cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Matt Sealey]

Is there any support for using the ICTC to reduce CPU power and so on
for PowerPC currently, in absense of a real PMU like on Macs, or the
dual-PLL/DFS/DFS4 stuff in newer G3 and G4 chips?

I was wondering if there were any definitive performance benchmarks
to see if the setting had any appreciable effect in the first place.
If it doesn't do much more than kill a couple of milliwatts and the
switch between ICTC settings has too high a latency, it would make
no sense in a desktop system.

I am basically trying to evaluate if we can do ANYTHING to reduce
power consumption of systems which are idle, as I have noticed that
for running firmware (sitting at a Forth prompt) and booting a
Linux kernel and doing some work (for instance a simple benchmark),
power consumption barely changes at all. The difference between an
idle G4 and a running G4 is negligible. RC5 makes a big difference
but it is heavily tuned. Most people won't run RC5 all the time to
stress the CPU to the level that it wants to draw a couple extra
watts, I think.

Comments? :)

-- 
Matt Sealey <matt@genesi-usa.com>
Manager, Genesi, Developer Relations

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Jon Loeliger]

So, like, the other day "Matt Sealey" mumbled:
> 
> I am basically trying to evaluate if we can do ANYTHING to reduce
> power consumption of systems which are idle, as I have noticed that
> ...
> 
> Comments? :)

Hi Matt,

There is an effort afoot to revitalize some of the Linux Power
Management issues, erm, currently going on over on the linux-pm
list these days.  No immediate results (yet), but there is some
concerted effort.  It should include some PowerPC presence.

Please feel free to contribute, of course. :-)

jdl

RE: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Matt Sealey]

I have nothing real to contribute other than I would like to see it :)

Before I asked I checked Google (as is expected of anyone these days)
and found some discussions on debian-powerpc from 2002 but nobody
really did anything and nothing really came of it. I know 4 years later
all we have is powernowd which pokes up cpufreq which only supports
DFS and certain kinds of Mac PMU. 

ICTC is such a simple thing to support and you can slow down the CPU
pretty comprehensively (from halving to 255x in theory) with an on and
off flag. I am surprised nobody implemented a cpufreq governer even if
it is totally useless and gives no perceivable benefits..

-- 
Matt Sealey <matt@genesi-usa.com>
Manager, Genesi, Developer Relations


> -----Original Message-----
> From: Jon Loeliger [mailto:jdl@jdl.com] 
> Sent: Friday, June 23, 2006 8:34 AM
> To: matt@genesi-usa.com
> Cc: linuxppc-dev@ozlabs.org
> Subject: Re: cpu power "management" for non-dfs chips with no 
> pmu (for instance, 750cxe and mpc7447 in pegasos) 
> 
> So, like, the other day "Matt Sealey" mumbled:
> > 
> > I am basically trying to evaluate if we can do ANYTHING to reduce 
> > power consumption of systems which are idle, as I have noticed that 
> > ...
> > 
> > Comments? :)
> 
> Hi Matt,
> 
> There is an effort afoot to revitalize some of the Linux 
> Power Management issues, erm, currently going on over on the 
> linux-pm list these days.  No immediate results (yet), but 
> there is some concerted effort.  It should include some 
> PowerPC presence.
> 
> Please feel free to contribute, of course. :-)
> 
> jdl
> 

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Jon Loeliger]

So, like, the other day "Matt Sealey" mumbled:
> 
> I have nothing real to contribute other than I would like to see it :)

Disbelief.  :-)

> Before I asked I checked Google (as is expected of anyone these days)
> and found some discussions on debian-powerpc from 2002 but nobody
> really did anything and nothing really came of it. I know 4 years later
> all we have is powernowd which pokes up cpufreq which only supports
> DFS and certain kinds of Mac PMU. 

Which is why it is all being revitalized now... :-)
There was a mini symposium a couple months ago that started
a few working groups to help kick start this back into viability.

> ICTC is such a simple thing to support and you can slow down the CPU
> pretty comprehensively (from halving to 255x in theory) with an on and
> off flag. I am surprised nobody implemented a cpufreq governer even if
> it is totally useless and gives no perceivable benefits..

See?  You _are_ contributing already.  I encourage
you to hit the linux-pm@lists.osdl.org list for a spell!

You could start here:
	https://lists.osdl.org/mailman/listinfo/linux-pm

Thanks,
jdl

RE: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Matt Sealey]

> You could start here:
> 	https://lists.osdl.org/mailman/listinfo/linux-pm

Hmm it says the certificate isn't valid :3

Maybe on Monday..

-- 
Matt Sealey <matt@genesi-usa.com>
Manager, Genesi, Developer Relations

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Raquel Velasco and Bill Buck]

Jon, we appreciate the effort that you, Kate and Becky are making. ;-)
R&B
On Jun 23, 2006, at 9:26 AM, Jon Loeliger wrote:

> So, like, the other day "Matt Sealey" mumbled:
>>
>> I have nothing real to contribute other than I would like to see  
>> it :)
>
> Disbelief.  :-)
>
>> Before I asked I checked Google (as is expected of anyone these days)
>> and found some discussions on debian-powerpc from 2002 but nobody
>> really did anything and nothing really came of it. I know 4 years  
>> later
>> all we have is powernowd which pokes up cpufreq which only supports
>> DFS and certain kinds of Mac PMU.
>
> Which is why it is all being revitalized now... :-)
> There was a mini symposium a couple months ago that started
> a few working groups to help kick start this back into viability.
>
>> ICTC is such a simple thing to support and you can slow down the CPU
>> pretty comprehensively (from halving to 255x in theory) with an on  
>> and
>> off flag. I am surprised nobody implemented a cpufreq governer  
>> even if
>> it is totally useless and gives no perceivable benefits..
>
> See?  You _are_ contributing already.  I encourage
> you to hit the linux-pm@lists.osdl.org list for a spell!
>
> You could start here:
> 	https://lists.osdl.org/mailman/listinfo/linux-pm
>
> Thanks,
> jdl
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Guennadi Liakhovetski]

On Fri, 23 Jun 2006, Jon Loeliger wrote:

> See?  You _are_ contributing already.  I encourage
> you to hit the linux-pm@lists.osdl.org list for a spell!
> 
> You could start here:
> 	https://lists.osdl.org/mailman/listinfo/linux-pm

Sorry, I looked at the list archives and didn't find any search, could 
you, please, post a link to the thread or the subject line?

Thanks
Guennadi
---
Guennadi Liakhovetski

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Jon Loeliger]

So, like, the other day Guennadi Liakhovetski mumbled:
> On Fri, 23 Jun 2006, Jon Loeliger wrote:
> 
> > See?  You _are_ contributing already.  I encourage
> > you to hit the linux-pm@lists.osdl.org list for a spell!
> > 
> > You could start here:
> > 	https://lists.osdl.org/mailman/listinfo/linux-pm
> 
> Sorry, I looked at the list archives and didn't find any search, could 
> you, please, post a link to the thread or the subject line?

Perhaps there was some misunderstanding of what I was
trying to say.

I am NOT saying that they have solved your question with
regard to the 7447 or pagasos directly.  I have no specific
thread to refer to you.

I AM saying that there is renewed interest in the topic of
Linux Power Managament.  Questions, suggestions and interested
parties should gather and discuss things on the linux-pm
list host on osdl.org.  That is likely the best place to ask
this question and get a feel for an answer.

Thanks,
jdl

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Guennadi Liakhovetski]

On Sat, 24 Jun 2006, Jon Loeliger wrote:

> I am NOT saying that they have solved your question with
> regard to the 7447 or pagasos directly.  I have no specific
> thread to refer to you.

Sorry, it wasn't I who originally asked the question, but thanks for the 
answer, anyway.

> I AM saying that there is renewed interest in the topic of
> Linux Power Managament.  Questions, suggestions and interested
> parties should gather and discuss things on the linux-pm
> list host on osdl.org.  That is likely the best place to ask
> this question and get a feel for an answer.

Ok, thanks, I'll subscribe and post my question there too.

Thanks
Guennadi
---
Guennadi Liakhovetski

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Benjamin Herrenschmidt]

> > ICTC is such a simple thing to support and you can slow down the CPU
> > pretty comprehensively (from halving to 255x in theory) with an on and
> > off flag. I am surprised nobody implemented a cpufreq governer even if
> > it is totally useless and gives no perceivable benefits..
> 
> See?  You _are_ contributing already.  I encourage
> you to hit the linux-pm@lists.osdl.org list for a spell!

His question is very powerpc specific... I don't see the point of
bringing ICTC related discussions to linux-pm...

Now, we _did_ some experiemnts in the past and didn't see that much
improvements. That might have changed though, since those 744x/745x CPUs
cannot DOZE and we cannot use their NAP mode neither when idle on
Pegasdos, in which case ICTC might have some use there...

It's easy to tweak it, just test and tell us. I'm not sure if cpufreq is
the right interface though as the cpu frequency isn't actually changing.

Ben.

Re: cpu power "management" for non-dfs chips with no pmu (for instance, 750cxe and mpc7447 in pegasos)

[Benjamin Herrenschmidt]

> I am basically trying to evaluate if we can do ANYTHING to reduce
> power consumption of systems which are idle, as I have noticed that
> for running firmware (sitting at a Forth prompt) and booting a
> Linux kernel and doing some work (for instance a simple benchmark),
> power consumption barely changes at all. The difference between an
> idle G4 and a running G4 is negligible. RC5 makes a big difference
> but it is heavily tuned. Most people won't run RC5 all the time to
> stress the CPU to the level that it wants to draw a couple extra
> watts, I think.

Well, I would expect the Idle G4 to have a fairly good PM in the first
place due to the NAP mode... though we may not enable that on Pegasos.
It depends if the bridge supports the proper protocol to handle
snooping. The Discovery II is bad enough as it is, I wouldn't expect it
to support that in a non-buggy way unfortunately.

Ben.