PowerOP vs OPpoint

PowerOP vs OPpoint

[Matthew Locke]

Unfortunately, there are two efforts underway that makes this confusing 
and I think require a bit more than the short summary requested.  A one 
paragraph summary can't address the why and how.  This email briefly 
describes the why and the differences.

There are two main reasons for both these efforts:
- existing power management interfaces do not enable the power 
management features on the latest SOC's used in embedded mobile  
devices
- existing power management interfaces do not provide the API necessary 
to build power managers (userspace and/or kernel space) that optimize 
power consumption to level required by embedded mobile devices

PowerOP
Focus is on a platform independent interface for selecting and creating 
operating points.  We want to get the basic power management block in 
place and build on it.  Integrating with other existing power 
management interfaces as it makes sense.  The first natural integration 
point is the cpufreq_driver layer in cpufreq and does not affect the 
userspace interface.

OPpoint
Goal is to show how all existing interfaces can use the operating point 
concept.  It is more than an interface for selecting and creating 
operating points.  It integrates with cpufreq and sleep states defining 
new userspace interfaces and using existing interfaces in different 
ways.  There are a lot of issues with the OPpoint operating point 
interface that was discussed here: 
http://lists.osdl.org/pipermail/linux-pm/2006-August/003541.html .  
Many objections to the sleep state integration.  Most of the negative 
comments about cpufreq are about the OPpoint patches.

I have not seen or heard any justification for the OPpoint patches to 
create a different operating point interface.

Matt

Re: [linux-pm] PowerOP vs OPpoint

[Jon Loeliger]

On Thu, 2006-09-14 at 04:22, Matthew Locke wrote:
> Unfortunately, there are two efforts underway that makes this confusing 
> and I think require a bit more than the short summary requested.  A one 
> paragraph summary can't address the why and how.  This email briefly 
> describes the why and the differences.
> 
> There are two main reasons for both these efforts:
> - existing power management interfaces do not enable the power 
> management features on the latest SOC's used in embedded mobile  
> devices
> - existing power management interfaces do not provide the API necessary 
> to build power managers (userspace and/or kernel space) that optimize 
> power consumption to level required by embedded mobile devices

So does it make sense to re-unify these two patch-sets
into one common, more general patch-set first?  Might
it make sense to do so in small, incremental steps that
everyone can agree on as we go along?

For example, maybe the very first thing to do is define
some notion of general "operating point" that is a super-set
of the cpufreq definition.   If we can define that structure
maybe we can progress towards introducing and using it.

Totally side-step the kernel-user level stuff for a bit...
Totally side-step the suspend/resume issues for a bit...

Thanks,
jdl

Re: PowerOP vs OPpoint

[Amit Kucheria]

On Mon, 2006-09-18 at 15:07 -0500, ext Jon Loeliger wrote:
> On Thu, 2006-09-14 at 04:22, Matthew Locke wrote:
> > Unfortunately, there are two efforts underway that makes this confusing 
> > and I think require a bit more than the short summary requested.  A one 
> > paragraph summary can't address the why and how.  This email briefly 
> > describes the why and the differences.
> > 
> > There are two main reasons for both these efforts:
> > - existing power management interfaces do not enable the power 
> > management features on the latest SOC's used in embedded mobile  
> > devices
> > - existing power management interfaces do not provide the API necessary 
> > to build power managers (userspace and/or kernel space) that optimize 
> > power consumption to level required by embedded mobile devices
> 
> So does it make sense to re-unify these two patch-sets
> into one common, more general patch-set first?  Might
> it make sense to do so in small, incremental steps that
> everyone can agree on as we go along?

That has been the idea. Maybe if you have better way to 'communicate'
with David Singleton (oppoint) since he refuses to be drawn into the
merge discussions.

> For example, maybe the very first thing to do is define
> some notion of general "operating point" that is a super-set
> of the cpufreq definition.   If we can define that structure
> maybe we can progress towards introducing and using it.

Yes, it is a good first step. Please comment on the PowerOP patches to
see if there is something in the OP notion that is missing in your
opinion.

> Totally side-step the kernel-user level stuff for a bit...
> Totally side-step the suspend/resume issues for a bit...

I believe the PowerOP patches from Eugeny and Matt do not touch
suspend-resume issues and make the kernel-userspace interface to define
OP as an optional patch.

Oppoing patches on the other hand touch those issues.

Regards,
Amit

-- 
Amit Kucheria <amit.kucheria@nokia.com>
Nokia