[PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[eric.piel-VkQ1JFuSMpfAbQlEx87xDw]

[-- Attachment #1: Type: text/plain, Size: 481 bytes --]

Hello,

Some processors seems to support only C1 and C3 states, for instance... mine ;-)
(Pentium III Coppermine). The problem is that, in this case, no promotion or
demotion are linking those two states, therefore the preocessor never goes to C3!

The attached patch (against kernel 2.6.1) should do the trick, that's mainly a
copy-paste of the C2/C3 code. It works fine on my computer (and the temperature
really goes down :-) ).

Hopefully you'll like it,
Eric Piel

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: acpi-processor-c1-c3-promotion.patch --]
[-- Type: text/x-patch; name="acpi-processor-c1-c3-promotion.patch", Size: 1174 bytes --]

--- drivers/acpi/processor.c.orig	2004-02-05 19:38:06.000000000 +0100
+++ drivers/acpi/processor.c	2004-02-08 22:46:38.728259984 +0100
@@ -586,6 +586,26 @@
 		pr->power.states[ACPI_STATE_C3].demotion.threshold.bm = 0x0F;
 		pr->power.states[ACPI_STATE_C3].demotion.state = ACPI_STATE_C2;
 	}
+	/*
+	 * C1/C3
+	 * -----
+	 * Set promotion and demotion between C1 and C3 in case there isn't
+	 * any C2 state available
+	 */
+	if ((!pr->power.states[ACPI_STATE_C2].valid) &&
+		(pr->power.states[ACPI_STATE_C3].valid)) {
+		pr->power.states[ACPI_STATE_C1].promotion.threshold.count = 40;
+		pr->power.states[ACPI_STATE_C1].promotion.threshold.ticks =
+			pr->power.states[ACPI_STATE_C3].latency_ticks;
+		pr->power.states[ACPI_STATE_C1].promotion.threshold.bm = 0x0F;
+		pr->power.states[ACPI_STATE_C1].promotion.state = ACPI_STATE_C3;
+
+		pr->power.states[ACPI_STATE_C3].demotion.threshold.count = 1;
+		pr->power.states[ACPI_STATE_C3].demotion.threshold.ticks =
+			pr->power.states[ACPI_STATE_C3].latency_ticks;
+		pr->power.states[ACPI_STATE_C3].demotion.threshold.bm = 0x0F;
+		pr->power.states[ACPI_STATE_C3].demotion.state = ACPI_STATE_C1;
+	}
 
 	return_VALUE(0);
 }

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Ow Mun Heng]

> -----Original Message-----
> From: eric.piel-VkQ1JFuSMpfAbQlEx87xDw@public.gmane.org [mailto:eric.piel@tremplin-utc.net]
> Sent: Monday, February 09, 2004 10:07 PM
> To: len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org
> Cc: acpi-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
> Subject: [ACPI] [PATCH] ACPI/Processor: C1/C3 states promotion and

> Some processors seems to support only C1 and C3 states, for 
> instance... mine ;-)
> (Pentium III Coppermine). The problem is that, in this case, 
> no promotion or
> demotion are linking those two states, therefore the 
> preocessor never goes to C3!

Hmmm so.. your processor only has C1 and C3 and no C2 right??
and you're saying since there's no C2, it will never go to c3


> 
> The attached patch (against kernel 2.6.1) should do the 
> trick, that's mainly a
> copy-paste of the C2/C3 code. It works fine on my computer 
> (and the temperature
> really goes down :-) ).

Any idea what is 'threshold_count' and 'latency_ticks'
and what it does? My Processor Pentium M doen't go to C3 much.

 


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[eric.piel-VkQ1JFuSMpfAbQlEx87xDw]

Quoting Ow Mun Heng <ow.mun.heng-Sjgp3cTcYWE@public.gmane.org>:
> > Some processors seems to support only C1 and C3 states, for 
> > instance... mine ;-)
> > (Pentium III Coppermine). The problem is that, in this case, 
> > no promotion or
> > demotion are linking those two states, therefore the 
> > preocessor never goes to C3!
> 
> Hmmm so.. your processor only has C1 and C3 and no C2 right??
> and you're saying since there's no C2, it will never go to c3
Exactly. This simple logic is coded in processor.c, the patch makes it a bit
more clever...

> Any idea what is 'threshold_count' and 'latency_ticks'
> and what it does? My Processor Pentium M doen't go to C3 much.
Well, not really, I just took (an modified) the code for C1/C2 :-/ Basically,
what I undertood is that 'threshold_count' is the number of time the idle
function should be called before the state changes to the promote one. The
'latency_ticks' is given by the bios and it's the time cost of going to this
more energy saving state.

If you want your processor to go to C3 state you'll need to have your bus master
activity null (you can check it in /proc/acpi/processor/CPU0/power). On my
laptop it requires not playing any sound and not having any usb device on... not
so convenient but I've a script that does all those requirements as soon as I
close the LID.

Eric



-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Ow Mun Heng]

> -----Original Message-----
> From: eric.piel-VkQ1JFuSMpfAbQlEx87xDw@public.gmane.org [mailto:eric.piel@tremplin-utc.net]

> Quoting Ow Mun Heng <ow.mun.heng-Sjgp3cTcYWE@public.gmane.org>:
> > > Some processors seems to support only C1 and C3 states, for 
> > > instance... mine ;-)
> > > (Pentium III Coppermine). The problem is that, in this case, 
> > > no promotion or
> > > demotion are linking those two states, therefore the 
> > > preocessor never goes to C3!
> > 
> > Hmmm so.. your processor only has C1 and C3 and no C2 right??
> > and you're saying since there's no C2, it will never go to c3
> Exactly. This simple logic is coded in processor.c, the patch 
> makes it a bit
> more clever...
> 
> If you want your processor to go to C3 state you'll need to 
> have your bus master
> activity null (you can check it in 
> /proc/acpi/processor/CPU0/power). On my
> laptop it requires not playing any sound and not having any 
> usb device on... not
> so convenient but I've a script that does all those 
> requirements as soon as I
> close the LID.

Can you post me the script? (off list maybe)
I just recompiled my kernel last night (no-usb, not even modules)
and no sound too since I use ALSA and that needs to be compiled 
separately.

I still can't really get into C3 much. [00000090] is all I get.
(I've made a separate post about this earlier)

You mentioned Bus master activity null.. what does that mean??

with usb aware kernel     -> bus master activity = fffffff
with NON usb aware kernel -> bus master activity = aa23as2
					(this value changes very sec!)


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[eric.piel-VkQ1JFuSMpfAbQlEx87xDw]

Quoting Ow Mun Heng <ow.mun.heng-Sjgp3cTcYWE@public.gmane.org>:
> Can you post me the script? (off list maybe)
Well, I don't have my laptop with me and the script is so terrible that I'd be
very shameful. Anyway, the idea is just to put those two lines in a script run
by acpid when the LID event is detected:
xmms -n 0 -u
service hotplug stop

 
> I still can't really get into C3 much. [00000090] is all I get.
> (I've made a separate post about this earlier)
The good news is that your computer can go to C3 , at least :-/

> You mentioned Bus master activity null.. what does that mean??
That means "bus master activity = 000000"

> with usb aware kernel     -> bus master activity = fffffff
> with NON usb aware kernel -> bus master activity = aa23as2
> 					(this value changes very sec!)
So this means there is another thing that is using the bus master. Try to unload
as many things as you can until you see the activity completly flat, that's just
what I did to understand ALSA and USB were the culprits on my laptop.

Eric



-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

Re: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Bruno Ducrot]

On Mon, Feb 09, 2004 at 03:07:19PM +0100, eric.piel-VkQ1JFuSMpfAbQlEx87xDw@public.gmane.org wrote:
> Hello,
> 
> Some processors seems to support only C1 and C3 states, for instance... mine ;-)
> (Pentium III Coppermine). The problem is that, in this case, no promotion or
> demotion are linking those two states, therefore the preocessor never goes to C3!
> 

There is no intel-like processors for which there is C3, but no C2.  So
you have actually to fake a C2 (which should be at a latency of 30 IIRC).


-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Yu, Luming]

> > Some processors seems to support only C1 and C3 states, for 
> instance... mine ;-)
> > (Pentium III Coppermine). The problem is that, in this 
> case, no promotion or
> > demotion are linking those two states, therefore the 
> preocessor never goes to C3!
> > 
> 
> There is no intel-like processors for which there is C3, but 
> no C2.  So
> you have actually to fake a C2 (which should be at a latency 
> of 30 IIRC).
> 

I'm not sure a faking C2 can work as expected, because
C2 power state is an optional ACPI clock state that needs 
chipset hardware support. (ACPI spec 2.0b  8.1.3)

To this issue, I need FADT or PTC object to verify 
processor P_BLK .

--Luming


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[eric.piel-VkQ1JFuSMpfAbQlEx87xDw]

Quoting "Yu, Luming" <luming.yu-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>:
> > There is no intel-like processors for which there is C3, but 
> > no C2.  So
> > you have actually to fake a C2 (which should be at a latency 
> > of 30 IIRC).
> 
> I'm not sure a faking C2 can work as expected, because
> C2 power state is an optional ACPI clock state that needs 
> chipset hardware support. (ACPI spec 2.0b  8.1.3)
> 
> To this issue, I need FADT or PTC object to verify 
> processor P_BLK .

What I already know is that the acpi reports 999 latency for C2 (>100 so it's
not a valid one) and 85 for C3.
Want my FADT? How do I dump it? My DSDT is already dumped but not the FADT...

Eric



-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

Re: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Bruno Ducrot]

On Wed, Feb 11, 2004 at 10:19:22AM +0800, Yu, Luming wrote:
> > > Some processors seems to support only C1 and C3 states, for 
> > instance... mine ;-)
> > > (Pentium III Coppermine). The problem is that, in this 
> > case, no promotion or
> > > demotion are linking those two states, therefore the 
> > preocessor never goes to C3!
> > > 
> > 
> > There is no intel-like processors for which there is C3, but 
> > no C2.  So
> > you have actually to fake a C2 (which should be at a latency 
> > of 30 IIRC).
> > 
> 
> I'm not sure a faking C2 can work as expected, because
> C2 power state is an optional ACPI clock state that needs 
> chipset hardware support. (ACPI spec 2.0b  8.1.3)
> 

We are talking about a PIII Coppermine that need to be put in sleep
mode (C2) before deep sleep (C3).

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Yu, Luming]

> On Wed, Feb 11, 2004 at 10:19:22AM +0800, Yu, Luming wrote:
> > > > Some processors seems to support only C1 and C3 states, for 
> > > instance... mine ;-)
> > > > (Pentium III Coppermine). The problem is that, in this 
> > > case, no promotion or
> > > > demotion are linking those two states, therefore the 
> > > preocessor never goes to C3!
> > > > 
> > > 
> > > There is no intel-like processors for which there is C3, but 
> > > no C2.  So
> > > you have actually to fake a C2 (which should be at a latency 
> > > of 30 IIRC).
> > > 
> > 
> > I'm not sure a faking C2 can work as expected, because
> > C2 power state is an optional ACPI clock state that needs 
> > chipset hardware support. (ACPI spec 2.0b  8.1.3)
> > 
> 
> We are talking about a PIII Coppermine that need to be put in sleep
> mode (C2) before deep sleep (C3).

According to ACPI spec, entering  C2 needs chipset hardware support.
So it is possible that only C1 and C3 get supported, despite CPU 
can support C1 , C2, and C3. 

--Luming


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

Re: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Bruno Ducrot]

On Wed, Feb 11, 2004 at 07:37:16PM +0800, Yu, Luming wrote:
> 
> > On Wed, Feb 11, 2004 at 10:19:22AM +0800, Yu, Luming wrote:
> > > > > Some processors seems to support only C1 and C3 states, for 
> > > > instance... mine ;-)
> > > > > (Pentium III Coppermine). The problem is that, in this 
> > > > case, no promotion or
> > > > > demotion are linking those two states, therefore the 
> > > > preocessor never goes to C3!
> > > > > 
> > > > 
> > > > There is no intel-like processors for which there is C3, but 
> > > > no C2.  So
> > > > you have actually to fake a C2 (which should be at a latency 
> > > > of 30 IIRC).
> > > > 
> > > 
> > > I'm not sure a faking C2 can work as expected, because
> > > C2 power state is an optional ACPI clock state that needs 
> > > chipset hardware support. (ACPI spec 2.0b  8.1.3)
> > > 
> > 
> > We are talking about a PIII Coppermine that need to be put in sleep
> > mode (C2) before deep sleep (C3).
> 
> According to ACPI spec, entering  C2 needs chipset hardware support.
> So it is possible that only C1 and C3 get supported, despite CPU 
> can support C1 , C2, and C3. 
> 

I am not aware of southbridge that are buggy in order to enter a PIII
coppermine in the sleep state whereas it can can put it in deep sleep
state.  Hmm, well, appart perhaps earlyer version of the ich2m ?
I don't remember.  Will check.

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Yu, Luming]

> Want my FADT? How do I dump it? My DSDT is already dumped but 
> not the FADT...

acpidmp


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id\x1356&alloc_id438&op=click

RE: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Yu, Luming]

> > > On Wed, Feb 11, 2004 at 10:19:22AM +0800, Yu, Luming wrote:
> > > > > > Some processors seems to support only C1 and C3 states, for 
> > > > > instance... mine ;-)
> > > > > > (Pentium III Coppermine). The problem is that, in this 
> > > > > case, no promotion or
> > > > > > demotion are linking those two states, therefore the 
> > > > > preocessor never goes to C3!
> > > > > > 
> > > > > 
> > > > > There is no intel-like processors for which there is C3, but 
> > > > > no C2.  So
> > > > > you have actually to fake a C2 (which should be at a latency 
> > > > > of 30 IIRC).
> > > > > 
> > > > 
> > > > I'm not sure a faking C2 can work as expected, because
> > > > C2 power state is an optional ACPI clock state that needs 
> > > > chipset hardware support. (ACPI spec 2.0b  8.1.3)
> > > > 
> > > 
> > > We are talking about a PIII Coppermine that need to be 
> put in sleep
> > > mode (C2) before deep sleep (C3).
> > 
> > According to ACPI spec, entering  C2 needs chipset hardware support.
> > So it is possible that only C1 and C3 get supported, despite CPU 
> > can support C1 , C2, and C3. 
> > 
> 
> I am not aware of southbridge that are buggy in order to enter a PIII
> coppermine in the sleep state whereas it can can put it in deep sleep
> state.  Hmm, well, appart perhaps earlyer version of the ich2m ?
> I don't remember.  Will check.
> 

The most possibility is buggy BIOS (including buggy ACPI table or DSDT
method) 
, or even this is an intended feature of BIOS. So I think facking a C2
is not
a good thing.

--Luming


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id\x1356&alloc_id438&op=click

Re: [PATCH] ACPI/Processor: C1/C3 states promotion and demotion

[Bruno Ducrot]

> The most possibility is buggy BIOS (including buggy ACPI table or DSDT
> method) 

Agreed (but perhaps there is a _CST with appropriate value,
hint feature request http://bugzilla.kernel.org/show_bug.cgi?id=1958).

> , or even this is an intended feature of BIOS. So I think facking a C2
> is not
> a good thing.

Agreed again.  Probably I should write in my first answer a disclaimer
then, sorry.

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click