2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Guntsche Michael]

[-- Attachment #1: Type: text/plain, Size: 2008 bytes --]

Hi list,

I recently upgraded one of my servers from 2.6.24.2 to 2.6.25. This is  
more or less a basic fileserver that also does some compilations from  
time to time.

cat /proc/cpuinfo from 2.6.24.2:

processor : 0
vendor_id : GenuineIntel
cpu family  : 15
model   : 1
model name  : Intel(R) Pentium(R) 4 CPU 1.70GHz
stepping  : 2
cpu MHz   : 425.000
cache size  : 256 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 pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pebs bts  
sync_rdtsc
bogomips  : 3401.38
clflush size  : 64

scaling_available_frequencies (p4-clockmod): 425000 637500 850000  
1062500 1275000 1487500 1700000

Since the server is idling most of the time I enabled the ondemand  
governor and installed p4-clockmod. I read that this is not the best  
solution but acpi-cpufreq is not working for me.

Everything was good with 2.6.24. After switching to to 2.6.25 I have  
the following output.

cat /proc/cpuinfo from 2.6.25.4:

processor	: 0
vendor_id	: GenuineIntel
cpu family	: 15
model		: 1
model name	: Intel(R) Pentium(R) 4 CPU 1.70GHz
stepping	: 2
cpu MHz		: 3450.000
cache size	: 256 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 pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pebs bts
bogomips	: 3402.06
clflush size	: 64

scaling_available_frequencies (p4-clockmod): 3450000 5175000 6900000  
8625000 10350000 12075000 13800000

The scaling is working but the frequencies are way of. Every value is  
multiplied by ~8.12.
I get the same result with 2.6.25 and 2.6.25.4.
I am not sure if this is only a display related problem, or if  
something more serious is causing this.

Please CC me on any replies since I am not subscribed to the list.

Kind regards,
Michael








[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 2417 bytes --]

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Arjan van de Ven]

On Fri, 30 May 2008 15:44:55 +0200
Guntsche Michael <mike@it-loops.com> wrote:

> Since the server is idling most of the time I enabled the ondemand  
> governor and installed p4-clockmod. I read that this is not the best  
> solution but acpi-cpufreq is not working for me.

... and it's not the best solution because it's not saving you any
energy...
(in fact it may cost you some)

p4-clockmod should go away.
(at least as cpufreq driver)
-- 
If you want to reach me at my work email, use arjan@linux.intel.com
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Guntsche Michael]

On May 30, 2008, at 15:57, Arjan van de Ven wrote:

> ... and it's not the best solution because it's not saving you any
> energy...
> (in fact it may cost you some)
>

I am most interested in keeping the temperature of my CPU down, which  
means slower fans, which means less noise.
The main "problem" I have is that I do not know if this is a simple  
display issue or if I am having a more fundamental problem here.

/Michael

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Guntsche Michael]

On May 30, 2008, at 16:07, Guntsche Michael wrote:

>
> I am most interested in keeping the temperature of my CPU down,  
> which means slower fans, which means less noise.
> The main "problem" I have is that I do not know if this is a simple  
> display issue or if I am having a more fundamental problem here.

Just FYI I found out the cause of my problem.

Reverting http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=ed9cbcd40004904dbe61ccc16d6106a7de38c998 
  this patch resultsin the correct numbers for me.

<cpuinfo snip>
processor	: 0
vendor_id	: GenuineIntel
cpu family	: 15
model		: 1
model name	: Intel(R) Pentium(R) 4 CPU 1.70GHz
stepping	: 2
cpu MHz		: 637.500
<snip>

This patch was reverted because other people seem to have problems  
with it, reverting the revert does not look like a good idea either.
But since I know now, that this is just a "display" problem and  
everything is working otherwise, I'll just patch this locally for my  
machine here.

Kind regards,
Michael

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Arjan van de Ven]

On Sun, 1 Jun 2008 23:57:14 +0200
Guntsche Michael <mike@it-loops.com> wrote:

> 
> On May 30, 2008, at 16:07, Guntsche Michael wrote:
> 
> >
> > I am most interested in keeping the temperature of my CPU down,  
> > which means slower fans, which means less noise.
> > The main "problem" I have is that I do not know if this is a
> > simple display issue or if I am having a more fundamental problem
> > here.
> 
> Just FYI I found out the cause of my problem.
> 
> Reverting
> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=ed9cbcd40004904dbe61ccc16d6106a7de38c998
> this patch resultsin the correct numbers for me.
> 
> <cpuinfo snip>
> processor	: 0
> vendor_id	: GenuineIntel
> cpu family	: 15
> model		: 1
> model name	: Intel(R) Pentium(R) 4 CPU 1.70GHz
> stepping	: 2
> cpu MHz		: 637.500
> <snip>
> 
> This patch was reverted because other people seem to have problems  
> with it, reverting the revert does not look like a good idea either.
> But since I know now, that this is just a "display" problem and  
> everything is working otherwise, I'll just patch this locally for my  
> machine here.
> 

just as a side note.. you do realize that with p4-clockmod, your cpu is
still running at 1.7 GHz right? (it's just doing less work '-)

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Roger Heflin]

Arjan van de Ven wrote:
> On Sun, 1 Jun 2008 23:57:14 +0200
> Guntsche Michael <mike@it-loops.com> wrote:
> 
>> On May 30, 2008, at 16:07, Guntsche Michael wrote:
>>
>>> I am most interested in keeping the temperature of my CPU down,  
>>> which means slower fans, which means less noise.
>>> The main "problem" I have is that I do not know if this is a
>>> simple display issue or if I am having a more fundamental problem
>>> here.
>> Just FYI I found out the cause of my problem.
>>
>> Reverting
>> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=ed9cbcd40004904dbe61ccc16d6106a7de38c998
>> this patch resultsin the correct numbers for me.
>>
>> <cpuinfo snip>
>> processor	: 0
>> vendor_id	: GenuineIntel
>> cpu family	: 15
>> model		: 1
>> model name	: Intel(R) Pentium(R) 4 CPU 1.70GHz
>> stepping	: 2
>> cpu MHz		: 637.500
>> <snip>
>>
>> This patch was reverted because other people seem to have problems  
>> with it, reverting the revert does not look like a good idea either.
>> But since I know now, that this is just a "display" problem and  
>> everything is working otherwise, I'll just patch this locally for my  
>> machine here.
>>
> 
> just as a side note.. you do realize that with p4-clockmod, your cpu is
> still running at 1.7 GHz right? (it's just doing less work '-)

 From my testing, I believe the only thing that p4-clockmod does is forces an 
idle call when it could otherwise do work on an active process, so fullspeed and 
idle uses *EXACTLY* the same amount of power as p4-clockmod slower speed and 
idle (and therefore generates exactly the same amount of heat), the only power 
difference would be that if you were using p4-clockmod to slow down the cpu when 
it had an active running process (force the cpu to be idle a lot of the time 
even though it has work).   Arjan's point is that if you are using  p4-clockmod 
to slow down an idle cpu in hopes of saving power when the cpu is not being 
used, then it is not going to make *ANY* difference in the power usage at all.

I tried it on my p4 here, and cannot see any power difference in idle/fullspeed 
and idle/slowspeed, this is unlike the later power saving stuff that actually 
does slow down the cpu frequency, and you can measure a different amount of 
power usage with the different clock speeds and an idle cpu.

                                    Roger

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Michael Guntsche]

On Sun, 01 Jun 2008 21:36:58 -0500, Roger Heflin <rogerheflin@gmail.com>
wrote:
> Arjan van de Ven wrote:
>>
>> just as a side note.. you do realize that with p4-clockmod, your cpu is
>> still running at 1.7 GHz right? (it's just doing less work '-)
>  From my testing, I believe the only thing that p4-clockmod does is
forces
> an
> idle call when it could otherwise do work on an active process, so
> fullspeed and
> idle uses *EXACTLY* the same amount of power as p4-clockmod slower speed
<snip>

Thanks for the detailed information everyone. The first time I tested this
I was under the impression that it indeed had an influence on the produced
heat of the CPU. I did a quick test this morning and of course you are
right, there is no change at all. 

This is a good thing though, since it gives me an excuse to upgrade this
machine to a C2D. :)
For the time being I will just disable the cpufreq stuff since it really
does not make a difference in my case.

Kind regards,
Michael

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Henrique de Moraes Holschuh]

On Sun, 01 Jun 2008, Roger Heflin wrote:
> From my testing, I believe the only thing that p4-clockmod does is forces 
> an idle call when it could otherwise do work on an active process, so 
> fullspeed and idle uses *EXACTLY* the same amount of power as p4-clockmod 
> slower speed and idle (and therefore generates exactly the same amount of 
> heat), the only power difference would be that if you were using 
> p4-clockmod to slow down the cpu when it had an active running process 
> (force the cpu to be idle a lot of the time even though it has work).

Indeed.  And if we had a generic interface for throttling (ACPI
T-states-like) and frequency changes (ACPI P-states-like), we could just
move p4-clockmod to the throttling one (and NOT letting it register with
the frequency changes interface), and be done with it.

Do we have these interfaces?

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Matthew Garrett]

On Mon, Jun 02, 2008 at 12:37:59PM -0300, Henrique de Moraes Holschuh wrote:

> Indeed.  And if we had a generic interface for throttling (ACPI
> T-states-like) and frequency changes (ACPI P-states-like), we could just
> move p4-clockmod to the throttling one (and NOT letting it register with
> the frequency changes interface), and be done with it.
> 
> Do we have these interfaces?

No.
-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: 2.6.25.x: Wrong CPU frequency (cpufreq table) with p4-clockmod

[Guntsche Michael]

Ok,

I enabled debugging and here is the output of the boot-up

cpufreq-core: trying to register driver p4-clockmod
cpufreq-core: adding CPU 0
p4-clockmod: has errata -- disabling low frequencies
speedstep-lib: x86: f, model: 1
speedstep-lib: P4 - MSR_EBC_FREQUENCY_ID: 0x8a01fe00 0x0
speedstep-lib: P4 - FSB 100000 kHz; Multiplier 138; Speed 13800000 kHz
freq-table: setting show_table for cpu 0 to e2950d60
freq-table: table entry 0 is invalid, skipping
freq-table: table entry 1 is invalid, skipping
freq-table: table entry 2: 3450000 kHz, 2 index
freq-table: table entry 3: 5175000 kHz, 3 index
freq-table: table entry 4: 6900000 kHz, 4 index
freq-table: table entry 5: 8625000 kHz, 5 index
freq-table: table entry 6: 10350000 kHz, 6 index
freq-table: table entry 7: 12075000 kHz, 7 index
freq-table: table entry 8: 13800000 kHz, 8 index
cpufreq-core: setting new policy for CPU 0: 3450000 - 13800000 kHz
freq-table: request for verification of policy (3450000 - 13800000  
kHz) for cpu 0
freq-table: verification lead to (3450000 - 13800000 kHz) for cpu 0
freq-table: request for verification of policy (3450000 - 13800000  
kHz) for cpu 0
freq-table: verification lead to (3450000 - 13800000 kHz) for cpu 0
cpufreq-core: new min and max freqs are 3450000 - 13800000 kHz
cpufreq-core: governor switch
cpufreq-core: __cpufreq_governor for CPU 0, event 1
cpufreq-core: governor: change or update limits
cpufreq-core: __cpufreq_governor for CPU 0, event 3
cpufreq-core: initialization complete
cpufreq-core: driver p4-clockmod up and running
p4-clockmod: P4/Xeon(TM) CPU On-Demand Clock Modulation available

FSB and the multiplier look a litte bit off to me.

/Michael