user undervolting

Message

bdz · Post by **bdz** » Fri Oct 21, 2005 6:51 pm

Well I was about to answer this:

I tried to compile with your .config and that I've had this:
Code: Select all
arch/i386/kernel/built-in.o(.text+0xb597): In function `store_user_voltage':
: undefined reference to `strtoul'
same error as you.
So maybe there is a kernel option that I have in my config and that is not in yours that causes the strtoul code to be linked into the kernel.
So I should add something to my patch to ensure that strtoul is linked whatever the kernel config is.

But then I saw your edit. So it is OK for you now.

But you wrote:

Martini wrote: I have deactivated X86_SPEEDSTEP_CENTRINO_ACPI in my kernel config

and it is already deactivated in your .config file I downloaded from your site:

Code: Select all

$ grep X86_SPEEDSTEP_CENTRINO_ACPI config.2.6.13.martini
# CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI is not set

Did you meant "activated" or did you updated you .config before I downloaded it?

Anyway,
If anybody knows what I should add to have strtoul linked into the kernel please let me know. I'm too n00b in kernel hacking

Edit:

Martini wrote:My speedstep_centrino.c is here

It's not the good file. My patch is not in there

Martini · Post by **Martini** » Fri Oct 21, 2005 7:30 pm

Hi

Yes, the config you have downloaded is the old one from yesterday. Now i have compared your
config with my one in acpi-related things. I've modified it after your download.
Sorry for this mistake.

I'm too n00b in kernel hacking

Thats not true.

I'm impressed!

BTW: the script from the wiki-page works here at the moment, thanks for it.
but i don't understand the "I= ... mv" what shows this exactly?

Thanks for your help
Martin

edit:

Martini hat folgendes geschrieben:
My speedstep_centrino.c is here

It's not the good file. My patch is not in there

Oh yes, this one is without your patch. I've modified by hand for a while with infos from this thread. I'm used it
for testing with voltage combinations.
I'm also sorry for this other mistake.

I need holiday....

bdz · Post by **bdz** » Fri Oct 21, 2005 7:56 pm

Oups!

it not mV but mA. It was too late when I wrote that (something like 1 or 2 AM)
I will correct the wiki page...

Edit:
It's done the wiki page is fixed.
As you now should have guessed it, these mA values are the current drawned from the battery. I use them to estimate how much power is used by the CPU. Well you can not know the absolute value because it is the current drawn by all the components of the PC. But it can give a good relative measure of CPU power usage at different voltage/frequency/load
This of course needs to have A/C disconnected. Or all you will see it the rate at which the battery is charging, or 0 if it is fully charged.

Martini · Post by **Martini** » Fri Oct 21, 2005 8:01 pm

bdz wrote:Oups! it not mV but mA. It was too late when I wrote that (something like 1 or 2 AM)
I will corect the wiki page...

... then i understand this.

thx

bdz · Post by **bdz** » Fri Oct 21, 2005 8:10 pm

Martini wrote:Yes, the config you have downloaded is the old one from yesterday. Now i have compared your
config with my one in acpi-related things. I've modified it after your download.
Sorry for this mistake.

So you mean you now have X86_SPEEDSTEP_CENTRINO_ACPI that is enabled? Or is it also time for me to take some huge hollydays?

Martini · Post by **Martini** » Fri Oct 21, 2005 8:19 pm

So you mean you now have X86_SPEEDSTEP_CENTRINO_ACPI that is enabled?

Yes. Now it is enabled and i can make my kernel.

Code: Select all

cat /sys/devices/system/cpu/cpu0/cpufreq/voltage_table
1084,1084,1084,1084,1084,1084,1084,988,908,860

Code: Select all

2005/10/21 22:19:31  T = 47.0/48.0 °C  Avg(T) = 49.45193 °C    I = 944/944 mV  Avg(I) = 943.635 mV
2005/10/21 22:19:42  T = 47.0/48.0 °C  Avg(T) = 49.43389 °C    I = 943/945 mV  Avg(I) = 943.750 mV
2005/10/21 22:19:52  T = 47.0/48.0 °C  Avg(T) = 49.41799 °C    I = 944/946 mV  Avg(I) = 944.126 mV
2005/10/21 22:20:02  T = 47.0/48.0 °C  Avg(T) = 49.39925 °C    I = 943/944 mV  Avg(I) = 943.933 mV
2005/10/21 22:20:13  T = 47.0/48.0 °C  Avg(T) = 49.38268 °C    I = 943/971 mV  Avg(I) = 945.724 mV

Code: Select all

martin-nb martin # cat /proc/cpuinfo
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 6
model           : 13
model name      : Intel(R) Pentium(R) M processor 1.60GHz
stepping        : 8
cpu MHz         : 798.351
cache size      : 2048 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 pae mce cx8 apic sep mtrr pge mca cmov pat clflush dts acpi mmx fxsr sse sse2 ss tm pbe nx est tm2
bogomips        : 1598.74

edit:
after a while:

but i dont know how stable is ist.

Code: Select all

martin-nb martin # echo "1084,1084,1084,1084,1084,1084,1000,900,850,800" > /sys/devices/system/cpu/cpu0/cpufreq/voltage_table
martin-nb martin # cat /sys/devices/system/cpu/cpu0/cpufreq/voltage_table
1084,1084,1084,1084,1084,1084,988,892,844,796

Code: Select all

2005/10/21 22:41:27  T = 47.0/47.0 °C  Avg(T) = 48.00907 °C    I = 945/954 mV  Avg(I) = 950.538 mV
2005/10/21 22:41:38  T = 47.0/47.0 °C  Avg(T) = 47.99998 °C    I = 945/947 mV  Avg(I) = 948.873 mV
2005/10/21 22:41:48  T = 47.0/47.0 °C  Avg(T) = 47.99098 °C    I = 946/947 mV  Avg(I) = 947.913 mV
2005/10/21 22:41:58  T = 47.0/47.0 °C  Avg(T) = 47.98206 °C    I = 946/946 mV  Avg(I) = 947.203 mV
2005/10/21 22:42:09  T = 47.0/47.0 °C  Avg(T) = 47.97321 °C    I = 944/946 mV  Avg(I) = 946.496 mV
2005/10/21 22:42:19  T = 47.0/47.0 °C  Avg(T) = 47.96444 °C    I = 943/947 mV  Avg(I) = 945.974 mV
2005/10/21 22:42:29  T = 47.0/47.0 °C  Avg(T) = 47.95575 °C    I = 944/947 mV  Avg(I) = 946.024 mV

dgaffuri · Post by **dgaffuri** » Fri Oct 21, 2005 8:55 pm

bdz wrote:

Code: Select all

arch/i386/kernel/built-in.o(.text+0xb597): In function `store_user_voltage':
: undefined reference to `strtoul'

I think the problem is here

Code: Select all

./include/acpi/platform/aclinux.h:#define strtoul simple_strtoul

this is included in speedstep_centrino.c only if X86_SPEEDSTEP_CENTRINO_ACPI is set. You should use

Code: Select all

simple_strtoul

in your patch.

bdz · Post by **bdz** » Fri Oct 21, 2005 9:21 pm

OK, thank you very much! I will try this... during the week-end.

dgaffuri · Post by **dgaffuri** » Fri Oct 21, 2005 9:29 pm

bdz wrote:OK, thank you very much! I will try this... during the week-end.

The weekend is already here

BTW, I'm interested in this thread because I'm successfully using undervolting in Windows with my Toshiba Tecra M2 notebook. I will try you patch as soon as I can (now my root fs is messed up) and I'll post results here.
Thank you for your work.

bdz · Post by **bdz** » Fri Oct 21, 2005 9:34 pm

Martini wrote:
after a while:
but i dont know how stable is ist.
Code: Select all
martin-nb martin # echo "1084,1084,1084,1084,1084,1084,1000,900,850,800" > /sys/devices/system/cpu/cpu0/cpufreq/voltage_table
martin-nb martin # cat /sys/devices/system/cpu/cpu0/cpufreq/voltage_table
1084,1084,1084,1084,1084,1084,988,892,844,796
Code: Select all
2005/10/21 22:41:27  T = 47.0/47.0 °C  Avg(T) = 48.00907 °C    I = 945/954 mV  Avg(I) = 950.538 mV
2005/10/21 22:41:38  T = 47.0/47.0 °C  Avg(T) = 47.99998 °C    I = 945/947 mV  Avg(I) = 948.873 mV
...
Hey! looks like it is working pretty nice!

I forgot to talk about my tests with mprime.
But first I must thank the guy who has mentioned it in the previous page of this thread.

Thank you very much Skystorm!!!

This tool is great to check CPU stabilty. It has detected problems for some settings in a few seconds where cpuburn was running with no error during hours

Martini I strongly recommend you to use it.
My initial settings were these ones:
Code: Select all
924,924,924,924,924,924,924,828,748,700
That is the min voltages I can set before my CPU freeze.
But after checking with mprime I had to increase to this:
Code: Select all
956,956,956,956,956,956,956,860,764,700
Because it was clear that my CPU was computing wrong values at 1.6, 1.3 and 1 GHz.

bdz · Post by **bdz** » Fri Oct 21, 2005 10:10 pm

dgaffuri wrote:
bdz wrote:OK, thank you very much! I will try this... during the week-end.
The weekend is already here

Yeah I know

. But I think I will go to sleep before I can rebuild the kernel, test it during enough time to be confident in the new patch version and update the wiki page

Well, actually I have already replaced strtoul by simple_strtoul and checked that it builds without eror when X86_SPEEDSTEP_CENTRINO_ACPI is not set.
Thank you again for your help

I will update the wiki page and the files on my web site tomorrow or sunday after having performed some regression tests. (You can never trust what you do after mid night

)

Edit:

dgaffuri wrote:BTW, I'm interested in this thread because I'm successfully using undervolting in Windows with my Toshiba Tecra M2 notebook. I will try you patch as soon as I can (now my root fs is messed up) and I'll post results here.

A second beta tester! Cool!

Martini · Post by **Martini** » Sat Oct 22, 2005 8:16 am

Good morning

Bdz, yes.. now this works nice. I know the first 6 values i've used are nonsense.

This was the first fast test before i put myself into deep-sleep-mode. Suspend to bed!

Martini I strongly recommend you to use it.
My initial settings were these ones:
Code:
924,924,924,924,924,924,924,828,748,700

That is the min voltages I can set before my CPU freeze.
But after checking with mprime I had to increase to this:
Code:
956,956,956,956,956,956,956,860,764,700

Because it was clear that my CPU was computing wrong values at 1.6, 1.3 and 1 GHz.

I will testing it over the weekend. But first i will reflash the fan controller with the original firmware.
The book becomes to warm completely without fan under 60C. I have never tested with undervolting before.
I think this works good with undervolting in combination with the original firmware.

Thanks bdz and dgaffuri
Martin

Skystorm · Post by **Skystorm** » Sat Oct 22, 2005 1:24 pm

bdz wrote:But first I must thank the guy who has mentioned it in the previous page of this thread.

Thank you very much Skystorm!!!

You are most welcome!

bdz wrote: This tool is great to check CPU stabilty. It has detected problems for some settings in a few seconds where cpuburn was running with no error during hours

Martini I strongly recommend you to use it.
My initial settings were these ones:
Code: Select all
924,924,924,924,924,924,924,828,748,700
That is the min voltages I can set before my CPU freeze.
But after checking with mprime I had to increase to this:
Code: Select all
956,956,956,956,956,956,956,860,764,700
Because it was clear that my CPU was computing wrong values at 1.6, 1.3 and 1 GHz.

Same for me (only done undervolting under Windows so far), it seemed stable at lower voltages but some minutes of Prime95 (equivalent of mprime) torture mode produced errors, so I had to increase the voltages a little.

Haven't gotten around to try this under Linux, but I hope eventually I will...

bdz · Post by **bdz** » Sun Oct 23, 2005 10:21 pm

After having tested it over the week-end I think the new patch version is ok. I have updated it on the wiki page

bdz · Post by **bdz** » Tue Oct 25, 2005 11:10 pm

I have added the procedure and the script I have used to find the minimum voltages that my CPU can acheive on the wiki page

Martini · Post by **Martini** » Wed Oct 26, 2005 7:07 pm

Hello undervolting heroes

bdz wrote:I have added the procedure and the script I have used to find the minimum voltages that my CPU can acheive on the wiki page

Wow, thats great, bdz. I will test it as soon as possible. I've tested with mprime but only at 800MHz@700mv only.
This works without errors over 2 hours.
Thanx

greets
Martin

edit
just one question:
Since your patch works with X86_SPEEDSTEP_CENTRINO_ACPI (right ?), must i switch this on or off in my kernel.
I'm a little bit confused about this. It works without this.
Thanx

bdz · Post by **bdz** » Wed Oct 26, 2005 7:26 pm

Same thing for me. 800 MHz @ 700 mV -> mprime has run one full night with no error.
However for the other frequencies I had to raise a little bit the voltage from the min.
I will add a second table with my safe voltages on the wiki page when I will have some free time to write the next chapter. (Or maybe someone else wants to write it?)

About X86_SPEEDSTEP_CENTRINO_ACPI: The patch doesn't care if you use the ACPI table or the builtin tables. As long as the speedstep-centrino driver has successfully initialized the frequency table structure it's fine.

Martini · Post by **Martini** » Wed Oct 26, 2005 8:20 pm

About X86_SPEEDSTEP_CENTRINO_ACPI: The patch doesn't care if you use the ACPI table or the builtin tables. As long as the speedstep-centrino driver has successfully initialized the frequency table structure it's fine.

Understand, thx for the enlightenment.

I will add a second table with my safe voltages on the wiki page when I will have some free time to write the next chapter. (Or maybe someone else wants to write it?)

Thats helpfull.
I would help with pleasure, but you point about you speak and i'm only a beta tester.

Sorry

uahh... crappy english. I hope this understand everyone.

Thx
Martin

bdz · Post by **bdz** » Wed Oct 26, 2005 8:48 pm

Martini wrote: I would help with pleasure, but you point about you speak and i'm only a beta tester. Sorry

Not sure I understand that correctly.
But I think I was also not very clear myself:

I would like that the wiki page contain sample voltages of several people, not only me.
And it's a wiki page. So the spirit is: anybody is welcome to modify it and add its contribution.
At the moment it contains only things from me. So the text is like "this is what I have done".
But it would be very nice if other guys would add add some stuff in it and that it becomes something like "this is what we have done" or something like a howto, not only my personal experience.

bdz

Martini · Post by **Martini** » Wed Oct 26, 2005 9:33 pm

Arrghh..

I'm sorry. I totally misunderstood that you have wrote. Now it's clear.
Please forget my previous posting.

Sorry
Martin

bdz · Post by **bdz** » Wed Oct 26, 2005 9:43 pm

No problemo

rschwarze · Post by **rschwarze** » Sun Nov 06, 2005 6:37 pm

Thx for all your work!

I have a few questions:

Does your patch also work for my Pentium M 1.4 Dothan?

The Link for the speedstep-centrino.c on the wiki is probably wrong. I only can get the patch there.

bdz · Post by **bdz** » Sun Nov 06, 2005 8:25 pm

I don't have a Pentium M 1.4 Dothan to test the patch on it but I think it should work. The only way to be sure is that you try it yourself

The link on the wiki page was wrong. I have fixed it.
Thank you for the information.

dgaffuri · Post by **dgaffuri** » Sun Nov 06, 2005 11:32 pm

Fiinally I did it. I've tortured test with mprime my Toshiba Tecra M2 (Pentium-M 750 2 Ghz) for 36 hours at all freqs. I've published safe voltages on the wiki. I tested this same voltages in Windows XP with prime95 in August, and from then I run with this voltages (controlled by CHC) 5 days out of 7.

In my Linux tests I got 10 degrees less at 600 and 2000 Mhz, with fan needed only at 1400 instead of 800. Battery charge after a 50 minutes test is 20% more. And I confirm that the wiki patch applies correctly to a 2.6.14 kernel.

If someone is interested in details I will post them here.

rschwarze · Post by **rschwarze** » Sun Nov 06, 2005 11:36 pm

i dont know what i did wrong.

i used the speedstep-centrino.c from the wiki and recompiled the kernel.

now i have the old (bad) voltages back (so far so good

) but i cant find the /sys/devices/system/cpu/cpu0/cpufreq/voltage_table.

is there anything else i have to change? maybe something in the kernel-config?