CPU - Power state (Cx) usage count not working

CPU - Power state (Cx) usage count not working

[Gerfried Maier]

Hi everybody!

According to the ACPI4Linux doc

cat /proc/acpi/processor/CPU0/power

should look like

active state:            C2
default state:           C1
bus master activity:     00004f00
states:
    C1:                  promotion[C2] demotion[--] latency[000]
usage[00033670]
   *C2:                  promotion[C3] demotion[C1] latency[084]
usage[00357060]
    C3:                  promotion[--] demotion[C2] latency[120]
usage[00000000]

Unfortunaltely on my system (P4m 1800Mhz; 2.4.31 or .32) the usage count
shows always 00000000. Unloading several modules (eg usb-ohci) has no
effect.

IIrc the count and the statistics were working in earlier times (arond
2.4.25+) All of them vanilla with the preemptive patch from
http://www.kernel.org/pub/linux/kernel/people/rml/preempt-kernel/v2.4/

Is there anything I can do to figure out more on the reason?
Of course I would like to know whether this is only a symptom for a
minor proble or wheher the whole power-state thing is not working,
resulting in shorter battery life.

Thanks for your help!

G. Maier

Re: CPU - Power state (Cx) usage count not working

[Niko Ehrenfeuchter]

Hi Gerfried,

On Mon, 2006-01-23 at 09:19 +0100, Gerfried Maier wrote:
> Hi everybody!
> 
> According to the ACPI4Linux doc
> 
> cat /proc/acpi/processor/CPU0/power
> 
> should look like
> 
> active state:            C2
> default state:           C1
> bus master activity:     00004f00
> states:
>     C1:                  promotion[C2] demotion[--] latency[000]
> usage[00033670]
>    *C2:                  promotion[C3] demotion[C1] latency[084]
> usage[00357060]
>     C3:                  promotion[--] demotion[C2] latency[120]
> usage[00000000]
> 
> Unfortunaltely on my system (P4m 1800Mhz; 2.4.31 or .32) the usage count
> shows always 00000000. Unloading several modules (eg usb-ohci) has no
> effect.

the counter normally shows the de facto usage, so I assume your system
never uses C3 in the above situation. Unloading the usb-modules on a 2.4
kernel _might_ help, but it also might not suffice. If you have anything
that causes busmaster activity, your system will never use C3. This
could be a module (as said above, the usb-modules on 2.4 do so - that's
fixed in the 2.6 series) as well as some process, e.g. sound output or
some monitoring stuff (I remember gkrellm  once preventing my system to
enter C3). To check this, you could try to boot into single-user mode,
unload all unnecessary modules and then recheck for C3 usage.

Some time ago, I wrote a couple of lines considering the c-states with
kernel 2.4 on the acpi-wiki. You can have a look there:

http://acpi.sourceforge.net/wiki/index.php/WhyMyCxPowerStateIsNotUsed

hth,
Niko

Re: CPU - Power state (Cx) usage count not working

[Gerfried Maier]

Hi Niko!

Thanks for your quick reply.

Niko Ehrenfeuchter wrote:
> [...]
>>active state:            C2
>>default state:           C1
>>bus master activity:     00004f00
>>states:
>>    C1:                  promotion[C2] demotion[--] latency[000]
>>usage[00033670]
>>   *C2:                  promotion[C3] demotion[C1] latency[084]
>>usage[00357060]
>>    C3:                  promotion[--] demotion[C2] latency[120]
>>usage[00000000]
>> [...]


> the counter normally shows the de facto usage, so I assume your system
> never uses C3 in the above situation. Unloading the usb-modules on a 2.4
> kernel _might_ help, but it also might not suffice. 

The above example was not a capture from my system (which i
unfortunately do not have available by now) but the example given in the
wiki-entry you referred below.

On my system all _3 counters_ C1, C2 and C3 always stay at [00000000]
and not only the C3 counter.

Having read my original posting again, I'm afraid that this ambiguity
was caused by me with my unclear formulation of the OP. Sorry for this.

> [...]
> 
> http://acpi.sourceforge.net/wiki/index.php/WhyMyCxPowerStateIsNotUsed

moali

Re: CPU - Power state (Cx) usage count not working

[Niko Ehrenfeuchter]

On Mon, 2006-01-23 at 12:39 +0100, Gerfried Maier wrote:
> Hi Niko!
> 
> Thanks for your quick reply.

[...]

> > the counter normally shows the de facto usage, so I assume your system
> > never uses C3 in the above situation. Unloading the usb-modules on a 2.4
> > kernel _might_ help, but it also might not suffice. 
> 
> The above example was not a capture from my system (which i
> unfortunately do not have available by now) but the example given in the
> wiki-entry you referred below.

:)

> On my system all _3 counters_ C1, C2 and C3 always stay at [00000000]
> and not only the C3 counter.

ok, now that's much clearer...  can you revert to the old kernel that
made use of the power states to check if it still works?

as far as i can see, the latest acpi patch for 2.4 is really old
(20040326), so i'd really recommend to try 2.6

regards,
niko

Re: CPU - Power state (Cx) usage count not working

[Gerfried Maier]

Niko Ehrenfeuchter wrote:
> 
>> On my system all _3 counters_ C1, C2 and C3 always stay at [00000000]
>> and not only the C3 counter.
>
> ok, now that's much clearer...  can you revert to the old kernel that
> made use of the power states to check if it still works?

I figured out, that the preemptive-patch from
http://www.kernel.org/pub/linux/kernel/people/rml/preempt-kernel/v2.4/
is responsible for all the couters staying at [00000000]. With this
patch not applied 2.4.31 works as expected. So, this was my fault at
all. Sorry for bothering the list.

> as far as i can see, the latest acpi patch for 2.4 is really old
> (20040326), so i'd really recommend to try 2.6

I'm afraid that I can't use 2.6 because of annoyances with the nvidia
X11-drivers. So I'm stuck to 2.4.

moali