[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Lukasz Majewski]

Dear all,

> Despite this patch set is working and applicable on top of 3.14-rc5, 
> please regard it solely as a pure RFC.
> 
> This patch provides support for LAB governor build on top of ondemand.
> Previous version of LAB can be found here:
> http://thread.gmane.org/gmane.linux.kernel/1484746/match=cpufreq
> 
> LAB short reminder:
> 
> LAB uses information about how many cores are in "idle" state (the
> core idleness is represented as the value between 0 and 100) and the
> overall load of the system (from 0 to 100) to decide about frequency
> to be set. It is extremely useful with SoCs like Exynos4412, which
> can set only one frequency for all cores.
> 
> Important design decisions:
> 
> - Reuse well established ondemand governor's internal code. To do this
>   I had to expose some previously static internal ondemand code. 
>   This allowed smaller LAB code when compared to previous version.
> 	
> - LAB works on top of ondemand, which means that one via device tree
>   attributes can specify if and when e.g. BOOST shall be enabled or
> if any particular frequency shall be imposed. For situation NOT
> important from the power consumption reduction viewpoint the ondemand
> is used to set proper frequency.
> 
> - It is only possible to either compile in or not the LAB into the
> kernel. There is no "M" option for Kconfig. It is done on purpose,
> since ondemand itself can be also compiled as a module and then it
> would be possible to remove ondemand when LAB is working on top of it.
> 
> - The LAB operation is specified (and thereof extendable) via device
> tree lab-ctrl-freq attribute defined at /cpus/cpu0.
> 
> 
> Problems:
> - How the governor will work for big.LITTLE systems (especially
> Global Task Scheduling).
> - Will there be agreement to expose internal ondemand code to be
> reused for more specialized governors.
> 
> Test HW:
> 	Exynos4412 - Trats2 board.
> Above patches were posted on top of Linux 3.14-rc5
> (SHA1: 3f9590c281c66162bf8ae9b7b2d987f0a89043c6)
> 

Any comments about those patches?

> Lukasz Majewski (5):
>   cpufreq:LAB:ondemand Adjust ondemand to be able to reuse its methods
>   cpufreq:LAB:cpufreq_governor Adjust cpufreq_governor.[h|c] to
> support LAB
>   cpufreq:LAB:lab Add LAB governor code
>   cpufreq:LAB:Kconfig Add LAB definitions to Kconfig
>   cpufreq:LAB:dts:trats2: Add DTS nodes for LAB governor
> 
>  arch/arm/boot/dts/exynos4412-trats2.dts |   29 ++
>  drivers/cpufreq/Kconfig                 |   28 ++
>  drivers/cpufreq/Makefile                |    1 +
>  drivers/cpufreq/cpufreq_governor.c      |    7 +
>  drivers/cpufreq/cpufreq_governor.h      |   12 +
>  drivers/cpufreq/cpufreq_lab.c           |  457
> +++++++++++++++++++++++++++++++
> drivers/cpufreq/cpufreq_ondemand.c      |   24 +- 7 files changed,
> 550 insertions(+), 8 deletions(-) create mode 100644
> drivers/cpufreq/cpufreq_lab.c
> 



-- 
Best regards,

Lukasz Majewski

Samsung R&D Institute Poland (SRPOL) | Linux Platform Group

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Viresh Kumar]

On 17 March 2014 21:08, Lukasz Majewski <l.majewski@samsung.com> wrote:
>> Despite this patch set is working and applicable on top of 3.14-rc5,
>> please regard it solely as a pure RFC.
>>
>> This patch provides support for LAB governor build on top of ondemand.
>> Previous version of LAB can be found here:
>> http://thread.gmane.org/gmane.linux.kernel/1484746/match=cpufreq
>>
>> LAB short reminder:
>>
>> LAB uses information about how many cores are in "idle" state (the
>> core idleness is represented as the value between 0 and 100) and the
>> overall load of the system (from 0 to 100) to decide about frequency
>> to be set. It is extremely useful with SoCs like Exynos4412, which
>> can set only one frequency for all cores.
>>
>> Important design decisions:
>>
>> - Reuse well established ondemand governor's internal code. To do this
>>   I had to expose some previously static internal ondemand code.
>>   This allowed smaller LAB code when compared to previous version.
>>
>> - LAB works on top of ondemand, which means that one via device tree
>>   attributes can specify if and when e.g. BOOST shall be enabled or
>> if any particular frequency shall be imposed. For situation NOT
>> important from the power consumption reduction viewpoint the ondemand
>> is used to set proper frequency.
>>
>> - It is only possible to either compile in or not the LAB into the
>> kernel. There is no "M" option for Kconfig. It is done on purpose,
>> since ondemand itself can be also compiled as a module and then it
>> would be possible to remove ondemand when LAB is working on top of it.
>>
>> - The LAB operation is specified (and thereof extendable) via device
>> tree lab-ctrl-freq attribute defined at /cpus/cpu0.
>>
>>
>> Problems:
>> - How the governor will work for big.LITTLE systems (especially
>> Global Task Scheduling).
>> - Will there be agreement to expose internal ondemand code to be
>> reused for more specialized governors.
>>
>> Test HW:
>>       Exynos4412 - Trats2 board.
>> Above patches were posted on top of Linux 3.14-rc5
>> (SHA1: 3f9590c281c66162bf8ae9b7b2d987f0a89043c6)
>>
>
> Any comments about those patches?

Sorry for being late on reviewing these..

I tried to go through the patches but didn't looked at the minutest
of the details. Its been a long time when you first sent this patchset.
And the memories have corrupted by now :) ..

To get context back, can we discuss again the fundamentals behind
this new governor you are proposing. And then we can discuss about
it again, its pros/cons, etc..

I tried to go to earlier threads but I think we better do it again..

People are reluctant in getting another governor in and want to give
existing governors a try if possible.

So, please explain the basics behind your governor again and then
we can put our arguments again..

--
viresh

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Lukasz Majewski]

Hi Viresh,

> On 17 March 2014 21:08, Lukasz Majewski <l.majewski@samsung.com>
> wrote:
> >> Despite this patch set is working and applicable on top of
> >> 3.14-rc5, please regard it solely as a pure RFC.
> >>
> >> This patch provides support for LAB governor build on top of
> >> ondemand. Previous version of LAB can be found here:
> >> http://thread.gmane.org/gmane.linux.kernel/1484746/match=cpufreq
> >>
> >> LAB short reminder:
> >>
> >> LAB uses information about how many cores are in "idle" state (the
> >> core idleness is represented as the value between 0 and 100) and
> >> the overall load of the system (from 0 to 100) to decide about
> >> frequency to be set. It is extremely useful with SoCs like
> >> Exynos4412, which can set only one frequency for all cores.
> >>
> >> Important design decisions:
> >>
> >> - Reuse well established ondemand governor's internal code. To do
> >> this I had to expose some previously static internal ondemand code.
> >>   This allowed smaller LAB code when compared to previous version.
> >>
> >> - LAB works on top of ondemand, which means that one via device
> >> tree attributes can specify if and when e.g. BOOST shall be
> >> enabled or if any particular frequency shall be imposed. For
> >> situation NOT important from the power consumption reduction
> >> viewpoint the ondemand is used to set proper frequency.
> >>
> >> - It is only possible to either compile in or not the LAB into the
> >> kernel. There is no "M" option for Kconfig. It is done on purpose,
> >> since ondemand itself can be also compiled as a module and then it
> >> would be possible to remove ondemand when LAB is working on top of
> >> it.
> >>
> >> - The LAB operation is specified (and thereof extendable) via
> >> device tree lab-ctrl-freq attribute defined at /cpus/cpu0.
> >>
> >>
> >> Problems:
> >> - How the governor will work for big.LITTLE systems (especially
> >> Global Task Scheduling).
> >> - Will there be agreement to expose internal ondemand code to be
> >> reused for more specialized governors.
> >>
> >> Test HW:
> >>       Exynos4412 - Trats2 board.
> >> Above patches were posted on top of Linux 3.14-rc5
> >> (SHA1: 3f9590c281c66162bf8ae9b7b2d987f0a89043c6)
> >>
> >
> > Any comments about those patches?
> 
> Sorry for being late on reviewing these..
> 
> I tried to go through the patches but didn't looked at the minutest
> of the details. Its been a long time when you first sent this
> patchset. And the memories have corrupted by now :) ..

Unfortunately memory is volatile ... since LAB governor is a follow up
of BOOST, which review and inclusion took considerable time, some
details could be forgotten. 

> 
> To get context back, can we discuss again the fundamentals behind
> this new governor you are proposing. And then we can discuss about
> it again, its pros/cons, etc..

Please consider following links:

The original implementation - threads:
http://thread.gmane.org/gmane.linux.power-management.general/32523/match=lab
http://thread.gmane.org/gmane.linux.kernel/1484746/match=lab


LAB justification data:
http://article.gmane.org/gmane.linux.kernel/1472381


> People are reluctant in getting another governor in and want to give
> existing governors a try if possible.

As I've stated in the covering letter, this code is an extension of
Ondemand.

This is totally different from what have been posted previously (v1,
v2).
The first LAB proposal was written with some parts copied from Ondemand.
It was a separate, standalone governor.


The approach proposed in those patches is very different. It simply
reuses Ondemand code as much as possible (timers, default attributes
exported to sysfs, etc.).

On top of the Ondemand we have the LAB, which thereof is its optional
extension. The existing code is reused and can be easily extracted as a
common code.

> 
> So, please explain the basics behind your governor again and then
> we can put our arguments again..
> 

I hope that provided overview is sufficient. More in depth
information can be found in posted patches or provided LKML archives.

> --
> viresh

-- 
Best regards,

Lukasz Majewski

Samsung R&D Institute Poland (SRPOL) | Linux Platform Group

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Lukasz Majewski]

Hi Viresh,

> Hi Viresh,
> 
> > On 17 March 2014 21:08, Lukasz Majewski <l.majewski@samsung.com>
> > wrote:
> > >> Despite this patch set is working and applicable on top of
> > >> 3.14-rc5, please regard it solely as a pure RFC.
> > >>
> > >> This patch provides support for LAB governor build on top of
> > >> ondemand. Previous version of LAB can be found here:
> > >> http://thread.gmane.org/gmane.linux.kernel/1484746/match=cpufreq
> > >>
> > >> LAB short reminder:
> > >>
> > >> LAB uses information about how many cores are in "idle" state
> > >> (the core idleness is represented as the value between 0 and
> > >> 100) and the overall load of the system (from 0 to 100) to
> > >> decide about frequency to be set. It is extremely useful with
> > >> SoCs like Exynos4412, which can set only one frequency for all
> > >> cores.
> > >>
> > >> Important design decisions:
> > >>
> > >> - Reuse well established ondemand governor's internal code. To do
> > >> this I had to expose some previously static internal ondemand
> > >> code. This allowed smaller LAB code when compared to previous
> > >> version.
> > >>
> > >> - LAB works on top of ondemand, which means that one via device
> > >> tree attributes can specify if and when e.g. BOOST shall be
> > >> enabled or if any particular frequency shall be imposed. For
> > >> situation NOT important from the power consumption reduction
> > >> viewpoint the ondemand is used to set proper frequency.
> > >>
> > >> - It is only possible to either compile in or not the LAB into
> > >> the kernel. There is no "M" option for Kconfig. It is done on
> > >> purpose, since ondemand itself can be also compiled as a module
> > >> and then it would be possible to remove ondemand when LAB is
> > >> working on top of it.
> > >>
> > >> - The LAB operation is specified (and thereof extendable) via
> > >> device tree lab-ctrl-freq attribute defined at /cpus/cpu0.
> > >>
> > >>
> > >> Problems:
> > >> - How the governor will work for big.LITTLE systems (especially
> > >> Global Task Scheduling).
> > >> - Will there be agreement to expose internal ondemand code to be
> > >> reused for more specialized governors.
> > >>
> > >> Test HW:
> > >>       Exynos4412 - Trats2 board.
> > >> Above patches were posted on top of Linux 3.14-rc5
> > >> (SHA1: 3f9590c281c66162bf8ae9b7b2d987f0a89043c6)
> > >>
> > >
> > > Any comments about those patches?
> > 
> > Sorry for being late on reviewing these..
> > 
> > I tried to go through the patches but didn't looked at the minutest
> > of the details. Its been a long time when you first sent this
> > patchset. And the memories have corrupted by now :) ..
> 
> Unfortunately memory is volatile ... since LAB governor is a follow up
> of BOOST, which review and inclusion took considerable time, some
> details could be forgotten. 
> 
> > 
> > To get context back, can we discuss again the fundamentals behind
> > this new governor you are proposing. And then we can discuss about
> > it again, its pros/cons, etc..
> 
> Please consider following links:
> 
> The original implementation - threads:
> http://thread.gmane.org/gmane.linux.power-management.general/32523/match=lab
> http://thread.gmane.org/gmane.linux.kernel/1484746/match=lab
> 
> 
> LAB justification data:
> http://article.gmane.org/gmane.linux.kernel/1472381
> 
> 
> > People are reluctant in getting another governor in and want to give
> > existing governors a try if possible.
> 
> As I've stated in the covering letter, this code is an extension of
> Ondemand.
> 
> This is totally different from what have been posted previously (v1,
> v2).
> The first LAB proposal was written with some parts copied from
> Ondemand. It was a separate, standalone governor.
> 
> 
> The approach proposed in those patches is very different. It simply
> reuses Ondemand code as much as possible (timers, default attributes
> exported to sysfs, etc.).
> 
> On top of the Ondemand we have the LAB, which thereof is its optional
> extension. The existing code is reused and can be easily extracted as
> a common code.
> 
> > 
> > So, please explain the basics behind your governor again and then
> > we can put our arguments again..
> > 
> 
> I hope that provided overview is sufficient. More in depth
> information can be found in posted patches or provided LKML archives.
> 

Viresh, will you find time for reviewing this RFC in a near future?

> > --
> > viresh
> 



-- 
Best regards,

Lukasz Majewski

Samsung R&D Institute Poland (SRPOL) | Linux Platform Group

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Viresh Kumar]

On 24 March 2014 12:17, Lukasz Majewski <l.majewski@samsung.com> wrote:
> Viresh, will you find time for reviewing this RFC in a near future?

Yes. I have been trying hard last week but couldn't find some time
for it. Will try this week for sure.. Sorry to keep you waiting :(

@Rafael: Please see if you can also give them a look..

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Viresh Kumar]

On 4 March 2014 15:57, Lukasz Majewski <l.majewski@samsung.com> wrote:
> Despite this patch set is working and applicable on top of 3.14-rc5,
> please regard it solely as a pure RFC.

Okay, I am trying to do a review here and because you have mentioned
how different it is from the earlier versions, I am trying with a fresh mind.
i.e. with zero memories of earlier discussions :)

LAB was: Legacy Application Boost ??

Probably mention that in your new threads as well, so that new readers
know the details. Also, like other governors, just name it "boost" governor.

> This patch provides support for LAB governor build on top of ondemand.
> Previous version of LAB can be found here:
> http://thread.gmane.org/gmane.linux.kernel/1484746/match=cpufreq
>
> LAB short reminder:
>
> LAB uses information about how many cores are in "idle" state (the core
> idleness is represented as the value between 0 and 100) and the overall
> load of the system (from 0 to 100) to decide about frequency to be set.
> It is extremely useful with SoCs like Exynos4412, which can set only one
> frequency for all cores.

Probably a description of how exactly these two values come into play
would have been more interesting here for all. Always think of new followers
of your patchset and so add all interesting things about it when you resend
it.

If I remember well the logic was more or less like this:
- More idle cores means run few running cores at high frequency
- Less idle cores means don't run them at very high frequencies

Right?

What about making it as simple as:
- changing the ondemand governor only instead of adding a new governor
- Keeping the bahavior as is for all platforms not publishing boost frequencies
- If more cores are idle, enable switching to boost frequencies and take them
into consideration all the time.
- If less cores are idle, disable boost frequencies..

Lets discuss this first and then I will get into the very details of your
implementation.

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Lukasz Majewski]

Hi Viresh,

> On 4 March 2014 15:57, Lukasz Majewski <l.majewski@samsung.com> wrote:
> > Despite this patch set is working and applicable on top of 3.14-rc5,
> > please regard it solely as a pure RFC.
> 
> Okay, I am trying to do a review here and because you have mentioned
> how different it is from the earlier versions, I am trying with a
> fresh mind. i.e. with zero memories of earlier discussions :)
> 
> LAB was: Legacy Application Boost ??

Yes, correct.

> 
> Probably mention that in your new threads as well, so that new readers
> know the details. Also, like other governors, just name it "boost"
> governor.

I think, that "LAB" name is with us for some time, so it would be a
pity to discard it.

> 
> > This patch provides support for LAB governor build on top of
> > ondemand. Previous version of LAB can be found here:
> > http://thread.gmane.org/gmane.linux.kernel/1484746/match=cpufreq
> >
> > LAB short reminder:
> >
> > LAB uses information about how many cores are in "idle" state (the
> > core idleness is represented as the value between 0 and 100) and
> > the overall load of the system (from 0 to 100) to decide about
> > frequency to be set. It is extremely useful with SoCs like
> > Exynos4412, which can set only one frequency for all cores.
> 
> Probably a description of how exactly these two values come into play
> would have been more interesting here for all. Always think of new
> followers of your patchset and so add all interesting things about it
> when you resend it.
> 
> If I remember well the logic was more or less like this:
> - More idle cores means run few running cores at high frequency
> - Less idle cores means don't run them at very high frequencies
> 
> Right?

This is correct. Also, the underlying SoC - Exynos4412 has 4 cores with
option to set frequency only on all of them.

> 
> What about making it as simple as:
> - changing the ondemand governor only instead of adding a new governor

My goal is to not touch the ondemand code. It has matured, so I would
like to leave it as it is.

> - Keeping the bahavior as is for all platforms not publishing boost
> frequencies

This is also done - you get the LAB configuration specified in the DT
for the particular platform/board.

> - If more cores are idle, enable switching to boost frequencies and
> take them into consideration all the time.

I'm not sure if I have understood you, but something like that is also
performed in the code.

> - If less cores are idle, disable boost frequencies..

As written above.

> 
> Lets discuss this first and then I will get into the very details of
> your implementation.

Discussion about above functionalities requires consulting the
implementation to be sure that our opinions are the same.

-- 
Best regards,

Lukasz Majewski

Samsung R&D Institute Poland (SRPOL) | Linux Platform Group

[RFC v3 0/5] cpufreq:LAB: Support for LAB governor.

[Viresh Kumar]

[Adding Linaro lists in cc as there are few people here working on power/thermal
stuff.]

On 24 March 2014 15:30, Lukasz Majewski <l.majewski@samsung.com> wrote:
>> On 4 March 2014 15:57, Lukasz Majewski <l.majewski@samsung.com> wrote:

> I think, that "LAB" name is with us for some time, so it would be a
> pity to discard it.

It doesn't matter with Mainline how you do naming initially for your code :)
We need to pick the right name now, and the decision should be made
now (after discussions obviously) :)

>> What about making it as simple as:
>> - changing the ondemand governor only instead of adding a new governor
>
> My goal is to not touch the ondemand code. It has matured, so I would
> like to leave it as it is.

Because the boost feature is already part of CPUFreq core, I think its
better if we enhance current governors to use it. So, I would like to
make this part of existing governors. Not only ondemand but maybe
conservative as well..

Also, I feel we maynot necessarily move this piece of code into cpufreq.
All you are doing is thermal management here :)

If we are sure we will not burn out our SoC (When many cores are idle),
run at max freq (if there is enough load of course :))..

And if there are chances that we might burn our chip (when very few
cores are idle), don't run on boost frequencies..

This is actually a 'cooling' device :)

Think of it this way: CPUFreq will provide a range of frequency which
SoC's can use. And then based on some conditions we may or may not
want to run on these frequencies.

@Zhang/Eduardo: Can we have your inputs here as well ?

This may look hard but we need to design things in the best possible
way for managing things better in future. Lets see what others have
to say on this.