the keyboard or mouse input. I would like the input to be >> >>>>reacted upon (not just used to wake up the system, and >> >>>>then forgotten). >> >>>> >> >>> >> >>>S1 is what you want... but... if you get what you asked for, >> >>>do you realize that for example clock would not update on the deskt= op >> >>>in this mode? >> >> >> >>Which hardware can power down the CPU on its own? >> > >> >I thought that CPU powered down is pretty much definition S1. >> > >> >> Please refer to http://acpi.sourceforge.net/documentation/sleep.html >> and ACPI spec 2.0a around Page 216 >> >> "Sleep States >> S1 - "Stopgrant" >> Power to cpu is maintained, but no instructions are executed. >> The CPU halts itself and may shut down many of its internal >> components. In Microsoft Windows, the "Standby" command >> is associated with this state by default. >> " > > Well, I think what I want is S2, from the same document: > > S2 > > While defined in the spec, this state is not currently in use. It > resembles S3 with the qualification that some devices are permitte= d to > remain on. "some devices" could include the CPU - fair enough! > > S3 - "Suspend to RAM" > > All power to the cpu is shut off, and the contents of its register= s are > flushed to RAM, which remains on. In Microsoft Windows, the "Stand= by" > command can be associated with this state if enabled in the BIOS. > Because it requires a high degree of coordination between the cpu, > chipset, devices, OS, BIOS, and OS device drivers, this system sta= te is > the most prone to errors and instability. > Pavel Machek has created a small document with some hints how to s= olve > problems with S3. You can find it in the kernel sources at > Documentation/power/tricks.txt. > S3 is currently _not_ supported by the 2.4.x kernel series in Linu= x. > > > I would like the CPU to power down, save things in memory, and then > still have the screen and associated hardware on, and possibly also the > keyboard and mouse if that is needed to capture an ACPI event, includin= g > the actual keystroke. This seems to be S2. I am not sure if all of S2 > needs to be implemented to do what I asked for, I just would like the > functionality I described. > >> If you look at the original post, S1 "Stopgrant" costs 5W on his celer= on. >> It takes a lot >> more on many 90nm chips. I read in the Centrino datasheet that it can = draw >> 15W >> stopgrant at _maximum_ vcore!!!. The maximum possible power consumpt= ion >> is only 25W!. >> Now, that is life with modern CMOS, leakage is most of the power >> consumption.... >> I think it will improve but 1:3 is the best you can expect for multi G= HZ >> CPU's. > > According to that, then what I ask for could also be beneficial to > centrino CPU's. It would be benefitial to most CPU's also desktops. > >> So again, which (mainboard/notebook) HW does depower the CPU and keeps= the >> rest running ? > > I thought this was something that most mainboards would support, given > that we make the appropiate SW in the kernel. If you want to depower the CPU, you have to depower its bridge connections as well, _No_ Mainboard especially the chipsets can do that. . ... also need a seperate set of main regulators for the CPU. The software side would be mostly CPU part of S3. > > Another thing is that if a mainboard/BIOS supports ACPI, then it should > actually support all of ACPI, including S2 support. Or am I wrong? All optional by the spec. You get S0, and S5 and S4 and S3 when you are lucky. Got a P4 with S1 and S3, but its DSDT is so buggy I run it with ACPI=3Doff and use swsusp 2 on 2.4. > >> As to ACPI there should be a S2.5 state, or a S3 extension to power do= wn >> the CPU and run the rest... > > I thought S2 was actually powering down the cpu, saving all registers > and things in memory, like S3, and then leaving some devices on. As I h= ave read it, > this is clean S2, and not a 2.5 level. But I am not an ACPI expert. It could be, provided someone writes a spec on how to power the CPU down while keeping the rest running and the manufacturers implement it. ACPI for Linux implements the spec and supports features supported by hardware. I suggest you discuss in a PC hardware forum. Best Regards Michael ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com