Why supports the speedstep-ich only 2 processor states

Why supports the speedstep-ich only 2 processor states

[Thomas Kisler]

Hello everybody,

i'm wondering why the speedstep-ich cpufreq driver only supports two
processor states in general.

I own an IBM Thinkpad X24 with a 1.13 GHz Mobile Pentium 3m (Tualatin)
which runs on 1130 MHz and 733 MHz with cpufreq.

But if I select in BIOS the 'fixed slow' option at the customized Power
Mode _and_ boot a kernel without cpufreq support at all, /proc/cpuinfo
reports me the cpu running at 191 MHz. Also the acpi reports a lower
discharging rate of the battery, as if i run at 733 MHz.

So I had a look at the source code of the speedstep-ich driver and noticed
that there are only a SPEEDSTEP_HIGH and a SPEEDSTEP_LOW frequency in the
speedstep_freqs struct. Can anybody tell me why?

If there is no specific reason, has anybody information about how the
available frequencies are calculated. So i can adapt the source code that
this 3rd power state is available.

Thanks in advance.

Greetings

Thomas Kisler

Re: Why supports the speedstep-ich only 2 processor states

[Hiroshi Miura]

Hi,

In early generation of speedstep, only two processor speeds are supported.
And speedstep-ich is based on reverse engineering basis.
It try to find proper register location and value by watching
difference of chipset registers  between AC powered and battery
powered state.
It might tell us -two- states of speedstep.

You may ask why BIOS can set to another frequency. Answer is that
you can set such a low frequency using speedstep _and_ acpi processor
throtlling feature.

The latter is configurable by setting  /proc/acpi/processor/CPU0/throttling

For example, if your acpi bios support 4 throttling state and set LOW
with  speedstep, then freq = 733 x 25% => 184Mhz.

I don't know how BIOS set chips to run such a state but it may help
you to understand what is happened.

Hiroshi

2006/8/3, Thomas Kisler <kay@rz.fh-augsburg.de>:
> Hello everybody,
>
> i'm wondering why the speedstep-ich cpufreq driver only supports two
> processor states in general.
>
> I own an IBM Thinkpad X24 with a 1.13 GHz Mobile Pentium 3m (Tualatin)
> which runs on 1130 MHz and 733 MHz with cpufreq.
>
> But if I select in BIOS the 'fixed slow' option at the customized Power
> Mode _and_ boot a kernel without cpufreq support at all, /proc/cpuinfo
> reports me the cpu running at 191 MHz. Also the acpi reports a lower
> discharging rate of the battery, as if i run at 733 MHz.
>
> So I had a look at the source code of the speedstep-ich driver and noticed
> that there are only a SPEEDSTEP_HIGH and a SPEEDSTEP_LOW frequency in the
> speedstep_freqs struct. Can anybody tell me why?
>
> If there is no specific reason, has anybody information about how the
> available frequencies are calculated. So i can adapt the source code that
> this 3rd power state is available.
>
> Thanks in advance.
>
> Greetings
>
> Thomas Kisler
>
>


-- 
HIroshi Miura

Re: Why supports the speedstep-ich only 2 processor states

[Erik Slagter]

[-- Attachment #1.1: Type: text/plain, Size: 493 bytes --]

On do, 2006-08-03 at 23:47 +0900, Hiroshi Miura wrote:
> The latter is configurable by setting  /proc/acpi/processor/CPU0/throttling
> 
> For example, if your acpi bios support 4 throttling state and set LOW
> with  speedstep, then freq = 733 x 25% => 184Mhz.
> 
> I don't know how BIOS set chips to run such a state but it may help
> you to understand what is happened.

Hmmm, I don't think that will buy you much. At least in my system
throttling doesn't give _any_ power saving.

[-- Attachment #1.2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 2771 bytes --]

[-- Attachment #2: Type: text/plain, Size: 147 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@lists.linux.org.uk
http://lists.linux.org.uk/mailman/listinfo/cpufreq

Re: Why supports the speedstep-ich only 2 processor states

[Hiroshi Miura]

Hi,

Tomas said 'as if i run at 733 MHz.' and it seems  that
state of speedstep is 733MHz(LOW) and some mechanism like throttling
are working but is not affect any power saving.

I think throttling looks like processor halt state.
CPU just not working some duration till timer intrupt coming or
 predefined ratio.
If cpu load is low, I think throttling may have no difference, because
cpu is halted or throttled in many of cpu time.

2006/8/3, Erik Slagter <erik@slagter.name>:
> On do, 2006-08-03 at 23:47 +0900, Hiroshi Miura wrote:
> > The latter is configurable by setting  /proc/acpi/processor/CPU0/throttling
> >
> > For example, if your acpi bios support 4 throttling state and set LOW
> > with  speedstep, then freq = 733 x 25% => 184Mhz.
> >
> > I don't know how BIOS set chips to run such a state but it may help
> > you to understand what is happened.
>
> Hmmm, I don't think that will buy you much. At least in my system
> throttling doesn't give _any_ power saving.
>

-- 
HIroshi Miura

Re: Why supports the speedstep-ich only 2 processor states

[kay]

Hi,

first, thanks for your answers.

Ok, I tried the throttling. And it's available two for the 191 MHz Mode.
And you are both right, there is no power saving if idle with the
throttling at both Modes.
Power Saving is about 200 mW at 191 MHz compared to 733 MHz. I know it's
not much, but I am not willing to accept that it's not possible :-).
Because it definitly is.

So I will try to adapt the code by hand for the 191 MHz state... and I
hope this will not damage anything.

Strange behavior i noticed:
- without cpufreq support in the kernel the processor has 191 MHz with
the fixed slow option in the BIOS. With cpufreq support, but without any
cpufreq modules loaded, it has 205 MHz reported by /proc/cpuinfo

- both the 205 and the 191 MHz mode are reported with 359 bogomips

- after booting up with cpufreq-support, but without loading the modules
(and reported 205 MHz) I loaded the modules by hand. Now /proc/cpuinfo
reports 1130 MHz at 556 bogomips

- without cpufreq-support and 1130 MHz /proc/cpuinfo reports 2240
bogomips, with cpufreq-support and loading the modules at startup, it
reports 3400 bogomips

Ok this got a little messy but I hope you understand what I wrote.

Greetz

Thomas
-------------------------------------------------------------------------------

On Fri, 2006-08-04 at 00:16 +0900, Hiroshi Miura wrote:
> Hi,
> 
> Tomas said 'as if i run at 733 MHz.' and it seems  that
> state of speedstep is 733MHz(LOW) and some mechanism like throttling
> are working but is not affect any power saving.
> 
> I think throttling looks like processor halt state.
> CPU just not working some duration till timer intrupt coming or
>  predefined ratio.
> If cpu load is low, I think throttling may have no difference, because
> cpu is halted or throttled in many of cpu time.
> 
> 2006/8/3, Erik Slagter <erik@slagter.name>:
> > On do, 2006-08-03 at 23:47 +0900, Hiroshi Miura wrote:
> > > The latter is configurable by setting  /proc/acpi/processor/CPU0/throttling
> > >
> > > For example, if your acpi bios support 4 throttling state and set LOW
> > > with  speedstep, then freq = 733 x 25% => 184Mhz.
> > >
> > > I don't know how BIOS set chips to run such a state but it may help
> > > you to understand what is happened.
> >
> > Hmmm, I don't think that will buy you much. At least in my system
> > throttling doesn't give _any_ power saving.
> >
> 

Re: Why supports the speedstep-ich only 2 processor states

[Bruno Ducrot]

On Thu, Aug 03, 2006 at 08:38:38PM +0100, kay@rz.fh-augsburg.de wrote:
> Hi,
> 
> first, thanks for your answers.
> 
> Ok, I tried the throttling. And it's available two for the 191 MHz Mode.
> And you are both right, there is no power saving if idle with the
> throttling at both Modes.
> Power Saving is about 200 mW at 191 MHz compared to 733 MHz. I know it's
> not much, but I am not willing to accept that it's not possible :-).
> Because it definitly is.
> 

Maybe something is broken with power states of the processor.
What say
cat /proc/acpi/processor/*/power

Cheers,

-- 
Bruno Ducrot

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

Re: Why supports the speedstep-ich only 2 processor states

[Thomas Kisler]

On Fri, 2006-08-04 at 10:54 +0200, Bruno Ducrot wrote:
> On Thu, Aug 03, 2006 at 08:38:38PM +0100, kay@rz.fh-augsburg.de wrote:
> > Hi,
> > 
> > first, thanks for your answers.
> > 
> > Ok, I tried the throttling. And it's available two for the 191 MHz Mode.
> > And you are both right, there is no power saving if idle with the
> > throttling at both Modes.
> > Power Saving is about 200 mW at 191 MHz compared to 733 MHz. I know it's
> > not much, but I am not willing to accept that it's not possible :-).
> > Because it definitly is.
> > 
> 
> Maybe something is broken with power states of the processor.
> What say
> cat /proc/acpi/processor/*/power
> 
> Cheers,
> 
Hi,

/proc/acpi/processor/CPU0/power says (at 191Mhz in Battery Mode):

active state:            C1
max_cstate:              C8
bus master activity:     68118057
states:
   *C1:                  type[C1] promotion[C2] demotion[--]
latency[000] usage[00047422]
    C2:                  type[C2] promotion[C3] demotion[C1]
latency[084] usage[00154405]
    C3:                  type[C3] promotion[--] demotion[C2]
latency[120] usage[00043993]

With the AC-Adapter plugged and running at 1130 MHz the state changes
about every 5 secs. Maybe this is caused by the cpufreqd.

Greetz Thomas Kisler

Re: Why supports the speedstep-ich only 2 processor states

[Hiroshi Miura]

Oh, I'm sorry I mistook the expression.

In early generaion of speedstep technology, a circuit which
changes cpu frequency and voltage is connected to south bridge
GPIO port or specific port.

It may be controlled by single flag(1/0), so it can serve
only two frequency value.

Recent ich driver use a specific register port to control
the circuit.(outb(value, (pmbase + 0x50));)

Hiroshi

2006/8/3, Hiroshi Miura <miura@da-cha.org>:
> And speedstep-ich is based on reverse engineering basis.
> It try to find proper register location and value by watching
> difference of chipset registers  between AC powered and battery
> powered state.
-- 
HIroshi Miura

Re: Why supports the speedstep-ich only 2 processor states

[Bruno Ducrot]

On Fri, Aug 04, 2006 at 12:02:21AM +0900, Hiroshi Miura wrote:
> Oh, I'm sorry I mistook the expression.
> 
> In early generaion of speedstep technology, a circuit which
> changes cpu frequency and voltage is connected to south bridge
> GPIO port or specific port.
> 
> It may be controlled by single flag(1/0), so it can serve
> only two frequency value.
> 
> Recent ich driver use a specific register port to control
> the circuit.(outb(value, (pmbase + 0x50));)

In the latter, only bit 0 will flip the 2 states btw.

-- 
Bruno Ducrot

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