acpi_processor.c - kernel 2.6.1

acpi_processor.c - kernel 2.6.1

[christian graf]

Hi,

reading the /usr/src/linux-2.6.1/drivers/acpi/processor.c
i thought the kernel should set the state dynamically, based on the
load. Maybe it is possible to set the processor-state of my centrino
manually. Can please anybody tell me how to get it right?

The reason I'm asking is, that once I do load the processor.ko module,
my Centrino switches immediately into C2-state. It never switches back
to C1 state, even under heavy load like compiling the kernel.

I think the kernel should switch back from C2 to C1 state from alone????

Please ask me whatever you need to help me swtiching back into C1 state.

christian



cat /proc/acpi/processor/CPU0/power
active state:            C2
default state:           C1
bus master activity:     ffffffff
states:
    C1:                  promotion[C2] demotion[--] latency[000]
usage[00000010]
   *C2:                  promotion[C3] demotion[C1] latency[001]
usage[00905291]
    C3:                  promotion[--] demotion[C2] latency[185]
usage[00000000]


behn-GKM3wtwTCu0@public.gmane.org:~$ cat /proc/cpuinfo
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 6
model           : 9
model name      : Intel(R) Pentium(R) M processor 1600MHz
stepping        : 5
cpu MHz         : 599.964
cache size      : 1024 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr mce cx8 apic sep mtrr pge mca
cmov pat clflush dts acpi mmx fxsr sse sse2 tm pbe tm2 est
bogomips        : 1185.79






-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[Ducrot Bruno]

On Sun, Jan 18, 2004 at 12:43:13AM +0100, christian graf wrote:
> Hi,
> 
> reading the /usr/src/linux-2.6.1/drivers/acpi/processor.c
> i thought the kernel should set the state dynamically, based on the
> load. Maybe it is possible to set the processor-state of my centrino
> manually. Can please anybody tell me how to get it right?
> 
> The reason I'm asking is, that once I do load the processor.ko module,
> my Centrino switches immediately into C2-state. It never switches back
> to C1 state, even under heavy load like compiling the kernel.
> 
> I think the kernel should switch back from C2 to C1 state from alone????
> 
> Please ask me whatever you need to help me swtiching back into C1 state.
> 
> christian
> 
> 
> 
> cat /proc/acpi/processor/CPU0/power
> active state:            C2
> default state:           C1
> bus master activity:     ffffffff
> states:
>     C1:                  promotion[C2] demotion[--] latency[000]
> usage[00000010]
>    *C2:                  promotion[C3] demotion[C1] latency[001]
> usage[00905291]
>     C3:                  promotion[--] demotion[C2] latency[185]
> usage[00000000]
> 

The latency for C2 is too low (and therefore it is a BIOS trouble).
Anyway, C2 is better than C1, so you want C2 ;)

Note that there is some missing code in the processor.c stuff, which may
give you more appropriate latency, and a C4 state.  The code for is under
developpement.

-- 
Ducrot Bruno

--  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: acpi_processor.c - kernel 2.6.1

[Yu, Luming]

> I think the kernel should switch back from C2 to C1 state 
> from alone????

This issue seems to be the opposite of thread
 "[ACPI] Processor C states and USB problem"

In that case CPU is unable to C3, here CPU is 
unable to C1.

Hmm,...


-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[Yury V. Umanets]

On Sun, 2004-01-18 at 02:43, christian graf wrote:
> Hi,
Hello,
> 
> reading the /usr/src/linux-2.6.1/drivers/acpi/processor.c
> i thought the kernel should set the state dynamically, based on the
> load. Maybe it is possible to set the processor-state of my centrino
> manually. Can please anybody tell me how to get it right?
> 

There is an user-space daemon, which adjusts CPU speed basing on current
loading. And it is able to change CPU frequency if userspace cpufreq
governor is turned on. I would like to see this functionality in kernel
though.

> The reason I'm asking is, that once I do load the processor.ko module,
> my Centrino switches immediately into C2-state. It never switches back
> to C1 state, even under heavy load like compiling the kernel.
> 
I guess, that there is no code in kernel, which does watch current
loading and adjust CPU speed. Also this is job for a cpufreq governor,
not for CPU driver. 

For now, there are three governors:
(1) powersave -- setting CPU frequency statically to lowest available
frequency.
(2) performance -- setting to highest available frequency.
(3) userspace -- used for setting CPU frequency manually.

Probably there also should be number (4), which will set it basing on
current loading.
> I think the kernel should switch back from C2 to C1 state from alone????
> 
> Please ask me whatever you need to help me swtiching back into C1 state.
> 
> christian

-- 
umka



-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[Luca Capello]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

on 01/18/04 12:18, Yury V. Umanets wrote:
> I guess, that there is no code in kernel, which does watch current
> loading and adjust CPU speed. Also this is job for a cpufreq governor,
> not for CPU driver.
>
> For now, there are three governors:
> (1) powersave -- setting CPU frequency statically to lowest available
> frequency.
> (2) performance -- setting to highest available frequency.
> (3) userspace -- used for setting CPU frequency manually.
>
> Probably there also should be number (4), which will set it basing on
> current loading.
some questions and/or suggestions for kernel 2.6.1:

- - 'Power management option (ACPI, APM)' should be changed in 'Power
management option (ACPI, APM, CPUFreq)'

- - why 'Power management option > CPU Frequency scaling > Default CPUFreq
governor' shows only 'performance|userspace' and not also 'powersave'?

- - here my config section for 'Power management option > CPU Frequency
scaling':
> #
> # CPU Frequency scaling
> #
> CONFIG_CPU_FREQ=y
> CONFIG_CPU_FREQ_PROC_INTF=y
> CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y
> # CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE is not set
> CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
> CONFIG_CPU_FREQ_GOV_POWERSAVE=m
> CONFIG_CPU_FREQ_GOV_USERSPACE=m
> # CONFIG_CPU_FREQ_24_API is not set
> CONFIG_CPU_FREQ_TABLE=y

as I didn't know (I had never read docs before and the help from 'make
menuconfig' isn't complete IMHO), I've never load the
'cpufreq_userspace' module, even if I was thinking to use the
'cpufreqd'. If I've correctly understood, 'cpufreqd' need the
'userspace' option/module to work, right? So, why even without the
'userspace' module loaded, is 'cpufreqd' working? I can see it via
/proc/cpuinfo plugging in/out the AC-Adapter.

- - what does loading 'cpufreq_' modules cause? I can see no differences
when I load 'cpufreq_[powersave|userspace]'...

I finished, for now :-)

Thx, bye,
Gismo / Luca
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Debian - http://enigmail.mozdev.org

iD8DBQFACoOAVAp7Xm10JmkRAs+aAJ9w5ZC7Nn5G6K5K5uemEvyb+Q7c9ACfYANJ
c9A6Rz2Sns701M4qsdi77iw=
=qN86
-----END PGP SIGNATURE-----



-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[Mattia Dongili]

[-- Attachment #1: Type: text/plain, Size: 667 bytes --]

On Sun, Jan 18, 2004 at 02:00:49PM +0100, Luca Capello wrote:
> Hello,
[...]
> 'cpufreqd'. If I've correctly understood, 'cpufreqd' need the
> 'userspace' option/module to work, right? So, why even without the

no, it doesn't. Other cpufreq related daemons need it, not cpufreqd.

> 'userspace' module loaded, is 'cpufreqd' working? I can see it via
> /proc/cpuinfo plugging in/out the AC-Adapter.
> 
> - what does loading 'cpufreq_' modules cause? I can see no differences
> when I load 'cpufreq_[powersave|userspace]'...

you only load the code into the kernel, you can then use procfs to select
frequencies ang governor.

ciao
-- 
mattia
:wq!

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: acpi_processor.c - kernel 2.6.1

[christian graf]

Hu Yuri, hi Yu

I have enabled all 3 governors:

cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors
powersave performance userspace


Actual after logging in I run the command:
TARGET=/sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed
F_LOW=powersave
echo $F_LOW > $TARGET

But it does not really matter to which governor I'm switching, my CPU
stays in C2 independant from load and used governor. My notebook in a
Siemens Lifebook E4010D, newest Bios.


But I thought setting a governor is totally independant from setting the
cpu_freq, using speedstep-technology?


Any other ideas for me?

Thanks

christian


Am So, den 18.01.2004 schrieb Yury V. Umanets um 12:18:
> On Sun, 2004-01-18 at 02:43, christian graf wrote:
> > Hi,
> Hello,
> > 
> > reading the /usr/src/linux-2.6.1/drivers/acpi/processor.c
> > i thought the kernel should set the state dynamically, based on the
> > load. Maybe it is possible to set the processor-state of my centrino
> > manually. Can please anybody tell me how to get it right?
> > 
> 
> There is an user-space daemon, which adjusts CPU speed basing on current
> loading. And it is able to change CPU frequency if userspace cpufreq
> governor is turned on. I would like to see this functionality in kernel
> though.
> 
> > The reason I'm asking is, that once I do load the processor.ko module,
> > my Centrino switches immediately into C2-state. It never switches back
> > to C1 state, even under heavy load like compiling the kernel.
> > 
> I guess, that there is no code in kernel, which does watch current
> loading and adjust CPU speed. Also this is job for a cpufreq governor,
> not for CPU driver. 
> 
> For now, there are three governors:
> (1) powersave -- setting CPU frequency statically to lowest available
> frequency.
> (2) performance -- setting to highest available frequency.
> (3) userspace -- used for setting CPU frequency manually.
> 
> Probably there also should be number (4), which will set it basing on
> current loading.
> > I think the kernel should switch back from C2 to C1 state from alone????
> > 
> > Please ask me whatever you need to help me swtiching back into C1 state.
> > 
> > christian



-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[christian graf]

Hi Bruno,

I thought for full performance the C1 state is better as the C2 (which
keeps good performance as well as powersavings)? Or why do you prefer
C2?

How can I avoid the bios - trouble?

christian

 
Am So, den 18.01.2004 schrieb Ducrot Bruno um 03:58:
> On Sun, Jan 18, 2004 at 12:43:13AM +0100, christian graf wrote:
> > Hi,
> > 
> > reading the /usr/src/linux-2.6.1/drivers/acpi/processor.c
> > i thought the kernel should set the state dynamically, based on the
> > load. Maybe it is possible to set the processor-state of my centrino
> > manually. Can please anybody tell me how to get it right?
> > 
> > The reason I'm asking is, that once I do load the processor.ko module,
> > my Centrino switches immediately into C2-state. It never switches back
> > to C1 state, even under heavy load like compiling the kernel.
> > 
> > I think the kernel should switch back from C2 to C1 state from alone????
> > 
> > Please ask me whatever you need to help me swtiching back into C1 state.
> > 
> > christian
> > 
> > 
> > 
> > cat /proc/acpi/processor/CPU0/power
> > active state:            C2
> > default state:           C1
> > bus master activity:     ffffffff
> > states:
> >     C1:                  promotion[C2] demotion[--] latency[000]
> > usage[00000010]
> >    *C2:                  promotion[C3] demotion[C1] latency[001]
> > usage[00905291]
> >     C3:                  promotion[--] demotion[C2] latency[185]
> > usage[00000000]
> > 
> 
> The latency for C2 is too low (and therefore it is a BIOS trouble).
> Anyway, C2 is better than C1, so you want C2 ;)
> 
> Note that there is some missing code in the processor.c stuff, which may
> give you more appropriate latency, and a C4 state.  The code for is under
> developpement.



-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[Luca Capello]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

on 01/18/04 18:10, Mattia Dongili wrote:
> no, it doesn't. Other cpufreq related daemons need it, not cpufreqd.
thx, I didn't know this :-) So, I read the 'man' page and I found that
it uses the ACPI/APM/PMU battery info and the 'cpufreq' driver...

> you only load the code into the kernel, you can then use procfs to select
> frequencies ang governor.
Well, stupid question, but I can't find it: where are the 'governor'
entries?

Ciao (for the Italians), bye (for all the others),
Gismo / Luca
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Debian - http://enigmail.mozdev.org

iD8DBQFACsrCVAp7Xm10JmkRAuRlAJ4ohELCGFFGukKIqY0vAuYPPUmT5gCdFCiw
8+7vNAb27svwZvFMm3390ao=
=+Jn0
-----END PGP SIGNATURE-----



-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[christian graf]

Configure your 2.6.1 kernel 
Power management options (ACPI,..)-> ACPI
	->CPU-Frequency scaling -> and there you see all the "governors".
Include powersave, performance and set the Default CPUFreq governor to
userspace.
DOwn in the make menuconfig screen you have to choose your cpu wich fits
your system !

load the newly added modules and you should find either one of the
following options (depends on your configured modules):
echo 600000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed 
or
echo 600000 > /proc/sys/cpu/0/speed 
or
echo powersave > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor 

the 600000 is just an example. check out your frequency range by:

if you have mounted your sysfs, you should see the dir (and your
frequency-range min, max):
/sys/devices/system/cpu/cpu0/cpufreq$ ls
cpuinfo_max_freq             scaling_driver    scaling_min_freq
cpuinfo_min_freq             scaling_governor  scaling_setspeed
scaling_available_governors  scaling_max_freq


	
hope this helps

christian

Am So, den 18.01.2004 schrieb Luca Capello um 19:04:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hello,
> 
> on 01/18/04 18:10, Mattia Dongili wrote:
> > no, it doesn't. Other cpufreq related daemons need it, not cpufreqd.
> thx, I didn't know this :-) So, I read the 'man' page and I found that
> it uses the ACPI/APM/PMU battery info and the 'cpufreq' driver...
> 
> > you only load the code into the kernel, you can then use procfs to select
> > frequencies ang governor.
> Well, stupid question, but I can't find it: where are the 'governor'
> entries?
> 
> Ciao (for the Italians), bye (for all the others),
> Gismo / Luca
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.2.4 (GNU/Linux)
> Comment: Using GnuPG with Debian - http://enigmail.mozdev.org
> 
> iD8DBQFACsrCVAp7Xm10JmkRAuRlAJ4ohELCGFFGukKIqY0vAuYPPUmT5gCdFCiw
> 8+7vNAb27svwZvFMm3390ao=
> =+Jn0
> -----END PGP SIGNATURE-----
> 
> 
> 
> -------------------------------------------------------
> 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
> _______________________________________________
> Acpi-devel mailing list
> Acpi-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
> https://lists.sourceforge.net/lists/listinfo/acpi-devel
> 



-------------------------------------------------------
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: acpi_processor.c - kernel 2.6.1

[Dominik Brodowski]

[-- Attachment #1: Type: text/plain, Size: 336 bytes --]

On Sun, Jan 18, 2004 at 06:41:51PM +0100, christian graf wrote:
> I thought for full performance the C1 state is better as the C2 (which
> keeps good performance as well as powersavings)? Or why do you prefer
> C2?

C1 and C2 have much less influence to both performance and powersaving than
the various P-States... 

	Dominik

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: acpi_processor.c - kernel 2.6.1

[Ducrot Bruno]

On Sun, Jan 18, 2004 at 02:18:48PM +0300, Yury V. Umanets wrote:
> On Sun, 2004-01-18 at 02:43, christian graf wrote:
> > Hi,
> Hello,
> > 
> > reading the /usr/src/linux-2.6.1/drivers/acpi/processor.c
> > i thought the kernel should set the state dynamically, based on the
> > load. Maybe it is possible to set the processor-state of my centrino
> > manually. Can please anybody tell me how to get it right?
> > 
> 
> There is an user-space daemon, which adjusts CPU speed basing on current
> loading. And it is able to change CPU frequency if userspace cpufreq
> governor is turned on. I would like to see this functionality in kernel
> though.
> 
> > The reason I'm asking is, that once I do load the processor.ko module,
> > my Centrino switches immediately into C2-state. It never switches back
> > to C1 state, even under heavy load like compiling the kernel.
> > 
> I guess, that there is no code in kernel, which does watch current
> loading and adjust CPU speed. Also this is job for a cpufreq governor,
> not for CPU driver. 
> 
> For now, there are three governors:
> (1) powersave -- setting CPU frequency statically to lowest available
> frequency.
> (2) performance -- setting to highest available frequency.
> (3) userspace -- used for setting CPU frequency manually.
> 
> Probably there also should be number (4), which will set it basing on
> current loading.
> > I think the kernel should switch back from C2 to C1 state from alone????
> > 
> > Please ask me whatever you need to help me swtiching back into C1 state.
> > 
> > christian

ACPI Cx states are not the same as ACPI Px states and Cx have nothing to
do with cpufreq.  C-states are the sleeping states for a processor.

ACPI P-state will lower the voltage of the CPU and the frequency, but
the processor is supposed to run, whereas ACPI C-state is a kind of
replacement of the asm 'hlt' (for a 386 compatible), mostly used in
the idle loop.

C1 is actually this asm 'hlt', whereas C2 require a read to a IO register,
and C3 require a read to a IO register plus a kind of control of the bus 
(on UP system).  More the x in Cx is high, more you get power saving,
but this have a price for the latency in order to wakeup the processor
when it have to serve an interrupt.

-- 
Ducrot Bruno

--  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