Colibri PXA320 Power management question

Colibri PXA320 Power management question

[Yuri Ludkevich]

Hi there :)

I have a question about power management of Colibri PXA320 module:
I found that bits PVE and FVE in PVCR register are disabled,
so at frequency change voltage will not change.

If I turn on PVE and FVE with pxaregs tool - voltage changes just fine.
Where are suitable place in the kernel for PVCR initialization?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20100812/2d01eea3/attachment-0001.html>

[Openpxa-users] Colibri PXA320 Power management question

[Marek Vasut]

Dne ?t 12. srpna 2010 12:02:17 Yuri Ludkevich napsal(a):
> Hi there :)
> 
> I have a question about power management of Colibri PXA320 module:
> I found that bits PVE and FVE in PVCR register are disabled,
> so at frequency change voltage will not change.
> 
> If I turn on PVE and FVE with pxaregs tool - voltage changes just fine.
> Where are suitable place in the kernel for PVCR initialization?

I believe kernel should handle all that. Eric CCed.

[Openpxa-users] Colibri PXA320 Power management question

[Eric Miao]

On Sat, Aug 14, 2010 at 2:53 PM, Marek Vasut <marek.vasut@gmail.com> wrote:
> Dne ?t 12. srpna 2010 12:02:17 Yuri Ludkevich napsal(a):
>> Hi there :)
>>
>> I have a question about power management of Colibri PXA320 module:
>> I found that bits PVE and FVE in PVCR register are disabled,
>> so at frequency change voltage will not change.
>>
>> If I turn on PVE and FVE with pxaregs tool - voltage changes just fine.
>> Where are suitable place in the kernel for PVCR initialization?
>
> I believe kernel should handle all that. Eric CCed.
>

Yep. Some of the design doesn't use the built-in power I2C at all,
they use some non-standard I2C power management IC. So this
should be handled in kernel indeed. The currently problem is that
we don't yet have a clean framework, some platforms are writing
directly to these registers to configure, which should be avoided.

[Openpxa-users] Colibri PXA320 Power management question

[Igor Grinberg]

On 08/15/10 05:06, Eric Miao wrote:
> On Sat, Aug 14, 2010 at 2:53 PM, Marek Vasut <marek.vasut@gmail.com> wrote:
>   
>> Dne ?t 12. srpna 2010 12:02:17 Yuri Ludkevich napsal(a):
>>     
>>> Hi there :)
>>>
>>> I have a question about power management of Colibri PXA320 module:
>>> I found that bits PVE and FVE in PVCR register are disabled,
>>> so at frequency change voltage will not change.
>>>
>>> If I turn on PVE and FVE with pxaregs tool - voltage changes just fine.
>>> Where are suitable place in the kernel for PVCR initialization?
>>>       
>> I believe kernel should handle all that. Eric CCed.
>>
>>     
> Yep. Some of the design doesn't use the built-in power I2C at all,
> they use some non-standard I2C power management IC. So this
> should be handled in kernel indeed. The currently problem is that
> we don't yet have a clean framework, some platforms are writing
> directly to these registers to configure, which should be avoided.
>   

If I recall correctly, the pxa3xx built-in power I2C control is designed
for use with 100% compatible PMIC chip and if for some reason
the hardware has another PMIC, which is not 100% compatible,
then the built-in control can do more harm then good.

I think we need a good and configurable DVFS driver to take care
of all the power management stuff.

> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>   

-- 
Regards,
Igor.

[Openpxa-users] Colibri PXA320 Power management question

[Eric Miao]

On Sun, Aug 15, 2010 at 5:13 PM, Igor Grinberg <grinberg@compulab.co.il> wrote:
> On 08/15/10 05:06, Eric Miao wrote:
>> On Sat, Aug 14, 2010 at 2:53 PM, Marek Vasut <marek.vasut@gmail.com> wrote:
>>
>>> Dne ?t 12. srpna 2010 12:02:17 Yuri Ludkevich napsal(a):
>>>
>>>> Hi there :)
>>>>
>>>> I have a question about power management of Colibri PXA320 module:
>>>> I found that bits PVE and FVE in PVCR register are disabled,
>>>> so at frequency change voltage will not change.
>>>>
>>>> If I turn on PVE and FVE with pxaregs tool - voltage changes just fine.
>>>> Where are suitable place in the kernel for PVCR initialization?
>>>>
>>> I believe kernel should handle all that. Eric CCed.
>>>
>>>
>> Yep. Some of the design doesn't use the built-in power I2C at all,
>> they use some non-standard I2C power management IC. So this
>> should be handled in kernel indeed. The currently problem is that
>> we don't yet have a clean framework, some platforms are writing
>> directly to these registers to configure, which should be avoided.
>>
>
> If I recall correctly, the pxa3xx built-in power I2C control is designed
> for use with 100% compatible PMIC chip and if for some reason
> the hardware has another PMIC, which is not 100% compatible,
> then the built-in control can do more harm then good.

Indeed. With some of those incompatible I2C power management ICs,
some are controlling the Power I2C manually with the bits turned off
in PVCR.

>
> I think we need a good and configurable DVFS driver to take care
> of all the power management stuff.
>

Yes.

Colibri PXA320 Power management question

[Mark Brown]

On Thu, Aug 12, 2010 at 02:02:17PM +0400, Yuri Ludkevich wrote:

> I have a question about power management of Colibri PXA320 module:
> I found that bits PVE and FVE in PVCR register are disabled,
> so at frequency change voltage will not change.

> If I turn on PVE and FVE with pxaregs tool - voltage changes just fine.
> Where are suitable place in the kernel for PVCR initialization?

The expectation in Linux is that voltage scaling will be done in
software using cpufreq rather than autonomously by the CPU.

Colibri PXA320 Power management question

[Mike Rapoport]

Mark Brown wrote:
> On Thu, Aug 12, 2010 at 02:02:17PM +0400, Yuri Ludkevich wrote:
> 
>> I have a question about power management of Colibri PXA320 module:
>> I found that bits PVE and FVE in PVCR register are disabled,
>> so at frequency change voltage will not change.
> 
>> If I turn on PVE and FVE with pxaregs tool - voltage changes just fine.
>> Where are suitable place in the kernel for PVCR initialization?
> 
> The expectation in Linux is that voltage scaling will be done in
> software using cpufreq rather than autonomously by the CPU.

PXA3xx has ability to automatically control certain regulators (almost) without 
software intervention. Setting the bits Yuri is talking about should enable CPU 
initiated i2c transactions upon frequency scaling events.

> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel


-- 
Sincerely yours,
Mike.

Colibri PXA320 Power management question

[Mark Brown]

On Thu, Sep 02, 2010 at 05:41:57PM +0300, Mike Rapoport wrote:
> Mark Brown wrote:

> >The expectation in Linux is that voltage scaling will be done in
> >software using cpufreq rather than autonomously by the CPU.

> PXA3xx has ability to automatically control certain regulators
> (almost) without software intervention. Setting the bits Yuri is
> talking about should enable CPU initiated i2c transactions upon
> frequency scaling events.

Yes, exactly - what I'm saying is that Linux is set up to use cpufreq
rather than rely on the built in stuff.

Colibri PXA320 Power management question

[Mike Rapoport]

Mark Brown wrote:
> On Thu, Sep 02, 2010 at 05:41:57PM +0300, Mike Rapoport wrote:
>> Mark Brown wrote:
> 
>>> The expectation in Linux is that voltage scaling will be done in
>>> software using cpufreq rather than autonomously by the CPU.
> 
>> PXA3xx has ability to automatically control certain regulators
>> (almost) without software intervention. Setting the bits Yuri is
>> talking about should enable CPU initiated i2c transactions upon
>> frequency scaling events.
> 
> Yes, exactly - what I'm saying is that Linux is set up to use cpufreq
> rather than rely on the built in stuff.

There's no contradiction here. The PXA3XX cpufreq implementation could trigger 
automatic voltage scaling.

-- 
Sincerely yours,
Mike.

[Openpxa-users] Colibri PXA320 Power management question

[Marek Vasut]

Dne ?t 2. z??? 2010 17:43:02 Mike Rapoport napsal(a):
> Mark Brown wrote:
> > On Thu, Sep 02, 2010 at 05:41:57PM +0300, Mike Rapoport wrote:
> >> Mark Brown wrote:
> >>> The expectation in Linux is that voltage scaling will be done in
> >>> software using cpufreq rather than autonomously by the CPU.
> >> 
> >> PXA3xx has ability to automatically control certain regulators
> >> (almost) without software intervention. Setting the bits Yuri is
> >> talking about should enable CPU initiated i2c transactions upon
> >> frequency scaling events.
> > 
> > Yes, exactly - what I'm saying is that Linux is set up to use cpufreq
> > rather than rely on the built in stuff.
> 
> There's no contradiction here. The PXA3XX cpufreq implementation could
> trigger automatic voltage scaling.

Considering the CPU and the regulator are compatible and you wont fry the CPU. 
In that case, you use cpufreq indeed.

Colibri PXA320 Power management question

[Mark Brown]

On Thu, Sep 02, 2010 at 06:43:02PM +0300, Mike Rapoport wrote:
> Mark Brown wrote:

> >Yes, exactly - what I'm saying is that Linux is set up to use cpufreq
> >rather than rely on the built in stuff.

> There's no contradiction here. The PXA3XX cpufreq implementation
> could trigger automatic voltage scaling.

That'd be slightly entertaining, though, since IIRC the hardware
implementation isn't really set up for being constrained - it doesn't
expect to be cooperating with software governors.

Colibri PXA320 Power management question

[Mike Rapoport]

Mark Brown wrote:
> On Thu, Sep 02, 2010 at 06:43:02PM +0300, Mike Rapoport wrote:
>> Mark Brown wrote:
> 
>>> Yes, exactly - what I'm saying is that Linux is set up to use cpufreq
>>> rather than rely on the built in stuff.
> 
>> There's no contradiction here. The PXA3XX cpufreq implementation
>> could trigger automatic voltage scaling.
> 
> That'd be slightly entertaining, though, since IIRC the hardware
> implementation isn't really set up for being constrained - it doesn't
> expect to be cooperating with software governors.

As far as I remember, PXA3xx power I2C can be setup to automatically send I2C 
command sequence to PMIC upon certain events, such as frequency change, power 
state transition and probably something else. You can enable this behavior and 
then when cpufreq decides to change frequency the voltage should be scaled 
appropriately. No idea, though, if anybody made it work.
Maybe Eric had tried it when he was with Marvell :)

-- 
Sincerely yours,
Mike.

Colibri PXA320 Power management question

[Eric Miao]

On Mon, Sep 6, 2010 at 6:13 PM, Mark Brown
<broonie@opensource.wolfsonmicro.com> wrote:
> On Thu, Sep 02, 2010 at 06:43:02PM +0300, Mike Rapoport wrote:
>> Mark Brown wrote:
>
>> >Yes, exactly - what I'm saying is that Linux is set up to use cpufreq
>> >rather than rely on the built in stuff.
>
>> There's no contradiction here. The PXA3XX cpufreq implementation
>> could trigger automatic voltage scaling.
>
> That'd be slightly entertaining, though, since IIRC the hardware
> implementation isn't really set up for being constrained - it doesn't
> expect to be cooperating with software governors.
>

It doesn't need to. There is actually an internal command table, and
not really sure about the mapping scheme though, but when software
then has only to care about the frequencies, and voltages will be
adjusted accordingly.

In the other hand, such hardcoded mechanism won't work very well
with incompatible power management ICs, so if there is such a
platform, we do need a way to fall back to using regulators.

> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>

Colibri PXA320 Power management question

[Mark Brown]

On Mon, Sep 06, 2010 at 07:09:44PM +0800, Eric Miao wrote:
> On Mon, Sep 6, 2010 at 6:13 PM, Mark Brown

> > That'd be slightly entertaining, though, since IIRC the hardware
> > implementation isn't really set up for being constrained - it doesn't
> > expect to be cooperating with software governors.

> It doesn't need to. There is actually an internal command table, and
> not really sure about the mapping scheme though, but when software
> then has only to care about the frequencies, and voltages will be
> adjusted accordingly.

Ah, I'm misremembering the PXA implementation for the i.MX one - the
i.MX version of this feature had a governor rather than just doing a
lookup from the frequency to a voltage.

Colibri PXA320 Power management question

[Haojian Zhuang]

On Mon, Sep 6, 2010 at 7:30 PM, Mark Brown
<broonie@opensource.wolfsonmicro.com> wrote:
> On Mon, Sep 06, 2010 at 07:09:44PM +0800, Eric Miao wrote:
>> On Mon, Sep 6, 2010 at 6:13 PM, Mark Brown
>
>> > That'd be slightly entertaining, though, since IIRC the hardware
>> > implementation isn't really set up for being constrained - it doesn't
>> > expect to be cooperating with software governors.
>
>> It doesn't need to. There is actually an internal command table, and
>> not really sure about the mapping scheme though, but when software
>> then has only to care about the frequencies, and voltages will be
>> adjusted accordingly.
>
> Ah, I'm misremembering the PXA implementation for the i.MX one - the
> i.MX version of this feature had a governor rather than just doing a
> lookup from the frequency to a voltage.
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>

I think that the key isn't scaling voltage. We can enable both
auto-scaling (hardware) and software-scaling voltage in cpufreq
driver. The key point is that device may occur fails while scaling
frequency in device working mode. I remember that cpufreq doesn't care
about this behavior.